Network architecture and networking method of optical fiber early warning system

Abstract

The application provides a network architecture and networking method of an optical fiber early warning system, relates to the field of optical fiber communication, and comprises the following steps: acquiring state information of sensing nodes in a distributed optical fiber sensing network, constructing a network state tensor and decomposing the network state tensor to acquire a coupling feature vector, constructing a hierarchical network topology based on a cooperative measurement value, calculating a service feature vector of early warning service data, adjusting a gradient according to a demand intensity and a path to perform local reconstruction on the network, and dynamically adjusting a service bearing weight, so that transmission path optimization and efficient resource utilization are achieved, and the system early warning precision and response speed are improved.

Description

Technical Field

[0001] This invention relates to optical fiber communication technology, and more particularly to a network architecture and networking method for an optical fiber early warning system. Background Technology

[0002] In existing technologies, fiber optic early warning systems typically employ a fixed topology, maintaining its structure largely unchanged after system deployment. The transmission paths for early warning data are generally pre-defined using static routing algorithms or dynamic routing algorithms based on the shortest path. Status monitoring of sensing nodes primarily focuses on their own operational status, rarely considering the mutual influence and collaborative relationships between nodes. During the processing of early warning operational data, the focus is mainly on the characteristics of the data itself, neglecting the intrinsic connection between data transmission and the network topology.

[0003] Traditional fixed topologies lack in-depth analysis of the coupling relationships between network nodes, failing to effectively identify and utilize the collaborative characteristics between sensing nodes. This leads to uneven distribution of network resources, with some nodes overloaded while others are idle, impacting the overall early warning efficiency of the system. Furthermore, the lack of a dynamic adaptation mechanism between early warning service data and the network topology means that when the early warning service load changes, the system cannot intelligently adjust the network structure according to service characteristics, resulting in increased data transmission latency or packet loss rate, reducing the real-time performance and reliability of the early warning system. Existing systems have overly simplistic response mechanisms to network state changes, lacking predictive capabilities and proactive adjustment strategies. When network congestion or partial node failures occur, timely network reconstruction and service reallocation are impossible, resulting in poor robustness in complex environments. Summary of the Invention

[0004] This invention provides a network architecture and networking method for a fiber optic early warning system, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a network architecture and networking method for a fiber optic early warning system, comprising:

[0006] Obtain the state information of sensing nodes in a distributed optical fiber sensor network, construct the network state tensor and decompose it to obtain the coupled feature vector;

[0007] The collaborative metric value of the sensing nodes is calculated based on the coupling feature vector. The sensing nodes are then projected according to the collaborative metric value to obtain the projected coordinates. Based on the projected coordinates, the sensing nodes are clustered to generate a hierarchical network topology.

[0008] Acquire early warning service data from sensing nodes, calculate the service spatial location based on the projection coordinates of the source node and target node of the early warning service data, calculate the correlation weight between the service spatial location and the service attribute features of the early warning service data, and generate a service feature vector.

[0009] The demand intensity of early warning service data on the transmission path in the hierarchical network topology is calculated based on the service feature vector. The path adjustment gradient is calculated based on the demand intensity and the carrying status of the transmission path. The hierarchical network topology is then partially reconstructed based on the path adjustment gradient.

[0010] The path adjustment gradient is backpropagated to the sensing node. The network state impact is predicted based on the coupled feature vector. The service carrying weight of the sensing node is dynamically adjusted. The early warning service data is allocated to the locally reconstructed transmission path according to the adjusted service carrying weight.

[0011] To obtain the state information of sensing nodes in a distributed fiber optic sensor network, the network state tensor is constructed and decomposed to obtain the coupled feature vector, including:

[0012] Collect optical signal parameters and service load parameters of each sensing node to generate a node state vector;

[0013] The fiber optic connection relationship and spatial location information of the sensing nodes are obtained. Adjacent sensing node pairs are determined based on the fiber optic connection relationship. The link hop count between each pair of adjacent sensing nodes is calculated. The physical distance between each pair of adjacent sensing nodes is calculated based on the spatial location information of the nodes.

[0014] The link hop count and physical distance are weighted and summed to obtain the inter-node coupling degree, and a node coupling matrix with the inter-node coupling degree as the element is constructed.

[0015] The node state vector is expanded into a state matrix according to the time sequence. The state matrix is ​​then multiplied with the node coupling matrix to generate a network state tensor containing node dimension, time dimension and coupling dimension.

[0016] The network state tensor is decomposed along the node dimension, time dimension, and coupling dimension, and node features, temporal features, and coupling features are extracted and fused to obtain the coupling feature vector.

[0017] The collaborative metric value of the sensing nodes is calculated based on the coupling feature vector. The sensing nodes are then projected according to the collaborative metric value to obtain projected coordinates. Based on the projected coordinates, the sensing nodes are clustered to generate a hierarchical network topology, including:

[0018] Construct a collaborative propagation graph of coupled feature vectors, treating each sensing node as a graph node and the similarity of coupled feature vectors between sensing nodes as graph edge weights;

[0019] Perform multi-hop cooperative propagation operation on the cooperative propagation graph, and iteratively update the cooperative propagation information received by each sensing node until convergence, and use the converged cooperative propagation information as the cooperative metric value of the current sensing node.

[0020] A nonlinear projection function is constructed based on the collaborative metric value to map the sensing nodes to the projection space, thereby obtaining the projected coordinates of each sensing node.

[0021] Calculate the spatial distance between projected coordinates, cluster the sensing nodes according to the spatial distance to obtain clusters, calculate the gradient direction and topological connection strength of the collaborative metric between each cluster and other clusters, and determine the collaborative flow relationship between clusters;

[0022] Based on the collaborative flow relationship, the clusters are divided into core layer clusters, convergence layer clusters, and access layer clusters to generate a hierarchical network topology.

[0023] Calculate the spatial distance between projected coordinates, cluster the sensing nodes based on the spatial distance to obtain clusters, calculate the gradient direction and topological connectivity strength of the cooperative metric between each cluster and other clusters, and determine the cooperative flow relationship between clusters, including:

[0024] Calculate the spatial distance between the projected coordinates, and construct a proximity matrix of the sensing nodes based on the spatial distance;

[0025] The proximity relation matrix is ​​subjected to spectral decomposition to extract a set of feature vectors. Based on the feature vectors in the set of feature vectors, the sensing nodes are spatially clustered to obtain clusters. The aggregated collaborative metric value of the cluster is calculated based on the collaborative metric value of all sensing nodes in each cluster.

[0026] Calculate the difference in the aggregation synergy metric between clusters, and determine the gradient direction of the synergy metric between clusters based on the difference;

[0027] The number of physical fiber optic connections between clusters is counted, and the ratio of the number of physical fiber optic connections to the total number of sensing nodes within a cluster is used as the topological connection strength between clusters.

[0028] A gradient field is constructed based on the gradient direction of the co-metric values ​​between clusters. In the gradient field, a sequence of clusters with decreasing co-metric values ​​is traced along the gradient direction of the co-metric values ​​to form a co-transmission chain.

[0029] The topological connection strength between each pair of adjacent clusters in the collaborative transmission chain is extracted, and the minimum topological connection strength is taken as the link carrying strength of the collaborative transmission chain. When the link carrying strength is greater than a preset carrying threshold, it is determined that there is a collaborative flow relationship between adjacent clusters in the collaborative transmission chain.

[0030] Acquire early warning service data from sensing nodes; calculate the service spatial location based on the projected coordinates of the source node and target node of the early warning service data; calculate the correlation weight between the service spatial location and the service attribute features of the early warning service data; and generate a service feature vector including:

[0031] Acquire early warning service data from sensing nodes, and extract the source node projection coordinates and target node projection coordinates of the early warning service data;

[0032] Construct a spatial path between the projected coordinates of the source node and the projected coordinates of the target node, collect historical service traffic data of the sensing nodes on the spatial path and perform time-series decomposition to obtain the evolution characteristics of the service traffic, calculate the degree of aggregation and diffusion of the service traffic based on the evolution characteristics, and determine the dwell position of the service traffic as the service spatial position based on the comparison difference between the degree of aggregation and diffusion.

