KR102493362B1 - Real-time accident detection system for waterworks using AI and method thereof - Google Patents

Abstract

According to one embodiment of the present invention, a model learning step of learning a pressure prediction model using a learning dataset including pressure and flow rate measured by a measuring instrument in a monitoring section for detecting whether an accident has occurred in the waterworks network, learning The pressure prediction step of predicting the pressure at each measurement point within the monitoring section by inputting the pressure or flow rate received in real time from the instrument into the pressure prediction model. Provides a real-time accident detection method and server for water supply using AI, including an accident detection step, detects pressure change when a pipe breakage accident occurs, recognizes the accident immediately, and causes a large pressure change at the measurement point adjacent to the pipe breakage accident It is possible to support rapid accident recovery by estimating the accident occurrence point based on the phenomenon and providing the accident location to the manager based on the GIS view map.

Description

Real-time accident detection system for waterworks using AI and method thereof}

The present invention relates to a water supply real-time accident detection system and method using AI.

A waterworks is a large-scale facility capable of supplying a large amount of water to a large number of areas. In the case of water supply, accidents occur in which pipes are damaged due to various causes. The main causes of pipeline damage include aging, water shock caused by sudden stop of pumps and sudden closing of valves, changes in topography in which pipelines are buried, and pipeline damage accidents caused by excavation works around pipelines. Due to the nature of pipelines buried underground, it is difficult to quickly identify the location of damage in the event of a damage accident. As a method for detecting whether or not the pipeline is damaged or the location of the damage, set individual upper and lower limits for flow or pressure gauges and monitor whether they are exceeded, or use pipe network analysis to detect abnormalities with the difference between the online analysis result and the measured value, Based on the statistical forecast of water demand, there are methods such as monitoring for abnormalities by the difference from the actual supply. However, these existing methods are difficult to distinguish between an increase in supply (demand) and an increase in flow due to pipeline damage due to the time variability of water supply. It is difficult to specify. In addition, if the pipe line is changed due to construction, the detection system must be current by reflecting the changes, and professional manpower and long-time work are required.

KR 10-1205103 B1

An object according to an embodiment of the present invention is the current flow rate, pressure, pump On / Off and electric valve opening degree measured at all measurement points existing in the determined analysis section of the sensor network map, and the pipeline system that affects the hydraulic behavior. To provide a system and method for detecting a pipe breakage accident in real time and estimating and providing the accident location as the location of the pressure gauge by using an artificial intelligence model that predicts the pressure at each pressure gauge installation point generated by learning the data. .

Water supply real-time accident detection method using AI according to an embodiment of the present invention, pressure, flow rate, pump On / Off and opening degree of electric valve measured by a measuring instrument in a monitoring section for monitoring whether an accident occurs in a water supply network A model learning step of learning a pressure prediction model using a learning data set including, inputting the pressure, flow rate, pump On/Off and the opening value of the motorized valve received in real time from the instrument into the learned pressure prediction model. It may include a pressure estimation step of predicting the pressure of each pressure measurement point in the monitoring section, and an accident detection step of detecting whether an accident has occurred by comparing the predicted pressure with the pressure measured in real time.

In addition, in the real-time water supply accident detection method using AI according to an embodiment of the present invention, before the model learning step, the pressure, flow rate, pump On / Off and opening of the electric valve received from the meter installed in the water supply network A data pre-processing step of correcting errors in real-time measurement data of values may be further included.

In addition, the real-time water supply accident detection method using AI according to an embodiment of the present invention may further include a section determining step of automatically determining a monitoring section based on a sensor network map before the model learning step.

In addition, the water supply real-time accident detection method using AI according to an embodiment of the present invention, when an accident is detected, automatically estimates the location where the accident occurred in the monitoring section, and reports the accident to the user terminal based on the GIS view. An accident notification step of notifying the accident location may be further included.

In addition, in the model learning step, the pressure and flow rate of a plurality of measurement points included in the monitoring section, the pump operating state, and the opening value of the electric valve are learning data under normal conditions where no accident has occurred, and included in the monitoring section It may include a learning data set generating step of generating a learning data set in which the pressure of any one measurement point is label data, and a training step of learning the pressure prediction model with the learning data set.

In addition, the pressure prediction step includes an input data generation step of generating input data in real time from the flow rate and pressure received from the measuring instrument, the operation state of the pump, and the opening value of the electric valve, and the learned pressure prediction model using the input data. It may include a model use step of predicting the pressure at the measurement point by inputting to.

In addition, the error means that there is a flow rate or pressure that is incorrectly measured or missing from the meter, and the data preprocessing step is a number determination step of determining whether the number of errors is greater than a predetermined number, when the number of errors is less than the predetermined number , an interpolation correction step of correcting errors using an interpolation method, and a period correction step of correcting errors using data of a previous period when the number of errors is greater than a predetermined number.

In addition, the section determining step may automatically determine the monitoring section as an area in which the flow rate flowing in and the flow rate flowing out of the sensor network map are the same.

A water supply real-time accident detection system using AI according to an embodiment of the present invention includes an accident detection server, wherein the accident detection server has a pressure measured by an instrument in a monitoring section for detecting whether an accident has occurred in the water supply network. and a model learning unit that learns a pressure prediction model using a learning data set including flow rate, and inputs the pressure or flow rate received in real time from the instrument to the learned pressure prediction model, thereby inputting the pressure at each measurement point within the monitoring section. It may include an accident detection unit that predicts and compares the predicted pressure with the pressure measured in real time to detect whether an accident has occurred.

In addition, the water supply real-time accident detection system using AI according to an embodiment of the present invention may further include a data collection unit for correcting an error in pressure or flow rate received from the meter installed in the water supply network.

In addition, the water supply real-time accident detection system using AI according to an embodiment of the present invention, when an accident is detected, estimates the location where the accident occurred in the monitoring section, and the accident occurs and the location of the accident to the user terminal based on the GIS view map. It may further include an accident notification unit notifying.

In addition, the model learning unit automatically determines the monitoring section based on the sensor network map, and the pressure and flow rate of a plurality of measurement points included in the monitoring section, the pump operating state, and the electric valve under normal conditions without an accident. A learning data set in which an opening degree is learning data and a pressure at any one measurement point included in the monitoring section is label data may be generated, and the pressure prediction model may be trained with the generated learning data set.

In addition, the accident detection unit predicts the pressure at the measuring point by inputting the flow rate and pressure received from the instrument in real time to the pressure prediction model, the operating state of the pump, and the opening value of the electric valve, and predicts the pressure at the measured point and the instrument It may be determined that an accident has occurred when the predicted pressure is greater than the received pressure by a predetermined range or more by comparing the received pressure.

