A communication navigation electromagnetic environment self-adaptive optimization method based on big data analysis

Abstract

The application discloses a communication navigation electromagnetic environment self-adaptive optimization method based on big data analysis, which comprises the following steps: S1, collecting electromagnetic environment data and operation data of a communication navigation system; S2, pre-processing the collected data; S3, generating node embedding representation by using a graph attention network, and adjusting an initial adjacency matrix and updating the node embedding representation in combination with a graph structure memory unit; S4, identifying interference behavior patterns by using a Gaussian mixture model with introduced spatial position constraints, and performing semantic mapping on performance degradation characteristics in a communication navigation system operation state; S5, generating optimal system parameter configurations from a preset system parameter configuration strategy set; and S6, deploying the optimal system parameter configurations to the communication navigation system. The application combines the graph attention network and the Gaussian mixture model, realizes electromagnetic interference identification and parameter optimization, and has the advantages of accurate pattern extraction, intelligent response and strong adaptability.

Description

Technical Field

[0001] This invention relates to the field of communication and navigation system technology, and in particular to an adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis. Background Technology

[0002] With the continuous integration and development of modern communication and navigation technologies, communication and navigation systems are playing an increasingly crucial role in complex application scenarios such as unmanned system control, integrated air-ground scheduling, tactical communication support, and intelligent urban mobility. However, these systems are usually deployed in highly dynamic and complex electromagnetic environments, frequently facing the combined effects of electromagnetic disturbances such as multi-source interference, spatial obstruction, and multipath propagation. This can easily lead to problems such as unstable communication links, decreased positioning accuracy, and task scheduling failures. In order to ensure the continuous and stable operation of communication and navigation systems in dynamic environments, it is urgent to develop intelligent optimization mechanisms with environmental perception and adaptive adjustment capabilities.

[0003] Currently, research on electromagnetic environment perception and optimization mainly focuses on two aspects: one is physical modeling-based methods, which infer the electromagnetic environment by constructing propagation models or interference source models. However, such methods rely heavily on prior knowledge, have poor adaptability, and are difficult to cope with dynamically changing scenarios. The other is static recognition methods based on traditional machine learning, such as support vector machines, K-means clustering, and random forests. Although these methods can achieve interference recognition and preliminary classification in specific scenarios, they usually ignore the temporal evolution characteristics and spatial structure coupling characteristics of electromagnetic disturbances. They cannot accurately depict the propagation path and behavior patterns of interference, and they also lack the ability to deeply correlate and model the operating status of communication and navigation systems, making it difficult to achieve refined system parameter tuning.

[0004] At the system optimization level, existing research has attempted to manually adjust system parameters (such as transmit power, frequency selection, modulation method, etc.) through rule bases or expert experience. However, this approach has a slow response speed, poor generalization ability, and cannot make optimal decisions in a timely manner when faced with sudden and complex interference events. Although some deep reinforcement learning-based methods have shown some potential in parameter tuning, they lack a structured understanding of the semantics of the electromagnetic environment and the ability to summarize behavioral patterns, resulting in their strategies lacking interpretability and transferability.

[0005] Furthermore, existing technologies generally neglect the spatial interference propagation relationships between different nodes in communication and navigation systems. They fail to fully utilize graph structure modeling techniques to model and dynamically adjust the interference propagation topology, and lack a unified framework to integrate multi-dimensional sensing data, topological evolution states, and performance degradation characteristics to form an interference-system impact semantic representation system that can be used for knowledge reasoning. This fragmented modeling approach directly limits the system's adaptive optimization capabilities under complex interference scenarios.

[0006] In summary, existing communication and navigation systems still face many bottlenecks in electromagnetic environment perception, interference behavior modeling, and adaptive optimization. There is a lack of a unified framework that can integrate spatial topology perception, temporal evolution modeling, and interference semantic understanding, making it impossible to achieve dynamic optimal adjustment of communication and navigation system performance while ensuring computational efficiency.

[0007] Therefore, there is an urgent need to propose an electromagnetic sensing and communication navigation optimization method oriented towards big data analysis capabilities, which integrates innovation in both model structure and decision-making mechanism to effectively improve the system's stability and anti-interference capability in complex electromagnetic environments. Summary of the Invention

[0008] One objective of this invention is to propose an adaptive optimization method for electromagnetic environments in communication and navigation based on big data analysis. This invention integrates graph attention networks and Gaussian mixture models to construct an electromagnetic interference propagation topology and interference behavior pattern recognition mechanism. Combined with semantic mapping of system performance degradation and parameter configuration strategy selection, it realizes closed-loop control of the communication and navigation system in complex electromagnetic environments, which has the advantages of high interference modeling accuracy, fast strategy response speed, and strong adaptive capability.

[0009] An adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to an embodiment of the present invention includes the following steps: S1. Collect electromagnetic environment data and operational data of communication and navigation systems; S2. Preprocess electromagnetic environment data and operational data to generate a unified fusion dataset; S3. Establish the initial adjacency matrix of the electromagnetic environment. Based on the fused dataset, use a graph attention network to extract the propagation topology of electromagnetic interference in the spatial structure, generate node embedding representations, and combine graph structure memory units to adjust the initial adjacency matrix and update the node embedding representations. S4. Based on the updated node embedding representation, a Gaussian mixture model with spatial position constraints is used to identify interference behavior patterns, and the interference behavior patterns are semantically mapped to the performance degradation features in the operation state of the communication and navigation system to generate a system performance degradation representation vector. S5. Based on the updated node embedding representation and system performance degradation representation vector, select the optimal solution from the preset system parameter configuration strategy set and output the optimal system parameter configuration; S6. Deploy the optimal system parameter configuration to the communication and navigation system.

