WO2021082464A1

WO2021082464A1 - Method and device for predicting destination of vehicle

Info

Publication number: WO2021082464A1
Application number: PCT/CN2020/096004
Authority: WO
Inventors: 汪亮; 张亚楠; 朱林
Original assignee: 华为技术有限公司
Priority date: 2019-10-31
Filing date: 2020-06-14
Publication date: 2021-05-06
Also published as: CN112749825B; CN112749825A

Abstract

A method for predicting a destination of a vehicle, relating to the field of intelligent transportation. The method comprises: acquiring trajectory data of a vehicle awaiting destination prediction travelling in a traffic region and travel data of the vehicle, and acquiring destination information of the vehicle in the traffic region according to the trajectory data, the travel data, and a target neural network model. The destination information comprises: a target sub-region of the vehicle and a target point of interest (POI) type of the vehicle. The travel data of the vehicle comprises one or more pieces of the following data: a vehicle type, a travel weather type, the vehicle travel count in a first time period, the vehicle travel frequency in a second time period, and a vehicle travel sub-time period in a third time period. The method improves the efficiency and accuracy of predicting a destination of a vehicle.

Description

Method and device for predicting destination of vehicle

Technical field

This application relates to the field of smart transportation, and more specifically, to a method and device for predicting the destination of a vehicle.

Background technique

Transportation is the basic need and prerequisite for economic development, and it promotes the progress and development of modern equipment. However, with the rapid development of urban economy, the contradiction between residents' travel demand and urban road traffic supply capacity is increasing. The ensuing traffic congestion problem has become a global "urban disease".

Traffic congestion will not only lead to the decline of various functions of the city, but also increase the cost of travel time for residents, and reduce the quality of life of residents. In addition, traffic accidents, air pollution, noise impact and other related problems caused by traffic congestion have severely hindered the economic and social development of the city.

In order to improve the quality of urban traffic management and traffic services, to effectively alleviate traffic congestion and reduce the cost of public travel time, the destination of traveling vehicles can be known in advance for traffic warning and diversion. In the prior art, the method of obtaining the destination of a vehicle is to use a questionnaire survey. This method surveys a group of vehicle owners by searching for passing vehicles in a certain traffic area or sharing a questionnaire link on the Internet to obtain the destination information of the vehicle. The efficiency of the destination data obtained by the method is low, and it is greatly affected by time and area. Therefore, how to predict the destination of the vehicle is a technical problem that needs to be solved urgently.

Summary of the invention

The present application provides a method, device and computing device for predicting the destination of a vehicle, which can improve the efficiency of predicting the destination of a vehicle.

In the first aspect, this application provides a method for predicting the destination of a vehicle. The method can be applied to a traffic area in which multiple monitoring devices and multiple POIs are distributed. The method includes: obtaining the trajectory data of the vehicle to be predicted in the travel process and the travel data of the vehicle to be predicted in the traffic area; and obtaining the vehicle to be predicted according to the trajectory data, the travel data, and the target neural network model. Predict the destination information of the vehicle in the traffic area, where the destination information includes: the destination sub-region of the vehicle to be predicted and the type of POI of the destination point of interest of the vehicle to be predicted; the travel of the vehicle to be predicted The data includes one or more of the following data: vehicle type, travel weather type, number of vehicle trips in the first time period, vehicle travel frequency in the second time period, and vehicle travel sub-time periods in the third time period .

The method of the present application predicts the sub-region and type of the target POI of the vehicle to be predicted based on the trajectory data and travel data of the vehicle to be predicted and the target neural network model trained on the trajectory data and travel data of a large number of vehicles. To obtain the destination of the vehicle to be predicted, the efficiency and accuracy of predicting the destination of the vehicle can be improved.

In some possible implementation manners, the method further includes: determining, according to the destination information of the vehicle to be predicted, that the destination is the traffic volume corresponding to the type of POI in the destination sub-area; The flow rate predicts the traffic state of the road in the destination sub-area.

Further, traffic guidance can also be given according to the traffic state of the road to relieve traffic pressure.

In some possible implementations, the target neural network model includes an embedded model, a first feature extraction model, a second feature extraction model, a fusion model, a first classification model, and a second classification model, wherein the embedded model Used to vectorize the data input to the embedding model, and the first feature extraction model and the second feature extraction model are respectively used to extract the data input to the first feature extraction model and the second feature extraction model. Feature extraction is performed on the data of the model, the fusion model is used to perform feature fusion on the data input to the fusion model, and the first classification model and the second classification model are respectively used to perform feature extraction according to the first classification model and The input data of the second classification model performs category prediction.

In some possible implementation manners, the obtaining the destination information of the vehicle to be predicted in the traffic area according to the trajectory data, the travel data and the target neural network model includes: inputting the trajectory data And the travel data to the embedded model to obtain the initial trajectory feature and initial travel feature of the vehicle to be predicted; input the initial trajectory feature to the first feature extraction model to obtain the trajectory feature of the vehicle to be predicted Input the initial travel feature to the second feature extraction model to obtain the travel feature of the vehicle to be predicted; input the trajectory feature and the travel feature to the fusion model to obtain the travel of the vehicle to be predicted Characteristics; input the driving characteristics to the first classification model to obtain the target sub-area of the vehicle to be predicted; input the driving characteristics to the second classification model to obtain the type of the target POI of the vehicle to be predicted .

In these implementations, the trajectory data and travel data of the vehicle to be predicted are first mapped into multi-dimensional vectors, and then the mapped multi-dimensional vectors are input to the feature extraction model to extract trajectory features and travel features with deep semantics. The destination predicted by travel characteristics is more accurate.

In some possible implementations, the acquiring the trajectory data of the vehicle to be predicted in the traffic area during the current travel includes: determining that the vehicle to be predicted is currently traveling based on the passing data in the traffic area Information of multiple monitoring devices that have passed; and determining the trajectory data of the vehicle to be predicted according to the information of the multiple monitoring devices.

In some possible implementation manners, the method further includes: acquiring sub-region information in the traffic area; wherein the determining the trajectory data of the vehicle to be predicted according to the information of the multiple monitoring devices includes : Determine the trajectory data according to the sub-region information and the information of the multiple monitoring devices, where the trajectory data includes the information of the sub-regions to which the multiple monitoring devices belong.

In other words, the trajectory of the vehicle to be predicted is represented by the location information of the sub-region to which the monitoring device belongs. In this way, when each sub-area contains only one or fewer monitoring devices, less data can be used to characterize the trajectory of the vehicle to be predicted, thereby reducing the amount of data calculation and the complexity of data calculation, and further improving the pre-stored data. The efficiency of the destination of the vehicle.

In some possible implementation manners, the trajectory data further includes time information when the vehicle to be predicted passes through the multiple monitoring devices. These implementations use more information to predict the destination of the vehicle, which can improve the accuracy of the predicted destination.

In some possible implementation manners, the trajectory data further includes the POI types included in the sub-regions to which the multiple monitoring devices belong. These implementations use more information to predict the destination of the vehicle, which can improve the accuracy of the predicted destination.

In some possible implementation manners, the target neural network model is a neural network model trained by training data, and the training data includes historical trajectory data of vehicles in the traffic area and travel data of the vehicles.

In a second aspect, the present application provides a device for predicting the destination of a vehicle. The device is applied to a geographic traffic area in which multiple monitoring devices and multiple points of interest POI are distributed. The device includes: an acquisition module, It is used to obtain the trajectory data of the vehicle to be predicted in the travel process and the travel data of the vehicle to be predicted in the traffic area; the prediction module is used to obtain the data according to the trajectory data, the travel data and the target neural network model. The destination information of the vehicle to be predicted in the traffic area, where the destination information includes: the destination sub-area of the vehicle to be predicted and the type of the destination POI of the vehicle to be predicted; the vehicle to be predicted The travel data of includes one or more of the following data: vehicle type, travel weather type, number of vehicle trips in the first time period, vehicle travel frequency in the second time period, and vehicle travel in the third time period period.

The device can predict the destination sub-area and destination POI type of the vehicle based on the current travel trajectory data and travel data of the vehicle, so as to know the destination of the vehicle. Compared with knowing the destination of the vehicle manually, the prediction efficiency and accuracy can be improved.

In some possible implementations, the prediction module is further configured to: according to the destination information of the vehicle to be predicted, determine that the destination is the traffic volume corresponding to the type of POI in the destination sub-area; The traffic flow predicts the traffic state of the road in the destination sub-area.

In some possible implementations, the prediction module is specifically configured to: input the trajectory data and the travel data to the embedded model to obtain the initial trajectory characteristics and initial travel characteristics of the vehicle to be predicted; and input the The initial trajectory feature to the first feature extraction model to obtain the trajectory feature of the vehicle to be predicted; input the initial travel feature to the second feature extraction model to obtain the travel feature of the vehicle to be predicted; input the Trajectory characteristics and the travel characteristics to the fusion model to obtain the driving characteristics of the vehicle to be predicted; input the driving characteristics to the first classification model to obtain the target subregion of the vehicle to be predicted; input the The driving feature is transferred to the second classification model, and the target POI type of the vehicle to be predicted is obtained.

In these implementations, the trajectory data of the vehicle to be predicted is first mapped to a multi-dimensional vector, and then the mapped multi-dimensional vector is input to the feature extraction model to extract trajectory features with deep semantics, which can make the destination predicted based on the trajectory feature more accurate.

In some possible implementation manners, the acquisition module is specifically configured to: according to the passing data in the traffic area, determine the information of multiple monitoring devices that the vehicle to be predicted has passed through during the current trip; The information of multiple monitoring devices determines the trajectory data of the vehicle to be predicted.

In some possible implementations, the acquisition module is specifically configured to: acquire sub-area information in the traffic area; determine the trajectory data according to the sub-area information and the information of the multiple monitoring devices, so The trajectory data includes information about the sub-regions to which the multiple monitoring devices belong.

In some possible implementation manners, the trajectory data further includes time information when the vehicle passes through each of the at least one location. These implementations use more information to predict the destination of the vehicle, which can improve the accuracy of the predicted destination.

In a third aspect, a computing device is provided. The computing device includes a processor and a memory, where computer instructions are stored in the memory, and the processor executes the computer instructions to implement the methods of the first aspect and possible implementation manners thereof.

In a fourth aspect, a computer-readable storage medium is provided, which is characterized in that the computer-readable storage medium stores computer instructions, and when the computer instructions in the computer-readable storage medium are executed by a computing device, the computing device executes the first Aspects and possible implementation manners thereof, or enable a computing device to implement the functions of the above-mentioned second aspect and possible implementation manners of the apparatus.

In a fifth aspect, a computer program product containing instructions is provided, which when running on a computing device, causes the computing device to execute the above-mentioned first aspect and its possible implementation methods, or causes the computing device to implement the above-mentioned second aspect The function of the device and its possible implementations.

Description of the drawings

In order to more clearly illustrate the technical methods of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments.