In addition, the error is that there is a flow rate or pressure that is incorrectly measured or missing from the meter, and the data collection unit determines whether the number of errors is greater than the predetermined number, and if the number of errors is less than the predetermined number, interpolation is used Errors are corrected, and if the number of errors is greater than a predetermined number, errors may be corrected using data at a specific point in time.

Features and advantages of the present invention will become more apparent from the following detailed description taken in conjunction with the accompanying drawings.

Prior to this, the terms or words used in this specification and claims should not be interpreted in a conventional and dictionary sense, and the inventor may appropriately define the concept of the term in order to explain his or her invention in the best way. It should be interpreted as a meaning and concept consistent with the technical idea of the present invention based on the principle that there is.

According to an embodiment of the present invention, it is possible to immediately recognize the occurrence of an accident by predicting the pressure at a measurement point existing in a defined monitoring section of the sensor network map and comparing it with the actually measured pressure value to detect a change in pressure when a pipe breakage accident occurs. can

In addition, according to an embodiment of the present invention, the accident occurrence point is estimated based on the phenomenon that the pressure difference between the pipeline breakage accident and the adjacent measurement point is greater than the pressure difference between the pipe breakage accident and the distant measurement point, and based on the GIS pipe network map By providing the location of the accident to the administrator, it is possible to support rapid accident recovery.

1 is a diagram showing a water supply real-time accident detection system using AI according to an embodiment of the present invention.
2 is a diagram showing the configuration of a water supply real-time accident detection system using AI according to an embodiment of the present invention.
3 is a diagram showing a sensor network map according to an embodiment of the present invention.
4 is a table showing an example of a first learning dataset generated to learn a first pressure prediction model for predicting pressure at a measurement point to which a first pressure gauge is connected.
Figure 5 is the input data generated by the accident detection unit to predict the pressure at the measuring point connected to the first pressure gauge in the first monitoring section, the output data predicted by the pressure prediction model, the pressure measured by the first pressure gauge, and the error rate It is a table showing exemplarily.
6 is a diagram showing the predicted pressure and the measured pressure for each measurement point in the first monitoring section according to an embodiment of the present invention.
7 is a flowchart showing each step of the real-time accident detection method for water supply using AI according to an embodiment of the present invention.
8 is a flowchart showing in detail a data pre-processing step according to an embodiment of the present invention.
9 is a graph showing a comparison between predicted pressure and measured pressure at a measurement point to which a fifth pressure gauge is connected according to an embodiment of the present invention.
10 is a graph showing an error between a predicted pressure and a measured pressure at a measurement point to which a fifth pressure gauge is connected according to an embodiment of the present invention.

Objects, advantages, and features of one embodiment of the present invention will become more apparent from the following description of one embodiment in conjunction with the accompanying drawings. In adding reference numerals to components of each drawing in this specification, it should be noted that the same components have the same numbers as much as possible, even if they are displayed on different drawings. In addition, terms such as "one side", "other side", "first", and "second" are used to distinguish one component from another, and the components are not limited by the above terms. no. Hereinafter, in describing an embodiment of the present invention, a detailed description of related known technologies that may unnecessarily obscure the subject matter of an embodiment of the present invention will be omitted.

Hereinafter, with reference to the accompanying drawings, an embodiment of the present invention will be described in detail.

1 is a diagram showing a water supply real-time accident detection system using AI according to an embodiment of the present invention.

As shown in FIG. 1, the water supply real-time accident detection system using AI according to an embodiment of the present invention is connected to the water supply network to measure the flow rate or pressure, and the meter 20 measures It may include an accident detection server 100 that detects whether an accident has occurred by receiving flow rate or pressure. The accident detection server 100 may inform the user terminal 30 of the occurrence and location of an accident when an accident occurs.

A waterworks network refers to a pipe network that transports purified water from a water source to a consumer. Waterworks include multi-regional waterworks that transport large amounts of water and between regions using large pipelines, local waterworks that are connected to multiregional waterworks and transport water to consumers, and water that is purified from simple water intakes and supplied to small areas. It is a concept that includes simple constants and the like.

A meter 20 is installed in the waterworks network to measure the flow of water. The measuring instrument 20 includes various types of measuring instruments 20 such as pressure gauges, flow meters, turbidity meters, and water thermometers. Meter 20 may measure more than one item at a point. For example, meter 20 may measure both pressure and flow at one point. In this specification, the instrument 20 may be understood to include a pressure gauge or a flowmeter. A plurality of measuring instruments 20 are installed in the water supply network. The measuring instrument 20 may transmit the measured pressure or flow rate to the accident detection server 100 . The meter 20 may measure pressure or flow rate at predetermined intervals, store the measured pressure or flow rate in the storage unit 160, and transmit the stored pressure or flow rate to the accident detection server 100 at predetermined intervals. For example, the meter 20 measures the pressure or flow rate once every 15 seconds, continuously stores the measured pressure or flow rate in the storage unit 160, and detects an accident of the pressure or flow rate stored for 5 minutes every 5 minutes. It can be transmitted to the server 100. The measuring instrument 20 may be connected to a wired or wireless network to transmit/receive data with the accident detection server 100 .

The accident detection server 100 may detect whether an accident such as damage to the water supply pipe has occurred using the flow rate or pressure measured by the meter 20 . The accident detection server 100 may be operated by institutions such as companies, public corporations, national governments, and local governments that manage waterworks networks. The accident detection server 100 may be composed of one or more server devices. For example, the accident detection device is in charge of detailed functions such as a server for collecting data, a database 120 for storing the collected data, a server for determining whether an accident has occurred, a server for notifying the occurrence of an accident to the user terminal 30, and the like. It may consist of a plurality of server devices that do. Even if the accident detection server 100 is composed of a plurality of server devices, as long as each server device performs an operation according to an embodiment of the present invention, it is entirely included in the scope of the present invention.

The user terminal 30 is an information processing device capable of receiving information about the occurrence and location of an accident from the accident detection server 100 and displaying the information to the user. The user terminal 30 may include a PC, a smart phone, a monitoring terminal, a tablet PC, a notebook computer, and the like. The user may include a manager who manages water supply, an operator who operates the accident detection server 100, or a construction site manager in charge of accident recovery. The user terminal 30 displays the occurrence and location of the accident to the user so that the user can immediately respond to the accident.