[0010] Optionally, the electromagnetic environment data includes power spectral density, electric field strength, location of interference source, type of interference signal, and duration of interference, while the operational data includes signal-to-noise ratio, bit error rate, carrier tracking offset, positioning error, system parameter configuration, and link dynamic characteristic indicators.

[0011] Optionally, the location of the interference source is calculated using the time difference of arrival (TDOA) method, the system parameter configuration includes transmit power, operating frequency and modulation scheme, and the link dynamic characteristic indicators include the number of link interruptions, hold duration and establishment delay duration.

[0012] Optionally, the preprocessing includes data cleaning, time synchronization, spatial coordinate transformation, outlier removal, normalization, and standardization.

[0013] Optionally, S3 specifically includes: S31. Construct an input tensor based on the fusion dataset, aggregate the frequency dimension to obtain the spatial node feature matrix of each time step. The fusion dataset is four-dimensional data, and the dimensions include: time step length, number of spatial observation points, number of frequency channels, and feature dimension of each spatiotemporal frequency point. S32. An initial adjacency matrix is ​​established based on the spatial physical relationships between communication and navigation nodes. An electromagnetic interference propagation graph structure is constructed by combining the initial adjacency matrix. A graph attention network is used to perform spatial modeling of the spatial node feature matrix to obtain the node embedding representation. Each node in the electromagnetic interference propagation graph structure represents a communication and navigation terminal, and each edge represents a pair of nodes that have interference propagation or mutual influence. ; in, This represents the embedding of nodes. Represents the initial adjacency matrix. Represents the identity matrix. Let represent the initial adjacency matrix with self-loops. Let represent the degree matrix of the initial adjacency matrix with self-loops. Represents the feature matrix of spatial nodes. This represents the graph convolution weight matrix. Indicates the activation function; S33. Input the initial adjacency matrix into the graph structure memory unit and output the structure evolution state vector. S34. For each node, concatenate the node embedding representation with the structural evolution state vector to obtain the joint input vector; S35. Input the joint input vector into the gated loop unit, perform time modeling, and output the timing state vector of the communication and navigation node; S36. Based on the temporal state vector of the previous time step, calculate the node state similarity: ; in, Represents a node and Normalized similarity score of states, Represents the vector dot product. Indicates the number of communication and navigation observation nodes This represents the natural exponential function. Represents a node At time step The temporal state vector, Represents a node At time step The temporal state vector, Represents a node At time step The temporal state vector; S37. Calculate the node structure perturbation rate and generate the perturbation fusion weight factor: ; in, This represents the perturbation fusion weighting factor. This represents the activation function. This represents the node structure perturbation rate. Indicates the weighting coefficient; S38. Construct a dynamically adjusted adjacency matrix based on the perturbation fusion weight factor: ; in, This represents the adjacency matrix after feedback weighting adjustment. This represents the initial adjacency matrix of the previous time step; S39. Input the adjusted adjacency matrix into the graph attention network and recalculate the node embedding representation.

[0014] Optionally, the graph structure memory unit includes a structure encoding layer, a splicing layer, and a structure evolution modeling layer. The structure encoding layer uses spectral embedding and principal component analysis methods, and the structure evolution modeling layer employs gated cyclic units.

[0015] Optionally, the node structure perturbation rate measures the normalized change frequency of the connection relationship between communication and navigation node pairs within a given time window, and is statistically analyzed based on whether a connection state change occurs in the adjacency matrix state sequence: ; in, This represents the node structure perturbation rate. This represents a Boolean conditional function. Indicates the length of the time sliding window.

[0016] Optionally, S33 specifically includes: S331. Obtain the initial adjacency matrix sequence of the graph structure within consecutive time steps. ,in, Indicates the length of the time sliding window; S332. Perform structural encoding on each adjacency matrix, and calculate the Laplacian matrix based on spectral embedding: ; in, Indicates the first The Laplace matrix at each time step; S333. Perform eigenvalue decomposition on the Laplacian matrix and extract the eigenvalues. The eigenvectors corresponding to the smallest eigenvalues ​​form an eigenvector matrix; S334. Reduce the dimensionality of the row vectors of the feature vector matrix by global pooling to generate spectral embedding feature vectors; S335. Center each adjacency matrix and extract the previous... Given principal components, obtain the principal component projection matrix, and then calculate the principal component eigenvectors: ; in, Represents the principal component eigenvector. Represents the principal component projection matrix. The vector form representing the initial adjacency matrix; S336. Concatenate the spectral embedding feature vector with the principal component feature vector to form a structure encoding vector, and obtain the structure encoding sequence. S337. Input the structure encoding sequence into the gated cyclic unit to model the evolution trend and output the structure evolution state vector.

