Power grid analysis reasoning method and system based on multi-source intelligent map enhancement

Abstract

The invention provides a power grid analysis and reasoning method and system based on multi-source intelligent map enhancement, the method is applied to a power grid analysis and reasoning platform, and a candidate fault reasoning chain is generated by inputting an input power grid fault text into a main reasoning model. Physical compliance verification is carried out on the candidate fault reasoning chain through a dynamic equation discriminator constructed based on a power system dynamic equation, a power grid fault analysis reasoning result is output, and a graph attention network is integrated in the main reasoning model, so that a power grid fault analysis reasoning result is obtained. The cross-modal feature enhancement module is used for carrying out cross-modal feature enhancement on text features of a power grid fault text according to the power multi-source intelligent causal atlas in the main inference model, and a side relationship of the power multi-source intelligent causal atlas at least comprises the following causal relationships: physical connection, triggering conditions and time sequence dependence. The problem that causal errors and physical violation are likely to occur in fault reasoning due to multi-source data splitting, marking scarcity and lack of dynamic physical constraints of an existing power grid analysis platform is solved.

Description

Technical Field

[0001] This invention relates to the field of computers, and specifically to a power grid analysis and reasoning method and system based on multi-source intelligent graph enhancement. Background Technology

[0002] Traditional power grid analysis primarily relies on numerical simulations based on differential-algebraic equations (such as electromagnetic transient simulations and electromechanical transient simulations). These methods are based on physical laws and have high computational reliability given an accurate model and parameters. However, their computational process is complex and time-consuming, and they are inherently unable to directly process and understand unstructured natural language fault report data.

[0003] In recent years, deep learning techniques, especially natural language processing methods based on pre-trained language models (such as BERT), have been introduced into the field of power grid fault text analysis for tasks such as fault classification and key information extraction. However, these models are essentially data-driven probabilistic models, whose optimization objective is to maximize the likelihood probability of a text sequence rather than capturing the inherent causal mechanisms of the physical system. This can lead to models that generate results that violate the causal timing and physical constraints of the actual power system (e.g., predicting that "circuit breaker tripping" occurs before "line overload").

[0004] Among existing improvements, the closest to this invention is the Graph Neural Network (GNN) method that combines a power grid topology knowledge graph. This method introduces domain knowledge to some extent by injecting static topological structures such as the connection relationships of power grid devices into the model in the form of a knowledge graph. However, its main limitations are: static limitation: the graph content mainly focuses on static information such as the physical connections of devices, failing to effectively embed causal rules reflecting the dynamic propagation process of events (such as triggering conditions for state transitions and temporal dependencies between events). Lack of physical constraints: it lacks explicit modeling and constraints on the core dynamic equations of the power system (such as differential-algebraic equations describing transient processes and temporal logic equations for protection actions), making it impossible to ensure the compliance of the model output at the physical level.

[0005] Therefore, it is of great significance to address the problems of fragmented multi-source data, scarce labeling, and lack of dynamic physical constraints in existing power grid analysis platforms, which lead to causal errors and physical violations in fault reasoning. Summary of the Invention

[0006] To address the problems of fragmented multi-source data, scarce annotations, and lack of dynamic physical constraints in existing power grid analysis platforms, which lead to causal errors and physical violations in fault reasoning, this invention proposes a power grid analysis reasoning method and system based on multi-source intelligent graph enhancement.

[0007] Firstly, a power grid analysis and reasoning method based on multi-source intelligent graph enhancement is provided, applied to a power grid analysis and reasoning platform, including: The input power grid fault text is fed into the main inference model to generate a candidate fault inference chain; The candidate fault reasoning chain is physically verified by a dynamic equation discriminator constructed based on the dynamic equation of the power system, and the power grid fault analysis reasoning results are output. The main inference model integrates a graph attention network, which is used to perform cross-modal feature enhancement on the text features of power grid fault text based on the power multi-source intelligent causal graph in the main inference model. The edge relationships of the power multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition and temporal dependency.

