Position sensing method, device and medium based on multi-signal similarity measurement
By constructing AP position heterogeneous graphs and using graph self-supervised learning algorithms, the feature vectors of multi-signal RSSI sequences are calculated, which solves the problem of comparing different signal similarity in the existing technology, and realizes high-precision multi-signal similarity measurements. The relative position of the mobile terminal.
Patent Information
- Application Number
- CN202211359537.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-02
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-11-02
AI Technical Summary
The prior art cannot effectively compare the RSSI sequence similarity of different signals, and due to device differences and environmental changes, the need for multi-signal position perception cannot be met.
By constructing AP position heterogeneous graphs, using pre-trained hierarchical graph neural network model and graph self-supervised learning algorithms, the feature vectors of multi-signal RSSI sequences are calculated to measure different signal similarity.
It improves the accuracy of multi-signal similarity measurement, overcomes the signal noise impact caused by device differences and environmental changes, meets the comparative needs of different signal similarity, and reflects the relative position of the mobile terminal position.
Smart Images

Figure CN115915392B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of positioning technology, and in particular to a location sensing method, device and medium based on multi-signal similarity measurement. Background Art
[0002] With the development of IoT and sensor technologies, a large number of signals, such as Wi-Fi signals and Bluetooth signals, are widely used in various positioning and sensing scenarios. RSSI sequences can be used to represent a position in space. Positions with similar distances have similar RSSI sequences because of the small signal differences. Therefore, the similarity of RSSI sequences reflects the distance of the corresponding positions in space. Measuring the similarity of RSSI sequences has a large number of applications in smart city scenarios, such as location tracking, indoor positioning, unmanned driving, etc.
[0003] The existing technology mainly compares the similarity of the same signal, and cannot meet the similarity comparison of the RSSI sequences of two different signals. In real scenarios, there are differences in the permissions enabled on the devices of different users. For example, the user has enabled the Wi-Fi permission but not the Bluetooth permission, and the user has enabled the Bluetooth permission but not the Wi-Fi permission. The use of existing technology cannot meet the need for similarity comparison through the RSSI sequences of two different signals. In addition, different devices have different signal perception capabilities; even if two different devices are in the same place, the number of APs scanned and the strength of the signal are different. Secondly, the strength of the signal varies with changes in the environment. Even if the same device is in the same place, the number of APs scanned and the strength of the signal are different at different times.
[0004] In summary, the existing technology lacks consideration of the correlation between access points and the correlation between locations and access points, and is unable to overcome the influence of signal noise caused by differences in equipment and environmental changes, and its performance is greatly limited. Summary of the invention
[0005] The present invention provides a location perception method based on multi-signal similarity measurement. By modeling the correlation between APs, between APs and locations, and between different signals, the characteristic expression of the location is learned, thereby realizing the measurement of the similarity of multi-signal RSSI sequences and determining the relative positions of different mobile terminals.
[0006] An embodiment of the present invention provides a location perception method based on multi-signal similarity measurement, including:
[0007] Acquire an RSSI sequence of the first mobile terminal and an RSSI sequence of the second mobile terminal;
[0008] Adding the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal to the RSSI layer, and updating the AP position heterogeneous graph; the AP position heterogeneous graph is obtained by connecting the first node isomorphic graph and the second node isomorphic graph;
[0009] Calculate the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal according to the pre-trained hierarchical graph neural network model;
[0010] The similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal is calculated to obtain the position similarity between the first mobile terminal and the second mobile terminal.
[0011] The first node isomorphic graph specifically includes:
[0012] Acquire an RSSI data set of a first signal, where the RSSI data set of the first signal includes at least one RSSI sequence; wherein an RSSI sequence of the first signal is defined as:
[0013]
[0014] in represents the kth node, Represents R i The signal strength of the kth node in;
[0015] A first node isomorphic graph is constructed according to the RSSI data set of the first signal.
[0016] The second node isomorphic graph specifically includes:
[0017] Acquire an RSSI data set of a second signal, where the RSSI data set of the second signal includes at least one RSSI sequence; wherein an RSSI sequence of the second signal is defined as:
[0018]
[0019] in represents the kth node, Represents R j The signal strength of the kth node in;
[0020] A second node isomorphic graph is constructed according to the RSSI data set of the second signal.