[0017] Optionally, S4 specifically includes: S41. Based on node embedding representation, extract the embedding representation vector of each node within the time window and obtain the corresponding spatial location vector to form an embedding representation set and a spatial location set. S42. Apply a Gaussian mixture model with spatial location constraints to the set of embedded representations to perform disturbance behavior clustering. The parameters of each disturbance pattern are defined as triples. ,in Indicates the first Embedding center of interference-like patterns, Indicates the first The covariance matrix of the interference-like mode, Indicates the first Spatial location center vector of the interference-like pattern; Based on the Gaussian mixture model, calculate the posterior probability that a node belongs to the interference mode: ; in, Represents a node Belongs to the Posterior probability of interference-like patterns Indicates the first Mixed weighting coefficients Embedded nodes represent nodes in a set. The embedding representation vector, Indicates the mean Covariance The Gaussian distribution under which the embedding represents the probability density of the vector is used. Represents nodes in a set of spatial locations Spatial position vector, Represents the number of clusters, Indicates the first Mixed weighting coefficients Indicates the first Spatial location center vector of the interference-like mode, Indicates the first Embedding center of interference-like patterns, Indicates the first The covariance matrix of the interference-like mode, Indicates the mean Covariance The Gaussian distribution under which the embedding represents the probability density of the vector is used. Represents a node Spatial position vector and position center vector Spatial consistency function; ; in, Indicates the spatial distance adjustment factor. Represents a node The Euclidean distance between the spatial location vector and the spatial center location vector of the pattern; Based on the posterior probability, the embedding center, covariance matrix, and spatial location center vector are iteratively optimized using the expectation-maximization algorithm: S43. Based on the posterior probability of the interference pattern, determine the interference pattern category label of the node according to the maximum a posteriori probability rule, and define the first... The set of nodes corresponding to the interference-like pattern: ; in, Represents a node Interference mode category labels, This represents the cluster index that maximizes the function value; S44. Construct an interference-system impact semantic mapping using known system performance degradation labels. Assume a set of labeled nodes, where each node in the labeled node set corresponds to a system performance degradation label. Let the nth node be an example node. The labeled subset of the interference pattern is the first The intersection of the node set corresponding to the interference pattern and the labeled node set is used to construct the system performance degradation representation vector corresponding to the interference pattern: ; in, Indicates the first The system performance degradation representation vector corresponding to the interference-like patterns. This indicates the number of supervised nodes in the labeled subset. Represents a node Performance degradation label.

[0018] Optionally, S5 specifically includes: S51. Construct a joint state vector for each node based on the node embedding representation set and the system performance degradation representation vector. S52. Define a set of system parameter configuration strategies, wherein each strategy in the set of system parameter configuration strategies includes transmit power, operating frequency and modulation method; S53. Evaluate the fitness score of each policy in the current state: ; in, In the joint state vector Select Parameter Configuration Strategy The rating value This represents a multilayer perceptron. The weight parameters of the scoring function are represented. Represents a node The joint state vector, This represents the parameter configuration strategy in the system parameter configuration strategy set. Indicates the transpose operation; S54. For each node, select the current optimal system parameter configuration based on the fitness score: ; in, Represents a node Select the optimal system parameter configuration. This represents the set of system parameter configuration strategies.

[0019] The beneficial effects of this invention are: First, this invention employs a graph attention network to model the spatial propagation topology of electromagnetic interference in communication and navigation systems. By introducing an attention mechanism, the model can dynamically allocate connection weights between different nodes, reflecting the intensity and range of interference propagation. Compared to traditional graph convolution methods, this invention exhibits stronger local perception and feature selection capabilities when dealing with non-uniform interference distributions. This mechanism enhances the spatial discriminability of node embedding representations, providing a more distinguishable representational basis for subsequent interference pattern clustering and state analysis, and effectively enhancing the system's structural adaptability to complex electromagnetic environments.

[0020] Secondly, this invention introduces a graph structure memory unit to model the structural evolution of the initial adjacency matrix between communication and navigation nodes. It estimates the connection change trend by combining the node structure perturbation rate and adjusts the adjacency relationship through perturbation fusion weight factor to construct a dynamically updated graph topology. This design can realize the time-series modeling and feedback adjustment of electromagnetic environment structural perturbations, enabling the graph structure to capture rapid connection changes and local perturbation paths in the electromagnetic environment within the time window. Compared with static graph structures, this invention improves the timeliness and accuracy of interference propagation modeling and enhances the system's ability to perceive and its robustness to dynamic interference states.

[0021] Furthermore, this invention employs a Gaussian mixture model with spatial location constraints during the interference behavior modeling process. It performs joint clustering analysis on node embedding representation and spatial location. By embedding the spatial consistency function into the posterior probability calculation, it enhances the clustering process's ability to perceive physical spatial information, thereby achieving accurate classification of spatially concentrated interference sources. This mechanism avoids the misclassification problem of traditional unconstrained clustering in high-dimensional space, making the identification results of interference patterns more interpretable and continuous in space, and improving clustering stability and the completeness of the expression of interference behavior patterns.

[0022] Furthermore, based on the system performance degradation labels of labeled nodes, this invention constructs a semantic correspondence between interference patterns and the performance degradation state of communication and navigation systems. By aggregating statistical features of supervised subsets of interference patterns, a system performance degradation representation vector for optimization inference is formed, realizing semantic modeling from interference behavior at the perception layer to operational risks at the system layer. This structured mapping mechanism opens up the semantic channel between interference pattern recognition and system state regulation, improves the pertinence and interpretability of the system tuning mechanism, and provides clear semantic path support for parameter optimization.

