Power line health status assessment and prediction method and system based on big data

Abstract

The present invention provides a method and system for evaluating and predicting the health status of power lines based on big data, which relates to the technical field of power line health assessment. The method comprises preprocessing historical operating data and generating standardized data, extracting fault features using a spatiotemporal heterogeneous graph network, constructing a basic hierarchical knowledge base for automatic labeling, and finally generating a training sample set; modeling a directed acyclic graph, identifying trunk transmission lines and dividing risk sections, optimizing the knowledge base, adaptively generating inspection plans based on a reinforcement learning inspection strategy model, evaluating the health status of power lines, and generating health status reports. The present invention improves the fault prediction accuracy and inspection efficiency of power lines.

Description

Technical Field

[0001] The present invention relates to the technical field of power line health assessment, and in particular to a method and system for assessing and predicting the health status of a power line based on big data. Background Art

[0002] As the scale of power systems continues to expand, the health of power lines directly impacts their safe and stable operation. Currently, power line health assessment primarily relies on regular inspections and real-time monitoring data. This assessment and prediction of line health is achieved by collecting heterogeneous data from multiple sources, including operating parameters, environmental parameters, and equipment status. The development of big data technology has provided new technical means for power line health assessment, enabling the extraction of fault characteristics and operating patterns from massive amounts of historical data, providing data support for health assessment.

[0003] However, existing technologies still have shortcomings. Traditional health status assessment methods often process each monitoring data independently, ignoring the temporal and spatial correlation between data, resulting in insufficient feature extraction; most existing inspection strategies adopt a fixed-cycle inspection mode, failing to dynamically adjust the inspection frequency according to the importance of the line and the risk of faults, resulting in a waste of inspection resources; there is a lack of in-depth analysis of the topological structure characteristics of power lines, which makes it impossible to accurately identify key transmission paths and difficult to achieve differentiated management.

[0004] In summary, the present invention aims to solve the above-mentioned technical problems and proposes a power line health status assessment and prediction method based on big data. It extracts the correlation features of multi-source data through a spatiotemporal heterogeneous graph network, establishes a hierarchical knowledge base to realize the effective organization of knowledge, adopts reinforcement learning method to optimize the inspection strategy, and combines the power line topology characteristics to realize differentiated inspection and evaluation, thereby improving the accuracy of power line health status assessment and inspection efficiency. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for evaluating and predicting the health status of power lines based on big data, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] A power line health status assessment and prediction method based on big data is provided, including:

[0008] Obtain historical operating data of power lines, pre-process the historical operating data according to preset data quality rules, and generate standardized operating data;

[0009] Extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, build a basic hierarchical knowledge base, automatically annotate the standardized operating data, and generate a training sample set;

[0010] The power line system is modeled as a directed acyclic graph, the transmission path eigenvalues ​​are calculated, the trunk transmission lines are identified, the risk sections are divided, and the risk sections are integrated into the basic hierarchical knowledge base to obtain an optimized hierarchical knowledge base.

[0011] Based on the optimized hierarchical knowledge base and trunk transmission lines, an inspection strategy model is established based on the reinforcement learning method, and an inspection execution plan is adaptively generated. The inspection execution plan is executed and the health status of the power line is evaluated to obtain the health status assessment results.

[0012] Generate a power line health status report based on the health status assessment results.

[0013] In an optional embodiment,

[0014] Extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, and build a basic hierarchical knowledge base including:

[0015] The standardized operation data includes equipment data, environmental data and historical fault data;

[0016] Based on standardized operation data, they are respectively constructed into equipment status nodes, environmental feature nodes and historical fault nodes;

[0017] Based on the physical connection relationship between device status nodes, the topological association edge between devices is established. Based on the influence relationship between device status nodes and environmental feature nodes, the device-environment association edge is established. Based on the correspondence between device status nodes and historical fault nodes, the device-fault association edge is established. Together, they form a heterogeneous information network.

[0018] For each node in the heterogeneous information network, type-aware weighted aggregation is performed based on a multi-head attention mechanism to extract fault features, obtaining a fault feature vector. Feature aggregation is then performed through differentiable pooling guided by node importance to obtain a subgraph feature representation.

[0019] Performing a sliding convolution operation on the subgraph feature representation using a causal convolution kernel, generating a time series feature vector through time series information processing, calculating the contrast loss between normal operating state samples and the boundary loss between abnormal operating state samples, and obtaining a comprehensive anomaly score of the fault feature;

[0020] The time series feature vector, the comprehensive anomaly score and the corresponding fault case are stored in the instance layer to establish a mapping relationship between the fault feature and the operating status; the fault cases whose Euclidean distance between the time series feature vectors corresponding to the fault cases is less than a first preset threshold are aggregated into fault modes and stored in the pattern layer; the evolution relationship between the fault modes whose time series correlation corresponding to the fault modes is greater than a second preset threshold is stored in the rule layer; the instance layer, the pattern layer and the rule layer constitute a basic hierarchical knowledge base.

[0021] In an optional embodiment,

[0022] For each node in the heterogeneous information network, type-aware weighted aggregation is performed based on the multi-head attention mechanism to extract fault features, obtaining a fault feature vector. Feature aggregation is performed through differentiable pooling guided by node importance, and the obtained subgraph feature representation includes:

[0023] Constructing a multi-head attention weight matrix based on different types of neighbor nodes of each node in the heterogeneous information network, wherein the multi-head attention weight matrix is ​​used to calculate the dynamic association strength between the node and the different types of neighbor nodes;

[0024] Multiplying the original feature vector of the node by the conversion matrix corresponding to the type of the node to obtain a type-aware feature vector, and performing a weighted aggregation operation on the type-aware feature vector based on the dynamic association strength to generate a fault feature vector;

[0025] The fault feature vector is concatenated with the global feature vector of the heterogeneous information network, and node importance scores are calculated using a preset multi-layer perceptron. Nodes are sampled based on the node importance scores to construct a local network structure.

[0026] A differentiable pooling operation is used to perform feature aggregation on the fault feature vector in the local network structure, and the aggregated features are updated through a gated recurrent unit to obtain a subgraph feature representation.

[0027] In an optional embodiment,

[0028] The power line system is modeled as a directed acyclic graph, the transmission path eigenvalues ​​are calculated, the trunk transmission lines are identified, the risk sections are divided, and the risk sections are integrated into the basic hierarchical knowledge base. The optimized hierarchical knowledge base includes:

[0029] Constructing a directed acyclic graph of the power line system, the directed acyclic graph including a set of operating characteristic nodes and a set of power transmission edges, obtaining an operating characteristic value of each node in a training sample set, and setting the power transmission efficiency corresponding to each edge as the edge characteristic value;

[0030] Calculating the sum of the connected edge eigenvalues ​​of each running feature node in the directed acyclic graph to obtain a node importance value, determining an electric energy transmission path based on the node importance value, and calculating the product of the edge eigenvalues ​​of each electric energy transmission path to obtain a path eigenvalue;

[0031] Obtain the line load threshold of each operating characteristic node, identify the trunk transmission line that meets the load threshold constraint through the iterative capacity-adjusted maximum flow algorithm, and calculate the corresponding edge cut set load ratio;

[0032] Obtain the failure probability of each operating characteristic node and the set of downstream affected nodes, calculate the node risk value, modify the node risk value based on the load ratio of the edge cut set, and divide the risk segments using a spectral clustering algorithm;

[0033] The trunk transmission line is written into the skeleton layer of the basic hierarchical knowledge base, the protection level is divided based on the risk section, the path characteristic value and the node risk value are recorded, a hierarchical association relationship including the edge cut set load ratio is established, and the control strategy priority of the basic hierarchical knowledge base is updated based on the hierarchical association relationship to obtain an optimized hierarchical knowledge base.

[0034] In an optional embodiment,

[0035] Obtain the line load threshold of each operating characteristic node, identify the trunk transmission line that meets the load threshold constraint through the iterative capacity adjustment maximum flow algorithm, and calculate the corresponding edge cut set load ratio including:

[0036] Obtain the line load threshold of each operating characteristic node in the power system network, construct an initial capacity matrix, and use the maximum flow algorithm to obtain the initial maximum flow value under the constraint that the actual load is less than or equal to the line load threshold. Based on the initial maximum flow value, determine the corresponding minimum cut set to obtain the initial trunk transmission line;

[0037] Apply a disturbance coefficient to the initial trunk transmission line, modify the initial capacity matrix to obtain a modified capacity matrix, calculate the maximum flow value after the disturbance, and calculate the ratio of the maximum flow value to the initial maximum flow value to obtain the network fitness index;

[0038] Based on the difference between the network fitness index and the preset fitness threshold, the modified capacity matrix is ​​updated to obtain the iterative capacity matrix. The iterative maximum flow value corresponding to the iterative capacity matrix is ​​calculated. The iteration is repeated until the rate of change of the maximum flow values ​​of two adjacent iterative cycles is less than the preset rate of change threshold. The final capacity matrix is ​​obtained. From the final capacity matrix, lines that meet the line load threshold are selected as the trunk transmission optimization lines, and the corresponding edge cut set load ratio is calculated.

[0039] In an optional embodiment,

[0040] Based on the optimized hierarchical knowledge base and backbone transmission lines, an inspection strategy model is established based on the reinforcement learning method. The inspection execution plan is adaptively generated, including:

[0041] Based on the fault characteristics in the optimized hierarchical knowledge base, the mapping relationship between the fault characteristics and the operating status, and the transmission path characteristic values ​​of the trunk transmission line, a regional characteristic function is constructed, and the boundary of the dynamic associated sub-region is determined according to the regional characteristic function;

[0042] A local state evaluator is constructed for the dynamically associated sub-regions. The risk factor score of the local state evaluator is multiplied by the weight coefficient to obtain the regional risk degree. The regional risk degree is then combined with the inspection time interval and load level to calculate the regional inspection urgency.

