Internet of Things anti-interference positioning method and training method of anti-interference positioning model

Abstract

The embodiment of the invention relates to the technical field of Internet of Things positioning, and provides an Internet of Things anti-interference positioning method and a training method of an anti-interference positioning model, and the method comprises the steps: respectively constructing a communication graph and an interference graph corresponding to all first network nodes; fusing the communication graph and the interference graph to obtain a fused topological graph; performing spatio-temporal joint modeling based on the fused topological graph and the label communication data of each time slot corresponding to the fused topological graph, and determining a target cooperative node participating in positioning from all the first network nodes; based on the label communication data corresponding to the target cooperative node, determining a target anti-interference positioning feature corresponding to the to-be-positioned label; inputting the target anti-interference positioning features into an anti-interference positioning model to obtain a positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is obtained by training based on a sample reward value and a sample penalty value. According to the embodiment of the invention, the positioning efficiency of the label can be improved, and the flexible expansion of the network topology is ensured through the dynamically optimized topological structure.

Description

Technical Field

[0001] This application relates to the field of IoT positioning technology, and in particular to an IoT anti-interference positioning method and a training method for an anti-interference positioning model. Background Technology

[0002] Tag positioning is an important application scenario for passive IoT. The working principle of tag positioning is based on the principle of radio wave reflection. A receiver captures the weak signal reflected back from the passive tag, and then, combined with the multipath propagation characteristics, a positioning algorithm is used to estimate the tag's position.

[0003] Currently, deep learning-based localization algorithms are gradually becoming a research hotspot in this field. These algorithms not only improve localization accuracy but also enhance the system's adaptability and robustness. However, deep learning-based localization algorithms suffer from the following problems: multiple reader polling and inventory management lead to wasted communication resources; the localization cycle is long, resulting in low localization efficiency; and traditional multi-node systems have fixed topologies, requiring complex array designs in advance to achieve optimal coverage, resulting in high costs, poor flexibility, and difficulty adapting to dynamic interference environments. Therefore, improving the localization efficiency and flexibility of passive tags is a pressing technical problem that needs to be solved. Summary of the Invention

[0004] This application provides an IoT anti-interference positioning method and an anti-interference positioning model training method to solve the problem of low positioning efficiency and flexibility of passive tags in the prior art.

[0005] In a first aspect, embodiments of this application provide an IoT anti-interference positioning method, applied to a management device, the method comprising the following steps.

[0006] Construct communication graphs and interference graphs for all first network nodes respectively; the communication graphs are used to represent the communication link connections between the first network nodes; the interference graphs are used to represent the degree of signal interference between the first network nodes. By fusing the communication graph and the interference graph, a fused topology graph is obtained; Spatiotemporal joint modeling is performed based on the fusion topology map and the tag communication data of each time slot corresponding to the fusion topology map, and the target cooperative node participating in the positioning is determined from all first network nodes; Based on the tag communication data corresponding to the target cooperative node, the target anti-interference positioning features corresponding to the tag to be located are determined. The target anti-interference positioning features are input into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on sample reward value and sample penalty value.

[0007] In one embodiment, constructing a communication graph corresponding to all first network nodes includes: For any two first network nodes, determine the probability of a successful link between the two first network nodes based on the physical distance between them. Taking each of the first network nodes as nodes, and in the case where there is an edge between any two first network nodes, the success probability of the link between the two first network nodes is determined as the communication weight of the edge between the two first network nodes, thus obtaining the communication graph corresponding to all the first network nodes.

[0008] In one embodiment, constructing the interference graph corresponding to all first network nodes includes: For any two first network nodes, based on the tag communication data corresponding to each first network node, determine the number of tag collection repetitions corresponding to each first network node; based on the number of tag collection repetitions corresponding to each of the two first network nodes, determine the tag collection repetition rate corresponding to the two first network nodes. By taking each of the first network nodes as nodes and using the label collection repetition rate between any two first network nodes as the interference weight of the edge between the two first network nodes, the interference graph corresponding to all first network nodes is obtained.

[0009] In one embodiment, fusing the communication graph and the interference graph to obtain a fused topology graph includes: For any two nodes, the fusion weight of the edge is determined based on the communication weight of the edge in the communication graph and the interference weight of the edge in the interference graph. By superimposing the nodes in the communication graph and the interference graph, the fused nodes in the fused topology graph are obtained; The fusion topology graph is constructed based on the fusion weights of all fusion nodes and the edges between any two fusion nodes.

[0010] In one embodiment, the step of performing spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, and determining the target cooperative node participating in the localization from all first network nodes, includes: Based on the fused topology map and tag communication data of each time slot, the graph state representation corresponding to each time slot is determined; the graph state representation is used to characterize the network state of the current time slot. Extract the spatial and temporal features corresponding to each of the graph state representations; By fusing the spatial and temporal features corresponding to each graph state representation, a spatiotemporal joint feature corresponding to each graph state representation is obtained; the spatiotemporal joint feature is used to characterize the global state of each fused node in the graph state representation in the current time slot; Based on the spatiotemporal joint features corresponding to any two fusion nodes, determine the collaborative attention score between any two fusion nodes; Based on the collaborative attention score and physical distance between any two fusion nodes, determine the final collaboration matrix between any two fusion nodes; Based on the final cooperation matrix, the target cooperation nodes participating in the localization are determined from all the first network nodes.

[0011] In one embodiment, extracting the spatial features corresponding to each of the graph state representations includes: Based on the fusion weights of all edges in the graph state representation of each time slot, the adjacency matrix corresponding to each graph state representation is determined; Based on the adjacency matrix and identity matrix corresponding to each graph state representation, determine the self-connection matrix corresponding to each graph state representation; Determine the degree matrix corresponding to each of the self-connection matrices; Based on each of the self-connection matrices and the corresponding degree matrices, determine the normalized adjacency matrix corresponding to each of the graph state representations; Based on the normalized adjacency matrices and graph convolutional networks, spatial features corresponding to the graph state representations are extracted.

[0012] In one embodiment, extracting the temporal features corresponding to each of the graph state representations includes: Based on the temporal convolutional network and the node feature sequences corresponding to each time window in each graph state representation, the temporal features corresponding to each graph state representation are extracted.

[0013] In one embodiment, determining the target cooperative node participating in the localization from all first network nodes based on the final cooperation matrix includes: Based on the comparison results between the collaborative element values ​​corresponding to any two fusion nodes in the final collaborative matrix and the first preset threshold, all collaborative nodes are determined. Based on the final cooperation matrix, the sum of the cooperation strengths of each fusion node with other fusion nodes is determined; All collaborative nodes corresponding to the maximum total collaborative strength are identified as the target collaborative nodes for location services.

[0014] In one embodiment, the tag communication data includes an initial phase or an end phase; The step of determining the target anti-interference positioning features corresponding to the tag to be located based on the tag communication data corresponding to the target cooperative node includes: Clustering is performed on the target phase of the target cooperative node at each frequency point to obtain the main phase interval of the target cooperative node at each frequency point; the target phase includes the initial phase or the ending phase. The target frequency is determined based on the main phase intervals corresponding to each of the adjacent frequency points; Based on the tag communication data corresponding to each tag to be located under the target frequency point, the target anti-interference positioning features corresponding to each tag to be located are determined.

[0015] In one embodiment, determining the target frequency point based on the main phase intervals corresponding to adjacent frequency points includes: The phase difference between adjacent frequency points is determined based on the main phase intervals corresponding to each adjacent frequency point. The correction triggering result is determined based on the phase difference and phase distribution corresponding to adjacent frequency points; Based on the correction triggering results, the target correction phase corresponding to each frequency point is determined; The target frequency is determined based on the target correction phase corresponding to each frequency point.

[0016] In one embodiment, determining the target correction phase corresponding to each frequency point based on the correction triggering result includes: If the correction trigger result is a trigger correction and the target phase is within the first phase interval, the difference between the target phase and the first preset phase value is determined as the target correction phase. If the correction trigger result is a trigger correction and the target phase is within the second phase interval, determine the number of first phases of the target cooperative node in the third phase interval and the number of second phases in the fourth phase interval at each frequency point; if the target phase satisfies the first preset condition or the second preset condition, determine the difference between the first preset phase value and the target phase as the target correction phase; The first preset condition is that the target phase is in the third phase interval, and the number of the second phase is greater than the number of the first phase; The second preset condition is that the target phase is in the fourth phase interval, and the number of the first phase is greater than the number of the second phase.

