CN112004203B - Indoor positioning method and device based on position prediction and error compensation - Google Patents
Indoor positioning method and device based on position prediction and error compensation Download PDFInfo
- Publication number
- CN112004203B CN112004203B CN202010961908.4A CN202010961908A CN112004203B CN 112004203 B CN112004203 B CN 112004203B CN 202010961908 A CN202010961908 A CN 202010961908A CN 112004203 B CN112004203 B CN 112004203B
- Authority
- CN
- China
- Prior art keywords
- indoor
- information
- fingerprint
- predicted
- fingerprint data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012937 correction Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims description 33
- 230000006870 function Effects 0.000 claims description 24
- 239000013598 vector Substances 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 6
- 238000012423 maintenance Methods 0.000 abstract description 6
- 238000001914 filtration Methods 0.000 description 10
- 230000008859 change Effects 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 239000002245 particle Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012887 quadratic function Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S11/00—Systems for determining distance or velocity not using reflection or reradiation
- G01S11/02—Systems for determining distance or velocity not using reflection or reradiation using radio waves
- G01S11/06—Systems for determining distance or velocity not using reflection or reradiation using radio waves using intensity measurements
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention discloses an indoor positioning method and device based on position prediction and error compensation, comprising the steps of obtaining signal intensity information of a position to be predicted; taking the signal intensity information as input information, and carrying out position prediction on a position to be predicted by combining a position prediction model to obtain first position information; taking the first position information as input information, and combining an error correction model to perform position prediction on a position to be predicted to obtain final position prediction information; the invention carries out the preliminary positioning of the user through the fingerprint library and the position estimation model before the indoor environment changes, and then establishes the error correction model through the new fingerprint library to carry out error correction on the preliminary positioning information, thereby achieving the accurate positioning of the indoor position of the user, avoiding the need of re-collecting a large amount of fingerprint information to re-establish the fingerprint library and re-establishing the position estimation model, reducing the data quantity of the re-collected fingerprint information, and further reducing the maintenance cost of the positioning device after the indoor environment changes.
Description
Technical Field
The invention belongs to the technical field of indoor positioning, and particularly relates to an indoor positioning method and device based on position prediction and error compensation.
Background
The indoor positioning technology is Location Based Services (LBS), which is a core support technology for important business applications. The fingerprint-based indoor positioning method is concerned by solving the problem of positioning accuracy reduction under the Non Line of Sight (NLOS) channel condition.
The positioning accuracy of the fingerprint positioning method is greatly influenced by a position fingerprint database (Radio Map). When the building structure, the room layout and the like of the indoor environment change, the indoor wireless propagation environment also changes, so that a large difference exists between the indoor wireless propagation environment and the established position fingerprint database, and a large positioning error is generated.
At this point, the RSS sample needs to be collected again to reconstruct or update the fingerprint library. However, rebuilding or updating a fingerprint library is a very time-consuming and laborious task, which increases the maintenance costs of the fingerprint library to a large extent.
Disclosure of Invention
The invention aims to provide an indoor positioning method and device based on position prediction and error compensation, which only needs to re-collect a small amount of data and does not reduce positioning accuracy when an indoor environment changes.
The invention adopts the following technical scheme: an indoor positioning method based on position prediction and error compensation comprises the following steps:
acquiring signal intensity information of a position to be predicted;
taking the signal intensity information as input information, and carrying out position prediction on a position to be predicted by combining a position prediction model to obtain first position information; the position prediction model is obtained by training a first fingerprint database before the indoor environment changes;
taking the first position information as input information, and combining an error correction model to perform position prediction on a position to be predicted to obtain second position information; the error correction model is obtained by training a second fingerprint library which is acquired again after the indoor environment changes, and the second position information is the final position prediction information of the position to be predicted.
Further, the acquiring the signal strength information of the position to be predicted comprises:
repeatedly acquiring a plurality of signal strength values of the position to be predicted relative to each indoor AP;
calculating the average value of a plurality of signal strength values of the position to be predicted relative to each indoor AP;
combining all the means into a signal strength vector; wherein the signal strength vector is signal strength information.
Further, the training method of the position prediction model comprises the following steps:
constructing a first fingerprint database; the first fingerprint database comprises a plurality of groups of first fingerprint data sets, and each group of first fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information;
dividing the first fingerprint data set into a first fingerprint data subset and a second fingerprint data subset; the first fingerprint data subset comprises an abscissa of an indoor physical position coordinate and signal intensity information corresponding to the abscissa, and the second fingerprint data subset comprises an ordinate of the indoor physical position coordinate and signal intensity information corresponding to the ordinate;
respectively predicting the abscissa and the ordinate by adopting an initial objective function of an XGboost algorithm to obtain a predicted abscissa and a predicted ordinate;
performing iterative optimization on an initial objective function by combining indoor physical position coordinates in each group of first fingerprint data sets and corresponding prediction abscissa and prediction ordinate through a greedy algorithm to obtain an optimal objective function; wherein, the optimal objective function is a position prediction model.