FIG. 1 is a schematic flowchart of a method for predicting the destination of a vehicle according to this application;

Fig. 2 is a schematic flow chart of a method for obtaining trajectory data of a vehicle according to this application;

FIG. 3 is another schematic flow chart of the method for obtaining vehicle trajectory data according to this application;

Figure 4 is a schematic structural diagram of the target neural network model of the application;

FIG. 5 is a schematic flowchart of a method for obtaining destination information of a vehicle according to this application;

FIG. 6 is another schematic flowchart of a method for obtaining destination information of a vehicle according to this application;

Fig. 7 is another schematic structural diagram of the target neural network model of the application;

FIG. 8 is another schematic flowchart of a method for obtaining destination information of a vehicle according to this application;

FIG. 9 is a schematic structural diagram of a device for predicting the destination of a vehicle in this application;

FIG. 10 is another schematic structural diagram of the device for predicting the destination of a vehicle in this application;

FIG. 11 is a schematic structural diagram of a computing device for predicting the destination of a vehicle in this application;

FIG. 12 is a schematic architecture diagram of a system to which the apparatus of the embodiment of the present application can be applied;

FIG. 13 is a schematic deployment diagram of a device to which an embodiment of the present application can be applied;

FIG. 14 is a schematic flow chart of the application for obtaining initial trajectory features;

Fig. 15 is another schematic flow chart for obtaining the initial trajectory feature according to the present application.

Detailed ways

The solutions in the embodiments provided in this application will be described below in conjunction with the drawings in this application.

To facilitate understanding, an explanation of concepts related to the present application is given below.

1. Point of interest (POI)

POI is a place that people are interested in and frequent in daily life. Generally speaking, a POI can be described from three aspects: name, location and type.

The name of the POI is used to identify the POI to distinguish it from other POIs. The type of POI is usually the result of dividing the POI according to the function or purpose of the POI. The location of the POI is usually expressed by the longitude and latitude of the location of the POI.

The types of POI can include: government departments, gas stations, department stores, supermarkets, restaurants, hotels, convenience stores, hospitals, tourist attractions, stations, parking lots, etc.

Examples of POI names include: Tiananmen Square, Oriental Pearl Tower, Terracotta Warriors and Horses, Wangfujing Department Store, and Baiyun Airport. Each POI name corresponds to one POI type. One POI type can correspond to multiple POI names, such as: Tiananmen Square, Oriental Pearl Tower, The POI type corresponding to the Terracotta Warriors and Horses of Qin Shihuang is "tourist attraction"; the POI type corresponding to Wangfujing Department Store is "shopping mall"; the POI type corresponding to Baiyun Airport is "transportation".

2. Monitoring system

The monitoring system is a system that monitors the driving information of vehicles in the traffic area, and further processes the driving information of the vehicles to obtain monitoring data.

The monitoring system includes monitoring equipment and processing systems. In the embodiments of the present application, the data obtained from the monitoring system is called monitoring data, and the monitoring data includes traffic passing data of multiple intersections or multiple road sections. The passing data of each intersection or each road section is the data recorded by the monitoring equipment installed at the intersection or the road section and analyzed by the processing system.

The passing data of a monitoring device includes the license plate information and model information of the vehicle passing the location of the monitoring device within a period of time, the time information when the monitoring device captures the vehicle, and the location information of the location of the monitoring device (such as latitude and longitude information) And the number information of the monitoring device. Wherein, the location information of the location where the monitoring device is located can also be understood as the location information of the location where the vehicle passes, and the time information of the vehicle captured by the monitoring device can also be understood as the time information of the vehicle passing the location.

The monitoring system in the embodiment of the present application may be a bayonet monitoring system. The bayonet monitoring system is used to monitor vehicles passing through specific places in the traffic area (such as toll stations, traffic or public security checkpoints, intersections, road sections, etc.). The bayonet monitoring system includes a bayonet device and a processing system. The bayonet device is set at a certain position of an intersection or road section to monitor vehicles passing by that position. The bayonet device is a device that can capture images or images, such as Camera, or camera, etc.; the processing system can obtain images or images captured by the bayonet device, and use deep learning algorithms to identify the license plate, model, and number of vehicles in the image or image captured by the bayonet device, and can also record the elapsed time And other information. The processing system can be a software system running on a computing device. The processing system can be deployed in a server close to the bayonet device or on a remote server. The data processed by the processing system in the bayonet monitoring system can be used as the monitoring data of the bayonet monitoring system.

In a traffic area, bayonet devices can be installed only at some intersections, such as trunk road sections in the traffic area, road sections prone to traffic jams, road sections with intensive accidents, and key road junctions. The bayonet device installed at an intersection can capture vehicles passing through all lanes of the intersection. For example, the angle of view (shooting range) of the bayonet device at the intersection can cover all lanes of the intersection; the bayonet device installed at the intersection can also It is possible to only photograph the vehicles passing through a part of the lane of the intersection. For example, the angle of view (the shooting range) of the bayonet device at the intersection may only cover the lane in the direction of the intersection.

It should be noted that, in the embodiments of the present application, the monitoring system is a bayonet monitoring system as an example for description. In fact, the monitoring system can also be an electronic police system, which can monitor vehicles passing through intersections in a traffic area, identify vehicle information, and further determine possible traffic violations and traffic accidents.

The electronic police system includes electronic police monitoring equipment and an analysis and processing system. The content of the data recorded by the electronic police monitoring equipment is similar to the content of the data captured by the bayonet device. The analysis and processing system analyzes and processes the data and the processing system of the bayonet monitoring system. The processed data is also similar. Specifically, the data analyzed and processed by the analysis and processing system can also include the license plate of the vehicle passing the intersection where the electronic police monitoring device is located, the recorded elapsed time and the entrance lane, and can also include the vehicle model, one or The number of vehicles passing through the intersection where the electronic police monitoring equipment is located in multiple time periods; the monitoring data of the electronic police system includes the data after the analysis and processing system analyzes and processes the data recorded by multiple electronic police monitoring equipment.

As a possible implementation manner, the data analyzed and processed by the analysis and processing system in the electronic police monitoring system and the data processed by the processing system of the bayonet monitoring system may be correspondingly merged, and the merged data may be used as monitoring data.

In the embodiments of this application, the monitoring system is the bayonet monitoring system as an example. For the monitoring system is an electronic police system (correspondingly, the monitoring data is the monitoring data of the electronic police system), or the monitoring system is monitored by the bayonet The situation of the system formed by the combination of the system and the electronic monitoring system (correspondingly, the monitoring data is the fused monitoring data) is similar to the situation where the monitoring system is a bayonet monitoring system, and will not be repeated here.

3. Parking lot data

The parking lot data refers to the parking records of the parking lot of each POI or the parking lot near each POI. For example, a camera at a parking lot bayonet can collect the parking data of the parking lot. The parking lot data can include: the license plate information of the vehicle, the time when the vehicle enters the parking lot, the time when the vehicle leaves the parking lot, and the parking duration, within a period of time The number of vehicles entering, the number of vehicles leaving within a period of time, the remaining number of vehicles that can be accommodated in the parking lot, etc.

4. Road traffic status

There are many ways to divide the traffic state of the road. For example, the road traffic state can be divided into three states: congested, slow, and unblocked, or the road traffic state can be divided into unblocked, lightly congested, congested, and severely congested.

5. Neural network model

Neural network model is a kind of mathematical calculation model that imitates the structure and function of biological neural network (animal's central nervous system). A neural network model can be composed of a combination of multiple sub-neural network models. Neural network models with different structures can be used in different scenes (for example, classification, recognition or image segmentation) or provide different effects when used in the same scene. Different neural network model structures specifically include one or more of the following: the number of network layers in the neural network model is different, the order of each network layer is different, and the weights, parameters or calculation formulas in each network layer are different.

There are many different neural network models with high accuracy for application scenarios such as weather prediction, image content prediction, and event probability prediction in the industry. Among them, some neural network models can be trained by a specific training set to complete a task alone or combined with other neural network models (or other functional modules) to complete a task. Some neural network models can also be used directly to complete a task alone or in combination with other neural network models (or other functional modules) to complete a task.

In real life, when a vehicle is traveling, it usually has a clear POI, such as going to a hospital in a certain area, or going to a primary school to send a child to school, or shopping in a mall. If during the journey of each vehicle, it is possible to predict in advance the destination that the vehicle will go to, and predict which sub-area in the traffic area can be reached and the POI type of the vehicle’s destination, then the entire vehicle can be predicted. The number of vehicles arriving at the same POI in the traffic area. Further, the future traffic state of the road near the POI can be predicted based on the number and the road network data near the POI, and the traffic management and prompts can be performed early based on the predicted future traffic state of the road.

Therefore, this application proposes a method for predicting the destination of a vehicle, by which the destination information of the traveling vehicle can be obtained in advance. In this application, the destination information of the vehicle includes the destination sub-area of the vehicle and the vehicle. The purpose of the POI type. In the method of predicting the destination of the vehicle, a neural network model that has been trained is used, called the target neural network model, based on the travel data (trajectory data and/or travel data) that the current traveling vehicle has generated during the trip. , Predict the destination sub-area and the type of the destination POI of the current traveling vehicle to achieve the purpose of predicting the destination of the current traveling vehicle. This method can improve the prediction accuracy and predicted speed of the destination of the vehicle. Further, the method can also predict the road traffic status of each sub-area in the traffic area based on the destinations of multiple traveling vehicles in the entire traffic area, so that the traffic management department can timely warn the road traffic status of the traffic area. And regulation.

Fig. 1 is a schematic flowchart of a method for predicting the destination of a vehicle according to the present application. The method may include S110 to S120. The device that executes this method is called a prediction device.

S110: Acquire trajectory data of a vehicle to be predicted in a traffic area, where the trajectory data includes location information of locations that the vehicle to be predicted has passed during this trip. The location information may be latitude and longitude information. The vehicle to be predicted is a traveling vehicle that is in the process of driving and has not yet reached the destination. The vehicle to be predicted that can be used for destination prediction in this solution is usually a vehicle that has passed several monitoring equipment, that is, a vehicle that has formed a driving track, for example: the driving track of the vehicle can be judged in the process of driving It is determined that the vehicle whose position information included in the driving track is greater than the preset threshold value can be predicted by this solution, that is, this kind of vehicle can be called the vehicle to be predicted. It should be understood that there are usually multiple vehicles to be predicted in a traffic area. The method for predicting the destination of a vehicle in this application can be executed on multiple vehicles to be predicted in the traffic area. For ease of understanding, this application will subsequently predict a vehicle to be predicted. Predict the destination of the vehicle as an example, and describe the method.

S120. Obtain destination information of the vehicle to be predicted in the traffic area according to the trajectory data and the target neural network model, where the destination information includes the destination sub-region of the vehicle to be predicted and the vehicle to be predicted The type of POI for the purpose of the vehicle. The POI of this type in the destination sub-area is the destination of the vehicle to be predicted.

According to the trajectory data and the target neural network model, obtaining the target sub-region and the target POI type of the vehicle to be predicted can be understood as: inputting the trajectory data to the target neural network model; obtaining all the output of the target neural network model Describe the target sub-area of the vehicle to be predicted and the type of the target POI.

The above step S110 will be described in detail below. Fig. 2 is an exemplary flow chart of a method for obtaining trajectory data of a vehicle to be predicted in this application. The method shown in FIG. 2 includes S210 to S220.

S210: Acquire passing traffic data in the traffic area.