[0017] In one embodiment, determining the target frequency point based on the target correction phase corresponding to each of the frequency points includes: Based on the target correction phase corresponding to each frequency point, a phase stability coefficient corresponding to each frequency point is determined; the phase stability coefficient is used to characterize the degree of phase fluctuation corresponding to each frequency point. The frequency point corresponding to the phase stability coefficient that is greater than the first preset threshold is determined as the target frequency point.

[0018] In one embodiment, the tag communication data further includes the RSSI of each tag to be located; The step of determining the target anti-interference positioning features corresponding to each tag to be located based on the tag communication data corresponding to each tag to be located under the target frequency point includes: Based on the initial and final phases of each tag to be located at the target frequency, the average phase value of each tag to be located is determined. Based on the average phase value of all tags to be located, the frequency point phase slope corresponding to the target frequency point is determined; the frequency point phase slope is used to characterize the distance between the tag to be located and the corresponding target cooperative node. Based on the RSSI, initial phase, and final phase of each of the tags to be located, a tag vector is constructed. Based on the Euclidean distance between the target label vector and the reference label vector, the label similarity between each target label and the reference label is determined; the reference label vector includes the RSSI, initial phase, and final phase corresponding to the reference label. Based on the node identifier, target frequency, power, RSSI, initial phase, ending phase, frequency phase slope, and tag similarity of each tag to be located, the target anti-interference positioning features corresponding to each tag to be located are determined.

[0019] Secondly, embodiments of this application provide a training method for an anti-interference localization model, comprising the following steps.

[0020] Construct a reward and penalty loss function based on sample reward value, sample penalty value, and error penalty factor; The initial anti-interference positioning model is trained based on the anti-interference positioning sample features and the reward / penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method described in the first aspect.

[0021] Thirdly, embodiments of this application provide an Internet of Things (IoT) anti-interference positioning device, which includes the following modules.

[0022] The first construction module is used to construct communication graphs and interference graphs corresponding to all first network nodes respectively; the communication graph is used to represent the communication link connection relationship between the first network nodes; the interference graph is used to represent the signal interference degree between the first network nodes. The fusion module is used to fuse the communication graph and the interference graph to obtain a fused topology graph; The first determining module is used to perform spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, and to determine the target cooperative node participating in the positioning from all the first network nodes; The second determining module is used to determine the target anti-interference positioning features corresponding to the tag to be located based on the tag communication data corresponding to the target cooperative node. The localization module is used to input the anti-interference localization features of the target into the anti-interference localization model to obtain the localization prediction result output by the anti-interference localization model; the anti-interference localization model is trained based on sample reward values ​​and sample penalty values.

[0023] Fourthly, embodiments of this application provide a training device for an anti-interference positioning model, comprising the following modules.

[0024] The second building module is used to construct a reward and penalty loss function based on the sample reward value, sample penalty value, and error penalty factor; The training module is used to train the initial anti-interference positioning model based on the anti-interference positioning sample features and the reward / penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method described in the first aspect.

[0025] Fifthly, embodiments of this application provide a management device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the IoT anti-interference positioning method as described in the first aspect above, or the anti-interference positioning model training method as described in the second aspect.

[0026] Sixthly, embodiments of this application provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the IoT anti-interference positioning method as described in the first aspect above, or the anti-interference positioning model training method as described in the second aspect.

[0027] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the IoT anti-interference positioning method as described in the first aspect above, or the anti-interference positioning model training method as described in the second aspect.

[0028] The IoT anti-interference positioning method and anti-interference positioning model training method provided in this application embodiment construct a communication graph representing the communication link connection relationship between all first network nodes and an interference graph representing the signal interference level between first network nodes. The communication graph and interference graph are then fused to obtain a fused topology graph. Based on the spatiotemporal joint features corresponding to the fused topology graph of each time slot, target cooperative nodes participating in positioning are determined from all first network nodes. Based on the tag communication data of the target cooperative nodes, target anti-interference positioning features corresponding to the tag to be positioned are determined. The target anti-interference positioning features are input into the anti-interference positioning model trained based on sample reward values ​​and sample penalty values ​​to obtain the positioning prediction result output by the anti-interference positioning model. In this application embodiment, spatiotemporal joint optimization is performed through the fused topology graph of each time slot, and the network state is analyzed in real time. Only necessary target cooperative nodes are activated to participate in positioning, avoiding simultaneous operation of all first network nodes, reducing communication resource waste, shortening the positioning cycle, and thus improving the positioning efficiency of tags. Simultaneously, a dynamically optimized topology structure is used instead of the traditional fixed topology structure to adapt to real-time changes in network state and ensure flexible expansion of the network topology. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating the IoT anti-interference positioning method provided in the embodiments of this application.

[0031] Figure 2 This is a flowchart illustrating the training method of the anti-interference localization model provided in the embodiments of this application.

[0032] Figure 3 This is a schematic diagram of the structure of the IoT anti-interference positioning device provided in the embodiments of this application.

[0033] Figure 4 This is a schematic diagram of the structure of the training device for the anti-interference positioning model provided in the embodiments of this application.

[0034] Figure 5 This is a schematic diagram of the structure of the management device provided in the embodiments of this application. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] Ultra-high frequency (UHF) passive RFID (Radio Frequency Identification) is widely used in retail, logistics, asset management, and other fields. It features zero power consumption, low cost, and easy deployment, making it a fundamental enabling technology for the Internet of Things (IoT). Traditional RFID system architecture consists of two parts: passive tags and receivers. Addressing the limitations of UHF passive RFID in terms of range, low system efficiency, and poor positioning accuracy, the industry began researching system optimization schemes with a new transmit / receive separation architecture in 2014 to achieve network deployment and efficiency improvements for UHF passive RFID. The new UHF passive RFID system adopts a separate forward link architecture, deploying the forward and reverse links separately at the receiver and exciter. This decoupling of the links improves the spatial isolation between transmitted and received signals, enhances the receiver's sensitivity to tag-reflected signals, and significantly expands the system's coverage area; this is called separate passive RFID. To further expand coverage and achieve regional and cross-process business management, passive IoT is introduced to expand system capabilities. New protocols, new architectures, new tags, and new labels leverage network infrastructure to build "end-to-end" coverage, enabling full lifecycle management of assets. This can be used in single / multi-gate areas in scenarios such as large-scale warehousing and asset management to achieve automated, large-scale, and efficient inventory and location tracking of goods.

[0037] To address the issues of low positioning efficiency and flexibility of existing passive tag technologies, this application provides an anti-interference positioning method for the Internet of Things (IoT), applied to management devices. Figure 1 This is a flowchart illustrating the IoT anti-interference positioning method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes steps 110 to 150.

[0038] Step 110: Construct communication graphs and interference graphs for all first network nodes respectively; the communication graphs are used to characterize the communication link connection relationships between the first network nodes; the interference graphs are used to characterize the signal interference levels between the first network nodes.

[0039] Specifically, the first network node is a passive IoT transceiver deployed within the monitoring area. For example, the first network node is an RFID reader / writer that integrates the functions of a receiver and an exciter. There are multiple first network nodes.

[0040] After receiving a location request command from the user, the management device controls all first network nodes to send excitation signals to the corresponding passive tag. Upon receiving the excitation signal, the tag reflects the signal back to the first network node that sent the excitation signal, returning its tag communication data. All first network nodes upload the tag communication data returned by their tags to the management device. Based on the communication link connections between all first network nodes, the management device constructs a communication graph, which determines which first network nodes can effectively transmit data. Simultaneously, based on the tag communication data uploaded by all first network nodes, an interference graph is constructed, reflecting the degree of signal interference between the first network nodes. This interference graph identifies which first network nodes would interfere with each other if operating simultaneously, resulting in wasted resources.

