A substation main equipment knowledge graph updating method and device

Abstract

The application provides a power transformation main equipment knowledge graph updating method and device, which aims to determine the power transformation main equipment state change amount by integrating the equipment physical topology logic and operation and maintenance operation rule data, and to update the knowledge graph by taking the equipment as the starting point when an event update occurs, so as to obtain an accurate power transformation main equipment knowledge graph. The power transformation main equipment knowledge graph updating method comprises the following steps: obtaining the knowledge graph of the power transformation main equipment, determining the equipment physical topology logic state data based on the equipment physical topology logic data in the knowledge graph and the message passing mechanism of the power physical topology layer graph neural network; determining the operation and maintenance operation state data based on the operation and maintenance operation rule data and the operation and maintenance operation dependency relationship rule; and after calculating the state change amount of the power transformation main equipment within a preset time based on the determined state data, updating the knowledge graph by using an incremental graph updating algorithm to obtain a target knowledge graph.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method and apparatus for updating a knowledge graph of substation main equipment. Background Technology

[0002] With the increasing complexity of power grid structures, the number of substation main equipment is growing exponentially. Currently, the ledger data, real-time operating status, and operating rules of different substation main equipment are scattered across multiple independent systems. During operation and maintenance, staff must repeatedly switch between these systems to query information, which is not only cumbersome but also makes it difficult to intuitively grasp equipment information, thus hindering the quality and speed of operation and maintenance. Therefore, utilizing knowledge graphs to construct multi-source data fusion and realize visualized operation and maintenance applications for equipment information has become a core direction for power system operation and maintenance upgrades. To ensure the accuracy of the substation main equipment knowledge graph data, the substation main equipment knowledge graph needs to be updated.

[0003] In existing technologies, methods for updating the knowledge graph of substation main equipment mostly rely on incremental updates based on the change range of single-dimensional data over a preset time period. However, the state changes of substation main equipment are multi-sourced, resulting from the collaborative effects of multi-dimensional data. If judgments are based solely on single-dimensional data, the following problems may arise: First, occasional fluctuations in a single dimension (such as short-term voltage fluctuations) may be misjudged as changes in the overall equipment state, triggering unnecessary graph updates; second, normal single-dimensional data (such as temperature not exceeding a threshold) may mask anomalies in multi-dimensional collaboration, leading to misjudgments of the equipment's true state. Both of these problems result in incorrect judgments of substation main equipment state changes, leading to inaccurate knowledge graph updates.

[0004] Therefore, there is an urgent need for a new method for updating the knowledge graph of substation main equipment in order to obtain an accurate knowledge graph of substation main equipment. Summary of the Invention

[0005] This application provides a method and apparatus for updating a knowledge graph of substation main equipment. The purpose is to integrate multimodal data of equipment physical topology logic and operation and maintenance rules, collaboratively judge changes in the status of substation main equipment, and incrementally update the knowledge graph starting from the equipment when an event update occurs, thereby obtaining an accurate knowledge graph of substation main equipment.

[0006] To address the aforementioned technical problems, the embodiments of this application provide the following technical solutions:

[0007] The first aspect of this application provides a method for updating a knowledge graph of substation main equipment, including:

[0008] Obtain a knowledge graph of the main substation equipment. The nodes of the knowledge graph are the main substation equipment entities, the feature vectors of the nodes contain multimodal data of the equipment, and the edges of the knowledge graph are the line connection relationships between the main substation equipment. The multimodal data includes equipment physical topology logical data and operation and maintenance rule data.

[0009] Based on the message passing mechanism of the device physical topology logic data and the power physical topology layer graph neural network, the device physical topology logic state data is determined.

[0010] Based on operation and maintenance rule data and operation and maintenance dependency rules, determine the operation and maintenance status data;

[0011] Based on the physical topology logical state data and operation and maintenance status data of the equipment, calculate the state change of the substation main equipment within a preset time.

[0012] When the state change amount is greater than the corresponding preset amount, it is determined that an event update has occurred in the substation main equipment;

[0013] When an event update occurs in the main substation equipment, the knowledge graph is updated using an incremental graph update algorithm, starting from the main substation equipment, to obtain the target knowledge graph.

[0014] A second aspect of this application provides an apparatus for updating a knowledge graph of substation main equipment, the apparatus comprising:

[0015] The acquisition unit is used to acquire the knowledge graph of the substation main equipment. The nodes of the knowledge graph are substation main equipment entities, the feature vectors of the nodes contain equipment multimodal data, the edges of the knowledge graph are the line connection relationships between substation main equipment, and the multimodal data includes equipment physical topology logical data and operation and maintenance rule data.

[0016] The determination unit is used to determine the physical topology logic state data of the equipment based on the message passing mechanism of the equipment physical topology logic data and the power physical topology layer graph neural network in the acquisition unit.

[0017] The determining unit is used to determine the operation and maintenance status data based on the operation and maintenance operation rule data and operation and maintenance operation dependency relationship rules in the acquisition unit;

[0018] The calculation unit is used to calculate the state change of the main substation equipment within a preset time period based on the equipment physical topology logical state data and operation and maintenance status data in the determination unit.

[0019] The unit is used to determine that an event update has occurred in the main substation equipment when the state change amount in the calculation unit is greater than the corresponding preset amount.

[0020] The obtaining unit is used to update the knowledge graph using an incremental graph update algorithm, starting from the substation main equipment, when an event update occurs in the substation main equipment, to obtain the target knowledge graph.

[0021] A third aspect of this application provides a storage medium comprising a stored program that, when the program is executed, controls the device where the storage medium is located to perform the aforementioned method for updating the knowledge graph of substation main equipment.

[0022] A fourth aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for updating the knowledge graph of substation main equipment.

[0023] Compared with existing technologies, the method for updating the knowledge graph of substation main equipment in this application firstly acquires a knowledge graph containing multimodal data of the equipment (physical topology logical data and operation and maintenance rule data), laying a comprehensive data foundation for accurately judging the status of substation main equipment and constructing an accurate knowledge topology graph. Secondly, it determines the physical topology logical state data based on the message passing mechanism of the power physical topology layer graph neural network, and determines the operation and maintenance state data by combining the operation and maintenance dependency rules. Then, it collaboratively calculates the state change amount, avoiding the problem of misjudgment by single-dimensional data. This prevents unnecessary updates triggered by accidental fluctuations and avoids the masking of multi-dimensional collaborative anomalies, ensuring the accuracy of substation main equipment status judgment and providing a reliable basis for knowledge topology graph updates. Finally, when the state change amount exceeds a preset amount, the graph is updated using an incremental algorithm starting from the target equipment to avoid invalid update interference and ensure that the content of the knowledge topology graph is consistent with the actual status of the equipment. In summary, this solution effectively solves the defects of existing technologies through precise control of multiple stages, and finally obtains an accurate knowledge topology graph, providing reliable support for power operation and maintenance. Attached Figure Description

[0024] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, with the same or corresponding reference numerals denoteing the same or corresponding parts, wherein:

[0025] Figure 1 A flowchart illustrating a method for updating a knowledge graph of substation main equipment is shown schematically.

[0026] Figure 2 A flowchart illustrating another method for updating the knowledge graph of substation main equipment is shown schematically.

[0027] Figure 3 A schematic diagram of the structure of a knowledge graph updating device for substation main equipment is shown.