For example, the predicting device receives the passing data in the traffic area periodically sent by the monitoring system, and the passing data in the traffic area includes the passing data recorded by multiple monitoring devices in the traffic area.

For another example, the prediction device sends a request message to the monitoring system to request the passing data in the traffic area, and the request message carries the name or area identification information of the traffic area. After receiving the request message, the monitoring system sends the passing data in the traffic field to the prediction device.

S220: Determine trajectory data of the vehicle to be predicted according to the passing data.

Since the passing data recorded by each monitoring device in the traffic area can include the license plate information, model information of the vehicle passing the location of the monitoring device within a period of time, the time information when the monitoring device captures the vehicle, and the location of the monitoring device. Location information (such as latitude and longitude information) and number information of the monitoring device. According to the passing data, the trajectory data of the vehicle to be predicted can be determined.

In the embodiment of the present application, the trajectory data of the vehicle to be predicted may include a variety of information, for example: 1. The trajectory data of the vehicle to be predicted includes: location information or grids of sub-regions in the traffic area that the vehicle to be predicted passes through Number; 2. The trajectory data of the vehicle to be predicted includes: the position information or grid number of the sub-area in the traffic area that the vehicle to be predicted passes through, and the time information when the vehicle to be predicted passes through one or more of these locations; 3. The trajectory data of the predicted vehicle includes: the position information or grid number of the sub-area in the traffic area that the vehicle to be predicted passes through, and the type of POI that the vehicle to be predicted passes through; 4. The trajectory data of the vehicle to be predicted includes: the vehicle to be predicted passes through Location information or grid numbers of sub-regions in the traffic area, the type of POI that the vehicle to be predicted passes through, and time information when the vehicle to be predicted passes one or more of these locations.

The following describes how to determine the trajectory data of the vehicle to be predicted based on the passing data.

Fig. 3 is an exemplary flow chart of an implementation method of determining the trajectory data of the vehicle to be predicted according to the passing data. The method shown in FIG. 3 includes S310 to S330.

S310: Acquire location information of sub-areas in the traffic area.

In some possible implementations, the prediction device divides the map covering the traffic area into grids with a specified accuracy or a specified number. The area covered by a grid is a sub-area, and the center point of each grid covers the location. Location information (for example, latitude and longitude) indicates the location information of the sub-region corresponding to the grid. The position information of multiple sub-areas in the traffic area forms a position information sequence.

The prediction device can use artificial division, Geohash method or other methods to divide the map of the traffic area.

The division accuracy of each grid can be determined by the application's accuracy requirements for the predicted target sub-region and the overall area of the traffic area. For example, the map of the traffic area can be divided into 10,000-meter, kilometer, or hundred-meter-level grids.

If the historical passing frequency in the sub-region corresponding to a grid is low, and the historical passing frequency in the sub-region corresponding to the adjacent grid is high, the grid can be merged with its adjacent grid, that is, it can be used. The center point of the adjacent grid is used as the center point of the grid. The less and more here can be based on a threshold. The threshold may be set according to the historical passing frequency of each grid after the historical passing frequency in the sub-region corresponding to each grid is counted. For example, the historical passing frequency may be sorted, and the number of the nth historical passing frequency in the historical passing frequency sorting is taken as the threshold, where n is a positive integer greater than 0.

When the map of the traffic area is gridded, only part of the map corresponding to the area where the vehicle has traveled can be gridded.

Under normal circumstances, when dividing the traffic area, you can try to meet the requirement that only one monitoring device is deployed in a sub-area, so as to ensure that different position information in the trajectory data of the vehicle to be predicted falls in different sub-areas. More meaningful trajectory features can be extracted from the trajectory data, and finally the accuracy of prediction can be improved.

In other possible implementations, the prediction device does not need to perform grid division of the traffic area, and the prediction device sends a message requesting location information of the sub-area to other devices, and the message may carry the name or area identification information of the traffic area. After receiving the message, other devices send the location information of the sub-area in the traffic area to the prediction device. For another example, the location information of the sub-area in the traffic area can be manually copied to the prediction device.

S320: Determine initial trajectory data of the vehicle to be predicted according to the passing data.

In some possible implementations, the prediction device obtains the location information (such as latitude and longitude information) of the target monitoring device from the passing data and the time information of the vehicle to be predicted recorded by the target monitoring device. The target monitoring device refers to the recorded vehicle. The monitoring equipment of the vehicle to be predicted (for example, the license plate number of the vehicle to be predicted); the time information of all the target monitoring equipment records the vehicle to be predicted is arranged in chronological order. Accordingly, the location information of all the target monitoring equipment is The target monitoring equipment records the time sequence of the target vehicle; according to the difference between the time indicated by the two adjacent time information in the time information sequence, the initial trajectory of the vehicle to be predicted for this trip is obtained from the position information sequence data.

Specifically, assuming that the length of the time information sequence is n, the following operations are performed to obtain the trajectory data of the current trip of the vehicle to be predicted from the position information sequence: (1) Initialize i=n; (2) Take out the time information The i-th time information and the i-1th time information in the sequence; (3) Calculate the difference between the times indicated by the two time information; (4) If the difference is less than or equal to the preset time threshold , Then calculate i=i-1, and repeat (2) and (3); (5) If the difference is greater than the preset time threshold, the behavior of the vehicle to be predicted passing the position corresponding to the i-th position information is explained, and The behavior that passes through the position corresponding to the i-1th position information does not belong to the same trip, that is, the position corresponding to the i-th position information can be regarded as the starting point of the current trip of the vehicle to be predicted, so it is obtained from the position information sequence From the i-th to the n-th position information, the initial trajectory data of the vehicle to be predicted is obtained. If there is no situation where the difference is greater than the preset time threshold after i is set to 2, it means that all the position information in the position information sequence constitutes the initial trajectory data of the vehicle to be predicted.

The above-mentioned time threshold may be determined according to the average speed of the vehicle to be predicted and the driving distance between the two target monitoring devices. For example, assuming that the driving distance between two target monitoring devices is 30 kilometers or 15 kilometers (there are two different driving routes), and the average speed of the vehicle to be predicted in the traffic area is 10 kilometers per hour, the time threshold can be The preset is 1.5 hours to 3 hours.

S330: Determine the trajectory data of the vehicle to be predicted according to the location information of the sub-area in the traffic area and the initial trajectory data of the vehicle to be predicted.

In a possible implementation manner, each position information in the initial trajectory data of the vehicle to be predicted is replaced with the position information of the subregion to which the position indicated by the position information belongs, so as to obtain the trajectory data of the vehicle to be predicted. Wherein, the position information in the initial trajectory data is actually the position information of the monitoring device that captured the vehicle to be predicted.

Optionally, after obtaining the trajectory data of the vehicle to be predicted, the trajectory data of the vehicle to be predicted may be deduplicated. That is, search for multiple adjacent and identical position information in the trajectory data of the vehicle to be predicted, delete the repeated position information in the multiple position information, and leave only one of them. This can reduce the amount of data, which can improve forecasting efficiency.

In another implementation manner, the position information sequence formed by the position information of the sub-areas in the traffic area is converted into a grid sequence.

For example, change

Converted to (g ₁ ,…,g _i ,…g _n ), where,

Represents the latitude and longitude coordinates of the i-th subregion, g _i represents the grid number of the i-th grid in the grid sequence, i is taken from 1 to n, and n is equal to the number of subregions.

After obtaining the grid sequence corresponding to the traffic area, each position information in the initial trajectory data of the vehicle to be predicted is replaced with the corresponding grid serial number, thereby obtaining the trajectory data of the vehicle to be predicted. Wherein, the grid sequence number corresponding to each location information refers to the grid sequence number corresponding to the sub-region to which the location indicated by the location information belongs. Since the grid sequence number can be represented by more concise information than the position information, indicating the trajectory of the vehicle to be predicted by the grid sequence number can reduce the amount of data calculation, thereby improving the prediction efficiency.

Optionally, after obtaining the trajectory data of the vehicle to be predicted, the trajectory data of the vehicle to be predicted may be deduplicated. That is, search for multiple adjacent and identical grid serial numbers in the trajectory data of the vehicle to be predicted, delete the repeated grid serial numbers among the multiple grid serial numbers, and keep only one of them. This can reduce the amount of data, which can improve forecasting efficiency.

In other implementation manners, the prediction device may directly use the initial trajectory data of the vehicle to be predicted as the trajectory data of the vehicle to be predicted.

In other implementation manners, after the prediction device determines the initial trajectory data of the vehicle to be predicted based on the passing data, it may determine the trajectory data of the vehicle to be predicted based on the initial trajectory data and the time information sequence corresponding to the current trip.

For example, the initial trajectory data and the time information sequence corresponding to the current trip form the trajectory data of the vehicle to be predicted.

For another example, the position information in the initial trajectory data is replaced with the position information of the subregion, and the position information sequence after the replacement and the time information sequence corresponding to the current trip form the trajectory data of the vehicle to be predicted.

For another example, the position information in the initial trajectory data is replaced with the grid serial number of the subregion, and the replaced grid serial number sequence and the time information sequence corresponding to the current trip form the trajectory data of the vehicle to be predicted.

For another example, replace the position information in the initial trajectory data with the position information of the subregion, and combine the position information sequence after the replacement and the first time information in the time information sequence corresponding to this trip to form the trajectory data of the vehicle to be predicted .

For another example, replace the position information in the initial trajectory data with the grid number of the sub-region, and combine the replaced grid number sequence and the first time information in the time information sequence corresponding to this trip to form the vehicle to be predicted Track data.

In other implementations, the prediction device may also obtain the correspondence between the POI and the POI type in the traffic area, and determine the trajectory data of the vehicle to be predicted based on the correspondence and the initial trajectory data of the vehicle to be predicted.

For example, the position information in the initial trajectory data is replaced with the position information of the sub-region, and the position information sequence after the replacement and the type of POI in each sub-region are combined into the trajectory data of the vehicle to be predicted.

For another example, the position information in the initial trajectory data is replaced with the grid serial number of the subregion, and the replaced grid serial number sequence and the type of POI in each subregion are combined into the trajectory data of the vehicle to be predicted.

For another example, replace the position information in the initial trajectory data with the position information of the sub-area, and replace the position information sequence after the replacement, the first time information in the time information sequence corresponding to this trip, and the information in each sub-area. The type of POI constitutes the trajectory data of the vehicle to be predicted.

For another example, replace the position information in the initial trajectory data with the grid number of the sub-area, and replace the grid sequence after the replacement, the first time information in the time information sequence corresponding to this trip, and the first time information in each sub-area. The type of POI constitutes the trajectory data of the vehicle to be predicted.

Each position information or each grid number in the trajectory data of the vehicle to be predicted may correspond to one or more POI types.

When the trajectory data to be predicted includes time information, the time information in the initial trajectory data can be converted from the format of year, month, day, hour, and minute to the format of month, week, day, hour, and hour. For example, 17:36 on December 1, 2018 can be expressed as [12,6,1,17,3], where "12" in "[]" means December, "6" means Saturday, and "1 "Means the 1st, "17" means 17:00, and "3" means 36 minutes is the third time in an hour. Month, week, day, hour, and moment can be called time elements of time information.