[0008] Secondly, a power grid analysis and inference system based on multi-source intelligent graph enhancement is provided, which is applied to a power grid analysis and inference platform, including: The generation module is used to input the power grid fault text into the main inference model and generate candidate fault inference chains; The output module is used to perform physical compliance verification on the candidate fault inference chain through a dynamic equation discriminator constructed based on the dynamic equation of the power system, and output the power grid fault analysis inference results. The main inference model integrates a graph attention network, which is used to perform cross-modal feature enhancement on the text features of power grid fault text based on the power multi-source intelligent causal graph in the main inference model. The edge relationships of the power multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition and temporal dependency.

[0009] In another aspect, this application also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a power grid analysis reasoning method based on multi-source intelligent graph enhancement as described above is implemented.

[0010] In another aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the power grid analysis and reasoning method based on multi-source intelligent graph enhancement as described above.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a power grid analysis and reasoning method and system based on multi-source intelligent graph enhancement. The method is applied to a power grid analysis and reasoning platform. By inputting power grid fault text into the main reasoning model, candidate fault reasoning chains are generated. Then, a dynamic equation discriminator based on the power system dynamic equations performs physical compliance verification on these candidate fault reasoning chains, outputting the power grid fault analysis and reasoning results. Furthermore, the main reasoning model integrates a graph attention network to perform cross-modal feature enhancement on the textual features of the power grid fault text based on the multi-source intelligent causal graph in the main reasoning model. The edge relationships in this multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition, and temporal dependency. This addresses the problems of fragmented multi-source data, scarce annotations, and lack of dynamic physical constraints in existing power grid analysis platforms, which easily lead to causal errors and physical violations in fault reasoning.

[0012] Specifically, the construction of multi-source power grid causal graphs results in fewer causal time reversals or logical contradictions in the generated event and fault propagation chains. The inference conclusions are semantically and temporally closer to the actual engineering situation, making them easier for dispatch and maintenance personnel to understand and adopt.

[0013] Enhanced data efficiency and robustness to small samples: Prior and cross-modal alignment enhanced by multi-source intelligent graphs reduce the dependence on large-scale labeled data, enabling the model to maintain stable performance in scenarios with scarce labels and cross-domain migration, and reducing the sample cost of platform deployment.

[0014] Physical compliance assurance: The dynamic equation discriminator verifies candidate chains on both electrical and temporal scales, effectively filtering out physically infeasible analytical conclusions and improving the engineering verifiability of the platform's output.

[0015] All of the above prediction results can be clearly traced back to the specific reasoning path in the causal graph and the physical verification process of the dynamic equation discriminator, which significantly enhances the transparency and credibility of the model decision and meets the requirements of traceability and verifiability for power fault analysis. Attached Figure Description

[0016] Figure 1 This is a flowchart of the power grid analysis and reasoning method based on multi-source intelligent graph enhancement of the present invention; Figure 2 This is a schematic diagram illustrating the specific process of the power grid analysis and reasoning method based on multi-source intelligent graph enhancement of the present invention. Figure 3 This is a schematic diagram of the multi-source intelligent causal graph construction process of the power grid analysis and reasoning method based on multi-source intelligent graph enhancement of the present invention; Figure 4 This is a schematic diagram of the graph enhancement training process for the power grid analysis and reasoning method based on multi-source intelligent graph enhancement of the present invention; Figure 5 This is a schematic diagram of the dynamic equation discriminator of the power grid analysis and reasoning method based on multi-source intelligent graph enhancement of the present invention; Figure 6 This is a schematic diagram of the power grid analysis and reasoning system based on multi-source intelligent graph enhancement according to the present invention; Figure 7 This is a schematic diagram of an electronic device structure according to the present invention. Detailed Implementation

[0017] While existing power grid graphics platforms can display equipment topology and operating status, they lack the ability to enhance multi-source intelligent graphs, making it difficult to uniformly capture and follow the causal logic of the power system under cross-system data fusion conditions. Furthermore, the amount of real-world labeled data for power grid events is limited, and traditional methods struggle to train robust analytical models with small sample sizes, affecting the platform's universality in complex scenarios. Current multi-source fusion and knowledge graph methods mostly remain at the level of static display, lacking constraints on power dynamic equations (such as transient equations and protection action logic), which may cause the analytical reasoning results generated by the platform to not conform to physical laws.