[0023] Finally, this invention constructs a node-level joint state vector, integrates the electromagnetic environment embedding representation with system performance degradation characteristics as input, and selects the optimal solution from a preset system parameter configuration strategy set using a multi-strategy scoring mechanism. This method uses a scoring function to calculate the adaptability score of each strategy in the current state, and selects the current optimal combination of transmit power, frequency, and modulation scheme accordingly, achieving precise parameter adjustment based on interference behavior and operating state. Compared with configuration methods that rely on rules or single index ranking, this invention achieves closed-loop optimization of state awareness and strategy evaluation during parameter configuration, improving the system's adaptive adjustment capability and anti-interference stability. Attached Figure Description

[0024] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis proposed in this invention. Figure 2 This is a schematic diagram of the structure of the electromagnetic propagation topology jointly modeled by graph attention network and graph structure memory unit in the communication and navigation electromagnetic environment adaptive optimization method based on big data analysis proposed in this invention. Figure 3 This is a schematic diagram of the graph structure spectrum embedding and principal component analysis processing flow of the adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis proposed in this invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0026] refer to Figure 1-3 An adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis includes the following steps: S1. Collect electromagnetic environment data and operational data of communication and navigation systems; S2. Preprocess electromagnetic environment data and operational data to generate a unified fusion dataset; S3. Establish the initial adjacency matrix of the electromagnetic environment. Based on the fused dataset, use a graph attention network to extract the propagation topology of electromagnetic interference in the spatial structure, generate node embedding representations, and combine graph structure memory units to adjust the initial adjacency matrix and update the node embedding representations. S4. Based on the updated node embedding representation, a Gaussian mixture model with spatial position constraints is used to identify interference behavior patterns, and the interference behavior patterns are semantically mapped to the performance degradation features in the operation state of the communication and navigation system to generate a system performance degradation representation vector. S5. Based on the updated node embedding representation and system performance degradation representation vector, select the optimal solution from the preset system parameter configuration strategy set and output the optimal system parameter configuration; S6. Deploy the optimal system parameter configuration to the communication and navigation system.

[0027] This invention proposes an adaptive optimization method for electromagnetic environments in communication and navigation systems. It constructs a complete process from data acquisition, graph modeling, interference clustering, semantic mapping to optimized configuration. By integrating big data analysis technology and structural modeling mechanism, it realizes situational awareness and optimal parameter adjustment of communication and navigation systems in complex electromagnetic environments. It has advantages such as handling diverse data types, strong dynamic changes in the environment, and real-time decision feedback. It significantly improves the system's interference identification accuracy, stable operation capability, and adaptive control efficiency, and is suitable for navigation and communication support tasks in various complex scenarios with strong interference.

[0028] In this embodiment, the electromagnetic environment data includes power spectral density, electric field strength, interference source location, interference signal type, and interference duration, while the operational data includes signal-to-noise ratio, bit error rate, carrier tracking offset, positioning error, system parameter configuration, and link dynamic characteristic indicators.

[0029] This invention clarifies the composition of electromagnetic environment data and communication and navigation operation data, covering key dimensions such as spectrum, location, signal type, and link indicators. It provides high-precision, multi-dimensional data support for subsequent model input. By jointly collecting interference source information and navigation error indicators, the system is able to identify the root causes of interference and the trend of dynamic perception performance degradation. This enhances the modeling foundation for the coupling relationship between the diversity of input variables and system behavior, effectively improves the accuracy and generalization ability of the overall situational awareness model, and provides a solid data foundation for system optimization.

[0030] In this embodiment, the location of the interference source is calculated using the time difference of arrival (TDOA) method. The system parameter configuration includes transmit power, operating frequency, and modulation scheme. The link dynamic characteristic indicators include the number of link interruptions, hold duration, and establishment delay duration.

[0031] This invention employs the time-of-arrival (TOA) method to accurately calculate the location of interference sources and refines the definition of system parameter configuration and link dynamic characteristics, effectively enhancing the model's ability to model the relationship between high-dimensional physical signals and operational states. Spatial localization of interference sources, combined with link dynamic indicators to reflect communication performance fluctuations, improves the system's ability to identify local anomaly propagation paths and vulnerable nodes in the link. The fine-grained parameter definitions make the model more interpretable and operable during policy optimization, providing rich state information support for subsequent adaptive parameter tuning mechanisms.

[0032] In this embodiment, the preprocessing includes data cleaning, time synchronization, spatial coordinate transformation, outlier removal, normalization, and standardization.

[0033] This invention employs a multi-step preprocessing process, including data cleaning, time synchronization, spatial coordinate transformation, outlier removal, normalization, and standardization, to unify the structural representation of data from different sources, ensuring the quality and consistency of input data. This multi-dimensional preprocessing workflow effectively eliminates spatiotemporal mismatches, scale inhomogeneities, and outlier interference, improving the stability and accuracy of the model's fusion analysis of electromagnetic data and system state data. The resulting highly consistent fusion dataset provides a standardized input foundation for subsequent graph structure modeling and behavior pattern recognition, enhancing the overall robustness of the model.