The following describes the implementation manner for the prediction device to obtain the correspondence between the POI and the POI type in the traffic area.

In one implementation, the prediction device can first obtain information about all POIs in the traffic area, and then use K-means clustering algorithm, hierarchical clustering algorithm, density-based clustering algorithm, Gaussian mixture model clustering algorithm, or mean shift Any one of the clustering algorithms performs clustering processing on all POIs in the traffic area, and establishes the corresponding relationship between the POI type and the POI; and then stores the corresponding relationship between the POI type and the POI. For example, the POIs used for accommodation in hotels, guesthouses, and inns can be clustered into one category, and the POIs used for providing cooked food such as Chinese restaurants, western restaurants, fast food restaurants, etc. can be clustered into one category.

In other possible implementation manners, the prediction device may obtain the correspondence between the POI and the POI type in the traffic area from other equipment. For example, the prediction device sends a message requesting the corresponding relationship to other devices, and the message may carry the name of the traffic area or the area identification information. After receiving the message, other devices send the corresponding relationship to the prediction device.

Fig. 4 is an exemplary structure diagram of the target neural network model of the application. As shown in Figure 4, the target neural network model of the present application may include an embedded model, a first feature extraction model, a first classification model, and a second classification model. The embedded model is used for vector mapping to obtain a multi-dimensional vector; The feature extraction model is used to obtain the trajectory features of the vehicle to be predicted; the first classification model is used to output the target subregion of the vehicle to be predicted according to the trajectory feature; the second classification model is used to output the target POI of the vehicle to be predicted according to the trajectory feature Types of.

An embedding layer can be included in the embedding model. The first feature extraction model can include any one of a long short term memory (LSTM) network, a bidirectional recurrent neural network (BRNN), and a memory network (Memory Networks). The first classification model or the second classification model may be an artificial neural network model. For example, the first classification model or the second classification model is an artificial neural network model that only includes a fully connected layer and an activation function.

With regard to the target neural network model shown in FIG. 4, a method for obtaining the target sub-region and the target POI type of the vehicle to be predicted in the aforementioned S120 of the present application will be introduced below in conjunction with FIG. 5. The method shown in FIG. 5 includes S510 to S540.

S510: Acquire an initial trajectory feature of the vehicle to be predicted according to the trajectory data and the embedded model of the vehicle to be predicted. An exemplary implementation of this step will be introduced in the subsequent content.

S520: Acquire the trajectory feature of the vehicle to be predicted according to the initial trajectory feature of the vehicle to be predicted and the first feature extraction model.

For example, the initial trajectory feature of the vehicle to be predicted is input into the first feature extraction model, and the feature output by the first feature extraction model can be used as the trajectory feature of the vehicle to be predicted.

S530: Obtain a target sub-region of the vehicle to be predicted according to the trajectory feature output by the feature extraction model and the first classification model.

For example, the trajectory feature output by the feature extraction model is input to the first classification model, and the first classification model outputs the target sub-region of the vehicle to be predicted.

S540: Acquire the target POI type of the vehicle to be predicted according to the trajectory feature output by the feature extraction model and the second classification sub-model.

For example, the trajectory feature output by the feature extraction model is input into the second classification model, and the second classification model outputs the type of the target POI of the vehicle to be predicted.

The following introduces several different implementation methods for obtaining the initial trajectory characteristics of the vehicle to be predicted according to the trajectory data of the vehicle to be predicted and the embedded model when the trajectory data of the vehicle to be predicted includes different information.

If the trajectory data of the vehicle to be predicted includes position information or grid serial numbers, the prediction device may first input the position information or grid serial numbers in the trajectory data of the vehicle to be predicted into the first embedding layer in the embedding model, and the first embedding layer pairs The position information or grid sequence number is mapped to obtain multiple multi-dimensional vectors.

Generally speaking, the dimension of the vector obtained by the mapping is preset, and the dimensions of the multiple vectors obtained from the mapping of the trajectory data of the vehicle to be predicted are all the same.

For example, if the trajectory data of the vehicle to be predicted includes n pieces of position information, and each position information is mapped to a v-dimensional vector, the trajectory data of the vehicle to be predicted can be mapped to n vectors, and these n vectors can form a n* The matrix of v, m and v are both positive integers.

For example, if the trajectory data of the vehicle to be predicted includes n grid numbers, and each grid number is mapped to a v-dimensional vector, the trajectory data of the vehicle to be predicted can be mapped to n vectors, and these n vectors can form one For the matrix of n*v, both m and v are positive integers.

After using the first embedding layer to obtain multiple vectors of the vehicle to be predicted, the multiple vectors can be merged to obtain the spatial feature vector of the vehicle to be predicted, and the spatial feature vector can be used as the initial trajectory feature of the vehicle to be predicted. For example, these multiple vectors can be spliced together in order to obtain the spatial feature vector of the vehicle to be predicted. For another example, a dot multiplication operation can be performed on these multiple vectors, and the result of the dot multiplication can be used as the spatial feature vector of the vehicle to be predicted.

As shown in Figure 14, the n grid numbers "g ₁ , ..., g _i , ..., g _n "in the trajectory data of the vehicle to be predicted are input into the first embedding layer in the embedding model, and the vectors "[a ₁₁ …a _1j …a _1n ]”…“[a _i1 …a _ij …a _in ]”…“[a _n1 …a _nj …a _nn ]”, where i and j are positive integers less than or equal to n; The vector “[a ₁₁ …a _1j …a _1n ]”…“[a _i1 …a _ij …a _in ]”…“[a _n1 …a _nj …a _nn ] stitched together to get the initial trajectory features of the vehicle to be predicted" a ₁₁ …a _1j …a _1n …a _i1 …a _ij …a _in …a _n1 …a _nj …a _nn .

If the trajectory data of the vehicle to be predicted includes position information or grid serial number, and time information, the prediction device may first input the position information or grid serial number in the trajectory data of the vehicle to be predicted into the first embedding layer in the embedding model to Obtain the spatial feature vector; and input each time element in the time information into the second embedding layer to the sixth embedding layer in the embedded model to obtain the temporal feature vector of the vehicle to be predicted; and combine the spatial feature vector and the temporal feature vector Fusion is the initial trajectory feature of the vehicle to be predicted.

The position information or the grid sequence number in the trajectory data of the vehicle to be predicted is input into the first embedding layer in the embedding model to obtain the implementation of the spatial feature vector, as described above, and will not be repeated here.

The following describes how to input each time element in the time information into the second embedding layer to the sixth embedding layer in the embedded model to obtain the time feature vector of the vehicle to be predicted.

For each time information, the "month" time element is input to the second embedding layer, and the second embedding layer outputs a multi-dimensional vector; the "week" time element is input to the third embedding layer, and the third embedding layer outputs a multi-dimensional vector; "day" time The element is input to the fourth embedding layer, and the fourth embedding layer is mapped to obtain a multi-dimensional vector; the “time” time element is input to the fifth embedding layer, and the fifth embedding layer is mapped to obtain a multi-dimensional vector; The six embedding layer outputs a multi-dimensional vector.

It is understandable that the dimensions of the vectors output by the second, third, fourth, fifth, and sixth embedding layers can be preset, and the vectors output by these five embedding layers The dimensions of can be the same or different.

After the above five embedding layers output five vectors corresponding to one time information, these five vectors can be fused to obtain a time feature vector of the vehicle to be predicted.

For example, the five vectors can be spliced together in order to form a time feature vector of the vehicle to be predicted; or, the five vectors can be dot-multiplied, and the result of the operation can be used as a time feature vector. It should be noted that the dimensions of these five vectors must be the same when performing dot multiplication operations.

After the prediction device obtains the time feature vector of the vehicle to be predicted, it can fuse the space feature vector and the time feature vector of the vehicle to be predicted to obtain the initial trajectory feature of the vehicle to be predicted.

Among them, when the time feature vector of the vehicle to be predicted is one, the spatial feature vector and time feature vector of the vehicle to be predicted can be spliced together to obtain the initial trajectory feature of the vehicle to be predicted; or, the spatial feature vector of the vehicle to be predicted can be obtained. The point multiplication operation is performed with the time feature vector, and the result of the operation is the initial trajectory feature of the vehicle to be predicted. This method requires that the dimensions of the space feature vector and the time feature vector are the same.

When there are multiple temporal feature vectors of the vehicle to be predicted, the spatial feature vector and the multiple temporal feature vectors can be spliced in sequence to obtain the initial trajectory feature of the vehicle to be predicted; or, the multiple temporal feature vectors can be selected first. Multiplication, and then splicing the calculated vector with the spatial feature vector to obtain the initial trajectory feature of the vehicle to be predicted; or, do a dot multiplication on the multiple temporal feature vectors and the spatial feature vector, and the result of the calculation is To predict the initial trajectory characteristics of the vehicle, this method requires that the dimensions of the temporal feature vector and the spatial feature vector are the same.

As shown in Figure 15, the network sequence numbers "g ₁ ,..., g _i ,..., g _n "in the trajectory data of the vehicle to be predicted are sequentially input into the first embedding layer and the splicing module, and the spatial feature vector "a ₁₁ …a _1j …a _1n …a _i1 …a _ij …a _in …a _n1 …a _nj …a _nn ".

Input the "month" time element "m" into the second embedding layer to obtain the multidimensional vector "[m ₁ …m _j …m _n ]"; input the "week" time element "w" into the third embedding layer to obtain the multidimensional vector " [w ₁ …w _j …w _n ]”; input the “day” time element “d” into the fourth embedding layer to obtain the multi-dimensional vector “[d ₁ …d _j …d _n ]”; replace the “time” time element “ h" is input to the fifth embedding layer, and the multi-dimensional vector “[h ₁ …j _j …h _n ]” is obtained; and the “engraved” time element “q” is input to the sixth embedding layer to obtain the multi-dimensional vector “[q ₁ …q _j … q _n ]"; input the multi-dimensional vector output from the second embedding layer to the sixth embedding layer into the splicing module to obtain the temporal feature vector "m ₁ …m _j …m _n w ₁ …w _j …w _n d ₁ …d _j …D _n h ₁ …j _j …h _n q ₁ …q _j …q _n ".

Combine the above spatial feature vector "a ₁₁ …a _1j …a _1n …a _i1 …a _ij …a _in …a _n1 …a _nj …a _nn "and the time feature vector "m ₁ …m _j …m _n w ₁ …w _j …w _n d ₁ …d _j …d _n h ₁ …j _j …h _n q ₁ …q _j …q _n ”Input the splicing module to obtain the initial trajectory feature of the vehicle to be predicted “a ₁₁ …a _1j …a _1n …A _i1 …a _ij …a _in …a _n1 …a _nj …a _nn m ₁ …m _j …m _n w ₁ …w _j …w _n d ₁ …d _j …d _n h ₁ …j _j …h _n q ₁ …q _j …q _n ".

If the trajectory data of the vehicle to be predicted includes location information or grid number, as well as time information and POI type, the prediction device can first input the location information or grid number in the trajectory data of the vehicle to be predicted into the first embedding in the embedded model Layer to obtain the spatial feature vector; input each time element in the time information into the second embedding layer to the sixth embedding layer in the embedding model to obtain the time feature vector of the vehicle to be predicted; enter the POI type into the seventh embedding Layer to obtain the POI feature vector; and fuse the spatial feature vector, the temporal feature vector, and the POI feature vector into the initial trajectory feature of the vehicle to be predicted.