[0028] Figure 4 A schematic diagram of another device for updating the knowledge graph of substation main equipment is shown. Detailed Implementation

[0029] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0030] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0031] In existing technologies, methods for updating the knowledge graph of substation main equipment mostly rely on incremental updates based on the change range of single-dimensional data over a preset time period. However, the state changes of substation main equipment are multi-sourced, resulting from the collaborative effects of multi-dimensional data. If judgments are based solely on single-dimensional data, the following problems may arise: First, occasional fluctuations in a single dimension (such as short-term voltage fluctuations) may be misjudged as changes in the overall equipment state, triggering unnecessary graph updates; second, normal single-dimensional data (such as temperature not exceeding a threshold) may mask anomalies in multi-dimensional collaboration, leading to misjudgments of the equipment's true state. Both of these problems result in incorrect judgments of substation main equipment state changes, leading to inaccurate knowledge graph updates.

[0032] Based on this, to overcome the aforementioned limitations, the applicant has devised the following method to obtain an accurate target knowledge graph. The specific steps of this method are as follows: Figure 1 As shown, it includes:

[0033] Step 101: Obtain the knowledge graph of the main substation equipment.

[0034] In this step, the nodes of the knowledge graph represent substation main equipment entities, and the feature vectors of the nodes contain multimodal data of the equipment. The edges of the knowledge graph represent the line connection relationships between substation main equipment. "Nodes" correspond to substation main equipment entities (such as transformers, circuit breakers, busbars, and other key equipment), and "edges" represent the line connection relationships between equipment (such as electrical connections and control connections). The node feature vectors integrate multi-source heterogeneous data to intuitively present equipment attributes, statuses, and relationships, supporting operation and maintenance decisions. The feature vectors contain two types of core attributes: static and dynamic. Static attributes store inherent parameters of the equipment, including persistent information such as equipment ledgers. Dynamic attributes maintain a timestamped state queue, recording the temporal changes in the equipment's operating status, completing the dynamic state binding of the equipment. When the substation main equipment status is updated, the system automatically triggers correlation analysis to ensure the logical consistency between the new status and historical data. Multimodal data refers to data related to substation main equipment collected from different systems and in different formats, including static data (equipment ledgers, models, rated parameters, etc.), dynamic data (real-time status monitoring data, time-series data such as voltage), and text data (document data such as operating rules, maintenance plans, and operating procedures). Quantization of each modal data can be achieved by calculating feature values ​​such as mean, standard deviation, and trend within a given window using feature engineering. Text data (such as operation records and procedure changes) is converted into semantic vectors through power industry terminology segmentation and semantic annotation. Line connection relationships refer to the physical connections and logical associations formed between substation main equipment through transmission lines, busbars, cables, etc., marking not only the connection method (e.g., overhead line connection, cable connection) but also parameters such as voltage level, rated current, and connection topology location.

[0035] In this embodiment, multimodal data includes equipment physical topology logical data and operation and maintenance rule data. The operation and maintenance rule data includes operation and maintenance text data, and the equipment physical topology logical data includes equipment status data and line topology data. This embodiment also cleans and formats the acquired raw data, formatting timestamp information and converting records from different time zones into a standard time format. The validity of the equipment status data is also verified, and abnormal data is filtered out. A sliding time window algorithm is used for the equipment status data to align data streams with different sampling frequencies. The equipment status data includes data fields such as equipment identifier, timestamp, operating status, and electrical measurement values. Equipment ledger data is imported synchronously from the equipment asset management system to obtain the static attribute information of the equipment. Deep analysis is performed on the operation and maintenance procedure text data to identify key entities and operational constraints. Network features from the equipment connection relationships and wiring diagrams in the line topology data are used, where equipment connection relationships are represented by an adjacency matrix, with matrix elements marking the physical connection status between devices. Simultaneously, a device feature matrix is ​​generated, containing key parameters such as voltage level, capacity, and operating status.

[0036] Step 102: Determine the physical topology logical state data of the equipment based on the message passing mechanism of the electrical physical topology layer graph neural network and the equipment physical topology logical data.

[0037] In this step, the physical topology logic data of the equipment consists of feature vectors representing the connectivity and operating status of the equipment within the physical topology. The message-passing mechanism of the power physical topology layer graph neural network is the process by which nodes in a graph neural network update their own states by sharing features, used to simulate electrical interactions between power equipment. Specifically, the physical topology logic data of the substation main equipment is acquired, including feature vectors such as rated voltage, current parameters, and line connection relationships. This data is then input into the power physical topology layer graph neural network, and through multiple iterations via the message-passing mechanism, the features of adjacent nodes are aggregated, outputting the physical topology logic state data.

[0038] Step 103: Determine the operation status data based on the operation and maintenance rule data and the operation and maintenance dependency relationship rules.

[0039] In this step, the operation and maintenance rule data includes operation and maintenance procedure texts, equipment operation records, and other data. Operation and maintenance operation dependency rules are rules that regulate the sequence and conditions of equipment operations, reflecting the logical relationships between operations. The constraint matrix is ​​a sparse matrix, where element values ​​represent the strength of the rule's constraint on the equipment, and the matrix structure reflects the operational dependencies between devices. Specifically, a semantic analysis model is used to extract logical features from the operation and maintenance rule data. A mapping function is used to associate the rules with the equipment states, and based on the equipment operation dependencies, the matching degree between the current state and the rules is analyzed. The output includes operation and maintenance operation status data that includes operational compliance and the influence weight of associated equipment.

[0040] Step 104: Based on the physical topology logical state data and operation and maintenance status data of the equipment, calculate the state change of the substation main equipment within a preset time period.

[0041] After the data is determined in steps 103 and 102, a first feature fusion value and a second feature fusion value are determined using a gate attention mechanism. The first feature fusion value represents the state information of the substation main equipment at the first time, which is the moment corresponding to the minimum value of the feature fusion value within a preset time. The second feature fusion value represents the state information of the substation main equipment at the second time, which is the moment corresponding to the maximum value of the feature fusion value within a preset time. The time interval between the first time and the second time is greater than a threshold. The difference between the first feature fusion value and the second feature fusion value is calculated, and the difference is determined to be the state change amount within a preset time. The threshold is set by the user based on the actual situation.

[0042] Step 105: When the state change amount is greater than the corresponding preset amount, it is determined that an event update has occurred in the substation main equipment.

[0043] In this step, the preset values ​​are set according to the substation main equipment operation and maintenance standards, safety procedures, and equipment characteristics, and serve as the boundary distinguishing between "normal equipment fluctuations" and "events requiring updates." Event updates refer to events that require updates to the knowledge graph due to abnormal status, attribute changes, or operational execution of the substation main equipment. These events are the "triggers" that trigger incremental updates to the knowledge graph. The process involves comparing the state change of the substation main equipment with its corresponding preset values. When the state change exceeds the corresponding preset value, an event update is determined to have occurred.

[0044] Step 106: When an event update occurs in the main substation equipment, starting from the main substation equipment, update the knowledge graph using an incremental graph update algorithm to obtain the target knowledge graph.

[0045] When a substation main equipment event update is determined in step 105, a breadth-first search algorithm is initiated to traverse the topology of the knowledge graph, starting with the substation main equipment experiencing the event update. Based on the line connection relationships between equipment, the affected associated equipment nodes and line edges are determined, forming a local subgraph that needs to be updated. Incremental updates to the local subgraph are then performed. Incremental updates to the local subgraph mean that only the node characteristics (such as updating the real-time status time-series data of abnormal equipment, and the equipment attribute data in the change log) and edge attributes (such as adjustments to line connection constraints due to equipment status changes) corresponding to the changed data are recalculated, reusing unchanged data and avoiding the resource waste of a full reconstruction.

[0046] Furthermore, based on the above Figure 1 The embodiments of the present invention shown provide a more detailed explanation of how to update the knowledge graph of substation main equipment, as detailed below. Figure 2 As shown:

[0047] Step 201: Construct a knowledge graph of the main substation equipment.