[0043] Constructing a feature similarity matrix of dynamically associated sub-regions, generating a regional knowledge transfer function based on the feature similarity matrix, and fusing the regional knowledge transfer function with historical inspection experience to obtain regional shared experience;

[0044] The regional inspection urgency, regional risk, and regional shared experience are used to construct a state vector, the inspection priority parameter is used to construct an action vector, and a reward function is established based on the state vector and the action vector;

[0045] A reinforcement learning method is used to optimize the reward function to obtain strategy model parameters, and an inspection execution plan is generated according to the strategy model parameters.

[0046] In an optional embodiment,

[0047] Constructing a feature similarity matrix of dynamically associated sub-regions, generating a regional knowledge transfer function based on the feature similarity matrix, and integrating the regional knowledge transfer function with historical inspection experience to obtain regional shared experience includes:

[0048] The dynamically associated sub-regions are constructed as a graph network model, where each dynamically associated sub-region corresponds to a region node, and the region node contains fault characteristics and operating status data. Edge connections are constructed based on the physical connection relationship between the dynamically associated sub-regions, and the topological association features between the region nodes are extracted through a graph convolutional network to obtain a feature similarity matrix;

[0049] Calculating the similarity between adjacent regional nodes to obtain a gating coefficient, and constructing a regional knowledge transfer function based on the gating coefficient and the feature similarity matrix, wherein the regional knowledge transfer function represents the knowledge transfer relationship between regional nodes;

[0050] Acquire features of the same dynamically associated sub-region in different time windows to construct positive sample pairs, acquire features of different dynamically associated sub-regions to construct negative sample pairs, and train a feature extraction model using a comparative loss function to obtain historical inspection experience;

[0051] The regional knowledge transfer function is integrated with the historical inspection experience to obtain regional shared experience.

[0052] According to a second aspect of the embodiments of the present invention,

[0053] Provided is a power line health status assessment and prediction system based on big data, including:

[0054] The first unit is used to obtain historical operation data of the power line, pre-process the historical operation data according to preset data quality rules, and generate standardized operation data;

[0055] The second unit is used to extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, build a basic hierarchical knowledge base, automatically annotate the standardized operating data, and generate a training sample set;

[0056] The third unit is used to model the power line system as a directed acyclic graph, calculate the transmission path characteristic values, identify the trunk transmission lines, divide the risk sections, and integrate them into the basic hierarchical knowledge base to obtain an optimized hierarchical knowledge base;

[0057] The fourth unit is used to establish an inspection strategy model based on the optimized hierarchical knowledge base and the trunk transmission line based on the reinforcement learning method, and adaptively generate an inspection execution plan; execute the inspection execution plan and evaluate the health status of the power line to obtain the health status assessment result;

[0058] The fifth unit is used to generate a power line health status report based on the health status assessment result.

[0059] According to a third aspect of the embodiments of the present invention,

[0060] An electronic device is provided, comprising:

[0061] processor;

[0062] a memory for storing processor-executable instructions;

[0063] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0064] According to a fourth aspect of the embodiments of the present invention,

[0065] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0066] In an embodiment of the present invention, by preprocessing and standardizing the historical operation data of the power line, the quality and consistency of the data can be improved, thereby providing a reliable data basis for subsequent fault feature extraction and status evaluation, and more accurately reflecting the actual operation status of the power line; based on the spatiotemporal heterogeneous graph network, fault features are extracted and a mapping relationship between features and operation status is established, which can effectively identify potential fault risks, improve the accuracy of fault detection, provide a scientific basis for the maintenance and management of the power line, and reduce the probability of fault occurrence; combined with the adaptive inspection strategy model generated by the reinforcement learning method, it can realize dynamic evaluation and real-time monitoring of the health status of the power line, optimize the configuration of inspection resources, improve inspection efficiency, and ensure the safe and stable operation of the power line. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a method for evaluating and predicting the health status of power lines based on big data according to an embodiment of the present invention;

[0068] Figure 2 This is the performance analysis diagram of the early warning time;

[0069] Figure 3 Comparison diagram of network topology and load distribution before and after optimization of power system trunk transmission lines;

[0070] Figure 4 A visualization diagram of the knowledge tree migration flow and experience accumulation. DETAILED DESCRIPTION

[0071] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0072] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0073] Figure 1 FIG. 1 is a flow chart of a method for evaluating and predicting the health status of power lines based on big data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0074] Obtain historical operating data of power lines, pre-process the historical operating data according to preset data quality rules, and generate standardized operating data;

[0075] Extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, build a basic hierarchical knowledge base, automatically annotate the standardized operating data, and generate a training sample set;

[0076] The power line system is modeled as a directed acyclic graph, the transmission path eigenvalues ​​are calculated, the trunk transmission lines are identified, the risk sections are divided, and the risk sections are integrated into the basic hierarchical knowledge base to obtain an optimized hierarchical knowledge base.

[0077] Based on the optimized hierarchical knowledge base and trunk transmission lines, an inspection strategy model is established based on the reinforcement learning method, and an inspection execution plan is adaptively generated. The inspection execution plan is executed and the health status of the power line is evaluated to obtain the health status assessment results.

[0078] Generate a power line health status report based on the health status assessment results.

[0079] In an optional embodiment, extracting fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establishing a mapping relationship between features and operating states, and constructing a basic hierarchical knowledge base includes:

[0080] The standardized operation data includes equipment data, environmental data and historical fault data;

[0081] Based on standardized operation data, they are respectively constructed into equipment status nodes, environmental feature nodes and historical fault nodes;

[0082] Based on the physical connection relationship between device status nodes, the topological association edge between devices is established. Based on the influence relationship between device status nodes and environmental feature nodes, the device-environment association edge is established. Based on the correspondence between device status nodes and historical fault nodes, the device-fault association edge is established. Together, they form a heterogeneous information network.

[0083] For each node in the heterogeneous information network, type-aware weighted aggregation is performed based on a multi-head attention mechanism to extract fault features, obtaining a fault feature vector. Feature aggregation is then performed through differentiable pooling guided by node importance to obtain a subgraph feature representation.

[0084] Performing a sliding convolution operation on the subgraph feature representation using a causal convolution kernel, generating a time series feature vector through time series information processing, calculating the contrast loss between normal operating state samples and the boundary loss between abnormal operating state samples, and obtaining a comprehensive anomaly score of the fault feature;

[0085] The time series feature vector, the comprehensive anomaly score and the corresponding fault case are stored in the instance layer to establish a mapping relationship between the fault feature and the operating status; the fault cases whose Euclidean distance between the time series feature vectors corresponding to the fault cases is less than a first preset threshold are aggregated into fault modes and stored in the pattern layer; the evolution relationship between the fault modes whose time series correlation corresponding to the fault modes is greater than a second preset threshold is stored in the rule layer; the instance layer, the pattern layer and the rule layer constitute a basic hierarchical knowledge base.

[0086] In one specific embodiment, standardized operating data is acquired, and the operating data includes equipment data, environmental data, and historical fault data. Equipment data may include parameters such as equipment temperature, pressure, and operating time; environmental data may include external environmental conditions such as humidity, temperature, and air pressure; and historical fault data includes the types of faults that have occurred in the past and the time they occurred.

[0087] The device status node represents the current state of the device, the environmental characteristic node represents the characteristics of the environment, and the historical fault node represents the device's past fault records. By analyzing the physical connection relationships between device status nodes, topological association edges are established between devices. For example, if two devices are connected by a cable, a topological association edge is established between them. Based on the influence relationship between the device status node and the environmental characteristic node, a device-environment association edge is established. If the operating state of a device is significantly affected by ambient temperature, an association edge is established between the device status node and the environmental characteristic node. Based on the corresponding relationship between the device status node and the historical fault node, a device-fault association edge is established. If a device has experienced a fault in a specific environment, an association edge is established between its status node and the corresponding fault node. This constructs a heterogeneous information network consisting of device status nodes, environmental characteristic nodes, and historical fault nodes.

[0088] In the constructed heterogeneous information network, a multi-head attention mechanism is used to perform type-aware weighted aggregation to extract fault features. For each node, based on its type (device, environment, or fault), different attention heads perform weighted aggregation of the features of adjacent nodes to obtain the node's fault feature vector. A differentiable pooling method guided by node importance is then used to aggregate these features to obtain a subgraph feature representation. By analyzing the importance of nodes in the network, the features of important nodes are selectively aggregated, thereby improving the effectiveness of the feature representation.

[0089] A sliding convolution operation is performed on the subgraph feature representation using a causal convolution kernel. By sliding the convolution kernel over the time series, time series features are extracted and a time series feature vector is generated. A comprehensive anomaly score for the fault signature is calculated by calculating the contrast loss between samples in normal operating conditions and the boundary loss between samples in abnormal operating conditions. The smaller the difference between the feature vectors of normal samples and the larger the difference between abnormal samples, the higher the comprehensive anomaly score, reflecting the severity of the fault.

[0090] The time series feature vectors, comprehensive anomaly scores, and corresponding fault cases are stored in the instance layer to establish a mapping between fault features and operating status. The instance layer stores detailed information about each fault case, including its time series feature vector and comprehensive anomaly score. This information is used for subsequent fault pattern identification and evolution analysis.

[0091] At the instance layer, the Euclidean distance between the time series feature vectors corresponding to the fault cases is calculated. If the distance is less than a first preset threshold, the fault cases are aggregated into fault patterns and stored in the pattern layer. Fault patterns represent a group of fault cases with similar characteristics and manifestations, facilitating subsequent analysis and processing.

[0092] In the pattern layer, the temporal correlation between fault modes is analyzed. If the temporal correlation between certain fault modes exceeds a second preset threshold, the evolutionary relationships between these fault modes are recorded and stored in the rule layer. The rule layer stores the evolutionary relationships between fault modes, helping to analyze potential fault trends and mutual impacts.

[0093] Through the above steps, a basic hierarchical knowledge base is constructed, consisting of an instance layer, a pattern layer, and a rule layer. This basic hierarchical knowledge base can provide an important basis for predicting and diagnosing equipment failures, helping maintenance personnel take timely measures to reduce equipment failure rates and improve equipment reliability and safety.

