Multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment

Abstract

The application discloses a multi-source heterogeneous data fusion management system based on a spectrum sensing detection device, a device anchoring unit configured to anchor a spectrum sensing device in a region to be sensed; a real-time feature acquisition unit configured to acquire real-time spectrum features of the spectrum sensing device; a real-time feature input unit configured to input the real-time spectrum features to a pre-constructed spectrum state recognition model; a state sequence output unit configured to output a spectrum state time sequence of each spectrum sensing device in the region to be sensed according to the spectrum state recognition model; and a spectrum device management unit configured to uniformly manage state data of multiple heterogeneous spectrum sensing devices according to the spectrum state time sequence of each spectrum sensing device. The application realizes uniform structural expression of devices with uneven spatial distribution and inconsistent frequency bands by combining geographical coordinate clustering and directionality edge building, so that the system has good heterogeneous adaptability and state fusion capability.

Description

Technical Field

[0001] This invention relates to a spectrum data management system, specifically a multi-source heterogeneous data fusion management system based on spectrum sensing and detection equipment. Background Technology

[0002] With the rapid development of wireless communication technology and the increasing scarcity of spectrum resources, spectrum sensing technology plays a crucial role in cognitive radio, interference detection, and spectrum resource scheduling. Traditional spectrum sensing systems mainly rely on each spectrum sensing device to independently collect and determine the status of its local frequency band. However, in actual deployments, due to objective conditions such as heterogeneous devices, uneven distribution, and different frequency band coverage, spectrum status data suffers from spatial fragmentation and incomplete observation, making it difficult to form a unified and continuous spectrum status view. Patent document CN115250473B discloses a spectrum sharing method, device, and computing equipment based on graph convolutional neural networks, which can conveniently obtain a better spectrum access scheme. However, because spectrum status has obvious spatial correlation and consistent behavioral patterns, it is impossible to accurately infer the status of other nodes by simply setting rules (such as the power threshold method) or a small number of labeled points. Especially in large-scale deployment scenarios with joint monitoring of multi-source heterogeneous devices, the incompleteness of spectrum status will seriously affect the real-time accuracy of spectrum scheduling strategies. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a multi-source heterogeneous data fusion management system based on spectrum sensing and detection equipment. This system addresses the technical problems mentioned in the background by constructing a graph-structured spectrum state identification of the region to be sensed.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment, the management system comprising:

[0006] The device anchoring unit is configured to anchor a spectrum sensing device in the area to be sensed.

[0007] The real-time feature acquisition unit is configured to acquire the real-time spectrum features of the spectrum sensing device.

[0008] The real-time feature input unit is configured to input real-time spectral features into a pre-built spectral state recognition model;

[0009] Among them, the spectrum state recognition model has fitted the temporal mapping between the spectrum characteristics and spectrum state of any spectrum sensing device in the region to be sensed;

[0010] The state sequence output unit is configured to output the time sequence of the spectrum state of each spectrum sensing device in the area to be sensed, based on the spectrum state recognition model.

[0011] The spectrum device management unit is configured to uniformly manage the status data of multiple heterogeneous spectrum sensing devices based on the time sequence of the spectrum status of each spectrum sensing device.

[0012] In some specific embodiments, the pre-construction step of the spectrum state identification model includes:

[0013] S1. Delineate the area to be sensed on the electronic map;

[0014] S2. In the area to be sensed, construct a graph structure sample of N spectrum sensing devices at the current timestamp;

[0015] S3. Collect graph structure samples with multiple timestamps sequentially on the time axis with a fixed step size time window to generate a graph structure sample sequence;

[0016] S4. Input the graph structure sample sequence into a time-series model based on a graph neural network for iterative training to fit the time-series mapping between the spectral features and spectral states of any node in the region to be sensed.

[0017] In some specific embodiments, step S2, which involves constructing graph structure samples of N spectrum sensing devices at the current timestamp, includes:

[0018] S2-1. Mark the geographic coordinates of N spectrum sensing devices in the area to be sensed, and define the geographic coordinates as N nodes of a graph structure.