[0021] Furthermore, the AP position heterogeneous graph specifically includes:
[0022] Using Node R kRepresenting an RSSI sequence, constructing an RSSI sequence layer from the RSSI sequences in all the data sets, and constructing an AP position heterogeneous graph using the first node isomorphic graph and the second node isomorphic graph;
[0023] Among them, if the node signal strength in the RSSI sequence exists s k,i ≠0, then R k With node a i Connected, the edge e(R k , a i ) is defined as follows:
[0024]
[0025] As an improvement to the above solution, the pre-trained hierarchical graph neural network model specifically includes:
[0026] Inputting the first node isomorphic graph into a first graph convolutional network, and using a graph self-supervised learning method to obtain a feature vector of the first node;
[0027] Inputting the second node isomorphic graph into a second graph convolutional network, and using a graph self-supervised learning method to obtain a feature vector of the second node;
[0028] Based on the learned feature vector of the first node, inputting the first node isomorphic graph into a third graph convolutional network to obtain a feature vector of the third node;
[0029] Based on the learned feature vector of the second node, inputting the second node isomorphic graph into a fourth graph convolutional network to obtain a feature vector of the fourth node;
[0030] Obtaining a feature vector of each RSSI sequence by feature aggregation at an RSSI sequence layer according to the AP position heterogeneous graph, the feature vector of the third node, and the feature vector of the fourth node;
[0031] Calculate the similarity of the feature vectors of two RSSI sequences;
[0032] The hierarchical graph neural network model is trained by constructing RSSI pairs.
[0033] The feature aggregation is calculated as follows:
[0034]
[0035] Among them, Neigh(R k ) represents R k Neighbor nodes in the AP location heterogeneous graph, w(R k , a i) represents the edge e(R k , a i ), W5 is a learnable parameter, is the RSSI sequence of node a i The feature vector of .
[0036] The formula for calculating the similarity is:
[0037]
[0038] Among them l i With l j They are respectively the characteristic vector of the RSSI sequence of the first mobile terminal and the characteristic vector of the RSSI sequence of the second mobile terminal.
[0039] Accordingly, an embodiment of the present invention further provides a location sensing device based on multi-signal similarity measurement, comprising:
[0040] A target sequence acquisition module, used to acquire the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal;
[0041] An AP position heterogeneous graph updating module is used to add the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal to the RSSI layer, and update the AP position heterogeneous graph; the AP position heterogeneous graph is obtained by connecting the first node isomorphic graph and the second node isomorphic graph;
[0042] A feature vector acquisition module is used to calculate the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal according to a pre-trained hierarchical graph neural network model;
[0043] The similarity calculation module is used to calculate the similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal to obtain the position similarity between the first mobile terminal and the second mobile terminal.
[0044] Correspondingly, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above-mentioned location perception method based on multi-signal similarity measurement are implemented.
[0045] Compared with the prior art, the present invention has the following beneficial effects:
[0046] The present invention constructs an AP position heterogeneous graph based on RSSI data sets of different types of signals, describes the complex relationships between APs, between APs and positions, and between different signals, and uses a graph self-supervised learning algorithm to pre-train the feature expression of APs to improve the expressiveness of AP feature expression; and learns the feature expression of APs and positions through a hierarchical graph neural network, obtains the feature data of RSSI sequences through feature aggregation, and finally uses similarity calculation to compare signal similarities, solves the signal noise caused by equipment differences and environmental changes, and realizes high-precision multi-signal similarity measurement; thereby meeting the needs of measuring the similarity of different types of signals, reflecting the distance of the positions corresponding to different mobile terminals in space, and thus determining their relative positions. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 is a flow chart of a location perception method based on multi-signal similarity measurement provided by an embodiment of the present invention;
[0048] Figure 2 is a schematic diagram of mobile phone signal scanning in an embodiment of the present invention;
[0049] Figure 3 is a WIFI node isomorphic graph in an embodiment of the present invention;
[0050] Figure 4 is a Bluetooth node isomorphic graph in an embodiment of the present invention;
[0051] Figure 5 It is a structural schematic diagram of a position sensing device based on multi-signal similarity measurement provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0053] See also Figure 1 , Figure 1 It is a flowchart of a location perception method based on multi-signal similarity measurement provided by an embodiment of the present invention.