As shown in FIG. 1, when a pipe breakage accident occurs at any location in the waterworks network, a pressure different from the pressure measured in a normal state will be measured in the measuring instrument 20 adjacent to the accident point. The accident detection server 100 predicts the pressure predicted to be measured by the instrument 20 in a normal state, compares the expected pressure with the actually measured pressure to determine whether an accident has occurred, and detects the occurrence of an accident. Accident occurrence and location may be provided to the terminal 30 .

2 is a diagram showing the configuration of a water supply real-time accident detection server 100 using AI according to an embodiment of the present invention.

As shown in Figure 2, the water supply real-time accident detection system using AI according to an embodiment of the present invention includes an accident detection server 100 as a real-time accident detection system, and the accident detection server 100 is a water supply network Model learning for learning a pressure prediction model using a learning dataset including the pressure, flow rate, operating state of the pump, and opening value of the electric valve measured by the instrument 20 in the monitoring section to detect whether an accident has occurred within Unit 130, and the pressure received in real time from the measuring instrument 20, the flow rate, the operating state of the pump, the opening value of the motorized valve into the pressure prediction model learned to predict the pressure at each pressure measurement point in the monitoring section , It may include an accident detection unit 140 that detects whether an accident has occurred by comparing the predicted pressure with the measured pressure in real time. In addition, the water supply real-time accident detection server 100 using AI according to an embodiment of the present invention includes a data collection unit 110 for correcting errors in pressure or flow rate received from the meter 20 installed in the water supply network, accident is detected, the accident notification unit 150 and the data collection unit 110, which estimate the location of the accident in the monitoring section and inform the user terminal 30 of the occurrence and location of the accident based on the GIS view map, collect and organize A database 120 for storing data such as pressure or flow rate may be further included.

The accident detection server 100 is connected to a wired or wireless network and includes a communication unit 170 capable of transmitting and receiving data to and from the measuring instrument 20 or the user terminal 30, data and data necessary to operate the accident detection server 100. A user who manages the storage unit 160 and the accident detection server 100 for storing program codes for performing the real-time accident detection method for water supply using AI according to an embodiment of the invention inputs data or commands or It may further include an input/output unit 180 that displays monitoring conditions and the like to the user.

The data collection unit 110 may receive measurement values from the measuring instrument 20 through a wired or wireless network. The data collection unit 110 receives the pressure or flow rate data received from the meter 20 in units of 1 minute or at other predetermined intervals, and transmits the received pressure or flow rate data in relation to the location of the meter 20 in the waterworks network. It can be organized and stored in the database 120. The data collection unit 110 may correct the error when there is an error in the data received from the measuring instrument 20 .

An error present in the data received from the meter 20 is the presence of a flow rate or pressure that is erroneously measured or missing from the meter 20 . The data collection unit 110 determines whether the number of errors is greater than the predetermined number, if the number of errors is less than the predetermined number, corrects the errors using an interpolation method, and if the number of errors is greater than the predetermined number, at a specific point in time Errors can be corrected using the data.

The data collection unit 110 may receive measurement values from various types of measuring instruments 20 such as pressure gauges, flow meters, and turbidimeters. When the data collection unit 110 collects data, misinformation or missing errors may occur due to various causes, such as a problem with the measuring instrument 20 or a problem with data transmission and reception.

A misstatement is an error in which data are delivered in excess or underestimation compared to the data normally measured. The misjudgement may be determined based on whether data is larger or smaller than a reference value generated based on data collected during a predetermined period. Specifically, a false positive may be determined as a false positive when data exceeding this value is received based on 50 times the value corresponding to the 95th order when the data collected for the past two weeks are arranged in ascending order.

Missing is an error in which data are not collected. This may occur for various reasons, such as failure to measure due to a problem with the measuring instrument 20, a measured value not being stored, or a measured value not being transmitted. Missing may be determined as missing when there is no data received by the data collection unit 110 at a predetermined time.

The data collection unit 110 uses an interpolation method to correct errors or missing data when the number is less than a predetermined number. The interpolation method generates data of the part where the error occurred by using data before and after the data where the error occurred in time. When correcting misleading or missing data by interpolation, the average of data before and after can be used.

The data collection unit 110 may correct errors using data of a specific point in time when the number of erroneous or missing data is greater than a predetermined number. Here, a specific point in time may be a cycle for forming a new pressure prediction model. When the pressure prediction model is updated every 2 weeks, the error can be corrected by taking the data of the same time period from 2 weeks ago and inserting it into the wrong or missing part. Alternatively, the error can be corrected by importing data from the same time zone of the previous day and inserting it into the wrong or missing part.

The database 120 may store the pipe connection structure of the waterworks network, the location of the pipe in a Geographic Information System (GIS), the location of the measuring instrument 20, and the measured value of the measuring instrument 20 according to time. Since the location of the pipe and the location of the measuring instrument 20 are stored in association with the GIS-based GIS pipe map, location information according to the GIS pipe map can be obtained by searching for any pipe or measuring instrument 20. The database 120 may store a sensor network map.

3 is a diagram showing a sensor network map according to an embodiment of the present invention. 3 illustratively shows a part of the entire sensor network map. In FIG. 3, the pressure gauge is indicated by P and the flow meter is indicated by F. The number displayed after P or F, such as 'P1' or 'F2', is to distinguish the pressure gauge. In FIG. 3, the conduit and the divergence of the conduit are indicated by lines.

See Figures 2 and 3 together. A sensor network map is a map reflecting the connection relationship between pipelines, branching of pipelines, measuring instruments 20 connected to pipelines, pumps connected to pipelines, and electric valves connected to pipelines. The data collection unit 110 automatically stores changes in the corresponding pipe in the database 120 when information such as new establishment, change, or closure of a pipeline, operation of a pump, opening/closing of an electric valve, or opening value of an electric valve is received. Update the sensor network map. The opening degree of the motorized valve may be a value between 0% when the motorized valve is closed and 100% when all are open. In other words, the sensor network map reflects elements for operating the waterworks network.

The sensor network map may be divided into arbitrary sections. It can be divided into a first monitoring area (area-1) indicated by a dotted line in FIG. 3 or a second monitoring area (area-2) indicated by a dashed-dot line. The monitoring period is a partial area of the sensor network map. The monitoring section can be determined by determining the hydraulic connectivity based on points where the upstream and downstream are hydraulically disconnected, such as reservoirs and pumping stations, and points where a flow meter is installed on the downstream main pipeline to measure the inflow and outflow flow rates in the section simultaneously. The monitoring period may be automatically determined by the model learning unit 130 .