Further, the optimal objective function is:
wherein T is the leaf node ordinal number of the kth tree in the XGboost algorithm, T is the total number of the leaf nodes of the kth tree, and Gt、HtBoth are constant values, and γ and λ are hyper-parameters that adjust the complexity of each tree.
Further, constructing the first fingerprint library includes:
dividing an indoor area before the indoor environment changes into a plurality of grids with equal size;
acquiring indoor physical position coordinates and corresponding signal intensity information thereof in each grid;
and combining the collected indoor physical position coordinates and the corresponding signal intensity information to form a first fingerprint database.
Further, the training method of the error correction model comprises the following steps:
constructing a second fingerprint library; the second fingerprint database comprises a plurality of groups of indoor physical position coordinates and corresponding position prediction coordinates;
and training the second fingerprint library by adopting an elastic net algorithm to construct an error correction model.
Further, constructing the second fingerprint library comprises:
acquiring a plurality of groups of second fingerprint data sets; each group of second fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information, and the number of the second fingerprint data sets is smaller than that of the first fingerprint data sets;
predicting the indoor physical position coordinates of each group of second fingerprint data sets based on the position prediction model to obtain predicted indoor position coordinates corresponding to each group of second fingerprint data sets;
generating an error value of the indoor physical position coordinate according to the indoor physical position coordinate and the predicted indoor position coordinate;
and combining the indoor physical position coordinates in each group of second fingerprint data sets and the corresponding error values to obtain a second fingerprint library.
Further, acquiring a plurality of sets of second fingerprint data comprises:
dividing an indoor area with changed indoor environment into a plurality of grids with equal size;
acquiring indoor physical position coordinates and corresponding signal intensity information thereof in each grid;
and combining the indoor physical position coordinates acquired in all the grids and the corresponding signal intensity information to obtain a plurality of groups of second fingerprint data sets.
The other technical scheme of the invention is as follows: an indoor positioning device based on XGboost position prediction and error compensation comprises:
the acquisition module is used for acquiring the signal intensity information of the position to be predicted;
the position prediction module is used for performing position prediction on a position to be positioned by taking the signal strength information as input information and combining a position prediction model to obtain first position information; the position prediction model is obtained by training through an existing first fingerprint database before the indoor environment changes;
the error correction module is used for performing position prediction on the position to be positioned by taking the first position information as input information and combining the error correction model to obtain second position information; the error correction model is obtained by training a second fingerprint library which is acquired again after the indoor environment changes, and the second position information is the final position prediction information of the position to be predicted.
The other technical scheme of the invention is as follows: an indoor positioning device based on XGboost position prediction and error compensation comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein when the processor executes the computer program, the indoor positioning device based on the XGboost position prediction and error compensation realizes any one of the indoor positioning method based on the XGboost position prediction and error compensation.
The invention has the beneficial effects that: the invention carries out the preliminary positioning of the user through the fingerprint library and the position estimation model before the indoor environment changes, and then establishes the error correction model through the new fingerprint library to carry out error correction on the preliminary positioning information, thereby achieving the accurate positioning of the indoor position of the user, avoiding the need of re-collecting a large amount of fingerprint information to re-establish the fingerprint library and re-establishing the position estimation model, reducing the data quantity of the re-collected fingerprint information, and further reducing the maintenance cost of the positioning device after the indoor environment changes.
Drawings
FIG. 1 is a flow chart of a prior art indoor positioning method combining machine learning and filtering;
FIG. 2 is a flowchart of an indoor positioning method according to an embodiment of the present invention;
FIG. 3 is a graph comparing the CDF performance of the method of the present invention with that of a prior art method in a validation example;
FIG. 4 is a graph comparing CDF performance of the method of the present invention with that of a prior art method when different numbers of APs are contaminated with the same noise in the validation embodiment;
FIG. 5 is a graph comparing CDF performance of the method of the present invention with that of the prior art method when the same number of APs receive different noise pollution in the validation embodiment;
FIG. 6 is a graph comparing the motion trajectories of the method of the present invention and the prior art method in the verification example.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
In the prior art, when the indoor environment changes, RSS samples need to be collected again to reconstruct or update the fingerprint database. However, rebuilding or updating a fingerprint library is a very time-consuming and laborious task, which increases the maintenance costs of the fingerprint library to a large extent.
To overcome this problem, a precise indoor positioning algorithm based on a Gradient Boosting Decision Tree (GBDT) machine learning algorithm combined with filtering has appeared, as shown in fig. 1. The algorithm comprises an off-line stage and an on-line stage, and the implementation steps are described as follows: an off-line stage, 1, dividing an indoor environment into a plurality of grids with equal size; 2. acquiring and recording physical position coordinates and a Received Signal Strength (RSS) value of a WIFI Signal aiming at each grid point, and establishing a position fingerprint database (Radio Map); 3. and training the position fingerprint database data by adopting a GBDT ensemble learning algorithm so as to establish a GBDT positioning model. An online stage: 4. a mobile phone user scans a received WiFi signal strength (RSS) value in the indoor environment; 5. predicting the position of a mobile phone user by using the trained GBDT model, wherein the predicted position can be reduced in positioning precision due to the change of a fingerprint database, and is called as position rough estimation in FIG. 1; 6. in order to improve the positioning accuracy, a filtering algorithm is further used for tracking compensation, such as the particle filtering method in fig. 1, so as to obtain a fine estimation of the position.