[0034] In this embodiment, S3 specifically includes: S31. Construct an input tensor based on the fusion dataset, aggregate the frequency dimension to obtain the spatial node feature matrix of each time step. The fusion dataset is four-dimensional data, and the dimensions include: time step length, number of spatial observation points, number of frequency channels, and feature dimension of each spatiotemporal frequency point. S32. An initial adjacency matrix is ​​established based on the spatial physical relationships between communication and navigation nodes. An electromagnetic interference propagation graph structure is constructed by combining the initial adjacency matrix. A graph attention network is used to perform spatial modeling of the spatial node feature matrix to obtain the node embedding representation. Each node in the electromagnetic interference propagation graph structure represents a communication and navigation terminal, and each edge represents a pair of nodes that have interference propagation or mutual influence. ; in, This represents the embedding of nodes. Represents the initial adjacency matrix. Represents the identity matrix. Let represent the initial adjacency matrix with self-loops. Let represent the degree matrix of the initial adjacency matrix with self-loops. Represents the feature matrix of spatial nodes. This represents the graph convolution weight matrix. Indicates the activation function; S33. Input the initial adjacency matrix into the graph structure memory unit and output the structure evolution state vector. S34. For each node, concatenate the node embedding representation with the structural evolution state vector to obtain the joint input vector; S35. Input the joint input vector into the gated loop unit, perform time modeling, and output the timing state vector of the communication and navigation node; S36. Based on the temporal state vector of the previous time step, calculate the node state similarity: ; in, Represents a node and Normalized similarity score of states, Represents the vector dot product. Indicates the number of communication and navigation observation nodes This represents the natural exponential function. Represents a node At time step The temporal state vector, Represents a node At time step The temporal state vector, Represents a node At time step The temporal state vector; S37. Calculate the node structure perturbation rate and generate the perturbation fusion weight factor: ; in, This represents the perturbation fusion weighting factor. This represents the activation function. This represents the node structure perturbation rate. Indicates the weighting coefficient; S38. Construct a dynamically adjusted adjacency matrix based on the perturbation fusion weight factor: ; in, This represents the adjacency matrix after feedback weighting adjustment. This represents the initial adjacency matrix of the previous time step; S39. Input the adjusted adjacency matrix into the graph attention network and recalculate the node embedding representation.

[0035] This invention constructs an electromagnetic interference propagation graph structure based on a graph attention network, extracts the non-uniform propagation relationship between nodes, and introduces a graph structure memory unit and a disturbance feedback mechanism to dynamically adjust the adjacency matrix, enhancing the model's ability to capture the evolution of the interference structure. By constructing a graph convolution embedding and temporal state joint modeling mechanism, it effectively characterizes the evolution law of interference in the spatiotemporal dimension, providing support for refined node state modeling. The dynamic evolution process of this graph structure improves the model's adaptability and expression accuracy when facing sudden interference, and has stronger real-time performance and anti-interference ability.

[0036] In this embodiment, the graph structure memory unit includes a structure encoding layer, a splicing layer, and a structure evolution modeling layer. The structure encoding layer uses spectral embedding and principal component analysis methods, and the structure evolution modeling layer adopts a gated cyclic unit.

[0037] In this embodiment, the node structure perturbation rate measures the normalized change frequency of the connection relationship between communication and navigation nodes within a given time window, and is statistically analyzed based on whether a connection state change occurs in the adjacency matrix state sequence: ; in, This represents the node structure perturbation rate. This represents a Boolean conditional function. Indicates the length of the time sliding window.

[0038] This invention measures the frequency of connection changes in a graph structure over time by defining a node structure perturbation rate. Combined with statistical analysis of adjacency matrix state sequence changes, it achieves a quantitative characterization of local structural perturbations in the system. This perturbation rate index is dynamic and interpretable, effectively identifying the impact of abrupt changes in inter-node connection relationships on system behavior patterns. This quantification mechanism helps introduce dynamic feedback factors into structural evolution modeling, improving the accuracy and adaptive adjustment efficiency of adjacency matrix adjustments, and enhancing the system's sensitivity and control over structurally unstable regions.

[0039] In this embodiment, S33 specifically includes: S331. Obtain the initial adjacency matrix sequence of the graph structure within consecutive time steps. ,in, Indicates the length of the time sliding window; S332. Perform structural encoding on each adjacency matrix, and calculate the Laplacian matrix based on spectral embedding: ; in, Indicates the first The Laplace matrix at each time step; S333. Perform eigenvalue decomposition on the Laplacian matrix and extract the eigenvalues. The eigenvectors corresponding to the smallest eigenvalues ​​form an eigenvector matrix; S334. Reduce the dimensionality of the row vectors of the feature vector matrix by global pooling to generate spectral embedding feature vectors; S335. Center each adjacency matrix and extract the previous... Given principal components, obtain the principal component projection matrix, and then calculate the principal component eigenvectors: ; in, Represents the principal component eigenvector. Represents the principal component projection matrix. The vector form representing the initial adjacency matrix; S336. Concatenate the spectral embedding feature vector with the principal component feature vector to form a structure encoding vector, and obtain the structure encoding sequence. S337. Input the structure encoding sequence into the gated cyclic unit to model the evolution trend and output the structure evolution state vector.

