A Data Visualization Modeling Method for Complex Systems Based on the Fusion of GIS and Topology Maps

Abstract

The application discloses a complex system data visualization modeling method based on GIS and topological graph fusion, relates to the technical field of data visualization modeling, and comprises data fusion modeling, space-time anomaly reasoning, state propagation modeling and multi-dimensional visual enhancement. Through unified access, cleaning and standardization processing of operation data, topological relationship data, geographic space data, function configuration data and environment data of a complex system, a unified data model combining geographic attributes and topological attributes is constructed; a convolution time sequence modeling method combined with a graph neural network is used to realize abnormal root cause reasoning and space constraint propagation modeling of a system state; a time gating mechanism is used to depict the dynamic evolution law of a node state, and interactive display is realized; the method can realize global visualization of a complex system operation state, dynamic backtracking of an abnormal propagation path and hierarchical analysis of a key node, and improves the intelligent monitoring, situation awareness and decision support capability of the system.

Description

Technical Field

[0001] This invention relates to the field of data visualization modeling technology, specifically to a data visualization modeling method for complex systems based on the fusion of GIS and topology maps. Background Technology

[0002] The data visualization modeling method for complex systems based on the fusion of GIS and topology maps is a comprehensive intelligent analysis and visualization technology for complex systems. Its core idea is to fuse GIS spatial data and network topology data of complex systems (such as communication systems or IoT systems) into a unified model, enabling multi-dimensional representation of the relationships between communication devices, links, service flows, and the external environment. This method allows for the intuitive display of network structure and operational status in a geographic coordinate system, dynamic tracking of fault propagation paths and service flows, and supports cross-level drill-down analysis, spatiotemporal reasoning, and visualized decision-making.

[0003] This technology breaks through the limitations of traditional network monitoring, which separates logical topology from geographical distribution. It enables intelligent management of the entire process, from node performance monitoring and fault root cause location to propagation impact prediction. It is widely applicable to the comprehensive operation monitoring and visualized operation and maintenance of metropolitan area networks, backbone networks, enterprise private networks, and emergency communication systems.

[0004] However, existing methods for visualizing data in complex systems suffer from technical problems such as the separation of geographic and topological information, the lack of visibility of state propagation paths, and the limited dimension of data presentation.

[0005] In existing anomaly reasoning processes for complex systems, there are technical problems such as separate modeling of temporal and topological features, low root cause localization accuracy, and insufficient expression of spatial constraint features.

[0006] Existing state propagation modeling methods for complex systems suffer from technical problems such as static propagation relationships, inability to reflect spatial influence range, and lack of ability to quantify multi-layer propagation paths. Summary of the Invention

[0007] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a data visualization modeling method for complex systems based on the fusion of GIS and topology maps. The technical solution adopted by this invention is as follows: The data visualization modeling method for complex systems based on the fusion of GIS and topology maps provided by this invention includes the following steps:

[0008] Step S1: Data fusion modeling;

[0009] Step S2: Spatiotemporal anomaly reasoning;

[0010] Step S3: State propagation modeling;

[0011] Step S4: Multidimensional visual enhancement.

[0012] Furthermore, in step S1, the data fusion modeling is used to collect and preprocess multi-source heterogeneous data of complex systems. Specifically, it involves unified access, cleaning and standardization of system operation data, relational topology data, geospatial data, functional configuration data and external environment data, and establishing a unified data model that integrates geographical attributes, topological attributes and functional attributes to obtain multi-source fused raw data.

[0013] The multi-source fusion raw data specifically includes performance index data, topology data, geospatial data, functional configuration data, and environmental perception data of complex systems.

[0014] Further, in step S2, the spatiotemporal anomaly inference is used to intelligently locate the root causes of anomalies in the operating state of complex systems. Specifically, based on the multi-source fused original data, an anomaly root cause localization model is improved by using convolutional temporal modeling combined with graph neural networks. The time series features and spatial feature vectors of system nodes are extracted, and spatiotemporal feature representation and causal inference analysis are performed to obtain spatiotemporal analysis data of anomaly root causes. This includes the following steps:

[0015] Step S21: Multimodal feature encoding, used to transform the multi-source fusion raw data into a unified multimodal feature representation that can be computed by the model. Specifically, based on the multi-source fusion raw data, temporal feature extraction, geospatial feature encoding and spatiotemporal feature fusion are performed to obtain a high-dimensional spatiotemporal feature input matrix.

[0016] The time-series feature extraction specifically uses the CPU utilization, module speed, data loss rate, latency, convergence latency, and signal fluctuation of each system node as time performance indicators. The time performance indicators are reconstructed into time-series vectors, and a one-dimensional convolutional feature extraction method is used to extract the features to obtain time-series feature data.

[0017] The generation of topological features specifically involves transforming the location information in the geospatial data into sine and cosine coordinates, and then mapping it through a multilayer perceptron to obtain a spatial feature vector.

[0018] The spatiotemporal feature fusion specifically involves concatenating the temporal feature data and the spatial feature vector to obtain a fused feature vector, which serves as the high-dimensional spatiotemporal feature input matrix.

[0019] Step S22: Graph temporal convolution fusion, specifically, based on the topological structure data, define the adjacency matrix between node data, and combine the latitude and longitude and spatial path information of the nodes in the geospatial data, construct spatial association weights by calculating the geographical distance between nodes, aggregate node neighborhood features through graph convolution operation, and extract the dynamic change features of system nodes using temporal convolution to obtain graph temporal convolution feature data;

[0020] Step S23: Spatiotemporal attention fusion, specifically, based on the graph temporal convolution feature data, calculate the spatial attention weight and temporal importance weight between nodes, and obtain the spatiotemporal attention fusion feature data through spatiotemporal attention weighted fusion;

[0021] Step S24: Multi-channel residual enhancement, specifically, constructing multi-layer adjacency graph convolution channels, extracting multi-channel convolution propagation features, and introducing a residual connection mechanism to perform residual enhancement weighted fusion on the propagation features to obtain multi-scale propagation feature data;

[0022] Step S25: Root cause localization analysis, specifically, based on the multi-scale propagation characteristic data, through probability estimation and confidence interval calculation, outputs the abnormal root cause score and prediction confidence of each node, and obtains abnormal root cause localization analysis data;

[0023] Step S26: Root cause localization model training, specifically, constructing an input layer through the multimodal feature encoding, constructing a convolutional layer through graph temporal convolution fusion, constructing an attention layer through spatiotemporal attention fusion, constructing a residual enhancement layer through multi-channel residual enhancement, and constructing a classification and regression layer through root cause localization analysis, thereby training the anomaly root cause localization model to obtain the anomaly root cause localization model;

[0024] Step S27: Anomaly reasoning, specifically, based on the multi-source fusion raw data, the anomaly root cause localization model is used to perform spatiotemporal feature representation and causal reasoning analysis to obtain anomaly root cause spatiotemporal analysis data;

[0025] The spatiotemporal analysis data of the abnormal root causes includes the identification of root cause nodes and spatiotemporal weight matrices.

[0026] Further, in step S3, the state propagation modeling is used to establish the propagation law of node states in a complex system within a multi-layered structure. Specifically, based on the spatiotemporal analysis data of the anomaly root causes and the multi-source fused raw data, a dynamic graph neural network propagation relationship modeling network combining geographical distance constraints and topological connection strength is used to perform node state updates and path evolution simulations to obtain system state propagation data, including the following steps:

[0027] Step S31: Propagation graph construction, used to establish a propagation graph containing topological structure and spatial location constraints. Specifically, it constructs node adjacency relationships based on topological structure data, calculates the geographical distance between nodes based on the latitude and longitude of nodes in geospatial data, and forms a propagation graph structure with spatial constraint attributes through spherical distance calculation and adjacency matrix mapping, thus obtaining geographic constraint dynamic propagation graph data.

[0028] The geographic constraint dynamic propagation map data is synthesized by normalizing the topological adjacency matrix and the spatial weight matrix, and the calculation formula is as follows:

[0029] ;

[0030] In the formula, It is the normalized spatial weight. It is a normalized topological adjacency matrix, k is the normalized node index, and Sc ij It is the spatial weight between node i and its neighboring node j. It is the spatial weight between node i and normalized node k. It is to prevent the parameter from being divided into zero. It is the topological adjacency matrix between node i and its neighboring node j. It is the identity matrix between node i and its neighboring node j. It is the topological adjacency matrix between node i and normalized node k. It is the identity matrix between node i and normalized node k;

[0031] Step S32: Generate a fusion weight matrix, which is used to integrate the propagation weighting relationship between topological connection strength and spatial distance. Specifically, based on the geographical distance between nodes in the geographic constraint dynamic propagation graph data, the spatial distance attenuation coefficient is calculated and fused with the topological weight to construct a spatial topological fusion propagation weight matrix, thereby obtaining propagation weight fusion data.

[0032] The spatial topology fusion propagation weight matrix is ​​specifically calculated using a weighted linear fusion model, and the calculation formula is as follows:

[0033] ;

[0034] In the formula, W fuse It is about propagating weighted data. It is the topological balance coefficient. It is a topological adjacency matrix. It is a spatial constraint matrix;

[0035] Step S33: Dynamic graph state propagation modeling, used to simulate the propagation and evolution of system node states in the time dimension. Specifically, based on the propagation weight fusion data, a dynamic graph neural network model with time gating structure is used to jointly model the information propagation and time dependency relationship between nodes to obtain node state evolution representation data.

[0036] The joint modeling is specifically calculated by introducing the propagation weight fusion data into the state update mechanism that combines graph convolution and gated recursive units;

[0037] Step S34: Propagation path simulation, used to simulate the propagation path and influence range of system state in complex system. Specifically, based on the node state evolution representation data, multi-path propagation simulation is performed on the abnormal source node, the node state change amount and propagation influence intensity are calculated, and system state propagation data are obtained through propagation path tracking and influence domain identification.

[0038] The propagation path tracking and influence domain identification specifically employs a joint tracking algorithm based on weighted shortest path and depth-first search to backtrack the propagation path of each abnormal source node, obtaining a path set. Then, by statistically analyzing the set of nodes in all path sets whose cumulative propagation intensity is greater than a set threshold, an influence domain node set is obtained. Finally, by labeling and mapping the nodes in the influence domain node set and their corresponding associated edges, system state propagation data is obtained.

[0039] The system state propagation data specifically includes node state parameters, propagation path weights, and spatial location information;

[0040] The node state parameters specifically include node number, state strength, state change amount, state time sequence vector, and time label; the propagation path weight specifically includes path identifier, path length, one-sided propagation weight, path cumulative influence degree, and path time window; the spatial location information specifically includes node latitude and longitude coordinates, path spatial length, influence domain node set, and node spatial proximity matrix.

[0041] Furthermore, in step S4, the multi-dimensional visual enhancement is used to realize the intelligent interactive visual display of complex system states. Specifically, based on the system state propagation data, hierarchical rendering is used to dynamically visualize the state evolution process, performance status and data flow to obtain a system visualization interface.

[0042] In a preferred embodiment, the multidimensional visual enhancement, based on the WebGL and D3.js visualization engine, realizes intelligent interactive visual display of the communication network status. Specifically, it performs visualization mapping based on the system status propagation data, mapping spatial coordinates to node positions, mapping state changes to color or brightness gradients, and mapping propagation intensity to line thickness or transparency. Through time axis control and interactive operation interface, it realizes dynamic playback, local zoom-in, and conditional filtering of the system status, thereby displaying the operating status, propagation path, and spatial distribution characteristics of complex systems in a unified three-dimensional view.

[0043] The beneficial effects achieved by the present invention using the above solution are as follows:

[0044] (1) In view of the technical problems in existing complex system data visualization methods, such as the separation of geographic information and topological information, the invisibility of state propagation paths, and the single dimension of data presentation, this solution creatively adopts a comprehensive visualization modeling path that integrates spatiotemporal anomaly reasoning, state propagation modeling, and multi-dimensional visualization enhancement. By introducing the fusion modeling of GIS spatial data and topological structure data, the system realizes the three-dimensional dynamic display of complex system states and multi-layer semantic linkage expression. The system can not only intuitively locate abnormal nodes in geographic coordinate space, but also dynamically track propagation paths at the topological level. Combined with hierarchical rendering and time axis control mechanism, the system can realize interactive visualization of cross-layer drilling, propagation backtracking, and state flow at the user end, which significantly improves the situational awareness and decision support capabilities of complex systems.

[0045] (2) In response to the technical problems of separate modeling of temporal and topological features, low root cause localization accuracy, and insufficient expression of spatial constraint features in the existing anomaly reasoning process for complex systems, this solution creatively adopts convolutional temporal modeling combined with graph neural networks to improve the reasoning method. By extracting temporal convolutional features of node running status and generating spatial features by combining geospatial sine and cosine coordinate mapping, the joint propagation of topological relationships and spatial constraints is realized in the graph temporal convolutional fusion layer. At the same time, a spatiotemporal attention mechanism is introduced to dynamically adjust the time weight and spatial weight, so that the model can identify potential causal chains, propagation trigger nodes and implicit coupling relationships in complex systems, significantly improving the accuracy of anomaly localization and spatiotemporal interpretability.

[0046] (3) To address the technical problems of existing state propagation modeling methods for complex systems, such as static propagation relationships, inability to reflect spatial influence range, and lack of multi-layer propagation path quantification capabilities, this solution creatively adopts a dynamic graph neural propagation relationship modeling network that combines geographical distance constraints and topological connection strength. A spatially constrained propagation graph is established through spherical distance calculation and adjacency matrix mapping, and topological connection strength and spatial attenuation coefficient are integrated to form a dynamic spatial-topological weighted propagation matrix. A time-gated structure is used to characterize the propagation rate change and time delay accumulation effect, thereby generating traceable multi-path propagation trajectories and dynamic influence domain prediction results. This model can continuously simulate node state evolution in the time dimension, significantly improving the accuracy, stability, and interpretability of anomaly propagation modeling in complex systems. Attached Figure Description

[0047] Figure 1 A flowchart illustrating the data visualization modeling method for complex systems based on the fusion of GIS and topology maps provided by this invention;

[0048] Figure 2 A flowchart illustrating the data fusion modeling process for step S1;

[0049] Figure 3 This is a flowchart illustrating the spatiotemporal anomaly reasoning in step S2.

[0050] Figure 4 A flowchart illustrating the state propagation model for step S3.

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0054] Example 1, see Figure 1 The present invention provides a data visualization modeling method for complex systems based on the fusion of GIS and topology maps. The method includes the following steps:

[0055] Step S1: Data fusion modeling;

[0056] Step S2: Spatiotemporal anomaly reasoning;

[0057] Step S3: State propagation modeling;

[0058] Step S4: Multidimensional visual enhancement.

[0059] By performing the above operations, this solution addresses the technical problems of existing complex system data visualization methods, such as the separation of geographic and topological information, the lack of visibility of state propagation paths, and the single dimension of data presentation. It creatively adopts a comprehensive visualization modeling path that integrates spatiotemporal anomaly reasoning, state propagation modeling, and multi-dimensional visualization enhancement. By introducing the fusion modeling of GIS spatial data and topological structure data, it achieves three-dimensional dynamic display and multi-layer semantic linkage expression of complex system states. The system can not only intuitively locate abnormal nodes in geographic coordinate space but also dynamically track propagation paths at the topological level. Combined with hierarchical rendering and timeline control mechanisms, it enables interactive visualization of cross-layer drill-down, propagation backtracking, and state flow on the user end, significantly improving the situational awareness and decision support capabilities of complex systems.

[0060] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the data fusion modeling is used to collect and preprocess multi-source heterogeneous data of complex systems. Specifically, it involves unified access, cleaning and standardization of system operation data, relational topology data, geospatial data, functional configuration data and external environment data to establish a unified data model that integrates geographical attributes, topological attributes and functional attributes, and obtain multi-source fused raw data.

[0061] The system operation data includes CPU utilization, module speed, loss rate, latency, convergence latency, and signal fluctuation.

[0062] The relational topology data includes system nodes, interaction edges, hierarchical structure, and carrying relationships;

[0063] The geospatial data includes node latitude and longitude, spatial path, and region polygon;

[0064] The functional configuration data includes parameter strategies, resource allocation strategies, and data flow directions;

[0065] The external environmental data includes meteorological information and geographical risk data;

[0066] Preferably, the specific steps of the unified access, cleaning, and standardization processing include:

[0067] Step S11: Performance index data generation, used to establish a performance index dataset for the complex system operation status. Specifically, the system operation data is identified and its features are extracted, data is associated based on global node identifiers and interface identifiers, and the operation logs and monitoring data of each module are parsed to extract performance-related index fields and obtain performance index data.

[0068] The performance metrics data specifically refers to a dataset indexed by timestamps after format standardization and outlier removal, including node operation metrics, interface operation metrics, and system quality metrics.

[0069] The node performance metrics include CPU utilization, memory usage, and cache hit rate.

[0070] The interface performance metrics include bandwidth utilization, input / output loss rate, interface speed, and average round-trip time.

[0071] The system quality indicators include signal fluctuation, convergence delay, and congestion trigger rate.

[0072] Step S12: Topology data generation, used to construct topology data reflecting the connection relationships and hierarchical structure of complex systems. Specifically, based on the node identification information in the performance index data, combined with the system configuration table and interaction relationship table, the physical connection relationship and logical carrying relationship between nodes are parsed to establish the node and edge relationship structure and obtain the topology data.

[0073] The topology data is used to describe the connection topology between system nodes, including node data and interaction data;

[0074] The node data includes node identifier, interface number, node type, region, and running status;

[0075] The interactive data includes source node identifier, target node identifier, relationship type, capacity parameter, and runtime latency;

[0076] Step S13: Geospatial data generation, used to establish a geospatial index structure for nodes and paths in complex systems. Specifically, the collected node latitude and longitude, location points and spatial path coordinates are standardized, and a hierarchical geographic grid is constructed using the H3 spatial indexing method to form a spatial mapping between nodes and paths. At the same time, the Haversine distance formula is used to calculate the geographic distance and spatial proximity relationship between nodes to form a geographic adjacency matrix that can be used for spatiotemporal analysis, thus obtaining geospatial data.

[0077] The geospatial data includes node spatial information, path spatial information, and regional coverage information;

[0078] The node spatial information includes the node's latitude and longitude, its station number, and its region code;

[0079] The path spatial information includes the path node coordinate sequence, path length, route area, and path direction angle;

[0080] The regional coverage information includes the system service coverage polygon, regional boundaries, and service overlap rate.

[0081] Step S14: Functional configuration data generation, used to construct a configuration dataset reflecting the operation and policy control information of complex systems. Specifically, it is based on the policy configuration file exported from the system management and scheduling platform, and is associated and mapped according to the global node identifier and interface identifier. Logical parameters and running configuration items are extracted to obtain functional configuration data.

[0082] The functional configuration data includes parameter configuration, strategy constraints, and data flow information;

[0083] Step S15: Environmental perception data generation, used to establish external environmental perception data corresponding to the operating area of ​​the complex system. Specifically, based on the geospatial data, meteorological and geographical risk information matching the system area is obtained through an external data interface, and mapping and matching are performed based on the spatial index of nodes and paths in the geospatial data.

[0084] The multi-source fusion raw data specifically includes performance index data, topology data, geospatial data, functional configuration data, and environmental perception data of complex systems;

[0085] In a preferred embodiment of the present invention, the complex system can be specifically applied to a communication network system. In this embodiment, the system node corresponds to the device node in the communication network, the interaction relationship corresponds to the communication link or logical bearer relationship, the functional configuration data corresponds to network configuration parameters and service strategy data, and the geospatial data corresponds to the geographical location and line spatial path information of the communication facility. By incorporating the operation data, topology data, geospatial data, and strategy configuration data of the communication network into a unified multi-source data fusion model, structured and visual modeling and analysis of the operating status of the communication network system can be achieved, providing data support for subsequent spatiotemporal anomaly reasoning and state propagation modeling.

[0086] This embodiment illustrates a typical application of the method of the present invention in a communication network system scenario, but the present invention is not limited thereto and can also be applied to other complex systems such as energy dispatching, traffic operation, industrial Internet of Things and ecological monitoring.

[0087] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the spatiotemporal anomaly inference is used to perform intelligent anomaly root cause localization on the operating state of complex systems. Specifically, based on the multi-source fused original data, an anomaly root cause localization model is improved by using convolutional temporal modeling combined with graph neural networks. The time series features and spatial feature vectors of system nodes are extracted, and spatiotemporal feature representation and causal inference analysis are performed to obtain spatiotemporal analysis data of anomaly root causes. The steps include:

[0088] Step S21: Multimodal feature encoding, used to transform the multi-source fusion raw data into a unified multimodal feature representation that can be computed by the model. Specifically, based on the multi-source fusion raw data, temporal feature extraction, geospatial feature encoding and spatiotemporal feature fusion are performed to obtain a high-dimensional spatiotemporal feature input matrix.

[0089] The aforementioned time-series feature extraction specifically uses the CPU utilization, module speed, data loss rate, latency, convergence latency, and signal fluctuation of each system node as time performance indicators. These time performance indicators are then reconstructed into time-series vectors, and a one-dimensional convolutional feature extraction method is used for feature extraction to obtain time-series feature data. The calculation formula is as follows:

[0090] ;

[0091] In the formula, This is temporal feature data, where t is the time index, i is the device node index, ReLU is a non-linear activation function, and Conv1D is a one-dimensional convolution operation. It is the time-series vector data after the performance metrics are reconstructed, where k is the index of the time window length;

[0092] The generation of the topological structure features specifically involves transforming the location information in the geospatial data using sine and cosine coordinates, followed by mapping via a multilayer perceptron to obtain a spatial feature vector. The calculation formula is as follows:

[0093] ;

[0094] In the formula, s i It is a spatial feature vector, MLP is a multilayer perceptron mapping function, sin is the sine transform of latitude and longitude values, cos is the cosine transform of latitude and longitude values, and lat is the spatial feature vector. i It is the dimensional information of node i, lon i It is the longitude information of node i;

[0095] The spatiotemporal feature fusion specifically involves concatenating the temporal feature data and the spatial feature vector to obtain a fused feature vector, which serves as the high-dimensional spatiotemporal feature input matrix.

[0096] Step S22: Graph temporal convolution fusion, specifically, based on the topological structure data, define the adjacency matrix between node data, and combine the latitude and longitude and spatial path information of the nodes in the geospatial data, construct spatial association weights by calculating the geographical distance between nodes, aggregate node neighborhood features through graph convolution operation, and extract the dynamic change features of system nodes using temporal convolution to obtain graph temporal convolution feature data;

[0097] Step S23: Spatiotemporal attention fusion, specifically, based on the graph temporal convolution feature data, calculate the spatial attention weight and temporal importance weight between nodes, and obtain the spatiotemporal attention fusion feature data through spatiotemporal attention weighted fusion;

[0098] The spatiotemporal attention-weighted fusion specifically employs attention-based weighted calculation methods to model the spatial correlation and temporal importance of nodes. Spatial attention calculates feature similarity between nodes based on adjacency relationships and performs normalization processing, while temporal attention determines time step weights based on feature correlation within a time window. Finally, the features of each node are dynamically fused according to the joint weighting result of spatial and temporal weights, calculated using the following formula:

[0099] ;

[0100] In the formula, Let j be the spatiotemporal attention fusion feature data of node i at time t, and j be the neighbor node index. It is the set of neighboring nodes. These are spatial attention weights, Ws is the spatial linear transformation matrix, and H... j,t This represents the spatial attention feature of node j at time t, where t is the temporal index. This is the attention time index, where w is the time window length. It is the time attention weight, W t It is a time linear transformation matrix. Is node i at time... Temporal attention characteristics;

[0101] Step S24: Multi-channel residual enhancement, specifically, constructing multi-layer adjacency graph convolution channels, extracting multi-channel convolution propagation features, and introducing a residual connection mechanism to perform residual enhancement weighted fusion on the propagation features to obtain multi-scale propagation feature data;

[0102] The calculation formula for the multi-channel convolutional propagation feature extraction is as follows:

[0103] ;

[0104] In the formula, It is the convolutional feature output of the m-th adjacency graph convolutional channel, and ReLU is a non-linear activation function. This is the correspondence degree matrix, calculated by summing the connection weights of each node in the m-th order adjacency matrix to form a diagonal matrix, which is used for normalization during feature aggregation. It is the m-th order adjacency matrix, used to characterize the propagation connection strength at different topological scales. It is spatiotemporal attention fusion feature data, W (m) It is the weight matrix of the m-th convolutional channel of the adjacency graph, where m is the convolutional channel index of the adjacency graph;

[0105] The calculation formula for the residual enhancement weighted fusion is as follows:

[0106] ;

[0107] In the formula, It is multi-scale propagation characteristic data. This is spatiotemporal attention fusion feature data, where M is the total number of convolutional channels in the adjacency graph, with a default value of 3. It is the channel fusion weight. It is the convolutional feature output of the m-th adjacency graph convolutional channel;

[0108] Step S25: Root cause localization analysis, specifically, based on the multi-scale propagation characteristic data, through probability estimation and confidence interval calculation, outputs the abnormal root cause score and prediction confidence of each node, and obtains abnormal root cause localization analysis data;

[0109] Preferably, the confidence interval is calculated using the Bootstrap method. The confidence interval is used for uncertainty assessment during model validation and online inference, and does not participate in parameter updates during the training process.

[0110] Step S26: Root cause localization model training, specifically, constructing an input layer through the multimodal feature encoding, constructing a convolutional layer through graph temporal convolution fusion, constructing an attention layer through spatiotemporal attention fusion, constructing a residual enhancement layer through multi-channel residual enhancement, and constructing a classification and regression layer through root cause localization analysis, thereby training the anomaly root cause localization model to obtain the anomaly root cause localization model;

[0111] Preferably, as shown in the table, key training parameters are illustrated, wherein: batch size and learning rate are used to control the stability and convergence speed of gradient updates; time window length and number of channels determine the feature capture range of spatiotemporal attention and multi-channel convolution; spatial decay scale, topological space balance coefficient, and temporal decay coefficient are used to adjust the balance between spatial distance decay, topological connection strength, and temporal delay effects; the number of attention heads and hidden layer dimensions affect the model's expressive power and computational complexity; regularization coefficient and sparsity constraint coefficient are used to prevent model overfitting; optimization algorithms generally use Adam and AdamW to balance stability and adaptability; the number of training epochs determines the extent of model iteration; the loss function adopts a joint classification and regression form to simultaneously optimize the accuracy of anomaly node identification and root cause scoring.

[0112]

[0113] As a further optimization of this embodiment, the training samples for training the root cause localization model are derived from the historical operation monitoring data and event work order records of the complex system, specifically including node operation logs, equipment alarm records, service interruption events and maintenance handling results; a labeled sample set is constructed by slicing and labeling the time series performance indicators, topological relationships and spatial locations of nodes in the multi-source fusion original data.

[0114] Positive samples correspond to abnormal root cause nodes and their occurrence time windows that have been confirmed manually or by the system, while negative samples correspond to other nodes that are running normally within the same time period. The sample labels include node number, time index, abnormal category label (0 / 1), and root cause score label ([0,1]).

[0115] The training data is divided according to the time dimension, with a preferred ratio of 6:2:2 for the training set, validation set, and test set. The validation set is used to adjust hyperparameters, and the test set is used to evaluate model performance. During model training, batch gradient descent is used to optimize parameters based on the joint loss function of classification and regression. The classification branch is used to identify root cause nodes, and the regression branch is used to predict root cause scores. The two are jointly optimized to improve the accuracy of root cause identification and the precision of scoring.

[0116] Step S27: Anomaly reasoning, specifically, based on the multi-source fusion raw data, the anomaly root cause localization model is used to perform spatiotemporal feature representation and causal reasoning analysis to obtain anomaly root cause spatiotemporal analysis data;

[0117] The spatiotemporal analysis data of the abnormal root causes includes the identification of root cause nodes and spatiotemporal weight matrices.

[0118] By performing the above operations, this solution addresses the technical problems in existing anomaly reasoning processes for complex systems, such as separate modeling of temporal and topological features, low root cause localization accuracy, and insufficient expression of spatial constraint features. It creatively adopts a convolutional temporal modeling approach combined with graph neural networks to improve the reasoning method. This involves extracting temporal convolutional features from node operating states and generating spatial features by combining geospatial sine and cosine coordinate mapping. The joint propagation of topological relationships and spatial constraints is achieved in the graph temporal convolutional fusion layer. Simultaneously, a spatiotemporal attention mechanism is introduced to dynamically adjust temporal and spatial weights, enabling the model to identify potential causal chains, propagation trigger nodes, and implicit coupling relationships in complex systems, significantly improving the accuracy and spatiotemporal interpretability of anomaly localization.

[0119] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, the state propagation modeling is used to establish the propagation law of node states in a complex system in a multi-layered structure. Specifically, based on the spatiotemporal analysis data of the anomaly root causes and the multi-source fused original data, a dynamic graph neural network propagation relationship modeling network combining geographical distance constraints and topological connection strength is adopted to perform node state update and path evolution simulation to obtain system state propagation data, including the following steps:

[0120] Step S31: Propagation graph construction, used to establish a propagation graph containing topological structure and spatial location constraints. Specifically, it constructs node adjacency relationships based on topological structure data, calculates the geographical distance between nodes based on the latitude and longitude of nodes in geospatial data, and forms a propagation graph structure with spatial constraint attributes through spherical distance calculation and adjacency matrix mapping, thus obtaining geographic constraint dynamic propagation graph data.

[0121] Preferably, the spatial location of the node is hierarchically encoded based on the H3 spatial index to ensure consistent mapping of the spatial proximity relationships of nodes at different levels in the propagation graph;

[0122] The geographical distance between the nodes is specifically calculated using the Haversine formula, which measures the shortest arc distance between two points on the Earth's surface. The formula is as follows:

[0123] ;

[0124] In the formula, d ij This represents the geographical distance between nodes, where i is the node index, j is the neighbor node index, and R is the Earth's radius, specifically 6371 kilometers. It is the difference in latitude between nodes. It is the latitude in radians of node i. It is the latitude in radians of the neighboring node j. It is the difference in longitude between nodes;

[0125] The spherical distance calculation specifically utilizes a spatial attenuation function to achieve a constrained mapping between geographical distance and propagation intensity. The calculation formula is as follows:

[0126] ;

[0127] In the formula, Sc ij It is the spatial weight between node i and its neighboring node j. It is the spatial attenuation scale factor. It is a spherical distance indicator function. It takes the value 1 when the distance between nodes is less than the radius of influence, otherwise it takes the value 0. R0 is the effective radius of influence, and the default value range is [5 km, 30 km].

[0128] The geographic constraint dynamic propagation map data is synthesized by normalizing the topological adjacency matrix and the spatial weight matrix, and the calculation formula is as follows:

[0129] ;

[0130] In the formula, It is the normalized spatial weight. It is a normalized topological adjacency matrix, k is the normalized node index, and Sc ij It is the spatial weight between node i and its neighboring node j. It is the spatial weight between node i and normalized node k. It is to prevent the parameter from being divided into zero. It is the topological adjacency matrix between node i and its neighboring node j. It is the identity matrix between node i and its neighboring node j. It is the topological adjacency matrix between node i and normalized node k. It is the identity matrix between node i and normalized node k;

[0131] Step S32: Generate a fusion weight matrix, which is used to integrate the propagation weighting relationship between topological connection strength and spatial distance. Specifically, based on the geographical distance between nodes in the geographic constraint dynamic propagation graph data, the spatial distance attenuation coefficient is calculated and fused with the topological weight to construct a spatial topological fusion propagation weight matrix, thereby obtaining propagation weight fusion data.

[0132] The spatial topology fusion propagation weight matrix is ​​specifically calculated using a weighted linear fusion model, and the calculation formula is as follows:

[0133] ;

[0134] In the formula, W fuse It is about propagating weighted data. This is the topological space balance coefficient, with a default value range of [0.4, 0.7]. It is a topological adjacency matrix. It is a spatial constraint matrix;

[0135] Step S33: Dynamic graph state propagation modeling, used to simulate the propagation and evolution of system node states in the time dimension. Specifically, based on the propagation weight fusion data, a dynamic graph neural network model with time gating structure is used to jointly model the information propagation and time dependency relationship between nodes to obtain node state evolution representation data.

[0136] The joint modeling specifically involves incorporating the propagation weight fusion data into the state update mechanism that combines graph convolution and gated recursive units. The calculation formula is as follows:

[0137] ;

[0138] In the formula, Z t It updates the gate output; sig is the sigmoid function; W fuse It is propagation weight fusion data, that is, the spatial topological propagation weight matrix obtained by fusion and normalization, which models the spatial diffusion of input features, thereby constraining the range and influence intensity of information transmission between nodes during the gating update process. This ensures that state updates are significantly amplified only on topologically reliable and spatially proximate connections, thus obtaining node state evolution representation data, X. t It is the input feature matrix, specifically referring to the node features, topological features, and spatial features in the spatiotemporal analysis data of anomaly root causes and the original data from multi-source fusion. U z It updates the gate weights, H. t-1 It is the hidden state from the previous moment, V z It updates the hidden state weight of the door, b z It updates the gate bias term, R t It resets the gate output, U r It resets the gate weights, V r It resets the weight of the hidden state of the door, b r It resets the door offset item. It is the output of the candidate hidden state, U h These are the candidate hidden state weights. It is the Hadamard product operator. It is the candidate hidden state fusion weight, b h It is the candidate hidden state bias term, H t It is the hidden state at the current moment;

[0139] Step S34: Propagation path simulation, used to simulate the propagation path and influence range of system state in complex system. Specifically, based on the node state evolution representation data, multi-path propagation simulation is performed on the abnormal source node, the node state change amount and propagation influence intensity are calculated, and system state propagation data are obtained through propagation path tracking and influence domain identification.

[0140] The change in node state is specifically calculated through the state difference between adjacent time steps, and the calculation formula is as follows:

[0141] ;

[0142] In the formula, S is the instantaneous change in the node state. t It is the node state evolution representation data at the current moment, S t-1 It represents the node state evolution data from the previous moment;

[0143] The intensity of the propagation impact is specifically calculated using a path-weighted cumulative model, and the calculation formula is as follows:

[0144] ;

[0145] In the formula, It represents the intensity of the propagation effect, and L represents the path length. It is the path propagation node index. It is the propagation weight between adjacent nodes on the path. This is the propagation time decay coefficient, with a default value of 0.1;

[0146] The propagation path tracking and influence domain identification specifically employs a joint tracking algorithm based on weighted shortest path and depth-first search to backtrack the propagation path of each abnormal source node, obtaining a path set. Then, by statistically analyzing the set of nodes in all path sets whose cumulative propagation intensity is greater than a set threshold, an influence domain node set is obtained. Finally, by labeling and mapping the nodes in the influence domain node set and their corresponding associated edges, system state propagation data is obtained.

[0147] The system state propagation data specifically includes node state parameters, propagation path weights, and spatial location information;

[0148] The node state parameters specifically include node number, state strength, state change amount, state time sequence vector, and time label; the propagation path weight specifically includes path identifier, path length, one-sided propagation weight, path cumulative influence degree, and path time window; the spatial location information specifically includes node latitude and longitude coordinates, path spatial length, influence domain node set, and node spatial proximity matrix.

[0149] By performing the above operations, this solution addresses the technical problems of existing state propagation modeling methods for complex systems, such as static propagation relationships, inability to represent spatial influence range, and lack of multi-layer propagation path quantification capabilities. It creatively employs a dynamic graph neural network for propagation relationship modeling, combining geographical distance constraints and topological connectivity strength. A spatially constrained propagation graph is established through spherical distance calculation and adjacency matrix mapping, and topological connectivity strength and spatial attenuation coefficients are integrated to form a dynamic spatial-topological weighted propagation matrix. A time-gated structure is used to characterize propagation rate changes and time delay accumulation effects, thereby generating traceable multi-path propagation trajectories and dynamic influence domain prediction results. This model can continuously simulate node state evolution in the time dimension, significantly improving the accuracy, stability, and interpretability of anomaly propagation modeling in complex systems.

[0150] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the multi-dimensional visual enhancement is used to realize the intelligent interactive visual display of complex system states. Specifically, based on the system state propagation data, hierarchical rendering is used to dynamically visualize the state evolution process, performance status and data flow to obtain the system visualization interface.

[0151] In a preferred embodiment, the multidimensional visual enhancement, based on the WebGL and D3.js visualization engine, realizes intelligent interactive visual display of the communication network status. Specifically, it performs visualization mapping based on the system status propagation data, mapping spatial coordinates to node positions, mapping state changes to color or brightness gradients, and mapping propagation intensity to line thickness or transparency. Through time axis control and interactive operation interface, it realizes dynamic playback, local zoom-in, and conditional filtering of the system status, thereby displaying the operating status, propagation path, and spatial distribution characteristics of complex systems in a unified three-dimensional view.

[0152] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0153] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0154] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A complex system data visualization modeling method based on GIS and topological graph fusion, characterized in that: The method comprises the following steps: Step S1: data fusion modeling, establishing a unified data model of fused geographic attributes, topological attributes and functional attributes, and obtaining multi-source fused original data; Step S2: spatio-temporal anomaly reasoning, according to the multi-source fused original data, using a convolutional time series modeling combined with a graph neural network to improve an anomaly root cause positioning model, extracting time series features and spatial feature vectors of system nodes, and performing spatio-temporal feature representation and causal reasoning analysis to obtain anomaly root cause spatio-temporal analysis data, including the following steps: multi-modal feature encoding, graph time series convolution fusion, spatio-temporal attention fusion, multi-channel residual enhancement, root cause positioning analysis, root cause positioning model training and anomaly reasoning; The root cause positioning analysis is specifically based on multi-scale propagation feature data, and outputs anomaly root cause scores and prediction confidence of each node through probability estimation and confidence interval calculation to obtain anomaly root cause positioning analysis data; The root cause positioning model training is specifically constructing an input layer through the multi-modal feature encoding, constructing a convolution layer through the graph time series convolution fusion, constructing an attention layer through the spatio-temporal attention fusion, constructing a residual enhancement layer through the multi-channel residual enhancement, and constructing a classification regression layer through the root cause positioning analysis, and performing anomaly root cause positioning model training to obtain an anomaly root cause positioning model; The anomaly reasoning is specifically using the anomaly root cause positioning model to perform spatio-temporal feature representation and causal reasoning analysis according to the multi-source fused original data to obtain anomaly root cause spatio-temporal analysis data; Step S3: state propagation modeling, according to the anomaly root cause spatio-temporal analysis data and the multi-source fused original data, using a dynamic graph neural propagation relationship modeling network combined with geographic distance constraints and topological connection strength to perform node state updating and path evolution simulation to obtain system state propagation data, including the following steps: propagation graph construction, fused weight matrix generation, dynamic graph state propagation modeling and propagation path simulation; The propagation graph construction is used to establish a propagation graph containing topological structure and spatial position constraints, and specifically, node adjacency relationships are constructed according to topological structure data, and geographic distances between nodes are calculated based on node longitude and latitude in geographic space data, and a propagation graph structure with spatial constraint attributes is formed through spherical distance calculation and adjacency matrix mapping to obtain geographic constraint dynamic propagation graph data; The fused weight matrix generation is used to comprehensively consider the propagation weighting relationship of topological connection strength and spatial distance, and specifically, a spatial topological fusion propagation weight matrix is constructed through spatial distance decay coefficient calculation and topological weight fusion according to the geographic distances between nodes in the geographic constraint dynamic propagation graph data to obtain propagation weight fusion data; The dynamic graph state propagation modeling is used to simulate the propagation and evolution rules of system node states in the time dimension, and specifically, a dynamic graph neural network model with a time gating structure is used to jointly model the information propagation and time dependence relationship between nodes according to the propagation weight fusion data to obtain node state evolution representation data; The propagation path simulation is used for simulating a propagation path and an influence range of a system state in a complex system, and specifically, a multi-path propagation simulation is performed on an abnormal source node according to node state evolution representation data, a node state change amount and a propagation influence intensity are calculated, and system state propagation data are obtained through propagation path tracking and influence domain identification. In step S4, the multi-dimensional visual enhancement is used for intelligently and interactively displaying a complex system state, and specifically, a state evolution process, a performance situation and a data flow direction are dynamically visually presented through hierarchical rendering according to the system state propagation data, and a system visual interface is obtained. 2.The GIS and topological graph fusion-based complex system data visualization modeling method according to claim 1, characterized in that: In step S1, the multi-source fusion raw data specifically include performance index data, topology structure data, geographic space data, function configuration data and environment perception data of the complex system. 3.The GIS and topological graph fusion-based complex system data visualization modeling method according to claim 2, characterized in that: In step S2, the multi-modal feature encoding is used for converting the multi-source fusion raw data into unified multi-modal feature representation available for model calculation, and specifically, time sequence feature extraction, geographic space feature encoding and space-time feature fusion are performed according to the multi-source fusion raw data, and a high-dimensional space-time feature input matrix is obtained. The graph time sequence convolution fusion is specifically that an adjacency matrix between node data is defined according to the topology structure data, a spatial correlation weight is constructed by calculating a geographic distance between nodes in combination with node longitude and latitude and spatial path information in the geographic space data, node neighborhood feature aggregation is performed through graph convolution operation, dynamic change features of system nodes are extracted by using time sequence convolution, and graph time sequence convolution feature data are obtained. The space-time attention fusion is specifically that spatial attention weights and time importance weights between nodes are calculated according to the graph time sequence convolution feature data, and space-time attention fusion feature data are obtained through space-time attention weighted fusion. The multi-channel residual enhancement is specifically that a multi-layer adjacency graph convolution channel is constructed, multi-channel convolution propagation feature extraction is performed, and residual enhancement weighted fusion is performed on propagation features by introducing a residual connection mechanism, and multi-scale propagation feature data are obtained.

4. The GIS and topological graph fusion-based complex system data visualization modeling method according to claim 3, characterized in that: The abnormal root cause space-time analysis data include an identified root cause node and a space-time weight matrix.

5. The GIS and topological graph fusion-based complex system data visualization modeling method according to claim 4, characterized in that: In step S3, the geographic constraint dynamic propagation graph data are specifically obtained by synthesizing a normalized topology adjacency matrix and a normalized space weight matrix, and a calculation formula is as follows: ； wherein is a normalized spatial weight, is a normalized topological adjacency matrix, k is a normalized node index, Sc ij is a spatial weight between node i and neighboring node j, is a spatial weight between node i and normalized node k, is a zero prevention parameter, is a topological adjacency matrix between node i and neighboring node j, is an identity matrix between node i and neighboring node j, is a topological adjacency matrix between node i and normalized node k, is an identity matrix between node i and normalized node k; The space topology fusion propagation weight matrix is specifically calculated by using a weighted linear fusion model, and a calculation formula is as follows: ； In the formula, W fuse is the propagation weight fusion data, is the topological space balance coefficient, is the topological adjacency matrix, is the space constraint matrix; The joint modeling is specifically performed by introducing the propagation weight fusion data into a state updating mechanism combining a graph convolution and a gated recurrent unit. 6.The GIS and topological graph fusion-based complex system data visualization modeling method according to claim 5, characterized in that: In step S4, the multi-dimensional visual enhancement is used for intelligently and interactively displaying a complex system state, and specifically, a state evolution process, a performance situation and a data flow direction are dynamically visually presented through hierarchical rendering according to the system state propagation data, and a system visual interface is obtained.

Images