[0019] S2-2, Obtain the spectral feature vectors of N nodes;

[0020] S2-3. Based on the spectral feature vectors and geographic coordinates of N nodes, anchor K cluster core nodes and non-cluster core nodes in the graph structure;

[0021] S2-4. Define the cluster center node as the starting point of the direction, and mark the upstream and downstream nodes between node pairs;

[0022] S2-5. For any pair of nodes, determine the weight of the directed edge based on its geographical coordinates and spectral feature vector.

[0023] S2-6. Obtain the spectral state of K cluster center nodes, and perform feature aggregation on non-cluster center nodes according to the directed edges and edge weights of node pairs until the spectral state of all nodes is obtained.

[0024] S2-7. Define the graph structure containing the spectrum states of all nodes as the graph structure sample of the current timestamp.

[0025] In some specific embodiments, obtaining the spectral feature vectors of N nodes in step S2-2 includes:

[0026] S2-2-1. Obtain several raw spectrum parameters collected by each node in the specified frequency band;

[0027] S2-2-2, Perform feature standardization on several original spectral parameters to generate several spectral features;

[0028] S2-2-3. Combine several spectral features to form the spectral feature vector of each node.

[0029] In some specific embodiments, anchoring K cluster core nodes and non-cluster core nodes in the graph structure in step S2-3 includes:

[0030] S2-3-1. Perform K-value clustering based on the geographic coordinates of nodes to divide N nodes into K spatial clusters;

[0031] S2-3-2. Based on the spectral feature vector of each node, determine the cluster center node and non-cluster center nodes in each spatial cluster.

[0032] In some specific embodiments, dividing the N nodes into K spatial clusters in step S2-2-1 includes:

[0033] S2-3-1-1, Obtain the coordinates of N nodes corresponding to N nodes;

[0034] S2-3-1-2. Randomly select K node coordinates from N node coordinates as the initial cluster center;

[0035] S2-3-1-3. For each node coordinate, calculate its coordinate distance to the K initial cluster centers to obtain the K coordinate distances of each node coordinate;

[0036] S2-3-1-4. Select the minimum coordinate distance from the K coordinate distances of each node.

[0037] S2-3-1-5. Distribute the coordinates of N nodes to the initial cluster centers with the minimum coordinate distance to obtain K initial spatial clusters;

[0038] S2-3-1-6. Calculate the centroid of the node coordinates in each initial spatial cluster, and determine K centroids;

[0039] S2-3-1-7, Use K centroids as the cluster centers for the next round;

[0040] S2-3-1-8. Calculate the change distance between the next cluster center and the initial cluster center;

[0041] S2-3-1-9. If the change distance is greater than the change threshold, then repeatedly traverse the coordinates of N nodes and assign the coordinates of N nodes to the next cluster center that is the distance from their minimum coordinates, until the change distance of the cluster center is less than the change threshold.

[0042] S2-3-1-10. Define the cluster centers whose change distance is less than the change threshold and their assigned node coordinates as the spatial clusters to generate K spatial clusters.

[0043] In some specific embodiments, determining the cluster center node and non-cluster center nodes in each spatial cluster in step S2-3-2 includes:

[0044] S2-3-2-1. Determine the mean vector of the spatial cluster based on the spectral feature vector in each spatial cluster;

[0045] S2-3-2-2, Define the mean vector of the spatial cluster as a virtual node;

[0046] S2-3-2-3, Calculate the vector distance between each node in the spatial cluster and the virtual node;

[0047] S2-3-2-4. Define the node corresponding to the minimum vector distance as the cluster center node of the spatial cluster, and the other nodes as non-cluster center nodes.

[0048] In some specific embodiments, marking the upstream and downstream nodes between node pairs in steps S2-4 includes:

[0049] S2-4-1. Select any pair of nodes in each spatial cluster;

[0050] S2-4-2. Calculate the vector distance between the two nodes in the node pair and their cluster center node;

[0051] S2-4-3. Compare the vector distances between two nodes and their cluster center nodes;

[0052] S2-4-4. Select the node with the smaller vector distance as the upstream node and the other node as the downstream node.