Input the position information or grid number in the trajectory data of the vehicle to be predicted into the first embedding layer in the embedding model to obtain the realization of the spatial feature vector, and input the time elements in the time information into the embedding model. The implementation of the second embedding layer to the sixth embedding layer to obtain the time feature vector of the vehicle to be predicted is as described above and will not be repeated here.

The following describes how to input the POI type into the seventh embedding layer to obtain the POI feature vector.

After each POI type is input to the seventh embedding layer, the seventh embedding layer outputs a multi-dimensional vector. The dimension of the vector can be preset. The dimensions of the vectors corresponding to different POI types are the same.

When the prediction device obtains that each position information or grid number in the trajectory data corresponds to multiple POI types, it can first splice or perform dot multiplication operations on multiple vectors corresponding to the multiple POI types to obtain the position information or grid The POI vector corresponding to the serial number.

When multiple POI vectors are obtained according to the trajectory data of the vehicle to be predicted, the multiple POI vectors can be spliced or dot multiplied to obtain the POI feature vector of the vehicle to be predicted. If a POI vector is obtained from the trajectory data of the vehicle to be predicted, this POI vector can be directly used as the POI feature vector of the vehicle to be predicted.

After the POI feature vector of the vehicle to be predicted is obtained, the POI feature vector and the spatial feature vector of the vehicle to be predicted can be spliced or dot multiplied, and the obtained vector can be used as the initial trajectory feature of the vehicle to be predicted; or, the POI can be used as the initial trajectory feature of the vehicle to be predicted. The feature vector is spliced or dot multiplied with the spatial feature vector and time feature vector of the vehicle to be predicted, and the obtained vector is used as the initial trajectory feature of the vehicle to be predicted. Among them, the dot multiplication method requires the dimensions of each feature vector to be the same.

When predicting the destination of a vehicle, in addition to predicting based on the trajectory data of the vehicle to be predicted, it can also be predicted based on the trajectory data and travel data of the predicted vehicle, and the travel data of the vehicle to be predicted can be added for the destination of the vehicle Prediction can improve the accuracy of predicted destination information.

Fig. 6 is an exemplary flowchart of another method for predicting the destination of a vehicle to be predicted in this application. The method shown in FIG. 6 includes S610 to S630.

S610: Acquire trajectory data of the vehicle to be predicted in the traffic area, where the trajectory data includes location information of locations that the vehicle to be predicted has passed during this trip.

The implementation of this step can refer to the implementation of S110, which will not be repeated here.

S620: Acquire travel data of the vehicle to be predicted.

In some implementations, the travel data of the vehicle to be predicted may include one or more of the following: the number of trips of the vehicle to be predicted in a period of time, the travel frequency of the vehicle to be predicted in a period of time, the type of vehicle to be predicted, and the type of vehicle to be predicted. Predict the weather type when the vehicle is traveling, the travel sub-time period of the vehicle to be predicted in a period of time, the number of trips of the vehicle of the vehicle type to be predicted in a period of time, and the number of trips of the vehicle type of the vehicle to be predicted in a period of time Travel frequency is the number of vehicles of the same type as the vehicle to be predicted that travel within a period of time.

For example, the travel data of the vehicle to be predicted may include one or more of the following information: the number of sunrise trips of the vehicle to be predicted, the frequency of monthly trips, the type of vehicle to be predicted, the weather type at the start time of the vehicle to be predicted, The sub-time period of the trip of the vehicle to be predicted in a day, the number of trips of the vehicle type of the vehicle to be predicted in a day, the trip frequency of the vehicle type of the vehicle to be predicted within a month, and the number of trips within a month , The number of vehicles of the same vehicle type as the vehicle to be predicted.

One way to obtain the number of trips of the vehicle to be predicted in a period of time is as follows: Obtain the historical passing data of the traffic area within the period of time, and then determine the number of trips of the vehicle to be predicted in the period of time based on the historical process data . For an implementation manner of determining the number of trips of the vehicle to be predicted during the period of time according to the historical process data, reference may be made to related content in S320. The difference is that the passing data in S320 is the passing data of the time period during which the vehicle to be predicted will travel this time, while the passing data in this step is historical passing data; and the i-th time information is determined in this step When the i-1th time information belongs to the time information of the two trips of the vehicle to be predicted, the prediction device continues to execute (2) and (3) until i=2, so that it can be known that the vehicle to be predicted is within the period of time The number of trips.

The travel frequency of the vehicle to be predicted within a period of time refers to the ratio of the number of sub-periods in which the vehicle to be predicted travels to the total number of sub-periods included in the period of time. For example, the travel frequency of the vehicle to be predicted in a period of time may refer to the ratio of the number of days the vehicle to be predicted travels to the total number of days in the month in a month.

The vehicle type of the vehicle to be predicted refers to classifying the vehicle in a certain way. For example, vehicles can be divided into different types such as taxis, passenger cars, private cars, and trucks.

The type of weather to be predicted when the vehicle travels may include sunny, cloudy, cloudy, rainy and snowy, etc. For example, the weather type when the vehicle to be predicted travels may be the weather type on the day of travel, or may be the weather type at the time period when the travel starts. The weather type of the vehicle to be predicted when traveling can be obtained from the weather station or weather software.

The number of trips of a vehicle of the vehicle type to which the vehicle to be predicted belongs within a period of time can be obtained by the following method: adding the number of trips of all vehicles in the vehicle type during the period of time.

The travel frequency of the vehicle of the vehicle type to be predicted within a period of time can be obtained by calculating the number of sub-periods in which vehicles of the vehicle type travel within a period of time and the total number of sub-periods included in the period of time. The ratio of the quantity.

Each piece of data in the travel data of the vehicle to be predicted can be encoded to form a vector, for example:

S630: Obtain a target sub-region and a target POI type of the vehicle to be predicted in the traffic area according to the trajectory data, the travel data, and the target neural network model. The POI of this type in the destination sub-area is the destination of the vehicle to be predicted.

For the method shown in FIG. 6, an exemplary structure of the target neural network model of the present application is shown in FIG. 7. The target neural network model may include an embedded model, a first feature extraction model, a second feature extraction model, a fusion model, a first classification model, and a second classification model, where the embedded model is used for vector mapping to obtain a multidimensional vector; The feature extraction model is used to extract the trajectory features of the vehicle to be predicted; the second feature extraction model is used to extract the travel features in the travel data; the fusion model is used to fuse the trajectory features and the travel features into driving features; the first classification The model is used to output the target sub-region of the vehicle to be predicted according to the driving feature; the second classification model is used to output the target POI type of the vehicle to be predicted according to the driving feature.

An embedding layer can be included in the embedding model. The first feature extraction model may include any one of LSTM network, BRNN, and memory network. The second extracted feature model may include an artificial neural network model. For example, the second extracted feature model may be a neural network model including one or more fully connected layers. The first classification model or the second classification model may be an artificial neural network model. For example, the first classification model or the second classification model is a neural network model that only includes a fully connected layer and an activation function.

With regard to the target neural network model shown in FIG. 7, a method for obtaining the target sub-region and the target POI type of the vehicle to be predicted in the aforementioned S630 will be introduced below in conjunction with FIG. 8. The method shown in FIG. 8 includes S810 to S870.

S810: Acquire an initial trajectory feature of the vehicle to be predicted according to the trajectory data and the embedded model of the vehicle to be predicted.

For this step, refer to S510, which will not be repeated here.

S820: Acquire the trajectory feature of the vehicle to be predicted according to the initial trajectory feature of the vehicle to be predicted and the first feature extraction model.

For this step, refer to S520, which will not be repeated here.

S830: Obtain the initial travel characteristics of the vehicle to be predicted based on the travel data and the embedded model.

For example, input each type of data in the travel data into the corresponding embedding layer in the embedding model, and the embedding layer maps the corresponding data into a multi-dimensional vector, where different types of data have different embedding layers, and the dimensions of the mapped vector can be the same , It can also be different, the dimension of the vector obtained by different data mapping is preset.

Optionally, before the number of trips of the vehicle to be predicted in a period of time is input into the embedding layer, the number of trips can be encoded first. The encoding method may be: specifying that the number of trips of the vehicle to be predicted in a period of time is 0 to n times is the first gear, and the corresponding code value is "1"; the number of trips from n+1 to n+2 is the second gear, and the corresponding code value is "2", and so on. In this way, the code value corresponding to the number of trips can be determined according to the gear division method and the code value corresponding to each gear, and then the corresponding code value is input into the corresponding embedded layer for mapping. This method can reduce the amount of calculation and the complexity of the calculation.

Similarly, before the travel frequency of the vehicle to be predicted in a period of time is input to the embedding layer, the travel frequency can be encoded first. The encoding method can be: specify the travel frequency of the vehicle to be predicted in a period of time from 0 to frequency 1. For the first gear, the corresponding value is "1"; Frequency 1 to Frequency 2 are for the second gear, and the corresponding value is "2", and so on. In this way, the value corresponding to the travel frequency can be determined according to the gear division method and the value corresponding to each gear, and then the corresponding value is input into the corresponding embedded layer for mapping. This method can reduce the amount of calculation and the complexity of the calculation.

In the same way, before entering the number of vehicles of the same vehicle type as the vehicle to be predicted that travel within a period of time into the embedding layer, you can first specify 0 to number 1 as the first gear, and the corresponding value is "1" ; Quantity 1 to quantity 2 are the second level, and the corresponding value is "2", and so on. In this way, the value corresponding to the number of traveling vehicles can be determined according to the gear division method and the value corresponding to each gear, and then the corresponding value is input into the corresponding embedded layer for mapping. This method can reduce the amount of calculation and the complexity of the calculation.

In the same way, before entering the embedded layer of the weather type when the vehicle to be predicted is traveling, the value corresponding to each weather type can be specified first, for example: the value "00" for sunny days, the value "01" for cloudy days, and the value "01" for cloudy days. 10", the corresponding value "11" in rainy and snowy days, and then find the value corresponding to the weather type when the vehicle to be predicted travels from these values, and then input the corresponding value into the corresponding embedding layer for mapping. ,

Similarly, before entering the type of vehicle to be predicted into the embedding layer, you can first specify the value corresponding to each type, and then find the value corresponding to the type of vehicle to be predicted from these values, and then enter the corresponding value into the corresponding embedding Mapping in the layer. In this way, the amount of data can be reduced, so that the amount of calculation and the complexity of the calculation can be reduced.

After the prediction device obtains the multi-dimensional vector corresponding to various historical travel data of the vehicle to be predicted according to the embedded model, the vector corresponding to the various historical travel data can be merged into a feature vector by splicing or dot multiplication. This feature vector is called The initial travel characteristics of the vehicle to be predicted.

S840: Acquire the travel feature of the vehicle to be predicted according to the initial travel feature of the vehicle to be predicted and the second feature extraction model.

For example, the initial travel characteristics of the vehicle to be predicted are input into the second feature extraction model, and the second feature extraction model outputs the travel characteristics of the vehicle to be predicted.

