Dual-mode carrier communication aggregation interference identification system and device based on graph neural network

Abstract

The invention discloses a dual-mode carrier communication aggregation interference identification system and device based on a graph neural network, and the system comprises the following modules: a communication graph construction module which is used for constructing a communication topological graph and configuring node dynamic attributes and disturbance direction labels; the graph neural network modeling module is used for constructing a multi-layer rotation sensing graph neural network, sensing a structure connection relation and direction change and generating an interference heat map; the interference path search module is used for performing jump search by adopting an improved fruit fly optimization algorithm to generate an interference path candidate set; the path evaluation and source point positioning module is used for determining an optimal interference path and an interference source position; the linkage control module is used for dynamically implementing interference suppression and path switching operation; and the self-learning updating module is used for collecting communication link signal quality data and updating an interference heat map generation mechanism and a drosophila jumping strategy. According to the method, the image neural network and the improved fruit fly optimization algorithm are fused, and the communication interference path is accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of communication interference identification and control, and particularly to a dual-mode carrier communication aggregation interference identification system and device based on a graph neural network. Background Art

[0002] In the context of the current widespread application of dual-mode carrier communication in industrial control, power transmission, and intelligent sensing networks, the problem of communication interference has become increasingly prominent. Especially in complex environments, communication links are affected by various non-linear and sudden electromagnetic interference sources, seriously threatening the transmission stability of signals and the reliable reception of information. Traditional interference identification and suppression methods are mostly based on means such as frequency-domain filtering, level detection, and channel equalization. These methods usually rely on preset thresholds or static rules for interference identification and cannot adapt to the dynamically changing interference characteristics and the structural complexity brought by multi-path signal propagation. In addition, existing methods have limited effectiveness in dealing with typical crosstalk, harmonic coupling, and frequency-hopping interference in power line communication systems. Especially when the interference signal has a sudden change in direction during propagation in the path or the cumulative attenuation of multiple nodes in series, traditional methods are difficult to effectively characterize the evolution process of the interference path, thereby affecting the subsequent deployment of suppression and compensation strategies.

[0003] In recent years, the development of artificial intelligence technology, especially graph neural networks (GNNs), has provided new ideas for communication modeling and interference identification under structured data. Some studies have attempted to model communication networks as graph structures and extract topological dependencies between nodes through graph convolution or graph attention mechanisms. However, most graph neural network models do not combine physical attributes such as channel perturbation direction and path angle change, and cannot accurately model the propagation trajectory of interference signals in the graph structure. In addition, existing graph neural network-based methods mostly focus on link quality assessment, node classification, or local interference detection, lacking the ability to trace and identify interference paths from the overall network structure and failing to effectively solve the problem of interference source localization.

[0004] In terms of interference path search, current mainstream methods are mostly based on heuristic search algorithms such as genetic algorithms, ant colony optimization, and simulated annealing. Although these algorithms have certain global search capabilities, due to their lack of perception ability for communication semantics such as perturbation directionality and node heat evolution trend in the graph structure, they often can only obtain locally optimal paths, and the search efficiency significantly decreases in the case of multi-source interference and high-dimensional path spaces. At the same time, most of these methods do not design a strategy update mechanism linked with neural network models and lack system-level closed-loop feedback capabilities, unable to dynamically adjust search behaviors according to the system operation results.

[0005] In addition, the existing technologies still mainly rely on fixed strategies in the control of communication hardware. Typical solutions include fixed-bandwidth filtering, amplifier gain control based on level thresholds, etc. Such control strategies have poor flexibility and are difficult to automatically adjust the compensation loop and filtering configuration according to the dynamic changes of the actual interference path, resulting in a disconnection between the control means and the recognition results, a lag in the overall system response, and a lack of end-to-end capabilities for accurate path-level interference positioning - adaptive suppression - stable recovery.

[0006] Facing the above problems, the traditional technologies have the following prominent defects: First, there is a lack of a unified modeling method that can integrate communication graph structure, channel state, and interference propagation characteristics; second, the graph neural network model fails to effectively perceive the changes in the interference propagation direction and multi-scale adjacency characteristics, and cannot form a deep understanding of the interference path; third, the interference path search lacks a dynamic perception mechanism and a feedback-driven strategy evolution ability; fourth, there is no close linkage between the recognition result and the hardware control, and the system lacks the closed-loop self-learning ability.

[0007] Therefore, how to provide a dual-mode carrier communication aggregation interference recognition system and device based on graph neural network is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0008] An object of the present invention is to propose a dual-mode carrier communication aggregation interference recognition system and device based on graph neural network. The present invention integrates a multi-layer rotating perception graph neural network and an improved fruit fly optimization algorithm to construct a communication graph with rich dynamic structure features, realizing accurate recognition of interference paths and interference source positioning in a dual-mode carrier communication system. The system has the advantages of fine interference modeling, efficient path search, accurate source point judgment, fast control response, and strong strategy adaptability, can effectively improve the stability and robustness of the communication link, is applicable to interference suppression application scenarios of power line communication, industrial field bus, and intelligent communication systems in complex environments, and has good engineering application prospects and promotion value.

[0009] The dual-mode carrier communication aggregation interference recognition system based on graph neural network according to an embodiment of the present invention includes the following modules:

[0010] A communication graph construction module, configured to identify various nodes and communication links in the dual-mode communication system through a channel detection device, construct a communication topology graph, and configure node dynamic attributes and perturbation direction labels;

[0011] A graph neural network modeling module, configured to construct a multi-layer rotating perception graph neural network, perceive the structural connection relationship and direction changes, output the node interference heat value, and generate an interference heat map;

[0012] The interference path search module is used to initialize multiple fruit fly individuals within the high-heat region of the heat map, and perform a jumping search using an improved fruit fly optimization algorithm to generate a candidate set of interference paths;

[0013] The path evaluation and source point positioning module is used to evaluate the heat coverage and propagation rationality of the candidate set of interference paths, and determine the optimal interference path and the position of the interference source;

[0014] The linkage control module is used to control the hardware devices of the filter, compensation circuit, and buffer amplifier, and dynamically implement interference suppression and path switching operations;

[0015] The self-learning update module is used to collect the signal quality data of the communication link, and feedback it to the multi-layer rotating perception graph neural network and the improved fruit fly optimization algorithm to update the interference heat map generation mechanism and the fruit fly jumping strategy.