[0094] For example, assume that a device is operating at a temperature of 70°C, a pressure of 1.5 MPa, and an ambient humidity of 60%. Historical fault data indicates that the device has previously experienced a fault under the same temperature and humidity conditions. Using the above method, a heterogeneous information network is constructed, with associated edges established between device status nodes and environmental feature nodes, forming a complete fault feature extraction system. Based on the device's operating status and historical fault records, the resulting composite anomaly score is 0.85, indicating a high risk of failure in the device's current state.

[0095] In an optional embodiment, for each node in the heterogeneous information network, type-aware weighted aggregation is performed based on a multi-head attention mechanism to extract fault features to obtain a fault feature vector, and feature aggregation is performed through differentiable pooling guided by node importance to obtain a subgraph feature representation including:

[0096] Constructing a multi-head attention weight matrix based on different types of neighbor nodes of each node in the heterogeneous information network, wherein the multi-head attention weight matrix is ​​used to calculate the dynamic association strength between the node and the different types of neighbor nodes;

[0097] Multiplying the original feature vector of the node by the conversion matrix corresponding to the type of the node to obtain a type-aware feature vector, and performing a weighted aggregation operation on the type-aware feature vector based on the dynamic association strength to generate a fault feature vector;

[0098] The fault feature vector is concatenated with the global feature vector of the heterogeneous information network, and node importance scores are calculated using a preset multi-layer perceptron. Nodes are sampled based on the node importance scores to construct a local network structure.

[0099] A differentiable pooling operation is used to perform feature aggregation on the fault feature vector in the local network structure, and the aggregated features are updated through a gated recurrent unit to obtain a subgraph feature representation.

[0100] In one specific implementation, a node type mapping table is established to assign a unique type identifier to each node in the power line: equipment nodes such as transmission towers and transformers are designated as Type 1, environmental sensor nodes for temperature, humidity, and wind speed are designated as Type 2, and nodes with historical fault records such as insulator loss and conductor breakage are designated as Type 3. For each node in the network, an adjacency table is established based on the topology, recording all directly connected neighboring nodes and their types.

[0101] For different types of node pairs, the calculation parameter matrices of the eight attention heads are initialized, and each attention head independently captures one aspect of the correlation characteristics between nodes. During attention calculation, the original feature vectors of the central node and each type of neighboring node are respectively transformed through the corresponding type of linear transformation matrix to generate a query vector, a key vector, and a value vector. The query vector and the key vector are scaled and dot-producted to obtain the original attention score. The original score is normalized using the softmax function to obtain the attention weight that reflects the strength of the association between the node pairs. After each attention head independently calculates a set of attention weights, the results of the eight heads are spliced ​​and linearly transformed to obtain the final multi-head attention weight matrix.

[0102] A dedicated feature transformation matrix is ​​defined for each node type, with the matrix dimensions matching the node's original feature dimensions. For device nodes, for example, multidimensional raw features, including runtime, load factor, maintenance history, and vibration data, are collected to form a feature vector. This feature vector is multiplied by the node-specific transformation matrix to obtain an initial feature representation that incorporates the node's type information.

[0103] For each node, its different types of neighboring nodes are processed separately. The type-aware features of the neighboring nodes are weighted based on the obtained multi-head attention weights. The feature vectors of the neighboring nodes are multiplied by the corresponding attention weights, and then the weighted features of all neighbors are summed to obtain a converged representation of the neighborhood information. The node's own type-aware features are concatenated with the neighborhood's converged features and processed using the nonlinear activation function LeakyReLU to extract fault features and generate a fault feature vector. This resulting feature preserves the node's own information while incorporating the influence of different types of neighboring nodes.

[0104] Perform global feature statistics on the entire heterogeneous information network, calculating topological features such as the average node degree, clustering coefficient, and eigenvalue distribution. The mean, variance, and distribution characteristics of equipment operating parameters are calculated, and the temporal and spatial distribution patterns of faults are analyzed. These global statistical features are then combined with the fault feature vector of each node to generate a comprehensive feature vector.

[0105] A three-layer perceptron network was constructed for importance assessment. The input layer dimensions matched the comprehensive feature vector. The first hidden layer used 256 neurons, and the second hidden layer used 128 neurons. Both hidden layers used the ReLU activation function to implement nonlinear transformations, and the Dropout mechanism was applied to prevent overfitting. The output layer used a single neuron with a sigmoid activation function, and the result was mapped to a range of 0–1 as the node importance score.

[0106] Nodes are ranked based on their importance scores, and the top 30% of nodes are selected as key nodes. For each key node, its second-order neighbors (neighbors of neighbors) are collected to construct a local network structure for subsequent feature aggregation. This importance-based sampling ensures that subsequent analysis focuses on key areas of the network.

[0107] For each local network structure, an affinity matrix is ​​constructed between nodes. The cosine similarity of the feature vectors between pairs of nodes is calculated and mapped to the 0-1 range using a Gaussian kernel function. Nodes are soft-assigned based on the affinity matrix, with features of nodes with high similarity assigned greater weights and aggregated to generate feature representations of several cluster centers.

[0108] A gated recurrent unit (GRU) is used to process temporal feature sequences, with features from 30 consecutive time steps fed into the GRU. At each time step, a reset gate determines how much historical information to forget based on the current input and the previous hidden state; an update gate determines the degree to which the current input information is incorporated. The final output is a feature representation that incorporates both historical information and the current state. To improve temporal modeling, a bidirectional GRU structure is employed, taking into account both forward and reverse temporal dependencies.

[0109] By combining the feature representations at different time steps through an attention mechanism, we obtain a subgraph representation that reflects the dynamic evolution of the local network. This hierarchical feature aggregation and time series modeling approach effectively captures the multi-scale spatiotemporal correlation patterns in power lines.

[0110] For example, the analysis process of the No. 1 tower node on a 500kV transmission line is as follows:

[0111] 1. Attention Calculation: This tower is connected to Tower 2 (equipment node), the temperature sensor (environmental node), and the primary insulator fallout record (fault node). Three sets of attention weights are calculated: For Tower 2, considering the similarity of their operating parameters and physical distance, a dot product operation yields an initial score of 0.85, which is normalized to 0.8. For the temperature sensor, a score of 0.65 is calculated based on the impact of temperature on equipment performance, which is normalized to 0.6. For the fault record, a score of 0.45 is calculated based on the correlation with the equipment status at the time of the fault, which is normalized to 0.4.

[0112] 2. Feature processing: The original features of the tower include values ​​such as [operating time: 5 years, load rate: 85%, number of maintenances: 3 times]. After conversion matrix mapping and neighbor feature weighting, the fault feature vector is obtained, which reflects the comprehensive information of equipment status, environmental impact and historical faults.

[0113] 3. Importance Assessment: The tower fault characteristics are combined with the overall statistical characteristics of the line (average load rate of 75%, average failure rate of 0.8 times / year, etc.). The importance score of 0.85 is calculated through a multi-layer perceptron, indicating that the tower is a key monitoring point.

[0114] 4. Feature aggregation: The tower and nearby important towers are combined into a local structure. The state sequence of the last 30 days is processed through a gated recurrent unit to generate a feature representation that reflects the evolution of the health status of the section.

[0115] The theoretical basis for the solution of this embodiment stems from recent research findings in the field of graph neural networks, particularly the cross-application of heterogeneous graph neural networks and time series graph modeling. Currently, power system fault prediction primarily utilizes traditional CNN time series methods, unidirectional GRU methods, and differentiable pooling methods. Traditional CNN time series methods segment power equipment data into time windows and convert them into two-dimensional matrices for processing. While they can capture local time series patterns, they struggle to model long-range dependencies and fail to effectively represent topological relationships between devices. This results in limited warning times, such as only providing a 21.5-hour lead time for insulator fallout. Unidirectional GRU methods directly process parameter sequences and can capture time series dependencies to a certain extent, but they ignore the spatial topological structure of the power network. While this improves prediction lead time, it remains limited, such as only providing a 39.1-hour lead time for equipment aging failures. Differentiable pooling methods extract network features through graph convolution and pooling operations, effectively capturing topological relationships between devices. However, they often treat different node types as homogeneous and lack sufficient modeling of time series evolution. For example, they only provide a 27.8-hour lead time for ground faults.

[0116] In response to the above limitations, the solution of this embodiment proposes a comprehensive improvement method based on differentiable pooling of heterogeneous information networks and bidirectional GRU time series modeling. First, unlike the traditional method of simplifying the power system into a homogeneous graph, this solution clearly distinguishes different types of nodes such as transmission equipment, environmental sensors, historical fault records, etc., and constructs a special attention calculation mechanism for each type of node pair. This design is based on the understanding that the power system is essentially a complex system composed of multiple heterogeneous entities, and there are significant differences in the interaction patterns between different types of nodes. Secondly, this solution designs special feature conversion matrices for different types of nodes, and performs differentiated aggregation of neighborhood information based on dynamically calculated attention weights. Taking into account that different types of power system components have different working characteristics and failure modes, their features need to be mapped to a feature space suitable for heterogeneous network learning through different transformations.

[0117] This solution also combines node features and global network characteristics, evaluates node importance through a multi-layer perceptron, and samples and constructs a local network based on this. This design stems from the understanding that not all nodes in the power system are equally important for fault prediction. Focusing on key nodes and their local structures can reduce noise interference and improve computational efficiency. The most innovative is the organic combination of the differentiable pooling operation in the spatial domain and the bidirectional GRU model in the time domain to construct a spatiotemporal fusion feature representation. This is because power system faults are usually complex events with spatiotemporal correlations, with both spatial topological propagation characteristics and temporal evolution laws. The bidirectional GRU can simultaneously consider historical trends and possible future developments, while the differentiable pooling effectively aggregates the structural information of the local network.

