CN114679683A - Indoor intelligent positioning method based on derivative fingerprint migration - Google Patents
Indoor intelligent positioning method based on derivative fingerprint migration Download PDFInfo
- Publication number
- CN114679683A CN114679683A CN202210330811.2A CN202210330811A CN114679683A CN 114679683 A CN114679683 A CN 114679683A CN 202210330811 A CN202210330811 A CN 202210330811A CN 114679683 A CN114679683 A CN 114679683A
- Authority
- CN
- China
- Prior art keywords
- reference point
- fingerprint
- point
- points
- value
- 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.)
- Pending
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/025—Services making use of location information using location based information parameters
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
-
- 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
-
- 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
- H04W64/006—Locating users or terminals or network equipment for network management purposes, e.g. mobility management with additional information processing, e.g. for direction or speed determination
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention discloses an indoor intelligent positioning method based on derivative fingerprint migration, which comprises the steps that firstly, 5G base stations collect RSS, incidence angles and position information of different indoor reference points, an edge server performs three types of fingerprint derivation by combining the RSS and the incidence angle information, and an offline derivative fingerprint library is constructed by combining the reference point position information and an optimal K value; and finally, indoor positioning is carried out through an optimal K value online selection method of the point to be measured and a KNN model migration mechanism considering environment time variation, and a better positioning result is sent to the device to be measured under the condition of low off-line data re-acquisition. The method improves the defects of common fingerprint characteristics by adopting derived fingerprints, determines the K value selection of the KNN positioning model by an online optimal selection method of the K value of the point to be measured so as to improve the positioning accuracy of the model, and effectively improves the indoor positioning accuracy of time-varying scenes by adopting a transfer learning positioning algorithm under the condition of low data acquisition cost, thereby having very wide application prospect.
Description
Technical Field
The invention belongs to the technical field of indoor positioning.
Background
With the development of location service industries such as intelligent sorting and disaster rescue, high-precision positioning becomes a popular research direction of wireless positioning technology. According to recent survey statistics, people are located indoors for 80% -90% of the time, so that indoor positioning research is increasingly focused. Simultaneously along with the continuous expansion of enterprise's scale, hardly prevent staff's safety, so the demand carries out accurate indoor location to personnel in specific region. The above results show that the accurate positioning of the indoor environment becomes a positioning research hotspot, and the research results not only bring great economic benefits, but also provide support for technical application in other fields.
The existing indoor positioning methods include technologies such as a proximity detection method, a centroid positioning method, a multilateral positioning method, a fingerprint positioning method, a dead reckoning method and the like, but the methods have more or less defects, for example, although the proximity detection method is simple and easy to implement, only approximate positioning information can be provided, and intensive base station deployment is needed for realizing high-precision positioning by the centroid positioning method, which causes huge economic cost. Comprehensively, indoor positioning generally adopts a multilateral positioning method and a fingerprint positioning method, and both positioning methods can obtain good positioning accuracy. The principle of the multilateration method is that the distance is calculated by the received signal strength and the attenuation model of the signal, and a linear equation set is established according to the distance to solve the positioning position. Indoor positioning based on a fingerprint positioning method is a matching or mapping method of an offline fingerprint library. For example, according to RSS, a classification algorithm is used for matching reference points in an offline fingerprint library, and then the positions of the reference points are weighted to obtain a positioning position; or using regression algorithm to construct mapping relation according to RSS to obtain the positioning position. Multilateration is affected by complex indoor environments and base station deployment locations, and the positioning accuracy is reduced by using a type of attenuation model to represent the RSS-distance relationship. In addition, fingerprint location using RSS data also results in poor location accuracy due to RSS signal fluctuations.
Disclosure of Invention
The purpose of the invention is as follows: in order to solve the problems in the prior art, the invention provides an indoor intelligent positioning method based on derivative fingerprint migration.
The technical scheme is as follows: the invention provides an indoor intelligent positioning method based on derivative fingerprint migration, which specifically comprises the following steps:
step 1: in an off-line stage, mobile equipment is adopted to communicate with a 5G base station at different indoor positions, the different positions are used as reference points, RSS information received by the mobile equipment at each reference point in a time sequence, an incident angle of a signal received by the mobile equipment at each reference point and position information of each reference point are obtained; the total number of the moments in the period of time sequence is U, the total number of the 5G base stations is M, the total number of the reference points is E0, and the set of the reference points is R;
and 2, step: calculating for each reference point derived fingerprint features including a cumulative mean fingerprint feature F1(RSS_θr) Differential fingerprint feature F2(RSS_θr) And hyperbolic fingerprint feature F3(RSS_θr):RSS_θrA joint fingerprint representing a reference point r based on the RSS and the incident angle information; calculating the optimal K value of each reference point according to the three fingerprint characteristics of each reference point; establishing an offline derivative fingerprint database, wherein the offline derivative fingerprint database comprises derivative fingerprint characteristics of each reference point, an optimal K value, RSS information, signal incidence angle and position information, and r is 1, 2.
And step 3: judging whether the time variation of the actual indoor scene exceeds the preset days, if so, turning to the step 4, and otherwise, calculating the position information of the points to be measured in the target domain according to the optimal K value of the reference points in the offline derivative fingerprint library; the target domain comprises a known point and a point to be measured, wherein the known point is RSS information in an actual indoor scene, and both the incident angle and the position information of a signal are known;
and 4, step 4: and migrating the off-line derivative fingerprint library by adopting migration learning so as to construct a radio map, and determining the position information of the point to be measured according to a new reference point in the radio map and the optimal K value of the reference point in the off-line derivative fingerprint library before migration.