[0048] In this step, this embodiment also provides a method for constructing a knowledge graph of substation main equipment. Specifically, it involves: acquiring operation and maintenance text data, equipment status data, and line topology data; inputting the operation and maintenance text data into a semantic analysis model to determine text semantic features; using a sliding window to analyze equipment status data to determine equipment temporal status features; determining graph structure features based on graph structure analysis algorithms combined with line topology data; mapping text semantic features, equipment temporal status features, and graph structure features to spatial vectors of the target dimension to obtain fused feature vectors; and constructing a knowledge graph based on the fused feature vectors, the connection relationships between substation main equipment in the line topology data, and the substation main equipment itself. Operation and maintenance text data refers to various textual information generated during the operation and maintenance of substation main equipment, including equipment maintenance reports, operation procedure documents, operation records, fault analysis reports, etc., containing key information such as equipment operation logic, fault types, and maintenance requirements. Equipment status data is the real-time collected operating status parameters of substation main equipment through sensors and monitoring devices, covering time-series data such as current, voltage, temperature, and switch opening and closing status, which is the basis for extracting equipment temporal status features. The semantic analysis model is a pre-trained model based on knowledge enhancement of substation main equipment. By adding an electrical feature encoding layer and fine-tuning with domain corpora, it can accurately understand the professional terminology, operational logic, and equipment relationships in operation and maintenance texts, and is used to extract semantic features from the text. The graph structure analysis algorithm is a power feature-enhanced graph convolutional network (GCN). By introducing electrical parameter correction factors and multi-scale convolution operations into the traditional GCN, it can accurately capture information such as connection characteristics and node importance in the topology network of substation main equipment, and is used to extract graph structure features.

[0049] Specifically, when acquiring semantic features from text, the operation and maintenance text data (including equipment maintenance reports, operating procedures, operation records, etc.) is first preprocessed: redundant spaces, special symbols, and meaningless annotations are removed, and typos in professional terminology are corrected and standardized. Secondly, based on a professional dictionary of substation main equipment operation and maintenance, a bidirectional maximum matching algorithm is used to complete domain-specific word segmentation, and then the segmentation results are labeled with part-of-speech tags to clarify attributes such as nouns and verbs. The semantic analysis model uses a pre-trained model for substation main equipment knowledge enhancement based on an improved BERT framework, adding an electrical feature encoding layer to convert parameters such as equipment rated voltage and current into vectors for semantic analysis. The model is then fine-tuned using a historical operation and maintenance text corpus in the substation field to ensure accurate understanding of operational logic. Finally, the preprocessed text is input into the model, and deep semantics are mined through multiple Transformer layers to extract text semantic feature vectors.

[0050] To determine the time-series status characteristics of the equipment, the sliding window parameters are first configured based on the characteristics of the equipment status data. This data includes time-series monitoring data such as current, voltage, temperature, and oil level. Simultaneously, the statistical indicators to be calculated within the window (mean, standard deviation, maximum value, and slope of the trend) are determined. Then, outliers are processed during the window sliding process. The 3σ principle is used to identify outlier data. If the outlier data is caused by sensor malfunction (e.g., a sudden temperature rise followed by immediate recovery), linear interpolation is used to repair it based on preceding and following normal data. If it is genuine equipment anomaly data (e.g., current continuously exceeding the rated value), the outlier is retained and marked. Finally, statistical indicators are calculated for each window's data, and these are integrated into a preliminary time-series feature vector. Then, a dynamic time warping algorithm is used to eliminate differences in sampling frequencies between different data points, resulting in a time-series status feature vector that reflects the changing patterns of the equipment status.

[0051] To determine the graph structure characteristics, the following steps are taken: First, the line topology data, sourced from the substation equipment asset management system, is preprocessed. This data includes equipment ID, type, connection ports, line resistance, voltage level, etc. Duplicate connection records are removed, erroneous connections are corrected, and missing line parameters are supplemented. Second, a power feature-enhanced graph neural network is selected as the graph structure analysis algorithm. This algorithm captures the connection characteristics of equipment at different voltage levels through multi-scale convolution, improves the message passing mechanism, and introduces an electrical parameter correction factor in the message weight calculation, enabling the algorithm to accurately analyze the electrical connection relationships between equipment. Finally, the initial topology graph is input into the algorithm to extract the graph structure feature vector. This vector characterizes the importance of a device's position in the topology network and the strength of its connections with other devices.

[0052] It is worth noting that textual semantic features, equipment temporal state features, and graph structure features are all represented by vectors. During vector representation, the dimensions of the vectors may be the same or different. Therefore, it is necessary to map the textual semantic features, equipment temporal state features, and graph structure features to spatial vectors of the target dimension to obtain fused feature vectors. Based on the fused feature vectors, the connection relationships between substation main equipment in the line topology data, and the substation main equipment, a knowledge graph is constructed. The construction of the knowledge graph can begin by building knowledge graph nodes, using substation main equipment entities as nodes, assigning each node a unique identifier, and including fused feature attributes in the node attributes. Next, knowledge graph edges are constructed based on the equipment connection relationships in the line topology data. The edge attributes include line type, length, resistance, voltage level, and connection status (on / off, represented by Boolean values). For equipment with operational logic constraints, "logical constraint edges" are constructed, with attributes including constraint rule descriptions and constraint priorities. Finally, a hybrid storage architecture of "graph database + vector database" is adopted. The graph database (such as Neo4j) stores the basic attributes, edge attributes, and topological / logical relationships of nodes, while the vector database (such as Milvus) stores the fused feature vectors of nodes. After storage, structural verification (checking connectivity to avoid isolated nodes) and attribute verification (verifying the consistency between fused features and basic attributes) are performed to ensure the accuracy and completeness of the knowledge graph. The fused feature vector can be represented as a device state vector = [basic attribute layer: static data such as device ledgers, operational state layer: time-series data of status monitoring, topological association layer: connection relationship encoding_phy, operational constraint layer: locking condition logic_log]. Alternatively, it can be represented as a triple.

[0053] In this step, this embodiment will also fuse feature vectors and input them into a two-layer graph neural network model to obtain data output from the physical topology layer and the logical constraint layer. The physical topology layer is configured with a graph convolution operation unit that enhances the features of the substation main equipment, and the data of the logical constraint layer is the predicate rule set generated by compiling the substation main equipment operation procedures. The physical topology layer and the logical constraint layer are connected through a gating attention mechanism in the dynamic decision layer. The physical topology layer uses a graph convolutional network with enhanced power features to improve the message passing mechanism, where the message weight W_phy introduces an electrical parameter correction factor. The logical constraint layer constructs a differentiable logic reasoning module, compiles the operation rules into a predicate rule set, and achieves self-optimization adjustment of the rules through gradient descent. The dual-channel information is exchanged through a gating fusion unit. The above steps first process the wiring network topology through iterative graph convolution operations, and combine message passing, constraint matrix generation, gating signals and feature fusion formulas to form stable topological features through multi-layer neighborhood information aggregation, thus completing physical topology reasoning; then, traverse the operation and maintenance rule set to match node conditions, generate operation constraint markers and spread them to realize logical constraint propagation; finally, use the attention mechanism to calculate the fusion weight, and the gating unit dynamically adjusts the information contribution to generate a comprehensive reasoning conclusion, thus completing the dynamic fusion decision-making for the operation and maintenance of substation main equipment.