[0040] The graph structure memory modeling method proposed in this invention uses spectral embedding and principal component analysis to jointly encode the adjacency matrix sequence, and combines gated cyclic units to model the temporal evolution trend of the adjacency structure to generate a structural evolution state vector. This multi-source structural representation mechanism can capture the evolution law and global change characteristics of the graph structure in the time dimension, effectively enhancing the model's ability to model the dynamic process of the structure. By fusing and modeling multiple structural features, it helps to improve the accuracy of the adjacency feedback adjustment mechanism and realize the adaptive evolution of the electromagnetic propagation graph in dynamic scenarios.

[0041] In this embodiment, S4 specifically includes: S41. Based on node embedding representation, extract the embedding representation vector of each node within the time window and obtain the corresponding spatial location vector to form an embedding representation set and a spatial location set. S42. Apply a Gaussian mixture model with spatial location constraints to the set of embedded representations to perform disturbance behavior clustering. The parameters of each disturbance pattern are defined as triples. ,in Indicates the first Embedding center of interference-like patterns, Indicates the first The covariance matrix of the interference-like mode, Indicates the first Spatial location center vector of the interference-like pattern; Based on the Gaussian mixture model, calculate the posterior probability that a node belongs to the interference mode: ; in, Represents a node Belongs to the Posterior probability of interference-like patterns Indicates the first Mixed weighting coefficients Embedded nodes represent nodes in a set. The embedding representation vector, Indicates the mean Covariance The Gaussian distribution under which the embedding represents the probability density of the vector is used. Represents nodes in a set of spatial locations Spatial position vector, Represents the number of clusters, Indicates the first Mixed weighting coefficients Indicates the first Spatial location center vector of the interference-like mode, Indicates the first Embedding center of interference-like patterns, Indicates the first The covariance matrix of the interference-like mode, Indicates the mean Covariance The Gaussian distribution under which the embedding represents the probability density of the vector is used. Represents a node Spatial position vector and position center vector Spatial consistency function; ; in, Indicates the spatial distance adjustment factor. Represents a node The Euclidean distance between the spatial location vector and the spatial center location vector of the pattern; Based on the posterior probability, the embedding center, covariance matrix, and spatial location center vector are iteratively optimized using the expectation-maximization algorithm: S43. Based on the posterior probability of the interference pattern, determine the interference pattern category label of the node according to the maximum a posteriori probability rule, and define the first... The set of nodes corresponding to the interference-like pattern: ; in, Represents a node Interference mode category labels, This represents the cluster index that maximizes the function value; S44. Construct an interference-system impact semantic mapping using known system performance degradation labels. Assume a set of labeled nodes, where each node in the labeled node set corresponds to a system performance degradation label. Let the nth node be an example node. The labeled subset of the interference pattern is the first The intersection of the node set corresponding to the interference pattern and the labeled node set is used to construct the system performance degradation representation vector corresponding to the interference pattern: ; in, Indicates the first The system performance degradation representation vector corresponding to the interference-like patterns. This indicates the number of supervised nodes in the labeled subset. Represents a node Performance degradation label.

[0042] This invention employs a Gaussian mixture model with spatial constraints to jointly cluster node embedding representations and spatial locations, identifying interference behavior patterns that combine spatial consistency and feature similarity. It also integrates labeled node tags to construct a system performance degradation vector. This method achieves end-to-end modeling of interference identification from embedding representation to semantic mapping, possessing good clustering interpretability and predictive ability. By constructing a semantic graph of interference-system impact through the correspondence between patterns and system performance, it provides structured knowledge support for subsequent parameter optimization, improving the intelligence and targeting of system response.

[0043] In this embodiment, S5 specifically includes: S51. Construct a joint state vector for each node based on the node embedding representation set and the system performance degradation representation vector. S52. Define a set of system parameter configuration strategies, wherein each strategy in the set of system parameter configuration strategies includes transmit power, operating frequency and modulation method; S53. Evaluate the fitness score of each policy in the current state: ; in, In the joint state vector Select Parameter Configuration Strategy The rating value This represents a multilayer perceptron. The weight parameters of the scoring function are represented. Represents a node The joint state vector, This represents the parameter configuration strategy in the system parameter configuration strategy set. Indicates the transpose operation; S54. For each node, select the current optimal system parameter configuration based on the fitness score: ; in, Represents a node Select the optimal system parameter configuration. This represents the set of system parameter configuration strategies.

[0044] This invention constructs a joint state vector to fuse node embedding representation and performance degradation representation into a model. It then evaluates the adaptability score of each strategy using a scoring function and selects the current optimal system parameter configuration. This mechanism achieves a deep correlation between strategy selection and node state, effectively enhancing the pertinence and dynamism of the strategy selection process. Compared with static rule matching, this method has advantages such as state-driven, autonomous adjustment, and local optimum approximation, improving the parameter response efficiency and stability of the system under electromagnetic disturbances and providing practical support for intelligent control strategies.

[0045] Example 1: To verify the feasibility of this invention in practice, it was applied to a company's "remote communication and navigation system" deployed in the southeastern coastal region. The company needs to ensure communication and positioning synchronization between the unmanned maritime inspection platform and the shore-based control center in a complex electromagnetic environment. During the morning peak hours when interference is dense, the system's operational stability fluctuates for a long time, especially manifested in problems such as navigation error deviation, increased communication delay, and high link error rate. The previous static frequency redistribution and fixed gain enhancement strategies are difficult to adapt to the spatiotemporal dynamic changes of electromagnetic interference, resulting in high operation and maintenance costs, long adjustment delays, and low support efficiency.