[0033] Extract the business attribute features of the early warning business data;

[0034] Determine the attribute sampling area around the business space location, acquire the business attribute features of historical business data within the attribute sampling area and construct the attribute distribution function, calculate the distribution position of the business attribute features in the attribute distribution function, and determine the association weight between the business space location and the business attribute features of the early warning business data based on the distribution position and the distribution shape of the attribute distribution function.

[0035] The business feature vector of early warning business data is generated by combining the business spatial location with the associated weight.

[0036] Calculate the demand intensity of early warning service data on transmission paths in the hierarchical network topology based on the service feature vector, calculate the path adjustment gradient based on the demand intensity and the carrying status of the transmission path, and perform local reconstruction of the hierarchical network topology based on the path adjustment gradient, including:

[0037] Extract business spatial location and association weights from business feature vectors;

[0038] In the hierarchical network topology, locate the target level containing the service spatial location, construct a set of candidate paths from the source node to the target node in the target level, obtain the path topology for each candidate path in the candidate path set, calculate the spatial deviation of the service spatial location in the path topology, and attenuate and modulate the association weights based on the spatial deviation to obtain the demand intensity of the candidate path.

[0039] Obtain the resource occupancy status of each candidate path in the candidate path set, and determine the carrying status of each candidate path based on the resource occupancy status;

[0040] For each candidate path in the candidate path set, a supply-demand balance relationship between demand intensity and carrying capacity is constructed. Deviation analysis is performed on the supply-demand balance relationship to obtain a supply-demand mismatch vector. The mismatch direction and mismatch intensity are extracted from the supply-demand mismatch vector and coupled to form the path adjustment gradient of the candidate path.

[0041] Based on the path adjustment gradient of each candidate path, the topology resource configuration of the corresponding path is adjusted to complete the local reconstruction of the hierarchical network topology.

[0042] The path adjustment gradient is backpropagated to the sensing nodes. The network state impact is predicted based on the coupled feature vector. The service carrying weights of the sensing nodes are dynamically adjusted. The early warning service data is then allocated to the locally reconstructed transmission path according to the adjusted service carrying weights.

[0043] Obtain the path adjustment gradient of each candidate path during the local reconstruction process;

[0044] Tracing back along the candidate path to the sensing node, the path adjustment gradients of each candidate path associated with each sensing node are aggregated according to the path topology connection strength to obtain the gradient response of the sensing node. The gradient response is then applied to the sensing node as a backpropagation stimulus to obtain the activated sensing node.

[0045] Extract the coupling feature vectors of the activated sensing nodes, perform temporal unfolding on the coupling feature vectors to obtain the state change trend, and perform topological correlation analysis on the coupling feature vectors to obtain the topological change trend;

[0046] The network state impact prediction value is obtained by coupling the state change trend with the topology change trend. The network state impact prediction value is then coupled with the gradient response value to calculate the weight adjustment value. The service carrying weight of the sensing node is then dynamically adjusted based on the weight adjustment value.

[0047] An allocation scheme is constructed based on the adjusted service carrying weights, and the early warning service data is allocated to the sensing nodes according to the allocation scheme and transmitted through the locally reconstructed transmission path.

[0048] A second aspect of the present invention provides an electronic device, comprising:

[0049] processor;

[0050] Memory used to store processor-executable instructions;

[0051] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0052] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0053] This invention constructs a network state tensor and extracts coupled feature vectors to achieve accurate collaborative measurement and reasonable clustering of fiber optic sensing nodes. This effectively constructs a hierarchical network topology adapted to the characteristics of early warning services, improving network resource utilization efficiency and system scalability. A correlation analysis mechanism between service spatial location and service attribute characteristics is introduced. By calculating the demand intensity of early warning services for transmission paths and path adjustment gradients, intelligent local reconstruction of the network topology is achieved, avoiding the system overhead of global reconstruction while ensuring the transmission quality of early warning services. Employing a network state prediction method based on coupled feature vectors and a dynamic service bearer weight adjustment strategy, optimized allocation of early warning service data is achieved, improving the system's adaptability to environmental changes and response speed to emergencies, and enhancing the reliability and real-time performance of the fiber optic early warning system. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the network architecture and networking method of the fiber optic early warning system according to an embodiment of the present invention.

[0055] Figure 2 This is a flowchart of the local network topology reconstruction process according to an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 1 This is a flowchart illustrating the network architecture and networking method of the fiber optic early warning system according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:

[0059] Obtain the state information of sensing nodes in a distributed optical fiber sensor network, construct the network state tensor and decompose it to obtain the coupled feature vector;

[0060] The collaborative metric value of the sensing nodes is calculated based on the coupling feature vector. The sensing nodes are then projected according to the collaborative metric value to obtain the projected coordinates. Based on the projected coordinates, the sensing nodes are clustered to generate a hierarchical network topology.

[0061] Acquire early warning service data from sensing nodes, calculate the service spatial location based on the projection coordinates of the source node and target node of the early warning service data, calculate the correlation weight between the service spatial location and the service attribute features of the early warning service data, and generate a service feature vector.

[0062] The demand intensity of early warning service data on the transmission path in the hierarchical network topology is calculated based on the service feature vector. The path adjustment gradient is calculated based on the demand intensity and the carrying status of the transmission path. The hierarchical network topology is then partially reconstructed based on the path adjustment gradient.

[0063] The path adjustment gradient is backpropagated to the sensing node. The network state impact is predicted based on the coupled feature vector. The service carrying weight of the sensing node is dynamically adjusted. The early warning service data is allocated to the locally reconstructed transmission path according to the adjusted service carrying weight.

[0064] In one optional implementation, acquiring the state information of sensing nodes in a distributed optical fiber sensor network, constructing a network state tensor, and decomposing it to obtain a coupled feature vector includes:

[0065] Collect optical signal parameters and service load parameters of each sensing node to generate a node state vector;

[0066] The fiber optic connection relationship and spatial location information of the sensing nodes are obtained. Adjacent sensing node pairs are determined based on the fiber optic connection relationship. The link hop count between each pair of adjacent sensing nodes is calculated. The physical distance between each pair of adjacent sensing nodes is calculated based on the spatial location information of the nodes.

[0067] The link hop count and physical distance are weighted and summed to obtain the inter-node coupling degree, and a node coupling matrix with the inter-node coupling degree as the element is constructed.

[0068] The node state vector is expanded into a state matrix according to the time sequence. The state matrix is ​​then multiplied with the node coupling matrix to generate a network state tensor containing node dimension, time dimension and coupling dimension.

[0069] The network state tensor is decomposed along the node dimension, time dimension, and coupling dimension, and node features, temporal features, and coupling features are extracted and fused to obtain the coupling feature vector.

[0070] In a distributed optical fiber sensor network, to acquire the state information of each sensing node and construct the network state tensor, state data is first collected from each sensing node. This data includes optical signal parameters and service load parameters. Optical signal parameters include indicators such as optical power, optical attenuation, signal-to-noise ratio (SNR), and wavelength drift; service load parameters include indicators such as data throughput, processing latency, queue length, and resource utilization. For example, for a given sensing node, the collected parameters might be: optical power -15dBm, optical attenuation 0.35dB / km, SNR 28dB, data throughput 125Mbps, and processing latency 15ms. These parameters are then combined in a preset order to form a node state vector describing the current state of that node.