[0018] The existing technology has the following specific problems: 1. Causal Mechanism Deficiencies: Existing data-driven models (including various pre-trained language models and statistical learning methods) typically learn the correlations between observed data, rather than the explicitly defined causal relationships within the power system. When applying these models to power grid events or fault inference, they tend to build event chains based solely on surface semantics or statistical correlations, neglecting the triggering conditions and temporal dependencies between events. This often leads to causal inversions, logical breaks, or omissions of crucial intermediate steps in the inference chain. For multi-source event streams presented on graphical platforms, this deficiency makes the event propagation paths provided by the platform lack engineering interpretability and decision-making value.

[0019] 2. Data scarcity amplifies shortcomings: Power grid operation-related data is typically scattered across multiple systems such as GIS, SCADA, protection device records, operation and maintenance logs, and procedural texts, resulting in heterogeneous formats and scarce labeled samples. Relying solely on one type of data or making minor adjustments to a pre-trained model is insufficient to cover the diverse fault modes and cross-domain causal chains in real-world scenarios. Consequently, the model's performance deteriorates sharply when encountering cross-regional linkages, atypical operating conditions, or small sample scenarios, leading to unstable or misleading analytical conclusions in platform-level applications.

[0020] 3. Inaccurate Physical Laws: Most existing methods for injecting topological information into models remain at the level of static connectivity, lacking the explicit conversion of key dynamic physical laws that determine fault evolution (such as power / energy conservation, transient process equations, protection action time limits and trigger thresholds) into constraint signals for training or inference. Therefore, even if the generated causal chain is semantically reasonable, it may violate physical feasibility in terms of electrical quantity changes or time scales, making the platform output unverifiable in engineering and reducing the platform's usability and reliability in operation, maintenance, and scheduling scenarios.

[0021] Based on the aforementioned deficiencies in the existing technology, the overall objective of this invention is to address the core shortcomings of the existing technology within the framework of a "multi-source smart grid graphical platform," namely, the lack of causal logic, poor adaptability due to data scarcity, and insufficient constraints from physical laws. This invention aims to propose an analytical reasoning method suitable for multi-source smart grid graphical platforms. Enhanced causal logic: Construct and maintain a power causal graph that includes rules for triggering devices, states, and events, and make fault propagation mechanisms, protection coordination logic, and timing dependencies explicit, serving as structured priors for platform reasoning, and ensuring that the reasoning chain conforms to engineering common sense in terms of semantics and timing.

[0022] Data adaptation and small sample augmentation: Through multi-source intelligent map augmentation, cross-modal alignment and multi-task training strategies, the generalization ability of the model in scenarios with scarce annotations and cross-domain scenarios is improved, enabling the platform to make full use of multi-source data from GIS, SCADA, protection records and other sources.

[0023] Physical constraints: The dynamic equations and protection / control timing equations of the power system are transformed into computable constraint modules (differentiable discriminators). During the model training and inference phases, physical consistency checks are performed on candidate inference chains and adversarial corrections are provided to ensure that the output meets the basic electrical quantity and time scale constraints.

[0024] The core objective of this framework is to significantly improve the causal accuracy, physical compliance, and data utilization efficiency of power grid fault reasoning models under low-labeled data conditions, through the constraints of structured causal knowledge and explicit physical equation discriminators.