[0053] In some specific embodiments, determining the edge weight of the directed edge in step S2-5 includes:

[0054] S2-5-1. Obtain the geographic coordinates of the upstream and downstream nodes in a node pair, and calculate the vector distance between the node pairs.

[0055] S2-5-2. Compare the vector distance between node pairs with a preset distance threshold, and filter out node pairs whose vector distance is less than the distance threshold, defining them as candidate node pairs;

[0056] S2-5-3. For candidate node pairs, establish directed edges from the upstream node to the downstream node;

[0057] S2-5-4. Obtain the spectral feature vectors of the upstream and downstream nodes in each established directed edge;

[0058] S2-5-5. Calculate the vector similarity of directed edges based on the spectral feature vectors of upstream and downstream nodes.

[0059] The formula for calculating the vector similarity is:

[0060] ;

[0061] in, This represents the spectral feature vector of the upstream node. Represents the spectral feature vector of the downstream node. Vector similarity is used to represent directed edges; The scaling factor represents the rate at which similarity decays. The vector Euclidean distance representing the spectral eigenvectors of upstream and downstream nodes. This represents the natural exponential function, used to map distance values ​​to similarity weights within the interval (0,1].

[0062] S2-5-6. Obtain the vector similarity of all directed edges;

[0063] S2-5-7. Normalize the vector similarity of any directed edge with the vector similarity of all directed edges to generate the edge weight of the corresponding directed edge.

[0064] This invention provides a multi-source heterogeneous data fusion management system based on spectrum sensing detection devices, effectively solving the problem that traditional rule-based methods cannot achieve state completion and unified management in complex deployment scenarios. Furthermore, by introducing a small number of cluster core nodes with known spectrum states in a multi-source heterogeneous spectrum sensing environment, and utilizing graph structure features to aggregate and complete the spectrum states of all nodes in the graph, the system improves spectrum state coverage under conditions of scarce annotations. Simultaneously, by combining geographic coordinate clustering and directional edge construction, a unified structural expression for devices with uneven spatial distribution and inconsistent frequency bands is achieved, giving the system good heterogeneous adaptability and state fusion capabilities. Attached Figure Description

[0065] Figure 1 This is a structural block diagram of the multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment of the present invention;

[0066] Figure 2 This is a schematic diagram of the management process of the multi-source heterogeneous data fusion management system based on spectrum sensing and detection equipment of the present invention;

[0067] Figure 3 This is a schematic diagram of the pre-construction process of the spectrum state recognition model described in this invention;

[0068] Figure 4 This is a schematic diagram illustrating the process of generating the edge weights of directed edges according to the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] First, the prior art and related concepts involved in the embodiments of the present invention will be described:

[0071] ST-GCN model: Spatio-Temporal Graph Convolutional Network. The ST-GCN model models both the spatial relationships and temporal dependencies of nodes by embedding temporal convolutional modules into a graph convolutional network.

[0072] Coordinate distance: used to represent the spatial distance between spectrum sensing devices calculated based on geographic coordinates, using two-dimensional Euclidean distance;

[0073] Vector distance: used to represent the similarity between spectral feature vectors in the feature space, using Euclidean distance.

[0074] Example 1: Please refer to Figures 1 to 2 This invention provides a multi-source heterogeneous data fusion management system based on spectrum sensing and detection equipment, the management system comprising the following units:

[0075] The device anchoring unit is configured to anchor a spectrum sensing device in the area to be sensed.

[0076] The real-time feature acquisition unit is configured to acquire the real-time spectrum features of the spectrum sensing device.

[0077] The real-time feature input unit is configured to input real-time spectral features into a pre-built spectral state recognition model;

[0078] Among them, the spectrum state recognition model has fitted the temporal mapping between the spectrum characteristics and spectrum state of any spectrum sensing device in the region to be sensed;

[0079] The state sequence output unit is configured to output the time sequence of the spectrum state of each spectrum sensing device in the area to be sensed, based on the spectrum state recognition model.