[0071] The process involves acquiring the fiber optic connections between sensing nodes, typically represented by an adjacency table that records which nodes have direct fiber optic connections. Simultaneously, the spatial location information of the nodes is acquired, including their three-dimensional coordinates. Adjacent sensing node pairs are determined based on these fiber optic connections; for example, in a network with 100 sensing nodes, there might be approximately 300 pairs of adjacent nodes. For each pair of adjacent sensing nodes in the network, the link hop count between them is calculated, representing the minimum number of connections required to travel from one node to another along the network topology. This can be achieved using breadth-first search or Dijkstra's algorithm. Furthermore, the physical distance between each pair of adjacent sensing nodes is calculated using the spatial coordinates of the nodes, employing a three-dimensional Euclidean distance calculation method.

[0072] The coupling degree between nodes is obtained by weighted summation of link hop count and physical distance. Specifically, the weight of link hop count can be set to 0.6 and the weight of physical distance to 0.4, and then the summation is performed after normalization. For example, if the link hop count between two nodes is 2 and the physical distance is 50 meters, after normalization, the normalized value of link hop count is 0.5 and the normalized value of physical distance is 0.25, then the coupling degree between nodes is 0.6 × 0.5 + 0.4 × 0.25 = 0.4. The coupling degree of all adjacent node pairs in the network is calculated in this way to construct a node coupling matrix, where each element of the matrix represents the strength of the coupling relationship between the corresponding node pairs.

[0073] After collecting node state data for a sufficient duration, the state vectors of each node are arranged in a time sequence to form a state matrix. If there are 50 sensing nodes in the network, each node's state vector contains 10 parameters, and data is collected at 100 time points, a 50×100×10 three-dimensional state data set can be formed. This state matrix is ​​then multiplied by the previously constructed node coupling matrix using tensor operations to generate the network state tensor. This tensor contains information in three dimensions: node dimension, time dimension, and coupling dimension, fully expressing the complex relationships between node states over time and the interactions between nodes.

[0074] The generated network state tensor is decomposed to extract features from the node, time, and coupling dimensions. Tensor decomposition techniques can be used to break down the original high-dimensional tensor into a combination of multiple low-dimensional core tensors and factor matrices. At the node dimension, node feature vectors characterizing different node properties are extracted; at the time dimension, temporal feature vectors reflecting the temporal evolution of the state are extracted; and at the coupling dimension, coupling feature vectors representing the interaction patterns between nodes are extracted.

[0075] Finally, the extracted node features, temporal features, and coupling features are fused to obtain the final coupling feature vector. Fusion methods can include weighted averaging or feature concatenation. For example, for a certain perceptual network, the final coupling feature vector might contain 50 elements, where the first 20 elements represent static coupling characteristics between nodes, the middle 15 elements represent time-varying coupling characteristics, and the last 15 elements represent group coupling characteristics among multiple nodes. These features can be used in applications such as network anomaly detection, fault diagnosis, and performance optimization.

[0076] In this embodiment, by simultaneously considering the state parameters, spatial topology, and connection characteristics of sensing nodes, a multi-dimensional characterization of the operational status of the distributed fiber optic sensor network is achieved. By constructing a network state tensor that integrates nodes, time, and coupling relationships and performing multi-dimensional decomposition, the potential collaborative change patterns and coupling characteristics between nodes can be effectively extracted, improving the accuracy and robustness of state monitoring, enabling early identification of abnormal fluctuations or failure trends, thereby enhancing the overall sensing performance and intelligent monitoring level of the system.

[0077] In one optional implementation, the process of calculating the collaborative metric value of the sensing nodes based on the coupling feature vector, projecting the sensing nodes according to the collaborative metric value to obtain projected coordinates, and clustering the sensing nodes based on the projected coordinates to generate a hierarchical network topology includes:

[0078] Construct a collaborative propagation graph of coupled feature vectors, treating each sensing node as a graph node and the similarity of coupled feature vectors between sensing nodes as graph edge weights;

[0079] A multi-hop cooperative propagation operation is performed on the cooperative propagation graph. The cooperative propagation information received by each sensing node is iteratively updated until convergence. The converged cooperative propagation information is used as the cooperative metric value of the current sensing node.

[0080] A nonlinear projection function is constructed based on the collaborative metric value to map the sensing nodes to the projection space, thereby obtaining the projected coordinates of each sensing node.

[0081] Calculate the spatial distance between projected coordinates, cluster the sensing nodes according to the spatial distance to obtain clusters, calculate the gradient direction and topological connection strength of the collaborative metric between each cluster and other clusters, and determine the collaborative flow relationship between clusters;

[0082] Based on the collaborative flow relationship, the clusters are divided into core layer clusters, convergence layer clusters, and access layer clusters to generate a hierarchical network topology.

[0083] The method for constructing a hierarchical network topology is achieved through collaborative measurement, projection mapping, and cluster analysis of sensing nodes. The implementation process starts with the coupled feature vectors of the sensing nodes, projects the nodes based on the collaborative measurement values, and constructs a hierarchical network topology based on the projection results.

[0084] When constructing the cooperative propagation graph of coupled feature vectors, each sensing node is considered as a node in the graph. For any two sensing nodes i and j, the similarity between their coupled feature vectors is calculated as the edge weight between the nodes. The similarity calculation can use the cosine similarity method. For the coupled feature vectors Vi and Vj of nodes i and j, their similarity can be expressed as the dot product of the two vectors divided by the product of their respective magnitudes. For example, if the coupled feature vectors of two sensing nodes are [0.5, 0.3, 0.8] and [0.4, 0.6, 0.7], the calculated similarity is 0.89, which is the weight of the corresponding edge in the cooperative propagation graph.

[0085] Multi-hop cooperative propagation is performed on the constructed cooperative propagation graph. This operation is achieved through the iterative transmission of information in the network. For each sensing node, its cooperative propagation information is initialized as a unit vector. In each iteration, the node updates its own cooperative propagation information by aggregating information from its neighboring nodes. The aggregation process considers edge weights, i.e., the similarity between nodes. The iteration process continues until the change in the cooperative propagation information of all nodes is less than a preset threshold, such as 0.001, indicating that the system has reached convergence. The converged cooperative propagation information is used as the cooperative metric value of the node. Taking a network containing 5 sensing nodes as an example, after 15 iterations, the system converges, and the cooperative metric values ​​of each node are [0.35, 0.42, 0.28, 0.51, 0.39].

[0086] Based on the acquired collaborative metric values, a nonlinear projection function is constructed to map the sensing nodes to the projection space. The projection function employs a t-distributed random neighbor embedding method. This method calculates the collaborative similarity distribution between sensing nodes in a high-dimensional space and reconstructs this similarity distribution in a t-distribution in a low-dimensional space. This ensures that similar nodes remain close to each other and dissimilar nodes remain far apart in the projection space, effectively revealing the overall topological relationships and hierarchical distribution characteristics while preserving the local neighborhood structure. This achieves visualization and clustering optimization of the collaborative characteristics of sensing nodes. During the projection process, the projection space dimension is set to 2 or 3, the learning rate is 0.01, and the number of iterations is 1000. The projection result is the coordinates of each sensing node in the projection space. Taking the aforementioned 5 nodes as an example, the coordinates after projection into the two-dimensional space might be: Node 1 (0.23, 0.45), Node 2 (0.15, 0.52), Node 3 (0.67, 0.33), Node 4 (0.78, 0.29), and Node 5 (0.41, 0.36).

[0087] The spatial distance between projected coordinates is calculated using Euclidean distance. For two nodes i and j, their projected coordinates are (x, j) and (j) respectively. i y i ) and (x j y j Spatial distance is calculated as the square root of the sum of the squares of the coordinate differences between two points. Based on the calculated spatial distance, a density peak clustering algorithm is used to cluster the sensing nodes. This algorithm first calculates the local density of each point and the minimum distance to high-density points, and then identifies density peak points as cluster centers. For each non-center point, it is assigned to the cluster containing the nearest higher-density point. Taking the aforementioned 5 nodes as an example, assume that the clustering result groups nodes 1 and 2 into one cluster, nodes 3 and 4 into another cluster, and node 5 into a separate cluster.