[0054] The embodiment of the present invention provides a location perception method based on multi-signal similarity measurement, including steps S101 to S104, which are specifically as follows:
[0055] S1, obtaining an RSSI sequence of a first mobile terminal and an RSSI sequence of a second mobile terminal;
[0056] S2, adding the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal to the RSSI layer, and updating the AP position heterogeneous graph; the AP position heterogeneous graph is obtained by connecting the first node isomorphic graph and the second node isomorphic graph;
[0057] S3, calculating a feature vector of the RSSI sequence of the first mobile terminal and a feature vector of the RSSI sequence of the second mobile terminal according to a pre-trained hierarchical graph neural network model;
[0058] S4, calculating the similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal, and obtaining the position similarity between the first mobile terminal and the second mobile terminal.
[0059] Exemplarily, in step S1, the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal are RSSI sequences of two different signals, which may be wireless transmission signals such as WIFI, Bluetooth, NFC or UWB.
[0060] In an optional embodiment, in step S2, the first node isomorphic graph specifically includes:
[0061] Acquire an RSSI data set of a first signal, where the RSSI data set of the first signal includes at least one RSSI sequence; wherein an RSSI sequence of the first signal is defined as:
[0062]
[0063] in represents the kth node, Represents R i The signal strength of the kth node in;
[0064] A first node isomorphic graph is constructed according to the RSSI data set of the first signal.
[0065] The second node isomorphic graph specifically includes:
[0066] Acquire an RSSI data set of a second signal, where the RSSI data set of the second signal includes at least one RSSI sequence; wherein an RSSI sequence of the second signal is defined as:
[0067]
[0068] in represents the kth node, Represents R j The signal strength of the kth node in;
[0069] A second node isomorphic graph is constructed according to the RSSI data set of the second signal.
[0070] It should be noted that the embodiment of the present invention collects data through a crowdsourcing mode. Specifically, the user terminal can collect signals of the user's environment through an application program or the like. For each signal, a node isomorphic graph can be constructed. In addition, a sequence of different signals collected simultaneously by the user terminal can be used to construct an AP position heterogeneous graph.
[0071] Specifically, in the node isomorphism graph, each node a i represents an AP if and only if two APs appear in the same RSSI sequence, two nodes a i with a j There is an edge between them, let e(a i , a j )=1, otherwise e(a i , a j )=0;
[0072] In specific embodiments, see Figure 2 Schematic diagram of mobile phone signal scanning. Assume that there are WIFI AP access points: (a1, a2, a3, a4) and Bluetooth AP access points: (a5, a6, a7) in the scene. If two APs are scanned by the same mobile terminal, they are connected to generate the following: Figure 3 The WIFI node isomorphism graph shown and Figure 4 The Bluetooth node isomorphism graph shown.
[0073] In the node isomorphism graph, for each edge e(a i , a j ), whose weight w(a i , a j ) can be defined as:
[0074]
[0075]
[0076] Among them, we stipulate that in R k In the case of i If it is not scanned, then s k,i = -100; m means i Number of connected APs.
[0077] Furthermore, the AP position heterogeneous graph specifically includes:
[0078] Using Node R kRepresenting an RSSI sequence, constructing an RSSI sequence layer from the RSSI sequences in all the data sets, and constructing an AP position heterogeneous graph using the first node isomorphic graph and the second node isomorphic graph;
[0079] Among them, if the node signal strength in the RSSI sequence exists s k,i ≠0, then R k With node a i Connect them and connect the edge e(R k , a i ) is defined as:
[0080]
[0081] In a specific embodiment, since some RSSI sequences in the RSSI data set may contain different signal types, in the RSSI sequence layer, different node isomorphic graphs are connected through these RSSI sequences, such as Figure 4 As shown, R2 makes different isomorphic graphs related;
[0082] In step S2, the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal are added to the RSSI layer, and the AP location heterogeneous map is updated;
[0083] Specifically, these two RSSIs are added as nodes to the RSSI layer, and each node is connected to the corresponding node in the node isomorphic graph according to which AP signals it contains.
[0084] Then, by defining the weight of each edge in the above AP position heterogeneous graph, the weights of the two newly added nodes and other nodes are calculated, thereby updating the AP position heterogeneous graph.