[0054] The formula for the message passing mechanism of a power physics topology layered graph neural network is as follows:

[0055] h_i^(l+1)=σ(∑(W_phy·h_j^(l)+b_phy))

[0056] Where h_i^(l+1) is the feature vector of node i in the (l+1)th layer network (i.e., the updated device state representation), h_j^(l) is the feature vector of node j adjacent to node i in the lth layer network (the state representation of neighboring devices), W_phy is a trainable weight matrix, which is the electrical parameter correction factor used to extract power topology features, b_phy is the bias term to enhance the model's expressive power, and σ is the activation function (such as ReLU) to introduce nonlinear transformation. Feature aggregation (∑ operation) is a weighted summation of the feature vectors of all neighboring nodes j of node i, aggregating local topology information. Linear transformation (W_phy·h_j^(l)) maps the neighbor features to a high-dimensional space through the weight matrix W_phy to capture the electrical interaction patterns between devices. Nonlinear activation (σ function) is a nonlinear activation of the aggregation result to generate a new feature representation of node i. The neighbor definition rule is that the range of neighboring nodes j is determined by the electrical connection relationship (rather than simple topological distance), ensuring that the aggregation conforms to the characteristics of the power system.

[0057] The special design of the weight matrix W_phy in the above formula allows for the addition of electrical parameter correction factors (such as equipment impedance and capacity) on the basis of standard graph convolution, making the message passing process conform to the laws of power physics. In addition, the above formula is used to design the message passing mechanism of the power physics topology layer. This formula is the core calculation unit of power physics topology analysis. It iteratively executes the following operations: (1) propagating equipment status information layer by layer: reflecting the diffusion process of electrical influence in the power grid. (2) dynamically correcting connection weights: adapting to the changes in topology characteristics under different operating conditions. (3) the output results are used for advanced applications such as electrical island partitioning, power flow calculation, and short-circuit analysis. The above formula is specifically used for the enhanced graph neural network model of power, improving the accuracy and efficiency of topology analysis.

[0058] The formula for generating the constraint matrix in the logical constraint layer is as follows:

[0059] constraint_matrix[i,j]=f(rule_i,state_j)

[0060] Wherein, rule_i is the i-th operation procedure, state_j is the current state of device j, the mapping function f implements the transformation from text rules to computable logic, and the generated constraint_matrix is ​​a sparse matrix, where the element value [i,j] represents the constraint strength of the i-th rule on device j (a continuous value from 0 to 1), and is used to represent the operation dependency relationship between devices reflected in the matrix structure.

[0061] Here's an example: For instance, `rule_i` means "The neutral point grounding switch must be closed before the main transformer is de-energized," and `state_j` means "Circuit breaker open position" and "Protection plate on / off status." The mapping function `f` converts text rules into computable logic. Rule parsing transforms natural language rules into formal logical expressions (such as first-order predicates). Example: `rule_i = "The neutral point grounding switch must be closed before operating the main transformer" → converted to the logical expression: `MustDo(close_neutral_switch) BEFORE Do(transformer_outage)`. The mapping function also performs state matching: checking if the current state of the equipment meets the rule triggering conditions. Example: When `state_j` detects "Main transformer operation command" and "Neutral point grounding switch not closed," the constraint is triggered. Through a semantic model enhanced with power knowledge, text rules are embedded as low-dimensional vectors `rule_embedding`, preserving logical semantic features.

[0062] In this step, the purpose of the gating attention mechanism in the dynamic decision layer is to dynamically balance the physical characteristics and operational logic of the power topology, achieving adaptive feature fusion. The gating unit automatically adjusts the contribution weights of the two types of features according to the current system state. The gating signal generation formula is: gate = σ(W_g·[h_phy||h_log]), where h_phy is the physical topology feature vector, h_log is the logical constraint feature vector, [h_phy||h_log] is the concatenation of the physical topology feature vector and the logical constraint feature vector, W_g is a trainable parameter matrix that learns the correlation between the two types of features, and σ is the Sigmoid activation function that compresses the output to the [0,1] interval. The feature fusion value is calculated as: h_final = gate⊙h_phy + (1-gate)⊙h_log, where ⊙ is element-wise multiplication. gate is the first weight, and 1-gate is the second weight.

[0063] The physical topology layer and logical constraint layer complete the integrated decision-making of substation main equipment operation and maintenance through the gating attention mechanism in the dynamic decision layer. When the physical topology features are more important, the gate approaches the preset value (0.8-1), thus strengthening h_phy. When the operation logic features are more critical, the gate approaches the preset value (0-0.3), thus emphasizing h_log. For example, the following scenario application examples are provided: (1) Normal operation: The weight of physical topology features is high (gate≈0.8), ensuring the accuracy of electrical connections. (2) Switching operation process: The weight of logical constraint features increases (gate≈0.3), prioritizing the fulfillment of operation procedures. (3) Fault handling stage: Dynamically balance the two types of features (gate≈0.5), taking into account both topology safety and handling specifications. This mechanism enables the system to dynamically adjust the inference strategy according to the real-time scenario, significantly improving the decision adaptability of complex power scenarios.

[0064] As can be seen, the dual-layer graph neural network architecture and gating attention mechanism organically integrate physical topology analysis and logical constraint reasoning, solving the pain point of the separation between "physical characteristics" and "operational rules" in traditional operation and maintenance. The enhanced graph convolutional units in the physical topology layer accurately capture the electrical connection characteristics and topological dependencies between devices, while the logical constraint layer transforms operating procedures into a computable set of predicate rules, ensuring that reasoning conforms to operation and maintenance specifications. The gating mechanism in the dynamic decision layer adaptively adjusts the weights of the two types of features according to the operating scenario (such as normal operation or switching operation), achieving flexible reasoning based on either "topology priority" or "rule priority." This integrated decision-making model significantly improves the scientific rigor and adaptability of operation and maintenance decisions, providing more reliable logical support for equipment status assessment and operational risk prediction.

[0065] Step 202: Determine if an event update has occurred in the main substation equipment.

[0066] In step 201, after determining the multimodal data of the substation main equipment in the knowledge graph, distributed event listening is used to monitor the multimodal data of the substation main equipment in the knowledge graph, and iterative graph convolution operation is used to determine the state change amount within a preset time, thereby determining the event update of the substation main equipment. Specific steps include: inputting the fused feature vector into a two-layer graph neural network model to obtain the data output by the physical topology layer and the data output by the logical constraint layer in the two-layer graph neural network model;

[0067] Based on the correlation between the data output from the physical topology layer and the data output from the logical constraint layer, a first weight and a second weight are determined. The first weight is the proportion of the device's physical topology logical data in the determination of device status, and the second weight is the proportion of the operation and maintenance rule data in the determination of the substation main equipment status. Based on the device's physical topology logical status data, the first weight, the operation and maintenance status data, and the second weight at a first time, a first feature fusion value is determined. The first feature fusion value represents the status information of the substation main equipment at the first time, which is the moment corresponding to the minimum value of the feature fusion value within a preset time. Based on the device's physical topology logical status data, the first weight, the operation and maintenance status data, and the second weight at a second time, a second feature fusion value is determined. The second feature fusion value represents the status information of the substation main equipment at the second time, which is the moment corresponding to the maximum value of the feature fusion value within a preset time, and the time interval between the first time and the second time is greater than a threshold. The difference between the first feature fusion value and the second feature fusion value is calculated, and the difference is determined to be the change in status within a preset time. The feature fusion value includes the first feature fusion value and the second feature fusion value, and the specific calculation method is as follows:

[0068] The formula for calculating the first weight is: gate = σ(W_g·[h_phy||h_log]). The formula for calculating the feature fusion value is h_final = gate⊙h_phy + (1-gate)⊙h_log. The difference between the first feature fusion value and the second feature fusion value is determined as the state change amount Δs within a preset time. When the change amount is greater than the corresponding preset amount, where the preset amount is θ, it is determined that an event update has occurred in the substation main equipment. That is, if Δs>θ||t_now-t_last>T_max, an update event is triggered, which means that the knowledge graph needs to be updated at this time. Here, Δs represents the change amplitude of the device's state in the first and second time periods, θ is the preset amount (e.g., a sudden change in the circuit breaker's open / close state, or the temperature exceeding the warning value), t_now-t_last>T_max prevents the state information from becoming outdated due to communication interruption and ensures data freshness, t_last is the first time, t_now is the second time, and T_max is the maximum state retention period (e.g., when the device has no state change for a long time, a forced refresh is performed).