[0088] When determining the co-current relationships between clusters, the co-current metric gradient and topological connectivity strength are calculated for each pair of clusters. The co-current metric gradient represents the direction and magnitude of change in the co-current metric between two clusters, while the topological connectivity strength measures the tightness of the connection between the two clusters. For example, a co-current metric gradient of 0.25 between cluster A and cluster B indicates an increase in the co-current metric from cluster A to cluster B; a topological connectivity strength of 0.7 between the two clusters indicates a strong connection between them.

[0089] Based on the cooperative flow relationship, the clusters are divided into core layer clusters, aggregation layer clusters, and access layer clusters. Clusters with the highest cooperative metric value and strong connections to other clusters are classified as core layer clusters; these clusters typically receive cooperative flows from multiple other clusters. Clusters with moderate cooperative metric values ​​and both receiving and output flows are classified as aggregation layer clusters; these clusters act as relays and convergents. Clusters with low cooperative metric values ​​and primarily output cooperative flows to other clusters are classified as access layer clusters; these clusters are mainly responsible for initial data collection and input. According to the aforementioned clustering results, if cluster 1 (containing nodes 3 and 4) has the highest cooperative metric value, it is classified as a core layer cluster; cluster 2 (containing node 5) is a aggregation layer cluster; and cluster 3 (containing nodes 1 and 2) is an access layer cluster. Thus, a hierarchical network topology containing a core layer, aggregation layer, and access layer is constructed.

[0090] In this embodiment, by combining cooperative propagation with nonlinear projection, the cooperative relationships between sensing nodes can be accurately characterized in a high-dimensional coupled feature space. Iterative propagation achieves global convergence of cooperative information between nodes, effectively suppressing noise interference and strengthening the correlation features between nodes. Based on clustering and hierarchical processing using projected coordinates, a multi-layered network topology conforming to actual business relationships can be adaptively constructed, achieving intelligent partitioning of the core, aggregation, and access layers, thereby improving the interpretability and dynamic optimization capabilities of the network structure.

[0091] In one optional implementation, the spatial distance between projected coordinates is calculated, and the sensing nodes are clustered according to the spatial distance to obtain clusters. The gradient direction of the cooperative metric and the topological connectivity strength between each cluster and other clusters are calculated, and the cooperative flow relationship between clusters is determined, including:

[0092] Calculate the spatial distance between the projected coordinates, and construct a proximity matrix of the sensing nodes based on the spatial distance;

[0093] The proximity relation matrix is ​​subjected to spectral decomposition to extract a set of feature vectors. Based on the feature vectors in the set of feature vectors, the sensing nodes are spatially clustered to obtain clusters. The aggregated collaborative metric value of the cluster is calculated based on the collaborative metric value of all sensing nodes in each cluster.

[0094] Calculate the difference in the aggregation synergy metric between clusters, and determine the gradient direction of the synergy metric between clusters based on the difference;

[0095] The number of physical fiber optic connections between clusters is counted, and the ratio of the number of physical fiber optic connections to the total number of sensing nodes within a cluster is used as the topological connection strength between clusters.

[0096] A gradient field is constructed based on the gradient direction of the co-metric values ​​between clusters. In the gradient field, a sequence of clusters with decreasing co-metric values ​​is traced along the gradient direction of the co-metric values ​​to form a co-transmission chain.

[0097] The topological connection strength between each pair of adjacent clusters in the collaborative transmission chain is extracted, and the minimum topological connection strength is taken as the link carrying strength of the collaborative transmission chain. When the link carrying strength is greater than a preset carrying threshold, it is determined that there is a collaborative flow relationship between adjacent clusters in the collaborative transmission chain.

[0098] In the specific implementation process, the projected coordinates of each sensing node are first obtained. These projected coordinates can be two-dimensional or three-dimensional coordinates obtained by dimensionality reduction of high-dimensional data, or they can be the physical location coordinates of the sensing node itself in space. Assuming there are 100 sensing nodes in total, each node has a corresponding two-dimensional coordinate (x, y), for example, the coordinates of node 1 are (1.2, 3.5) and the coordinates of node 2 are (1.5, 3.8).

[0099] Calculate the spatial distance between the projected coordinates of each sensing node. Taking two-dimensional coordinates as an example, the spatial distance between two nodes i and j can be calculated using Euclidean distance. For example, the spatial distance between node 1 and node 2 mentioned above is approximately 0.42. Based on the calculated spatial distances between all node pairs, construct a proximity matrix. This matrix can be a binary matrix, where the corresponding element is 1 when the distance between two nodes is less than a predetermined threshold (e.g., 0.5), and 0 otherwise. Alternatively, a weighted proximity matrix can be constructed, where the element values ​​are the reciprocal of the distance or the Gaussian kernel function value.

[0100] Spectral decomposition is performed on the proximity matrix to extract a set of eigenvectors. Spectral decomposition generates multiple eigenvectors; typically, the eigenvectors corresponding to the smallest few non-zero eigenvalues ​​are selected for subsequent clustering. In practice, the Laplacian matrix of the proximity matrix can be calculated first, and then eigenvalue decomposition can be performed on the Laplacian matrix. Assume the first three eigenvectors are selected, each with the same dimension as the number of nodes, i.e., 100 dimensions.

[0101] Spatial clustering of the sensing nodes is performed based on the extracted feature vector set. Clustering algorithms such as K-means can be used to divide 100 nodes into several clusters. Assuming the number of clusters is set to 5, 5 clusters are obtained, containing 20, 25, 15, 18, and 22 nodes respectively. Nodes within each cluster are close to each other in the feature space, indicating that they have similar topological relationships.

[0102] For each cluster, its aggregated co-operation metric is calculated. The aggregated co-operation metric is calculated based on the co-operation metrics of all sensing nodes within the cluster. The co-operation metric can be a performance parameter of a node, such as data transmission efficiency, signal strength, or processing capability. The calculation method can be the average, median, or weighted sum of the co-operation metrics of the nodes within the cluster. Assume the aggregated co-operation metrics for the five clusters are 85.6, 72.3, 90.1, 68.7, and 79.5, respectively.

[0103] Calculate the difference in the aggregation co-metric value between each cluster to determine the gradient direction of the co-metric value. For example, the difference between cluster 1 and cluster 2 is 13.3, and the difference between cluster 1 and cluster 3 is -4.5. When the difference is positive, the gradient direction points from cluster 1 to cluster 2; when the difference is negative, the gradient direction points from cluster 3 to cluster 1. In this way, the gradient direction between any two adjacent clusters can be determined.

[0104] Count the number of physical fiber optic connections between each cluster. The number of physical fiber optic connections reflects the physical communication capability between two clusters. Assuming there are 10 fiber optic connections between cluster 1 and cluster 2, with 20 nodes in cluster 1 and 25 nodes in cluster 2, the topology connection strength between cluster 1 and cluster 2 is calculated to be 10 / 45 ≈ 0.22. Similarly, calculate the topology connection strength between all cluster pairs.

[0105] A gradient field is constructed based on the gradient direction of the synergy metric between clusters. Within this gradient field, a sequence of clusters with decreasing synergy metrics is traced along the gradient direction to form a synergy propagation chain. For example, if cluster 3 → cluster 1 → cluster 5 → cluster 2 → cluster 4 is a path with gradually decreasing synergy metrics, then they constitute a synergy propagation chain.

[0106] Extract the topological connection strength between each pair of adjacent clusters in the cooperative transit chain, and take the minimum value as the link carrying strength of the chain. Taking the above cooperative transit chain as an example, if the topological connection strength between cluster 3 and cluster 1 is 0.30, between cluster 1 and cluster 5 it is 0.25, between cluster 5 and cluster 2 it is 0.18, and between cluster 2 and cluster 4 it is 0.20, then the link carrying strength of this cooperative transit chain is 0.18.