Further, the step 2 is as follows: the calculation of the cumulative average fingerprint feature of the reference point r is specifically:
the calculation formula for setting the cumulative average is as follows:
wherein the content of the first and second substances,represents the cumulative average of the data from time 1 to time U, l (U +1) represents the actual value of the data at time U +1, U is 1,2, …, U;
calculating the cumulative average fingerprint characteristic of the reference point r based on the calculation formula of the cumulative average:
wherein the content of the first and second substances,the accumulated average value of the derivative fingerprints of the Mth 5G base station to the reference point r at the U-1 th moment is represented; RSS _ thetarThe expression of (a) is:
wherein M represents the mth 5G base station, and M is 1,2, …, M;representing RSS data collected by the mth 5G base station at time u with respect to reference point r,representing signal incidence angle data collected by the mth 5G base station to the reference point r at the moment u;
differential fingerprint of reference point rCharacteristic F2(RSS_θr) Is expressed as:
F2(RSS_θr)=[Δψr(1),…,Δψr(u),…,Δψr(U)]
wherein, Delta psirThe expression of (u) is:
wherein m is1Denotes the m-th1Base station, m2Denotes the m-th2Base station, m1=1,2,...,M,m21,2, M, and M1≠m2;The expression of (c) is as follows:
hyperbolic fingerprint feature F of reference point r3(RSS_θr) The expression of (a) is:
F3(RSS_θr)=[lr(1),…,lr(u),…,lr(U)]
wherein lrThe expression of (u) is:
further, the step 2 of calculating the optimal K value of each reference point specifically includes: calculating any two reference points r1And r2Has a European style distance of fingerprint between
Wherein r is1=1,2,...,E0,r21,2, E0, and r1≠r2,||·||2Expressing a two-norm formula;representing a weight coefficient of fingerprint features corresponding to the e value;
the reference point r is calculated from the following objective function1Optimum K value of (c):
wherein the content of the first and second substances,andrepresents a reference point r1The actual position information of the mobile station (ms),andrespectively represent reference points r1The abscissa and the ordinate of the predicted position of (a),to representReference point r1The value of (a) of (b),represents a reference point r1K value ofAt the reference point r1Selected for KNN positioningA set of reference points, solving the objective function to obtain a reference point r1Optimum K value of Representing reference points r based on optimal K values1Selected for KNN positioningA set of reference points;
wherein the content of the first and second substances,representation collectionThe 1 st reference point in the list of (c),representation collectionTo middleThe number of the reference points is,represents a reference point r1And collectionsReference point r'1The euclidean distance of the fingerprint of (c),representation collectionReference point r'1Actual location information of.
Further, the step 3 of calculating the position information of the point to be measured in the target domain specifically includes:
using a nearest neighbor algorithm for reference point matching:
wherein EDr,*Representing the Euclidean distance of the fingerprint of the point to be measured from the reference point r, Fe(RSS_θ*) Representing fingerprint characteristics of points to be measured corresponding to the values of e;
selecting the optimal K value corresponding to the reference point with the closest Euclidean distance as the optimal K value of the point to be measuredSorting the Euclidean distances of the fingerprints of each reference point and the point to be measured from small to large, and taking the firstReference point composition set corresponding to European style distance of individual fingerprintCombination of Chinese herbsThereby calculating the position (x) of the point to be measured*,y*):
Wherein the content of the first and second substances,representation collectionIs the first of the reference points in (a),representation collectionTo middleThe number of the reference points is,representing points to be measured and the setReference point r*Of the Euclidean fingerprint distance therebetween, saidRepresentation collectionReference point r*Actual location information of.
Further, the step 4 specifically includes: taking the offline derivative fingerprint library as a source field, and establishing a transfer learning objective function:
and (3) carrying out Lagrange multiplier method solving on the target function of the transfer learning to obtain:
wherein, χ1Hexix-2Are all parameters, λ is Lagrangian multiplier, V isCharacteristic vector of Representing the source domain DsAnd a target domain DtThe inter-domain inter-class divergence matrix of (c),representing the source domain DsAnd a target domain DtThe inter-domain intra-class divergence matrix of (c),source domain DsThe inter-class divergence matrix within the domain,source domain DsAn intra-domain intra-class divergence matrix;andthe expression of (a) is:
wherein, wo,rRepresenting the difference between the known point o and the reference point r, o being a known point, 1,2, …, E1, E1
The total number of the chips is,σeis an adjustable parameter;represents a reference point r1And a reference point r2Inter-difference, lo (r) and lo (o) represent position information of the reference point r and the known point o, respectively;representing the mean value of the fingerprint feature vectors of the known point o after the mapping of the Gaussian kernel function,representing the mean value of the fingerprint characteristic vectors of the reference point r after the Gaussian kernel function mapping,andthe expression of (a) is as follows:
wherein the content of the first and second substances,andrepresenting the fingerprint feature vectors of the known point o and the reference point r respectively,a kernel matrix which is obtained by mapping the fingerprint characteristic vectors of all known points in the target domain through a Gaussian kernel function,a kernel matrix which is obtained by mapping the fingerprint characteristic vectors of all the reference points in the source field through a Gaussian kernel function,andthe expression of (a) is as follows:
will be provided withA plurality of generalized characteristic values are arranged from large to small and selectedBefore selectionA generalized characteristic value, establishing a dimension ofThe new mapping space of (a) is,
establishing a source sub-domain according to the following rules:
wherein the content of the first and second substances,the representation is based on a reference point r2Optimum K value ofTo the reference point r2Selected for KNN positioningA set of reference points; taking a known point with the same position as a reference point in the same source subdomain as the same target subdomain; dividing each source into domainsEach reference point r insaMapping to a dimension ofIn the new mapping space of (2), obtaining a mapped representation