[0041] It should be noted that after receiving the tag communication data uploaded by all the first network nodes, the management device performs preprocessing operations such as filtering and normalization on the tag communication data to obtain the node feature vector corresponding to the standardized tag communication data.

[0042] Furthermore, the tag communication data includes the tag EPC (Electronic Product Code), frequency, transmit power, RSSI (Received Signal Strength Indication), and phase, wherein the phase includes an initial phase and an ending phase. Each first network node can collect tag communication data of at least one tag, and can also collect tag communication data of tags for which no tags have been collected. In this embodiment of the invention, the number of tags corresponding to the tag communication data collected by each first network node is not limited.

[0043] Step 120: Merge the communication graph and the interference graph to obtain a fused topology graph.

[0044] Specifically, after constructing the communication graph and the interference graph, the communication graph and the interference graph are merged to obtain a fused topology graph, which quantifies and unifies the communication capabilities and interference risks in the topology structure.

[0045] Step 130: Perform spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, and determine the target cooperative node participating in the positioning from all first network nodes; Specifically, after determining the fusion topology, continuous time is discretized into time slots, that is, each t second is defined as a time slot. Spatiotemporal joint feature modeling is performed based on the fusion topology of each time slot and the node feature vectors corresponding to the tag communication data collected by all fusion nodes in the fusion topology. This facilitates the selection of necessary target cooperative nodes for the current time slot from all first network nodes to participate in the subsequent positioning of passive tags, avoiding polling all first network nodes and wasting resources.

[0046] Step 140: Based on the tag communication data corresponding to the target cooperative node, determine the target anti-interference positioning features corresponding to the tag to be located.

[0047] Specifically, after identifying the target collaborative node, anti-interference frequency points are filtered based on the tag communication data uploaded by the target collaborative node to avoid interference frequency bands, thereby obtaining the target anti-interference positioning features corresponding to the tag to be located and improving the accuracy of subsequent positioning.

[0048] Step 150: Input the target anti-interference positioning features into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on sample reward value and sample penalty value.

[0049] Specifically, after determining the anti-interference positioning features of the target, the anti-interference positioning features are input into the anti-interference positioning model. Through the anti-interference positioning model's ability to identify anti-interference features, the positioning prediction result output by the anti-interference positioning model is obtained.

[0050] The IoT anti-interference positioning method provided in this application constructs a communication graph representing the communication link connection relationship between all first network nodes and an interference graph representing the signal interference level between first network nodes. The communication graph and interference graph are then fused to obtain a fused topology graph. Based on the spatiotemporal joint features corresponding to the fused topology graph of each time slot, target cooperative nodes participating in positioning are determined from all first network nodes. Based on the tag communication data of the target cooperative nodes, target anti-interference positioning features corresponding to the tag to be positioned are determined. These target anti-interference positioning features are input into an anti-interference positioning model trained based on sample reward and penalty values ​​to obtain the positioning prediction result output by the anti-interference positioning model. In this application embodiment, spatiotemporal joint optimization is performed through the fused topology graph of each time slot, and the network state is analyzed in real time. Only necessary target cooperative nodes are activated to participate in positioning, avoiding simultaneous operation of all first network nodes, reducing communication resource waste, shortening the positioning cycle, and thus improving the positioning efficiency of tags. Simultaneously, a dynamically optimized topology structure is used instead of the traditional fixed topology structure to adapt to real-time changes in network state and ensure flexible expansion of the network topology.

[0051] In one embodiment, constructing a communication graph corresponding to all first network nodes includes: For any two first network nodes, determine the success probability of the link between the two first network nodes based on the physical distance between them; Taking each of the first network nodes as nodes, the communication weight of the edge between any two first network nodes is determined based on the success probability of the link between any two first network nodes, thus obtaining the communication graph corresponding to all first network nodes.

[0052] Specifically, all first network nodes are designated as nodes V in the communication graph G1, which is an undirected graph. When a communication link exists between any two first network nodes, the success probability of the link between them is calculated using equation (1) based on the physical distance and reference distance between the two nodes. Equation (1) is: .

[0053] Among them, P link (i,j) represents the success probability of the link between the first network node i and the first network node j, d ij d represents the physical distance between the first network node i and the first network node j. ref This represents the reference distance between the first network node i and the first network node j, which is typically 1 meter.

[0054] Next, the success probability of the link is compared with a third preset threshold. If the success probability is greater than or equal to the third preset threshold, an edge E1 is constructed between the first network node i and the first network node j. This edge is undirected, and the success probability is used as the communication weight of the edge. If the success probability is less than the third preset threshold, there is no edge between the first network node i and the first network node j. Repeating the above operation, a communication graph can be constructed, which is represented as G1(V, E1).

[0055] In this embodiment of the application, a communication graph is constructed to objectively reflect the link communication capabilities between nodes and eliminate unreliable links.

[0056] In one embodiment, constructing the interference graph corresponding to all first network nodes includes: For any two first network nodes, based on the tag communication data corresponding to each first network node, determine the number of tag collection repetitions corresponding to each first network node; based on the number of tag collection repetitions corresponding to each of the two first network nodes, determine the tag collection repetition rate corresponding to the two first network nodes. By taking each of the first network nodes as nodes and using the label collection repetition rate between any two first network nodes as the interference weight of the edge between the two first network nodes, the interference graph corresponding to all first network nodes is obtained.

[0057] Specifically, after receiving the tag communication data uploaded by all first network nodes, all first network nodes are treated as nodes V of the interference graph G2, which is an undirected graph. When there is a communication link between any two first network nodes, an edge E2 is constructed between these two first network nodes. This edge is undirected. Simultaneously, the number of tag EPCs in the tag communication data collected by each of the two first network nodes is counted; this number of tag EPCs represents the number of tags collected. Then, based on the tag EPCs, the number of tag collection overlaps between the two first network nodes is determined. Next, using equation (2), based on the number of tag collection overlaps between the two first network nodes and the number of tags collected by each of the two first network nodes, the tag collection overlap rate between the two first network nodes is determined, and this tag collection overlap rate is used as the interference weight of the edge to quantify signal interference. Equation (2) is: .

[0058] Among them, P overlap NUM represents the tag collection repetition rate corresponding to the first network node i and the first network node j. ij NUM represents the number of duplicate label collections between the first network node i and the first network node j. i NUM represents the number of tags collected by the first network node i. j This represents the number of tags collected by the first network node j. Repeating the above steps will construct the interference map, denoted as G2(V, E2).

[0059] In one embodiment, fusing the communication graph and the interference graph to obtain a fused topology graph includes: For any two nodes, the fusion weight of the edge is determined based on the communication weight of the edge in the communication graph and the interference weight of the edge in the interference graph. By superimposing the nodes in the communication graph and the interference graph, the fused nodes in the fused topology graph are obtained; The fusion topology graph is constructed based on the fusion weights of all fusion nodes and the edges between any two fusion nodes.

[0060] Specifically, since the nodes corresponding to the communication graph and the interference graph are the same, the nodes in the communication graph and the interference graph are superimposed to obtain the fused nodes in the fused topology graph. Then, using equation (3), the fusion weights of the edges in the fused topology graph are calculated based on the communication weights and interference weights corresponding to each edge. Equation (3) is: .

[0061] Where P represents the fusion weight of the edge, α represents the coefficient of the link success probability (i.e., communication weight), and β represents the coefficient of the tag collection repetition rate (i.e., interference weight). The fusion topology can be represented as G(V, E1∪E2).

[0062] In one embodiment, the step of performing spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, and determining the target cooperative node participating in the localization from all first network nodes, includes: Based on the fused topology map and tag communication data of each time slot, the graph state representation corresponding to each time slot is determined; the graph state representation is used to characterize the network state of the current time slot. Extract the spatial and temporal features corresponding to each of the graph state representations; By fusing the spatial and temporal features corresponding to each graph state representation, a spatiotemporal joint feature corresponding to each graph state representation is obtained; the spatiotemporal joint feature is used to characterize the global state of each fused node in the graph state representation in the current time slot; Based on the spatiotemporal joint features corresponding to any two fusion nodes, determine the collaborative attention score between any two fusion nodes; Based on the collaborative attention score and physical distance between any two fusion nodes, determine the final collaboration matrix between any two fusion nodes; Based on the final cooperation matrix, the target cooperation nodes participating in the localization are determined from all the first network nodes.