[0080] Specifically, the spectrum state time sequence refers to the set of spectrum states identified within a continuous time window for each spectrum sensing device in the area to be sensed, arranged in the order of timestamps, which is used to reflect the trend of spectrum state changes of several spectrum sensing devices in the monitored frequency band within the area to be sensed.

[0081] The spectrum device management unit is configured to uniformly manage the status data of multiple heterogeneous spectrum sensing devices based on the time sequence of the spectrum status of each spectrum sensing device.

[0082] Unified management can be achieved by classifying and managing spectrum usage status, detecting interference, dynamically scheduling or allocating resources based on preset rules, so as to realize intelligent fusion management of multi-source heterogeneous spectrum data.

[0083] In this embodiment, by setting up a device anchoring unit, a real-time feature acquisition unit, a real-time feature input unit, a state sequence output unit, and a spectrum device management unit, the acquisition of the temporal information of the spectrum state of any spectrum sensing device within the sensing area is realized, constructing a unified management path for multi-source heterogeneous spectrum data. This embodiment completes the identification and temporal representation of spectrum state based on a spectrum state recognition model, and manages the spectrum state of multiple devices through a unified spectrum device management unit, improving the centralization of spectrum state management.

[0084] Example 2: See Figures 2 to 4 The technical solution of Embodiment 2 differs from Embodiment 1 in that it discloses a pre-construction step for the spectrum state identification model described in the embodiment. This pre-construction step includes:

[0085] S1. Delineate the area to be sensed on the electronic map;

[0086] In this context, the electronic map refers to a Geographic Information System (GIS) with spatial coordinates, used to mark the geographical location of spectrum sensing devices. The area to be sensed refers to the target area for spectrum observation selected in the electronic map, typically a two-dimensional area within a communication coverage area, a local urban grid, or a specified scene boundary, used to deploy spectrum sensing devices.

[0087] S2. In the area to be sensed, construct a graph structure sample of N spectrum sensing devices at the current timestamp;

[0088] S3. Collect graph structure samples with multiple timestamps sequentially on the time axis with a fixed step size time window to generate a graph structure sample sequence;

[0089] S4. Input the graph structure sample sequence into a time-series model based on a graph neural network for iterative training to fit the time-series mapping between the spectral features and spectral states of any node in the region to be sensed.

[0090] The temporal mapping refers to taking the spectral features of nodes in the graph structure sample sequence and the relationship between adjacent nodes in the graph structure as input, and learning the prediction of the spectral state of any node at the current timestamp; wherein, any node indicates that the model has the spatial generalization ability of the spectral state of all nodes in the region to be perceived.

[0091] It should be noted that the "time series model based on graph neural networks" in this embodiment refers to a deep learning model used to simultaneously model the adjacency relationships and time series dependencies of graph structures. It can utilize the node features, edge connections, and edge weights in the graph structure sample sequence to achieve dynamic prediction or identification of node states in the time dimension. The model can be the ST-GCN model.

[0092] In this embodiment, by constructing a graph structure sample containing the spatial distribution and adjacency relationship of spectrum sensing devices, and generating a graph structure sample sequence on the time axis, and using it as input to train the graph neural network model, the spectrum state recognition model is able to fit the temporal mapping of the spectrum state of any node at any time stamp, thereby establishing a spectrum state temporal prediction model that can cover all spectrum sensing devices in the area to be sensed.

[0093] For example, step S2 in this embodiment further includes:

[0094] S2-1. Mark the geographic coordinates of N spectrum sensing devices in the area to be sensed, and define the geographic coordinates as N nodes of a graph structure.

[0095] S2-2, Obtain the spectral feature vectors of N nodes;

[0096] S2-3. Based on the spectral feature vectors and geographic coordinates of N nodes, anchor K cluster core nodes and non-cluster core nodes in the graph structure;

[0097] S2-4. Define the cluster center node as the starting point of the direction, and mark the upstream and downstream nodes between node pairs;

[0098] S2-5. For any pair of nodes, determine the weight of the directed edge based on its geographical coordinates and spectral feature vector.