Wherein the content of the first and second substances,representing source domainsAnd target subdomainA mapping matrix mapped into the new mapping space,representing source subdomainsReference point rsaThe fingerprint feature vector of (2);representing source subdomainsProjecting to a kernel matrix in the new mapping space;
dividing each target into domainsEach reference point o intaMapping to a dimension ofIn the new mapping space of (2), obtaining a mapped representation
Wherein, the first and the second end of the pipe are connected with each other,representing target domainsReference point otaThe fingerprint feature vector of (2);representing target domainsProjecting to a kernel matrix in the new mapping space;
forming a new offline derivative fingerprint library by each known point in the new mapping space and the residual reference points in the new mapping space; the residual reference points are reference points in the source field except for the reference points corresponding to the known points;
constructing a radio map according to the fingerprint feature vectors of the new off-line derivative fingerprint library in the new mapping space, wherein the set of new reference points in the radio map is Represents the E0 th new reference point in the wireless map, E0 ═ 1, 2.., E0; and a new reference pointThe optimal K value is the same as the e0 th reference point in the source field; calculating the point to be measured and the new reference pointFingerprint euclidean distance between:
wherein the content of the first and second substances,representing the fingerprint characteristics corresponding to the points to be measured and the e values in the new mapping space,Indicating a new reference pointThe representation of the fingerprint features corresponding to the e values in the new mapping space;
selecting the optimal K value corresponding to the new reference point closest to the Euclidean distance of the fingerprint of the point to be measured as the optimal K value of the point to be measuredArranging the Euclidean distances of fingerprints between all new reference points and points to be measured from small to large, before selectingNew reference point composition set corresponding to individual fingerprint Euclidean distanceThus obtaining the position of the point to be measured:
whereinRepresentation collectionThe first of the new reference points in the first,representation collectionFirst, the Representation collectionZhongxin reference point r'*The actual position information of the mobile station (ms),representing points to be measured and a new reference point r'*The euclidean distance of the fingerprints therebetween.
Has the advantages that: the method can improve the defects of common fingerprint characteristics by adopting derived fingerprints, and in addition, the K value selection of the KNN positioning model is determined by an online optimal selection method of the K value of the point to be measured so as to improve the positioning accuracy of the model.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a block diagram of an online optimal selection positioning method for K values of points to be measured;
fig. 3 is a flow chart of a sectored radio map construction.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate an embodiment of the invention and, together with the description, serve to explain the invention and not to limit the invention.
The invention provides a 5G indoor intelligent positioning method based on derivative fingerprint migration, as shown in figure 1, the 5G indoor intelligent positioning method based on derivative fingerprint migration provided by the invention comprises the steps that firstly, a 5G base station collects RSS information, signal incidence angles and position information received by indoor mobile equipment at different reference points, and uploads the collected data to an edge server; secondly, the edge server performs three types of fingerprint derivation by combining RSS and incidence angle information, and establishes a derived fingerprint-optimal K value-position fingerprint database (offline derived fingerprint database) by combining reference point position information and optimal K value; and finally, indoor positioning is carried out through an online optimal selection method of the K value of the point to be measured and a KNN model migration mechanism considering environmental time variation, and a better positioning result is sent to the equipment to be measured under the condition of low off-line data re-acquisition.
The invention mainly comprises three contents: firstly, generating derivative fingerprints, and respectively generating three types of fingerprint characteristics according to RSS and incidence angle information; secondly, an online optimal selection method of K values of the points to be measured is provided to improve the KNN positioning model; and thirdly, the edge server performs indoor positioning according to a KNN model migration mechanism considering environmental time variation, and predicts the position of the equipment to be tested under the condition of less off-line data re-acquisition.
1. And (3) generating a derivative fingerprint:
the derived fingerprint in this embodiment is a fingerprint feature derived by combining fingerprints based on RSS and incident angle information, and includes: on the basis of the accumulated average fingerprint characteristics, data acquisition is considered to be carried out in a short time, and the data have larger correlation, so that the currently acquired data are continuously corrected by utilizing historical data. The fingerprint joint characteristic based on difference can effectively solve the problem of device heterogeneous caused by hardware or software problems. The joint fingerprint characteristic based on the hyperbola is used for solving the problem that the difference between the absolute values of the signals acquired based on different users is large, and the data based on the proportion is used for being more reliable.
In the off-line stage, the set of reference points collected in the positioning scene is R, the total number is E0, U time samples are collected for each reference point, and the number of 5G base stations participating in positioning is M. For the reference point r, the corresponding position information is (x)r,yr). Based on the RSS and the angle of incidence information, the joint fingerprint of the reference point r is represented as:
wherein the content of the first and second substances,representing RSS data collected by 5G base station m at time index U (at time U, U is 1,2, …, U) to reference point r,representing angle data acquired by the 5G base station m for the reference point r at the time index u; m is 1,2, …, M.