S850: Determine the driving characteristics of the vehicle to be predicted according to the trajectory characteristics of the vehicle to be predicted, the travel characteristics of the vehicle to be predicted, and the fusion model.

For example, the fusion model merges the trajectory characteristics of the vehicle to be predicted and the travel characteristics of the vehicle to be predicted by splicing, so as to obtain the driving characteristics of the vehicle to be predicted.

For example, the fusion model fuses the trajectory characteristics of the vehicle to be predicted and the travel characteristics of the vehicle to be predicted by a point multiplication method, so as to obtain the driving characteristics of the vehicle to be predicted. However, this method requires that the trajectory characteristics of the vehicle to be predicted and the travel characteristics of the vehicle to be predicted have the same dimensions.

S860: Acquire a target subregion of the vehicle to be predicted according to the driving characteristics of the vehicle to be predicted and the first classification model.

For example, the driving characteristics of the vehicle to be predicted are input into the first classification model, and the first classification model outputs the target sub-region of the vehicle to be predicted.

S870: Acquire the target POI type of the vehicle to be predicted according to the driving characteristics of the vehicle to be predicted and the second classification model.

For example, the driving characteristics of the vehicle to be predicted are input into the second classification model, and the second classification model outputs the type of the target POI of the vehicle to be predicted.

It is understandable that, in the various embodiments of the present application, the trajectory data and/or travel data of the vehicle to be predicted may be obtained by the prediction apparatus from other equipment.

It should be understood that the target neural network model used in the foregoing embodiments of the present application is a neural network model obtained by training the initial neural network model. Since the target neural network model has been trained, the target neural network model has the ability to predict the target sub-region and the target POI type of the vehicle based on the trajectory data (and/or travel data) of the vehicle, so that the target neural network can be used for prediction in this application The destination method of the vehicle.

It should also be understood that the process of training the initial neural network model, in terms of time, before the target neural network model obtained by the initial neural network model training is used to predict the destination of the vehicle, in some embodiments, the initial neural network model The operation of training can be performed by the training module in the prediction device in this application. In other embodiments, the operation of training the initial neural network model can be performed by a third-party device or by an independent training device. The prediction device can use a third-party device or training device before predicting the destination of the vehicle. Obtain the trained target neural network model.

The following takes the training of the initial neural network performed by the training device as an example to introduce the training method of the neural network model of the present application. In the method for training a neural network model proposed in this application, a large amount of trajectory data and travel data obtained according to the historical travel conditions of a large number of vehicles (for example, thousands of vehicles) in a traffic area are used as training data for the initial neural network The model is trained, and the trained target neural network model can be used as the target neural network model in the method for predicting the destination of the vehicle proposed in this application, and is used to predict the target sub-area of the current traveling vehicle in the traffic area and the type of the target POI .

It should be understood that in practical applications, in the training phase of the neural network model, the training data is the historical trajectory data and travel data of vehicles in a traffic area. In the prediction phase, the trained target neural network model can be used To predict the destination of the current traveling vehicle in the traffic area.

It should be understood that before training the initial neural network model, the initial neural network model needs to be selected or designed in advance. For example, the initial neural network model that is suitable for this application for vehicle destination prediction is selected from the neural network models that have been built in the industry. Model, or construct an initial neural network model suitable for the application to predict the destination of the vehicle according to the needs, such as: design the structure of the initial neural network model (the number of layers of the initial neural network model, the type of sub-models in the initial neural network model , The number and types of neurons in each layer, the type of loss function, etc.), the structure of the initial neural network model used in this application is as mentioned above, and for different embodiments, the type of the initial neural network model is slightly different.

Taking the training data for training the neural network model of this application including historical trajectory data and travel data of a large number of vehicles as an example, a method for training the neural network model of this application may include step 8100 to step 8200. The device that performs this method is called a training device.

Step 8100: Obtain training data. The training data includes historical trajectory data and travel data of a large number of vehicles, and each training data also corresponds to label data corresponding to each vehicle. Among them, the trajectory data of each vehicle and the labeling data are in one-to-one correspondence. The trajectory data includes the location information of multiple locations that the vehicle passes. The labeling data records the POI type and the purpose of the real destination of the corresponding vehicle. The sub-region to which the land belongs.

The POI type of the destination of the vehicle is also referred to as the destination POI type of the vehicle, and the traffic sub-area described by the destination of the vehicle is also referred to as the destination sub-area of the vehicle.

Step 8200: Train the initial neural network model according to the training data, the neural network model obtained by training is the target neural network model, and the initial neural network model is used to predict the vehicle's trajectory data in the traffic area according to the vehicle's trajectory data. Destination sub-area and destination POI type.

In the method of this embodiment, the training device trains the initial neural network model used to predict the target sub-area and the target POI type of the vehicle in the traffic area through a large amount of historical trajectory data and travel data of the vehicle, so that the target neural network model obtained by training It can more accurately predict the destination sub-area and destination POI type of the vehicle in the traffic area.

Generally speaking, the more historical trajectory data included in the training data, the better. Specifically, the more historical trajectory data included in the training data, the higher the accuracy of the trained target neural network model for predicting the target sub-region and target POI type of the vehicle.

The method of obtaining the trajectory data in the training data can refer to the method of obtaining the trajectory data in the aforementioned method of predicting the destination of the vehicle, which will not be repeated here. The difference is that the trajectory data in this application is the historical trajectory data of the vehicle in the traffic area, that is, the trajectory data of the trip that has ended. In addition, the label data corresponding to the trajectory data needs to be obtained in this application. An exemplary method for obtaining annotation data in the present application may include step 9100 to step 9300.

Step 9100: Obtain map information of the traffic area, and divide sub-areas of the traffic area according to the map to obtain location information of the sub-areas of the traffic area. For this step, refer to S320, which will not be repeated here.

Step 9200: Obtain the POI information of the traffic area, and determine the correspondence between the POI and the POI type in the traffic area according to the POI information.

For this step, reference may be made to the implementation manner of determining the corresponding relationship between the POI and the POI type in the traffic area described earlier, which will not be repeated here.

Step 9300: Obtain parking lot data in the traffic area, and determine the label data corresponding to the vehicle based on the parking lot data and the correspondence between the POI and POI types in the traffic area.

For example, for a trajectory data of a vehicle, the target parking lot data is searched from the parking lot data, and the parking lot corresponding to the target parking lot data is located at the last location recorded in the trajectory data (that is, the vehicle corresponding to the trajectory data) In the vicinity of the last location captured by the surveillance system during the trip, for example, the distance between the last location and the parking lot is less than or equal to a preset distance threshold. An example of the distance threshold is 100 M; According to the correspondence between the POI and the POI type determined in step 920, the POI type of the POI to which the parking lot belongs is used as the destination POI type corresponding to the vehicle, and the subregion to which the POI belongs is the destination subregion corresponding to the vehicle Area; the corresponding relationship between the destination POI type and the destination sub-area and the vehicle is generated, and the destination POI type and the destination sub-area are the label data corresponding to the vehicle.

In some designs, if the target parking lot data corresponding to the trajectory data cannot be found in the parking lot data, the last location recorded in the trajectory data can be determined first (that is, the vehicle corresponding to the vehicle trajectory data is in this trip. The last location photographed by the monitoring system) nearby POI, for example, the distance between the last location and the parking lot is less than or equal to a preset distance threshold. An example of the distance threshold is one hundred meters; The corresponding relationship between the POI and the POI type determined in step 920, the POI type corresponding to the POI is determined as the destination POI type of the trajectory data, and the subarea to which the POI belongs is taken as the destination subarea corresponding to the vehicle; The destination POI type and the corresponding relationship between the destination sub-area and the vehicle, and the destination POI type and the destination sub-area are the label data corresponding to the vehicle.

It is understandable that step 9100 is only an implementation manner for the training device to obtain sub-area information in the traffic area, and other methods may also be used to obtain the sub-area information in the traffic area in this application. For example, the training device may send a request message to others to request sub-area information in the traffic area, and the request message may carry the name or area identification information of the traffic area. After receiving the request message, other devices can send the sub-area information in the traffic area to the training device. For another example, the sub-region information in the traffic area can be manually copied to the training device.

It is understandable that step 9200 is only an implementation manner for the training device to obtain the POI type in the traffic area, and the POI type in the traffic area may also be obtained in other ways in this application. For example, the training device may send a request message to others to request the POI type in the traffic area, and the request message may carry the name or area identification information of the traffic area. After receiving the request message, the other device performs the operation in step 920 or other operations, and sends the POI type in the traffic area to the training device. For another example, the POI type information in the traffic area can be manually copied to the training device.

It is understandable that step 9200 and step 9300 are only an implementation manner for the training device to obtain the label data corresponding to the vehicle, and the label data may also be obtained in other ways in this application. For example, the training device may send a request message to others to request the annotation data, and the request message may carry the trajectory data of the vehicle. After receiving the request message, other devices perform the operations in step 9200 and step 9300, or perform other operations, and send the annotation data to the training device.

It is understandable that the foregoing implementation of acquiring training data is only an example, and training data may also be acquired in other ways in this application. For example, the training device may send a request message to other devices to request training data of the traffic area, and the request message may carry the name or area identification information of the traffic area; after receiving the request message, the other device sends the request message to the training device The training data. For another example, the training data can be copied to the training device manually.

The target neural network model trained in this application can be used in the aforementioned method of predicting the destination of a vehicle. Normally, the data used to predict the destination of the vehicle to be predicted should be the same type of data used when training the target neural network model.

For example, the trajectory data used in training only includes the location information of the monitoring device, and the trajectory data in the method for predicting the destination of the vehicle only includes the location information of the monitoring device.

For another example, if the trajectory data used in training includes the location information of the sub-region or the grid number corresponding to the sub-region, the trajectory data in the method of predicting the destination of the vehicle includes the location information of the sub-region or the corresponding sub-region. The number of the grid.

For another example, if the trajectory data used during training includes location information and time information, the trajectory data in the method for predicting the destination of a vehicle includes location information and time information.

The difference between the method of training the target neural network model in this application and the method of predicting the destination of the vehicle based on the target neural network model is that after the target neural network model predicts the target sub-region and the type of the target POI of the vehicle each time, More steps need to be performed. For example, after performing step 1001 and step 1002, step 1003 to step 1007 need to be performed.

Step 1001: Obtain training data. Obtaining training data may include obtaining historical trajectory data. Optionally, obtaining training data may also include obtaining historical travel data.

To obtain historical trajectory data, refer to the aforementioned method of pre-storing the target sub-area of the vehicle to be predicted and the type of the target POI to obtain the trajectory data of the vehicle to be predicted. To obtain historical travel data, refer to the corresponding implementation method for obtaining travel data.

Step 1002: Input the training data to the initial neural network model, and the initial neural network model outputs the predicted target sub-region and the target POI type

In the above step S1002, if it is the first time to train the initial neural network model, the initial neural network model needs to be initialized. The initial neural network model is to initialize the parameters in the constructed or selected neural network model. . Input the training data to the initialized initial neural network model, and the initialized initial neural network model maps the input data according to the model structure, and then performs feature extraction on the mapped vector, then performs feature fusion, and finally performs the target POI classification And target sub-region classification. This process is similar to the steps of S510-S540 (or S810-S870 in another embodiment) described above. However, since the initial neural network model after only initialization has not learned the rules in the input training data and the corresponding label data, the target sub-region and the target POI type of the vehicle output in step S1002 are in the label data of the vehicle. There is a big difference between the true target sub-region and the target POI type, that is, the prediction result is not accurate. Therefore, the following step S1003 and subsequent steps need to be performed.