[0099] S2-6. Obtain the spectral state of K cluster center nodes, and perform feature aggregation on non-cluster center nodes according to the directed edges and edge weights of node pairs until the spectral state of all nodes is obtained.

[0100] The spectrum status refers to the discrete signal usage status of the spectrum sensing device in a specified frequency band, which typically includes, but is not limited to, "idle status", "occupied status", "interference status" or "unknown status". It is used to indicate whether the device has signal activity and its signal type characteristics in the frequency band, and is the classification target label of the spectrum status identification model.

[0101] The spectrum status can be extracted from historical labeled data, or it can be set according to rules and conditions such as the collected spectrum power threshold, occupancy rate, or interference intensity.

[0102] It should be noted that feature aggregation of non-cluster core nodes based on directed edges and edge weights refers to: taking a non-cluster core node as the target node, collecting the spectral states of all its upstream neighboring nodes, and combining this with the edge weights of the corresponding directed edges. Using feature aggregation (message passing mechanism) in graph neural networks, the state features of upstream nodes are weighted and aggregated according to edge weights, and then propagated to downstream nodes, thus completing state propagation. This feature aggregation process reflects the conditional dependencies of node states in the graph structure; the larger the edge weight, the more significant the state influence, ultimately achieving the gradual propagation of state features from the cluster core node to the entire graph structure.

[0103] S2-7. Define the graph structure containing the spectrum states of all nodes as the graph structure sample of the current timestamp.

[0104] In this embodiment, the geographic coordinates of the spectrum sensing device are constructed as nodes of a graph structure, and directed edges and their weights are constructed based on the spatial location and spectrum feature relationship between the devices, thereby realizing graph structure modeling of the spectrum state at the current timestamp; by calibrating the spectrum state of the cluster center node and performing feature aggregation for message passing based on the edge weights, the spectrum state results of all nodes at the current timestamp are generated, thus completing the propagation of the node spectrum state within the graph structure.

[0105] For example, step S2-2 in this embodiment further includes:

[0106] S2-2-1. Obtain several raw spectrum parameters collected by each node in the specified frequency band;

[0107] Several spectral parameters include, but are not limited to, quantifiable spectral statistical characteristics such as signal power, bandwidth utilization, energy spectral density, burst behavior indicators, spectral stability, and spectral usage duration.

[0108] S2-2-2, Perform feature standardization on several original spectral parameters to generate several spectral features;

[0109] It should be noted that standardization is based on the parameter type of each spectral parameter, with different processing methods selected accordingly. Specifically, for continuously varying energy parameters (such as average power and peak power), Z-score standardization can be used to preserve their relative deviation and directionality; while for proportional parameters such as occupancy ratio and bandwidth utilization, min-max normalization or scaling normalization methods are preferable. For example, for the above-mentioned different types of spectral parameters, the final result must be a uniform spectral characteristic of several spectral parameters under the same dimensional scale through an appropriate normalization method to meet the input consistency requirements.

[0110] S2-2-3. Combine several spectral features to form the spectral feature vector of each node.

[0111] Feature splicing refers to arranging several standardized spectral features corresponding to each node in a preset order and combining them to form a fixed-dimensional one-dimensional vector structure, which serves as the input feature representation of that node in the graph structure.

[0112] For example, in this embodiment, step S2-3 further includes:

[0113] S2-3-1. Perform K-value clustering based on the geographic coordinates of nodes to divide N nodes into K spatial clusters;

[0114] S2-3-2. Based on the spectral feature vector of each node, determine the cluster center node and non-cluster center nodes in each spatial cluster.

[0115] For example, in this embodiment, the step of dividing N nodes into K spatial clusters further includes:

[0116] S2-3-1-1, Obtain the coordinates of N nodes corresponding to N nodes;

[0117] S2-3-1-2. Randomly select K node coordinates from N node coordinates as the initial cluster center;

[0118] S2-3-1-3. For each node coordinate, calculate its coordinate distance to the K initial cluster centers to obtain the K coordinate distances of each node coordinate;

[0119] S2-3-1-4. Select the minimum coordinate distance from the K coordinate distances of each node.