[0063] Specifically, by combining the fused topology graph of each time slot and the node feature vectors corresponding to the tag communication data, a graph state representation for each time slot is constructed, which is denoted as G. t = (V, E) t X t ), where G t This represents the graph state representation of time slot t, where V represents the fusion node, and E... t X represents the fusion weight for time slot t. t This represents the node feature vector corresponding to the tag communication data in time slot t.

[0064] Next, Graph Convolutional Networks (GCNs) are used to extract spatial features from the graph state representation, and Temporal Convolutional Networks (TCNs) are used to extract temporal features. The spatial features characterize the spatial dependencies between fused nodes, while the temporal features characterize the changes in the fused node state over time. Finally, the spatial and temporal features are concatenated to obtain the spatiotemporal joint features corresponding to the graph state representation. These spatiotemporal joint features characterize the global state of the fused node in the current time slot.

[0065] Then, using equation (4), the collaborative attention score corresponding to the spatiotemporal joint features of any two fusion nodes is calculated. Equation (4) is: .

[0066] Among them, H i H represents the spatiotemporal joint feature corresponding to fusion node i. j Let W represent the spatiotemporal joint feature corresponding to fusion node j, W represent the weight, ReLU represent the activation function, and Attention represent the collaborative attention score between fusion node i and fusion node j.

[0067] Then, using Equation (5), the final collaboration matrix is ​​calculated based on the collaborative attention score and the physical distance between the two fusion nodes. Equation (5) is: .

[0068] in, Let represent the value of the collaborative element in the i-th row and j-th column of the final collaborative matrix, and let distance(i,j) represent the physical distance between fused node i and fused node j. This represents the attenuation term, which helps avoid unnecessary collaboration between distant devices. Repeating this process until all fusion nodes have been traversed yields the final collaboration matrix.

[0069] Subsequently, based on the final cooperation matrix, a node cooperation strategy is planned to select highly cooperative and necessary first network nodes from all first network nodes as target cooperative nodes. These target cooperative nodes form a dynamic topology structure that adapts to the current network state or environmental changes, so as to participate in subsequent tag localization, avoid polling each node one by one, shorten the localization cycle, and thus improve localization efficiency.

[0070] In one embodiment, extracting the spatial features corresponding to each of the graph state representations includes: Based on the fusion weights of all edges in the graph state representation of each time slot, the adjacency matrix corresponding to each graph state representation is determined; Based on the adjacency matrix and identity matrix corresponding to each graph state representation, determine the self-connection matrix corresponding to each graph state representation; Determine the degree matrix corresponding to each of the self-connection matrices; Based on each of the self-connection matrices and the corresponding degree matrices, determine the normalized adjacency matrix corresponding to each of the graph state representations; Based on the normalized adjacency matrices and graph convolutional networks, spatial features corresponding to the graph state representations are extracted.

[0071] Specifically, after determining the fusion topology graph for each time slot, an adjacency matrix A corresponding to this fusion topology graph is constructed. That is, if there is an edge between fusion node i and fusion node j, the fusion weight P[i,j] corresponding to that edge is determined as the corresponding element A[i,j] in the adjacency matrix; if there is no edge between fusion node i and fusion node j, then P[i,j] = A[i,j] = 0. Repeating the above operation determines all element values ​​in the adjacency matrix. Then, a self-join is added to this adjacency matrix, resulting in a self-join matrix as follows: ,in, Let represent the self-join matrix, and I represent the identity matrix. Then, the degree matrix of the self-join matrix is ​​calculated. Based on the degree matrix and the self-connection matrix, the normalized adjacency matrix is ​​determined, which is expressed as follows: ,in, Let represent the normalized adjacency matrix. Then, using equation (6), this normalized adjacency matrix is ​​input into the graph convolutional network to obtain the spatial features corresponding to the graph state representation. Equation (6) is: .

[0072] in, W represents the spatial matrix of time slot t. space This represents the weight matrix of a graph convolutional network. This represents the activation function.

[0073] In one embodiment, extracting the temporal features corresponding to each of the graph state representations includes: Based on the temporal convolutional network and the node feature sequences corresponding to each time window in each graph state representation, the temporal features corresponding to each graph state representation are extracted.

[0074] Specifically, the node feature sequence corresponding to the time window is extracted from the node feature vector, and this node feature sequence is represented as {X}. t-k ,……,X t The time window length of the node feature sequence is k. Then, the node feature sequence is input into a temporal convolutional network, and using equation (7), the temporal features output by the temporal convolutional network are obtained. Equation (7) is: .

[0075] in, X represents the temporal characteristics of time slot t. t-i Let represent the feature vector of the i-th node in the node feature sequence, and 0≤i≤k, wi represent the convolution kernel weights of the temporal convolutional network, and b represent the bias term.

[0076] In one embodiment, determining the target cooperative node participating in the localization from all first network nodes based on the final cooperation matrix includes: Based on the comparison results between the collaborative element values ​​corresponding to any two fusion nodes in the final collaborative matrix and the first preset threshold, all collaborative nodes are determined. Based on the final cooperation matrix, the sum of the cooperation strengths of each fusion node with other fusion nodes is determined; All collaborative nodes corresponding to the maximum total collaborative strength are identified as the target collaborative nodes for location services.

[0077] Specifically, after determining the final cooperation matrix, if the value of a cooperation element in the final cooperation matrix is ​​greater than a first preset threshold, it indicates that the two fusion nodes corresponding to that cooperation element value can cooperate. All fusion nodes are traversed to determine all cooperation nodes. Then, the row sum of each fusion node is calculated, that is, the sum of C[i,:] is calculated, where ":" is a wildcard, and C[i,:] represents any cooperation element value in the i-th row of the final cooperation matrix. This row sum represents the total cooperation intensity of fusion node i with other fusion nodes. The larger the row sum, the more tasks fusion node i undertakes in the entire fusion topology. All row sums are sorted in descending order. The earlier the sorted position, the higher the cooperation degree of all cooperation nodes corresponding to that cooperation intensity sum, and resources can be allocated preferentially to cooperation nodes with high cooperation degrees. Therefore, all cooperation nodes corresponding to the maximum cooperation intensity sum can be determined as target cooperation nodes for localization.

[0078] In one embodiment, determining the target anti-interference positioning features corresponding to the tag to be located based on the tag communication data corresponding to the target cooperative node includes: Clustering is performed on the target phase of the target cooperative node at each frequency point to obtain the main phase interval of the target cooperative node at each frequency point; the target phase includes the initial phase or the ending phase. The target frequency is determined based on the main phase intervals corresponding to each of the adjacent frequency points; Based on the tag communication data corresponding to each tag to be located under the target frequency point, the target anti-interference positioning features corresponding to each tag to be located are determined.

[0079] Specifically, after identifying the target cooperative nodes, cluster analysis is performed on the target phase in the tag communication data collected by the target cooperative nodes to automatically identify the main phase interval of all target cooperative nodes at each frequency point. This target phase includes the initial phase or the ending phase. Then, based on the main phase intervals corresponding to adjacent frequency points, anti-interference frequency points are selected to avoid interference bands and obtain anti-interference and phase-stable target frequency points. Next, tag communication data located at the target frequency points are selected, and target anti-interference positioning features for subsequent tag positioning are extracted based on the selected tag communication data.

[0080] In this embodiment, tag communication data at interfered frequency points are removed to provide high-quality target anti-interference positioning features for subsequent tag positioning, thereby improving the accuracy of tag positioning.

[0081] In one embodiment, determining the target frequency point based on the main phase intervals corresponding to adjacent frequency points includes: The phase difference between adjacent frequency points is determined based on the main phase intervals corresponding to each adjacent frequency point. The correction triggering result is determined based on the phase difference and phase distribution corresponding to adjacent frequency points; Based on the correction triggering results, the target correction phase corresponding to each frequency point is determined; The target frequency is determined based on the target correction phase corresponding to each frequency point.