[0016] The dual-mode carrier communication aggregation interference recognition method based on a graph neural network according to an embodiment of the present invention includes the following steps:

[0017] S1. Construct a communication topology graph through a channel detection device, and set dynamic attributes and perturbation direction labels to form a communication graph;

[0018] S2. Construct a multi-layer rotating perception graph neural network, and input the communication graph into the multi-layer rotating perception graph neural network to perceive the connection relationship and direction changes in the communication graph, extract multi-scale perturbation propagation features, and output an interference heat map;

[0019] S3. Use an improved fruit fly optimization algorithm to initialize multiple fruit fly individuals in the high-heat area of the interference heat map, and perform a jumping search based on the heat value, path direction, and perturbation characteristics to generate a candidate set of interference paths;

[0020] S4. According to the interference path heat coverage and propagation rationality indicators, evaluate the candidate set of interference paths to determine the optimal interference path and the position of the interference source;

[0021] S5. According to the optimal interference path and the position of the interference source, control the communication hardware in a linkage manner, dynamically adjust the compensation circuit and the buffer amplifier, and achieve interference suppression and stable maintenance of the communication link;

[0022] S6. Real-time collect the signal quality data of the communication link, and feedback it to the multi-layer rotating perception graph neural network and the improved fruit fly optimization algorithm for updating the interference heat map generation mechanism and the fruit fly jumping strategy.

[0023] Optionally, the S1 specifically includes:

[0024] S11. Perform an initial scan on the dual-mode communication system through a channel detection device, identify all terminal nodes, relay nodes, and communication coupling devices, and construct a node set;

[0025] S12. Identify the effective communication links between each pair of nodes according to the physical connection topology relationship, represent the effective communication link between each pair of nodes as an edge, and form a communication topology graph;

[0026] S13. Collect and label the channel parameters on each edge of the communication topology graph, including real-time instantaneous attenuation, interference intensity, frequency stability, and path propagation delay;

[0027] S14. Configure the dynamic working characteristics of each node during the communication process. The dynamic working characteristics include the real-time voltage, current, harmonic amplitude index of the node, and the signal-to-noise ratio of the connected path, which are used to reflect the operating state of the node and its sensitivity to interference;

[0028] S15. Further define the perturbation direction label on each edge to quantify the direction change of the interference signal on the propagation path, and record the perturbation direction angle as the perturbation direction label:

[0029]

[0030] where, θ ij represents the perturbation direction angle of edge e ij , represents the direction estimation of the interference electric field vector on edge e ij at time t + Δt, represents the direction estimation of the interference electric field vector on edge e ij at time t, ||·|| represents the norm, represents the spatial position vector of the j-th node, represents the spatial position vector of the i-th node, ∈ represents the smoothing factor, λ represents the weight coefficient, S ij represents the channel stability index on edge e ij , |·| represents the absolute value operation;

[0031] S16. Obtain a communication graph that simultaneously includes node physical states, edge channel attributes, and perturbation direction labels.

[0032] Optionally, the specific steps of S2 include:

[0033] S21. Construct a multi-layer rotation-aware graph neural network. The multi-layer rotation-aware graph neural network consists of L graph convolutional units. Each graph convolutional unit contains an adjacency awareness module and a direction awareness module. The adjacency awareness module is used to extract the structural connection relationships in the graph, and the direction awareness module is used to capture the edge direction change and perturbation propagation angle features;

[0034] S22. Initialize the node attribute set of the communication graph as the input feature matrix X (0), where each row represents the original dynamic features of a node and is input into the first-layer graph convolution unit;

[0035] S23. In the l-th layer graph convolution unit, let the neighbor set of node v i be N(v i ). The node features are updated using a rotation-aware convolution operator combined with the perturbation direction:

[0036]

[0037] where, represents the feature vector of the i-th node in the l-th layer, σ represents the activation function, W (l) represents the adjacency-aware module weight matrix, represents the feature vector of the j-th node in the (l - 1)-th layer, γ (l) represents the direction adjustment coefficient, θ ij represents the perturbation direction angle of edge e ij , D ij represents the direction embedding vector of the edge, α ij represents the perturbation intensity normalization coefficient;

[0038] S24. In each layer, a direction-aware module is used to capture the edge direction change and perturbation propagation angle features and update them, and finally the node feature matrix X (L) of the L-th layer is output;

[0039] S25. The node features output by multiple network layers are concatenated and weighted aggregated to form a final fusion feature matrix to capture the interference propagation patterns at different scales;

[0040] S26. Based on the final fusion feature matrix, the interference heat values of each node are obtained using a fully connected mapping and normalization operation, and an interference heat map is generated to characterize the possibility of each node in the communication graph being interfered with.

[0041] Optionally, the specific content of S3 includes:

[0042] S31. Based on the interference heat map, the node set with a heat value higher than the threshold τ h is extracted as the candidate interference source area, and multiple fruit fly individuals are initialized in the candidate interference source area, and each fruit fly individual corresponds to a search starting point of a potential interference path;

[0043] S32. Using an improved fruit fly optimization algorithm, the improvement lies in introducing a multi-factor perception function to replace the search mechanism based on a single value judgment of the objective function:

[0044]

[0045] where, F kdenotes the k-th fruit fly individual, and J(F k ) represents the comprehensive perception function value calculated by the k-th fruit fly individual in the current jump decision. η represents the jump amplitude adjustment factor, and h k represents the interference heat of the node where the k-th fruit fly individual is currently located, and ρ kj represents the consistency coefficient between the current path direction and the historical perturbation direction, and I kj represents the interference intensity, and SNR kj represents the signal-to-noise ratio, and θ kj represents the perturbation direction angle of edge e kj ;

[0046] S33. The fruit fly individual selects a local neighborhood based on the multi-factor perception function and performs a jumping search operation, jumping to the adjacent node that maximizes J(F k ), and at the same time updates the trajectory sequence and direction history information of the fruit fly individual;

[0047] S34. To overcome the problem that the traditional fruit fly algorithm has slow path convergence or gets stuck in the hot spot center in the high-dimensional graph structure, an adaptive perturbation adjustment strategy is set. When the heat value oscillates or the perturbation direction suddenly reverses during the continuous jumps of the fruit fly individual, the direction perturbation update mechanism is triggered:

[0048]

[0049] where, represents the individual's current direction perception variable, α represents the perturbation amplitude adjustment factor, and rand(·) represents the random number generation function;

[0050] S35. The jumping trajectory of each fruit fly individual is recorded as a path set in sequence, and the path sequences of all individuals form a candidate set of interference paths.