The model learning unit 130 automatically determines the monitoring section based on the sensor network map, and in normal conditions without an accident, the pressure and flow rate of a plurality of measurement points included in the monitoring section, the pump operating state, and the electric valve A learning data set in which the opening degree is the learning data and the pressure at any one measurement point included in the monitoring section is the label data can be created, and the pressure prediction model can be trained with the created learning data set.

The model learning unit 130 may determine a partial area of the sensor network map as a monitoring zone. The model learning unit 130 may automatically determine a monitoring period based on hydraulic connectivity. In the sensor network map, there may be locations where only pressure gauges are connected, only flow meters are connected, or both pressure and flow meters are connected. The model learning unit 130 includes a flowmeter capable of measuring the amount of water flowing into the pipe network system including the pipe line when an arbitrary pipe line is selected, and a flow meter capable of measuring the amount of water flowing out of the pipe network system. The monitoring section can be automatically determined based on the flow meter. In other words, the monitoring section may be automatically determined as an area where the inflow and outflow flows are the same in the sensor network map.

For example, the model learning unit 130 selects an arbitrary first pipe (Pipe-1) from the sensor network map of FIG. 3 and measures the flow rate flowing into the first pipe (pipe-1). (F1), an area partitioned by the second to sixth flow meters (F2 to F6) measuring the flow rate flowing out of the first pipeline (pipe-1) may be determined as the first monitoring section (area-1). The model learning unit 130 may automatically include a pump or motorized valve that may affect the pressure or flow rate of the pipeline in the monitoring section. Therefore, the first monitoring section (area-1) includes a pump for pumping water flowing into the first pipe (pipe-1).

For example, the model learning unit 130 may select an arbitrary second pipe (pipe-2) not included in the first monitoring section (area-1) in the sensor network map of FIG. 3 . In addition, the model learning unit 130 includes a fourth flow meter F4 measuring the flow rate flowing into the second pipe-2 and a seventh flow meter measuring the flow rate flowing out of the second pipe-2 (pipe-2). F7), the eighth flowmeter (F8), and the ninth flowmeter (F9) are recognized on the sensor network map, and whether the motorized valve connected to the second pipe (pipe-2) is opened in real time is checked. The area including the 5th flow meter (F5) measuring the flow rate flowing into pipe-3 and the 10th flow meter (F10) measuring the flow rate flowing out of the 3rd pipeline (pipe-3) is included in the second monitoring area (area). -2) can be determined. The model learning unit 130 may automatically include a pump or motorized valve that may affect the pressure or flow rate of the pipeline in the monitoring section. Therefore, the second monitoring section (area-2) includes an electric valve capable of connecting or blocking the second pipe (pipe-2) and the third pipe (pipe-3).

The model learning unit 130 may generate pressure prediction models as many as the number of pressure gauges included in the monitoring section. For example, there are 6 pressure gauges (P1 to P6) included in the first monitoring section (area-1), and 6 pressure prediction models for predicting pressures at measurement points to which each pressure gauge is connected can be generated. That is, the first pressure prediction model predicts the pressure at the location where the first pressure gauge P1 is connected, the second pressure prediction model predicts the pressure at the location where the second pressure gauge P2 is connected, and the third pressure gauge P3 is connected. A third pressure prediction model predicting the pressure at the location, a fourth pressure prediction model predicting the pressure at the location where the fourth pressure gauge P4 is connected, and a fifth pressure prediction model predicting the pressure at the location where the fifth pressure gauge P5 is connected. A sixth pressure prediction model for predicting the pressure at a location where the model, the sixth pressure gauge P6 is connected, may be generated. The pressure prediction model may use a deep neural network (DNN) model. The pressure prediction model includes an input layer, a hidden layer, and an output layer, and the number of nodes of the input layer and hidden layer or the number of layers of the hidden layer may be automatically adjusted according to the performance evaluation result. When the surveillance section of the sensor network map is changed due to the new construction, change, or closure of a pipeline, the input layer, hidden layer, and output layer of the pressure prediction model can be automatically changed to suit the changed monitoring section.

The model learning unit 130 retrieves data of pressure gauges, flowmeters, pumps, and electric valves included in the monitoring section collected and stored in the database 120 in real time (by 1 minute) to create a training data set. can create The training data set may include training data and labeled data. The learning data may include the pressure measured by the pressure gauge included in the monitoring section, the flow rate measured by the flowmeter, whether the pump is operating, and the opening value of the motorized valve. The label data is the pressure measured by the pressure gauge connected to the position where the pressure is to be predicted. A learning dataset may be created using data collected during a predetermined period. Learning data may be generated using data collected for the last two weeks.

The model learning unit 130 may generate as many learning data sets as the number of pressure gauges included in the monitoring section. For example, there are 6 pressure gauges included in the first monitoring period (area-1), and 6 learning datasets need to be created to learn each of the 6 pressure prediction models that predict the pressure at the location where each pressure gauge is connected. can In addition, the learning data includes the pressure gauge, flowmeter, pump, and electric valve data except for the pressure gauge at the point where the pressure is to be predicted, and the label data is the pressure measured by the pressure gauge at the point where the pressure is to be predicted.

FIG. 4 is a table showing an example of a first training data set generated to learn a first pressure prediction model for predicting the pressure at a measurement point to which the first pressure gauge P1 is connected.

As shown in FIG. 4, the first learning dataset for learning the first pressure prediction model for predicting the pressure at the measuring point to which the first pressure gauge P1 is connected is the second to sixth pressure gauges P2 to P6 It may include learning data consisting of data of the first to sixth flowmeters F1 to F6 and pumps, and label data consisting of measured values of the first pressure gauge P1. In FIG. 4, the pump value of 1 represents a state in which the pump is operating. The model learning unit 130 may generate a second learning dataset for learning a second pressure prediction model for predicting the pressure at a location where the second pressure gauge P2 is connected in the form shown in FIG. A dataset, a fourth learning dataset, a fifth learning dataset, and a sixth learning dataset may also be created.

The model learning unit 130 may learn a pressure prediction model generated according to a point to predict pressure using a learning data set generated according to a point to predict pressure. When the model learning unit 130 completes the process of learning the six pressure prediction models included in the first monitoring period (area-1), the accident detection unit 140 determines whether an accident has occurred by using the learned pressure prediction model. detect

The accident detection unit 140 predicts the pressure at a point where any one pressure gauge is located in the monitoring section using the learned pressure prediction model, and compares the predicted pressure with the actually measured pressure to detect whether an accident has occurred. . The accident detection unit 140 predicts the pressure at the measuring point by inputting the flow rate and pressure received from the instrument 20, the operation state of the pump, and the opening value of the electric valve in real time to the pressure prediction model, and the predicted pressure and the instrument By comparing the pressure received from (20), it can be determined that an accident has occurred if the predicted pressure differs significantly from the received pressure by more than a predetermined range. The measuring point at which the accident detection unit 140 predicts the pressure is a point where any one pressure gauge included in the monitoring section is connected.