Step 1003: Calculate the predicted loss value of the predicted target sub-region compared to the target sub-region in the label data, and calculate the predicted loss value of the predicted target POI type compared to the target POI type in the label data.

For example, the loss value of the predicted target sub-region compared to the target sub-region in the labeled data is calculated according to the loss function, and this loss value is called the first predicted loss value; the predicted target POI type calculated based on the loss function is compared to the labeled data The loss value of the target POI type in the target POI, which is called the second predicted loss value.

The first prediction loss value and the second prediction loss value are calculated by two loss functions respectively, and the obtained first prediction loss value represents the difference between the target sub-region predicted by the initial neural network model during the training process and the actual target sub-region of the vehicle. The degree of error between the two; the obtained second prediction loss value represents the degree of error between the target POI type predicted by the initial neural network model in the training process and the actual target POI type of the vehicle.

Step 1004, according to the first prediction loss value and the second prediction loss value, update the parameters in the initial neural network model, for example, update each embedding layer in the embedding model, the first feature extraction model, the second feature extraction model, and the first classification The parameters in the model and the second classification model. The implementation of updating the parameters in the initial neural network model according to the loss value can refer to the prior art, which will not be repeated here.

Step 1005: It is judged whether the training termination condition is satisfied.

For example, it is judged whether the number of training times has reached a preset threshold value. If it has been reached, it means that the training termination condition is met; otherwise, it means that the training termination condition is not met.

For another example, obtain training data that has not been used to train the initial neural network model, which is called test data; input the trajectory data in the test data into the initial neural network model, and calculate the target POI type predicted by the initial neural network model Compare the loss value of the POI type in the test data, and calculate the loss value of the target subregion predicted by the initial neural network model compared to the target subregion in the test data; if these two loss values are less than or equal to the preset The threshold value of, it means that the training termination condition is met, otherwise, it means that the training termination condition is not met.

In step 1006, if the training termination condition is not met, steps 1001 to S1005 are repeated.

Step 1007: If the training termination condition is met, output the trained neural network model, and the trained neural network model is used as the target neural network model for predicting the destination of the vehicle.

Optionally, according to the above-mentioned embodiment, the prediction device can learn the destination sub-area and the type of the destination POI of a large number of vehicles to be predicted in the traffic area, and the prediction device can count the vehicles of the same destination after learning the destination of a large number of vehicles to be predicted. flow.

Further, the predicting device can predict the number of vehicles arriving at the same destination in the same time period. The length of the time can be preset, for example, it can be half an hour or one hour.

For example, when half an hour is used as a prediction time period, the prediction device can calculate the time for each vehicle with the same destination to arrive at the destination from its current location according to the average vehicle speed in the traffic area and according to the conventional route, and Count the traffic volume that will arrive at the destination in the next half an hour, one hour, or one and a half hours in the future.

After the predicting device learns the traffic volume when a POI of one type in a sub-area is used as a destination in a time period in the future, it can also determine the traffic state of the road near the POI according to the traffic volume.

For example, you can preset the following thresholds for traffic flow when the traffic state of the road is severely congested, congested, lightly congested, and unblocked: traffic volume greater than 400 means severe congestion, traffic volume between 200 and 400 means congestion, traffic volume It is lightly crowded if it is between 100 and 200, and it is unblocked if it is less than 100.

After the prediction device learns the traffic state of the road near the POI, it can also send the road traffic state information to the traffic management platform. It enables the traffic management platform to notify the traffic status of the roads near each type of POI in each sub-area in real time through traffic radio stations or news information, or enables the traffic management platform to formulate a series of traffic diversion strategies based on the road traffic status. Or, after the prediction device learns the traffic status of the road near the POI, it can also send road traffic status information to the driving vehicle, and the driving vehicle receives the road traffic status in real time, so that it can decide to continue to the destination according to its own travel situation. Give up traveling or make a detour.

Optionally, after the predicting device learns the traffic state of the road near the POI, it may generate a traffic travel suggestion according to the traffic state of the road. The prediction device can also send the traffic travel advice to the driving vehicle, so that the vehicle can make a travel decision based on the obtained traffic travel advice.

Fig. 9 is a structural diagram of a device for predicting a destination of a vehicle provided by an embodiment of the present application. The device can be implemented as part or all of the device through software, hardware or a combination of the two. The device 900 includes an acquisition module 910 and a prediction module 920. The device 900 can implement the method for predicting the destination of the vehicle in this application.

The obtaining module 910 is used to obtain trajectory data of the vehicle to be predicted in the traffic area during travel.

The prediction module 920 is configured to obtain destination information of the vehicle to be predicted in the traffic area according to the trajectory data and the target neural network model, where the destination information includes: the destination sub-region of the vehicle to be predicted And the type of the POI of the destination point of interest of the vehicle to be predicted.

In some possible implementation manners, the target neural network model includes an embedded model, a first feature extraction model, a first classification model, and a second classification model, wherein the embedded model is used to input to the embedded model Vectorization of the data of the first feature extraction model, the first feature extraction model is used for feature extraction of the data input to the first feature extraction model, and the fusion model is used for feature fusion of the data input to the fusion model, The first classification model and the second classification model are respectively used for class prediction based on input data of the first classification model and the second classification model.

Wherein, the prediction module 920 is specifically configured to: input the trajectory data into the embedded model to obtain the initial trajectory feature of the vehicle to be predicted, and the initial trajectory feature includes the multi-dimensional vector corresponding to the trajectory data; The initial trajectory feature is input to the first feature extraction model to obtain the trajectory feature of the vehicle to be predicted; the trajectory feature is input to the first classification model to obtain the target subregion of the vehicle to be predicted; The trajectory feature is input to the second classification model, and the target POI type of the vehicle to be predicted is obtained.

In some possible implementation manners, the obtaining module 910 is further configured to obtain travel data of the vehicle to be predicted. Wherein, the prediction module 920 is specifically configured to obtain destination information of the vehicle to be predicted in the traffic area according to the trajectory data, the travel data, and the target neural network model.

In some possible implementations, the travel data of the vehicle to be predicted includes one or more of the following data: vehicle type, travel weather type, number of vehicle trips in the first time period, and travel data in the second time period The frequency of vehicle travel, and the sub-period of vehicle travel in the third time period.

In some possible implementations, the prediction module 920 is also used to: determine the destination as the traffic flow of the type of POI in the destination sub-area according to the destination information of the vehicle to be predicted; The traffic flow determines the traffic state of the road in the destination sub-area.

In some possible implementation manners, the acquiring module 910 is specifically configured to: determine information about multiple locations that the vehicle to be predicted passes through during travel according to the passing data in the traffic area; and acquire the traffic Information about the sub-regions in the area; determine the trajectory data of the vehicle to be predicted in the travel process according to the information of the multiple locations that the vehicle to be predicted passes through during the travel and the sub-region information in the traffic area.

In some possible implementation manners, the trajectory data includes position information and time information of the vehicle to be predicted passing in the traffic area.

In some possible implementation manners, the trajectory data further includes the type of POI that the vehicle to be predicted passes through in the traffic area.

As shown in FIG. 10, the device 900 further includes a training module 940, the training module 940 is used to: determine the initial neural network model; according to the historical trajectory data of the vehicles in the traffic area to perform the initial neural network model Training to obtain the target neural network model.

The training module 940 may also be used to determine an initial neural network model; train the initial neural network model according to historical trajectory data and travel data of vehicles in the traffic area to obtain the target neural network model.

In some possible implementation manners, the device 900 may further include an output module for outputting destination information of the vehicle to be predicted. Optionally, the output module can also be used for traffic flow. Optionally, the output module can also be used to output road traffic status.

In some possible implementation manners, the device 900 may further include a traffic guidance module, which is used to perform traffic guidance according to the traffic state of the road to relieve traffic pressure.

The embodiment of the present application also provides a computing device for predicting the destination of a vehicle. FIG. 11 exemplarily provides a possible architecture diagram of the computing device 1100.

The computing device 1100 includes a memory 1101, a processor 1102, and a communication interface 1103. Among them, the memory 1101, the processor 1102, and the communication interface 1103 implement communication connections between each other through a bus.

The memory 1101 may be a read only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1101 may store a program. When the program stored in the memory 1101 is executed by the processor 1102, the processor 1102 and the communication interface 1103 are used to execute the method of predicting the destination of the vehicle. The memory 1101 can also store a data set. For example, a part of the storage resources in the memory 1101 is divided into a data set storage module for storing the data set required to execute the method of predicting the destination of the vehicle, and a part of the storage resources in the memory 1101 It is divided into a neural network model storage module, which is used to store the target neural network model shown in Figure 4 or Figure 7.

The processor 1102 may adopt a general-purpose central processing unit (central Processing Unit, CPU), microprocessor, application specific integrated circuit (application specific integrated circuit, ASIC), graphics processing unit (graphics processing unit, GPU), or one or more integrated circuit.

The processor 1102 may also be an integrated circuit chip with signal processing capabilities. In the implementation process, part or all of the functions of the device for predicting the destination of the vehicle of the present application can be completed by hardware integrated logic circuits in the processor 1102 or instructions in the form of software. The aforementioned processor 1102 may also be a general-purpose processor, a digital signal processing (digital signal processing, DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (field programmable gate array, FPGA) or other programmable logic devices , Discrete gates or transistor logic devices, discrete hardware components. The methods, steps, and logical block diagrams disclosed in the foregoing embodiments of the present application can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, registers. The storage medium is located in the memory 1101, and the processor 1102 reads the information in the memory 1101 and completes part of the functions of the device for predicting the destination of the vehicle in the embodiment of the present application in combination with its hardware.

The communication interface 1103 uses a transceiver module such as but not limited to a transceiver to implement communication between the computing device 1100 and other devices or a communication network. For example, the data set can be obtained through the communication interface 1103.

The bus may include a path for transferring information between various components of the computing device 1100 (for example, the memory 1101, the processor 1102, and the communication interface 1103).

When there are multiple computing devices 1100, each of the foregoing computing devices 1100 establishes a communication path through a communication network. Each computing device 1100 runs any one or more of the acquisition module 910, the prediction module 920, the determination module 930, or the training module 940. Any computing device 1100 may be a computing device (for example, a server) in a cloud data center, or a computing device in an edge data center, or a terminal computing device.

The descriptions of the processes corresponding to each of the above drawings have their respective focuses. For parts that are not described in detail in a certain process, please refer to the related descriptions of other processes.

FIG. 12 is a schematic architecture diagram of a system to which the apparatus of the embodiment of the present application can be applied. As shown in FIG. 12, the system 1200 includes a prediction device 1210, a training device 1220, a database 1230, a data storage system 1250, and a data collection device 1260.