[0118] This significant increase in advance warning time brings significant practical value to power system maintenance, enabling operations teams to pre-deploy resources and plan optimal maintenance strategies, thereby reducing outage duration, lowering maintenance costs, and improving grid reliability. In particular, for severe faults such as insulator loss and conductor breakage, this solution can provide approximately two days of advance warning, fully meeting the time requirements for on-site repairs. This solution's balanced performance across different fault types demonstrates its robust modeling and generalization capabilities for diverse power system fault modes, providing a powerful decision-making support tool for preventive maintenance of power systems.

[0119] like Figure 2 The figure shows the advance warning time performance of different network structure feature aggregation and time series modeling methods for power line fault prediction. The horizontal axis represents different types of power line faults, and the vertical axis represents the average time (in hours) that the system can predict a fault. The numbers above the data points indicate the advance warning time achieved by the corresponding method for that fault type. The figure clearly shows that our solution (which combines differentiable pooling with a bidirectional GRU) performs best for all fault types. In particular, for insulator fall-off faults, our solution can predict the fault 47.3 hours (approximately 2 days) in advance, nearly 26 hours earlier than the traditional CNN time series method (21.5 hours) and approximately 15 hours earlier than the method using only differentiable pooling (32.6 hours). For conductor breakage faults, our solution achieves a warning time of 42.8 hours, 9.3 hours longer than the next best method (using only differentiable pooling). In lightning fault prediction, while the warning times of various methods are generally short (due to the sudden nature of lightning events), this technical solution still achieved an 11.6-hour lead time, outperforming other methods. For three types of progressive faults: equipment aging, ground faults, and connector damage, this technical solution achieved lead times of 58.7 hours, 38.2 hours, and 45.6 hours, respectively, maintaining a significant advantage. Particularly noteworthy is that for equipment aging, this technical solution achieved an astonishing 58.7 hours (over 2 days) of warning time, nearly 30 hours earlier than traditional methods, which has significant practical value for scheduling maintenance personnel and resources. Overall, by combining the advantages of differentiable pooling (effectively aggregating local network structure information) and bidirectional GRU (capturing temporal dependencies), this technical solution significantly improved the lead time for fault prediction, averaging 24.3 hours earlier than traditional methods, providing ample response time for power line maintenance.

[0120] In this embodiment, a multi-head attention mechanism is used to dynamically weight different types of neighbor nodes, so that the model can fully explore the multi-dimensional correlation features between various types of nodes, thereby improving the granularity and accuracy of feature expression; by multiplying the original features with the exclusive transformation matrix, a type-aware feature representation is generated, thereby better retaining the specific information of each type of node, effectively enhancing the complementarity of the node's own information and the neighborhood information; combining global statistical features with local key node sampling, it not only reflects the overall network topology characteristics, but also focuses on local key areas, so that the model has a global perspective when processing large-scale networks, and can accurately capture local anomalies or important patterns; through importance-based sampling and subsequent affinity matrix soft allocation and differentiable pooling operations, efficient aggregation of neighborhood features is achieved, redundant information is reduced, and data utilization efficiency is improved; the use of bidirectional GRU and attention mechanism to model continuous time series features can capture the multi-scale spatiotemporal evolution laws in power lines, providing a more reliable basis for fault prediction and maintenance decisions.

[0121] In an optional embodiment, the power line system is modeled as a directed acyclic graph, the transmission path characteristic values ​​are calculated, the trunk transmission lines are identified, the risk sections are divided, and the risk sections are integrated into the basic hierarchical knowledge base. The optimized hierarchical knowledge base includes:

[0122] Constructing a directed acyclic graph of the power line system, the directed acyclic graph including a set of operating characteristic nodes and a set of power transmission edges, obtaining an operating characteristic value of each node in a training sample set, and setting the power transmission efficiency corresponding to each edge as the edge characteristic value;

[0123] Calculating the sum of the connected edge eigenvalues ​​of each running feature node in the directed acyclic graph to obtain a node importance value, determining an electric energy transmission path based on the node importance value, and calculating the product of the edge eigenvalues ​​of each electric energy transmission path to obtain a path eigenvalue;

[0124] Obtain the line load threshold of each operating characteristic node, identify the trunk transmission line that meets the load threshold constraint through the iterative capacity-adjusted maximum flow algorithm, and calculate the corresponding edge cut set load ratio;

[0125] Obtain the failure probability of each operating characteristic node and the set of downstream affected nodes, calculate the node risk value, modify the node risk value based on the load ratio of the edge cut set, and divide the risk segments using a spectral clustering algorithm;

[0126] The trunk transmission line is written into the skeleton layer of the basic hierarchical knowledge base, the protection level is divided based on the risk section, the path characteristic value and the node risk value are recorded, a hierarchical association relationship including the edge cut set load ratio is established, and the control strategy priority of the basic hierarchical knowledge base is updated based on the hierarchical association relationship to obtain an optimized hierarchical knowledge base.

[0127] In one specific embodiment, a directed acyclic graph (DAG) of a power line system is constructed. The DAG consists of a set of operating characteristic nodes and a set of power transmission edges. The operating characteristic nodes represent individual devices or nodes in the power system, such as transformers and switches. The power transmission edges represent the paths of power flow between nodes. The operating characteristic values ​​of each node can be obtained through historical data or real-time monitoring data. Common characteristic values ​​include voltage, current, and power. For each edge, the power transmission efficiency, as the characteristic value of the edge, can be determined by the ratio of the edge's rated capacity to the actual transmitted power.

[0128] The sum of the eigenvalues ​​of the connected edges of each operational characteristic node is calculated to determine the node's importance. This value reflects the node's criticality in power transmission; a higher value indicates a greater impact on the overall power system. Based on the node's importance, the power transmission path is determined. A power transmission path consists of a series of connected nodes and edges. The product of the edge eigenvalues ​​of each power transmission path represents the path's eigenvalue. A higher eigenvalue indicates a higher transmission efficiency.

[0129] During the identification of trunk transmission lines, the line load threshold is determined for each characteristic operating node. This threshold represents the maximum load a node can withstand under normal operation. Using an iterative capacity-adjusted maximum flow algorithm, trunk transmission lines that meet the load threshold constraint are identified. By continuously adjusting the flow rate on each edge, paths that maximize power flow are sought. After identifying the trunk transmission lines, the corresponding edge cut set load ratio is calculated. An edge cut set is a set of edges that effectively separate power flows under specific conditions. The load ratio is the ratio of the load on an edge in this set to its carrying capacity.

[0130] To assess node risk, we obtain the failure probability and downstream impact node set for each operating characteristic node. The failure probability refers to the likelihood of a node failing within a specific timeframe, while the downstream impact node set refers to the other nodes affected by the node's failure. The calculated node risk value reflects the impact of a node failure on the entire system. The node risk value is then corrected based on the edge cut set load ratio to more accurately reflect the node's risk profile under varying load conditions.

[0131] To identify risk segments, a spectral clustering algorithm is used. By clustering node risk values, nodes with similar risks are grouped together, forming different risk segments. Each risk segment can then be tailored to its specific characteristics, enabling the development of a corresponding protection strategy to reduce overall system risk.

[0132] The identified trunk transmission lines are stored in the skeleton layer of the basic hierarchical knowledge base, a multi-layered knowledge management system that encompasses the system's basic structure and main components. Protection levels are divided based on risk zones, path characteristics and node risk values ​​are recorded, and hierarchical relationships are established, including edge-cut set load ratios. These relationships help understand the interactions between individual nodes and edges in the power system.

[0133] The control strategy priorities within the underlying hierarchical knowledge base are updated based on hierarchical relationships. Control strategy priorities refer to the nodes and paths that are prioritized in the event of a fault or anomaly. By optimizing the hierarchical knowledge base, the stability and reliability of the power system can be improved, ensuring a rapid response and appropriate measures when problems arise.

[0134] For example, let's take a power company as an example. Assume that the company has a power line system with ten nodes. The operating characteristic value of each node is as follows:

[0135] Node A: voltage 220V, current 100A, power 22kW;

[0136] Node B: voltage 220V, current 150A, power 33kW;

[0137] Node C: voltage 220V, current 80A, power 17.6kW;

[0138] Node D: voltage 220V, current 120A, power 26.4kW;

[0139] Node E: voltage 220V, current 90A, power 19.8kW;

[0140] Node F: voltage 220V, current 110A, power 24.2kW;

[0141] Node G: voltage 220V, current 130A, power 28.6kW;

[0142] Node H: voltage 220V, current 70A, power 15.4kW;

[0143] Node I: voltage 220V, current 140A, power 30.8kW;

[0144] Node J: voltage 220V, current 160A, power 35.2kW.

[0145] Using this data, we can calculate the sum of the edge eigenvalues ​​of each node, and thus the node's importance. Assume that node A is connected to nodes B and C, node B is connected to nodes D and E, and so on. Ultimately, the power transmission path is determined.

[0146] When identifying the trunk transmission line, assuming the load threshold for node A is 150A and the load threshold for node B is 200A, the maximum flow algorithm identifies the trunk line as ABD. The edge-cut set load ratio is 0.8, indicating the stability of the line under load.

[0147] By obtaining the node's failure probability and the set of downstream impacting nodes, assuming that the failure probability of node A is 0.01 and the impacting node set is {B, C}, the risk value of node A is 0.02. Using the spectral clustering algorithm, the nodes are divided into high-risk and low-risk segments.

[0148] The identified trunk transmission lines and risk section information are written into the basic hierarchical knowledge base, and the priority of the updated control strategy is optimized to ensure the efficient operation of the power system.