The method mainly solves the problems of position fingerprint library change and positioning accuracy improvement through filtering, and the most common filtering methods are Kalman filtering and particle filtering. The linear filter generated by kalman filtering is optimal when the basic data statistics are gaussian, but when the statistics deviate from gaussian, the optimality disappears, and particle filtering causes particle degradation to degrade the positioning accuracy.
The invention provides an indoor positioning method (XGboost-EC) based on XGboost initial positioning and Error Compensation (EC) combined with an Elastic Net (Elastic Net). The method only needs to update fewer fingerprint database samples, can effectively solve the problem of positioning accuracy reduction caused by environment change through an error compensation method,
the method models initial positioning and error compensation as nonlinear machine learning problems respectively, high-precision positioning can be realized by utilizing the parallel processing advantage and good generalization capability of XGboost and the sparse model learning capability and stability advantage of an elastic network, and the defect that the applicable condition of a filter is limited is overcome.
The invention discloses an indoor positioning method based on position prediction and error compensation, which comprises the following steps as shown in figure 2:
acquiring signal intensity information of a position to be predicted; taking the signal intensity information as input information, and carrying out position prediction on a position to be predicted by combining a position prediction model to obtain first position information; the position prediction model is obtained by training a first fingerprint database before the indoor environment changes; taking the first position information as input information, and combining an error correction model to perform position prediction on a position to be predicted to obtain second position information; the error correction model is obtained by training a second fingerprint library which is acquired again after the indoor environment changes, and the second position information is the final position prediction information of the position to be predicted.
The invention carries out the preliminary positioning of the user through the fingerprint library and the position estimation model before the indoor environment changes, and then establishes the error correction model through the new fingerprint library to carry out error correction on the preliminary positioning information, thereby achieving the accurate positioning of the indoor position of the user, avoiding the need of re-collecting a large amount of fingerprint information to re-establish the fingerprint library and re-establishing the position estimation model, reducing the data quantity of the re-collected fingerprint information, and further reducing the maintenance cost of the positioning device after the indoor environment changes.
In an embodiment of the present invention, acquiring the signal strength information of the position to be predicted includes:
repeatedly acquiring a plurality of signal strength values of the position to be predicted relative to each indoor AP; calculating the average value of a plurality of signal strength values of the position to be predicted relative to each indoor AP; combining all the means into a signal strength vector; wherein the signal strength vector is signal strength information.
In practical applications, it is assumed that the method is applied to a mobile phone, and the mobile phone receives RSS values (signal strength values) of a plurality of APs at a certain location, in this embodiment, the RSS values form a row vector, and the RSS value of each AP corresponds to an element in the row vector. Of course, the RSS value of each AP generally has time-varying property, and in order to reduce this randomness, the present embodiment is configured to scan the AP five times in succession, and then take the average value as the RSS value of the AP.
In the embodiment of the present invention, as a specific implementation manner, a training method of a position prediction model includes:
constructing a first fingerprint database; the first fingerprint database comprises a plurality of groups of first fingerprint data sets, and each group of first fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information; dividing the first fingerprint data set into a first fingerprint data subset and a second fingerprint data subset; the first fingerprint data subset comprises an abscissa of an indoor physical position coordinate and signal intensity information corresponding to the abscissa, and the second fingerprint data subset comprises an ordinate of the indoor physical position coordinate and signal intensity information corresponding to the ordinate; respectively predicting the abscissa and the ordinate by adopting an initial objective function of an XGboost algorithm to obtain a predicted abscissa and a predicted ordinate; performing iterative optimization on an initial objective function by combining indoor physical position coordinates in each group of first fingerprint data sets and corresponding prediction abscissa and prediction ordinate through a greedy algorithm to obtain an optimal objective function; wherein, the optimal objective function is a position prediction model.
Specifically, the optimal objective function is:
wherein T is the leaf node ordinal number of the kth tree in the XGboost algorithm, and T is the leaf of the kth treeTotal number of nodes, Gt、HtBoth are constant values, and γ and λ are hyper-parameters that adjust the complexity of each tree.
In the embodiment of the present invention, constructing the first fingerprint database includes:
dividing an indoor area before the indoor environment changes into a plurality of grids with equal size; acquiring indoor physical position coordinates and corresponding signal intensity information thereof in each grid; and combining the collected indoor physical position coordinates and the corresponding signal intensity information to form a first fingerprint database.