The data collection device 1260 is used to collect training data. After the training data is collected, the data collection device 1260 stores the training data in the database 1230, and the training device 1220 trains a preselected neural network model based on the training data maintained in the database 1230 to obtain the target neural network model 1201. The trained target neural network model 1201 has the function of predicting the sub-region to which the destination of the vehicle belongs and predicting the POI type of the destination of the vehicle.

It should be noted that in practical applications, the training data maintained in the database 1230 may not all come from the collection of the data collection device 1260, and may also be received from other devices. In addition, it should be noted that the training device 1220 does not necessarily train the target neural network model 1201 completely based on the training data maintained by the database 1230. It may also obtain training data from the cloud or other places for model training, or generate training data by itself. The description should not be taken as a limitation to the embodiments of the present application.

The target neural network model 1201 obtained by training according to the training device 1220 can be applied to different systems or devices, such as the prediction device 1210.

After the data collection device 1260 collects the trajectory data of the vehicle to be predicted, the trajectory data can be stored in the database 1230, and the prediction device 1210 makes predictions based on the trajectory data maintained in the database 1230. Alternatively, after the data collection device 1260 collects the trajectory data and travel data of the vehicle to be predicted, the trajectory data and travel data can be stored in the database 1230, and the prediction device 1210 makes predictions based on the trajectory data and travel data maintained in the database 1230.

In the process that the prediction device 1210 predicts the target sub-region and the target POI type of the vehicle, the prediction device 1210 can call the data, codes, etc. in the data storage system 1250 for the corresponding prediction processing, and can also use the data obtained from the corresponding processing, Instructions and the like are stored in the data storage system 1250.

It is understandable that FIG. 12 is only a schematic system architecture diagram, and the positional relationship between the devices, devices, modules, etc. shown in FIG. 12 does not constitute any limitation. For example, in FIG. 12, the data storage system 1250 relatively predicts The device 1210 is an external memory. In other cases, the data storage system 1250 can also be placed in the prediction device 1210. For example, the prediction device 1210 and the training device 1220 may be the same device.

In some designs, the prediction device may be deployed in a cloud environment, which is an entity that uses basic resources to provide cloud services to users in a cloud computing mode. The cloud environment includes a cloud data center and a cloud service platform. The cloud data center includes a large number of basic resources (including computing resources, storage resources, and network resources) owned by a cloud service provider. The computing resources included in the cloud data center can be a large number of computing resources. Device (for example, server).

The prediction device can be a server in a cloud data center; the prediction device can also be a virtual machine created in a cloud data center; the prediction device can also be a server or a software device deployed on a virtual machine in a cloud data center. It can be distributed on multiple servers, or distributed on multiple virtual machines, or distributed on virtual machines and servers. For example, multiple modules in the forecasting apparatus may be distributed on multiple servers, or distributed on multiple virtual machines, or distributed on virtual machines and servers.

When the prediction device is a software device, the prediction device can be logically divided into multiple parts, and each part has a different function. In this scenario, several parts of the prediction device can be deployed in different environments or devices. Taking Figure 13 as an example, part of the forecasting device is deployed in terminal computing equipment, and the other part is deployed in the data center (specifically deployed on the server or virtual machine in the data center). The data center can be a cloud data center or a data center. It is an edge data center. An edge data center is a collection of edge computing devices that are deployed closer to the terminal computing device.

It is understandable that this application does not restrict which parts of the prediction device are deployed in the terminal computing equipment and which parts are deployed in the data center. In actual applications, it can be adapted according to the computing capabilities of the terminal computing equipment or specific application requirements. deploy. It is worth noting that in some possible implementations, the prediction device can be deployed in three parts, of which one part is deployed in the terminal computing device, one part is deployed in the edge data center, and the other part is deployed in the cloud data center.

It is understandable that the division of modules in the embodiments of the present application is illustrative, and is only a logical function division. In actual implementation, there may be other division methods. In addition, the functional modules in the various embodiments of the present application It can be integrated in a processor, it can be a separate physical presence, or two or more modules can be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or software function modules.

If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solution of the present application essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , Including several instructions to make a terminal device (which may be a personal computer, a mobile phone, or a network device, etc.) or a processor (processor) execute all or part of the steps of the method in each embodiment of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disks or optical disks and other media that can store program codes. .

In the above-mentioned embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, it can be implemented in the form of a computer program product in whole or in part. The computer program product for video similarity detection includes one or more computer instructions for video similarity detection. When these computer program instructions are loaded and executed on the computer, the process or function described in FIG. 6 according to the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website, computer, server, or data center. Transmission to another website, computer, server, or data center via wired (such as coaxial cable, optical fiber, digital subscriber line, or wireless (such as infrared, wireless, microwave, etc.)). The computer-readable storage medium stores the video A readable storage medium of similarly detected computer program instructions. The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center integrated with one or more available media. The usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, an SSD).

Claims

A method for predicting the destination of a vehicle, characterized in that the method includes:

Acquiring the trajectory data of the vehicle to be predicted in the travel process and the travel data of the vehicle to be predicted in the traffic area;

According to the trajectory data, the travel data and the target neural network model, the destination information of the vehicle to be predicted in the traffic area is obtained, and the destination information includes: the destination sub-area of the vehicle to be predicted and The type of the POI of the destination point of interest of the vehicle to be predicted; the travel data of the vehicle to be predicted includes one or more of the following data: vehicle type, travel weather type, number of vehicle travels in the first time period, The frequency of vehicle travel in the second time period, and the sub-time period of vehicle travel in the third time period.
The method of claim 1, wherein the method further comprises:

According to the destination information of the vehicle to be predicted, determining that the destination is the traffic volume corresponding to the type of POI in the destination sub-area;

Predict the traffic state of the road in the target sub-region according to the traffic volume.
The method of claim 1 or 2, wherein the target neural network model includes an embedded model, a first feature extraction model, a second feature extraction model, a fusion model, a first classification model, and a second classification model , Wherein the embedding model is used to vectorize data input to the embedding model, and the first feature extraction model and the second feature extraction model are respectively used to perform data input to the first feature extraction model. Perform feature extraction with the data of the second feature extraction model, the fusion model is used to perform feature fusion on the data input to the fusion model, and the first classification model and the second classification model are respectively used for The input data of the first classification model and the second classification model performs category prediction.
The method of claim 3, wherein the obtaining the destination information of the vehicle to be predicted in the traffic area according to the trajectory data, the travel data and the target neural network model comprises:

Input the trajectory data and the travel data to the embedded model to obtain the initial trajectory characteristics and initial travel characteristics of the vehicle to be predicted;

Input the initial trajectory feature to the first feature extraction model to obtain the trajectory feature of the vehicle to be predicted;

Input the initial travel feature to the second feature extraction model to obtain the travel feature of the vehicle to be predicted;

Input the trajectory feature and the travel feature to the fusion model to obtain the driving feature of the vehicle to be predicted;

Input the driving feature to the first classification model to obtain the target sub-region of the vehicle to be predicted;

Input the driving feature to the second classification model to obtain the target POI type of the vehicle to be predicted.
The method according to any one of claims 1 to 4, wherein the acquiring trajectory data of the vehicle to be predicted in the traffic area during travel includes:

According to the passing data in the traffic area, determine information about multiple locations that the to-be-predicted vehicle passes through during travel;

Acquiring sub-area information in the traffic area;

Determine the trajectory data of the vehicle to be predicted in the travel process according to the information of the multiple locations that the vehicle to be predicted passes through during the travel and the sub-region information in the traffic area.
The method according to claim 5, wherein the trajectory data includes position information and time information of the sub-area that the vehicle to be predicted passes through in the traffic area.
The method according to claim 5 or 6, wherein the trajectory data further includes the type of POI that the vehicle to be predicted passes through in the traffic area.
The method according to any one of claims 1-7, wherein the target neural network model is a neural network model trained by training data, and the training data includes the history of vehicles in the traffic area Trajectory data and travel data of the vehicle.
A device for predicting the destination of a vehicle, characterized in that the device comprises:

An acquisition module for acquiring the trajectory data of the vehicle to be predicted in the travel process and the travel data of the vehicle to be predicted in the traffic area;

The prediction module is used to obtain the destination information of the vehicle to be predicted in the traffic area according to the trajectory data, the travel data and the target neural network model, and the destination information includes: the vehicle to be predicted The destination sub-area of the vehicle to be predicted and the type of the POI of the destination point of interest of the vehicle to be predicted; the travel data of the vehicle to be predicted includes one or more of the following data: vehicle type, travel weather type, and data in the first time period The number of vehicle trips, the frequency of vehicle trips in the second time period, and the sub-time periods of vehicle trips in the third time period.
The device of claim 9, wherein the prediction module is further configured to:

According to the destination information of the vehicle to be predicted, determining that the destination is the traffic volume corresponding to the type of POI in the destination sub-area;

Predict the traffic state of the road in the target sub-region according to the traffic volume.
The device according to claim 9 or 10, wherein the target neural network model includes an embedded model, a first feature extraction model, a second feature extraction model, a fusion model, a first classification model, and a second classification model , Wherein the embedding model is used to vectorize data input to the embedding model, and the first feature extraction model and the second feature extraction model are respectively used to perform data input to the first feature extraction model. Perform feature extraction with the data of the second feature extraction model, the fusion model is used to perform feature fusion on the data input to the fusion model, and the first classification model and the second classification model are respectively used for The input data of the first classification model and the second classification model performs category prediction.
The device according to claim 11, wherein the prediction module is specifically configured to:

Input the trajectory data and the travel data to the embedded model to obtain the initial trajectory characteristics and initial travel characteristics of the vehicle to be predicted;

Input the initial trajectory feature to the first feature extraction model to obtain the trajectory feature of the vehicle to be predicted;

Input the initial travel feature to the second feature extraction model to obtain the travel feature of the vehicle to be predicted;

Input the trajectory feature and the travel feature to the fusion model to obtain the driving feature of the vehicle to be predicted;

Input the driving feature to the first classification model to obtain the target sub-region of the vehicle to be predicted;

Input the driving feature to the second classification model to obtain the target POI type of the vehicle to be predicted.
The device according to any one of claims 9-12, wherein the acquisition module is specifically configured to:

According to the passing data in the traffic area, determine information about multiple locations that the to-be-predicted vehicle passes through during travel;

Acquiring sub-area information in the traffic area;

Determine the trajectory data of the vehicle to be predicted in the travel process according to the information of the multiple locations that the vehicle to be predicted passes through during the travel and the sub-region information in the traffic area.
The device according to claim 13, wherein the trajectory data includes position information and time information of the sub-regions that the vehicle to be predicted passes through in the traffic area.
The device according to claim 13 or 14, wherein the trajectory data further includes the type of POI that the vehicle to be predicted passes through in the traffic area.
The device according to any one of claims 9-15, wherein the target neural network model is a neural network model trained by training data, and the training data includes the history of vehicles in the traffic area Trajectory data and travel data of the vehicle.
A computing device for predicting the destination of a vehicle, characterized in that the computing device includes a processor and a memory, wherein:

Computer instructions are stored in the memory;

The processor executes the computer instructions to implement the method of any one of claims 1-8.
A computer-readable storage medium, characterized by comprising instructions, when the instructions run on a processor, the processor executes the method according to any one of claims 1-8.