Based on the accumulated and averaged fingerprint joint characteristics, the characteristics can observe the collected historical data and modify the currently collected data by combining the occurrence probability of the collected historical data. The specific formula for the cumulative average is as follows:
wherein the content of the first and second substances,represents the cumulative average of the data at times 1 to u, and l (u +1) represents the actual value of the data collected at time u + 1. Accordingly, the cumulative average fingerprint feature for reference point r is represented as:
wherein, F1(·) denotes the cumulative average fingerprint feature, wherein,representing the accumulated average value of the joint fingerprint of the Mth 5G base station to the reference point r at the U-1 th moment;representing the cumulative average of the 5G base station M over a certain time sequence for the reference point r combined fingerprint.
Difference-based fingerprint joint feature, which is to perform joint fingerprints received from different base stations under the same time sequence numberDifferenced from each other to construct all fingerprints under that time sequence number. For reference point r, 5G base station m1And 5G base station m2The differential fingerprint of the received data at time u is:
wherein m is1=1,2,...,M,m21,2, M, and M1≠m2。
The joint feature of the reference point r based on the difference fingerprint is expressed as:
F2(RSS_θr)=[Δψr(1),…,Δψr(u),…,Δψr(U)]
wherein, F2(. represents a differential fingerprint,. DELTA.. psir(u) is expressed as:
based on the combined characteristics of the fingerprints of the hyperbola, the fingerprint has better stability than the data of the direct use signal.
For reference point r, 5G base station m1And 5G base station m2The hyperbolic fingerprint of the received data at time sequence u is:
so the hyperbolic-based joint fingerprint feature of the reference point r is represented as:
F3(RSS_θr)=[lr(1),…,lr(u),…,lr(U)]
wherein, F3Denotes the change of the fingerprint into a hyperbolic fingerprint,/r(u) is expressed as:
2. the online optimal selection method of the K value of the point to be measured comprises the following steps:
in the KNN indoor positioning method, the positioning accuracy may be degraded due to noise. In addition, due to different position environments of the points to be measured, selecting a fixed K value cannot ensure that the points to be measured at each position obtain better positioning accuracy. In order to find the optimal self-adaptive K value of each initial positioning condition, the invention provides an online optimal selection method of the K value of the point to be measured, as shown in FIG. 2.
In order to determine the optimal K value for each off-line reference point, a KNN algorithm is performed for all reference points in the off-line phase. Calculating any two reference points r1And r2Has a European style distance of fingerprint between
Wherein r is1=1,2,...,E0,r21,2, E0, and r1≠r2,||·||2The formula of the two-norm is expressed,and representing the fingerprint characteristic weight coefficient corresponding to the value of e. Smaller Euclidean distance of the fingerprint indicates larger similarity of the fingerprint. Note that while KNN locating reference points, no match itself in the off-line fingerprint library. The specific matching process comprises the following steps: calculating a reference point r1And a reference point r2The Euclidean distance of the fingerprints between the first and second images is sorted from small to large, and the front with the smallest Euclidean distance of the fingerprints is selectedA reference point forming a setWherein the content of the first and second substances,representation collectionReference points with the middle sequence number of lr1, and the position information of these reference points is used for weighted cumulative summation.
Reference point r1The predicted position of (a) is expressed as:
wherein the content of the first and second substances,representation collectionReference point r'1The actual position information of the mobile station (ms),representing a reference point r1And collectionsReference point r'1The euclidean distance of the fingerprint.
To ensure the minimum positioning error of each reference point, the optimal K value of each reference point needs to be reasonably considered to be the reference point r1For example, the following optimization objectives may be formulated:
wherein the content of the first and second substances,the upper limit of (2) is becauseLarger will result in more computation consumption. Obtaining a reference point r by optimizing the target1Optimum K value of Representing reference points r based on optimal K values1Selected for KNN positioningA set of reference points.
In the online stage, if the time variation of the actual indoor scene does not exceed the preset days, calculating the optimal K value and position of the point to be measured in the target domain by adopting the following method; the target domain comprises a known point and a point to be measured, wherein the known point is RSS information in an actual indoor scene, and both the incident angle and the position information of a signal are known, and the point to be measured is RSS information in the actual indoor scene, and the incident angle of the signal is known but the position information is unknown.
Calculating the Euclidean distance of the fingerprint between the point to be measured and the reference point r;
and then selecting a corresponding K value according to the matched reference point, wherein the specific mode is as follows: calculating Euclidean distance of the point to be measured and all the reference points, and selecting the optimal K value of the reference point with the minimum Euclidean distance as the K value of the point to be measured (namely) Then, continuously selecting the front with minimum European distance of fingerprintA reference point, the number of the finally formed elements isSet of reference pointsSo as to realize the online selection method of the optimal K value of the point to be measured. The final positioning result of the point to be measured is as follows:
wherein the content of the first and second substances,representation collectionReference point r*The actual position information of the mobile station (bs),represents a reference point r*The euclidean distance of the fingerprint from the test point,representation collectionIs the first of the reference points in (a),representation collectionTo middleA reference point.
3. KNN model migration mechanism considering environmental time variation:
according to the KNN positioning principle, the position of any point is obtained by selecting the positions of K adjacent points according to the fingerprint similarity and calculating. But the environmental time variation causes the K value required by the KNN positioning model and the off-line derivative fingerprint library to change. To ensure the availability of the optimal K value under ambient time variation (actual indoor ambient time variation exceeds a preset number of days). Thus, as shown in fig. 3, the invention employs known point acquisition based on offline optimal K-values for subsequent construction of a radio map.