Specifically, in this embodiment, the positioning model based on the XGBoost is adopted when the signal strength information is used as the input information and the position prediction model is combined to perform the position prediction on the position to be positioned.
For WIFI fingerprint data set containing N samples
Wherein r isi＝[ri 1，ri 2,…,ri M]For M-dimensional received signal strength vectors, pi＝[xi，yi]Is the physical location coordinate of the ith reference point.
The data set D may be decomposed into D1And D2The two data sets being modelled separately, i.e.
For D1And D2The XGboost algorithm is respectively adopted by the two data sets to carry out the physical position piAbscissa x ofiAnd ordinate yiMaking a prediction to obtain piIs estimated value of
In a data set D2For example, assuming that K trees have been trained, the predicted values for the ith sample are:
in the formula (f)k(ri) For the kth tree pair sample riThe predicted value of (2). The objective function can be modeled as:
in the formula,
denotes the loss function, Ω (f)k) Representing the complexity of the kth tree, N being the number of samples. According to the feature of Additive Training, when Training the kth tree, the first k-1 trees are known, so the optimization problem of equation (2) can be expressed as:
further, the objective function is simplified by a second order taylor series approximation:
wherein,
gi，hiare all constants.
To solve, f needs to be calculatedk(ri) Expressed by parameterization, the value of the t leaf node of the kth tree is thus defined as ωtThe number of all leaf nodes is T, and the sample riThe position of the leaf node is q (r)i) All sample sets of the t-th node of the k-th tree are It＝{ri|q(ri) T }. The complexity Ω of a tree can be represented by two parts:
where γ and λ are hyper-parameters of the tuning complexity. The new objective function parameterized by equation (4) can be expressed as:
further simplification is:
wherein G ist，HtIs a constant value, therefore, equation (7) is a problem of solving the optimal solution for the quadratic function, and when the tree structure is fixed, the optimal objective function value under the current tree structure is:
the best structure of the segmentation point and the tree can be found through a greedy algorithm, so that the XGboost optimal model is trained to obtain
Predict value, for D1Modeling of the data set may result in
To thereby complete the estimation of the physical position coordinates piIs estimated by
When the environment changes, the accuracy of the indoor positioning model based on XGboost is reduced, and in order to improve the positioning accuracy, a polynomial regression error compensation algorithm is provided, and the algorithm can realize great improvement of the positioning accuracy by only updating a small amount of databases.
Specifically, the training method of the error correction model comprises the following steps:
constructing a second fingerprint library; the second fingerprint database comprises a plurality of groups of indoor physical position coordinates and corresponding position prediction coordinates; and training the second fingerprint library by adopting an elastic net algorithm to construct an error correction model. In this embodiment, constructing the second fingerprint library includes:
acquiring a plurality of groups of second fingerprint data sets; each group of second fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information, and the number of the second fingerprint data sets is smaller than that of the first fingerprint data sets; predicting the indoor physical position coordinates of each group of second fingerprint data sets based on the position prediction model to obtain predicted indoor position coordinates corresponding to each group of second fingerprint data sets; generating an error value of the indoor physical position coordinate according to the indoor physical position coordinate and the predicted indoor position coordinate; and combining the indoor physical position coordinates in each group of second fingerprint data sets and the corresponding error values to obtain a second fingerprint library.
More specifically, acquiring the plurality of sets of second fingerprint data sets includes:
dividing an indoor area with changed indoor environment into a plurality of grids with equal size; acquiring indoor physical position coordinates and corresponding signal intensity information thereof in each grid; and combining the indoor physical position coordinates acquired in all the grids and the corresponding signal intensity information to obtain a plurality of groups of second fingerprint data sets.
In this embodiment, the training method of the error correction model is as follows:
sampling a small amount of fingerprint data to construct a new data set for a changing environment
Wherein,
defined as the estimated vector of the physical location coordinates of the ith reference point by the XGBoost algorithm. e.g. of the typei＝[Δxi,Δyi]Is defined asIn the directions of the x axis and the y axis of the ith reference point position, the estimated value and the real error vector are as follows:
decomposing a data set S into S1And S2The two data sets being modelled separately, i.e.
For S1And S2The error values deltax of the two data sets are respectively realized by adopting an elastic network algorithmiAnd Δ yiPredicting to obtain corresponding predicted value
And
with S2For data set example, the objective function can be modeled as:
wherein,
represents Δ yiPredicted value of (2)
Is about
And
the polynomial of (a) is determined,θ＝[θ1,θ2,…θJ]representing the polynomial coefficient, alpha and beta are model hyper-parameters.
Can be expressed as:
by training the elastic net regression model, the model parameter theta can be solved, so as to obtain the predicted value
For the same reason, for data set S1The predicted value can be obtained by adopting the elastic net algorithm
The estimation precision is further improved by error compensation, namely:
at this moment, after preliminary position prediction of XGboost algorithm and Error Compensation (EC) of Elastic Net algorithm, the ith physical position coordinate p in the indoor plane is subjected toiIs estimated as
The method of the invention is different from the original method in that: when the indoor environment changes, the original method is to acquire the fingerprint again to reconstruct the fingerprint database, because the fingerprint database is updated again although the environment changes, the positioning accuracy can be ensured even by adopting the original algorithm. The method has the defects that the re-acquisition of the fingerprint data is a laborious work which is time-consuming and labor-consuming, and the maintenance cost of the fingerprint database is greatly increased.