The accident detection unit 140 inputs the data collected by the data collection unit 110 in real time to the pressure prediction model learned based on the data collected by the data collection unit 110 for the last about two weeks to connect the pressure gauge. The pressure at the measuring point can be predicted. Since the data of the pressure gauge, flowmeter, pump, and electric valve included in the monitoring section are updated in real time by the data collection unit 110 and stored in the database 120, the accident detection unit 140 inputs them in the same form as the learning data. It is possible to obtain output data in the same form as the label data by generating data and inputting the data to the learned pressure prediction model.

5 is input data generated by the accident detection unit 140 to predict the pressure at the measuring point to which the first pressure gauge P1 of the first monitoring section (area-1) is connected, and output data predicted by the pressure prediction model And, a table showing the pressure measured by the first pressure gauge P1 and an error by way of example. The error may be calculated according to Equation 1 below.

(Predicted pressure = pressure at the measuring point predicted by the pressure prediction model, measured pressure = pressure actually measured at the measuring point by the pressure gauge)

As shown in FIG. 5, the first input data for predicting the pressure at the measuring point to which the first pressure gauge P1 is connected is the second to sixth pressure gauges P2 to P6, the first to sixth flowmeters F1 to F6), may include pump data. The first input data is input to the first pressure prediction model, and the first pressure prediction model outputs first output data. The first output data is the predicted pressure at the measurement point to which the first pressure gauge P1 is connected. In the same way, it is possible to predict the pressure at each measurement point to which the second to sixth pressure gauges P2 to P6 are connected.

The accident detection unit 140 may determine whether an accident has occurred by comparing the output data and the measured data. The output data is the predicted value of the pressure at the measuring point to which any one pressure gauge among the plurality of pressure gauges included in the monitoring section is connected. The measurement data is the pressure actually measured by the corresponding pressure gauge. The accident detecting unit 140 compares the predicted pressure with the measured pressure, and can estimate that an accident has occurred when the predicted pressure is significantly different from the measured pressure by more than a predetermined range.

6 is a diagram showing the predicted pressure and the measured pressure for each measurement point in the first monitoring period (area-1) according to an embodiment of the present invention.

As shown in FIG. 6, the accident detection unit 140 may compare the pressure predicted by the pressure prediction model and the pressure measured by each pressure gauge at each measurement point where the first to sixth pressure gauges P1 to P6 are connected. . 6 exemplarily shows a case in which an accident occurred at 2020-01-24 15:17. The accident point (AP) passes through the divergence of the pipe leading from the first pipe (pipe-1) to the fourth pressure gauge (P4), and the pipe connecting the fifth pressure gauge (P5) and the sixth pressure gauge (P6) is divided. This is the position before losing. Before the accident occurred (2020-01-24 15:14~14:16), it can be seen that the error between the predicted pressure and the measured pressure at all measurement points within the monitoring section is within the range of -2% or more and 2% or less. there is. It can be seen that the pressure prediction model works well because the predicted pressure and the measured pressure in a steady state without an accident are almost identical.

At the time of the accident (2020-01-24 15:17), a large difference occurs when the predicted pressure and the measured pressure are compared at each measurement point. An error of 9.47% occurs at the measuring point of the first pressure gauge (P1), an error of 8.32% occurs at the measuring point of the second pressure gauge (P2), and an error of -60.97% occurs at the measuring point of the third pressure gauge (P3). 23.64% error occurs at the measuring point of the fourth pressure gauge P4, 42.74% error occurs at the measuring point of the fifth pressure gauge P5, and -10.29 error occurs at the measuring point of the sixth pressure gauge P6. % Occurs. Overall, it can be seen that a large error occurs at the time of the accident. The accident detection unit 140 may determine that an accident has occurred when the difference between the predicted pressure and the measured pressure at each measurement point is significantly greater than a predetermined range. The accident detecting unit 140 may determine that an accident has occurred when the result of calculating the error is greater than a predetermined range (the error is 5% or 10% or more).

When an accident occurs in which the pipe is broken and water leaks at a certain point in the pipe, the pressure in the pipe is lowered, so the measured pressure is smaller than the predicted pressure at the measuring point to which the pressure gauge is connected. The accident detecting unit 140 may recognize that an accident has occurred when the pressure predicted at each measurement point within the monitoring section is significantly greater than the measured pressure by a predetermined range or more. For example, in FIG. 6, the measured pressure is greater than the predicted pressure at the measurement points of the first pressure gauge P1, the second pressure gauge P2, the fourth pressure gauge P4, and the fifth pressure gauge P5, so an accident may occur. can be judged to have occurred. As the distance from the point of occurrence of the accident increases, the influence of pressure fluctuations becomes smaller, and the difference between the predicted pressure and the measured pressure becomes smaller.

There may also be measurement points where the error is judged as a (-) value. For example, the measurement point where the third pressure gauge P3 and the sixth pressure gauge P6 in FIG. 6 are connected has an error determined as a (-) value. The waterworks network operates pumps alternately from time to time to control the amount of supply and optimize operation, and the customer side also controls valves from time to time to control the valve level. Since the above control generates a water shock indicating a rapid decrease in pressure temporarily, and a pipe breakage accident may also generate a temporary water shock, there may be a measurement point where an error appears as (-) when an accident occurs. The accident detecting unit 140 does not use the measurement point where the error is (-) as a measurement point for determining whether an accident has occurred. The accident detecting unit 140 may determine that an accident has occurred if there is a measurement point having a larger error than a predetermined range among the measurement points having an error of (+).

The accident detection unit 140 compares the predicted pressure and the measured pressure to determine whether an accident has occurred, but does not determine whether an accident has occurred by comparing the predicted flow rate and the measured flow rate. The water usage patterns of customers differ depending on the customers, and the usage patterns of large users, which have a large impact on flow rate fluctuations, are caused by irregular valve operation, making it difficult to pattern water usage. Therefore, predicting the flow rate and comparing it with the measured flow rate It is not an appropriate judgment method to judge pipeline accidents due to the nature of water supply network operation. Therefore, the conventional technique of predicting the flow rate and performing accident detection has low reliability. However, the technology for detecting an accident by predicting the pressure proposed in the present invention is a pressure prediction model in which the pressure balance inside the pipe is momentarily broken when a pipe breakage accident occurs, so that the measured pressure of the pressure gauge adjacent to the accident point (AP) becomes very low. Reliability is high because it is detected by comparing with the predicted pressure.