[0051] Optionally, the specific steps of S4 include:

[0052] S41. Evaluate each interference path in the candidate set of interference paths P = {P1, P2,..., P N}, where P i represents the i-th interference path, N represents the total number of interference paths, and calculate the interference heat coverage and perturbation propagation rationality score for each interference path respectively;

[0053] S42. The interference heat coverage is defined as the normalized mean of the interference heat values of each node on the path:

[0054]

[0055] where, Ω i represents the interference heat coverage of the i-th interference path, and |P i| represents the length of the i-th interference path, h j represents the interference heat value of node v j on the interference path, and max(h j ) represents the maximum interference heat value of all nodes in the i-th interference path;

[0056] S43. Comprehensively evaluate the path direction continuity, consistency, and the change trend of the perturbation intensity, and calculate the perturbation propagation rationality score:

[0057]

[0058] where Ψ i represents the perturbation propagation rationality score of the i-th interference path, θ mn represents the perturbation direction angle of edge e mn , I mn represents the interference intensity of edge e mn , and δf mn represents the frequency drift;

[0059] S44. Normalize the interference heat coverage and the perturbation propagation rationality score and combine them into a path comprehensive score function:

[0060] Φ i = β1·Ω i + β2·Ψ i ;

[0061] where Φ i represents the comprehensive score of the i-th interference path, and β1 and β2 represent the weight coefficients;

[0062] S45. Sort all the interference paths P i according to the comprehensive score, select the interference path with the highest comprehensive score as the optimal interference path, and record the starting node of the optimal interference path as the current interference source location;

[0063] S46. Output the optimal interference path and the interference source location.

[0064] Optionally, the S5 specifically includes:

[0065] S51. Activate the hardware linkage control interface according to the optimal interference path and the interference source location, and send control instructions to the hardware modules of the communication circuit through the hardware control bus to activate the filters, compensation circuits, and buffer amplifiers along the interference path;

[0066] S52. Based on the interference signal spectrum characteristics on the optimal interference path, select the filtering parameters corresponding to the interference frequency band to suppress the interference signals in the interference frequency band from entering the communication receiving unit;

[0067] S53. Monitor the node voltage, current, and harmonic component data on the optimal interference path in real time. According to the node impedance characteristics, automatically adjust the capacitance and inductance units of the compensation circuit to meet the impedance balance requirements during signal transmission, and suppress signal reflection and channel distortion.

[0068] S54. Activate the feedback-type compensation circuit, measure the signal phase and amplitude characteristics in the communication link, dynamically adjust the transistors, switching diodes, and phase shift circuits in the compensation circuit, and generate a compensation signal in real time that is similar in amplitude and opposite in phase to the interference signal.

[0069] S55. Collect the signal quality data on the optimal interference path in real time. When it is found that the communication signal level drops, the waveform is distorted, or the bit error rate increases, automatically trigger the buffer amplifier adjustment program, and restore the interfered low-level communication signal by adjusting the amplifier gain parameter, and perform waveform repair and signal standardization output.

[0070] S56. If the communication link quality does not reach the preset index after the interference suppression control in steps S51 - S55, according to the predefined alternative communication path data, automatically execute the communication path switching mechanism and switch to the sub-optimal path for signal transmission.

[0071] Optionally, the specific content of S6 includes:

[0072] S61. Collect the signal quality data of the communication link after the hardware linkage control in real time, and establish a real-time communication quality database.

[0073] S62. Use the latest data in the real-time communication quality database to calculate the actual interference response deviation value of the communication node, and compare it with the heat value of the corresponding node in the interference heat map generated in step S2 to obtain the node heat error set.

[0074] S63. According to the node heat error set, dynamically adjust the weight matrix and direction adjustment coefficient of the adjacency perception module of the multi-layer rotating perception graph neural network through the backpropagation mechanism.

[0075] S64. Analyze the path perturbation direction change characteristics in the real-time communication quality data, calculate the deviation between the actual direction angle change of path propagation and the perturbation direction angle during the fruit fly jumping search process, and adjust the direction perception variable and jump amplitude adjustment factor in the fruit fly individual jumping strategy based on the direction deviation, and dynamically update and improve the multi-factor perception function J(F k )

[0076] S65. According to the updated multi-layer rotating perception graph neural network parameters and the improved fruit fly optimization algorithm strategy, re-execute steps S2 and S3 to generate a new interference heat map and an interference path candidate set, and realize the self-learning and strategy iteration ability.

[0077] Optionally, the communication link signal quality data includes node signal-to-noise ratio, bit error rate, interference intensity, voltage, and current variation.

[0078] The dual-mode carrier communication aggregation interference recognition device based on a graph neural network according to an embodiment of the present invention is used to execute each module in the dual-mode carrier communication aggregation interference recognition system based on a graph neural network, and realizes communication graph construction, graph neural network modeling, interference path search, path evaluation and source point positioning, linkage control, and self-learning update operations in a hardware or software manner, so as to complete the recognition and suppression of interference paths in a dual-mode carrier communication environment.

[0079] The beneficial effects of the present invention are as follows:

[0080] First, the present invention realizes high-fidelity modeling of the communication structure in a complex electromagnetic interference environment by constructing a communication topology graph including dynamic node attributes, channel states, and perturbation direction labels. Compared with traditional modeling methods that rely on static topologies or channel thresholds, the present invention introduces key features such as the perturbation propagation direction angle at the node and edge levels, which can more comprehensively depict the propagation trajectory of interference signals in physical paths and provide an accurate structural basis for subsequent recognition.

[0081] Second, relying on a multi-layer rotation perception graph neural network model, the present invention realizes in-depth perception of connection relationships and direction changes in the communication graph. By setting an adjacency perception module and a direction perception module, this model can fuse topological information between nodes and perturbation angle features of edges at multiple scales, thereby more effectively extracting pattern features of interference propagation. The output interference heat map not only reflects the possibility of each node being interfered with but also provides an accurate heat zone positioning basis for the path recognition process, significantly improving the focusing ability in the area around the interference source.

[0082] In addition, in terms of the path search mechanism, the present invention introduces an improved fruit fly optimization algorithm. By comprehensively considering information such as node heat values, direction consistency, channel interference intensity, and signal-to-noise ratio through a multi-factor perception function, a jumping search strategy with direction guidance and environmental adaptability is constructed. This method overcomes problems such as slow convergence and easy entrapment in local optima of traditional heuristic algorithms, enabling search individuals to quickly converge to the true interference path in a complex graph structure. During the path evaluation process, heat coverage and propagation rationality scores are fused, and the global optimality and physical rationality of the recognition result are ensured through a comprehensive evaluation mechanism, effectively improving the accuracy of path recognition.