[0107] When the link carrying strength is greater than a preset carrying threshold, a cooperative flow relationship is determined to exist between adjacent clusters on the cooperative transmission chain. Assuming the preset carrying threshold is 0.15, the link carrying strength of the aforementioned cooperative transmission chain, 0.18, is greater than the threshold, thus confirming a cooperative flow relationship between all adjacent cluster pairs on this chain. This means that cooperative flow relationships exist between the adjacent cluster pairs: cluster 3→cluster 1, cluster 1→cluster 5, cluster 5→cluster 2, and cluster 2→cluster 4.

[0108] In this embodiment, by combining spectral clustering and collaborative metric analysis, the precise identification of multi-level collaborative relationships among sensing nodes is achieved. By extracting potential cluster structures through spectral decomposition of the proximity matrix, node groups with similar collaborative characteristics can be accurately classified. Furthermore, by comprehensively determining the collaborative flow direction between clusters using the gradient direction of the aggregated collaborative metric value and the topological connection strength, the information transmission path and energy coupling direction can be dynamically revealed. This enables quantifiable modeling and hierarchical display of internal network collaboration, thereby improving the structural analysis and operational optimization capabilities of complex fiber optic sensor networks.

[0109] In one optional implementation, the early warning service data of the sensing node is acquired, the service spatial location is calculated based on the projected coordinates of the source node and the target node of the early warning service data, the correlation weight between the service spatial location and the service attribute features of the early warning service data is calculated, and a service feature vector is generated, including:

[0110] Acquire early warning service data from sensing nodes, and extract the source node projection coordinates and target node projection coordinates of the early warning service data;

[0111] Construct a spatial path between the projected coordinates of the source node and the projected coordinates of the target node, collect historical service traffic data of the sensing nodes on the spatial path and perform time-series decomposition to obtain the evolution characteristics of the service traffic, calculate the degree of aggregation and diffusion of the service traffic based on the evolution characteristics, and determine the dwell position of the service traffic as the service spatial position based on the comparison difference between the degree of aggregation and diffusion.

[0112] Extract the business attribute features of the early warning business data;

[0113] Determine the attribute sampling area around the business space location, acquire the business attribute features of historical business data within the attribute sampling area and construct the attribute distribution function, calculate the distribution position of the business attribute features in the attribute distribution function, and determine the association weight between the business space location and the business attribute features of the early warning business data based on the distribution position and the distribution shape of the attribute distribution function.

[0114] The business feature vector of early warning business data is generated by combining the business spatial location with the associated weight.

[0115] In this context, sensing nodes can be devices such as routers, switches, and firewalls in the network, which collect data traffic information. Early warning service data refers to network traffic data detected by these sensing nodes that may pose security risks. In practical applications, early warning service data can be obtained through sensors installed in the network. For example, if a security gateway deployed in a regional network detects abnormal traffic, the generated early warning data may include the source IP address 192.168.1.100 and the destination IP address 192.168.2.200.

[0116] Extract the source node's projected coordinates and the target node's projected coordinates from the early warning service data. Projected coordinates refer to the position coordinates of a network node mapped to a two-dimensional or three-dimensional space. In practical applications, the position on the network topology map can be used as the node's projected coordinates. For example, the projected coordinates of the source IP address 192.168.1.100 are (10, 15), and the projected coordinates of the target IP address 192.168.2.200 are (40, 35).

[0117] Construct the spatial path between the projected coordinates of the source node and the projected coordinates of the target node. A spatial path refers to the data transmission path from the source node to the target node. In practical applications, routing information in the network can be used to determine the spatial path. For example, the spatial path from the source node (10, 15) to the target node (40, 35) might pass through intermediate nodes (20, 20) and (30, 30).

[0118] Historical traffic data from sensing nodes along the spatial path is collected and decomposed into time-series components. Time-series decomposition refers to breaking down time-series data into trend, seasonal, and random components. In practical applications, hourly traffic data from the past 24 hours can be collected. For example, the traffic data from the source node to the intermediate node (20, 20) is [100, 120, 150, 180, 200, 220, 250, 280, 300, 320, 350, 380, 400, 420, 450, 480, 500, 520, 550, 580, 600, 620, 650, 680], and the traffic data from the intermediate node (20, 20) to the intermediate node (30, 30) is [90, 110, 140, 170, 190, 20 ... 210, 240, 270, 290, 310, 340, 370, 390, 410, 440, 470, 490, 510, 540, 570, 590, 610, 640, 670], the traffic data from intermediate node (30, 30) to the target node is [80, 100, 130, 160, 180, 200, 230, 260, 280, 300, 330, 360, 380, 400, 430, 460, 480, 500, 530, 560, 580, 600, 630, 660].

[0119] The evolutionary characteristics of business traffic are obtained, and the aggregation and diffusion degrees of business traffic are calculated based on these characteristics. Evolutionary characteristics refer to the features of business traffic changes over time. Aggregation degree refers to the degree to which business traffic is concentrated at a certain node, and diffusion degree refers to the degree to which business traffic spreads out from a certain node. The aggregation degree can be represented by calculating the difference between the inflow and outflow traffic at each node. For example, for the intermediate node (20, 20), the inflow traffic is the traffic from the source node to the intermediate node (20, 20), and the outflow traffic is the traffic from the intermediate node (20, 20) to the intermediate node (30, 30). The aggregation degree for the most recent hour is 680-670=10. Similarly, the aggregation degree for the intermediate node (30, 30) for the most recent hour is 670-660=10.

[0120] The location where traffic resides is determined by comparing the difference between the degree of clustering and the degree of diffusion. The node with the highest degree of clustering can be selected as the location where traffic resides. Based on the calculations above, the intermediate nodes (20, 20) and (30, 30) have the same degree of clustering (10). Therefore, the intermediate node (20, 20), which is closer to the source node, can be selected as the location of the traffic in the traffic space. Thus, the traffic location is (20, 20).

[0121] Extract the business attribute features of the early warning business data. Business attribute features refer to the characteristics of the early warning business data, such as protocol type, port number, and data packet size. The business attribute features of early warning business data may include: protocol type is TCP, source port is 8080, destination port is 80, data packet size is 1024 bytes, number of data packets is 100, and duration is 60 seconds.

[0122] Define an attribute sampling region around the business space location. The attribute sampling region is an area centered on the business space location used to collect business attribute features from historical business data. In practical applications, a circular area with a radius of 5 centered at the business space location (20, 20) can be used as the attribute sampling region, i.e., the area with coordinates [(15, 15), (25, 25)].

[0123] The system acquires the business attribute features of historical business data within the attribute sampling area and constructs an attribute distribution function. An attribute distribution function is a function that describes the distribution of business attribute features. In practical applications, the system can acquire the business attribute features of historical business data from the past 24 hours within the attribute sampling area and statistically analyze the distribution of each feature. For example, the distribution of protocol types is as follows: TCP 80%, UDP 15%, ICMP 5%; the distribution of source ports is: 8080 60%, 8081 20%, 8082 20%; the distribution of destination ports is: 80 70%, 443 20%, 22 10%; the distribution of packet size is: 1024 bytes 50%, 2048 bytes 30%, 4096 bytes 20%; the distribution of packet quantity is: 100 packets 40%, 200 packets 30%, 300 packets 30%; and the distribution of duration is: 60 seconds 45%, 120 seconds 35%, 180 seconds 20%.

[0124] The distribution position of business attribute features within the attribute distribution function is calculated. Based on this distribution position and the distribution pattern of the attribute distribution function, the correlation weight between the business spatial position and the business attribute features of the early warning business data is determined. The percentage position of each business attribute feature of the early warning business data within the attribute distribution function can be calculated. For example, if the protocol type is TCP, it accounts for 80% of the protocol type distribution; the source port is 8080, accounting for 60% of the source port distribution; the destination port is 80, accounting for 70% of the destination port distribution; the packet size is 1024 bytes, accounting for 50% of the packet size distribution; the number of packets is 100, accounting for 40% of the packet number distribution; and the duration is 60 seconds, accounting for 45% of the duration distribution. The correlation weight can be defined as the average of these percentages, i.e., (80%+60%+70%+50%+40%+45%) / 6=57.5%.