[0069] In this embodiment, the detection logic for judging the state change of the main substation equipment is as follows: Define the "check_trigger" function, and input the main substation equipment data at the first time and the second time, as well as the first time and the second time. First, calculate Δs using the "calculate_state_deviation" function, and then calculate the interval "time_elapsed" between the first time and the second time. If Δs is greater than a preset value θ, or "time_elapsed" is greater than the configured parameter T_max, and either condition is met, call the "trigger_update_event" function to trigger an update event. Here, θ represents the sensitivity to state changes, and its setting can be designed according to the equipment type and safety level of the main substation equipment. For example, θ = 1 for circuit breakers (state change) and θ = 5% for transformers (temperature change), which is not limited here. T_max is the data validity period, and its setting can be set according to the equipment state inertia and business response requirements. For example, T_max = 5min for critical equipment and T_max = 30min for ordinary equipment. It is worth noting that the preset time in this application can be T_max or a value less than T_max. If the preset time is greater than T_max, it is determined that the knowledge graph needs to be updated.

[0070] It is evident that calculating changes based on a two-layer network output and a gated attention mechanism solves the problems of easy misjudgment and difficulty in quantification in single-modal data monitoring. By characterizing the change by calculating the difference between the feature fusion values ​​at the first and second time points, the dual influence of physical topology and logical constraints is integrated, providing a more comprehensive and accurate reflection of the actual degree of change in equipment status, avoiding invalid update triggers caused by fluctuations in single parameters such as voltage and current. This quantification method takes into account the correlation of multimodal data and ensures the timeliness of change monitoring through time window comparison, providing an objective and unified standard for determining events where "the change is greater than the preset value," effectively balancing monitoring accuracy and computational efficiency, and reducing the false positive and false negative rates.

[0071] Step 203: When an event update occurs in the main substation equipment, the knowledge graph is updated using an incremental graph update algorithm, starting from the main substation equipment, to obtain the target knowledge graph.

[0072] When it is determined in step 202 that an event update has occurred in the main substation equipment, the target knowledge graph is obtained by: determining the influence radius and subgraph based on the line connection relationship between the main substation equipment and the main substation equipment in the knowledge graph, wherein the subgraph is a local knowledge graph determined based on the influence radius; performing local recalculation based on the subgraph to obtain the subgraph reconstruction result; and generating the target knowledge graph based on the subgraph reconstruction and the knowledge graph.

[0073] In this step, the core of the incremental graph update algorithm is to calculate the influence radius of the event update. Searching for the influence radius involves determining the main substation equipment and its line connections. These line connections are obtained based on line topology data. After identifying the affected equipment, the set of substation main equipment and the affected equipment (the state change event set) is used to determine the influence radius and subgraph. Based on the subgraph, local recalculation is performed to obtain the subgraph reconstruction result. The specific formula is as follows: procedure INCREMENTAL_UPDATE(G,ΔE):

[0074] Calculate the influence radius = R(v_i, Δs), extract the subgraph = G_sub = {v∈R}

[0075] Local recalculation = ΔG_sub, G' = G⊕ΔG_sub

[0076] Wherein, G is the current complete power knowledge graph, ΔE is the set of state change events (including at least the changed device, state difference, timestamp, etc.), R(v_i,Δs) is the influence radius calculation function with the changed device v_i as the center and the state change Δs as the input, G_sub is the subgraph to be updated extracted based on the influence radius, ΔG_sub is the subgraph reconstruction result, G' is the updated target knowledge graph, and ⊕ is a custom operator that represents the difference merging operation based on topological features. This operation ⊕ includes state attribute coverage update, topological connection relationship reconstruction, and logical constraint rules. The specific implementation in its synchronization process involves the subgraph isomorphic matching algorithm.

[0077] It is worth noting that, firstly, the radius of influence is calculated based on the characteristics of the power topology and the magnitude of state changes. First, the type of the changing device v_i and the Δs value in ΔE are analyzed, and its topological connection relationship in G is retrieved to form a dynamic radius formula. Specifically, a breadth-first search is used to traverse the topology network, filtering out directly connected devices, nodes associated with the same electrical island, and entities related to operation interlocking, constructing an influence set to ensure coverage of all devices and relationships affected by state changes. For example, in a 110kV substation, circuit breaker CB1 trips (Δs = 1, state changes from closed to open). First, the basic influence range of CB1 is determined to be the directly connected 220kV busbar B1 and line L1. Due to the large magnitude of the tripping operation's impact, the radius coefficient is adjusted to 1.5, extending to transformer T1 and disconnector QS2 on the same electrical island. Simultaneously, protection device P1 and control unit M2 associated with operation interlocking are included, ultimately forming an influence radius R containing 7 nodes and 9 connection relationships, which serves as the basis for extracting G_sub.

[0078] Secondly, extracting the subgraph to be updated involves first locating the change source device v_i in ΔE, and determining the boundary based on the calculated influence radius R. The knowledge graph G is recursively extracted from v_i and all device nodes within the range of R, while preserving all connections between these nodes, including physical connections, logical associations, and constraint rules. A topological index is used to quickly match associated edges, filtering redundant nodes and relationships to form a subgraph G_sub containing node attributes, connection types, and weights. A mapping relationship with the original graph is established to ensure the integrity and independence of the subgraph. For example, in a substation, the influence radius R of a circuit breaker CB2 tripping event includes busbar B2, line L3, disconnector QS5, and protection device P3. When extracting the subgraph, CB2 is first located as the change source, and its associated edges are traversed to find B2 and QS5; L3 on the same island is obtained through the electrical island topological index; and P3 is associated according to the blocking rules. Ultimately, G_sub contains 5 nodes, covering the physical connection between CB2 and B2, the topological relationship between L3 and QS5, and the protection logic of P3 for CB2, with a total of 7 relational edges, fully preserving the device association characteristics within the scope of influence.

[0079] Secondly, local subgraph recalculation involves reconstructing the topology matrix based on node connectivity during topology analysis of the extracted subgraph G_sub. Electrical connectivity components are recalculated using breadth-first search, split or merged electrical islands are marked, and new interlocking rules are derived based on equipment state changes, such as operational restrictions on equipment in power outage areas. The state attributes of equipment within the subgraph are updated item by item, including operating status and measured values. Finally, ΔG_sub is generated, containing the changed node attributes, updated connectivity relationships, and new constraint rules, forming a structured difference dataset. For example, after line L5 in a substation trips, the subgraph contains circuit breaker CB5, busbar B3, transformer T2, and protection device P5. In the topology analysis, recalculating connectivity reveals that B3 and T2 are electrically isolated; interlocking rules are derived: CB5 closing operation requires P5 alarm clearance; T2's state is updated to "standby," and B3's voltage value is 0. ΔG_sub records these changes: T2 attribute update, B3-T2 connection status marked as "disconnected", and new CB5 operation latching condition, providing a clear basis for subsequent merging.

[0080] Finally, the target knowledge graph is obtained by first locating the overlapping area between ΔG_sub and the original graph G through subgraph isomorphic matching. For state attributes, the "change set coverage" principle is adopted, replacing the old attributes of the corresponding nodes in G with the updated equipment attributes (such as operating status and measurement values) in ΔG_sub. For topology connections, based on the connection change records in ΔG_sub, invalid connections are deleted, valid connections are added, and the relationships between nodes are rebuilt. The association index is updated synchronously to ensure the consistency and integrity of the merged graph, and finally, the updated G' (target knowledge graph) is generated. For example: ΔG_sub of a substation contains the state change of circuit breaker CB3 (open → closed) and the newly added connection relationship between CB3 and bus B4. During the merging process, first match the position of CB3 in G and update its status attribute to "closed"; based on the ΔG_sub record, delete the "disconnected" marker between CB3 and B4 and add a "closed" connection relationship; synchronously update the electrical island topology index, incorporate the originally separated B4 and line L6 into the same connected component, complete the construction of the main graph G', and ensure that the topology relationship is consistent with the equipment status.