According to the optimal K value solved by the offline reference points in the source field (taking the offline derivative fingerprint library as the source field), the feature vectors of the three types of fingerprints of a certain reference point can be confirmed to have larger similarity with the feature vectors of other K-1 reference points; the one-way connection relation matrix between different position reference points is as follows:
wherein the content of the first and second substances,represents a reference point r1To a reference point r2The unidirectional connection relation of (1) is generally 0 and 1, 0 represents unidirectional connection, 1 represents unidirectional disconnection,wherein the content of the first and second substances,representation collectionThe number of elements (c); representation is based on the bestOptimal K value versus reference point r1The reference point set matched by KNN is carried out; according to the unidirectional connection relation matrix between the reference points, a reference point set can be constructed according to the following rules:
wherein the content of the first and second substances,the representation is based on a reference point r2Optimum K value ofTo the reference point r1Selected for KNN positioningA set of reference points.
According to the formula, the reference points in the source field can be partitioned, one partition is used as a source sub-field, then the known points of the target field with the same position needing to be collected are selected by combining the position information corresponding to the reference point of each partition, and finally the known points of the target field are subjected to partition migration learning to realize the extension of the radio map of the target field. And partitioning and migrating the source field by using the constructed reference point set.
For a time-varying indoor environment, the distribution of information features obtained at the same position in an offline stage (source domain) and an online stage (target domain) may be inconsistent, thereby causing the accuracy of fingerprint positioning to be reduced. If the fingerprint database is collected again for the target domain, the positioning cost is increased sharply and the subsequent positioning cannot be assisted. However, for a time-varying environment, the location of the positioning base station is substantially fixed, so that a migration learning method can be adopted to migrate the source domain data onto the target domain.
In the off-line phase, the off-line derivative fingerprint library is used as the source domain. Source domain DsThe fingerprint feature vector corresponding to the reference point r is expressed as:
source domain DsThe representation contains the set of all reference points, namely:
wherein E0 denotes the source domain DsTotal number of reference points.
In an online stage, a known position label and a known point of a known fingerprint characteristic vector and a point to be measured of an unknown position label and a known fingerprint characteristic vector are collected in indoor space to serve as a target domain DtThe fingerprint feature vector constructed by the known point o is represented as:
wherein, o is 1,2, …, E1, E1 is the total number of known points.
Target domain DtThe fingerprint feature vector constructed by the middle point to be measured o' is expressed as:
wherein, o' 1,2, E2, E2 are the total number of the points to be measured.
In the partition migration learning, a reference point and a known point of the same region are selected for the migration learning. The fingerprint feature vectors are first mapped to other dimensional spaces using a mapping function. Suppose thatRepresenting functions mapped to other dimensions, taking into account the mapping functionThe specific form of (c) is not well determined, so the kernel function Ke (-) is adopted as a substitute for other dimension spaces and mapping functions, namely:
wherein o is1And o2Represents a known point o1And a known point o2,o1=1,2,…,E1,o21,2, …, E1, and o1≠o2;o'1And o'2Represents a point to be measured o'1And to-be-measured point o'2,o'1=1,2,…,E2,o'21,2, …, E2 and o'1≠o'2,σkeRepresents the tunable parameters of a gaussian kernel function.
Accordingly, the source domain DsMean of fingerprint feature vectors with reference point rExpressed as:
wherein the content of the first and second substances,representing the representation of the source domain after kernel function mapping,passing of fingerprint feature vectors representing source domain reference points through a Gaussian kernel functionThe number of the mapped kernel matrix is counted, representation matrixMiddle r1Line r2The elements of the column.
The mean of the fingerprint feature vectors for the known point o in the target domain is represented as:
wherein the content of the first and second substances,the representation target domain is mapped to other dimensional spaces,a kernel matrix obtained by mapping the fingerprint feature vector of the known point of the target domain through a Gaussian kernel function, representation matrixO in1Line o2The elements of the column.
Source domain DsAnd a target domain DtThe inter-domain inter-class divergence matrix of (a) is represented as:
wherein, lo (r) andlo (o) indicates position information of the reference point r and the known point o,representing the difference, σ, between the position of the reference point r and the position of the known point oeIndicating an adjustable parameter.
Source domain DsAnd a target domain DtThe inter-domain intra-class divergence matrix of (a) is represented as:
source domain DsThe intra-domain inter-class divergence matrix is represented as:
wherein, lo (r)1) And lo (r)2) Represents a reference point r1And a reference point r2The category of the location information of (a),represents a reference point r1And a reference point r2The difference between them.
Source domain DsThe intra-domain intra-class divergence matrix is represented as:
the constructed objective function of the transfer learning is as follows:
wherein, χ1Hexix-2The method belongs to two adjustable parameters and is used for controlling the importance degree of the inter-domain characteristics and the intra-domain characteristics when calculating the mapping matrix; the objective function is related toAndthe generalized Rayleigh entropy is solved by a Lagrange multiplier method of the target function of the transfer learning, and finally the generalized Rayleigh entropy can be obtained:
where λ is the lagrange multiplier, defined for convenience:
If newly built a dimension isAnd (4) new mapping space of dimension, the expression form of the mapping matrix is as follows:
wherein the content of the first and second substances,to representTo (1) aAnd the feature vector matrix corresponding to the large generalized eigenvalue.