[0149] In an optional embodiment, obtaining a line load threshold of each operating characteristic node, identifying a trunk transmission line that meets the load threshold constraint through an iterative capacity-adjusted maximum flow algorithm, and calculating a corresponding edge-cut set load ratio includes:

[0150] Obtain the line load threshold of each operating characteristic node in the power system network, construct an initial capacity matrix, and use the maximum flow algorithm to obtain the initial maximum flow value under the constraint that the actual load is less than or equal to the line load threshold. Based on the initial maximum flow value, determine the corresponding minimum cut set to obtain the initial trunk transmission line;

[0151] Apply a disturbance coefficient to the initial trunk transmission line, modify the initial capacity matrix to obtain a modified capacity matrix, calculate the maximum flow value after the disturbance, and calculate the ratio of the maximum flow value to the initial maximum flow value to obtain the network fitness index;

[0152] Based on the difference between the network fitness index and the preset fitness threshold, the modified capacity matrix is ​​updated to obtain the iterative capacity matrix. The iterative maximum flow value corresponding to the iterative capacity matrix is ​​calculated. The iteration is repeated until the rate of change of the maximum flow values ​​of two adjacent iterative cycles is less than the preset rate of change threshold. The final capacity matrix is ​​obtained. From the final capacity matrix, lines that meet the line load threshold are selected as the trunk transmission optimization lines, and the corresponding edge cut set load ratio is calculated.

[0153] Construct the network topology of the power system, setting substations as network nodes and transmission lines as network edges. For each transmission line, collect the following operating parameters: rated capacity, real-time load, historical maximum load, voltage level, line length, conductor type, and other basic data. Based on historical operating data, calculate the average load level and load fluctuation range for each line. Considering equipment safety margins and system stability requirements, set graded load thresholds: for 500kV trunk lines, 80% of the rated capacity is used as the threshold; for 220kV regional lines, 75% of the rated capacity is used as the threshold; for 110kV and below distribution lines, 70% of the rated capacity is used as the threshold.

[0154] When constructing the initial capacity matrix, the matrix dimensions are total number of nodes × total number of nodes. Matrix elements represent the load thresholds of the lines between corresponding nodes, and matrix elements between unconnected nodes are zero. Based on the direction of power system flow, the capacity matrix is ​​converted into a directed graph structure, with power supply-side nodes designated as sources and load-side nodes as sinks.

[0155] Using the Ford-Fulkerson maximum flow algorithm, a residual network is constructed, with the flow on all edges initialized to 0. Using a breadth-first search, the network nodes are scanned layer by layer to find a feasible path from the source to the sink. For each feasible path, the residual capacity of all edges along the path is calculated, and the minimum value is selected as the augmented flow. The residual capacity of each edge in the residual network is updated, and the augmented flow is added to the total flow. The search-augmentation process is repeated until no new feasible paths can be found. The total flow at this point is the initial maximum flow value.

[0156] Starting from the source, all nodes reachable through the residual network are marked as set S; unmarked nodes are grouped into set T. All edges connecting set S and set T constitute a minimum cut set, and the transmission lines corresponding to these edges are the initial trunk transmission lines. Information such as the load threshold, actual load, and node connectivity of these lines is also recorded for subsequent optimization.

[0157] Design a multi-level perturbation test scheme, including single-line perturbations, combined perturbations on linked lines, and regional perturbations. For each trunk line, construct a perturbation coefficient sequence within the range of [-5%, +5%], with a perturbation step size of 1%. For linked line groups, consider the load transfer effect and design a complementary perturbation scheme.

[0158] For each trunk line, its initial capacity is multiplied by the corresponding disturbance coefficient to obtain a revised capacity value. Load transfer constraints between lines are also considered to ensure that the revised capacity distribution meets system stability requirements. For non-trunk lines, corresponding capacity adjustment coefficients are set based on their degree of correlation with the trunk line.

[0159] Under the modified capacity constraint, the maximum flow algorithm is repeated, using the same Ford-Fulkerson algorithm as the initial calculation, but with the modified capacity values. The maximum flow value for each disturbance is recorded and compared with the initial maximum flow value to calculate the network fitness index. Statistical analysis of the results of multiple disturbance tests is performed to obtain an evaluation index of the system's adaptability to different types of disturbances.

[0160] Based on the network fitness assessment results, a capacity optimization strategy is established. When the fitness index falls below the threshold, critical line capacity is prioritized; when the index rises above the threshold, non-critical line capacity is considered reduced to achieve optimal resource allocation. The specific steps of each iterative adjustment include: calculating the fitness difference, determining the adjustment target, generating an adjustment plan, verifying the feasibility of the plan, and updating the capacity matrix.

[0161] The maximum flow rate change rate between two consecutive iterations is calculated, while simultaneously monitoring convergence indicators across multiple dimensions, such as the capacity adjustment range and the trend of network fitness indicators. The optimization process is considered converged when all indicators meet the preset threshold requirements.

[0162] Based on the final capacity matrix, select lines that meet load threshold constraints and have high network importance. Calculate the characteristics of the edge cut sets formed by these lines, including: cut set capacity ratio, load distribution uniformity, reliability index, etc.

[0163] As an example, the optimization process for the main transmission lines of a specific power grid system, 500kV backbone grid, is described in detail. This grid system consists of 12 500kV substations and 18 transmission lines. These include three power supply nodes (S1 thermal power plant, S2 hydropower plant, and S3 nuclear power plant), four load center nodes (S9 to S12), and five relay transmission nodes (S4 to S8). In the system's initial operating state, the operating parameters of the main transmission lines L1 to L5 connecting the power source side to the relay node vary: Line L1, connecting the thermal power plant, has a rated capacity of 2400MW and an actual operating load of 1850MW; Line L2, connecting the hydropower station, has a rated capacity of 2200MW and an actual load of 1680MW; Line L3, connecting the nuclear power plant, has the largest rated capacity of 2600MW and a relatively high actual load of 2050MW; while Lines L4 and L5, connecting the relay nodes, have relatively small rated capacities of 2000MW and 1800MW, respectively, and correspondingly lower actual loads of 1520MW and 1380MW, respectively.

[0164] In the first round of iterative optimization, the system first determined the source point set and the sink point set, and calculated the initial maximum flow value of 5580MW through the maximum flow algorithm, and identified L1, L2, and L3 as key transmission channels. To test the adaptability of the system, different degrees of capacity disturbance were imposed on these three key lines: L1 increased by 5%, L2 increased by 3%, and L3 increased by 4%. After the disturbance, the maximum flow of the system increased to 5750MW, and the network fitness index reached 1.03. Based on the test results, the capacity matrix was adjusted for the first time: while increasing the capacity of key lines, the capacity configuration of non-trunk lines was appropriately reduced. Among them, the capacity of L1 increased to 1968MW, L2 increased to 1787MW, and L3 increased to 2122MW, while the capacity of L4 and L5 decreased to 1584MW and 1426MW, respectively.

[0165] In the second iteration, the system's maximum flow rate further increased to 5680MW, a 1.79% increase from the initial state. Perturbation tests within a ±4% range showed that the system's maximum flow rate could fluctuate between 5580MW and 5820MW, with the network adaptability maintained at 1.02. Based on this, capacity optimization continued: L1 increased by another 1.5% to 1998MW, L2 and L3 increased by 1% and 1.2% to 1805MW and 2147MW, respectively. Meanwhile, L4 and L5 capacities were slightly reduced to 1571MW and 1419MW.

[0166] The third iteration saw significant changes. The system's maximum flow reached 5720 MW, with the increase reduced to 0.70%. The minimum cut set was expanded to include the L6 line. The perturbation test range was reduced to ±3%, and the maximum flow fluctuation range narrowed to between 5650 MW and 5780 MW. Based on the test results, smaller adjustments were made to the trunk lines, keeping the capacity increases from L1 to L3 within 1%. The capacity of the newly included L6 line was also increased by 1%.

[0167] After entering the fourth iteration, system optimization began to show a clear convergence trend. The maximum flow value increased only slightly to 5735MW, and the rate of change dropped to 0.26%. Perturbation testing showed that the system operation was more stable, and the fluctuation range of the maximum flow value further narrowed to 5700MW to 5760MW. At this time, the capacity adjustment range of each line did not exceed 0.5%. By the fifth iteration, the system's maximum flow value reached 5742MW, and the rate of change between two consecutive rounds dropped to 0.12%, below the preset 0.2% threshold, indicating that the optimization process had reached convergence.

[0168] After five rounds of iterative optimization, the system has formed a clear layered transmission architecture: L1, L2, and L3 form the core backbone, L6 and L8 serve as important support, and L12 and L15 serve as local key channels. In terms of the edge cut set load ratio, L3 reaches the highest, reaching 0.95, making it the system's most critical transmission channel; L1 is second at 0.92, and L2 is 0.88, both at a relatively high level; L6 and L8 have load ratios between 0.80 and 0.83, providing good support. Overall, the optimized system's maximum transmission capacity has increased by 2.9%, the load distribution of key lines has become more balanced, and the network fitness has remained stable above 0.95, achieving the expected optimization goal.

[0169] Optimization, analysis, and enhancement of power system trunk transmission lines are crucial for the efficient operation of current power grids. Traditional methods for identifying power grid trunk lines rely primarily on static topology analysis and expert judgment, such as power flow marginal sensitivity analysis and N-1 security constraint analysis. These methods are inadequate for complex scenarios such as dynamic load fluctuations and the integration of new energy sources. Existing network flow theory, while widely used in communications and transportation networks, is limited in its application to power systems by the specific physical constraints of power flows.

[0170] Traditional methods typically treat line capacity as a fixed parameter, ignoring the coupling relationships between system operating parameters and their potential for dynamic adjustment. For example, analysis methods based on node importance focus on node connectivity and centrality but cannot accurately characterize load transfer relationships between lines. Methods based on power flow sensitivity can identify critical lines but struggle to quantify the impact of line capacity adjustments on overall system performance. Furthermore, most existing optimization methods employ a single objective function, making it difficult to balance the multi-dimensional requirements of system safety, economy, and flexibility.

[0171] The steps in this embodiment are based on the network maximum flow-minimum cut theory and introduce an iterative capacity matrix perturbation and adaptive optimization mechanism. Compared with traditional methods, the main improvements are reflected in three aspects: first, it implements hierarchical load threshold setting, differentially configuring safety margins based on line voltage level and importance, which better meets actual operational needs; second, it designs a multi-level perturbation testing system, which reveals the quantitative relationship between line capacity adjustment and the system's maximum transmission capacity through systematic perturbation analysis, overcoming the subjectivity of traditional methods in line importance assessment; third, it establishes an iterative optimization framework based on network fitness indicators, achieving adaptive adjustment and dynamic convergence of capacity allocation.