[0082] Specifically, after determining the main phase interval corresponding to each frequency point, the phase difference between adjacent frequency points is calculated. If the phase difference is greater than 90° and the phase distribution exhibits a bimodal distribution, it indicates a significant jump, and the correction trigger result is determined to be a triggered correction. Then, under the condition that the correction trigger result is a triggered correction, the initial phase and ending phase of each tag under that frequency point are corrected respectively to obtain the target correction phase corresponding to each tag under that frequency point. This target correction phase includes either the initial correction phase or the ending correction phase. Then, using the target correction phase corresponding to each frequency point, anti-interference and phase-stable target frequency points are selected from all frequency points.

[0083] In one embodiment, determining the target correction phase corresponding to each frequency point based on the correction triggering result includes: If the correction trigger result is a trigger correction and the target phase is within the first phase interval, the difference between the target phase and the first preset phase value is determined as the target correction phase. If the correction trigger result is a trigger correction and the target phase is within the second phase interval, determine the number of first phases of the target cooperative node in the third phase interval and the number of second phases in the fourth phase interval at each frequency point; if the target phase satisfies the first preset condition or the second preset condition, determine the difference between the first preset phase value and the target phase as the target correction phase; The first preset condition is that the target phase is in the third phase interval, and the number of the second phase is greater than the number of the first phase; The second preset condition is that the target phase is in the fourth phase interval, and the number of the first phase is greater than the number of the second phase.

[0084] For example, taking a first preset phase value of 180°, a first phase interval of (180, 360°), a second phase interval of (0, 180°), a third phase interval of (0, 90°), a fourth phase interval of (90, 180°), and the target phase as the initial phase, when phase correction is triggered, using equation (8), if the initial phase is within the first phase interval, the difference between the initial phase and 180° is calculated, and this difference is the initial correction phase. Equation (8) is: .

[0085] in, Indicates the initial correction phase. Indicates the initial phase.

[0086] If the initial phase is within the second phase interval, the initial phases of each cooperating node at different power and frequency points are statistically analyzed. The number of first phases in the third phase interval and the number of second phases in the fourth phase interval are counted respectively. Using equation (9), if the initial phase is in the third phase interval but the number of second phases is greater than the number of first phases, the difference between 180 and the initial phase in the third phase interval is calculated, and this difference is the initial correction phase. Similarly, if the initial phase is in the fourth phase interval but the number of second phases is less than the number of first phases, the difference between 180 and the initial phase in the fourth phase interval is calculated, and this difference is the initial correction phase. Equation (9) is: .

[0087] Where, N 90-180 N represents the number of second phases in the fourth phase interval. 0-90 This indicates the number of first phases that are in the third phase interval.

[0088] When the target phase is the ending phase, the correction process of the initial phase described above can be referred to, and will not be repeated here in the embodiments of this application.

[0089] In one embodiment, determining the target frequency point based on the target correction phase corresponding to each of the frequency points includes: Based on the target correction phase corresponding to each frequency point, determine the phase stability coefficient corresponding to each frequency point; The frequency point corresponding to the phase stability coefficient that is greater than the second preset threshold is determined as the target frequency point.

[0090] Specifically, after obtaining the target correction phase corresponding to each frequency point, the phase standard deviation at each frequency point is calculated based on all target correction phases at each frequency point. This phase standard deviation is used to characterize the phase fluctuation degree corresponding to each frequency point. Then, using equation (10), the phase stability coefficient corresponding to each frequency point is calculated based on the phase standard deviation at each frequency point. Equation (10) is: .

[0091] in, This represents the phase standard deviation corresponding to frequency point k. This represents the phase stability coefficient corresponding to frequency point k.

[0092] Next, the phase stability coefficient corresponding to each frequency point is compared with the second preset threshold. If the phase stability coefficient is less than or equal to the second preset threshold, it indicates that the phase of the frequency point is unstable. If the phase stability coefficient is greater than the second preset threshold, it indicates that the phase of the frequency point is stable. The frequency point can be identified as the target frequency point to participate in subsequent tag positioning.

[0093] In one embodiment, the tag communication data further includes the RSSI of each tag to be located; The step of determining the target anti-interference positioning features corresponding to each tag to be located based on the tag communication data corresponding to each tag to be located under the target frequency point includes: Based on the initial and final phases of each tag to be located at the target frequency, the average phase value of each tag to be located is determined. Based on the average phase value of all tags to be located, the frequency point phase slope corresponding to the target frequency point is determined; the frequency point phase slope is used to characterize the distance between the tag to be located and the corresponding target cooperative node. Based on the RSSI, initial phase, and final phase of each of the tags to be located, a tag vector is constructed. Based on the Euclidean distance between the target label vector and the reference label vector, the label similarity between each target label and the reference label is determined; the reference label vector includes the RSSI, initial phase, and final phase corresponding to the reference label. Based on the node identifier, target frequency, power, RSSI, initial phase, ending phase, frequency phase slope, and tag similarity of each tag to be located, the target anti-interference positioning features corresponding to each tag to be located are determined.

[0094] Specifically, for each tag to be located collected by the cooperating nodes at the target frequency, the average initial and final phases of each tag at different power levels and the target frequency are calculated. This average is the average phase value of each tag to be located. Then, based on the average phase values ​​of different cooperating nodes at different power levels, a straight line is fitted between all target frequencies and their corresponding average phase values. The slope of this fitted line is determined as the frequency-point phase slope, which reflects the distance characteristics between the tag to be located and the cooperating nodes.

[0095] Next, all tags to be located that are on the same cooperative node, have the same power, and are at the same target frequency are statistically analyzed. Based on the RSSI, initial phase, and final phase of each tag, a tag vector for each tag to be located is constructed. Simultaneously, a reference tag vector is constructed based on the RSSI, initial phase, and final phase of a reference tag. The Euclidean distance between each tag vector to be located and the reference tag vector is calculated, and this Euclidean distance is used as the tag similarity between the tag to be located and the reference tag. Then, based on the node identifier (EPC), target frequency, power, RSSI, initial phase, final phase, frequency phase slope, and tag similarity of each tag to be located, target anti-interference positioning features for each tag to be located are constructed.

[0096] In addition, the anti-interference positioning features of each target are normalized to avoid problems such as unstable positioning prediction caused by excessive differences in the order of magnitude of feature values ​​during the model prediction process.

[0097] This application also provides a method for training an anti-interference localization model. Figure 2 This is a flowchart illustrating the training method of the anti-interference localization model provided in the embodiments of this application, as shown below. Figure 2 As shown, the method includes steps 210 and 220.

[0098] Step 210: Construct a reward and penalty loss function based on the sample reward value, sample penalty value, and error penalty factor.

[0099] Before performing step 210, an initial anti-interference localization model can be constructed using machine learning algorithms or deep learning algorithms. For example, the initial anti-interference localization model can be constructed using random forest algorithms, support vector machines (SVM) algorithms, or feedforward neural networks (FNN). This application embodiment does not limit this.

[0100] Simultaneously, historical tag communication data corresponding to all first network nodes is acquired. Referring to the specific steps for determining the target anti-interference positioning features corresponding to any of the above embodiments, anti-interference positioning sample features are determined based on the historical tag communication data and all first network nodes. Then, these anti-interference positioning sample features and the true tags are used as the training dataset for the initial anti-interference positioning model.

[0101] After constructing the initial anti-interference localization model and determining the training dataset, the training dataset is input into the initial anti-interference localization model to obtain the predicted probability distribution output by the initial anti-interference localization model. The probability of each anti-interference localization sample feature corresponding to the correct label is extracted from this predicted probability distribution, and the correct prediction probability is used as the sample reward value conf. i The sample reward value is used to characterize the model's confidence in the correct classification. Meanwhile, the difference between 1 and the sample reward value is used as the sample penalty value, denoted as 1-conf. i The sample penalty value is used to characterize the uncertainty of the model in classifying the sample.

[0102] Then, the reward-penalty loss function shown in Equation (11) is constructed based on the sample reward value, sample penalty value, and error penalty factor. Equation (11) is: .

[0103] Where Loss represents the reward / penalty loss function, n represents the number of samples, and y i Let represent the true label corresponding to the i-th sample, and , ,and λ represents the error penalty factor, which is used to strengthen the penalty for difficult samples with strong interference and ambiguous features. The default value of λ can be set to 0.2 or 0.25, etc., and this application embodiment does not limit it.