Partitioning through a reference point set of a source field, and comparing with the reference point setTaking a known point with the same reference point position in the combined partition as one partition in the target domain, and taking the partition as the target partition; the target domain radio map reconstruction based on the partitions is realized through the migration learning mapping space calculated by each partition set (source partition and target partition); according to a mapping matrixThe source can be divided into domainsReference point rsaMapping to a new mapping space, which is represented as:
wherein the content of the first and second substances,representing source subdomainsIs projected to the representation of the new mapping space,representing source subdomainsAnd target subdomainThe mapping matrix of (a) is obtained,andrepresenting source subdomainsScoping with a targetAnd (5) projection kernel matrix.
Likewise, according to the mapping matrixObject sub-domains at the same position in the object domainKnown point otaMapping to a new space, which is represented as:
wherein, the first and the second end of the pipe are connected with each other,representing target domainsKnown point of (a)taA representation projected into the new mapping space, wherein,representing target subdomainsReference point otaThe fingerprint feature vector of (2);representing target domainsThe kernel matrix in the new mapping space is projected.
Forming a new offline derivative fingerprint database by each known point in the new mapping space and the residual reference points in the new mapping space; the remaining reference points are reference points in the source domain other than the reference points corresponding to the known points.
In the new off-line derivative fingerprint database, the position isThe fingerprint feature vector of the point in the new mapping space is:
constructing a radio map according to the fingerprint feature vectors of the new offline derivative fingerprint library in the new mapping space, wherein the set of the new reference points in the constructed radio map isWherein the content of the first and second substances,denotes a new reference point with sequence number e0, andand (3) carrying out reference point matching by using a nearest neighbor algorithm, wherein the optimal K value is the same as the e0 th reference point in the source field:
wherein the content of the first and second substances,representing new reference points in a constructed radiomapThe Euclidean distance of the fingerprint between the measuring point and the measuring point through other dimension transformation,fingerprint characteristics corresponding to points to be measured and values of e are mapped in a new spaceIs shown in a schematic view of a display device,indicating a new reference pointRepresentation of fingerprint features corresponding to the e-value in the new mapping space; calculating other dimensional transformation of all new reference points and points to be measuredThe Euclidean distances of the fingerprints are sorted from small to large, and the optimal K value of the new reference point with the minimum distance is selected as the optimal K value of the point to be measuredThen, the front with minimum Euclidean distance of fingerprints transformed in other dimensions is continuously selectedA new reference point, the number of elements formed isSet of (2)By adopting the method for realizing the online selection of the optimal K value of the point to be measured, the final positioning result of the point to be measured is as follows:
wherein, thereinRepresentation collectionThe first of the new reference points in the first,representation collectionFirst, theThe new reference point is then used to determine the new reference point,representation collectionZhongxin reference point r'*The actual position information of the mobile station (bs),representing points to be measured and a new reference point r'*The euclidean distance of the fingerprints therebetween.
It should be noted that the various features described in the above embodiments may be combined in any suitable manner without departing from the scope of the invention. The invention is not described in detail in order to avoid unnecessary repetition.
Claims (5)
1. An indoor intelligent positioning method based on derivative fingerprint migration is characterized in that: the method specifically comprises the following steps:
step 1: in an off-line stage, mobile equipment is adopted to communicate with a 5G base station at different indoor positions, the different positions are taken as reference points, RSS information received by the mobile equipment at each reference point in a time sequence, an incident angle of a signal received by the mobile equipment at each reference point and position information of each reference point are obtained; the total number of the moments in the period of time sequence is U, the total number of the 5G base stations is M, the total number of the reference points is E0, and the set of the reference points is R;
step 2: calculating for each reference point a derived fingerprint feature comprising a cumulative mean fingerprint feature F1(RSS_θr) Differential fingerprintCharacteristic F2(RSS_θr) And hyperbolic fingerprint feature F3(RSS_θr):RSS_θrA joint fingerprint representing a reference point r based on the RSS and the incident angle information; calculating the optimal K value of each reference point according to the three fingerprint characteristics of each reference point; establishing an offline derivative fingerprint database, wherein the offline derivative fingerprint database comprises derivative fingerprint characteristics of each reference point, an optimal K value, RSS information, signal incidence angle and position information, and r is 1, 2.
And step 3: judging whether the time variation of the actual indoor scene exceeds the preset days, if so, turning to the step 4, and otherwise, calculating the position information of the points to be measured in the target domain according to the optimal K value of the reference points in the offline derivative fingerprint library; the target domain comprises a known point and a point to be measured, wherein the known point is RSS information in an actual indoor scene, and both the incident angle and the position information of a signal are known;
and 4, step 4: and migrating the off-line derivative fingerprint library by adopting migration learning so as to construct a radio map, and determining the position information of the point to be measured according to a new reference point in the radio map and the optimal K value of the reference point in the off-line derivative fingerprint library before migration.