[0046] In a real-world environment, this invention first comprehensively collects historical operational data from communication and navigation terminals, along with electromagnetic environment monitoring data, covering multiple operational indicators such as power spectral density, electric field strength, interference type, channel error rate, and carrier tracking, and updates this data dynamically in real-time at 5-minute intervals. Subsequently, this multi-dimensional data is input into the graph attention network architecture proposed in this invention to establish spatial topological relationships between nodes, and the interference propagation and evolution trends are identified through a graph structure memory mechanism. During the interference identification phase, the system utilizes a Gaussian mixture model with spatial location constraints to form cluster centers for various interference behaviors, and through mapping learning, forms a system performance degradation representation corresponding to the interference patterns.

[0047] During the parameter optimization phase, the system jointly inputs the embedded representation and semantic vectors into the policy recommendation network. A multi-layered scoring mechanism selects the optimal combination from a policy pool that includes transmit power, modulation scheme, and operating frequency. The final output is written back to the communication and navigation system control module in real time, completing the closed-loop control deployment.

[0048] Table 1 Comparison of Total Experimental Data

[0049] Regarding the signal-to-noise ratio (SNR), the average SNR of the system before optimization was 7.6 dB, which was improved to 11.9 dB after optimization, representing an improvement of approximately 56.6%. This improvement indicates that the present invention can accurately identify interference sources and effectively reconstruct communication paths between nodes during the interference perception and topology modeling stages, thereby significantly improving the link signal quality.

[0050] Regarding the bit error rate, the average bit error rate of the system decreased from 0.0087 to 0.0037, a reduction of 57.5%. This result indicates that the present invention, by identifying interference behavior patterns through Gaussian mixture clustering and optimizing parameters based on their impact semantics, significantly reduced the bit error rate in the communication link, effectively ensuring the reliability of data transmission. The average carrier tracking offset decreased from 14.3 Hz to 7.1 Hz, a reduction of 50.3%, demonstrating that the system can still stably complete carrier locking under complex frequency interference, and its anti-interference frequency locking capability is significantly enhanced.

[0051] The overall latency of the communication system was reduced from 156.8 ms to 110.6 ms, an average reduction of about 29.5%. This optimization effect demonstrates the role of the parameter configuration strategy recommendation mechanism in improving system operating efficiency. In particular, it can effectively alleviate scheduling bottlenecks and improve response speed under high load and high interference conditions. More importantly, the system interference response has changed from relying on manual judgment to fully automatic identification and strategy deployment within 6 seconds, and parameter adjustment has been reduced from an average of 45 seconds to within 4 seconds. The response time has been shortened by more than 91%, which greatly improves the automation and intelligence level of the system.

[0052] In summary, this invention achieves comprehensive optimization of multiple performance indicators of communication and navigation systems without relying on external intervention. It has advantages such as strong signal quality, high system stability, and fast control response, providing a practical adaptive optimization solution for communication and navigation systems in highly dynamic and strongly interfering electromagnetic environments.

[0053] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for adaptive optimization of electromagnetic environment for communication and navigation based on big data analysis, characterized in that, Includes the following steps: S1. Collect electromagnetic environment data and operational data of communication and navigation systems; S2. Preprocess electromagnetic environment data and operational data to generate a unified fusion dataset; S3. Establish the initial adjacency matrix of the electromagnetic environment. Based on the fused dataset, use a graph attention network to extract the propagation topology of electromagnetic interference in the spatial structure, generate node embedding representations, and combine graph structure memory units to adjust the initial adjacency matrix and update the node embedding representations. S4. Based on the updated node embedding representation, a Gaussian mixture model with spatial position constraints is used to identify interference behavior patterns, and the interference behavior patterns are semantically mapped to the performance degradation features in the operation state of the communication and navigation system to generate a system performance degradation representation vector. S5. Based on the updated node embedding representation and system performance degradation representation vector, select the optimal solution from the preset system parameter configuration strategy set and output the optimal system parameter configuration; S6. Deploy the optimal system parameter configuration to the communication and navigation system.

2. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 1, characterized in that, The electromagnetic environment data includes power spectral density, electric field strength, location of interference source, type of interference signal, and duration of interference. The operational data includes signal-to-noise ratio, bit error rate, carrier tracking offset, positioning error, system parameter configuration, and link dynamic characteristic indicators.

3. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 2, characterized in that, The location of the interference source is calculated using the time difference of arrival (TDOA) method. The system parameter configuration includes transmit power, operating frequency, and modulation scheme. The link dynamic characteristic indicators include the number of link interruptions, hold duration, and establishment delay duration.

4. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 1, characterized in that, The preprocessing includes data cleaning, time synchronization, spatial coordinate transformation, outlier removal, normalization, and standardization.

5. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 1, characterized in that, S3 specifically includes: S31. Construct an input tensor based on the fused dataset, aggregate the frequency dimension, and obtain the spatial node feature matrix for each time step; S32. Establish an initial adjacency matrix based on the spatial physical relationships between communication and navigation nodes. Combine the initial adjacency matrix to construct an electromagnetic interference propagation graph structure. Use a graph attention network to perform spatial modeling of the spatial node feature matrix to obtain the node embedding representation: ； in, This represents the embedding of nodes. Represents the initial adjacency matrix. Represents the identity matrix. Let represent the initial adjacency matrix with self-loops. Let represent the degree matrix of the initial adjacency matrix with self-loops. Represents the feature matrix of spatial nodes. This represents the graph convolution weight matrix. Indicates the activation function; S33. Input the initial adjacency matrix into the graph structure memory unit and output the structure evolution state vector. S34. For each node, concatenate the node embedding representation with the structural evolution state vector to obtain the joint input vector; S35. Input the joint input vector into the gated loop unit, perform time modeling, and output the timing state vector of the communication and navigation node; S36. Calculate the node state similarity based on the temporal state vector of the previous time step; S37. Calculate the node structure perturbation rate and generate the perturbation fusion weight factor: ； in, This represents the perturbation fusion weighting factor. This represents the activation function. This represents the node structure perturbation rate. Indicates the weighting coefficient; S38. Construct a dynamically adjusted adjacency matrix based on the perturbation fusion weight factor: ； in, This represents the adjacency matrix after feedback weighting adjustment. This represents the initial adjacency matrix of the previous time step; S39. Input the adjusted adjacency matrix into the graph attention network and recalculate the node embedding representation.

6. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 5, characterized in that, The graph structure memory unit includes a structure encoding layer, a splicing layer, and a structure evolution modeling layer. The structure encoding layer uses spectral embedding and principal component analysis methods, and the structure evolution modeling layer adopts gated cyclic units.

7. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 5, characterized in that, The node structure perturbation rate measures the normalized change frequency of the connection relationship between communication and navigation nodes within a given time window, and is statistically analyzed based on whether a connection state change occurs in the adjacency matrix state sequence. ； in, This represents the node structure perturbation rate. This represents a Boolean conditional function. Indicates the length of the time sliding window.

8. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 5, characterized in that, Specifically, S33 includes: S331. Obtain the initial adjacency matrix sequence of the graph structure within a continuous time step; S332. Perform structural encoding on each adjacency matrix and calculate the Laplacian matrix based on spectral embedding. S333. Perform eigenvalue decomposition on the Laplacian matrix and extract the eigenvalues. The eigenvectors corresponding to the smallest eigenvalues ​​form an eigenvector matrix; S334. Reduce the dimensionality of the row vectors of the feature vector matrix by global pooling to generate spectral embedding feature vectors; S335. Center each adjacency matrix and extract the previous... One principal component is obtained, the principal component projection matrix is ​​obtained, and the principal component eigenvectors are calculated based on it; S336. Concatenate the spectral embedding feature vector with the principal component feature vector to form a structure encoding vector, and obtain the structure encoding sequence. S337. Input the structure encoding sequence into the gated cyclic unit to model the evolution trend and output the structure evolution state vector.

9. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 1, characterized in that, S4 specifically includes: S41. Based on node embedding representation, extract the embedding representation vector of each node within the time window and obtain the corresponding spatial location vector to form an embedding representation set and a spatial location set. S42. Apply a Gaussian mixture model with spatial location constraints to the set of embedded representations to perform disturbance behavior clustering. The parameters of each disturbance pattern are defined as triples. ,in Indicates the first Embedding center of interference-like patterns Indicates the first The covariance matrix of the interference-like mode, Indicates the first Spatial location center vector of the interference-like pattern; Based on the Gaussian mixture model, calculate the posterior probability that a node belongs to the interference mode: ； in, Represents a node Belongs to the Posterior probability of interference-like patterns Indicates the first Mixed weighting coefficients Embedded nodes represent nodes in a set. The embedding representation vector, Indicates the mean Covariance The Gaussian distribution under which the embedding represents the probability density of the vector is used. Represents nodes in a set of spatial locations Spatial position vector, Represents the number of clusters, Indicates the first Mixed weighting coefficients Indicates the first Spatial location center vector of the interference-like mode, Indicates the first Embedding center of interference-like patterns Indicates the first The covariance matrix of the interference-like mode, Indicates the mean Covariance The Gaussian distribution under which the embedding represents the probability density of the vector is used. Represents a node Spatial position vector and position center vector Spatial consistency function; Based on the posterior probability, the embedding center, covariance matrix, and spatial location center vector are iteratively optimized using the expectation-maximization algorithm: S43. Based on the posterior probability of the interference pattern, determine the interference pattern category label of the node according to the maximum a posteriori probability rule, and define the first... The set of nodes corresponding to the interference pattern; S44. Construct an interference-system impact semantic mapping using known system performance degradation labels. Assume a set of labeled nodes, where each node in the labeled node set corresponds to a system performance degradation label. Let the nth node be an example node. The labeled subset of the interference pattern is the first The intersection of the node set corresponding to the interference pattern and the labeled node set is used to construct the system performance degradation representation vector corresponding to the interference pattern.

10. The adaptive optimization method for communication and navigation electromagnetic environment based on big data analysis according to claim 1, characterized in that, S5 specifically includes: S51. Construct a joint state vector for each node based on the node embedding representation set and the system performance degradation representation vector. S52. Define a set of system parameter configuration strategies, wherein each strategy in the set of system parameter configuration strategies includes transmit power, operating frequency and modulation method; S53. Evaluate the fitness score of each policy in the current state; S54. For each node, select the current optimal system parameter configuration based on the adaptability score.

Images