The method of the invention is to keep the original fingerprint database and the original prediction algorithm, and on the basis, an error compensation fingerprint database is newly established, the error compensation fingerprint database is established by collecting a small amount of fingerprints, and a new elastic network algorithm is introduced for training, so that the trained elastic network model is used for error compensation estimation, and the estimated error value is preferably used for correcting the estimation information of the original algorithm. In a word, when the indoor environment changes, only a small number of fingerprint samples need to be updated, an error compensation fingerprint library is established, the elastic network algorithm is used for training so as to predict errors, the errors are used for correcting the positioning estimation of the original XGboost algorithm, and the positioning accuracy and the positioning robustness are improved.
For varying environments, the original fingerprint library may cause large positioning estimation errors. If the re-acquisition and establishment of a new fingerprint database are time-consuming and labor-consuming heavy work, one idea for avoiding the re-acquisition and establishment of the new fingerprint database is to acquire and update a small amount of fingerprint data on the basis of the original fingerprint database, construct an error compensation database by using the updated fingerprint data, realize error compensation through an error compensation database training algorithm, and then compensate for a large positioning error predicted by the original fingerprint database, so that the positioning precision is improved. Aiming at an error compensation database, considering that modeling can be taken as a nonlinear supervised learning problem, therefore, various machine learning algorithms can be adopted for processing, such as a polynomial regression algorithm, a support vector machine algorithm and the like.
The invention aims to improve the positioning precision by updating a small amount of fingerprints, and for the purpose, a small amount of fingerprints are collected to establish an error fingerprint database, and the error fingerprint database is trained through an elastic network algorithm. Of course, there may be many samples in the error fingerprint library, which is more beneficial for training the elastic net model, but the acquisition is a difficult task, which deviates from the object of the present invention.
In summary, the method of the present invention is different from the original method in that: when the environment changes, the original method is to acquire the fingerprint again to reconstruct the fingerprint database, because the fingerprint database is updated again although the environment changes, the positioning accuracy can be ensured even by adopting the original algorithm. The method of the invention is to keep the original fingerprint database and the original prediction algorithm, on the basis, an error compensation fingerprint database is newly established, the error compensation fingerprint database is established by collecting a small amount of fingerprints, a new elastic network algorithm is introduced for carrying out error compensation estimation, and the estimated error value is further utilized to correct the estimation information of the original algorithm.
Because the collected information is processed sequentially through the position prediction model and the error correction model, the position prediction model has certain precision when used for position prediction, and the position prediction precision is reduced only due to the change of indoor environment, so that the position prediction information only needs to be corrected through the error correction model. Because the data source for error correction has certain precision, the error correction model does not need to have very high precision, and further can collect less data fingerprints for training, thereby avoiding the process of collecting a large amount of data fingerprints for training,
the method can also be divided into an off-line phase and an on-line phase. In the off-line stage, the RSS value and coordinate information are collected at each Reference Point (RP) to establish a fingerprint database. And training and establishing an XGboost positioning model by adopting an XGboost algorithm based on the fingerprint database, and realizing the preliminary prediction of the position by utilizing the XGboost positioning model.
When the environment changes, only a few RSS fingerprint library samples need to be updated, an error fingerprint library is established by using the updated samples, and an error compensation model is established by using an Elastic network (Elastic Net) algorithm based on the error database (the Elastic network adds an L1 paradigm and an L2 paradigm on the basis of linear regression to be used as penalty terms).
And in the online stage, accurate positioning is realized by utilizing the trained XGboost prediction model and an Elastic Net (Elastic Net) error compensation model in the offline stage.
The other technical scheme of the invention is as follows: an indoor positioning device based on XGboost position prediction and error compensation comprises:
the acquisition module is used for acquiring the signal intensity information of the position to be predicted;
the position prediction module is used for performing position prediction on a position to be positioned by taking the signal strength information as input information and combining a position prediction model to obtain first position information; the position prediction model is obtained by training through an existing first fingerprint database before the indoor environment changes;
the error correction module is used for performing position prediction on the position to be positioned by taking the first position information as input information and combining the error correction model to obtain second position information; the error correction model is obtained by training a second fingerprint library which is acquired again after the indoor environment changes, and the second position information is the final position prediction information of the position to be predicted.
The other technical scheme of the invention is as follows: an indoor positioning device based on XGboost position prediction and error compensation comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein when the processor executes the computer program, the indoor positioning device based on the XGboost position prediction and error compensation realizes any one of the indoor positioning method based on the XGboost position prediction and error compensation.