2. The indoor intelligent positioning method based on derivative fingerprint migration according to claim 1, wherein the step 2 is: the calculation of the cumulative average fingerprint feature of the reference point r is specifically:
the calculation formula for setting the cumulative average is as follows:
wherein the content of the first and second substances,represents the cumulative average of the data from time 1 to time u, l (u +1) represents the actual value of the data at time u +1, u=1,2,…,U;
Calculating the cumulative average fingerprint characteristic of the reference point r based on a cumulative average calculation formula:
wherein the content of the first and second substances,the accumulated average value of the derivative fingerprints of the Mth 5G base station to the reference point r at the U-1 th moment is represented; RSS _ thetarThe expression of (a) is:
wherein M represents the mth 5G base station, and M is 1,2, …, M;representing RSS data collected by the mth 5G base station at time u with respect to reference point r,representing signal incidence angle data collected by the mth 5G base station to the reference point r at the moment u;
differential fingerprint feature F of reference point r2(RSS_θr) Is expressed as:
F2(RSS_θr)=[Δψr(1),…,Δψr(u),…,Δψr(U)]
wherein, Delta psirThe expression of (u) is:
wherein m is1Represents the m-th1Base station, m2Denotes the m-th2Base station, m1=1,2,...,M,m21,2, M, and M1≠m2;The expression of (a) is as follows:
hyperbolic fingerprint feature F of reference point r3(RSS_θr) The expression of (a) is:
F3(RSS_θr)=[lr(1),…,lr(u),…,lr(U)]
wherein lrThe expression of (u) is:
3. the method according to claim 2, wherein the step 2 of calculating the optimal K value of each reference point specifically comprises: calculating any two reference points r1And r2Has a European style distance of fingerprint between
Wherein r is1=1,2,...,E0,r21,2, E0, and r1≠r2,||·||2Expressing a two-norm formula;representing a weight coefficient of fingerprint features corresponding to the e value;
calculating the reference point r according to the following objective function1Optimum K value of (c):
wherein the content of the first and second substances,andrepresents a reference point r1The actual position information of the mobile station (ms),andrespectively represent reference points r1The abscissa and the ordinate of the predicted position of (c),representing a reference point r1The value of (a) of (b),represents a reference point r1K value ofAt the reference point r1Selected for KNN positioningA set of reference points, solving the objective function to obtain a reference point r1Optimum K value ofRepresenting reference points r based on optimal K values1Selected for KNN positioningA set of reference points;
wherein the content of the first and second substances,representation collectionThe 1 st reference point in the list of (c),representation collectionTo middleThe number of the reference points is,represents a reference point r1And collectionsReference point r'1The euclidean distance of the fingerprint of (c),representation collectionReference point r'1Actual location information of.
4. The method according to claim 3, wherein the step 3 of calculating the position information of the points to be measured in the target domain specifically comprises:
using a nearest neighbor algorithm for reference point matching:
wherein EDr,*Representing the Euclidean distance of the fingerprint of the point to be measured from the reference point r, Fe(RSS_θ*) Representing fingerprint characteristics of points to be measured corresponding to the values of e;
selecting the optimal K value corresponding to the reference point with the closest Euclidean distance as the optimal K value of the point to be measuredSorting the Euclidean distances of the fingerprints of each reference point and the point to be measured from small to large, and taking the firstCorresponding to European distance of individual fingerprintReference points constitute a setThereby calculating the position (x) of the point to be measured*,y*):
Wherein the content of the first and second substances,representation collectionIs the first of the reference points in (a),representation collectionTo middleA reference point, EDr*,*Representing points to be measured and the setReference point r*Of the Euclidean fingerprint distance therebetween, said (x)r*,yr*) Representation collectionReference point r*Actual location information of.
5. The method according to claim 3, wherein the step 4 specifically comprises: taking the offline derivative fingerprint library as a source field, and establishing a transfer learning objective function:
and (3) carrying out Lagrange multiplier method solving on the target function of the transfer learning to obtain:
wherein, χ1Hexix-2Are all parameters, λ is Lagrangian multiplier, V isCharacteristic vector of (2) to Representing the source domain DsAnd a target domain DtThe inter-domain inter-class divergence matrix of (c),representing the source domain DsAnd a target domain DtThe inter-domain intra-class divergence matrix of (c),source domain DsThe inter-class divergence matrix within the domain,source domain DsAn intra-domain intra-class divergence matrix;andthe expression of (a) is:
wherein, wo,rThe difference between the known point o and the reference point r is shown, wherein o is 1,2, …, E1, E1 is the total number of the known points,σeis an adjustable parameter;represents a reference point r1And a reference point r2The differences between lo (r) and lo (o) represent the position information of the reference point r and the known point o, respectively;representing the mean value of the fingerprint feature vectors of the known point o after the mapping of the Gaussian kernel function,representing the mean value of the fingerprint characteristic vectors of the reference point r after the Gaussian kernel function mapping,andthe expression of (c) is as follows:
wherein the content of the first and second substances,andrepresenting the fingerprint feature vectors of the known point o and the reference point r respectively,a kernel matrix which is obtained by mapping the fingerprint characteristic vectors of all known points in the target domain through a Gaussian kernel function,a kernel matrix which is obtained by mapping the fingerprint characteristic vectors of all reference points in the source field through a Gaussian kernel function,andthe expression of (a) is as follows:
will be provided withA plurality of generalized eigenvalues are arranged from large to small before selectionA generalized characteristic value, establishing a dimension ofThe new mapping space of (2);
establishing a source subdomain according to the following rules:
wherein the content of the first and second substances,the representation is based on a reference point r2Optimum K value ofTo the reference point r2Selected for KNN positioningA set of reference points; taking a known point with the same position as a reference point in the same source subdomain as the same target subdomain; dividing each source into domainsEach reference point r insaMapping to a dimension ofIn the new mapping space of (2), obtaining a mapped representation
Wherein the content of the first and second substances,representing source subdomainsAnd target subdomainA mapping matrix mapped into the new mapping space,representing source domainsReference point rsaThe fingerprint feature vector of (2);representing source subdomainsProjecting to a kernel matrix in the new mapping space;
dividing each target into domainsEach reference point o intaMapping to a dimension ofIn the new mapping space of (2), obtaining a mapped representation
Wherein the content of the first and second substances,representing target domainsReference point otaThe fingerprint feature vector of (2);representing target subdomainsProjecting to a kernel matrix in the new mapping space;
forming a new offline derivative fingerprint database by each known point in the new mapping space and the residual reference points in the new mapping space; the residual reference points are reference points in the source field except for the reference points corresponding to the known points;
constructing a radio map according to the fingerprint feature vectors of the new off-line derivative fingerprint library in the new mapping space, wherein the set of new reference points in the radio map is Represents the E0 th new reference point in the wireless map, E0 ═ 1, 2.., E0; and a new reference pointThe optimal K value is the same as the e0 th reference point in the source field; calculating the point to be measured and the new reference pointFingerprint euclidean distance between:
wherein, the first and the second end of the pipe are connected with each other,representing the representation of the fingerprint characteristics corresponding to the points to be measured and the values of e in the new mapping space,indicating a new reference pointThe representation of the fingerprint features corresponding to the e values in the new mapping space;
selecting the optimal K value corresponding to the new reference point closest to the Euclidean distance of the fingerprint of the point to be measured as the optimal K value of the point to be measuredArranging the Euclidean distances of fingerprints between all new reference points and points to be measured from small to large, before selectingNew reference point composition set corresponding to individual fingerprint Euclidean distanceThus obtaining the position of the point to be measured:
whereinRepresentation collectionThe first of the new reference points in the first,representation collectionTo middleThe new reference point is the point of reference,representation collectionZhongxin reference point r'*The actual position information of the mobile station (ms),representing points to be measured and a new reference point r'*The euclidean distance of the fingerprints therebetween.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210330811.2A CN114679683A (en) | 2022-03-30 | 2022-03-30 | Indoor intelligent positioning method based on derivative fingerprint migration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210330811.2A CN114679683A (en) | 2022-03-30 | 2022-03-30 | Indoor intelligent positioning method based on derivative fingerprint migration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114679683A true CN114679683A (en) | 2022-06-28 |
Family
ID=82076752
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210330811.2A Pending CN114679683A (en) | 2022-03-30 | 2022-03-30 | Indoor intelligent positioning method based on derivative fingerprint migration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114679683A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116233747A (en) * | 2023-03-09 | 2023-06-06 | 江苏亨鑫科技有限公司 | BLE positioning method and device based on transfer learning and fingerprint library updating |
-
2022
- 2022-03-30 CN CN202210330811.2A patent/CN114679683A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116233747A (en) * | 2023-03-09 | 2023-06-06 | 江苏亨鑫科技有限公司 | BLE positioning method and device based on transfer learning and fingerprint library updating |
CN116233747B (en) * | 2023-03-09 | 2023-12-22 | 江苏亨鑫科技有限公司 | BLE positioning method and device based on transfer learning and fingerprint library updating |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110012428B (en) | Indoor positioning method based on WiFi | |
CN110856112B (en) | Crowd-sourcing perception multi-source information fusion indoor positioning method and system | |
CN108989976B (en) | Fingerprint positioning method and system in intelligent classroom | |
CN106793086B (en) | Indoor positioning method | |
KR102116824B1 (en) | Positioning system based on deep learnin and construction method thereof | |
CN107318084A (en) | A kind of fingerprint positioning method and device based on optimal similarity | |
CN110536257B (en) | Indoor positioning method based on depth adaptive network | |
CN106912105A (en) | 3-D positioning method based on PSO_BP neutral nets | |
CN106597363A (en) | Pedestrian location method in indoor WLAN environment | |
CN109348416B (en) | Fingerprint indoor positioning method based on binary k-means | |
CN108882172B (en) | Indoor moving trajectory data prediction method based on HMM model | |
CN108896962B (en) | Iterative positioning method based on sound position fingerprint | |
CN110636437B (en) | Fingerprint pair-based auxiliary positioning method | |
CN109379711B (en) | positioning method | |
Yang et al. | Research on Wi-Fi indoor positioning in a smart exhibition hall based on received signal strength indication | |
CN104581945B (en) | The WLAN indoor orientation methods of semi-supervised APC clustering algorithms based on distance restraint | |
CN111901749A (en) | High-precision three-dimensional indoor positioning method based on multi-source fusion | |
CN109143161B (en) | High-precision indoor positioning method based on mixed fingerprint quality evaluation model | |
CN110401977A (en) | A kind of more floor indoor orientation methods returning more Classification and Identification devices based on Softmax | |
Xie et al. | BPNN based indoor fingerprinting localization algorithm against environmental fluctuations | |
CN114679683A (en) | Indoor intelligent positioning method based on derivative fingerprint migration | |
CN106197418B (en) | A kind of indoor orientation method merged based on the fingerprint technique of sliding window with sensor | |
Huan et al. | Indoor location fingerprinting algorithm based on path loss parameter estimation and bayesian inference | |
CN114449652B (en) | WIFI indoor positioning method based on reliable AP selection | |
CN109633531A (en) | Wireless sensor network node positioning system under composite noise condition |
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 |