[0125] The business feature vector of early warning business data is generated by combining the business spatial location with the association weight. The coordinates of the business spatial location and the association weight can be combined into a vector. For example, if the business spatial location is (20, 20) and the association weight is 57.5%, then the business feature vector is [20, 20, 0.575].

[0126] In this embodiment, spatial correlation modeling of early warning service data is achieved by fusing the projected coordinates of sensing nodes with service attribute features. By performing temporal decomposition and clustering analysis on the historical service traffic along the path between the source and target nodes, the residence area of ​​service activities can be accurately located, reflecting the dynamic evolution characteristics of the service flow. Combining the attribute distribution function to calculate the correlation weight quantifies the coupling strength between the spatial location of the service and its attribute features, thereby enabling multi-dimensional feature representation and anomaly correlation identification of early warning services, improving the accuracy of service situation awareness and early warning.

[0127] like Figure 2 The diagram illustrates the network topology local reconstruction process in this embodiment.

[0128] In one optional implementation, the process of calculating the demand intensity of early warning service data for transmission paths in the hierarchical network topology based on service feature vectors, calculating path adjustment gradients based on the demand intensity and the carrying status of the transmission paths, and performing local reconstruction of the hierarchical network topology based on the path adjustment gradients includes:

[0129] Extract business spatial location and association weights from business feature vectors;

[0130] In the hierarchical network topology, locate the target level containing the service spatial location, construct a set of candidate paths from the source node to the target node in the target level, obtain the path topology for each candidate path in the candidate path set, calculate the spatial deviation of the service spatial location in the path topology, and attenuate and modulate the association weights based on the spatial deviation to obtain the demand intensity of the candidate path.

[0131] Obtain the resource occupancy status of each candidate path in the candidate path set, and determine the carrying status of each candidate path based on the resource occupancy status;

[0132] For each candidate path in the candidate path set, a supply-demand balance relationship between demand intensity and carrying capacity is constructed. Deviation analysis is performed on the supply-demand balance relationship to obtain a supply-demand mismatch vector. The mismatch direction and mismatch intensity are extracted from the supply-demand mismatch vector and coupled to form the path adjustment gradient of the candidate path.

[0133] Based on the path adjustment gradient of each candidate path, the topology resource configuration of the corresponding path is adjusted to complete the local reconstruction of the hierarchical network topology.

[0134] In this invention, the extraction of business spatial location and correlation weight from the business feature vector is achieved through a feature parsing module. For example, for early warning business data with multi-dimensional features, its feature vector can be represented as {location coordinates (x=125.36, y=43.88), business type="network security early warning", priority=8, timeliness="high", correlation weight=0.85}. The feature parsing module will extract the business spatial location (125.36, 43.88) and its correlation weight 0.85 from this vector as the basic data for subsequent processing.

[0135] Locating the target layer containing the spatial location of services in a hierarchical network topology is accomplished through a hierarchical mapping algorithm. This algorithm compares the spatial location of services with the coverage areas of each layer of the hierarchical network topology. For example, a four-layer network structure may include a backbone layer (L1), a regional network layer (L2), a metropolitan area network layer (L3), and an access network layer (L4), each covering a different geographical area. For the location coordinates (125.36, 43.88), spatial positioning determines that it is within the coverage area of ​​the L3 metropolitan area network; therefore, the target layer is L3.

[0136] Constructing a set of candidate paths from the source node to the target node at the target layer involves path search techniques. Assuming that in layer L3, the source node is S (125.30, 43.85) and the target node is D (125.42, 43.91), topology analysis yields three possible transmission paths: path P1: S→A→B→D, path P2: S→C→D, and path P3: S→E→F→D. These three paths constitute the candidate path set.

[0137] Obtaining the path topology for candidate paths refers to extracting the connection relationships and attribute information of all nodes on the path. Taking path P1 as an example, its topology information includes the location coordinates of nodes S, A, B, and D, as well as the bandwidth, delay, and other attributes of the three links SA, AB, and BD. Specific data might be: S (125.30, 43.85), A (125.33, 43.87), B (125.38, 43.89), D (125.42, 43.91), link SA bandwidth 100Mbps, delay 5ms, link AB bandwidth 80Mbps, delay 8ms, link BD bandwidth 120Mbps, delay 4ms.

[0138] The spatial deviation of a service location from its position within the path topology is calculated using a spatial distance method. For the service location (125.36, 43.88), the distances to each node on path P1 are calculated: distance S is 0.072, distance A is 0.038, distance B is 0.022, and distance D is 0.068. The minimum distance of 0.022 is taken as the spatial deviation of the service location from path P1. Similarly, the spatial deviations from paths P2 and P3 are calculated to be 0.031 and 0.045, respectively.

[0139] The correlation weights are attenuated and modulated based on the degree of spatial deviation using an inverse proportional attenuation function. For an initial correlation weight of 0.85, adjustments are made according to the spatial deviation of each path: the attenuation coefficient for path P1 is 1 / (1+0.022×10)=0.82, and the adjusted demand intensity is 0.85×0.82=0.697; the attenuation coefficient for path P2 is 1 / (1+0.031×10)=0.76, and the adjusted demand intensity is 0.85×0.76=0.646; the attenuation coefficient for path P3 is 1 / (1+0.045×10)=0.69, and the adjusted demand intensity is 0.85×0.69=0.587.

[0140] The resource utilization status of each candidate path in the candidate path set is obtained through the network monitoring module. This module collects parameters such as the current bandwidth utilization and latency of the links on each path. For example, the current bandwidth utilization of the three links on path P1 is 60%, 75%, and 45%, with an average utilization of 60%; the bandwidth utilization of the two links on path P2 is 85% and 70%, with an average utilization of 77.5%; and the bandwidth utilization of the three links on path P3 is 40%, 35%, and 50%, with an average utilization of 41.7%.

[0141] Determining the bearer status of each candidate path based on resource occupancy involves bearer capacity assessment. Resource occupancy is converted into bearer status values ​​by setting threshold ranges. For example, the bearer status value is 1.0 for bandwidth utilization between 0-50%, 0.8 for 51-70%, 0.6 for 71-85%, and 0.4 for above 85%. Based on this, the bearer status value for path P1 is calculated to be 0.8, for path P2 to be 0.6, and for path P3 to be 1.0.

[0142] Establishing a supply-demand balance between demand intensity and carrying capacity refers to setting a benchmark for comparison between the two. For path P1, the demand intensity is 0.697 and the carrying capacity value is 0.8, forming a supply-demand comparison of (0.697, 0.8); for path P2 it is (0.646, 0.6); and for path P3 it is (0.587, 1.0).

[0143] The supply-demand mismatch vector is obtained by performing deviation analysis on the supply-demand balance relationship. This is achieved by calculating the difference between demand intensity and carrying capacity value. The mismatch vector for path P1 is 0.8 - 0.697 = 0.103, indicating that supply exceeds demand; the mismatch vector for path P2 is 0.6 - 0.646 = -0.046, indicating that supply is less than demand; and the mismatch vector for path P3 is 1.0 - 0.587 = 0.413, indicating that supply far exceeds demand.

[0144] The mismatch direction and intensity are extracted from the supply-demand mismatch vector. The sign of the mismatch vector is taken as the direction, and the absolute value is taken as the intensity. The mismatch direction of path P1 is positive (supply exceeds demand), and the mismatch intensity is 0.103; the mismatch direction of path P2 is negative (supply is less than demand), and the mismatch intensity is 0.046; the mismatch direction of path P3 is positive, and the mismatch intensity is 0.413.

[0145] The path adjustment gradient for coupling to form candidate paths takes into account both the mismatch direction and the mismatch intensity. For example, by multiplying the direction coefficient by the intensity value, the adjustment gradient for path P1 is +1 × 0.103 = 0.103, the adjustment gradient for path P2 is -1 × 0.046 = -0.046, and the adjustment gradient for path P3 is +1 × 0.413 = 0.413.

[0146] Adjusting the topology resource allocation of candidate paths based on their gradients refers to adjusting network resources according to the magnitude and direction of the gradient values. For paths P1 and P3 with positive gradients, resource allocation can be appropriately reduced; for path P2 with negative gradients, resource allocation should be increased. Specifically, 10% of the link bandwidth of path P1 can be reallocated to path P2, and 15% of the link bandwidth of path P3 can be reallocated to path P2, thereby completing a partial reconstruction of the hierarchical network topology and making network resource allocation more in line with business needs.

[0147] In this embodiment, by combining service feature vectors with a hierarchical network topology, dynamic matching of service demands and network resources is achieved. Through supply-demand balance analysis of demand intensity and carrying capacity, the resource suitability between different paths can be quantified, and the direction and intensity of mismatch can be extracted to form a path adjustment gradient. Based on this gradient, local topology reconstruction can optimize resource allocation and path utilization while ensuring service continuity, achieving adaptive adjustment and dynamic optimization of the network structure, thereby improving the operational efficiency and stability of the distributed fiber optic sensor network.

[0148] In one optional implementation, the path adjustment gradient is backpropagated to the sensing node, the network state impact is predicted based on the coupling feature vector, the service bearer weight of the sensing node is dynamically adjusted, and the early warning service data is allocated to the locally reconstructed transmission path according to the adjusted service bearer weight, including:

[0149] Obtain the path adjustment gradient of each candidate path during the local reconstruction process;

[0150] Tracing back along the candidate path to the sensing node, the path adjustment gradients of each candidate path associated with each sensing node are aggregated according to the path topology connection strength to obtain the gradient response of the sensing node. The gradient response is then applied to the sensing node as a backpropagation stimulus to obtain the activated sensing node.

[0151] Extract the coupling feature vectors of the activated sensing nodes, perform temporal unfolding on the coupling feature vectors to obtain the state change trend, and perform topological correlation analysis on the coupling feature vectors to obtain the topological change trend;

[0152] The network state impact prediction value is obtained by coupling the state change trend with the topology change trend. The network state impact prediction value is then coupled with the gradient response value to calculate the weight adjustment value. The service carrying weight of the sensing node is then dynamically adjusted based on the weight adjustment value.

[0153] An allocation scheme is constructed based on the adjusted service carrying weights, and the early warning service data is allocated to the sensing nodes according to the allocation scheme and transmitted through the locally reconstructed transmission path.

[0154] In this embodiment, the path adjustment gradient of each candidate path during the local reconstruction process is first obtained. The path adjustment gradient represents the trend and degree of adjustment of the path during the reconstruction process, and can be calculated by monitoring parameters such as path bandwidth changes, latency fluctuations, and packet loss rate. For example, for candidate path P1, its path adjustment gradient G1 can be calculated based on the bandwidth change rate (from 100Mbps to 80Mbps, a change rate of -20%), latency increase rate (from 5ms to 8ms, an increase rate of 60%), and packet loss rate change (from 0.1% to 0.3%), and the final path adjustment gradient G1 is 0.45; similarly, the path adjustment gradient G2 of candidate path P2 is 0.32, and the path adjustment gradient G3 of P3 is 0.28.

[0155] Tracing back along the candidate paths to the sensing nodes, the path adjustment gradients of each candidate path associated with each sensing node are weighted and aggregated according to the path topology connection strength to obtain the gradient response of the sensing node. Topology connection strength can be determined by indicators such as the reliability of the connection between the node and the path, the degree of bandwidth sharing, and the correlation with historical services. Taking sensing node S1 as an example, it connects to candidate paths P1 and P2. The topology connection strength between path P1 and S1 is 0.7, and the topology connection strength between path P2 and S1 is 0.5. Therefore, the gradient response of S1 is calculated as 0.45×0.7+0.32×0.5=0.475. For sensing node S2, it connects to candidate paths P1, P2, and P3. The topology connection strengths between each path and S2 are 0.6, 0.4, and 0.8, respectively. Therefore, the gradient response of S2 is 0.45×0.6+0.32×0.4+0.28×0.8=0.554. The gradient response is then applied to the sensing node as a backpropagation stimulus to activate the sensing node. The activation process is achieved by setting a threshold function. When the gradient response exceeds the preset threshold of 0.4, the sensing node enters the activated state and begins to participate in subsequent calculations.

[0156] The coupling feature vector of the activated sensing node is extracted. This vector contains multi-dimensional features such as node processing capacity, cache status, number of connection paths, and business processing history. Taking S1 as an example, its coupling feature vector includes processing capacity indicators (85 points out of 100), current cache utilization (40%), number of connection paths (2), and historical business success rate (98%). Time-series expansion of the coupling feature vector and analysis of the feature change trend over the past 10 minutes reveal that S1's processing capacity remains stable, while the cache utilization rate increases linearly from 25% to 40%. Simultaneously, topology correlation analysis shows that the connection stability between S1 and surrounding nodes is 0.92, higher than the network average of 0.85, indicating a relatively stable topology.

[0157] The trend of state change is coupled with the trend of topology change for extrapolation. Based on the cache utilization growth rate of S1 and the current network load, it is estimated that the cache utilization rate of S1 may reach 55% within the next 5 minutes, which is still within a safe range (the threshold is 80%). Combined with topology stability analysis, it is predicted that the connection reliability of S1 will not decrease in the short term. Taking all these factors into account, the predicted value of the network state impact of S1 is 0.78 (out of 1, indicating minimal impact). The predicted value of the network state impact is coupled with the gradient response value for calculation: 0.78 × (1 - 0.475) = 0.4095, resulting in a weight adjustment amount of 0.4095. The original service carrying weight is 0.5, and the adjusted service carrying weight is 0.5 + 0.4095 = 0.9095. Similarly, the weight adjustment amount of S2 is calculated to be 0.3335, and the adjusted service carrying weight is 0.5 + 0.3335 = 0.8335.

[0158] An allocation scheme is constructed based on the adjusted service carrying weights. Assuming the total amount of early warning service data is 100 units, the data is allocated according to the adjusted service carrying weight ratios of S1 and S2: the allocation ratio of S1 is 0.9095 / (0.9095+0.8335)=0.522, and the allocated data volume is 52.2 units; the allocation ratio of S2 is 0.8335 / (0.9095+0.8335)=0.478, and the allocated data volume is 47.8 units.

[0159] The early warning service data is allocated to the sensing nodes according to the allocation scheme. After receiving 52.2 units of data, S1 transmits 31.3 units of data (60%) through path P1 and 20.9 units of data (40%) through path P2, based on the transmission path of the local reconstruction. After receiving 47.8 units of data, S2 transmits 19.1 units of data (40%) through path P1, 14.3 units of data (30%) through path P2, and 14.4 units of data (30%) through path P3.

[0160] By backpropagating the path adjustment gradient to the sensing nodes, the coordinated optimization of local network reconstruction and adaptive node adjustment is achieved. Through the dynamic fusion of gradient response quantities and coupled feature vectors, the trend of network state changes can be accurately predicted, quantifying the impact of topology reconstruction on the service carrying capacity of nodes. Adaptively adjusting the service carrying weights of sensing nodes based on the prediction results enables intelligent allocation and dynamic balancing of service flows, thereby improving the network's self-organizing adjustment capability and global stability under complex service loads.

[0161] A second aspect of the present invention provides an electronic device, comprising:

[0162] processor;

[0163] Memory used to store processor-executable instructions;

[0164] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0165] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0166] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The network architecture and networking method of a fiber optic early warning system, characterized in that, include: Obtain the state information of sensing nodes in a distributed optical fiber sensor network, construct the network state tensor and decompose it to obtain the coupled feature vector; The collaborative metric value of the sensing nodes is calculated based on the coupling feature vector. The sensing nodes are then projected according to the collaborative metric value to obtain the projected coordinates. Based on the projected coordinates, the sensing nodes are clustered to generate a hierarchical network topology. Acquire early warning service data from sensing nodes, calculate the service spatial location based on the projection coordinates of the source node and target node of the early warning service data, calculate the correlation weight between the service spatial location and the service attribute features of the early warning service data, and generate a service feature vector. The demand intensity of early warning service data on the transmission path in the hierarchical network topology is calculated based on the service feature vector. The path adjustment gradient is calculated based on the demand intensity and the carrying status of the transmission path. The hierarchical network topology is then partially reconstructed based on the path adjustment gradient. Obtain the path adjustment gradient of each candidate path during the local reconstruction process; Tracing back along the candidate path to the sensing node, the path adjustment gradients of each candidate path associated with each sensing node are aggregated according to the path topology connection strength to obtain the gradient response of the sensing node. The gradient response is then applied to the sensing node as a backpropagation stimulus to obtain the activated sensing node. Extract the coupling feature vectors of the activated sensing nodes, perform temporal unfolding on the coupling feature vectors to obtain the state change trend, and perform topological correlation analysis on the coupling feature vectors to obtain the topological change trend; The network state impact prediction value is obtained by coupling the state change trend with the topology change trend. The network state impact prediction value is then coupled with the gradient response value to calculate the weight adjustment value. The service carrying weight of the sensing node is then dynamically adjusted based on the weight adjustment value. An allocation scheme is constructed based on the adjusted service carrying weights, and the early warning service data is allocated to the sensing nodes according to the allocation scheme and transmitted through the locally reconstructed transmission path.

2. The method according to claim 1, characterized in that, To obtain the state information of sensing nodes in a distributed fiber optic sensor network, the network state tensor is constructed and decomposed to obtain the coupled feature vector, including: Collect optical signal parameters and service load parameters of each sensing node to generate a node state vector; The fiber optic connection relationship and spatial location information of the sensing nodes are obtained. Adjacent sensing node pairs are determined based on the fiber optic connection relationship. The link hop count between each pair of adjacent sensing nodes is calculated. The physical distance between each pair of adjacent sensing nodes is calculated based on the spatial location information of the nodes. The link hop count and physical distance are weighted and summed to obtain the inter-node coupling degree, and a node coupling matrix with the inter-node coupling degree as the element is constructed. The node state vector is expanded into a state matrix according to the time sequence. The state matrix is ​​then multiplied with the node coupling matrix to generate a network state tensor containing node dimension, time dimension and coupling dimension. The network state tensor is decomposed along the node dimension, time dimension, and coupling dimension, and node features, temporal features, and coupling features are extracted and fused to obtain the coupling feature vector.

3. The method according to claim 1, characterized in that, The collaborative metric value of the sensing nodes is calculated based on the coupling feature vector. The sensing nodes are then projected according to the collaborative metric value to obtain projected coordinates. Based on the projected coordinates, the sensing nodes are clustered to generate a hierarchical network topology, including: Construct a collaborative propagation graph of coupled feature vectors, treating each sensing node as a graph node and the similarity of coupled feature vectors between sensing nodes as graph edge weights; Perform multi-hop cooperative propagation operation on the cooperative propagation graph, and iteratively update the cooperative propagation information received by each sensing node until convergence, and use the converged cooperative propagation information as the cooperative metric value of the current sensing node. A nonlinear projection function is constructed based on the collaborative metric value to map the sensing nodes to the projection space, thereby obtaining the projected coordinates of each sensing node. Calculate the spatial distance between projected coordinates, cluster the sensing nodes according to the spatial distance to obtain clusters, calculate the gradient direction and topological connection strength of the collaborative metric between each cluster and other clusters, and determine the collaborative flow relationship between clusters; Based on the collaborative flow relationship, the clusters are divided into core layer clusters, convergence layer clusters, and access layer clusters to generate a hierarchical network topology.

4. The method according to claim 3, characterized in that, Calculate the spatial distance between projected coordinates, cluster the sensing nodes based on the spatial distance to obtain clusters, calculate the gradient direction and topological connectivity strength of the cooperative metric between each cluster and other clusters, and determine the cooperative flow relationship between clusters, including: Calculate the spatial distance between the projected coordinates, and construct a proximity matrix of the sensing nodes based on the spatial distance; The proximity relation matrix is ​​subjected to spectral decomposition to extract a set of feature vectors. Based on the feature vectors in the set of feature vectors, the sensing nodes are spatially clustered to obtain clusters. The aggregated collaborative metric value of the cluster is calculated based on the collaborative metric value of all sensing nodes in each cluster. Calculate the difference in the aggregation synergy metric between clusters, and determine the gradient direction of the synergy metric between clusters based on the difference; The number of physical fiber optic connections between clusters is counted, and the ratio of the number of physical fiber optic connections to the total number of sensing nodes within a cluster is used as the topological connection strength between clusters. A gradient field is constructed based on the gradient direction of the co-metric values ​​between clusters. In the gradient field, a sequence of clusters with decreasing co-metric values ​​is traced along the gradient direction of the co-metric values ​​to form a co-transmission chain. The topological connection strength between each pair of adjacent clusters in the collaborative transmission chain is extracted, and the minimum topological connection strength is taken as the link carrying strength of the collaborative transmission chain. When the link carrying strength is greater than a preset carrying threshold, it is determined that there is a collaborative flow relationship between adjacent clusters in the collaborative transmission chain.

5. The method according to claim 1, characterized in that, Acquire early warning service data from sensing nodes; calculate the service spatial location based on the projected coordinates of the source node and target node of the early warning service data; calculate the correlation weight between the service spatial location and the service attribute features of the early warning service data; and generate a service feature vector including: Acquire early warning service data from sensing nodes, and extract the source node projection coordinates and target node projection coordinates of the early warning service data; Construct a spatial path between the projected coordinates of the source node and the projected coordinates of the target node, collect historical service traffic data of the sensing nodes on the spatial path and perform time-series decomposition to obtain the evolution characteristics of the service traffic, calculate the degree of aggregation and diffusion of the service traffic based on the evolution characteristics, and determine the dwell position of the service traffic as the service spatial position based on the comparison difference between the degree of aggregation and diffusion. Extract the business attribute features of the early warning business data; Determine the attribute sampling area around the business space location, acquire the business attribute features of historical business data within the attribute sampling area and construct the attribute distribution function, calculate the distribution position of the business attribute features in the attribute distribution function, and determine the association weight between the business space location and the business attribute features of the early warning business data based on the distribution position and the distribution shape of the attribute distribution function. The business feature vector of early warning business data is generated by combining the business spatial location with the associated weight.

6. The method according to claim 1, characterized in that, Calculate the demand intensity of early warning service data on transmission paths in the hierarchical network topology based on the service feature vector, calculate the path adjustment gradient based on the demand intensity and the carrying status of the transmission path, and perform local reconstruction of the hierarchical network topology based on the path adjustment gradient, including: Extract business spatial location and association weights from business feature vectors; In the hierarchical network topology, locate the target level containing the service spatial location, construct a set of candidate paths from the source node to the target node in the target level, obtain the path topology for each candidate path in the candidate path set, calculate the spatial deviation of the service spatial location in the path topology, and attenuate and modulate the association weights based on the spatial deviation to obtain the demand intensity of the candidate path. Obtain the resource occupancy status of each candidate path in the candidate path set, and determine the carrying status of each candidate path based on the resource occupancy status; For each candidate path in the candidate path set, a supply-demand balance relationship between demand intensity and carrying capacity is constructed. Deviation analysis is performed on the supply-demand balance relationship to obtain a supply-demand mismatch vector. The mismatch direction and mismatch intensity are extracted from the supply-demand mismatch vector and coupled to form the path adjustment gradient of the candidate path. Based on the path adjustment gradient of each candidate path, the topology resource configuration of the corresponding path is adjusted to complete the local reconstruction of the hierarchical network topology.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images