[0081] Therefore, the incremental process of "determining the radius of influence - extracting subgraphs - local recalculation - merging and updating" solves the problems of long processing time and high resource consumption in full updates. Starting with the changed equipment, the radius of influence is determined based on line connection relationships. The local area requiring update is delineated and a subgraph is extracted. Topology analysis and state recalculation are performed only on the subgraph, avoiding redundant processing of the entire map, thus shortening processing time and reducing resource consumption. The subgraph reconstruction result is merged with the original map to generate the target map. Difference updates ensure map integrity and consistency, while also ensuring the validity of unchanged parts of the original map and precise synchronization of changed areas, meeting the real-time and efficient map update requirements of dynamic power grid operation and maintenance.

[0082] Step 204: Verify the target knowledge graph.

[0083] In this step, after obtaining the target knowledge graph in step 203, the target knowledge graph also needs to be verified. Specific verification includes: structural verification, logical verification, and state verification. At least one verification method can be used. Structural verification includes: extracting the substation main equipment entity nodes and the line connection edges between nodes from the target knowledge graph; constructing a device topology adjacency matrix based on graph structure analysis algorithms, combining the substation main equipment entity nodes and the line connection edges between nodes. Elements in the device topology adjacency matrix mark the physical connection status between substation main equipment and the substation main equipment directly associated with each substation main equipment entity; traversing the device topology network in the target knowledge graph using a breadth-first search algorithm, starting with any substation main equipment as the starting node, and recording the set of substation main equipment entity nodes traversed; comparing the set of substation main equipment entity nodes with the device topology adjacency matrix to determine if there are any isolated nodes that have not been traversed, and simultaneously checking for illegal connection paths during the traversal process to determine if the connections between devices in the target knowledge graph are abnormal. If there are no isolated nodes that have not been traversed, check that there are no illegal connection paths during the traversal process to determine that the connections between devices in the target knowledge graph are normal.

[0084] Specifically, during the verification of the target knowledge graph structure, the following steps are taken: First, extract the entity nodes of the substation main equipment and the line connection edge data between the nodes from the knowledge graph: clarify the unique identifier of the equipment entity node, the equipment type, the start / end node identifier of the line connection, the physical connection status, and other core information to ensure that the data completely covers all main equipment and connection relationships within the graph. Second, construct an equipment topology adjacency matrix based on the graph structure analysis algorithm. The matrix dimension is determined by the number of equipment nodes, and the matrix elements are assigned values ​​according to rules. If two nodes have a direct physical connection and the status is "connected", the element value is marked as "1" and associated with the directly associated equipment information; if the connection status is "disconnected", it is marked as "0", and if there is no direct connection, it is marked as "∞". Then, select any substation main equipment as the starting node and start the breadth-first search algorithm to traverse the equipment topology network: initialize the queue and the set of traversed nodes, enqueue the starting node and mark it as traversed, dequeue nodes in turn and find their directly connected nodes according to the adjacency matrix, and add untraversed nodes to the queue and set until the queue is empty to obtain the complete set of traversed nodes. Finally, the traversed node set is compared with all device nodes covered by the adjacency matrix to determine if any isolated nodes exist. Simultaneously, based on the device connection rules in the substation operation and maintenance regulations (such as compliant connection requirements for devices of different voltage levels), the traversal path is checked for illegal connections. If neither isolated nodes nor illegal paths are found, the connections between devices in the target knowledge graph are considered normal; otherwise, abnormal node and path information is output to support subsequent corrections. This verification method provides clear anomaly location and basis for graph correction, ensuring the consistency between the target knowledge graph topology and the actual power grid device connections, and improving the reliability of the graph.

[0085] In this step, the logical verification includes: based on the target knowledge graph, extracting the predicate rule set generated by compiling the substation main equipment operation procedures, and the operating status data of each substation main equipment entity; matching each operation rule in the predicate rule set with the corresponding substation main equipment entity's operating status data one by one according to the constraints of each operation rule, and determining whether the current status of the equipment meets the preconditions and postconditions of the operation rule; marking the equipment status and operation records that do not meet the constraints of the operation rule, generating a logical verification anomaly report, and completing the logical verification of the target knowledge graph.

[0086] This step also includes status verification. Specifically, it involves extracting real-time operating status data (such as voltage, current, temperature, and circuit breaker status) and historical status time sequences of each substation main equipment entity based on the target knowledge graph; constructing a "real-time-historical" status comparison model to calculate the deviation rate between the current time and the historical average status within a preset third time period, while setting dual verification standards based on equipment rated parameter thresholds (such as rated voltage range and maximum allowable temperature). Equipment with a deviation rate exceeding a preset probability or real-time data exceeding the rated threshold is marked as a "status anomaly candidate." Further, by calling the sensor verification logs associated with the equipment, it is checked whether the data anomaly is caused by sensor failure. If it is a true equipment status anomaly, the anomaly type is matched and a preliminary diagnostic suggestion is generated by combining the fault case library in the maintenance text data. Finally, a status verification report is output, clearly specifying the list of normal equipment, the ID of abnormal equipment, the abnormal parameters, and the diagnostic basis, ensuring the accuracy and reliability of the equipment status data in the target knowledge graph.

[0087] Furthermore, as a response to the above Figure 1-2 The implementation of the method embodiment shown in this invention provides an update device for a knowledge graph of substation main equipment. This device embodiment corresponds to the foregoing method embodiments. For ease of reading, this embodiment will not repeat the details of the foregoing method embodiments one by one, but it should be clear that the device in this embodiment can implement all the contents of the foregoing method embodiments. For example... Figure 3 As shown, it includes:

[0088] The acquisition unit 31 is used to acquire the knowledge graph of the substation main equipment. The nodes of the knowledge graph are substation main equipment entities, the feature vectors of the nodes contain equipment multimodal data, the edges of the knowledge graph are the line connection relationships between substation main equipment, and the multimodal data includes equipment physical topology logical data and operation and maintenance rule data.

[0089] The determination unit 32 is used to determine the physical topology logic state data of the equipment based on the message passing mechanism of the equipment physical topology logic data and the power physical topology layer diagram neural network in the acquisition unit 31.

[0090] The determining unit 32 is used to determine the operation and maintenance status data based on the operation and maintenance operation rule data and operation and maintenance operation dependency relationship rules in the acquisition unit;

[0091] The calculation unit 33 is used to calculate the state change of the main substation equipment within a preset time based on the equipment physical topology logical state data and operation and maintenance state data in the determination unit 32.

[0092] Unit 34 is used to determine that an event update has occurred in the main substation equipment when the state change amount in the calculation unit is greater than the corresponding preset amount.

[0093] The obtaining unit 34 is used to update the knowledge graph using an incremental graph update algorithm, starting from the substation main equipment, when an event update occurs in the substation main equipment, to obtain the target knowledge graph.

[0094] Furthermore, the operation and maintenance rule data includes operation and maintenance text data, the device physical topology logical data includes device status data and line topology data, and the acquisition unit 31 includes:

[0095] The determination module 311 is used to input the operation and maintenance text data into the semantic analysis model to determine the semantic features of the text;

[0096] The determining module 311 is used to analyze the device status data using a sliding window to determine the device timing status characteristics.