[0104] After each training round, calculate the average prediction confidence of all training samples in that round. The average prediction confidence level is expressed as The update trend of the error penalty factor can also be adjusted by using the average prediction confidence over a preset number of rounds. For example, taking a preset number of rounds as 10 consecutive rounds, the average prediction confidence over these 10 rounds is determined. This average prediction confidence is compared with a fourth preset threshold. If the average prediction confidence is less than the fourth preset threshold, it indicates that the model's overall understanding of sample classification is insufficient, and that most of the samples are difficult to classify due to strong label interference and ambiguous features. In this case, the error penalty factor can be increased by a preset step size to strengthen the penalty for low-confidence samples in the reward-penalty loss function, forcing the initial anti-interference localization model to strengthen the learning of ambiguous features caused by interference. If the average is greater than or equal to the fourth preset threshold, then most of the samples are those corresponding to anti-interference features. In this case, the error penalty factor can be maintained at its default value or appropriately reduced.

[0105] Optionally, the preset step size can be 0.05 or 0.1, etc., and this application embodiment does not limit it.

[0106] Step 220: Train the initial anti-interference positioning model based on the anti-interference positioning sample features and the reward / penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method described in any of the above embodiments.

[0107] Specifically, the initial anti-interference localization model is iteratively trained based on the features of the anti-interference localization sample and the reward / penalty loss function. The reward / penalty loss function is gradually reduced until convergence, at which point the trained anti-interference localization model is obtained. At this point, the anti-interference localization model tends to improve the prediction probability of the correct label.

[0108] The IoT anti-interference positioning device provided in this application is described below. The IoT anti-interference positioning device described below can be referred to in correspondence with the IoT anti-interference positioning method described above.

[0109] This application provides an IoT anti-interference positioning device. Figure 3 This is a schematic diagram of the structure of the IoT anti-interference positioning device provided in the embodiments of this application, as shown below. Figure 3 As shown, the IoT anti-interference positioning device 300 includes: a first construction module 310, a fusion module 320, a first determination module 330, a second determination module 340, and a positioning module 350.

[0110] The first construction module 310 is used to construct communication graphs and interference graphs corresponding to all first network nodes respectively; the communication graph is used to characterize the communication link connection relationship between the first network nodes; the interference graph is used to characterize the signal interference degree between the first network nodes.

[0111] The fusion module 320 is used to fuse the communication graph and the interference graph to obtain a fused topology graph.

[0112] The first determining module 330 is used to perform spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, and to determine the target cooperative node participating in the positioning from all the first network nodes.

[0113] The second determining module 340 is used to determine the target anti-interference positioning features corresponding to the tag to be located based on the tag communication data corresponding to the target cooperative node.

[0114] The positioning module 350 is used to input the anti-interference positioning features of the target into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on the sample reward value and the sample penalty value.

[0115] The IoT anti-interference positioning device provided in this application constructs a communication graph representing the communication link connection relationship between all first network nodes and an interference graph representing the signal interference level between first network nodes. The communication graph and interference graph are then fused to obtain a fused topology graph. Based on the spatiotemporal joint features corresponding to the fused topology graph of each time slot, target cooperative nodes participating in positioning are determined from all first network nodes. Based on the tag communication data of the target cooperative nodes, target anti-interference positioning features corresponding to the tag to be positioned are determined. These target anti-interference positioning features are input into an anti-interference positioning model trained based on sample reward and penalty values ​​to obtain the positioning prediction result output by the anti-interference positioning model. In this application embodiment, spatiotemporal joint optimization is performed through the fused topology graph of each time slot, and the network state is analyzed in real time. Only necessary target cooperative nodes are activated to participate in positioning, avoiding simultaneous operation of all first network nodes, reducing communication resource waste, shortening the positioning cycle, and thus improving the positioning efficiency of tags. Simultaneously, a dynamically optimized topology structure is used instead of the traditional fixed topology structure to adapt to real-time changes in network state and ensure flexible expansion of the network topology.

[0116] Optionally, the first building module 310 is specifically used for: For any two first network nodes, determine the probability of a successful link between the two first network nodes based on the physical distance between them. Taking each of the first network nodes as nodes, and in the case where there is an edge between any two first network nodes, the success probability of the link between the two first network nodes is determined as the communication weight of the edge between the two first network nodes, thus obtaining the communication graph corresponding to all the first network nodes.

[0117] Optionally, the first building module 310 is specifically used for: For any two first network nodes, based on the tag communication data corresponding to each first network node, determine the number of tag collection repetitions corresponding to each first network node; based on the number of tag collection repetitions corresponding to each of the two first network nodes, determine the tag collection repetition rate corresponding to the two first network nodes. By taking each of the first network nodes as nodes and using the label collection repetition rate between any two first network nodes as the interference weight of the edge between the two first network nodes, the interference graph corresponding to all first network nodes is obtained.

[0118] Optionally, the fusion module 320 is specifically used for: For any two nodes, the fusion weight of the edge is determined based on the communication weight of the edge in the communication graph and the interference weight of the edge in the interference graph. By superimposing the nodes in the communication graph and the interference graph, the fused nodes in the fused topology graph are obtained; The fusion topology graph is constructed based on the fusion weights of all fusion nodes and the edges between any two fusion nodes.

[0119] Optionally, the first determining module 330 is specifically used for: Based on the fused topology map and tag communication data of each time slot, the graph state representation corresponding to each time slot is determined; the graph state representation is used to characterize the network state of the current time slot. Extract the spatial and temporal features corresponding to each of the graph state representations; By fusing the spatial and temporal features corresponding to each graph state representation, a spatiotemporal joint feature corresponding to each graph state representation is obtained; the spatiotemporal joint feature is used to characterize the global state of each fused node in the graph state representation in the current time slot; Based on the spatiotemporal joint features corresponding to any two fusion nodes, determine the collaborative attention score between any two fusion nodes; Based on the collaborative attention score and physical distance between any two fusion nodes, determine the final collaboration matrix between any two fusion nodes; Based on the final cooperation matrix, the target cooperation nodes participating in the localization are determined from all the first network nodes.

[0120] Optionally, the first determining module 330 is specifically used for: Based on the fusion weights of all edges in the graph state representation of each time slot, the adjacency matrix corresponding to each graph state representation is determined; Based on the adjacency matrix and identity matrix corresponding to each graph state representation, determine the self-connection matrix corresponding to each graph state representation; Determine the degree matrix corresponding to each of the self-connection matrices; Based on each of the self-connection matrices and the corresponding degree matrices, determine the normalized adjacency matrix corresponding to each of the graph state representations; Based on the normalized adjacency matrices and graph convolutional networks, spatial features corresponding to the graph state representations are extracted.

[0121] Optionally, the first determining module 330 is specifically used for: Based on the temporal convolutional network and the node feature sequences corresponding to each time window in each graph state representation, the temporal features corresponding to each graph state representation are extracted.

[0122] Optionally, the first determining module 330 is specifically used for: Based on the comparison results between the collaborative element values ​​corresponding to any two fusion nodes in the final collaborative matrix and the first preset threshold, all collaborative nodes are determined. Based on the final cooperation matrix, the sum of the cooperation strengths of each fusion node with other fusion nodes is determined; All collaborative nodes corresponding to the maximum total collaborative strength are identified as the target collaborative nodes for location services.

[0123] Optionally, the tag communication data includes an initial phase or an end phase.

[0124] Optionally, the second determining module 340 is specifically used for: Clustering is performed on the target phase of the target cooperative node at each frequency point to obtain the main phase interval of the target cooperative node at each frequency point; the target phase includes the initial phase or the ending phase. The target frequency is determined based on the main phase intervals corresponding to each of the adjacent frequency points; Based on the tag communication data corresponding to each tag to be located under the target frequency point, the target anti-interference positioning features corresponding to each tag to be located are determined.

[0125] Optionally, the second determining module 340 is specifically used for: The phase difference between adjacent frequency points is determined based on the main phase intervals corresponding to each adjacent frequency point. The correction triggering result is determined based on the phase difference and phase distribution corresponding to adjacent frequency points; Based on the correction triggering results, the target correction phase corresponding to each frequency point is determined; The target frequency is determined based on the target correction phase corresponding to each frequency point.