[0085] In an optional embodiment, in step S3, the pre-trained hierarchical graph neural network model is specifically:
[0086] S31, inputting the first node isomorphic graph into a first graph convolutional network, and obtaining a feature vector of the first node using a graph self-supervised learning method;
[0087] S32, inputting the second node isomorphic graph into a second graph convolutional network, and obtaining a feature vector of the second node using a graph self-supervised learning method;
[0088] It should be noted that the embodiment of the present invention uses a graph self-supervised learning method to pre-train the feature expressions of the nodes in each node isomorphism graph of each signal.
[0089] In a specific embodiment, for any node a in the node isomorphism graphi , we learn through a multi-layer graph convolutional network, whose k-th layer feature expression The calculation of is as follows:
[0090]
[0091] Among them, Neigh(a I ) represents node a i Neighbor nodes in the node isomorphism graph, w(a i , a u ) represents the edge e(a i , a u ), W1 is a learnable parameter, and σ(·) is the relu activation function.
[0092] Next, the feature expressions of nodes in the node isomorphism graph are pre-trained by predicting whether any two given nodes are adjacent in the node isomorphism graph.
[0093] We first select several pairs of adjacent nodes as positive samples, and then randomly select the same number of non-adjacent nodes as negative samples, and make predictions through a fully connected neural network. i with a j , use the following method to make predictions:
[0094] y i,j =W3·σ(W2·(h i ||h j )),
[0095] where || represents the connection between two feature expressions, W2 and W3 are learnable parameters, and σ(·) represents the relu activation function; i,j Indicates whether two nodes i and j are adjacent, y i,j =1 means adjacent, y i,j =0 means not adjacent;
[0096] In pre-training, we choose the cross entropy loss function as the loss function of the pre-training model, use the loss function to calculate the error, and update the parameters of the network.
[0097] It should be noted that the graph self-supervised learning method uses the information in the graph, that is, whether two nodes are adjacent, for training. It enables the model to better learn the information of the graph structure and make the model more stable.
[0098] S33, based on the learned feature vector of the first node, inputting the first node isomorphic graph into a third graph convolutional network to obtain a feature vector of a third node;
[0099] S34, based on the learned feature vector of the second node, inputting the second node isomorphic graph into a fourth graph convolutional network to obtain a feature vector of a fourth node;
[0100] S35, obtaining a feature vector of each RSSI sequence by feature aggregation at an RSSI sequence layer according to the AP position heterogeneous graph, the feature vector of the third node, and the feature vector of the fourth node;
[0101] In a specific embodiment, based on the learned node features and RSSI sequence layer, the features of each RSSI sequence are obtained by feature aggregation. The specific steps are as follows:
[0102] For any node a in the node isomorphism graph i Based on the node features learned above, we learn through a multi-layer graph convolutional network, whose k-th layer features The expression is calculated as follows:
[0103]
[0104] Among them, Neigh(a i ) represents node a i Neighbor nodes in the node isomorphism graph, w(a i , a u ) represents the edge e(a i , a u ), W4 is a learnable parameter, and σ(·) is the activation function.
[0105] For each set of RSSI sequences in the RSSI sequence layer, the characteristic expression R k , we use feature aggregation to scan the node a i The feature expression is aggregated and the calculation method is as follows:
[0106]
[0107] Among them, Neigh(R k ) represents R k Neighbor nodes in the AP location heterogeneous graph, w(R k , a i ) represents the edge e(R k , a i ), W5 is a learnable parameter, is access node a i The feature vector of .
[0108] S36, calculate the similarity of the feature vectors of the two RSSI sequences:
[0109]
[0110] Among them l i With l j They are R i and R j The feature vector of .
[0111] In the embodiment of the present invention, R i With R j It may be an RSSI sequence containing the same signal type or an RSSI sequence containing different signal types.
[0112] S37, train the hierarchical graph neural network model by constructing RSSI pairs.
[0113] If a pair of positive sample RSSI sequences R i , R j belong to the same position, then the pair of sequences R i , R j The similarity Sim(R i , R j ) is greater than other negative samples Sim(R i , R k ), use the pairwise loss function to train the model. The loss function is as follows:
[0114]
[0115] Where R represents the set of RSSIs and 1(·) represents an exponential function.