[0083] More critically, the system of the present invention tightly couples the recognition result with the hardware response process through the linkage control module, achieving dynamic regulation of devices such as filters, compensation circuits, and buffer amplifiers. This linkage control ability can not only implement targeted suppression measures for different types of interference signals but also support automatic switching to a sub-optimal path when the communication quality deteriorates, ensuring the continuous and stable operation of the communication link. Finally, with the help of the self-learning update module, the system can continuously collect communication quality data, dynamically update the parameters of the graph neural network model and the fruit fly jumping strategy, gradually evolve the recognition model and control strategy, and achieve rapid adaptation and response to new interference patterns, thereby constructing an intelligent communication interference recognition platform with learning ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0085] Figure 1 is a schematic structural diagram of a dual-mode carrier communication aggregation interference recognition system based on a graph neural network proposed by the present invention;

[0086] Figure 2 is an overall flowchart of a dual-mode carrier communication aggregation interference recognition method based on a graph neural network proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0087] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0088] Referring to Figure 1 , a dual-mode carrier communication aggregation interference recognition system based on a graph neural network includes the following modules:

[0089] A communication graph construction module, configured to identify various nodes and communication links in a dual-mode communication system through a channel detection device, construct a communication topology graph, and configure node dynamic attributes and perturbation direction labels;

[0090] A graph neural network modeling module, configured to construct a multi-layer rotation perception graph neural network, perceive the structural connection relationship and direction change, output the node interference heat value, and generate an interference heat map;

[0091] An interference path search module, configured to initialize multiple fruit fly individuals in the high-heat area of the heat map, and perform a jumping search using an improved fruit fly optimization algorithm to generate a candidate set of interference paths;

[0092] A path evaluation and source location module, which is used to evaluate the heat coverage and propagation rationality of the interference path candidate set, and determine the optimal interference path and the location of the interference source;

[0093] A linkage control module, which is used to control the filter, compensation circuit and buffer amplifier hardware devices, and dynamically implement interference suppression and path switching operations;

[0094] A self-learning and updating module, which is used to collect the communication link signal quality data, and feedback it to the multi-layer rotating perception graph neural network and the improved fruit fly optimization algorithm, and update the interference heat map generation mechanism and the fruit fly jumping strategy.

[0095] By integrating the six major modules of communication graph construction, graph neural network modeling, path search, path evaluation, linkage control and self-learning update into the same system architecture, this system constructs a dual-mode carrier communication interference recognition system with complete functions, clear structure and closed-loop iterative optimization. In the graph modeling stage, the system fully considers the node dynamic characteristics and perturbation directions, and combines the multi-layer graph neural network and the improved fruit fly optimization algorithm to realize the full-process automation from interference heat map generation to optimal path recognition and then to accurate interference source location, significantly improving the recognition accuracy and system adaptability. Especially in the face of actual working conditions such as high noise, frequency drift and complex path interference, the system can autonomously optimize the recognition strategy and control mechanism based on communication data, and has high deployability and engineering application value.

[0096] Reference Figure 2 , the dual-mode carrier communication aggregation interference recognition method based on graph neural network includes the following steps:

[0097] S1. Construct a communication topology graph through a channel detection device, and set dynamic attributes and perturbation direction labels to form a communication graph;

[0098] S2. Construct a multi-layer rotating perception graph neural network, and input the communication graph into the multi-layer rotating perception graph neural network to perceive the connection relationship and direction change in the communication graph, extract multi-scale perturbation propagation features, and output an interference heat map;

[0099] S3. Use the improved fruit fly optimization algorithm to initialize multiple fruit fly individuals in the high-heat area of the interference heat map, and perform a jumping search according to the heat value, path direction and perturbation characteristics to generate an interference path candidate set;

[0100] S4. According to the interference path heat coverage and propagation rationality indicators, evaluate the interference path candidate set to determine the optimal interference path and the location of the interference source;

[0101] S5. According to the optimal interference path and the location of the interference source, link and control the communication hardware, and dynamically adjust the compensation circuit and buffer amplifier to achieve interference suppression and stable maintenance of the communication link;

[0102] S6. Collect the signal quality data of the communication link in real time and feed it back to the multi-layer rotating perception graph neural network and the improved fruit fly optimization algorithm for updating the interference heat map generation mechanism and the fruit fly jumping strategy.

[0103] This method clarifies the collaborative operation logic between functional modules, constructs an end-to-end interference recognition and response method process. Starting from communication graph construction, through graph neural network perception and heat map output, fruit fly optimization search path, path score evaluation, linkage hardware control, to final self-learning feedback update, a complete closed-loop execution chain is formed. This method not only improves the real-time performance and traceability of interference recognition, but also highly integrates recognition and control, enabling the system to have the continuous operation ability from detection to decision-making and then to response, greatly enhancing the system's autonomous adjustment and stability maintenance performance in the face of complex dynamic interference environments.

[0104] In this embodiment, the S1 specifically includes:

[0105] S11. Initialize the scan of the dual-mode communication system through a channel detection device, identify all terminal nodes, relay nodes and communication coupling devices, and construct a node set;

[0106] S12. Identify the effective communication links between each node according to the physical connection topology relationship, represent the effective communication link between each pair of nodes as an edge, and form a communication topology graph;

[0107] S13. Collect and label the channel parameters on each edge of the communication topology graph, including real-time instantaneous attenuation, interference intensity, frequency stability and path propagation delay;

[0108] S14. Configure the dynamic working characteristics of each node during communication. The dynamic working characteristics include the node's real-time voltage, current, harmonic amplitude index and the signal-to-noise ratio of the connected path, which are used to reflect the operating state of the node and its sensitivity to interference;

[0109] S15. Further define the perturbation direction label on each edge to quantify the direction change of the interference signal on the propagation path, and record the perturbation direction angle as the perturbation direction label:

[0110]

[0111] where θ ij represents the perturbation direction angle of edge e ij , represents the direction estimation of the interference electric field vector on edge e ij at time t + Δt, represents the direction of the interference electric field vector on edge e ijDirection estimation on it, ||·|| represents the norm, represents the spatial position vector of the j-th node, represents the spatial position vector of the i-th node, ∈ represents the smoothing factor, λ represents the weight coefficient, S ij represents the edge e ij on the channel stability index, |·| represents the absolute value operation;

[0112] S16. Obtain a communication graph that simultaneously includes node physical states, edge channel attributes, and disturbance direction labels.

[0113] Steps S11 - S16 systematically refine the communication graph construction steps. By defining the node set, edge set, channel parameters, and disturbance direction angle labels, a structured communication graph with dynamic attributes and direction propagation characteristics is constructed, providing high-quality data support for subsequent graph neural network model training and interference path modeling. Different from traditional methods that only use static topology graphs or channel strength graphs, this communication graph introduces multi-dimensional attributes such as disturbance direction angles, frequency stability, and path propagation delay in its structure, which can accurately express the spatio-temporal trajectory of interference propagation in the system and effectively improve the model's ability to identify interference path with sudden direction changes.