[0172] The starting point for these improvements stems from a deeper understanding of the inherent characteristics of power systems. As critical infrastructure, power systems must maintain stable and reliable operation while efficiently utilizing existing resources to cope with load growth. Traditional static safety margin configurations often lead to wasted resources, while overly aggressive optimization can compromise system security. This approach, by organically integrating network flow theory with the characteristics of power systems, explores a technical approach to improving system transmission capacity while ensuring safety.

[0173] Practice has demonstrated that this embodiment can effectively identify key trunk lines in the power grid and scientifically and rationally adjust line load thresholds. In the test system, the maximum flow value increased by 2.96% after optimization, significantly improving the system's transmission capacity under the same network topology. Core backbone lines and load aggregation lines were accurately identified as key edge cut sets, providing clear guidance for operation, maintenance, and investment decisions. Furthermore, the network fitness index reached 0.97, demonstrating that the optimized system has strong anti-disturbance capabilities and operational resilience.

[0174] Compared to traditional methods, this implementation is more practical and adaptable. It applies not only to static planning scenarios but also supports real-time operational adjustments. It not only focuses on optimizing a single line but also considers overall system coordination. It not only provides optimization results but also reveals the dynamic characteristics of the system during the optimization process. This comprehensive technical solution provides new analytical tools and decision support for power system planning and operation, and is of great value in improving grid resource utilization efficiency and operational reliability.

[0175] like Figure 3The figure shows the network topology and load distribution of the optimized 500kV power grid system's trunk transmission lines. The entire power system consists of 12 nodes and 18 lines, including three power source nodes (circles), five relay transfer nodes (diamonds), and four load center nodes (squares). The thickness of the lines in the figure represents their importance in the system and load level, and the line labels display "Actual Load / Optimized Threshold (MW)." Through optimization, the system has formed a clear hierarchical transmission architecture: L1, L2, and L3 lines form the core backbone (the thickest lines), connecting the three power source nodes (S1 thermal power plant, S2 hydropower plant, and S3 nuclear power plant) to the relay substation. Their load ratios reach 92.6% (1850 / 1998 MW), 93.1% (1680 / 1805 MW), and 95.5% (2050 / 2147 MW), respectively, fully utilizing the transmission channel capacity. Lines L6, L8, L10, L12, and L15 form a critical support layer (medium-load lines), with load ratios ranging from 83.8% to 95.1%. L6 (1450 / 1730 MW) and L8 (1320 / 1580 MW) serve as key connections between relay nodes, supporting efficient energy allocation in the central part of the system. The remaining lines form a general transmission layer and a low-load backup layer, providing diverse path options for the system. Overall, the optimized system exhibits distinct "hub-and-spoke" structural characteristics, with the S8 substation as the core hub, connecting upstream power sources with downstream loads. Through this optimized configuration, the system's maximum flow value increased from the initial 5580MW to 5745MW, an increase of 2.96%, and the network fitness index reached 0.97. At the same time, the main edge cut set composed of L1+L2+L3 and the secondary edge cut set composed of L9+L10+L11 both achieved a high level of load balancing, indicating that this optimization scheme not only improves the overall transmission capacity of the system, but also enhances the system's stability in the face of complex operating conditions.

[0176] In this embodiment, through the maximum flow algorithm and minimum cut set analysis, the trunk transmission lines in the power network can be accurately identified, providing a clear basis for subsequent optimization; the disturbance coefficient and fitness index are introduced to realize the iterative update of the capacity matrix, so that the optimization process can adaptively adjust the network status and improve transmission efficiency and reliability; under the constraint of the line load threshold, the maximum flow value is repeatedly iteratively calculated to optimize the network load distribution, thereby reducing the risk of local overload and enhancing system stability; the edge cut set load ratio of the trunk line is calculated to further quantify the load situation of the key transmission path of the network, providing data support for power system operation and maintenance decisions.

[0177] In an optional embodiment, based on the optimized hierarchical knowledge base and the trunk transmission line, an inspection strategy model is established based on a reinforcement learning method, and the adaptive generation of an inspection execution plan includes:

[0178] Based on the fault characteristics in the optimized hierarchical knowledge base, the mapping relationship between the fault characteristics and the operating status, and the transmission path characteristic values ​​of the trunk transmission line, a regional characteristic function is constructed, and the boundary of the dynamic associated sub-region is determined according to the regional characteristic function;

[0179] A local state evaluator is constructed for the dynamically associated sub-regions. The risk factor score of the local state evaluator is multiplied by the weight coefficient to obtain the regional risk degree. The regional risk degree is then combined with the inspection time interval and load level to calculate the regional inspection urgency.

[0180] Constructing a feature similarity matrix of dynamically associated sub-regions, generating a regional knowledge transfer function based on the feature similarity matrix, and fusing the regional knowledge transfer function with historical inspection experience to obtain regional shared experience;

[0181] The regional inspection urgency, regional risk, and regional shared experience are used to construct a state vector, the inspection priority parameter is used to construct an action vector, and a reward function is established based on the state vector and the action vector;

[0182] A reinforcement learning method is used to optimize the reward function to obtain strategy model parameters, and an inspection execution plan is generated according to the strategy model parameters.

[0183] An optimized hierarchical knowledge base is constructed, containing mappings between fault characteristics, operating status, and transmission path characteristic values ​​for trunk transmission lines. By analyzing historical fault data, characteristics of different fault types are extracted and associated with the equipment's operating status. Operating status includes parameters such as temperature, pressure, and current. Regional characteristic functions are then constructed based on these characteristics, effectively reflecting the operating status and fault risk of different regions.

[0184] Based on the regional characteristic function, the boundaries of dynamically associated subregions are determined. Dynamically associated subregions are areas where fault characteristics are highly correlated with operating status under specific conditions. By monitoring operating data in real time and incorporating changes in fault characteristics, the boundaries of subregions are dynamically adjusted to ensure the effectiveness of inspection strategies.

[0185] For each dynamically associated sub-region, a local status evaluator is constructed. Its primary function is to assess the risk of the local region. The regional risk level is determined by combining the risk factor scores with weight coefficients. Risk factors include equipment failure history and the degree of abnormality in the current operating state. The regional risk level reflects the inspection priority of the region under its current state.

[0186] Based on this, the regional risk level is combined with the inspection interval and load level to calculate the urgency of regional inspections. The inspection interval refers to the time between the last inspection and the current moment, while the load level refers to the equipment workload at the time of the inspection. By comprehensively considering these factors, inspection plans can be more rationally arranged to ensure that high-risk areas receive timely inspections.

[0187] A feature similarity matrix for dynamically associated sub-regions is further constructed. This matrix compares the features of each region and assesses their similarity. Regions with high similarity can share inspection experience, thereby improving inspection efficiency. A regional knowledge transfer function is generated, integrating the feature similarity matrix with historical inspection experience to obtain regional shared experience. This process effectively utilizes existing inspection data, reduces duplication of effort, and improves the scientific nature and accuracy of inspections.

[0188] The state vector is constructed by integrating regional inspection urgency, regional risk, and regional shared experience. This vector contains all important information about the current inspection task and provides a basis for subsequent decision-making. Furthermore, the inspection priority parameter is constructed as an action vector, representing the different inspection strategies that can be adopted under the current state.

[0189] Based on the constructed state vector and action vector, a reward function is established. This reward function is used to evaluate the effectiveness of different inspection strategies and guide the reinforcement learning algorithm to optimize these strategies. By using feedback from historical inspection data, the parameters of the reward function are adjusted to more accurately reflect the strengths and weaknesses of the inspection strategies.

[0190] Reinforcement learning is used to optimize the reward function and determine the policy model parameters. The reinforcement learning algorithm adjusts the policy model parameters through trial and error to achieve the optimal inspection strategy. Through multiple rounds of learning and optimization, a highly adaptable and efficient inspection execution plan is ultimately generated.

[0191] For example, a power company's substation experienced multiple equipment failures over the past year, including transformer overheating and switchgear malfunctions. By analyzing historical fault data, an optimized hierarchical knowledge base containing fault characteristics and operating conditions was constructed. Based on this knowledge base, the boundaries of the dynamically associated subregions were determined to be specific areas around the transformers.

[0192] During actual inspections, the local state evaluator scored the risk factor for this area, determining a risk score of 0.8, an inspection interval of 30 days, and a load level of 70%. Combining these data, the area's inspection urgency was calculated to be high. The feature similarity matrix showed that this area was highly similar to neighboring areas, allowing for the sharing of historical inspection experience.

[0193] Based on the state vector and action vector, the established reward function guides the reinforcement learning algorithm to optimize the policy model parameters. The generated inspection execution plan is to conduct a special inspection in the area once a week to ensure that high-risk equipment receives timely attention.

[0194] In an optional embodiment, constructing a feature similarity matrix of dynamically associated sub-regions, generating a regional knowledge transfer function based on the feature similarity matrix, and fusing the regional knowledge transfer function with historical inspection experience to obtain regional shared experience includes:

[0195] The dynamically associated sub-regions are constructed as a graph network model, where each dynamically associated sub-region corresponds to a region node, and the region node contains fault characteristics and operating status data. Edge connections are constructed based on the physical connection relationship between the dynamically associated sub-regions, and the topological association features between the region nodes are extracted through a graph convolutional network to obtain a feature similarity matrix;

[0196] Calculating the similarity between adjacent regional nodes to obtain a gating coefficient, and constructing a regional knowledge transfer function based on the gating coefficient and the feature similarity matrix, wherein the regional knowledge transfer function represents the knowledge transfer relationship between regional nodes;

[0197] Acquire features of the same dynamically associated sub-region in different time windows to construct positive sample pairs, acquire features of different dynamically associated sub-regions to construct negative sample pairs, and train a feature extraction model using a comparative loss function to obtain historical inspection experience;

[0198] The regional knowledge transfer function is integrated with the historical inspection experience to obtain regional shared experience.