It should be noted that, because the contents of information interaction, execution process, and the like between the modules are based on the same concept as the method embodiment of the present invention, specific functions and technical effects thereof may be referred to specifically in the method embodiment section, and are not described herein again.
It will be clear to those skilled in the art that, for convenience and simplicity of description, the foregoing division of the functional modules is merely illustrated, and in practical applications, the above function distribution may be performed by different functional modules according to needs, that is, the internal structure of the apparatus is divided into different functional modules to perform all or part of the above described functions. Each functional module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated unit may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional modules are only used for distinguishing one functional module from another, and are not used for limiting the protection scope of the application. The specific working process of the modules in the apparatus may refer to the corresponding process in the foregoing method embodiment, and is not described herein again.
Verification of the examples:
selecting a parking lot under a certain ground in an experimental environment, considering a rectangular area with the length of 40m (y axis) and the width of 5.5m (x axis), placing 10 wireless Access Points (AP) with the height of 2 meters, wherein the plane coordinates (x, y) of the wireless Access points are respectively (0,0), (5.5,0), (0,10), (5.5,10), (0,20), (5.5,20), (0,30), (5.5,30), (0,40) and (5.5, 40); dividing the rectangular test area into 880 meshes of 0.5m × 0.5m, setting the vertex of each mesh as a Reference Point (RP), and obtaining 960 RP points; the mobile phone APP software developed in the embodiment is adopted to test and record the received signal strength RSS value of each AP of each RP point for constructing an off-line position fingerprint database (Radio Map).
The positioning accuracy is an important index for evaluating the positioning algorithm, based on the above environmental position fingerprint database, fig. 3 compares the performance of the estimation error Cumulative Distribution Function (CDF) including the method of the embodiment of the present invention and other positioning algorithms based on different machine learning, and for further more intuitive description of the experimental result, statistical results (average positioning accuracy and positioning accuracy at 80% percentile) based on the experimental data in the figure are shown in table 1.
As can be seen from fig. 3 and table 1: the XGboost-EC provided by the invention adopts an elastic network to carry out error compensation on positioning precision on the basis of the XGboost algorithm, so that the positioning precision is further improved, and the performance of the XGboost-EC is superior to that of the positioning algorithm based on the XGboost.
TABLE 1 precision comparison of several positioning algorithms
An important application scenario of the algorithm provided by the invention is that after the environment changes, only a small amount of off-line position fingerprint databases need to be updated, and the loss of the positioning precision can be effectively compensated.
Fig. 4 shows CDF performance comparison of positioning using the XGBoost algorithm and the proposed XGBoost-EC algorithm when RSS values of 2 and 8 APs are contaminated with noise having a standard deviation of 2, respectively, which shows that the positioning performance of 8 APs contaminated with noise is reduced compared to 2 APs contaminated with noise, and in each case, the proposed algorithm is superior to the XGBoost positioning algorithm.
Fig. 5 is a comparison of the performance of the positioning accuracy CDF when the standard deviation of noise is 1, 3, 5, and 7 respectively in the environment of 4 APs polluted by noise, and it is shown that the positioning accuracy of the proposed algorithm is affected by the standard deviation of noise, and the performance of the positioning accuracy CDF decreases as the standard deviation increases.
Further, a section of motion track is selected, and a real track, a positioning track adopting an XGboost algorithm and positioning error comparison of the XGboost-EC positioning track are simulated in an experiment in fig. 6, so that the whole XGboost-EC positioning track is closer to the real track than XGboost positioning estimation, and the positioning accuracy is higher.
Claims (9)
1. An indoor positioning method based on position prediction and error compensation is characterized by comprising the following steps:
acquiring signal intensity information of a position to be predicted;
taking the signal intensity information as input information, and combining a position prediction model to carry out position prediction on the position to be predicted to obtain first position information; the position prediction model is obtained by training through an existing first fingerprint database before the indoor environment changes;
taking the first position information as input information, and combining an error correction model to predict the position to be predicted to obtain second position information; the error correction model is obtained by training a second fingerprint library which is acquired again after the indoor environment changes, and the second position information is the final position prediction information of the position to be predicted;
constructing the second fingerprint library comprises:
acquiring a plurality of groups of second fingerprint data sets; each group of second fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information, and the number of the second fingerprint data sets is smaller than that of the first fingerprint data sets;
predicting the indoor physical position coordinates of each group of second fingerprint data sets based on the position prediction model to obtain predicted indoor position coordinates corresponding to each group of second fingerprint data sets;
generating an error value of the indoor physical position coordinate according to the indoor physical position coordinate and the predicted indoor position coordinate;
and combining the indoor physical position coordinates in each group of second fingerprint data sets and the corresponding error values to obtain a second fingerprint library.
2. The method as claimed in claim 1, wherein the obtaining the signal strength information of the position to be predicted comprises:
repeatedly acquiring a plurality of signal strength values of the position to be predicted relative to each indoor AP;
calculating the average value of a plurality of signal strength values of the position to be predicted relative to each AP in a room;
combining all of the means into a signal strength vector; wherein the signal strength vector is the signal strength information.