[0126] Optionally, the second determining module 340 is specifically used for: If the correction trigger result is a trigger correction and the target phase is within the first phase interval, the difference between the target phase and the first preset phase value is determined as the target correction phase. If the correction trigger result is a trigger correction and the target phase is within the second phase interval, determine the number of first phases of the target cooperative node in the third phase interval and the number of second phases in the fourth phase interval at each frequency point; if the target phase satisfies the first preset condition or the second preset condition, determine the difference between the first preset phase value and the target phase as the target correction phase; The first preset condition is that the target phase is in the third phase interval, and the number of the second phase is greater than the number of the first phase; The second preset condition is that the target phase is in the fourth phase interval, and the number of the first phase is greater than the number of the second phase.

[0127] Optionally, the second determining module 340 is specifically used for: Based on the target correction phase corresponding to each frequency point, a phase stability coefficient corresponding to each frequency point is determined; the phase stability coefficient is used to characterize the degree of phase fluctuation corresponding to each frequency point. The frequency point corresponding to the phase stability coefficient that is greater than the first preset threshold is determined as the target frequency point.

[0128] Optionally, the tag communication data may also include the RSSI of each tag to be located.

[0129] Optionally, the second determining module 340 is specifically used for: Based on the initial and final phases of each tag to be located at the target frequency, the average phase value of each tag to be located is determined. Based on the average phase value of all tags to be located, the frequency point phase slope corresponding to the target frequency point is determined; the frequency point phase slope is used to characterize the distance between the tag to be located and the corresponding target cooperative node. Based on the RSSI, initial phase, and final phase of each of the tags to be located, a tag vector is constructed. Based on the Euclidean distance between the target label vector and the reference label vector, the label similarity between each target label and the reference label is determined; the reference label vector includes the RSSI, initial phase, and final phase corresponding to the reference label. Based on the node identifier, target frequency, power, RSSI, initial phase, ending phase, frequency phase slope, and tag similarity of each tag to be located, the target anti-interference positioning features corresponding to each tag to be located are determined.

[0130] This application also provides a training device for an anti-interference positioning model. Figure 4 This is a schematic diagram of the structure of the training device for the anti-interference localization model provided in the embodiments of this application, as shown below. Figure 4 As shown, the training device 400 for the anti-interference positioning model includes a second building module 410 and a training module 420.

[0131] The second construction module 410 is used to construct a reward and penalty loss function based on the sample reward value, sample penalty value and error penalty factor; The training module 420 is used to train the initial anti-interference positioning model based on the anti-interference positioning sample features and the reward and punishment loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method described in any of the above embodiments.

[0132] Figure 5 This is a schematic diagram of the structure of the management device provided in the embodiments of this application, such as... Figure 5 As shown, the management device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other through the communications bus 540. The processor 510 can call logic instructions in the memory 530 to execute an IoT anti-interference positioning method or an anti-interference positioning model training method. The IoT anti-interference positioning method includes: constructing a communication graph and an interference graph corresponding to all first network nodes respectively; the communication graph is used to characterize the communication link connection relationship between the first network nodes; the interference graph is used to characterize the signal interference degree between the first network nodes; fusing the communication graph and the interference graph to obtain a fused topology graph; performing spatiotemporal joint modeling based on the fused topology graph and the tag communication data of each time slot corresponding to the fused topology graph to determine the target cooperative node participating in positioning from all first network nodes; determining the target anti-interference positioning features corresponding to the tag to be positioned based on the tag communication data corresponding to the target cooperative node; inputting the target anti-interference positioning features into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on sample reward values ​​and sample penalty values.

[0133] The training method of the anti-interference positioning model includes: constructing a reward and penalty loss function based on sample reward value, sample penalty value and error penalty factor; training the initial anti-interference positioning model based on anti-interference positioning sample features and the reward and penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method of any embodiment.

[0134] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0135] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the IoT anti-interference positioning method or the training method of the anti-interference positioning model provided by the above methods. The IoT anti-interference positioning method includes: constructing a communication graph and an interference graph corresponding to all first network nodes respectively; the communication graph is used to characterize the communication link connection relationship between the first network nodes; the interference graph is used to characterize the signal interference degree between the first network nodes; fusing the communication graph and the interference graph to obtain a fused topology graph; performing spatiotemporal joint modeling based on the fused topology graph and the tag communication data of each time slot corresponding to the fused topology graph to determine the target cooperative node participating in the positioning from all first network nodes; determining the target anti-interference positioning features corresponding to the tag to be positioned based on the tag communication data corresponding to the target cooperative node; inputting the target anti-interference positioning features into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on sample reward value and sample penalty value.

[0136] The training method of the anti-interference positioning model includes: constructing a reward and penalty loss function based on sample reward value, sample penalty value and error penalty factor; training the initial anti-interference positioning model based on anti-interference positioning sample features and the reward and penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method of any embodiment.

[0137] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a training method for the IoT anti-interference positioning method or anti-interference positioning model provided by the above methods. The IoT anti-interference positioning method includes: constructing a communication graph and an interference graph corresponding to all first network nodes respectively; the communication graph is used to characterize the communication link connection relationship between the first network nodes; the interference graph is used to characterize the signal interference degree between the first network nodes; fusing the communication graph and the interference graph to obtain a fused topology graph; performing spatiotemporal joint modeling based on the fused topology graph and the tag communication data of each time slot corresponding to the fused topology graph to determine the target cooperative node participating in the positioning from all first network nodes; determining the target anti-interference positioning features corresponding to the tag to be positioned based on the tag communication data corresponding to the target cooperative node; inputting the target anti-interference positioning features into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on sample reward values ​​and sample penalty values.

[0138] The training method of the anti-interference positioning model includes: constructing a reward and penalty loss function based on sample reward value, sample penalty value and error penalty factor; training the initial anti-interference positioning model based on anti-interference positioning sample features and the reward and penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method of any embodiment.

[0139] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An anti-interference positioning method for the Internet of Things, characterized in that, Applied to the management of equipment, the method includes: Construct communication graphs and interference graphs for all first network nodes respectively; the communication graphs are used to represent the communication link connections between the first network nodes; the interference graphs are used to represent the degree of signal interference between the first network nodes. By fusing the communication graph and the interference graph, a fused topology graph is obtained; Spatiotemporal joint modeling is performed based on the fusion topology map and the tag communication data of each time slot corresponding to the fusion topology map, and the target cooperative node participating in the positioning is determined from all first network nodes; Based on the tag communication data corresponding to the target cooperative node, the target anti-interference positioning features corresponding to the tag to be located are determined. The target anti-interference positioning features are input into the anti-interference positioning model to obtain the positioning prediction result output by the anti-interference positioning model; the anti-interference positioning model is trained based on sample reward value and sample penalty value.

2. The IoT anti-interference positioning method according to claim 1, characterized in that, Construct the communication graph corresponding to all first network nodes, including: For any two first network nodes, determine the probability of a successful link between the two first network nodes based on the physical distance between them. Taking each of the first network nodes as nodes, and in the case where there is an edge between any two first network nodes, the success probability of the link between the two first network nodes is determined as the communication weight of the edge between the two first network nodes, thus obtaining the communication graph corresponding to all the first network nodes.

3. The IoT anti-interference positioning method according to claim 2, characterized in that, Construct interference graphs for all first network nodes, including: For any two first network nodes, based on the tag communication data corresponding to each first network node, determine the number of tag collection repetitions corresponding to each first network node; based on the number of tag collection repetitions corresponding to each of the two first network nodes, determine the tag collection repetition rate corresponding to the two first network nodes. By taking each of the first network nodes as nodes and using the label collection repetition rate between any two first network nodes as the interference weight of the edge between the two first network nodes, the interference graph corresponding to all first network nodes is obtained.