[0199] In one specific implementation, an adaptive clustering algorithm is used to divide dynamically associated sub-regions based on the physical connectivity and electrical characteristics of the devices. First, the electrical distance between any two monitoring points is calculated, including impedance distance and power flow coupling. An affinity matrix is ​​then constructed based on this electrical distance, and preliminary zoning is performed using spectral clustering. Finally, the zoning results are manually corrected and optimized, taking into account factors such as geographic location and operating environment. The boundaries of each sub-region are dynamically adjusted based on real-time operating status, ensuring that the zoning consistently reflects the current system operating characteristics.

[0200] When building a graph network model, each dynamically associated subregion corresponds to a regional node. The node's feature vector consists of multiple components: operational status data includes statistical characteristics of electrical parameters such as voltage level, load factor, and power factor; equipment status data includes temporal characteristics of physical parameters such as temperature distribution, vibration characteristics, and insulation properties; and fault characteristic data includes historical statistical information such as fault frequency, type distribution, and severity. Edge connections between nodes are established based on physical topological relationships, and edge attributes include information on multiple dimensions such as line impedance, power flow magnitude, and fault correlation.

[0201] A multi-layer graph convolutional network (GCN) is used to extract topological correlation features between nodes. The first layer of graph convolution performs spatial feature aggregation, combining the features of each node with those of its first-order neighbors. Weight coefficients are calculated using an attention mechanism, taking into account the similarity of node features and the importance of edge attributes. The second layer of graph convolution expands the receptive field, aggregating information from second-order neighbor nodes to capture a wider range of correlation patterns. The third layer of graph convolution introduces skip connections to fuse feature representations at different levels. After multiple layers of nonlinear transformation and feature aggregation, a feature similarity matrix is ​​ultimately obtained, reflecting the deep correlations between regional nodes.

[0202] Calculating the strength of knowledge transfer between regional nodes requires comprehensive consideration of multiple factors. Based on the feature similarity matrix, the Euclidean distance and cosine similarity between node pairs are calculated to obtain an initial similarity score. Expert knowledge rules are then introduced to weight the similarity score based on factors such as regional operating environment similarity, device type consistency, and failure mode relevance. For pairs of nodes in adjacent regions, the gating coefficient is calculated between them, which determines the degree of knowledge transfer.

[0203] When constructing the regional knowledge transfer function, a dynamic weight matrix is ​​used. Each element of the matrix represents the strength of knowledge transfer from the source region to the target region. The matrix consists of three components: a basic transfer weight reflects the strength of the physical connection between regions; a similarity weight reflects the degree of similarity in regional characteristics; and a gating weight controls the selectivity of knowledge transfer. The combination of these three weights enables adaptive transfer of knowledge between regions. Furthermore, a time-series decay mechanism is designed to assign higher transfer weights to newer knowledge, ensuring the timeliness of knowledge transfer.

[0204] In the temporal dimension, for each dynamically associated subregion, feature sequences from different time windows are selected to construct positive sample pairs. The length of the time window is determined by the periodicity of the data and can be hourly, daily, or weekly. For each time window, a multidimensional feature vector is extracted, including device status, operating parameters, and environmental conditions. The construction of positive sample pairs takes into account temporal correlation, so feature sequences from adjacent time windows are more likely to form a positive sample pair.

[0205] In the spatial dimension, feature sequences from different subregions are selected to construct negative sample pairs. The selection of negative samples must ensure sufficient discrimination between samples, so pairs of regions with significant physical distance and operational differences are prioritized. Furthermore, given the cyclical nature of the power system, a sufficient number of negative sample pairs must be constructed at different time scales to ensure that the model learns stable feature representations.

[0206] The feature extraction model is trained using a contrastive learning framework. The designed loss function consists of two parts: a term that brings positive pairs of samples closer together, and a term that pushes negative pairs further apart. By minimizing this loss function, the model learns feature representations that maintain temporal continuity while reflecting spatial diversity. During training, a hard-sample mining strategy is employed to dynamically adjust sample weights, allowing the model to prioritize pairs of samples that are difficult to distinguish.

[0207] The trained feature extraction model is combined with the regional knowledge transfer function to generate regional shared experience. This process first uses the feature extraction model to encode each region's historical data, generating an experience vector that reflects the region's characteristics. The knowledge transfer function then calculates the strength of experience transfer between different regions. The resulting regional shared experience not only incorporates the unique operating characteristics of each region but also enables complementary experience between similar regions, providing a reliable knowledge foundation for subsequent inspection decisions.

[0208] For example, a 500kV transmission line system with a total length of 180 kilometers and six substations along the line is divided into four dynamically associated sub-areas based on equipment distribution and operational characteristics. Area A includes two substations and 50 kilometers of transmission lines, primarily located in mountainous terrain; Area B includes one substation and 40 kilometers of transmission lines, crossing plains; Area C includes two substations and 50 kilometers of transmission lines, passing through coastal areas; and Area D includes one substation and 40 kilometers of transmission lines, located in a densely industrialized area.

[0209] Each sub-area is equipped with intelligent sensors to continuously collect operational data. For example, in Area A, its node features include operating parameters such as current load factor (75%-85%), conductor temperature (45°C-55°C), and insulator contamination level (0.15-0.25) at five key monitoring points, as well as records of 12 equipment failures that occurred over the past three years. When constructing the graph network, Area A is connected to its adjacent Areas B and C by edges with weights of 0.8 and 0.6, respectively, reflecting the electrical distance between them.

[0210] Graph convolutional network analysis revealed that the feature similarity between regions A and B is 0.85. This is because both regions face common challenges posed by mountainous terrain, such as high lightning strike risk and severe mechanical stress. The calculated gating coefficient is 0.82, indicating strong knowledge transfer between the two regions. In contrast, the feature similarity between regions A and C is only 0.45, with a gating coefficient of 0.38, reflecting the significant differences in operating environments and fault characteristics between mountainous and coastal areas.

[0211] During the model training phase, we selected seasonal operational data from Region A to construct positive sample pairs, such as characteristic sequences from winter and spring, to capture seasonal variations. We also selected data from Regions A and C over the same period to construct negative sample pairs, helping the model identify characteristic differences across geographical environments. Through comparative learning, the model successfully extracted historical inspection experience that reflects the characteristics of each region.

[0212] The resulting regional shared experience shows that Areas A and B can share their experience in lightning protection and mechanical stress monitoring to optimize inspection strategies; Area C's experience in salt spray corrosion prevention is primarily used within its own region, with limited migration to Area D; and Area D's experience in industrial pollution prevention can provide some reference for other areas. This differentiated knowledge sharing mechanism significantly improves inspection efficiency, increasing fault warning accuracy by an average of 15% across all regions.

[0213] like Figure 4The tree-like knowledge flow structure shown in the figure visually illustrates the knowledge transfer process and its evolving effects between dynamically linked sub-regions. The thick black line in the center represents the timeline of system experience, from the initial state to 60 weeks. The size of the circular nodes directly reflects the growth of the accumulated experience of this technical solution, gradually increasing from an initial 57.2% to a final 96.3%. The square nodes on the right side of the figure also show the accumulated experience of the traditional method at the same time point, clearly showing slower growth, ultimately reaching only 77%. The four branches extending from the main trunk represent four typical inter-regional knowledge transfer paths, with the line thickness directly reflecting the strength of the knowledge transfer. The migration path in region A→B is the strongest, with a transfer efficiency of 0.90. This is primarily due to the high similarity in mountainous terrain characteristics between the two regions, ultimately contributing to a significant 19.8% improvement in fault warning capabilities. The migration path in region C→D is second, with a transfer efficiency of 0.75, primarily reflected in a 15.4% improvement in salt spray corrosion detection capabilities. In contrast, the A→C migration path is thinner, reflecting the differences in environmental characteristics between mountainous and coastal areas. Its migration efficiency is only 0.45, but even so, it still leads to a 12.3% improvement in inspection optimization. The figure visually demonstrates the accelerated accumulation of system experience over time. In particular, after 36 weeks, the gap between this technical solution and traditional methods widens further, highlighting the cumulative advantages of the knowledge transfer mechanism based on graph convolutional networks in long-term operation. The figure also clearly shows the positive correlation between higher knowledge transfer efficiency and significant performance improvements, confirming the effectiveness of the regional knowledge transfer function design in this technical solution.

[0214] In this embodiment, by constructing a graph network model and a graph convolutional network, the physical connections and topological features between dynamically associated sub-regions are effectively extracted, thereby improving the understanding and modeling capabilities of inter-regional relationships; the knowledge transfer function is constructed using gating coefficients and feature similarity to achieve effective transfer of experience and information between regions, providing support for fault prediction and operational decision-making; the feature extraction model is trained using positive and negative sample pairs and contrast loss functions to learn and extract key experiences from historical inspection data, thereby improving the model's ability to identify anomalies and faults; the regional knowledge transfer function is integrated with historical inspection experience to generate regional shared experience, achieve cross-regional information complementarity, and enhance the intelligent decision-making capabilities of the overall system.

[0215] The power line health status assessment and prediction system based on big data in an embodiment of the present invention includes:

[0216] The first unit is used to obtain historical operation data of the power line, pre-process the historical operation data according to preset data quality rules, and generate standardized operation data;

[0217] The second unit is used to extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, build a basic hierarchical knowledge base, automatically annotate the standardized operating data, and generate a training sample set;

[0218] The third unit is used to model the power line system as a directed acyclic graph, calculate the transmission path characteristic values, identify the trunk transmission lines, divide the risk sections, and integrate them into the basic hierarchical knowledge base to obtain an optimized hierarchical knowledge base;

[0219] The fourth unit is used to establish an inspection strategy model based on the optimized hierarchical knowledge base and the trunk transmission line based on the reinforcement learning method, and adaptively generate an inspection execution plan; execute the inspection execution plan and evaluate the health status of the power line to obtain the health status assessment result;

[0220] The fifth unit is used to generate a power line health status report based on the health status assessment result.