[0116] After the model training is completed, the first node isomorphism graph, the second node isomorphism graph and the feature vectors of their nodes have been determined; according to the trained model, the feature aggregation formula is:
[0117]
[0118] Among them, Neigh(R k ) represents R k Neighbor nodes in the AP location heterogeneous graph, w(R k , a i ) represents the edge e(R k , a i ), W5 is a learnable parameter, is the RSSI sequence of node a i The feature vector of .
[0119] The characteristic vector of the RSSI sequence of the first mobile terminal and the characteristic vector of the RSSI sequence of the second mobile terminal are calculated respectively.
[0120] Furthermore, in step S4, the similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal is calculated to obtain the position similarity between the first mobile terminal and the second mobile terminal.
[0121] The formula for calculating the similarity is:
[0122]
[0123] Among them l i With l j They are respectively the characteristic vector of the RSSI sequence of the first mobile terminal and the characteristic vector of the RSSI sequence of the second mobile terminal.
[0124] In summary, the location perception method based on multi-signal similarity measurement provided by the embodiment of the present invention constructs an AP location heterogeneous graph based on RSSI data sets of different types of signals, describes the complex relationships between APs, between APs and locations, and between different signals, and pre-trains the feature expression of APs using a graph self-supervised learning algorithm to improve the expressiveness of AP feature expression; and learns the feature expression of AP and location through a hierarchical graph neural network, obtains feature data of RSSI sequence through feature aggregation, and finally uses similarity calculation to compare signal similarity, solves the signal noise caused by device differences and environmental changes, and realizes high-precision multi-signal similarity measurement; thereby meeting the needs of measuring the similarity of different types of signals, reflecting the distance of the positions corresponding to different mobile terminals in space, and thus determining their relative positions.
[0125] An embodiment of the present invention also provides a location sensing device based on multi-signal similarity measurement, which can implement all processes of the location sensing method based on multi-signal similarity measurement provided in any of the above embodiments. The functions of each module and unit in the device and the technical effects achieved are respectively the same as the functions and technical effects achieved by the location sensing method based on multi-signal similarity measurement provided in the above embodiments, and will not be repeated here.
[0126] See also Figure 4 As shown, Figure 4 It is a structural schematic diagram of a position sensing device based on multi-signal similarity measurement provided by an embodiment of the present invention.
[0127] The position sensing device based on multi-signal similarity measurement includes:
[0128] A target sequence acquisition module 11 is used to acquire the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal;
[0129] The AP position heterogeneous graph updating module 12 is used to add the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal to the RSSI layer, and update the AP position heterogeneous graph; the AP position heterogeneous graph is obtained by connecting the first node isomorphic graph and the second node isomorphic graph;
[0130] A feature vector acquisition module 13 is used to calculate the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal according to a pre-trained hierarchical graph neural network model;
[0131] The similarity calculation module 14 is used to calculate the similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal to obtain the position similarity between the first mobile terminal and the second mobile terminal.
[0132] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the location perception method based on multi-signal similarity measurement described in any of the above embodiments.
[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary hardware platform, and of course can also be implemented entirely by hardware. Based on such an understanding, all or part of the contribution of the technical solution of the present invention to the background technology can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.
[0134] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A location perception method based on multi-signal similarity measurement, characterized in that: include: Acquire an RSSI sequence of the first mobile terminal and an RSSI sequence of the second mobile terminal; Adding the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal to the RSSI layer, and updating the AP position heterogeneous graph; The AP position heterogeneous graph is obtained by connecting the first node isomorphic graph and the second node isomorphic graph; The first node isomorphic graph and the second node isomorphic graph are constructed according to different types of signals; Calculate the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal according to the pre-trained hierarchical graph neural network model; The similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal is calculated to obtain the position similarity between the first mobile terminal and the second mobile terminal.
2. A location sensing method based on multi-signal similarity measurement as claimed in claim 1, characterized in that: The first node isomorphic graph specifically includes: Acquire an RSSI data set of a first signal, where the RSSI data set of the first signal includes at least one RSSI sequence; wherein an RSSI sequence of the first signal is defined as: ; in Indicates nodes, express Middle The signal strength of each node, represents the length of the RSSI sequence of the first signal, ; w is the identifier of the first signal to distinguish other different signals, An RSSI sequence of the i-th first signal in the RSSI data set representing the first signal; A first node isomorphic graph is constructed according to the RSSI data set of the first signal.