[0097] The determining module 311 is used to determine the graph structure features based on the graph structure analysis algorithm combined with the line topology data;

[0098] The module 312 is used to map the text semantic features, the device temporal state features and the graph structure features in the determining module 311 to a spatial vector of the target dimension to obtain a fused feature vector.

[0099] The obtained module 312 is used to construct a knowledge graph based on the fused feature vector, the connection relationship between substation main equipment in the line topology data, and the substation main equipment.

[0100] Furthermore, such as Figure 4 As shown, the computing unit 33 further includes:

[0101] The fused feature vector is input into a two-layer graph neural network model to obtain the data output by the physical topology layer and the data output by the logical constraint layer in the two-layer graph neural network model.

[0102] Based on the correlation between the data output from the physical topology layer and the data output from the logical constraint layer, a first weight and a second weight are determined. The first weight is the proportion of the device's physical topology logical data in the determination of the device's status, and the second weight is the proportion of the operation and maintenance rule data in the determination of the substation main equipment's status. Based on the device's physical topology logical status data at a first time, the first weight, the operation and maintenance status data, and the second weight, a first feature fusion value is determined. The first feature fusion value represents the status information of the substation main equipment at the first time, and the first time is the moment corresponding to the minimum value of the feature fusion value within a preset time. Based on the device's physical topology logical status data at a second time, the first weight, the operation and maintenance status data, and the second weight, a second feature fusion value is determined. The second feature fusion value represents the status information of the substation main equipment at a second time, and the second time is the moment corresponding to the maximum value of the feature fusion value within a preset time, and the time interval between the first time and the second time is greater than a threshold.

[0103] Calculate the difference between the first feature fusion value and the second feature fusion value, and determine that the difference is the state change amount within a preset time period.

[0104] Furthermore, such as Figure 4 As shown, the device further includes:

[0105] The formula for the message passing mechanism of the power physics topology graph neural network is as follows:

[0106] h_i^(l+1)=σ(∑(W_phy·h_j^(l)+b_phy))

[0107] Where h_i^(l+1) is the feature vector of node i in the (l+1)th layer network, h_j^(l) is the feature vector of node j adjacent to node i in the lth layer network, W_phy is a trainable weight matrix, which is the electrical parameter correction factor used to extract power topology features, b_phy is the bias term to enhance the model's expressive power, σ is the activation function, which performs nonlinear activation on the aggregation result, and W_phy·h_j^(l) maps the neighbor features to a high-dimensional space through the weight matrix W_phy to capture the electrical interaction patterns between devices;

[0108] The formula for the operation and maintenance dependency rule is as follows:

[0109] constraint_matrix[i,j]=f(rule_i,state_j)

[0110] Where rule_i is the i-th operation procedure, state_j is the current state of device j, the mapping function f implements the transformation from text rules to computable logic, and constraint_matrix is ​​a sparse matrix, the element value [i,j] represents the constraint strength of the i-th rule on device j, and is used to represent the operation dependency relationship between devices reflected in the matrix structure;

[0111] The formula for calculating the first weight is:

[0112] gate=σ(W_g·[h_phy||h_log])

[0113] The formula for calculating the first feature fusion value is:

[0114] h_final=gate⊙h_phy+(1-gate)⊙h_log

[0115] Where gate is the first weight, the second weight is 1-gate, h_phy is the data output by the physical topology layer, h_log is the data output by the logical constraint layer, [h_phy||h_log] is the concatenation of the data output by the physical topology layer and the data output by the logical constraint layer, W_g is the trainable parameter matrix that learns the correlation between the two types of features, σ is the Sigmoid activation function that compresses the output to the [0,1] interval, and ⊙ is the element-wise multiplication.

[0116] Furthermore, such as Figure 4 As shown, the obtaining unit 34 includes:

[0117] The data module 341 is used to determine the influence radius and subgraph based on the line connection relationship between the main substation equipment and the main substation equipment in the knowledge graph. The subgraph is a local knowledge graph determined based on the influence radius.

[0118] The data determination module 341 is used to perform local recalculation based on the subgraph to obtain the subgraph reconstruction result;

[0119] The generation module 342 is used to generate a target knowledge graph based on the subgraph reconstruction combination and knowledge graph in the determined data module 341.

[0120] Furthermore, such as Figure 4 As shown, the device further includes a verification unit 35, which includes:

[0121] The data acquisition module 351 is used to extract the main substation equipment entity nodes and the line connection edges between nodes in the target knowledge graph.

[0122] The data acquisition module 351 is used to construct a device topology adjacency matrix based on a graph structure analysis algorithm, combining the entity nodes of the substation main equipment and the line connection edges between the nodes. The elements in the device topology adjacency matrix mark the physical connection status between the substation main equipment and the substation main equipment directly associated with the substation main equipment entity.

[0123] The data acquisition module 351 is used to traverse the device topology network in the target knowledge graph using a breadth-first search algorithm, with any substation main equipment as the starting node, and record the set of substation main equipment entity nodes traversed.

[0124] The verification module 352 is used to compare the set of substation main equipment entity nodes in the data acquisition module 351 with the equipment topology adjacency matrix to determine whether there are any isolated nodes that have not been traversed, and at the same time check whether there are any illegal connection paths during the traversal process to determine whether the connection between devices in the target knowledge graph is abnormal.

[0125] Furthermore, such as Figure 4 As shown, the device further includes a verification unit 35, which includes:

[0126] The data acquisition module 351 is used to extract the predicate rule set generated by compiling the substation main equipment operation procedures and the operating status data of each substation main equipment entity based on the target knowledge graph.

[0127] The verification module 352 is used to match the operating status data of the corresponding power supply main equipment entity one by one with the constraint conditions of each operation rule in the predicate rule set of the data acquisition module 351, and to determine whether the current status of the equipment meets the preconditions and postconditions of the operation rule.

[0128] The verification module 352 is used to mark the device status and operation records that do not meet the constraints of the operation rules, generate a logic verification anomaly report, and complete the logic verification of the target knowledge graph.

[0129] Furthermore, embodiments of the present invention also provide a computing device, the computing device comprising: at least one processor, and a memory, wherein the memory stores instructions executable by the processor, the instructions being executed by the processor, thereby enabling the processor to perform the above-described operations. Figure 1-2 The method for updating the knowledge graph of substation main equipment as described in any one of the following.

[0130] Furthermore, embodiments of the present invention also provide a readable storage medium for storing a computer program, wherein the computer program, when running, controls the device where the storage medium is located to perform the above-described actions. Figure 1-2 The method for updating the knowledge graph of substation main equipment as described in any one of the following.

[0131] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0132] It is understood that the relevant features in the above methods and apparatus can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between embodiments and do not represent the superiority or inferiority of any particular embodiment.

[0133] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0134] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention. Additionally, the memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory may include at least one memory chip.

[0135] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0140] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0141] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for updating a knowledge graph of substation main equipment, characterized in that, include: Obtain a knowledge graph of the main substation equipment. The nodes of the knowledge graph are the main substation equipment entities, the feature vectors of the nodes contain multimodal data of the equipment, and the edges of the knowledge graph are the line connection relationships between the main substation equipment. The multimodal data includes equipment physical topology logical data and operation and maintenance rule data. Based on the message passing mechanism of the device physical topology logic data and the power physical topology layer graph neural network, the device physical topology logic state data is determined. Based on operation and maintenance rule data and operation and maintenance dependency rules, determine the operation and maintenance status data; Based on the physical topology logical state data and operation and maintenance status data of the equipment, calculate the state change of the substation main equipment within a preset time. When the state change amount is greater than the corresponding preset amount, it is determined that an event update has occurred in the substation main equipment; When an event update occurs in the main substation equipment, starting from the main substation equipment, the affected equipment is determined based on the line connection relationship between the main substation equipment. The main substation equipment and the affected equipment are combined into a state change event set. The line connection relationship is obtained based on the line topology data. Based on the state change event set, the influence radius and subgraph are determined. The subgraph is a local knowledge graph determined based on the influence radius. Based on the subgraph, local recalculation is performed to obtain the subgraph reconstruction result. Based on the subgraph reconstruction result and the knowledge graph, a target knowledge graph is generated.