[0114] In this embodiment, the specific content of S2 includes:

[0115] S21. Construct a multi-layer rotation-aware graph neural network, which is composed of L layers of graph convolution units. Each graph convolution unit contains an adjacency perception module and a direction perception module. The adjacency perception module is used to extract the structural connection relationships in the graph, and the direction perception module is used to capture the edge direction changes and disturbance propagation angle characteristics;

[0116] S22. Initialize the node attribute set of the communication graph as the input feature matrix X (0) , where each row represents the original dynamic features of a node, and input it into the first layer of graph convolution unit;

[0117] S23. In the l-th layer of graph convolution unit, let the neighbor set of node v i be N(v i ), and use the rotation-aware convolution operator combined with the disturbance direction to update the node features:

[0118]

[0119] Among them, represents the feature vector of the i-th node in the l-th layer, σ represents the activation function, W (l) represents the weight matrix of the adjacency perception module, represents the feature vector of the j-th node in the (l - 1)-th layer, γ (l)Represents the direction adjustment coefficient, θ ij Represents the edge e ij The disturbance direction angle of, D ij Represents the direction embedding vector of the edge, α ij Represents the disturbance intensity normalization coefficient;

[0120] S24. In each layer, a direction-aware module is adopted to capture the edge direction change and disturbance propagation angle features, and update them. Finally, the node feature matrix X of the L-th layer is output (L) ;

[0121] S25. Concatenate and weighted aggregate the node features output by multiple network layers to form the final fusion feature matrix, and capture the interference propagation patterns at different scales;

[0122] S26. Based on the final fusion feature matrix, use the fully connected mapping and normalization operations to obtain the interference heat values of each node, and generate the interference heat map, which is used to characterize the possibility of each node in the communication graph being interfered with.

[0123] The above steps introduce a multi-layer rotation-aware graph neural network structure at the graph neural network modeling level, innovatively combine the adjacency-aware and direction-aware modules, and realize considering both the topological connectivity of the graph structure and the characteristics of edge disturbance direction changes during the node feature extraction process. This model can fully fuse multi-scale neighborhood information, improve the modeling ability of direction deflection, signal attenuation, and propagation mode evolution in multi-hop interference paths. The output interference heat map has higher accuracy and more reasonable hot zone determination, provides more reliable initial guidance data for the subsequent fruit fly search algorithm, and significantly improves the upstream and downstream collaboration performance of the overall recognition system.

[0124] In this embodiment, the specific content of S3 is as follows:

[0125] S31. Based on the interference heat map, extract the node set with heat values higher than the threshold τ h as the candidate interference source area, and initialize multiple fruit fly individuals in the candidate interference source area. Each fruit fly individual corresponds to a search starting point of a potential interference path;

[0126] S32. Use the improved fruit fly optimization algorithm, and the improvement lies in introducing a multi-factor perception function to replace the search mechanism based on a single value judgment of the objective function:

[0127]

[0128] Among them, F k represents the k-th fruit fly individual, J(F k ) represents the comprehensive perception function value calculated by the k-th fruit fly individual in the current jump decision, η represents the jump amplitude adjustment factor, h kDenote the interference heat of the current node where the k-th fruit fly individual is located, ρ kj Denote the consistency coefficient of the current path direction and the historical perturbation direction, I kj Denote the interference intensity, SNR kj Denote the signal-to-noise ratio, θ kj Denote the edge e kj The perturbation direction angle of;

[0129] S33. The fruit fly individuals perform local neighborhood selection based on the multi-factor perception function and execute a leapfrog search operation, leaping to the adjacent node that maximizes J(F k ), and at the same time update the trajectory sequence and direction history information of the fruit fly individuals;

[0130] S34. To overcome the problem that the traditional fruit fly algorithm has slow path convergence or gets stuck in the hot spot center in the high-dimensional graph structure, an adaptive perturbation adjustment strategy is set. When the heat value oscillates or the perturbation direction reverses sharply during the continuous leaping of the fruit fly individuals, the direction perturbation update mechanism is triggered:

[0131]

[0132] Among them, Denote the individual's current direction perception variable, α denotes the perturbation amplitude adjustment factor, and rand(·) denotes the random number generation function;

[0133] S35. The leaping trajectories of each fruit fly individual are recorded in sequence as a path set, and the path sequences of all individuals form an interference path candidate set.

[0134] By improving the fruit fly optimization algorithm and introducing a multi-factor perception function to integrate key features such as heat value, path direction consistency, interference intensity, and signal-to-noise ratio, a more efficient and stronger direction-guided leapfrog path search mechanism is realized. Compared with the traditional search algorithm, this method not only avoids the problem of getting stuck in local optima, but also improves the path diversity and convergence speed by introducing a perturbation adjustment mechanism. The search process of fruit fly individuals is deeply combined with the communication heat map, effectively realizing the rapid locking of the real interference propagation path in the graph structure and providing a more feasible candidate set for subsequent path evaluation.

[0135] In this embodiment, the S4 specifically includes:

[0136] S41. Evaluate each interference path in the interference path candidate set P = {P1, P2,..., P N}, where P i Denote the i-th interference path, N denotes the total number of interference paths, and calculate the interference heat coverage and perturbation propagation rationality score for each interference path respectively;

[0137] S42. The interference heat coverage is defined as the normalized mean of the interference heat values of each node on the path:

[0138]

[0139] Among them, Ω i represents the interference heat coverage of the i-th interference path, |P i | represents the length of the i-th interference path, h j represents the interference heat value of the node v j on the interference path, and max(h j ) represents the maximum interference heat value of all nodes in the i-th interference path;

[0140] S43. Comprehensively evaluate the path direction continuity, consistency and the change trend of the perturbation intensity, and calculate the perturbation propagation rationality score:

[0141]

[0142] Among them, Ψ i represents the perturbation propagation rationality score of the i-th interference path, θ mn represents the perturbation direction angle of the edge e mn , and I mn represents the interference intensity of the edge e mn , and δf mn represents the frequency drift;

[0143] S44. Normalize the interference heat coverage and the perturbation propagation rationality score and combine them into a path comprehensive score function:

[0144] Φ i = β1·Ω i + β2·Ψ i ;

[0145] Among them, Φ i represents the comprehensive score of the i-th interference path, and β1 and β2 represent the weight coefficients;

[0146] S45. Sort all the interference paths P i according to the comprehensive score, select the interference path with the highest comprehensive score as the optimal interference path, and record the starting node of the optimal interference path as the current interference source location;

[0147] S46. Output the optimal interference path and the interference source location.

[0148] The above steps propose a path evaluation mechanism based on interference heat coverage and propagation rationality scores, which can scientifically quantify the advantages and disadvantages of different candidate paths in terms of interference intensity, direction consistency, and structural coherence. By forming a comprehensive scoring function through weighted combination, the optimal interference path is selected. Different from traditional algorithms that use the shortest path or the path with the maximum local heat as the goal, this method comprehensively considers the global quality of the path and propagation rationality, making the selected path not only match the actual logic of interference propagation but also facilitate subsequent execution of suppression control. The introduction of this step significantly improves the stability and credibility of interference recognition results.