However, when the accident detecting unit 140 compares the predicted pressure with the measured pressure and determines that an accident has occurred, additionally counts the flow rate of the water flowing into the pipeline and the water flowing out of the pipeline, and if an error of more than a predetermined range occurs It can be judged with more certainty that an accident has occurred. The resin table in FIG. 6 shows the difference between the amount of water flowing into the first pipe-1 through the first flow meter F1 and the amount of water flowing out through the second to sixth flow meters F2 to F6. it is shown If the resin is a (-) value, it means that the amount of water flowing out is large, and if the resin is a (+) value, it means that the amount of water flowing in is large. The resin may have a small value depending on the measurement error range of the flowmeter. If the balance is a value within a predetermined range, it is not determined that an accident has occurred, and if a value outside the predetermined range is displayed, it can be determined that an accident has occurred. It can be seen that the resin in FIG. 6 shows a small value before the accident, but shows a large value at the time of the accident. Since the amount of water flowing in the accident detection unit 140 is much greater than the amount of water flowing out of the first pipe-1, any part of the first pipe-1 is damaged and the amount corresponding to the resin is damaged. It can be seen that the water is leaking.

Reference is made to FIGS. 2 and 6 again. When the accident detection unit 140 determines that an accident has occurred, the accident notification unit 150 may provide the location of the accident location and the accident point (AP) to the user terminal 30 . The accident notification unit 150 may inform the user who manages the accident detection server 100 of the occurrence and location of an accident through the input/output unit 180 . Accident notification unit 150, when it is determined that an accident has occurred, based on the GIS stored in the database 120, the location of the actual map of the pipeline, the location of the measurement point where it is determined that the accident occurred, and the point where the accident is estimated to have occurred The location of may be provided to the user terminal 30 .

The accident notification unit 150 selects a measurement point where the error between the predicted pressure and the measured pressure is a (+) value within the monitoring section, and obtains the location of the measurement point based on the GIS stored in the database 120. there is. The accident notification unit 150 may determine which measurement point the accident point AP is close to based on the location of the corresponding measurement point acquired based on the GIS and the size of the error. For example, in FIG. 6, the measurement point at which the error is a (+) value is a point where the first pressure gauge P1, the second pressure gauge P2, the fourth pressure gauge P4, and the fifth pressure gauge P5 are connected. It can be determined that a pipe breakage accident has occurred between the measurement point where the fifth pressure gauge P5 with the largest error (42.74) is connected and the measurement point where the fourth pressure gauge P4 with the next largest error (23.64) is connected.

7 is a flowchart showing each step of the real-time accident detection method for water supply using AI according to an embodiment of the present invention.

As shown in FIG. 7, the real-time water supply accident detection method using AI according to an embodiment of the present invention includes a data collection step (S10) of receiving measurement data from the meter 20 installed in the water supply network, A data pre-processing step of correcting errors in pressure or flow rate received from the installed measuring instrument 20 (S20), a section determination step of automatically determining a monitoring section based on the sensor network map (S30), whether an accident has occurred in the waterworks network A model learning step (S40) of learning a pressure prediction model using a learning data set including the pressure and flow rate measured by the instrument 20 in the monitoring period for monitoring, from the instrument 20 to the learned pressure prediction model. A pressure prediction step (S50) of predicting the pressure at each measurement point within the monitoring section by inputting the pressure or flow rate received in real time, and an accident detection step of detecting whether an accident has occurred by comparing the predicted pressure with the measured pressure in real time. (S60), an accident notification step (S70) of estimating the location where an accident occurred in the monitoring section when an accident is detected and informing the user terminal 30 of the location of the accident based on the GIS may be included.

In the data collection step (S10), the data collection unit 110 of the accident detection server 100 collects data measured by the meter 20 from the meter 20 installed in the waterworks network. There are various types of measuring instruments 20, and each measuring instrument 20 may have a different method of providing data. Measurement data provided by the instrument 20 may be collected using a Supervisory Control And Data Acquisition (SCADA) system. The data collection unit 110 may collect measurement data such as flow rate or pressure measured by the meter 20 through the SCADA system.

8 is a flowchart showing in detail the data pre-processing step (S20) according to an embodiment of the present invention.

As shown in FIG. 8, in the data pre-processing step (S20), the data collection unit 110 of the accident detection server 100 corrects errors present in the pressure or flow rate data received from the measuring instrument 20 installed in the waterworks network. is to do The error may be that there is a misleading or missing flow rate or pressure from meter 20 .

The data preprocessing step (S20) includes a number determination step (S21) of determining whether the number of errors is greater than the predetermined number, an interpolation correction step (S22) of correcting errors using an interpolation method when the number of errors is less than the predetermined number, and If the number of errors is greater than the predetermined number, a periodic correction step (S23) of correcting errors using data at a specific time point may be included.

The data collection step (S10) and the data preprocessing step (S20) may be performed before the model learning step (S40), and may be repeatedly performed at predetermined intervals. The data collected and corrected by the data collection step ( S10 ) and the data preprocessing step ( S20 ) are stored in the database 120 . The operations performed by the data collection unit 110 in the data collection step (S10) and the data preprocessing step (S20) have been described above and will therefore be omitted.

The section determining step (S30) may be performed in the model learning unit 130 of the accident detection server 100. In the section determination step (S30), the monitoring section is automatically determined based on the sensor network map, but the monitoring section can be determined based on whether the sensor network map is hydraulically disconnected by reservoirs and pumping stations or whether the main inflow and outflow flow rate can be measured. . The monitoring section can be automatically determined based on the division of the hydraulically disconnected section through the analysis of hydraulic connectivity and the presence or absence of a flowmeter installed on the downstream main pipeline after the water purification plant outflow flowmeter. The section determining step (S30) may be performed at predetermined intervals. For example, if the period for learning the pressure prediction model is 2 weeks, the section determining step (S30) may be performed every 2 weeks. Alternatively, the section determining step (S30) may be performed when there is a change such as new construction, change, or closure of elements such as pipelines, measuring instruments 20, pumps, and electric valves in the waterworks network. When the interval determining step (S30) is performed and the monitoring interval is changed, the model learning step (S40) may be performed again to re-learn the pressure prediction model corresponding to the monitoring interval. Since the operation performed by the model learning unit 130 in the section determining step (S30) has been described above, it will be omitted.