2. The method according to claim 1, characterized in that, The operation and maintenance rule data includes operation and maintenance text data; the equipment physical topology logical data includes equipment status data and line topology data; and the acquisition of the knowledge graph of the substation main equipment includes: The operation and maintenance text data is input into a pre-trained model for enhancing the knowledge of substation main equipment based on the BERT framework to determine the semantic features of the text. The device status data is analyzed using a sliding window to determine the device's timing status characteristics. Based on the power feature enhancement graph neural network combined with the line topology data, the graph structure features are determined; The text semantic features, the device temporal state features, and the graph structure features are mapped to a spatial vector of the target dimension to obtain a fused feature vector. Based on the fused feature vectors, the connection relationships between substation main equipment in the line topology data, and the substation main equipment, a knowledge graph is constructed.

3. The method according to claim 2, characterized in that, The calculation of the state change of the main substation equipment within a preset time period based on the physical topology logical state data and operation and maintenance status data of the equipment includes: The fused feature vector is input into a two-layer graph neural network model to obtain the data output by the physical topology layer and the data output by the logical constraint layer in the two-layer graph neural network model. Based on the correlation between the data output by the physical topology layer and the data output by the logical constraint layer, a first weight and a second weight are determined. The first weight is the proportion of the physical topology logical data of the device in the determination of the device status, and the second weight is the proportion of the operation and maintenance rule data in the determination of the substation main equipment status. Based on the device physical topology logical state data, first weight, operation and maintenance status data and second weight, a first feature fusion value is determined. The first feature fusion value represents the status information of the substation main equipment at the first time. The first time is the moment corresponding to the minimum value of the feature fusion value within a preset time. Based on the device physical topology logical state data, the first weight, the operation and maintenance status data, and the second weight at the second time, a second feature fusion value is determined. The second feature fusion value represents the state information of the substation main equipment at the second time. The second time is the moment corresponding to the maximum value of the feature fusion value within a preset time, and the time interval between the first time and the second time is greater than a threshold. Calculate the difference between the first feature fusion value and the second feature fusion value, and determine that the difference is the state change amount within a preset time period.

4. The method according to claim 3, characterized in that, The feature fusion value includes a first feature fusion value and a second feature fusion value, and the method includes: The formula for the message passing mechanism of the power physics topology graph neural network is as follows: h_i^(l+1) = σ( ) in, Let h_i^(l+1) be the set of all neighboring nodes directly electrically connected to node i, h_j^(l) be the feature vector of node i in the (l+1)th layer of the network, h_j^(l) be the feature vector of node j adjacent to node i in the lth layer of the network, W_phy be the trainable weight matrix, which is the electrical parameter correction factor used to extract power topology features, b_phy be the bias term to enhance the expressive power of the model, σ be the activation function, which performs nonlinear activation on the aggregation result, and W_phy•h_j^(l) be the mapping of neighbor features to a high-dimensional space through the weight matrix W_phy to capture the electrical interaction patterns between devices. The formula for the operation and maintenance dependency rule is as follows: constraint_matrix[i,j] = f(rule_i, state_j) Wherein, rule_i is the i-th operation procedure, state_j is the current state of device j, the mapping function f implements the transformation from text rules to computable logic, and constraint_matrix is ​​a sparse matrix, the element value [i,j] represents the constraint strength of the i-th rule on device j, and is used to represent the operation dependency relationship between devices reflected in the matrix structure; The formula for calculating the first weight is: gate = σ(W_g•[h_phy||h_log]) The formula for calculating the feature fusion value is: h_final = gate⊙h_phy + (1-gate)⊙h_log Where gate is the first weight, the second weight is 1-gate, h_phy is the data output by the physical topology layer, h_log is the data output by the logical constraint layer, [h_phy||h_log] is the concatenation of the data output by the physical topology layer and the data output by the logical constraint layer, W_g is the trainable parameter matrix that learns the correlation between the two types of features, σ is the Sigmoid activation function that compresses the output to the [0,1] interval, and ⊙ is the element-wise multiplication.

5. The method according to claim 1, characterized in that, After generating the target knowledge graph, the method further includes: Extract the main substation equipment entity nodes and the line connection edges between nodes from the target knowledge graph; Based on graph structure analysis algorithm, a device topology adjacency matrix is ​​constructed by combining the entity nodes of substation main equipment and the line connection edges between nodes. The elements in the device topology adjacency matrix mark the physical connection status between substation main equipment and the substation main equipment directly associated with the substation main equipment entity. Starting with any substation main equipment, a breadth-first search algorithm is used to traverse the equipment topology network in the target knowledge graph, and the set of substation main equipment entity nodes traversed is recorded. The set of entity nodes of the main substation equipment is compared with the equipment topology adjacency matrix to determine whether there are any isolated nodes that have not been traversed. At the same time, the illegal connection paths are checked during the traversal process to determine whether the connection between devices in the target knowledge graph is abnormal.

6. The method according to claim 1, characterized in that, After generating the target knowledge graph, the method further includes: Based on the target knowledge graph, extract the predicate rule set generated by compiling the substation main equipment operation procedures and the operation status data of each substation main equipment entity; According to the constraints of each operation rule in the predicate rule set, match them one by one with the operating status data of the corresponding main equipment entity to determine whether the current status of the equipment meets the preconditions and postconditions of the operation rule. The device status and operation records that do not meet the constraints of the operation rules are marked, a logic verification anomaly report is generated, and the logic verification of the target knowledge graph is completed.

7. A device for updating a knowledge graph of substation main equipment, characterized in that, include: The acquisition unit is used to acquire the knowledge graph of the substation main equipment. The nodes of the knowledge graph are substation main equipment entities, the feature vectors of the nodes contain equipment multimodal data, the edges of the knowledge graph are the line connection relationships between substation main equipment, and the multimodal data includes equipment physical topology logical data and operation and maintenance rule data. The determination unit is used to determine the physical topology logic state data of the equipment based on the message passing mechanism of the equipment physical topology logic data and the power physical topology layer graph neural network in the acquisition unit. The determining unit is used to determine the operation and maintenance status data based on the operation and maintenance operation rule data and operation and maintenance operation dependency relationship rules in the acquisition unit; The calculation unit is used to calculate the state change of the main substation equipment within a preset time period based on the equipment physical topology logical state data and operation and maintenance status data in the determination unit. The unit is used to determine that an event update has occurred in the main substation equipment when the state change amount in the calculation unit is greater than the corresponding preset amount. The obtaining unit is used to, when an event update occurs in the substation main equipment, determine the affected equipment based on the line connection relationship between the substation main equipment, starting from the substation main equipment, and form a state change event set by combining the substation main equipment and the affected equipment. The line connection relationship is obtained based on line topology data. Based on the state change event set, the influence radius and subgraph are determined. The subgraph is a local knowledge graph determined based on the influence radius. Based on the subgraph, local recalculation is performed to obtain the subgraph reconstruction result. Based on the subgraph reconstruction result and the knowledge graph, a target knowledge graph is generated.

8. A storage medium comprising a stored program, characterized in that, When the program is running, it controls the device where the storage medium is located to execute the method for updating the knowledge graph of the substation main equipment as described in any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for updating the knowledge graph of substation main equipment as described in any one of claims 1 to 6.

Images