4. The IoT anti-interference positioning method according to claim 3, characterized in that, The process of fusing the communication graph and the interference graph to obtain a fused topology graph includes: For any two nodes, the fusion weight of the edge is determined based on the communication weight of the edge in the communication graph and the interference weight of the edge in the interference graph. By superimposing the nodes in the communication graph and the interference graph, the fused nodes in the fused topology graph are obtained; The fusion topology graph is constructed based on the fusion weights of all fusion nodes and the edges between any two fusion nodes.

5. The IoT anti-interference positioning method according to any one of claims 1-4, characterized in that, The spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, determines the target cooperative nodes participating in the localization from all first network nodes, including: Based on the fused topology map and tag communication data of each time slot, the graph state representation corresponding to each time slot is determined; the graph state representation is used to characterize the network state of the current time slot. Extract the spatial and temporal features corresponding to each of the graph state representations; By fusing the spatial and temporal features corresponding to each graph state representation, a spatiotemporal joint feature corresponding to each graph state representation is obtained; the spatiotemporal joint feature is used to characterize the global state of each fused node in the graph state representation in the current time slot; Based on the spatiotemporal joint features corresponding to any two fusion nodes, determine the collaborative attention score between any two fusion nodes; Based on the collaborative attention score and physical distance between any two fusion nodes, determine the final collaboration matrix between any two fusion nodes; Based on the final cooperation matrix, the target cooperation nodes participating in the localization are determined from all the first network nodes.

6. The IoT anti-interference positioning method according to claim 5, characterized in that, The extraction of spatial features corresponding to each of the graph state representations includes: Based on the fusion weights of all edges in the graph state representation of each time slot, the adjacency matrix corresponding to each graph state representation is determined; Based on the adjacency matrix and identity matrix corresponding to each graph state representation, determine the self-connection matrix corresponding to each graph state representation; Determine the degree matrix corresponding to each of the self-connection matrices; Based on each of the self-connection matrices and the corresponding degree matrices, determine the normalized adjacency matrix corresponding to each of the graph state representations; Based on the normalized adjacency matrices and graph convolutional networks, spatial features corresponding to the graph state representations are extracted.

7. The IoT anti-interference positioning method according to claim 5, characterized in that, Extracting the temporal features corresponding to each of the graph state representations, including: Based on the temporal convolutional network and the node feature sequences corresponding to each time window in each graph state representation, the temporal features corresponding to each graph state representation are extracted.

8. The IoT anti-interference positioning method according to claim 5, characterized in that, The step of determining the target cooperative node participating in the localization from all first network nodes based on the final cooperative matrix includes: Based on the comparison results between the collaborative element values ​​corresponding to any two fusion nodes in the final collaborative matrix and the first preset threshold, all collaborative nodes are determined. Based on the final cooperation matrix, the sum of the cooperation strengths of each fusion node with other fusion nodes is determined; All collaborative nodes corresponding to the maximum total collaborative strength are identified as the target collaborative nodes for location services.

9. The IoT anti-interference positioning method according to claim 5, characterized in that, The tag communication data includes an initial phase or an ending phase; The step of determining the target anti-interference positioning features corresponding to the tag to be located based on the tag communication data corresponding to the target cooperative node includes: Clustering is performed on the target phase of the target cooperative node at each frequency point to obtain the main phase interval of the target cooperative node at each frequency point; the target phase includes the initial phase or the ending phase. The target frequency is determined based on the main phase intervals corresponding to each of the adjacent frequency points; Based on the tag communication data corresponding to each tag to be located under the target frequency point, the target anti-interference positioning features corresponding to each tag to be located are determined.

10. The IoT anti-interference positioning method according to claim 9, characterized in that, The determination of the target frequency point based on the main phase intervals corresponding to adjacent frequency points includes: The phase difference between adjacent frequency points is determined based on the main phase intervals corresponding to each adjacent frequency point. The correction triggering result is determined based on the phase difference and phase distribution corresponding to adjacent frequency points; Based on the correction triggering results, the target correction phase corresponding to each frequency point is determined; The target frequency is determined based on the target correction phase corresponding to each frequency point.

11. The IoT anti-interference positioning method according to claim 10, characterized in that, The step of determining the target correction phase corresponding to each frequency point based on the correction triggering result includes: If the correction trigger result is a trigger correction and the target phase is within the first phase interval, the difference between the target phase and the first preset phase value is determined as the target correction phase. If the correction trigger result is a trigger correction and the target phase is within the second phase interval, determine the number of first phases of the target cooperative node in the third phase interval and the number of second phases in the fourth phase interval at each frequency point; if the target phase satisfies the first preset condition or the second preset condition, determine the difference between the first preset phase value and the target phase as the target correction phase; The first preset condition is that the target phase is in the third phase interval, and the number of the second phase is greater than the number of the first phase; The second preset condition is that the target phase is in the fourth phase interval, and the number of the first phase is greater than the number of the second phase.

12. The IoT anti-interference positioning method according to claim 10, characterized in that, Determining the target frequency point based on the target correction phase corresponding to each frequency point includes: Based on the target correction phase corresponding to each frequency point, a phase stability coefficient corresponding to each frequency point is determined; the phase stability coefficient is used to characterize the degree of phase fluctuation corresponding to each frequency point. The frequency point corresponding to the phase stability coefficient that is greater than the first preset threshold is determined as the target frequency point.

13. The IoT anti-interference positioning method according to claim 10, characterized in that, The tag communication data also includes the RSSI of each tag to be located; The step of determining the target anti-interference positioning features corresponding to each tag to be located based on the tag communication data corresponding to each tag to be located under the target frequency point includes: Based on the initial and final phases of each tag to be located at the target frequency, the average phase value of each tag to be located is determined. Based on the average phase value of all tags to be located, the frequency point phase slope corresponding to the target frequency point is determined; the frequency point phase slope is used to characterize the distance between the tag to be located and the corresponding target cooperative node. Based on the RSSI, initial phase, and final phase of each of the tags to be located, a tag vector is constructed. Based on the Euclidean distance between the target label vector and the reference label vector, the label similarity between each target label and the reference label is determined; the reference label vector includes the RSSI, initial phase, and final phase corresponding to the reference label. Based on the node identifier, target frequency, power, RSSI, initial phase, ending phase, frequency phase slope, and tag similarity of each tag to be located, the target anti-interference positioning features corresponding to each tag to be located are determined.

14. A training method for an anti-interference localization model, characterized in that, include: Construct a reward and penalty loss function based on sample reward value, sample penalty value, and error penalty factor; The initial anti-interference positioning model is trained based on the anti-interference positioning sample features and the reward / penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method according to any one of claims 1-13.

15. An anti-interference positioning device for the Internet of Things, characterized in that, include: The first construction module is used to construct communication graphs and interference graphs corresponding to all first network nodes respectively; the communication graph is used to represent the communication link connection relationship between the first network nodes; the interference graph is used to represent the signal interference degree between the first network nodes. The fusion module is used to fuse the communication graph and the interference graph to obtain a fused topology graph; The first determining module is used to perform spatiotemporal joint modeling based on the fused topology map and the tag communication data of each time slot corresponding to the fused topology map, and to determine the target cooperative node participating in the positioning from all the first network nodes; The second determining module is used to determine the target anti-interference positioning features corresponding to the tag to be located based on the tag communication data corresponding to the target cooperative node. The localization module is used to input the anti-interference localization features of the target into the anti-interference localization model to obtain the localization prediction result output by the anti-interference localization model; the anti-interference localization model is trained based on sample reward values ​​and sample penalty values.

16. A training device for an anti-interference positioning model, characterized in that, include: The second building module is used to construct a reward and penalty loss function based on the sample reward value, sample penalty value, and error penalty factor; The training module is used to train the initial anti-interference positioning model based on the anti-interference positioning sample features and the reward-penalty loss function to obtain the trained anti-interference positioning model; the anti-interference positioning model is used to execute the IoT anti-interference positioning method according to any one of claims 1-13.

17. A management device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the IoT anti-interference positioning method as described in any one of claims 1 to 13, or the training method for the anti-interference positioning model as described in claim 14.

18. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the IoT anti-interference positioning method as described in any one of claims 1 to 13, or the training method for the anti-interference positioning model as described in claim 14.

19. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the IoT anti-interference positioning method as described in any one of claims 1 to 13, or the training method for the anti-interference positioning model as described in claim 14.

Images