[0149] In this embodiment, S5 specifically includes:

[0150] S51. Activate the hardware linkage control interface according to the optimal interference path and the interference source location, and send control instructions to the hardware modules of the communication circuit through the hardware control bus to activate the filters, compensation circuits, and buffer amplifiers along the interference path;

[0151] S52. Based on the spectral characteristics of the interference signals on the optimal interference path, select the filtering parameters corresponding to the interference frequency band to suppress the interference signals in the interference frequency band from entering the communication receiving unit;

[0152] S53. Real-time monitor the node voltage, current, and harmonic component data on the optimal interference path, and automatically adjust the capacitance and inductance units of the compensation circuit according to the node impedance characteristics to meet the impedance balance requirements during signal transmission and suppress signal reflection and channel distortion;

[0153] S54. Activate the feedback-type compensation circuit, measure the signal phase and amplitude characteristics in the communication link, and dynamically adjust the transistors, switching diodes, and phase shift circuits in the compensation circuit to generate compensation signals with an amplitude similar to and a phase opposite to that of the interference signals in real time;

[0154] S55. Real-time collect the signal quality data on the optimal interference path. When it is found that the communication signal level drops, the waveform is distorted, or the bit error rate increases, automatically trigger the buffer amplifier adjustment program, and restore the interfered low-level communication signal by adjusting the amplifier gain parameters for waveform repair and signal standardization output;

[0155] S56. If the communication link quality does not reach the preset index after the interference suppression control in steps S51 - S55, automatically execute the communication path switching mechanism according to the predefined alternative communication path data and switch to the sub-optimal path for signal transmission.

[0156] The above steps propose a linkage control mechanism based on the optimal interference path and source point information. The system can dynamically adjust the filter frequency band parameters, the capacitance and inductance values of the compensation circuit, and the gain configuration of the buffer amplifier, forming a directional suppression control ability for the interference path. The system can real-time sense the interference situation in the communication link and automatically implement interference cancellation and waveform repair operations based on the current path state. If the path control fails to be recognized, the communication path switching mechanism can also be triggered to ensure the overall communication continuity of the system.

[0157] In this embodiment, step S6 specifically includes:

[0158] S61. Real-time collect the communication link signal quality data after hardware linkage control and establish a real-time communication quality database;

[0159] S62. Use the latest data in the real-time communication quality database to calculate the actual interference response deviation value of the communication node, and compare it with the heat value of the corresponding node in the interference heat map generated in step S2 to obtain a node heat error set;

[0160] S63. According to the node heat error set, dynamically adjust the weight matrix and direction adjustment coefficient of the adjacency perception module of the multi-layer rotating perception graph neural network through the backpropagation mechanism;

[0161] S64. Analyze the path perturbation direction change characteristics in the real-time communication quality data, calculate the deviation between the actual direction angle change of path propagation and the perturbation direction angle in the fruit fly jumping search process, and based on the direction deviation, adjust the direction perception variable and jump amplitude adjustment factor in the fruit fly individual jumping strategy, and dynamically update and improve the multi-factor perception function J(F k ) of the fruit fly optimization algorithm;

[0162] S65. According to the updated multi-layer rotating perception graph neural network parameters and the improved fruit fly optimization algorithm strategy, re-execute step S2 and step S3 to generate a new interference heat map and an interference path candidate set, realizing the self-learning and strategy iteration ability.

[0163] The above steps establish a system self-learning feedback mechanism, enabling the communication quality data to be not only used for performance monitoring but also as a dynamic input source for model optimization. By feeding back the node heat prediction error to the graph neural network and calibrating the fruit fly jumping strategy in combination with the direction deviation, the system realizes the functional leap from static inference to dynamic evolution. This mechanism ensures that the system can continuously adapt to new interference patterns, path changes, and network topology adjustments during long-term operation, truly realizing the evolutionary ability of the interference recognition strategy, and significantly enhancing the robustness and generalization ability of the system in complex scenarios.

[0164] In this embodiment, the communication link signal quality data includes node signal-to-noise ratio, bit error rate, interference intensity, voltage, and current changes.

[0165] A dual-mode carrier communication aggregation interference recognition device based on a graph neural network is used to execute each module in the dual-mode carrier communication aggregation interference recognition system based on a graph neural network, and realizes communication graph construction, graph neural network modeling, interference path search, path evaluation and source point positioning, linkage control, and self-learning update operations in a hardware or software manner, so as to complete the recognition and suppression of interference paths in the dual-mode carrier communication environment.

[0166] Example 1:

[0167] To verify the feasibility of the present invention in implementation, the present invention is applied to the communication network of a certain power substation. This site is located in the East China region and is an important node for load dispatching in the region. It uses dual-mode carrier communication to collect remote data and send control instructions. Due to the complex on-site equipment, dense cables, and frequent operation of high-voltage equipment, the communication link is frequently interfered with, including harmonic disturbances from inverters, electromagnetic transient pulses from switchgear, and occasional crosstalk signals, resulting in obvious fluctuations in communication reliability. Traditional interference processing methods based on fixed thresholds and rule libraries are difficult to effectively identify interference paths and even more difficult to locate the interference source point in a timely manner, causing frequent abnormal alarms in the system and a high data delay rate, seriously affecting the scheduling stability of the station control system.

[0168] In this field, we deployed the complete recognition and control process of the system of the present invention. First, channel detection devices are connected to the dual-mode communication network connecting the main control layer of the substation and primary equipment to collect the connection relationships and dynamic operating parameters between each terminal node, relay device, and communication coupler in real time, and construct a communication topology map with information such as node voltage, current, and harmonic amplitude. At the same time, disturbance direction labels are introduced on each path edge in the graph to depict the propagation trajectory characteristics of signal direction jumps under multi-source interference conditions. This topology map is then input into the multi-layer rotating perception graph neural network of the present system, and the network model extracts the connection pattern and the relationship between the disturbance propagation direction in the graph structure and outputs an interference heat map of the entire station. Through the visualization interface, the areas where interference may occur currently or will occur soon can be intuitively identified.

[0169] On the interference heat map, the system automatically identifies multiple heat-aggregated areas, initializes fruit fly individuals in these areas, and performs jump path searches based on factors such as heat value, directional continuity, and signal-to-noise ratio changes. In a sudden communication packet loss event, the system quickly locked the link between the No. 1 switch cabinet and the main communication machine in the control room as a high-heat path, and determined through the path evaluation module that the interference heat coverage of the path was 0.78, and the disturbance propagation rationality score was 0.85. The comprehensive score was the highest among all paths. The system determined it as the main interference path and marked its starting point as a candidate area for the interference source. Immediately, the system linkage control module was started, dynamically adjusting the filter parameters on the path, setting the interference frequency band to 8~20kHz, and adjusting the inductance group and capacitance group in the compensation loop to meet the current channel impedance balance requirements. The system also automatically activated the buffer amplifier to repair the signal distortion caused by interference. After one round of adjustment, the communication signal-to-noise ratio increased from the original 19.2dB to 31.5dB, and the bit error rate increased from 1.02×10 -3 Down to 4.37×10 -5 , the communication link resumed stable operation.