In the model learning step (S40), the pressure and flow rate of a plurality of measurement points included in the monitoring section under normal conditions without an accident, the operating state of the pump, and the opening value of the electric valve are learning data, and any one included in the monitoring section It may include a learning data set generating step of generating a learning data set in which the pressure of the measurement point of is label data, and a training step of learning a pressure prediction model with the learning data set.

The model learning step (S40) may be performed in the model learning unit 130. The model learning step (S40) may be performed at predetermined intervals. For example, when the pressure prediction model is updated once every two weeks, the model learning step (S40) may be performed once every two weeks. Alternatively, when changes such as new establishment, change, or closure of elements such as pipelines, measuring instruments 20, pumps, electric valves, etc. of the water supply network are reflected in the sensor network map and the monitoring section is newly determined, corresponding to the newly determined monitoring section A model learning step (S40) may be performed to learn the pressure prediction model. Since the operation performed by the model learning unit 130 in the model learning step (S40) has been described above, it will be omitted.

The pressure prediction step (S50) may be performed by the accident detection unit 140 of the accident detection server 100. The pressure prediction step (S50) is an input data generation step of generating the flow rate and pressure received from the measuring instrument 20 in real time, the operating state of the pump, and the opening value of the electric valve as input data, and A model use step of predicting the pressure at the measurement point by inputting the input data into the learned pressure prediction model may be included.

The step of generating input data is to generate, as input data, pressure and flow data received by the data receiving unit in real time and error-corrected, whether the pump is operating in real time, and the opening value of the motorized valve in real time. At this time, the pressure measured at the measurement point to predict the pressure is excluded from the input data. That is, input data is generated in the same form as the learning data.

The model use step is to predict the pressure at any one measurement point among a plurality of measurement points included in the monitoring section by inputting the input data generated in the input data generation step to the learned pressure prediction model.

The input data generation step and the model use step may be performed in parallel for each of a plurality of pressure prediction models, since a pressure prediction model is generated for each measurement point to which the pressure gauge included in the monitoring section is connected. The pressure prediction step (S50) is repeatedly performed in real time, and when the learned pressure prediction model is updated according to a predetermined cycle, it is performed according to the updated pressure prediction model and monitoring section. Since the operation performed by the accident detection unit 140 in the model use step has been described above, it will be omitted.

The accident detection step (S60) may be performed by the accident detection unit 140 of the accident detection server 100. In the accident detection step (S60), it may be determined that an accident has occurred when the predicted pressure is significantly greater than the received pressure by comparing the predicted pressure with the pressure received from the measuring device. If it is determined that an accident has occurred by comparing the pressure, the accident detection step (S60) calculates the balance between the amount of water flowing into the monitoring section and the amount of water flowing out of the monitoring section, and the amount of water flowing in is within a predetermined range greater than the amount of water flowing out. In many cases above, it is possible to more accurately determine whether an accident has occurred by further performing a process of determining that an accident has occurred. The accident detection step (S60) can be performed immediately after the pressure prediction step (S50) is performed to predict the pressure. Since the operation performed by the accident detection unit 140 in the accident detection step (S60) has been described above, it will be omitted.

The accident notification step (S70) may be performed in the accident notification unit 150 of the accident detection server 100. In the accident notification step (S70), when it is determined that an accident has occurred in the accident detection unit 140, the location of the accident location and the location of the accident are provided to the user terminal 30. In the accident notification step (S70), the user provides the fact that the accident occurred through the input/output unit 170 or the user terminal 30 and the location of the pipeline 10 estimated as the accident point (AP) determined based on the GIS view map. Therefore, rapid response to accidents is possible. Since the operation performed by the accident notification unit 150 in the accident notification step (S70) has been described above, it will be omitted.

9 is a graph showing a comparison between predicted pressure and measured pressure at a measurement point to which a fifth pressure gauge P5 is connected according to an embodiment of the present invention. The enlarged view of FIG. 9 shows an enlarged error between the predicted pressure and the measured pressure.

Referring to FIG. 9 , it can be seen that the predicted pressure and the measured pressure are almost the same for the measurement point to which the fifth pressure gauge P5 is connected until the time of the accident. And, at the time of the accident (2020-01-24 15:17), an error occurred between the predicted pressure (P5 (prediction)) and the measured pressure (P5 (measurement)) at the measurement point where the fifth pressure gauge (P5) was connected. can know that At the time of the accident, the predicted pressure (P5 (predicted)) appears larger than the measured pressure (P5 (measured)). This is because the pressure was lowered due to the momentary outflow of water due to the pipe breakage, and although the expected pressure was also lowered, the measured pressure was measured lower than that, confirming that an error occurred.

10 is a graph showing an error between a predicted pressure and a measured pressure at a measurement point to which a fifth pressure gauge P5 is connected according to an embodiment of the present invention.

Referring to FIG. 10 , the error between the predicted pressure and the measured pressure at the measurement point to which the fifth pressure gauge P5 is connected before the accident occurs is measured as a value lower than 1%. That is, it can be confirmed that the predicted pressure and the measured pressure are almost the same. In addition, it can be seen that the error is 42.74% (see FIG. 6) at the time of the accident (2020-01-24 15:17), so it appears as a nearly vertical straight line in the graph of FIG. 10. Therefore, the accident detecting unit 140 may determine that an accident has occurred at a point in time when the error rapidly increases. Sudden errors are further measured even after the point of occurrence of the accident, but these secondary errors and tertiary errors are errors that occur ex post facto due to pipe breakage, which is the cause of the first primary error. Therefore, the accident detecting unit 140 may determine the time when a large error first occurs as the time of occurrence of the accident in the accident detection step (S60).

The present invention has been described in detail through specific examples, and the examples are intended to specifically explain the present invention, the present invention is not limited thereto, and those skilled in the art within the technical spirit of the present invention It will be clear that the modification or improvement is possible by.

All simple modifications or changes of the present invention fall within the scope of the present invention, and the specific protection scope of the present invention will be clarified by the appended claims.