3. The indoor positioning method based on location prediction and error compensation as claimed in claim 2, wherein the training method of the location prediction model is:
constructing the first fingerprint library; the first fingerprint database comprises a plurality of groups of first fingerprint data sets, and each group of first fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information;
dividing the first fingerprint data set into a first fingerprint data subset and a second fingerprint data subset; wherein the first fingerprint data subset comprises an abscissa of the indoor physical location coordinate and the signal strength information corresponding thereto, and the second fingerprint data subset comprises an ordinate of the indoor physical location coordinate and the signal strength information corresponding thereto;
predicting the abscissa and the ordinate by adopting an initial objective function of an XGboost algorithm to obtain a predicted abscissa and a predicted ordinate;
iteratively optimizing the initial objective function by combining the indoor physical position coordinates in each first fingerprint data set and the corresponding predicted abscissa and predicted ordinate through a greedy algorithm to obtain an optimal objective function; wherein the optimal objective function is a location prediction model.
4. The method as claimed in claim 3, wherein the optimal objective function is:
wherein T is the leaf node ordinal number of the kth tree in the XGboost algorithm, T is the total number of the leaf nodes of the kth tree, and Gt、HtBoth are constant values, and γ and λ are hyper-parameters that adjust the complexity of each tree.
5. The method as claimed in claim 3 or 4, wherein the step of constructing the first fingerprint database comprises:
dividing an indoor area before the indoor environment changes into a plurality of grids with equal size;
acquiring indoor physical position coordinates and corresponding signal intensity information thereof in each grid;
and combining the collected indoor physical position coordinates and the corresponding signal intensity information to form a first fingerprint database.
6. The method as claimed in claim 3, wherein the training method of the error correction model comprises:
constructing the second fingerprint library; the second fingerprint database comprises a plurality of groups of indoor physical position coordinates and corresponding position prediction coordinates;
and training the second fingerprint library by adopting an elastic net algorithm to construct an error correction model.
7. The method as claimed in claim 6, wherein said obtaining multiple sets of second fingerprint data sets comprises:
dividing an indoor area with changed indoor environment into a plurality of grids with equal size;
acquiring indoor physical position coordinates and corresponding signal intensity information thereof in each grid;
and combining the indoor physical position coordinates acquired in all the grids and the corresponding signal intensity information to obtain a plurality of groups of second fingerprint data sets.
8. An indoor positioning device based on position prediction and error compensation, comprising:
the acquisition module is used for acquiring the signal intensity information of the position to be predicted;
the position prediction module is used for performing position prediction on the position to be predicted by taking the signal strength information as input information and combining a position prediction model to obtain first position information; the position prediction model is obtained by training through a first fingerprint database before the indoor environment changes;
the error correction module is used for performing position prediction on the position to be predicted by taking the first position information as input information and combining an error correction model to obtain second position information; the error correction model is obtained by training a second fingerprint library which is acquired again after the indoor environment changes, and the second position information is the final position prediction information of the position to be predicted;
constructing the second fingerprint library comprises:
acquiring a plurality of groups of second fingerprint data sets; each group of second fingerprint data sets comprises indoor physical position coordinates and corresponding signal intensity information, and the number of the second fingerprint data sets is smaller than that of the first fingerprint data sets;
predicting the indoor physical position coordinates of each group of second fingerprint data sets based on the position prediction model to obtain predicted indoor position coordinates corresponding to each group of second fingerprint data sets;
generating an error value of the indoor physical position coordinate according to the indoor physical position coordinate and the predicted indoor position coordinate;
and combining the indoor physical position coordinates in each group of second fingerprint data sets and the corresponding error values to obtain a second fingerprint library.