[0120] S2-3-1-5. Distribute the coordinates of N nodes to the initial cluster centers with the minimum coordinate distance to obtain K initial spatial clusters;

[0121] S2-3-1-6. Calculate the centroid of the node coordinates in each initial spatial cluster, and determine K centroids;

[0122] It should be noted that the process of calculating the centroid is as follows: for each spatial cluster, the geographic coordinates of all nodes in the cluster are counted, and the arithmetic mean of the values ​​of all nodes in each dimension of the coordinates is taken. The resulting average coordinates are the centroid coordinates of the spatial cluster.

[0123] S2-3-1-7, Use K centroids as the cluster centers for the next round;

[0124] S2-3-1-8. Calculate the change distance between the next cluster center and the initial cluster center;

[0125] S2-3-1-9. If the change distance is greater than the change threshold, then repeatedly traverse the coordinates of N nodes and assign the coordinates of N nodes to the next cluster center that is the distance from their minimum coordinates, until the change distance of the cluster center is less than the change threshold.

[0126] S2-3-1-10. Define the cluster centers whose change distance is less than the change threshold and their assigned node coordinates as the spatial clusters to generate K spatial clusters.

[0127] In this embodiment, K-value clustering is performed based on the geographic coordinates of the spectrum sensing device. The cluster center iterative update and distance convergence judgment mechanism are adopted to adaptively divide N nodes into K spatial clusters, thereby realizing the structured aggregation of spectrum sensing devices in geographic space.

[0128] For example, in this embodiment, determining the cluster center node and non-cluster center nodes in each spatial cluster in step S2-3-2 includes:

[0129] S2-3-2-1. Determine the mean vector of the spatial cluster based on the spectral feature vector in each spatial cluster;

[0130] S2-3-2-2, Define the mean vector of the spatial cluster as a virtual node;

[0131] S2-3-2-3, Calculate the vector distance between each node in the spatial cluster and the virtual node;

[0132] S2-3-2-4. Define the node corresponding to the minimum vector distance as the cluster center node of the spatial cluster, and the other nodes as non-cluster center nodes.

[0133] In this embodiment, by calculating the mean vector of the spectral feature vectors of all nodes in each spatial cluster, and constructing virtual nodes with the mean vector as a reference, and then selecting the node with the smallest distance as the cluster center node, the cluster center node with the most representative spectral behavior is selected, providing a feature aggregation representative for the propagation starting point of the spectral state.

[0134] For example, steps S2-4 further include:

[0135] S2-4-1. Select any pair of nodes in each spatial cluster;

[0136] S2-4-2. Calculate the vector distance between the two nodes in the node pair and their cluster center node;

[0137] S2-4-3. Compare the vector distances between two nodes and their cluster center nodes;

[0138] S2-4-4. Select the node with the smaller vector distance as the upstream node and the other node as the downstream node.

[0139] In this embodiment, by comparing the vector distances between the two nodes in any node pair and their respective cluster center nodes, the node closer is determined to be the upstream node and the node farther away is determined to be the downstream node. This realizes the rule-based partitioning of directed edge direction based on spatial geometric position, thereby constructing edges with propagation directionality and ensuring that the spectral state is transmitted in an orderly manner along the path of increasing distance.

[0140] For example, in this embodiment, step S2-5 further includes:

[0141] S2-5-1. Obtain the geographic coordinates of the upstream and downstream nodes in a node pair, and calculate the vector distance between the node pairs.

[0142] S2-5-2. Compare the vector distance between node pairs with a preset distance threshold, and filter out node pairs whose vector distance is less than the distance threshold, defining them as candidate node pairs;

[0143] S2-5-3. For candidate node pairs, establish directed edges from the upstream node to the downstream node;

[0144] S2-5-4. Obtain the spectral feature vectors of the upstream and downstream nodes in each established directed edge;

[0145] S2-5-5. Calculate the vector similarity of directed edges based on the spectral feature vectors of upstream and downstream nodes.