3. A location sensing method based on multi-signal similarity measurement as claimed in claim 2, characterized in that: The second node isomorphic graph specifically includes: Acquire an RSSI data set of a second signal, where the RSSI data set of the second signal includes at least one RSSI sequence; wherein an RSSI sequence of the second signal is defined as: ; in Indicates nodes, express Middle The signal strength of each node, represents the length of the RSSI sequence of the second signal, ; b is the identifier of the second signal to distinguish other different signals, An RSSI sequence of the jth second signal in the RSSI data set representing the second signal; A second node isomorphic graph is constructed according to the RSSI data set of the second signal.
4. A location sensing method based on multi-signal similarity measurement as claimed in claim 1, characterized in that: The AP position heterogeneous graph specifically includes: Use the kth mobile terminal node Representing an RSSI sequence, constructing an RSSI sequence layer from the RSSI sequences in all the data sets, and constructing an AP position heterogeneous graph using the first node isomorphic graph and the second node isomorphic graph; Among them, if the node signal strength in the RSSI sequence exists , then The i-th node in the graph isomorphic to the node Connect the edges in the AP position heterogeneous graph The weights are defined as follows: ; Represents the mobile terminal node represents the strength of the ith signal in the RSSI sequence, n represents the length of the RSSI sequence, .
5. A location sensing method based on multi-signal similarity measurement as claimed in claim 1, characterized in that: The pre-trained hierarchical graph neural network model specifically includes: Inputting the first node isomorphic graph into a first graph convolutional network, and using a graph self-supervised learning method to obtain a feature vector of the first node; Inputting the second node isomorphic graph into a second graph convolutional network, and using a graph self-supervised learning method to obtain a feature vector of the second node; Based on the learned feature vector of the first node, inputting the first node isomorphic graph into a third graph convolutional network to obtain a feature vector of the third node; Based on the learned feature vector of the second node, inputting the second node isomorphic graph into a fourth graph convolutional network to obtain a feature vector of the fourth node; Obtaining a feature vector of each RSSI sequence by feature aggregation at an RSSI sequence layer according to the AP position heterogeneous graph, the feature vector of the third node, and the feature vector of the fourth node; Calculate the similarity of the feature vectors of two RSSI sequences; The hierarchical graph neural network model is trained by constructing RSSI pairs.
6. A location sensing method based on multi-signal similarity measurement as claimed in claim 5, characterized in that: The feature aggregation is calculated as follows: , in, represents the kth mobile terminal node, is the i-th node in the node isomorphism graph, Represents a mobile terminal node Neighbor nodes in the AP location heterogeneous graph, Represents edge The weight of is a learnable parameter, It is the node in the RSSI sequence The characteristic vector of The feature vector representing the RSSI sequence represented by the kth mobile terminal node.
7. A location sensing method based on multi-signal similarity measurement as claimed in claim 3, characterized in that: The formula for calculating the similarity is: ; in and They are respectively the characteristic vector of the RSSI sequence of the first mobile terminal and the characteristic vector of the RSSI sequence of the second mobile terminal.
8. A location sensing device based on multi-signal similarity measurement, characterized in that: include: A target sequence acquisition module, used to acquire the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal; An AP position heterogeneous map updating module is used to add the RSSI sequence of the first mobile terminal and the RSSI sequence of the second mobile terminal to the RSSI layer, and update the AP position heterogeneous map; The AP position heterogeneous graph is obtained by connecting the first node isomorphic graph and the second node isomorphic graph; The first node isomorphic graph and the second node isomorphic graph are constructed according to different types of signals; A feature vector acquisition module is used to calculate the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal according to a pre-trained hierarchical graph neural network model; The similarity calculation module is used to calculate the similarity between the feature vector of the RSSI sequence of the first mobile terminal and the feature vector of the RSSI sequence of the second mobile terminal to obtain the position similarity between the first mobile terminal and the second mobile terminal.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the location perception method based on multi-signal similarity measurement according to any one of claims 1 to 7.