[0221] According to a third aspect of the embodiments of the present invention,

[0222] An electronic device is provided, comprising:

[0223] processor;

[0224] a memory for storing processor-executable instructions;

[0225] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0226] According to a fourth aspect of the embodiments of the present invention,

[0227] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0228] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0229] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power line health status assessment and prediction method based on big data, characterized by: include: Obtain historical operating data of power lines, pre-process the historical operating data according to preset data quality rules, and generate standardized operating data; Extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, build a basic hierarchical knowledge base, automatically annotate the standardized operating data, and generate a training sample set; The standardized operation data includes equipment data, environmental data and historical fault data; Based on standardized operation data, they are respectively constructed into equipment status nodes, environmental feature nodes and historical fault nodes; Based on the physical connection relationship between device status nodes, the topological association edge between devices is established. Based on the influence relationship between device status nodes and environmental feature nodes, the device-environment association edge is established. Based on the correspondence between device status nodes and historical fault nodes, the device-fault association edge is established. Together, they form a heterogeneous information network. For each node in the heterogeneous information network, type-aware weighted aggregation is performed based on a multi-head attention mechanism to extract fault features, obtaining a fault feature vector. Feature aggregation is then performed through differentiable pooling guided by node importance to obtain a subgraph feature representation. Performing a sliding convolution operation on the subgraph feature representation using a causal convolution kernel, generating a time series feature vector through time series information processing, calculating the contrast loss between normal operating state samples and the boundary loss between abnormal operating state samples, and obtaining a comprehensive anomaly score of the fault feature; The time series feature vector, the comprehensive anomaly score, and the corresponding fault case are stored in the instance layer, and a mapping relationship between the fault feature and the operating status is established; the fault cases whose Euclidean distance between the time series feature vectors corresponding to the fault cases is less than a first preset threshold are aggregated into fault modes and stored in the pattern layer; the evolution relationship between the fault modes whose time series correlation degree corresponding to the fault modes is greater than a second preset threshold is stored in the rule layer; The instance layer, the pattern layer and the rule layer constitute a basic hierarchical knowledge base; The power line system is modeled as a directed acyclic graph, the transmission path eigenvalues ​​are calculated, the trunk transmission lines are identified, the risk sections are divided, and the risk sections are integrated into the basic hierarchical knowledge base to obtain an optimized hierarchical knowledge base. Based on the optimized hierarchical knowledge base and backbone transmission lines, an inspection strategy model is established based on the reinforcement learning method to adaptively generate inspection execution plans; Execute the inspection plan and evaluate the health status of the power lines to obtain the health status evaluation results; Generate a power line health status report based on the health status assessment results.

2. The method according to claim 1, characterized in that For each node in the heterogeneous information network, type-aware weighted aggregation is performed based on the multi-head attention mechanism to extract fault features, obtaining a fault feature vector. Feature aggregation is performed through differentiable pooling guided by node importance, and the obtained subgraph feature representation includes: Constructing a multi-head attention weight matrix based on different types of neighbor nodes of each node in the heterogeneous information network, wherein the multi-head attention weight matrix is ​​used to calculate the dynamic association strength between the node and the different types of neighbor nodes; Multiplying the original feature vector of the node by the conversion matrix corresponding to the type of the node to obtain a type-aware feature vector, and performing a weighted aggregation operation on the type-aware feature vector based on the dynamic association strength to generate a fault feature vector; The fault feature vector is concatenated with the global feature vector of the heterogeneous information network, and node importance scores are calculated using a preset multi-layer perceptron. Nodes are sampled based on the node importance scores to construct a local network structure. A differentiable pooling operation is used to perform feature aggregation on the fault feature vector in the local network structure, and the aggregated features are updated through a gated recurrent unit to obtain a subgraph feature representation.

3. The method according to claim 1, characterized in that The power line system is modeled as a directed acyclic graph, the transmission path eigenvalues ​​are calculated, the trunk transmission lines are identified, the risk sections are divided, and the risk sections are integrated into the basic hierarchical knowledge base. The optimized hierarchical knowledge base includes: Constructing a directed acyclic graph of the power line system, the directed acyclic graph including a set of operating characteristic nodes and a set of power transmission edges, obtaining an operating characteristic value of each node in a training sample set, and setting the power transmission efficiency corresponding to each edge as the edge characteristic value; Calculating the sum of the connected edge eigenvalues ​​of each running feature node in the directed acyclic graph to obtain a node importance value, determining an electric energy transmission path based on the node importance value, and calculating the product of the edge eigenvalues ​​of each electric energy transmission path to obtain a path eigenvalue; Obtain the line load threshold of each operating characteristic node, identify the trunk transmission line that meets the load threshold constraint through the iterative capacity-adjusted maximum flow algorithm, and calculate the corresponding edge cut set load ratio; Obtain the failure probability of each operating characteristic node and the set of downstream affected nodes, calculate the node risk value, modify the node risk value based on the load ratio of the edge cut set, and divide the risk segments using a spectral clustering algorithm; The trunk transmission line is written into the skeleton layer of the basic hierarchical knowledge base, the protection level is divided based on the risk section, the path characteristic value and the node risk value are recorded, a hierarchical association relationship including the edge cut set load ratio is established, and the control strategy priority of the basic hierarchical knowledge base is updated based on the hierarchical association relationship to obtain an optimized hierarchical knowledge base.

4. The method according to claim 3, characterized in that Obtain the line load threshold of each operating characteristic node, identify the trunk transmission line that meets the load threshold constraint through the iterative capacity adjustment maximum flow algorithm, and calculate the corresponding edge cut set load ratio including: Obtain the line load threshold of each operating characteristic node in the power system network, construct an initial capacity matrix, and use the maximum flow algorithm to obtain the initial maximum flow value under the constraint that the actual load is less than or equal to the line load threshold. Based on the initial maximum flow value, determine the corresponding minimum cut set to obtain the initial trunk transmission line; Apply a disturbance coefficient to the initial trunk transmission line, modify the initial capacity matrix to obtain a modified capacity matrix, calculate the maximum flow value after the disturbance, and calculate the ratio of the maximum flow value to the initial maximum flow value to obtain the network fitness index; Based on the difference between the network fitness index and the preset fitness threshold, the modified capacity matrix is ​​updated to obtain the iterative capacity matrix. The iterative maximum flow value corresponding to the iterative capacity matrix is ​​calculated. The iteration is repeated until the rate of change of the maximum flow values ​​of two adjacent iterative cycles is less than the preset rate of change threshold. The final capacity matrix is ​​obtained. From the final capacity matrix, lines that meet the line load threshold are selected as the trunk transmission optimization lines, and the corresponding edge cut set load ratio is calculated.

5. The method according to claim 1, wherein Based on the optimized hierarchical knowledge base and backbone transmission lines, an inspection strategy model is established based on the reinforcement learning method. The inspection execution plan is adaptively generated, including: Based on the fault characteristics in the optimized hierarchical knowledge base, the mapping relationship between the fault characteristics and the operating status, and the transmission path characteristic values ​​of the trunk transmission line, a regional characteristic function is constructed, and the boundary of the dynamic associated sub-region is determined according to the regional characteristic function; A local state evaluator is constructed for the dynamically associated sub-regions. The risk factor score of the local state evaluator is multiplied by the weight coefficient to obtain the regional risk degree. The regional risk degree is then combined with the inspection time interval and load level to calculate the regional inspection urgency. Constructing a feature similarity matrix of dynamically associated sub-regions, generating a regional knowledge transfer function based on the feature similarity matrix, and fusing the regional knowledge transfer function with historical inspection experience to obtain regional shared experience; The regional inspection urgency, regional risk, and regional shared experience are used to construct a state vector, the inspection priority parameter is used to construct an action vector, and a reward function is established based on the state vector and the action vector; A reinforcement learning method is used to optimize the reward function to obtain strategy model parameters, and an inspection execution plan is generated according to the strategy model parameters.

6. The method according to claim 5, characterized in that Constructing a feature similarity matrix of dynamically associated sub-regions, generating a regional knowledge transfer function based on the feature similarity matrix, and integrating the regional knowledge transfer function with historical inspection experience to obtain regional shared experience includes: The dynamically associated sub-regions are constructed as a graph network model, where each dynamically associated sub-region corresponds to a region node, and the region node contains fault characteristics and operating status data. Edge connections are constructed based on the physical connection relationship between the dynamically associated sub-regions, and the topological association features between the region nodes are extracted through a graph convolutional network to obtain a feature similarity matrix; Calculating the similarity between adjacent regional nodes to obtain a gating coefficient, and constructing a regional knowledge transfer function based on the gating coefficient and the feature similarity matrix, wherein the regional knowledge transfer function represents the knowledge transfer relationship between regional nodes; Acquire features of the same dynamically associated sub-region in different time windows to construct positive sample pairs, acquire features of different dynamically associated sub-regions to construct negative sample pairs, and train a feature extraction model using a comparative loss function to obtain historical inspection experience; The regional knowledge transfer function is integrated with the historical inspection experience to obtain regional shared experience.

7. A power line health status assessment and prediction system based on big data, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain historical operation data of the power line, pre-process the historical operation data according to preset data quality rules, and generate standardized operation data; The second unit is used to extract fault features from standardized operating data based on a spatiotemporal heterogeneous graph network, establish a mapping relationship between features and operating status, build a basic hierarchical knowledge base, automatically annotate the standardized operating data, and generate a training sample set; The third unit is used to model the power line system as a directed acyclic graph, calculate the transmission path characteristic values, identify the trunk transmission lines, divide the risk sections, and integrate them into the basic hierarchical knowledge base to obtain an optimized hierarchical knowledge base; The fourth unit is used to establish an inspection strategy model based on the optimized hierarchical knowledge base and trunk transmission lines based on the reinforcement learning method, and adaptively generate an inspection execution plan; Execute the inspection plan and evaluate the health status of the power lines to obtain the health status evaluation results; The fifth unit is used to generate a power line health status report based on the health status assessment result.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images