[0146] The formula for calculating the vector similarity is:

[0147] ;

[0148] in, This represents the spectral feature vector of the upstream node. Represents the spectral feature vector of the downstream node. Vector similarity is used to represent directed edges; The scaling factor represents the rate at which similarity decays. The vector Euclidean distance (Euclidean distance) represents the spectral feature vectors of upstream and downstream nodes. This represents the natural exponential function, used to map distance values ​​to similarity weights within the interval (0,1].

[0149] Specifically, by negatively compressing the vector distance using the natural exponential function, the goal of smaller distance, greater similarity, and stronger edge weights can be achieved, which conforms to the logic of "stronger information coupling between similar nodes" in dependency relationships.

[0150] S2-5-6. Obtain the vector similarity of all directed edges;

[0151] S2-5-7. Normalize the vector similarity of any directed edge with the vector similarity of all directed edges to generate the edge weight of the corresponding directed edge. The edge weight is used to characterize the strength of the state dependency between nodes.

[0152] In this embodiment, the similarity between spectral feature vectors of nodes with vector distance less than a threshold is calculated, and the negative exponential function of vector distance is used to map it to continuous values ​​in the interval (0,1). Then, all similarities are normalized to obtain the edge weights for directed edges in the graph structure. This enables state dependency modeling based on behavioral similarity within the graph structure, quantifies the transmission strength of spectral state between nodes, and allows each node to reasonably aggregate the spectral state transmitted from upstream.

[0153] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means.

[0154] The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g.,...), etc. DVD ( ), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0155] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0156] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-source heterogeneous data fusion management system based on spectrum sensing and detection equipment, including: The device anchoring unit is configured to anchor a spectrum sensing device in the area to be sensed. The real-time feature acquisition unit is configured to acquire the real-time spectrum features of the spectrum sensing device. The real-time feature input unit is configured to input real-time spectral features into a pre-built spectral state recognition model; Among them, the spectrum state recognition model has fitted the temporal mapping between the spectrum characteristics and spectrum state of any spectrum sensing device in the region to be sensed; The state sequence output unit is configured to output the time sequence of the spectrum state of each spectrum sensing device in the area to be sensed, based on the spectrum state recognition model. The spectrum device management unit is configured to uniformly manage the status data of multiple heterogeneous spectrum sensing devices based on the time sequence of the spectrum status of each spectrum sensing device.

2. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 1, characterized in that, The pre-construction steps of the spectrum state recognition model include: S1. Delineate the area to be sensed on the electronic map; S2. In the area to be sensed, construct a graph structure sample of N spectrum sensing devices at the current timestamp; S3. Collect graph structure samples with multiple timestamps sequentially on the time axis with a fixed step size time window to generate a graph structure sample sequence; S4. Input the graph structure sample sequence into a time-series model based on a graph neural network for iterative training to fit the time-series mapping between the spectral features and spectral states of any node in the region to be sensed.

3. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 2, characterized in that, The step S2, which involves constructing graph structure samples of N spectrum sensing devices at the current timestamp, includes: S2-1. Mark the geographic coordinates of N spectrum sensing devices in the area to be sensed, and define the geographic coordinates as N nodes of a graph structure. S2-2, Obtain the spectral feature vectors of N nodes; S2-3. Based on the spectral feature vectors and geographic coordinates of N nodes, anchor K cluster core nodes and non-cluster core nodes in the graph structure; S2-4. Define the cluster center node as the starting point of the direction, and mark the upstream and downstream nodes between node pairs; S2-5. For any pair of nodes, determine the weight of the directed edge based on its geographical coordinates and spectral feature vector. S2-6. Obtain the spectral state of K cluster center nodes, and perform feature aggregation on non-cluster center nodes according to the directed edges and edge weights of node pairs until the spectral state of all nodes is obtained. S2-7. Define the graph structure containing the spectrum states of all nodes as the graph structure sample of the current timestamp.

4. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 3, characterized in that, The step S2-2 of obtaining the spectral feature vectors of N nodes includes: S2-2-1. Obtain several raw spectrum parameters collected by each node in the specified frequency band; S2-2-2, Perform feature standardization on several original spectral parameters to generate several spectral features; S2-2-3. Combine several spectral features to form the spectral feature vector of each node.

5. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 3, characterized in that, In step S2-3, anchoring K cluster core nodes and non-cluster core nodes in the graph structure includes: S2-3-1. Perform K-value clustering based on the geographic coordinates of nodes to divide N nodes into K spatial clusters; S2-3-2. Based on the spectral feature vector of each node, determine the cluster center node and non-cluster center nodes in each spatial cluster.

6. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 5, characterized in that, The step S2-3-1, which involves dividing the N nodes into K spatial clusters, includes: S2-3-1-1, Obtain the coordinates of N nodes corresponding to N nodes; S2-3-1-2. Randomly select K node coordinates from N node coordinates as the initial cluster center; S2-3-1-3. For each node coordinate, calculate its coordinate distance to the K initial cluster centers to obtain the K coordinate distances of each node coordinate; S2-3-1-4. Select the minimum coordinate distance from the K coordinate distances of each node. S2-3-1-5. Distribute the coordinates of N nodes to the initial cluster centers with the minimum coordinate distance to obtain K initial spatial clusters; S2-3-1-6. Calculate the centroid of the node coordinates in each initial spatial cluster, and determine K centroids; S2-3-1-7, Use K centroids as the cluster centers for the next round; S2-3-1-8. Calculate the change distance between the next cluster center and the initial cluster center; S2-3-1-9. If the change distance is greater than the change threshold, then repeatedly traverse the coordinates of N nodes and assign the coordinates of N nodes to the next cluster center that is the distance from their minimum coordinates, until the change distance of the cluster center is less than the change threshold. S2-3-1-10. Define the cluster centers whose change distance is less than the change threshold and their assigned node coordinates as the spatial clusters to generate K spatial clusters.

7. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 5, characterized in that, In step S2-3-2, determining the cluster center node and non-cluster center nodes in each spatial cluster includes: S2-3-2-1. Determine the mean vector of the spatial cluster based on the spectral feature vector in each spatial cluster; S2-3-2-2, Define the mean vector of the spatial cluster as a virtual node; S2-3-2-3, Calculate the vector distance between each node in the spatial cluster and the virtual node; S2-3-2-4. Define the node corresponding to the minimum vector distance as the cluster center node of the spatial cluster, and the other nodes as non-cluster center nodes.

8. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 3, characterized in that, The step S2-4, which involves marking upstream and downstream nodes between node pairs, includes: S2-4-1. Select any pair of nodes in each spatial cluster; S2-4-2. Calculate the vector distance between the two nodes in the node pair and their cluster center node; S2-4-3. Compare the vector distances between two nodes and their cluster center nodes; S2-4-4. Select the node with the smaller vector distance as the upstream node and the other node as the downstream node.

9. The multi-source heterogeneous data fusion management system based on spectrum sensing detection equipment according to claim 3, characterized in that, Determining the edge weight of the directed edge in step S2-5 includes: S2-5-1. Obtain the geographic coordinates of the upstream and downstream nodes in a node pair, and calculate the vector distance between the node pairs. S2-5-2. Compare the vector distance between node pairs with a preset distance threshold, and filter out node pairs whose vector distance is less than the distance threshold, defining them as candidate node pairs; S2-5-3. For candidate node pairs, establish directed edges from the upstream node to the downstream node; S2-5-4. Obtain the spectral feature vectors of the upstream and downstream nodes in each established directed edge; S2-5-5. Calculate the vector similarity of directed edges based on the spectral feature vectors of upstream and downstream nodes. The formula for calculating the vector similarity is: ; in, This represents the spectral feature vector of the upstream node. Represents the spectral feature vector of the downstream node. Vector similarity is used to represent directed edges; The scaling factor represents the rate at which similarity decays. The vector Euclidean distance representing the spectral eigenvectors of upstream and downstream nodes. This represents the natural exponential function, used to map distance values ​​to similarity weights within the interval (0,1]. S2-5-6. Obtain the vector similarity of all directed edges; S2-5-7. Normalize the vector similarity of any directed edge with the vector similarity of all directed edges to generate the edge weight of the corresponding directed edge.

Images