10: conduit
20: instrument
30: user terminal
100: accident detection server
110: data collection unit
120: database
130: model learning unit
140: accident detection unit
150: accident notification unit
160: storage unit
170: communication department
180: input/output unit

Claims (14)

The model learning unit is a sensor network map that is a map that reflects the connection relationship between the water supply network’s water purification plant, reservoir, pipeline, branching of the pipeline, flowmeter and pressure gauge, which are instruments connected to the pipeline, pumps connected to the pipeline, and electric valves connected to the pipeline. Determine the monitoring section based on, but the monitoring section is based on the point where the upstream and downstream are hydraulically disconnected and the point where the flow meter is installed to measure the inflow and outflow flow rate in the section simultaneously, affecting the pressure of the monitoring section A section determination step determined to include a pump or electric valve, a pressure gauge installed in the monitoring section, and a flow meter;
The pressure measured by the pressure gauge at the point where the model learning unit tries to predict the pressure within the monitoring section is label data, and the pressure measured by the remaining pressure gauges included in the monitoring section, the flow rate measured by the flow meter, the operating state of the pump, and the motorized valve A pressure prediction model that generates a learning dataset whose opening value is learning data by the number of pressure gauges included in the monitoring section, and predicts the pressure at the pressure measuring point to which the pressure gauge is connected using the learning dataset is created in the monitoring section. A model learning step of generating pressure prediction models as many as the number of pressure gauges included in the monitoring section by learning each pressure measurement point included in the monitoring section;
A pressure prediction step of predicting the pressure at all pressure measurement points within the monitoring section by inputting the pressure, flow rate, operating state of the pump, and opening value of the electric valve received in real time from the measuring instrument into the pressure prediction model learned by the accident detection unit. ; and
Real-time water supply accident using AI, including an accident detection step of comparing the predicted pressure and the pressure measured in real time by the accident detection unit and determining that an accident has occurred if the predicted pressure is greater than the measured pressure by a predetermined range or more detection method. The method of claim 1,
Before the model learning step, the water supply real-time accident detection method using AI, further comprising a data pre-processing step of correcting an error in the pressure or flow rate received from the meter installed in the water supply network. The method of claim 1,
The accident detection step is
When it is determined that an accident has occurred because the predicted pressure is greater than the measured pressure by more than a predetermined range, the balance between the amount of water flowing into the monitoring section and the amount of water flowing out of the monitoring section is calculated, and the amount of water flowing in is equal to the amount of water flowing out. Water supply real-time accident detection method using AI, which further performs the process of determining that an accident has occurred if there are more than a specified range than the amount. The method of claim 1,
If an accident is detected, the accident notification unit estimates the location where the accident occurred in the monitoring section, selects a measurement point where the error obtained by subtracting the predicted pressure from the measured pressure is a positive value, and selects the measurement point with the largest error and the next Water supply real-time accident detection method using AI, further comprising an accident notification step of determining that a pipe breakage accident has occurred between measurement points with a large error, and informing the user terminal of the accident occurrence and location of the accident based on the GIS pipe map. delete The method of claim 1,
The pressure prediction step is
An input data generation step of generating input data in real time from the flow rate and pressure, the operating state of the pump, and the opening value of the electric valve received from the measuring instrument; and
Water supply real-time accident detection method using AI, including a model use step of predicting the pressure at the measurement point by inputting the input data into the learned pressure prediction model. The method of claim 2,
The above error is
There is an erroneous or missing flow rate or pressure from the meter,
The data preprocessing step is
a number determination step of determining whether the number of errors is greater than a predetermined number;
an interpolation correction step of correcting errors using an interpolation method when the number of errors is less than the predetermined number; and
If the number of errors is greater than the predetermined number, a method for detecting real-time water supply accidents using AI, including a periodic correction step of correcting errors using data at a specific point in time. delete In the water supply real-time accident detection system,
The water supply real-time accident detection system includes an accident detection server,
The accident detection server,
A model learning unit for learning a pressure prediction model for detecting whether an accident occurs in the waterworks network; and
An accident detection unit for predicting pressure using the pressure prediction model,
The model learning unit
Based on the sensor network map, which is a map that reflects the connection relationship between the water supply network's water purification plant, reservoir, conduit, branching of the conduit, flowmeter and pressure gauge, which are instruments connected to the conduit, pumps connected to the conduit, and electric valves connected to the conduit The monitoring section is determined, but the monitoring section is based on the point where the upstream and downstream are hydraulically disconnected and the point at which the flow meter is installed to measure the inflow and outflow flow rate in the section simultaneously, and the pump or electric motor that affects the pressure in the monitoring section It is determined to include a valve, a pressure gauge installed in the monitoring section, and a flow meter,
The pressure measured by the pressure gauge at the point where the pressure is to be predicted within the monitoring section is the label data, and the pressure measured by the remaining pressure gauges included in the monitoring section, the flow rate measured by the flow meter, the operating state of the pump, and the opening value of the motorized valve. A learning data set, which is the learning data, is created as many as the number of pressure gauges included in the monitoring section, and a pressure prediction model for predicting the pressure at the pressure measuring point to which the pressure gauge is connected using the learning data set is included in the monitoring section. Learning each pressure at every pressure measurement point to generate as many pressure prediction models as the number of pressure gauges included in the monitoring section,
The accident detection department
The pressure, flow rate, operating state of the pump, and opening value of the motorized valve received in real time from the measuring instrument are input into the learned pressure prediction model to predict the pressure at all pressure measurement points within the monitoring section, respectively, and to compare the predicted pressure and real-time Water supply real-time accident detection system using AI, which compares the measured pressure and determines that an accident has occurred if the predicted pressure is greater than the measured pressure by a predetermined range or more. The method of claim 9,
Water supply real-time accident detection system using AI, further comprising a data collection unit for correcting errors in pressure or flow rate received from the meter installed in the water supply network. The method of claim 9,
If an accident is detected, estimate the location where the accident occurred in the monitoring section, but select a measurement point where the error obtained by subtracting the predicted pressure from the measured pressure is a positive value, and the measurement point with the largest error and the next largest error are selected. Water supply real-time accident detection system using AI, further comprising an accident notification unit that determines that a pipe breakage accident has occurred between the measurement points and informs the user terminal of the accident occurrence and location of the accident based on the GIS pipe map. delete The method of claim 9,
The accident detection department
When it is determined that an accident has occurred because the predicted pressure is greater than the measured pressure by more than a predetermined range, the balance between the amount of water flowing into the monitoring section and the amount of water flowing out of the monitoring section is calculated, and the amount of water flowing in is equal to the amount of water flowing out. A water supply real-time accident detection system using AI that further performs the process of determining that an accident has occurred if there are more than a specified range than the amount. The method of claim 10,
The above error is
There is an erroneous or missing flow rate or pressure from the meter,
The data collection department
It is determined whether the number of errors is greater than the predetermined number, and if the number of errors is less than the predetermined number, the error is corrected using an interpolation method. If the number of errors is greater than the predetermined number, data at a specific time point is used to A real-time accident detection system for water supply using AI that corrects errors.

Images