9. Indoor positioning apparatus based on location prediction and error compensation, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor when executing the computer program implements an indoor positioning method based on location prediction and error compensation according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010961908.4A CN112004203B (en) | 2020-09-14 | 2020-09-14 | Indoor positioning method and device based on position prediction and error compensation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010961908.4A CN112004203B (en) | 2020-09-14 | 2020-09-14 | Indoor positioning method and device based on position prediction and error compensation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112004203A CN112004203A (en) | 2020-11-27 |
| CN112004203B true CN112004203B (en) | 2022-04-12 |
Family
ID=73469259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010961908.4A Active CN112004203B (en) | 2020-09-14 | 2020-09-14 | Indoor positioning method and device based on position prediction and error compensation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112004203B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113596985B (en) * | 2021-07-07 | 2022-05-20 | 湘潭大学 | Signal intensity positioning device and method based on path loss correction model |
| CN113723240B (en) * | 2021-08-19 | 2023-08-22 | 武汉大学 | General fingerprint positioning method and system based on Boosting and sample difference |
| CN114286306B (en) * | 2021-12-27 | 2024-02-20 | 南京中孚信息技术有限公司 | Indoor signal positioning method, device, computer equipment and storage medium |
| CN114554395A (en) * | 2022-03-03 | 2022-05-27 | 上海维智卓新信息科技有限公司 | Positioning accuracy prediction method and device, and signal source layout determination method and device |
| CN116193574B (en) * | 2023-02-22 | 2023-10-13 | 中电建建筑集团有限公司 | 5g network-based observation information fusion positioning key technical method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106162868A (en) * | 2016-06-08 | 2016-11-23 | 南京理工大学 | High efficiency indoor localization method based on location fingerprint |
| CN109121081A (en) * | 2018-09-11 | 2019-01-01 | 电子科技大学 | A kind of indoor orientation method based on position Candidate Set Yu EM algorithm |
| CN109951798A (en) * | 2019-03-13 | 2019-06-28 | 南京邮电大学 | Merge the enhancing location fingerprint indoor orientation method of Wi-Fi and bluetooth |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107426687B (en) * | 2017-04-28 | 2019-08-09 | 重庆邮电大学 | Towards the method for adaptive kalman filtering for merging positioning in the room WiFi/PDR |
| US10257658B1 (en) * | 2017-11-30 | 2019-04-09 | Mapsted Corp. | Self-learning localization data repository |
| CN111277946A (en) * | 2018-12-04 | 2020-06-12 | 重庆邮电大学 | Fingerprint database self-adaptive updating method in Bluetooth indoor positioning system |
| CN111356082B (en) * | 2020-03-10 | 2021-06-08 | 西安电子科技大学 | Indoor mobile terminal positioning method based on WIFI and visible light communication |
-
2020
- 2020-09-14 CN CN202010961908.4A patent/CN112004203B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106162868A (en) * | 2016-06-08 | 2016-11-23 | 南京理工大学 | High efficiency indoor localization method based on location fingerprint |
| CN109121081A (en) * | 2018-09-11 | 2019-01-01 | 电子科技大学 | A kind of indoor orientation method based on position Candidate Set Yu EM algorithm |
| CN109951798A (en) * | 2019-03-13 | 2019-06-28 | 南京邮电大学 | Merge the enhancing location fingerprint indoor orientation method of Wi-Fi and bluetooth |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112004203A (en) | 2020-11-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112004203B (en) | Indoor positioning method and device based on position prediction and error compensation | |
| US11714943B2 (en) | Parallel analog circuit optimization method based on genetic algorithm and machine learning | |
| CN111540193B (en) | Traffic data restoration method for generating countermeasure network based on graph convolution time sequence | |
| Bonnet et al. | Necessary optimality conditions for optimal control problems in Wasserstein spaces | |
| CN110688393A (en) | Query statement optimization method and device, computer equipment and storage medium | |
| Yeong et al. | Particle filters with nudging in multiscale chaotic systems: With application to the Lorenz’96 atmospheric model | |
| CN114037140A (en) | Prediction model training method, prediction model training device, prediction model data prediction method, prediction model data prediction device, prediction model data prediction equipment and storage medium | |
| CN110889079A (en) | Air quality forecasting system based on dynamic inversion emission source data | |
| CN113361774A (en) | Ecological space intensive utilization optimization method and system | |
| Ge et al. | Reservoir-based surrogate modeling of dynamic user equilibrium | |
| Carlone et al. | A first-order solution to simultaneous localization and mapping with graphical models | |
| CN115169426B (en) | Anomaly detection method and system based on similarity learning fusion model | |
| CN113723006B (en) | LS-SVM (least squares-support vector machine) -based single-station earth change magnetic field modeling prediction method and system | |
| CN113112075B (en) | Internal combustion engine noise prediction method based on VMD and NARX | |
| CN114970698A (en) | Metering equipment operation performance prediction method based on improved LWPLS | |
| CN114676586A (en) | Construction method based on multidimensional, multi-space-time digital simulation and emulation | |
| Ge et al. | Reservoir-based surrogate modeling of dynamic user equilibrium | |
| CN114070438A (en) | Detection method and system for 5G base station electromagnetic signals | |
| CN113886451A (en) | Multi-view-integrated POI recommendation method based on self-attention mechanism | |
| Kim et al. | A clustering algorithm for path planning of SMT inspection machines | |
| CN118296095B (en) | Method and system for determining regional boundary in homeland space planning | |
| CN117521580B (en) | Circuit optimization method and device based on generalization model and storage medium | |
| Ruszczynski et al. | On the regularized decomposition method for two stage stochastic linear problems | |
| CN118586341B (en) | Automatic design method for analog integrated circuit size based on improved size pruning | |
| CN115018477B (en) | Big data analysis method and equipment based on enterprise OA system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Kang Xiaofei Inventor after: Yao Meng Inventor after: Liu Zihui Inventor after: Zeng Xuan Inventor after: Qiao Wei Inventor before: Kang Xiaofei Inventor before: Zeng Xuan Inventor before: Yao Meng Inventor before: Liu Zihui Inventor before: Qiao Wei |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |