An AI prediction-based urban communication node fault early warning system

Abstract

The application discloses a kind of city communication node failure early warning systems based on AI prediction, including following module: data acquisition and pre-processing module, for collecting and pre-processing city communication node original data;Feature extraction module, for through ChebNet graph convolutional neural network, using Chebyshev polynomial carries out feature extraction;City communication graph construction module, for calculating the Euclidean distance between nodes to build city communication graph;Information gain calculation module, for using improved ID3 algorithm to calculate the information gain of each node feature, generates fault classification decision tree;Fault classification graph construction module, for obtaining fault classification graph according to split rule;Fault early warning information generation module, for generating fault early warning information according to fault classification graph.The application is based on ChebNet and improved ID3 algorithm, realizes the efficient and accurate city communication node failure early warning system.

Description

Technical Field

[0001] This invention relates to the field of fault analysis technology, and in particular to an AI-based predictive early warning system for urban communication node faults. Background Technology

[0002] With the rapid development of urban communication networks, communication nodes, as crucial components, are vital to the overall system operation due to their reliability and stability. However, many current urban communication networks often face the problem of untimely early warning of communication node failures when dealing with complex environmental factors and massive data processing. Existing technologies largely rely on traditional rule-based monitoring methods, often monitoring node status through simple threshold settings or experience-based rules. This approach is slow to react to node anomalies and struggles to adapt to complex network environments and dynamically changing communication data. Therefore, existing technologies have significant limitations in fault early warning, failing to predict potential failures of individual nodes in urban communication networks in real time and accurately, leading to communication interruptions or performance degradation, and impacting the reliability and stability of urban communication networks.

[0003] Currently, fault prediction methods based on machine learning and artificial intelligence technologies are gradually becoming effective means to solve this problem. Through big data analysis and deep learning algorithms, effective features can be extracted from massive amounts of communication data to identify potential fault signs. However, most existing machine learning-based methods have certain limitations. For example, many existing methods rely on traditional deep neural networks or support vector machines (SVMs). While these can achieve fault prediction, they often face problems such as excessive computational complexity and long training times when dealing with high-dimensional, complex data, and the generalization ability and accuracy of the models cannot be fully guaranteed. Furthermore, traditional methods often ignore the relationships between nodes and the influence of network topology in communication networks, making it difficult to comprehensively and accurately reflect the spatiotemporal correlations between nodes. Graph convolutional networks (GCNNs) can effectively extract node features from the network topology and capture the spatiotemporal dependencies between nodes, providing a new approach for accurately predicting node faults. However, existing fault prediction methods based on GCNNs have not yet fully solved the problem of how to perform real-time, intelligent fault prediction based on the dynamic state of nodes and various heterogeneous data (such as environmental data and node operating status data).

[0004] Therefore, how to provide an AI-based predictive early warning system for urban communication node failures is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an AI-based predictive urban communication node fault early warning system. This invention combines AI prediction, ChebNet graph convolutional neural network, and an improved ID3 algorithm to accurately extract features of urban communication nodes and predict fault risks in real time. Through multi-source heterogeneous data fusion and an intelligent response mechanism, the system can efficiently generate fault early warning information and automatically adjust the early warning level and response measures according to the fault type. This significantly improves the accuracy of fault prediction, response efficiency, and stability of the urban communication network, providing an efficient and accurate solution for the operation and maintenance of urban communication networks.

[0006] An AI-based predictive urban communication node fault early warning system according to an embodiment of the present invention includes the following modules: The data acquisition and preprocessing module is used to collect and preprocess raw data of urban communication nodes from multiple heterogeneous data sources to generate a unified urban communication node dataset. The feature extraction module is used to extract features from the urban communication node dataset using the ChebNet graph convolutional neural network and Chebyshev multinomials, to obtain the urban communication node feature vector. The city communication map construction module is used to construct a city communication map by calculating the Euclidean distance between the nodes based on the feature vectors of the city communication nodes. The information gain calculation module is used to calculate the information gain of each node feature based on the city communication map using the improved ID3 algorithm, and select the feature with the largest information gain as the split node of the decision tree to generate a fault classification decision tree. The fault classification graph construction module is used to convert each node and branch into a graph structure according to the splitting rules, starting from the root node of the fault classification decision tree, to obtain the fault classification graph. The fault warning information generation module is used to generate fault warning information based on the fault classification map, combined with the current status of the node and the fault classification information, including the fault type of the node, the warning level, and the relevant response.

[0007] A method for early warning of urban communication node failures based on AI prediction according to an embodiment of the present invention includes the following steps: Step 1: Collect and preprocess raw data of urban communication nodes from multiple heterogeneous data sources to generate a unified urban communication node dataset; Step 2: Input the city communication node dataset into the ChebNet graph convolutional neural network, and obtain the feature vector of the city communication nodes through Chebyshev polynomials; Step 3: Based on the feature vectors of the city communication nodes, calculate the Euclidean distance between the nodes, and construct the city communication map based on the Euclidean distance; Step 4: Based on the city communication map, use the improved ID3 algorithm to calculate the information gain of each node's features, and select the feature with the largest information gain as the splitting node of the decision tree to generate a fault classification decision tree; Step 5: Starting from the root node of the fault classification decision tree, convert each node and branch into a graph structure according to the splitting rules to obtain the fault classification graph; Step Six: Based on the fault classification diagram, and combining the current status of the node with the fault classification information, generate fault warning information, including the fault type of the node, the warning level, and the relevant response.

[0008] Optionally, the raw data of the urban communication nodes includes environmental monitoring data, operational status data, and network topology data of the urban communication nodes; the network topology data includes the connectivity relationships and topology information between the urban communication nodes; the preprocessing step includes outlier removal, missing value imputation, timestamp alignment, and normalization processing of the raw data of different types of urban communication nodes to obtain a unified urban communication node dataset.

[0009] Optionally, the ChebNet graph convolutional neural network is specifically: The city communication node dataset is input into the graph convolutional layer of the ChebNet graph convolutional neural network. Combining the features of city communication nodes and the features of neighboring city communication nodes, Chebyshev multinomials are applied to approximate the city communication node dataset, resulting in the convolutional node feature matrix, specifically: Extract the features of the corresponding city communication nodes from the city communication node dataset to obtain the input feature matrix; Calculate the adjacency matrix and degree matrix of the input feature matrix; The Laplace matrix of the city communication node dataset is obtained by subtracting the adjacency matrix and the degree matrix. The convolutional node feature matrix is ​​obtained by aggregating node features using Chebyshev polynomials based on the Laplacian matrix. The Chebyshev polynomial is twice the Laplace matrix divided by the difference between the largest eigenvalue and the identity matrix; The convolutional node feature matrix is ​​input into the activation layer, and a nonlinear transformation is performed using the Sigmoid nonlinear activation function to obtain the activation node feature matrix. The aggregation layer aggregates the local information of urban communication nodes by performing max pooling on the feature matrix of activated nodes, thereby obtaining the global feature representation of urban communication nodes. The max pooling operation is to select the maximum value of each feature and aggregate the feature values ​​of each urban communication node. The global features are represented as the city communication node feature vector for each city communication node.

[0010] Optionally, step three specifically includes: The feature vector of each city communication node is used as a node in the city communication graph to obtain the node set of the city communication graph. Calculate the Euclidean distance between each pair of nodes in the node set of the city communication map; Based on the calculated Euclidean distance, if the Euclidean distance between two nodes is within a preset distance threshold, then the two corresponding nodes are considered to have a correlation in the city communication graph, and an edge connection is established to obtain the edge set of the city communication graph. The weight of the edges in the edge set of the city communication graph is the Euclidean distance between the corresponding nodes. The city communication graph is obtained based on the node set and edge set of the city communication graph.

[0011] Optionally, the improved ID3 algorithm is specifically as follows: Calculate the Pearson correlation coefficient for each pair of features in each node of the urban communication graph to obtain the Pearson correlation coefficient matrix; If the correlation coefficient between two features is greater than the set threshold, they are considered to be highly correlated, one of the features is retained, and redundant features are removed. If the correlation coefficient between features is lower than the set threshold, it indicates that the two features are independent and both features are retained. Remove all redundant features from the city communication map to obtain the filtered feature matrix, and calculate the entropy of the filtered feature matrix as the total entropy; The feature filtering matrix is ​​divided into multiple sub-matrices according to the different values ​​of each feature; The entropy of each feature in each submatrix is ​​calculated to obtain the conditional entropy of the corresponding feature under different value conditions. The conditional entropy represents the degree of disorder of the submatrix under the constraint of the feature. Calculate the information gain for each feature, where the information gain is the difference between the total entropy of the selected feature matrix and the conditional entropy of each feature in the submatrix; For each feature in the selected feature matrix, calculate the information gain; The feature with the largest information gain value is selected as the splitting node of the decision tree. If the current node meets the stopping condition, the further splitting of the current node is stopped, and the current node is designated as a leaf node. The stopping condition is reaching the maximum tree depth. The selection feature matrix is ​​recursively split using the features of the current split node. Based on the information gain, the selection feature matrix is ​​divided into multiple sub-matrices. For each sub-matrix, the information gain is calculated again, and the feature with the largest information gain is selected for splitting. The structure of the decision tree is recursively constructed to obtain the subtree splitting structure. All recursively generated subtree split structures are integrated to form a complete fault classification decision tree. Each leaf node of the fault classification decision tree corresponds to a different fault category. The status of urban communication nodes is classified according to the characteristics of the leaf nodes.

[0012] Optionally, step five specifically includes: The generated fault classification decision tree is converted into a graph structure. Each node in the fault classification decision tree represents a decision point, each branch represents a specific decision path, each node divides the dataset according to the splitting features of the fault classification decision tree, and each edge represents a condition judgment path from the parent node to the child node. Starting from the root node of the fault classification decision tree, nodes and edges of the fault classification graph are generated according to the splitting rules, which include: Each node in the fault classification graph represents a specific decision point in the decision tree, and the content of the node includes the feature information of the current node and the corresponding fault category. Each edge in the fault classification graph connects a parent node and a child node, representing the process by which a parent node splits into child nodes based on specific conditions. The edge labels describe the conditional judgments of the feature values. For each split node, the node's feature value and judgment condition are used as the label of the edge and connected to the child node; Each split point in the fault classification decision tree is converted into a graph node, and each split node is recursively converted into a node in the fault classification graph until the leaf nodes of the fault classification decision tree have been converted.

[0013] Optionally, step six specifically includes: Based on real-time monitoring data from the urban communication network, obtain the current status information of each node in the fault classification diagram; Based on the decision path in the fault classification diagram, determine which fault category the node currently belongs to; Based on the current state of the node and the fault classification information obtained from the fault classification map, specific fault warning information is generated, specifically as follows: Based on the node classification results, the fault type of the node is determined, including no fault, minor fault, and serious fault; Based on the node's fault type and current status, set the alert level and related response, specifically including: For nodes without faults, the warning level is set to normal. For nodes with minor faults, the warning level is set to warning, and the corresponding response is to manually inspect the nodes with minor faults. For severely faulty nodes, the warning level is set to emergency, and the relevant responses are switching communication paths, isolating the faulty node, and scheduling backup nodes. The beneficial effects of this invention are: This invention, by combining ChebNet graph convolutional neural network and multimodal data fusion, can deeply mine and analyze the spatiotemporal correlations of nodes in urban communication networks. It not only enables real-time monitoring of the status of communication nodes but also accurately predicts potential fault risks through deep learning models. Compared with traditional rule-based monitoring methods, this invention significantly improves the timeliness and accuracy of early warnings, avoids fault occurrences and delayed responses, and reduces the impact of urban communication network faults on urban operations. In terms of data processing, this invention integrates and preprocesses multi-source heterogeneous data, unifying and structuring environmental monitoring data, operational status data, and network topology information of urban communication nodes. This allows for the effective combination of different types of data, enhancing data usability and comprehensiveness. The feature extraction module uses ChebNet graph convolutional neural network, employing Chebyshev multinomials to accurately aggregate and process node features, precisely capturing the spatiotemporal dependencies between nodes. This makes the fault warning model more adaptable and accurate to complex network environments and changes in node behavior.

[0014] Furthermore, based on the improved ID3 algorithm, we introduced feature weights when calculating information gain, enhancing the algorithm's classification ability. We also generated an efficient decision tree structure through recursive splitting, further improving the accuracy of fault classification and the generalization ability of the decision tree. The fault classification graph provided by this invention can more intuitively display the fault decision path and classification process of nodes, helping maintenance personnel quickly identify and locate the source of faults. By combining the current state of nodes and fault classification information, the system can automatically generate detailed fault warning information and set appropriate warning levels and response measures according to different fault types. For "fault-free" nodes, the system maintains normal operation; for "minor fault" nodes, the system provides manual maintenance suggestions; and for "serious fault" nodes, the system can automatically switch communication paths, isolate faulty nodes, or schedule backup nodes, greatly improving the automation and efficiency of fault response. This invention not only effectively improves the predictive ability and timely warning of urban communication network faults, but also reduces the need for manual intervention through intelligent and automated response mechanisms, improving the overall system stability and reliability, thus providing a more efficient and accurate solution for the operation and maintenance management of urban communication networks. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of an AI-based predictive urban communication node fault early warning system proposed in this invention; Figure 2This is an overall flowchart of an AI-based predictive method for early warning of urban communication node failures proposed in this invention. Detailed Implementation

[0016] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0017] refer to Figure 1 An AI-based predictive urban communication node fault early warning system includes the following modules: The data acquisition and preprocessing module is used to collect and preprocess raw data of urban communication nodes from multiple heterogeneous data sources to generate a unified urban communication node dataset. The feature extraction module is used to extract features from the urban communication node dataset using the ChebNet graph convolutional neural network and Chebyshev multinomials, to obtain the urban communication node feature vector. The city communication map construction module is used to construct a city communication map by calculating the Euclidean distance between the nodes based on the feature vectors of the city communication nodes. The information gain calculation module is used to calculate the information gain of each node feature based on the city communication map using the improved ID3 algorithm, and select the feature with the largest information gain as the split node of the decision tree to generate a fault classification decision tree. The fault classification graph construction module is used to convert each node and branch into a graph structure according to the splitting rules, starting from the root node of the fault classification decision tree, to obtain the fault classification graph. The fault warning information generation module is used to generate fault warning information based on the fault classification map, combined with the current status of the node and the fault classification information, including the fault type of the node, the warning level, and the relevant response.

[0018] refer to Figure 2 A method for early warning of urban communication node failures based on AI prediction includes the following steps: Step 1: Collect and preprocess raw data of urban communication nodes from multiple heterogeneous data sources to generate a unified urban communication node dataset; Step 2: Input the city communication node dataset into the ChebNet graph convolutional neural network, and obtain the feature vector of the city communication nodes through Chebyshev polynomials; Step 3: Based on the feature vectors of the city communication nodes, calculate the Euclidean distance between the nodes, and construct the city communication map based on the Euclidean distance; Step 4: Based on the city communication map, use the improved ID3 algorithm to calculate the information gain of each node's features, and select the feature with the largest information gain as the splitting node of the decision tree to generate a fault classification decision tree; Step 5: Starting from the root node of the fault classification decision tree, convert each node and branch into a graph structure according to the splitting rules to obtain the fault classification graph; Step Six: Based on the fault classification diagram, and combining the current status of the node with the fault classification information, generate fault warning information, including the fault type of the node, the warning level, and the relevant response.

[0019] In this embodiment, the original data of the urban communication nodes includes environmental monitoring data, operational status data, and network topology data of the urban communication nodes; the network topology data includes the connectivity relationships and topology information between the urban communication nodes; the preprocessing step includes outlier removal, missing value imputation, timestamp alignment, and normalization processing of the original data of different types of urban communication nodes to obtain a unified urban communication node dataset.

[0020] In this embodiment, the ChebNet graph convolutional neural network is specifically: The city communication node dataset is input into the graph convolutional layer of the ChebNet graph convolutional neural network. Combining the features of city communication nodes and the features of neighboring city communication nodes, Chebyshev multinomials are applied to approximate the city communication node dataset, resulting in the convolutional node feature matrix, specifically: Extract the features of the corresponding city communication nodes from the city communication node dataset to obtain the input feature matrix; Calculate the adjacency matrix and degree matrix of the input feature matrix; The Laplace matrix of the city communication node dataset is obtained by subtracting the adjacency matrix and the degree matrix. The convolutional node feature matrix is ​​obtained by aggregating node features using Chebyshev polynomials based on the Laplacian matrix. The Chebyshev polynomial is twice the Laplace matrix divided by the difference between the largest eigenvalue and the identity matrix; The convolutional node feature matrix is ​​input into the activation layer, and a nonlinear transformation is performed using the Sigmoid nonlinear activation function to obtain the activation node feature matrix. The aggregation layer aggregates the local information of urban communication nodes by performing max pooling on the feature matrix of activated nodes, thereby obtaining the global feature representation of urban communication nodes. The max pooling operation is to select the maximum value of each feature and aggregate the feature values ​​of each urban communication node. The global features are represented as the city communication node feature vector for each city communication node.

[0021] This step introduces the ChebNet graph convolutional neural network and combines it with Chebyshev polynomials to process urban communication node datasets, effectively improving the accuracy and efficiency of the fault early warning system. Through feature extraction from graph convolutional layers and relationship modeling between nodes, the system can fully capture the spatiotemporal dependencies between nodes, reducing the limitations of traditional methods when processing complex data. The application of Chebyshev polynomials not only accelerates computation but also effectively avoids direct calculation of the global graph Laplacian matrix, improving the efficiency of large-scale data processing. Furthermore, through nonlinear activation functions and max pooling operations, the system can adaptively aggregate local information of nodes to generate a global feature vector for each urban communication node, making the system's prediction of node faults more accurate and possessing better generalization ability. This method effectively improves the response speed and predictive ability of urban communication node fault early warning, and is particularly suitable for real-time processing and fault classification of high-dimensional complex data.

[0022] In this embodiment, step three specifically includes: The feature vector of each city communication node is used as a node in the city communication graph to obtain the node set of the city communication graph. Calculate the Euclidean distance between each pair of nodes in the node set of the city communication map; Based on the calculated Euclidean distance, if the Euclidean distance between two nodes is within a preset distance threshold, then the two corresponding nodes are considered to have a correlation in the city communication graph, and an edge connection is established to obtain the edge set of the city communication graph. The weight of the edges in the edge set of the city communication graph is the Euclidean distance between the corresponding nodes. The city communication graph is obtained based on the node set and edge set of the city communication graph.

[0023] In this embodiment, the improved ID3 algorithm is specifically as follows: Calculate the Pearson correlation coefficient for each pair of features in each node of the urban communication graph to obtain the Pearson correlation coefficient matrix; If the correlation coefficient between two features is greater than the set threshold, they are considered to be highly correlated, one of the features is retained, and redundant features are removed. If the correlation coefficient between features is lower than the set threshold, it indicates that the two features are independent and both features are retained. Remove all redundant features from the city communication map to obtain the filtered feature matrix, and calculate the entropy of the filtered feature matrix as the total entropy; The feature filtering matrix is ​​divided into multiple sub-matrices according to the different values ​​of each feature; The entropy of each feature in each submatrix is ​​calculated to obtain the conditional entropy of the corresponding feature under different value conditions. The conditional entropy represents the degree of disorder of the submatrix under the constraint of the feature. Calculate the information gain for each feature, where the information gain is the difference between the total entropy of the selected feature matrix and the conditional entropy of each feature in the submatrix; For each feature in the selected feature matrix, calculate the information gain; The feature with the largest information gain value is selected as the splitting node of the decision tree. If the current node meets the stopping condition, the further splitting of the current node is stopped, and the current node is designated as a leaf node. The stopping condition is reaching the maximum tree depth. The selection feature matrix is ​​recursively split using the features of the current split node. Based on the information gain, the selection feature matrix is ​​divided into multiple sub-matrices. For each sub-matrix, the information gain is calculated again, and the feature with the largest information gain is selected for splitting. The structure of the decision tree is recursively constructed to obtain the subtree splitting structure. All recursively generated subtree split structures are integrated to form a complete fault classification decision tree. Each leaf node of the fault classification decision tree corresponds to a different fault category. The status of urban communication nodes is classified according to the characteristics of the leaf nodes.

[0024] This step effectively optimizes the construction process of the fault classification decision tree through an improved ID3 algorithm, significantly enhancing the accuracy and robustness of the decision tree, especially when processing node features in urban communication maps. By calculating the Pearson correlation coefficient for each pair of features, redundant features can be effectively identified and removed, avoiding interference from high correlations between features and ensuring the effectiveness and conciseness of feature selection. Furthermore, by calculating information gain and combining it with conditional entropy, the algorithm can accurately evaluate the contribution of each feature to node classification, thus selecting the most discriminative feature for splitting when constructing the decision tree, greatly improving the accuracy of fault classification. The recursive feature splitting process and the selection of the feature with the maximum information gain make the generated fault classification decision tree structure more concise, accurate, and possessing good generalization ability. In addition, the improved ID3 algorithm avoids overfitting problems through depth limits and stopping conditions, thereby enhancing the stability and predictive performance of the model. Overall, the decision tree construction method provided by this invention not only effectively improves the accuracy of fault prediction but also provides an efficient and flexible solution for fault early warning systems in complex urban communication networks, greatly improving the system's intelligence level and response speed.

[0025] In this embodiment, step five specifically includes: The generated fault classification decision tree is converted into a graph structure. Each node in the fault classification decision tree represents a decision point, each branch represents a specific decision path, each node divides the dataset according to the splitting features of the fault classification decision tree, and each edge represents a condition judgment path from the parent node to the child node. Starting from the root node of the fault classification decision tree, nodes and edges of the fault classification graph are generated according to the splitting rules, which include: Each node in the fault classification graph represents a specific decision point in the decision tree, and the content of the node includes the feature information of the current node and the corresponding fault category. Each edge in the fault classification graph connects a parent node and a child node, representing the process by which a parent node splits into child nodes based on specific conditions. The edge labels describe the conditional judgments of the feature values. For each split node, the node's feature value and judgment condition are used as the label of the edge and connected to the child node; Each split point in the fault classification decision tree is converted into a graph node, and each split node is recursively converted into a node in the fault classification graph until the leaf nodes of the fault classification decision tree have been converted.

[0026] In this embodiment, step six specifically includes: Based on real-time monitoring data from the urban communication network, obtain the current status information of each node in the fault classification diagram; Based on the decision path in the fault classification diagram, determine which fault category the node currently belongs to; Based on the current state of the node and the fault classification information obtained from the fault classification map, specific fault warning information is generated, specifically as follows: Based on the node classification results, the fault type of the node is determined, including no fault, minor fault, and serious fault; Based on the node's fault type and current status, set the alert level and related response, specifically including: For nodes without faults, the warning level is set to normal. For nodes with minor faults, the warning level is set to warning, and the corresponding response is to manually inspect the nodes with minor faults. For severely faulty nodes, the warning level is set to emergency, and the relevant responses are switching communication paths, isolating the faulty node, and scheduling backup nodes. Example 1: To verify the feasibility of this invention in practice, it was applied to the communication network of a large city, which contains hundreds of communication nodes. These nodes cover different areas and different types of communication infrastructure, including fiber optic switching equipment, routers, base stations, and monitoring equipment. These devices are often exposed to complex environments, such as extreme weather, power fluctuations, physical damage, and network attacks, leading to a diverse frequency and type of faults, posing a significant challenge to the stability of the communication network. Traditional fault detection systems typically rely on simple threshold settings and experience-based rules. These methods cannot effectively cope with the diversity and complexity of node faults, resulting in faults often being unpredictable and unhandled after they occur, and may even lead to severe communication outages.

[0027] In this context, the system of this embodiment is applied to the city's communication network. Using the method of this invention, the system can accurately analyze and predict communication node failures, reducing network outages and maintenance costs. The system performs data acquisition and preprocessing through multi-source heterogeneous data fusion, including environmental monitoring data (such as temperature and humidity), node operating status data (such as bandwidth, traffic, and latency), and network topology data (such as inter-node connectivity). Then, the system inputs this data into a ChebNet graph convolutional neural network, processing and aggregating node features using Chebyshev multinomials to obtain the feature vector for each node. Next, based on these feature vectors, the Euclidean distance between nodes is calculated to construct a city communication map. Subsequently, an improved ID3 algorithm is used to calculate the information gain of each node's features, generating a fault classification decision tree. Finally, fault warning information is generated through the fault classification map.

[0028] The application of this system significantly improves fault detection and response speed. In real-world scenarios, the system can make accurate predictions based on a node's current state and historical data before a node fails. Compared to traditional methods, the system demonstrates significant advantages in accuracy and response time. For example, in a node failure early warning process, the system predicted an impending node failure one hour in advance, while traditional methods could only detect and issue a warning approximately 0.5 hours and minutes after the failure occurred. This early warning effectively avoids communication interruptions and subsequent high repair costs.

[0029] Taking a base station node in a certain area as an example, there are 10 communication nodes in this area. The system successfully extracted the node feature vectors using a ChebNet graph convolutional neural network and generated a city communication map. The system predicted the fault type of the nodes, including three types: "no fault," "minor fault," and "serious fault." Using the ID3 algorithm, the system calculated the feature information gain of each node and generated a fault classification decision tree. In this process, node A was predicted as "minor fault," node B was predicted as "no fault," and node C was predicted as "serious fault." Based on the early warning information generated by the fault classification map, the system promptly sent handling suggestions to the maintenance personnel and switched the communication path of node A through an automated response mechanism, successfully preventing the potential spread of the fault.

[0030] Table 1 Performance Comparison of Urban Communication Node Fault Early Warning Methods

[0031] Table 1 shows the following: Threshold-based monitoring methods detect node status based on preset thresholds. A node is considered faulty once its monitoring indicators exceed or fall below a certain threshold. For node A, this method only detected the fault at 10:30 AM, without any prior warning. For fault handling, this method relies on manual intervention, which cannot provide a rapid response or prevent fault propagation. Rule-based monitoring methods rely on predetermined rules to determine whether a node is faulty, such as using historical experience or standards set by business rules to detect node health. For node B, this method detected a fault-free state with no operational requirements. Although it handles fault-free nodes well, its fault prediction capability is poor, failing to issue warnings before faults occur in nodes A or C. Traditional machine learning-based methods build machine learning models through training data, enabling fault prediction by learning from historical node data. While this method identified a fault half an hour in advance in the prediction of node C, it still has a significant response delay (zero-hour advance warning) compared to this invention, and it failed to provide an early warning in the fault prediction of node A.

[0032] This invention extracts features from urban communication nodes using a ChebNet graph convolutional neural network and combines this with multimodal data analysis to accurately predict the failures of nodes A and C one hour in advance. Particularly in the prediction of node C, the system effectively assesses the node's state using a fault classification map and the ID3 algorithm, providing early warning and automatic response, effectively preventing fault propagation. Furthermore, for node B in a fault-free state, this invention can accurately confirm its health status without requiring additional processing.

[0033] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AI-based predictive urban communication node fault early warning system, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect and preprocess raw data of urban communication nodes from multiple heterogeneous data sources to generate a unified urban communication node dataset. The feature extraction module is used to extract features from the urban communication node dataset using the ChebNet graph convolutional neural network and Chebyshev multinomials, to obtain the urban communication node feature vector. The city communication map construction module is used to construct a city communication map by calculating the Euclidean distance between the nodes based on the feature vectors of the city communication nodes. The information gain calculation module is used to calculate the information gain of each node feature based on the city communication map using the improved ID3 algorithm, and select the feature with the largest information gain as the split node of the decision tree to generate a fault classification decision tree. The fault classification graph construction module is used to convert each node and branch into a graph structure according to the splitting rules, starting from the root node of the fault classification decision tree, to obtain the fault classification graph. The fault warning information generation module is used to generate fault warning information based on the fault classification map, combined with the current status of the node and the fault classification information, including the fault type of the node, the warning level, and the relevant response.

2. The urban communication node fault early warning system based on AI prediction according to claim 1, characterized in that, The modules are connected in the following way: Step 1: Collect and preprocess raw data of urban communication nodes from multiple heterogeneous data sources to generate a unified urban communication node dataset; Step 2: Input the city communication node dataset into the ChebNet graph convolutional neural network, and obtain the feature vector of the city communication nodes through Chebyshev polynomials; Step 3: Based on the feature vectors of the city communication nodes, calculate the Euclidean distance between the nodes, and construct the city communication map based on the Euclidean distance; Step 4: Based on the city communication map, use the improved ID3 algorithm to calculate the information gain of each node's features, and select the feature with the largest information gain as the splitting node of the decision tree to generate a fault classification decision tree; Step 5: Starting from the root node of the fault classification decision tree, convert each node and branch into a graph structure according to the splitting rules to obtain the fault classification graph; Step Six: Based on the fault classification diagram, and combining the current status of the node with the fault classification information, generate fault warning information, including the fault type of the node, the warning level, and the relevant response.

3. The urban communication node fault early warning system based on AI prediction according to claim 2, characterized in that, The raw data of the urban communication nodes includes environmental monitoring data, operational status data, and network topology data of the urban communication nodes; the network topology data includes the connectivity relationships and topology information between urban communication nodes; the preprocessing steps include outlier removal, missing value imputation, timestamp alignment, and normalization of the raw data of different types of urban communication nodes to obtain a unified urban communication node dataset.

4. The urban communication node fault early warning system based on AI prediction according to claim 2, characterized in that, The ChebNet graph convolutional neural network is specifically as follows: The city communication node dataset is input into the graph convolutional layer of the ChebNet graph convolutional neural network. Combining the features of city communication nodes and the features of neighboring city communication nodes, Chebyshev multinomials are applied to approximate the city communication node dataset, resulting in the convolutional node feature matrix, specifically: Extract the features of the corresponding city communication nodes from the city communication node dataset to obtain the input feature matrix; Calculate the adjacency matrix and degree matrix of the input feature matrix; The Laplace matrix of the city communication node dataset is obtained by subtracting the adjacency matrix and the degree matrix. The convolutional node feature matrix is ​​obtained by aggregating node features using Chebyshev polynomials based on the Laplacian matrix. The Chebyshev polynomial is twice the Laplace matrix divided by the difference between the largest eigenvalue and the identity matrix; The convolutional node feature matrix is ​​input into the activation layer, and a nonlinear transformation is performed using the Sigmoid nonlinear activation function to obtain the activation node feature matrix. The aggregation layer aggregates the local information of urban communication nodes by performing max pooling on the feature matrix of activated nodes, thereby obtaining the global feature representation of urban communication nodes. The max pooling operation is to select the maximum value of each feature and aggregate the feature values ​​of each urban communication node. The global features are represented as the city communication node feature vector for each city communication node.

5. The urban communication node fault early warning system based on AI prediction according to claim 2, characterized in that, Step three specifically involves: The feature vector of each city communication node is used as a node in the city communication graph to obtain the node set of the city communication graph. Calculate the Euclidean distance between each pair of nodes in the node set of the city communication map; Based on the calculated Euclidean distance, if the Euclidean distance between two nodes is within a preset distance threshold, then the two corresponding nodes are considered to have a correlation in the city communication graph, and an edge connection is established to obtain the edge set of the city communication graph. The weight of the edges in the edge set of the city communication graph is the Euclidean distance between the corresponding nodes. The city communication graph is obtained based on the node set and edge set of the city communication graph.

6. The urban communication node fault early warning system based on AI prediction according to claim 2, characterized in that, The improved ID3 algorithm is specifically as follows: Calculate the Pearson correlation coefficient for each pair of features in each node of the urban communication graph to obtain the Pearson correlation coefficient matrix; If the correlation coefficient between two features is greater than the set threshold, they are considered to be highly correlated, one of the features is retained, and redundant features are removed. If the correlation coefficient between features is lower than the set threshold, it indicates that the two features are independent and both features are retained. Remove all redundant features from the city communication map to obtain the filtered feature matrix, and calculate the entropy of the filtered feature matrix as the total entropy; The feature filtering matrix is ​​divided into multiple sub-matrices according to the different values ​​of each feature; The entropy of each feature in each submatrix is ​​calculated to obtain the conditional entropy of the corresponding feature under different value conditions. The conditional entropy represents the degree of disorder of the submatrix under the constraint of the feature. Calculate the information gain for each feature, where the information gain is the difference between the total entropy of the selected feature matrix and the conditional entropy of each feature in the submatrix; For each feature in the selected feature matrix, calculate the information gain; The feature with the largest information gain value is selected as the splitting node of the decision tree. If the current node meets the stopping condition, the further splitting of the current node is stopped, and the current node is designated as a leaf node. The stopping condition is reaching the maximum tree depth. The selection feature matrix is ​​recursively split using the features of the current split node. Based on the information gain, the selection feature matrix is ​​divided into multiple sub-matrices. For each sub-matrix, the information gain is calculated again, and the feature with the largest information gain is selected for splitting. The structure of the decision tree is recursively constructed to obtain the subtree splitting structure. All recursively generated subtree split structures are integrated to form a complete fault classification decision tree. Each leaf node of the fault classification decision tree corresponds to a different fault category. The status of urban communication nodes is classified according to the characteristics of the leaf nodes.

7. The urban communication node fault early warning system based on AI prediction according to claim 2, characterized in that, Step five specifically involves: The generated fault classification decision tree is converted into a graph structure. Each node in the fault classification decision tree represents a decision point, each branch represents a specific decision path, each node divides the dataset according to the splitting features of the fault classification decision tree, and each edge represents a condition judgment path from the parent node to the child node. Starting from the root node of the fault classification decision tree, nodes and edges of the fault classification graph are generated according to the splitting rules, which include: Each node in the fault classification graph represents a specific decision point in the decision tree, and the content of the node includes the feature information of the current node and the corresponding fault category. Each edge in the fault classification graph connects a parent node and a child node, representing the process by which a parent node splits into child nodes based on specific conditions. The edge labels describe the conditional judgments of the feature values. For each split node, the node's feature value and judgment condition are used as the label of the edge and connected to the child node; Each split point in the fault classification decision tree is converted into a graph node, and each split node is recursively converted into a node in the fault classification graph until the leaf nodes of the fault classification decision tree have been converted.

8. The urban communication node fault early warning system based on AI prediction according to claim 2, characterized in that, Step six specifically involves: Based on real-time monitoring data from the urban communication network, obtain the current status information of each node in the fault classification diagram; Based on the decision path in the fault classification diagram, determine which fault category the node currently belongs to; Based on the current state of the node and the fault classification information obtained from the fault classification map, specific fault warning information is generated, specifically as follows: Based on the node classification results, the fault type of the node is determined, including no fault, minor fault, and serious fault; Based on the node's fault type and current status, set the alert level and related response, specifically including: For nodes without faults, the warning level is set to normal. For nodes with minor faults, the warning level is set to warning, and the corresponding response is to manually inspect the nodes with minor faults. For severely faulty nodes, the warning level is set to emergency, and the relevant responses are switching communication paths, isolating the faulty node, and scheduling a backup node.

Images