[0170] More importantly, the system feeds back the communication quality change data collected in the above process to the graph neural network model and the fruit fly optimization algorithm module in real time, automatically corrects the heat prediction error of some paths, and updates the jump strategy in the next round of interference identification, so that the path identification time is shortened from the initial 2.8 seconds to 1.6 seconds, and the interference source positioning accuracy is improved to 94.3%. The analysis of the field deployment data for 7 consecutive days shows that after the system of the present invention is activated, the number of abnormal alarms of the communication system has decreased by 71.5%, the data delay rate has decreased by an average of 48.9%, and in extreme cases, it has been compressed from more than 3 seconds to less than 1 second, and the overall operation stability of the system has been significantly improved.

[0171] It can be verified through this embodiment that the present invention can not only realize the accurate identification of interference paths in carrier communication links in complex environments, but also can link physical control equipment to implement dynamic interference suppression. At the same time, it has self-learning ability and can continuously optimize the identification strategy during operation. It is particularly suitable for deployment in power systems, rail transit, industrial sites and other scenarios where electromagnetic interference is complex but communication reliability requirements are extremely high.

[0172] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A dual-mode carrier communication aggregation interference recognition system based on a graph neural network, characterized in that It includes the following modules: A communication graph construction module, which is used to identify various nodes and communication links in a dual-mode communication system through a channel detection device, construct a communication topology graph, and configure dynamic node attributes and disturbance direction labels; A graph neural network modeling module, which is used to construct a multi-layer rotation-aware graph neural network, perceive the structural connection relationship and direction changes, output node interference heat values, and generate an interference heat map; An interference path search module, which is used to initialize multiple fruit fly individuals in the high-heat area of the heat map, and use an improved fruit fly optimization algorithm to perform a jumping search to generate a candidate set of interference paths; A path evaluation and source point location module, which is used to evaluate the heat coverage and propagation rationality of the candidate set of interference paths, and determine the optimal interference path and the location of the interference source; A linkage control module, which is used to control hardware devices such as filters, compensation circuits, and buffer amplifiers, and dynamically implement interference suppression and path switching operations; A self-learning update module, which is used to collect communication link signal quality data and feedback it to the multi-layer rotation-aware graph neural network and the improved fruit fly optimization algorithm to update the interference heat map generation mechanism and the fruit fly jumping strategy.

2. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 1, wherein The modules are implemented through the following methods: S1. Construct a communication topology graph through a channel detection device, and set dynamic attributes and disturbance direction labels to form a communication graph; S2. Construct a multi-layer rotation-aware graph neural network, and input the communication graph into the multi-layer rotation-aware graph neural network to perceive the connection relationship and direction changes in the communication graph, extract multi-scale disturbance propagation features, and output an interference heat map; S3. Use the improved fruit fly optimization algorithm to initialize multiple fruit fly individuals in the high-heat area of the interference heat map, and perform a jumping search based on heat values, path directions, and disturbance characteristics to generate a candidate set of interference paths; S4. According to the heat coverage and propagation rationality indicators of the interference paths, evaluate the candidate set of interference paths to determine the optimal interference path and the location of the interference source; S5. According to the optimal interference path and the location of the interference source, control the communication hardware in a linkage manner, dynamically adjust the compensation circuit and the buffer amplifier, and achieve interference suppression and stable maintenance of the communication link; S6. Real-time collect communication link signal quality data and feedback it to the multi-layer rotation-aware graph neural network and the improved fruit fly optimization algorithm for updating the interference heat map generation mechanism and the fruit fly jumping strategy.

3. The dual-mode carrier communication aggregation interference recognition system and device based on a graph neural network according to claim 2, characterized in that, The specific content of S1 includes: S11. Perform an initial scan on the dual-mode communication system through a channel detection device, identify all terminal nodes, relay nodes, and communication coupling devices, and construct a node set; S12. Identify the effective communication links between each pair of nodes according to the physical connection topology relationship, represent the effective communication links between each pair of nodes as edges, and form a communication topology graph; S13. Collect and label channel parameters on each edge of the communication topology graph, including real-time instantaneous attenuation, interference intensity, frequency stability, and path propagation delay; S14. Configure dynamic working characteristics for each node during the communication process. The dynamic working characteristics include the real-time voltage, current, harmonic amplitude index of the node, and the signal-to-noise ratio of the connected path, which are used to reflect the operating state of the node and its sensitivity to interference; S15. Further define a perturbation direction label on each edge to quantify the direction change of the interference signal on the propagation path, and record the perturbation direction angle as the perturbation direction label: Among them, θ ij represents the perturbation direction angle of edge e ij , represents the direction estimation of the interference electric field vector on edge e ij at time t + Δt, represents the direction estimation of the interference electric field vector on edge e ij at time t, ||·|| represents the norm, represents the spatial position vector of the j-th node, represents the spatial position vector of the i-th node, ∈ represents the smoothing factor, λ represents the weight coefficient, S ij represents the channel stability index on edge e ij , |·| represents the absolute value operation; S16. Obtain a communication graph that simultaneously includes the physical state of nodes, edge channel attributes, and perturbation direction labels.

4. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 2, characterized in that, The specific steps of S2 are as follows: S21. Construct a multi-layer rotation-aware graph neural network, which consists of L graph convolution units. Each graph convolution unit contains an adjacency-aware module and a direction-aware module. The adjacency-aware module is used to extract the structural connection relationships in the graph, and the direction-aware module is used to capture the edge direction change and perturbation propagation angle features; S22. Initialize the node attribute set of the communication graph as the input feature matrix X (0) , where each row represents the original dynamic features of a node and is input into the first-layer graph convolution unit; S23. In the l-th layer graph convolutional unit, let the neighbor set of node v i be N(v i ), and use the rotation-aware convolutional operator combined with the perturbation direction to update the node features: Among them, represents the feature vector of the $i$-th node in the $l$-th layer, $\sigma$ represents the activation function, and $W$ (l) represents the weight matrix of the adjacency perception module, represents the feature vector of the $j$-th node in the $(l - 1)$-th layer, and $\gamma$ (l) represents the direction adjustment coefficient, and $\theta$ ij represents the perturbation direction angle of the edge $e$ ij , and $D$ ij represents the direction embedding vector of the edge, and $\alpha$ ij represents the perturbation intensity normalization coefficient; S24. In each layer, a direction-aware module is used to capture the edge direction change and the disturbance propagation angle feature, and update them, and finally output the node feature matrix X of the L-th layer (L) ; S25. Concatenate and weighted aggregate the node features output by multiple network layers to form a final fusion feature matrix, and capture the interference propagation patterns at different scales; S26. Based on the final fusion feature matrix, use a fully connected mapping and normalization operation to obtain the interference heat value of each node, and generate an interference heat map to represent the probability of each node in the communication graph being interfered with.

5. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 2, wherein The specific steps of S3 are as follows: S31. Based on the interference heat map, extract the node set with a heat value higher than the threshold τ h as the candidate interference source area, and initialize multiple fruit fly individuals within the candidate interference source area, where each fruit fly individual corresponds to a search starting point of a potential interference path; S32. Use an improved fruit fly optimization algorithm, the improvement lies in introducing a multi-factor perception function to replace the search mechanism based on a single value judgment of the objective function: Among them, F k represents the k-th fruit fly individual, and J(F k ) represents the comprehensive perception function value calculated by the k-th fruit fly individual in the current jump decision. η represents the jump amplitude adjustment factor, and h k represents the interference heat of the node where the k-th fruit fly individual is currently located. ρ kj represents the consistency coefficient between the current path direction and the historical perturbation direction. I kj represents the interference intensity, SNR kj represents the signal-to-noise ratio, and θ kj represents the perturbation direction angle of edge e kj . S33. The Drosophila individual performs local neighborhood selection based on the multi-factor perception function and executes a skip search operation, skipping to the adjacent node that maximizes J(F k ), while updating the trajectory sequence and direction history information of the Drosophila individual; S34. Set an adaptive perturbation adjustment strategy. When the heat value oscillates or the perturbation direction reverses sharply during the continuous jumps of fruit fly individuals, trigger the direction perturbation update mechanism: Among them, represents the individual's current direction perception variable, α represents the disturbance amplitude adjustment factor, and rand(·) represents the random number generation function; S35. The jumping trajectory of each fruit fly individual is recorded as a path set in sequence, and the path sequences of all individuals constitute an interference path candidate set.

6. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 2, wherein The specific steps of S4 are as follows: S41. Evaluate each interference path in the interference path candidate set \(P = \{P_1, P_2, \ldots, P N \}\), where \(P i \) represents the \(i\)-th interference path, \(N\) represents the total number of interference paths, and calculate the interference heat coverage and the perturbation propagation rationality score for each interference path respectively; S42. The interference heat coverage is defined as the normalized mean of the interference heat values of each node on the path: Among them, Ω i represents the interference heat coverage of the i-th interference path, |P i | represents the length of the i-th interference path, h j represents the interference heat value of the node v j on the interference path, and max(h j ) represents the maximum interference heat value of all nodes in the i-th interference path; S43. Comprehensively evaluate the path direction continuity, consistency, and perturbation intensity change trend, and calculate the perturbation propagation rationality score: Among them, Ψ i represents the perturbation propagation rationality score of the i-th interference path, and θ mn represents the perturbation direction angle of edge e mn , I mn represents the interference intensity of edge e mn , and δf mn represents the frequency drift; S44. Normalize and combine the interference heat coverage and the perturbation propagation rationality score into a path comprehensive score function: Φ i = β1·Ω i + β2·Ψ i ; Among them, Φ i represents the comprehensive score of the i-th interference path, and β1 and β2 represent the weight coefficients; S45. Sort all interference paths P according to the comprehensive score i and select the interference path with the highest comprehensive score as the optimal interference path, and record the starting node of the optimal interference path as the current interference source location; S46. Output the optimal interference path and the interference source location.

7. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 2, characterized in that, The specific steps of S5 are as follows: S51. According to the optimal interference path and the interference source location, activate the hardware linkage control interface, and send control instructions to the hardware modules of the communication circuit through the hardware control bus to activate the filters, compensation circuits, and buffer amplifiers along the interference path; S52. Based on the interference signal spectrum characteristics on the optimal interference path, select the filtering parameters corresponding to the interference frequency band to suppress the interference signals in the interference frequency band from entering the communication receiving unit; S53. Real-time monitor the node voltage, current, and harmonic component data on the optimal interference path. According to the node impedance characteristics, automatically adjust the capacitance and inductance units of the compensation circuit to meet the impedance balance requirements during signal transmission, and suppress signal reflection and channel distortion; S54. Activate the feedback-type compensation circuit, measure the signal phase and amplitude characteristics in the communication link, and dynamically adjust the transistors, switching diodes, and phase shift circuits in the compensation circuit to generate a compensation signal with an amplitude similar to and a phase opposite to that of the interference signal in real time; S55. Real-time collect the signal quality data on the optimal interference path. When it is found that the communication signal level drops, the waveform is distorted, or the bit error rate increases, automatically trigger the buffer amplifier adjustment program. By adjusting the amplifier gain parameter, restore the interfered low-level communication signal, and perform waveform repair and signal standardization output; S56. If after the interference suppression control in steps S51 - S55, the communication link quality does not reach the preset index, according to the predefined alternative communication path data, automatically execute the communication path switching mechanism and switch to the sub-optimal path for signal transmission.

8. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 2, wherein The specific content of S6 includes: S61. Real-time collect the communication link signal quality data after hardware linkage control and establish a real-time communication quality database; S62. Use the latest data in the real-time communication quality database to calculate the actual interference response deviation value of the communication node, and compare it with the heat value of the corresponding node in the interference heat map generated in step S2 to obtain a node heat error set; S63. According to the node heat error set, dynamically adjust the weight matrix and direction adjustment coefficient of the adjacency perception module of the multi-layer rotating perception graph neural network through the backpropagation mechanism; S64. Analyze the characteristics of the change in the path perturbation direction in the real-time communication quality data, calculate the deviation between the actual direction angle change of the path propagation and the perturbation direction angle during the Drosophila jumping search process, and adjust the direction perception variable and the jump amplitude adjustment factor in the Drosophila individual jumping strategy based on the direction deviation, and dynamically update and improve the multi-factor perception function J(F k ) S65. Based on the updated multi-layer rotating perception graph neural network parameters and the improved fruit fly optimization algorithm strategy, re-execute step S2 and step S3 to generate a new interference heat map and an interference path candidate set, and realize the self-learning and strategy iteration ability.

9. The dual-mode carrier communication aggregation interference recognition system based on a graph neural network according to claim 8, wherein The communication link signal quality data includes node signal-to-noise ratio, bit error rate, interference intensity, voltage, and current changes.

10. A dual-mode carrier communication aggregation interference recognition device based on a graph neural network, characterized in that The device is used to execute each module in the dual-mode carrier communication aggregation interference recognition system based on the graph neural network as claimed in claim 1, and realize communication graph construction, graph neural network modeling, interference path search, path evaluation and source point positioning, linkage control, and self-learning update operations through hardware or software methods, so as to complete the recognition and suppression of interference paths in the dual-mode carrier communication environment.

Images