[0025] To better understand the present invention, the following description, in conjunction with the accompanying drawings and embodiments, will further illustrate the content of the present invention.

[0026] Example 1: A power grid analysis and reasoning method based on multi-source intelligent graph enhancement is applied to a power grid analysis and reasoning platform, such as... Figure 1 As shown, it includes: Step 1: Input the power grid fault text into the main inference model to generate candidate fault inference chains; Step 2: Using a dynamic equation discriminator built based on the dynamic equations of the power system, the physical compliance of the candidate fault reasoning chain is verified, and the power grid fault analysis reasoning results are output. The main inference model integrates a graph attention network, which is used to perform cross-modal feature enhancement on the text features of power grid fault text based on the power multi-source intelligent causal graph in the main inference model. The edge relationships of the power multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition and temporal dependency.

[0027] In this embodiment, before executing the power grid analysis and reasoning method based on multi-source intelligent graph enhancement of the present invention, it is necessary to construct a power grid analysis and reasoning platform, which specifically includes: The input historical power grid fault text is input into the main inference model of the power grid analysis and inference platform, and the initial training features are generated after being encoded by the pre-trained language model in the main inference model. Based on the initial training features, cross-modal feature interaction and enhancement are performed through the graph attention network in the main inference model and the pre-constructed multi-source intelligent causal graph to generate candidate training fault causal chains. The physical consistency of the candidate training fault causal chain is verified by using a pre-built dynamic equation discriminator, and the parameters of the main inference model are corrected by adversarial training until the power grid fault analysis inference training result that simultaneously satisfies the constraints of causal logic and physical equation is output, thus obtaining the corrected main inference model. The power grid analysis and reasoning platform is constructed based on the modified master inference model and the dynamic equation discriminator.

[0028] Specifically, the construction of a multi-source intelligent knowledge graph integrates the correlation of power grid equipment, fault propagation mechanisms, and protection action logic, and introduces multi-source information such as cross-modal text, time-series records, and procedural knowledge to achieve the construction of an enhanced knowledge graph with multi-level causal relationships. This includes: Obtain the physical connection relationships of equipment from the power grid GIS system as the basic topology; By integrating fault mode and effect analysis data, transient analysis theoretical texts, and protection coordination logic in scheduling procedures, dynamic causal rules are extracted. Based on the aforementioned basic topology and dynamic causal rules, an attribute graph model is used to represent the graph. Nodes record device type, identifier, and state variables, while edge relationships are labeled with causal type and weight parameters.

[0029] The process of extracting dynamic causal rules by integrating failure mode and effect analysis data, transient analysis theoretical texts, and protection coordination logic in scheduling procedures specifically includes: The system automatically analyzes equipment ledgers and monitoring data to generate nodes, which are then verified by human experience. By using text causal extraction, temporal correlation analysis, and topology matching algorithms, edge relationships are automatically discovered and associated with propagation probability or typical time delay parameters. The nodes and the edge relationships constitute dynamic causal rules.

[0030] In one specific embodiment, the multi-source intelligent causal graph transforms power domain knowledge into a structured representation. The content not only includes traditional device connection topologies but, more importantly, introduces dynamic causal chains for fault propagation. Graph nodes encompass primary devices (such as circuit breakers and busbars), secondary devices (such as relay protection devices), and state variables (such as voltage and current). Edge relationship definitions include three types of causal labels: "physical connection," "triggering condition," and "temporal dependency," used to accurately describe the causal relationships between devices and events.

[0031] In this embodiment, during the construction of the power grid analysis and inference platform, cross-modal feature interaction and enhancement can be performed through the pre-trained graph attention network in the main inference model and the multi-source intelligent causal graph constructed in the aforementioned steps, thereby obtaining candidate training fault causal chains. This provides a basis for optimizing the parameters of the main inference model by performing physical consistency checks through a pre-constructed dynamic equation discriminator. Specifically, this includes: Based on the initial training features, the pre-built multi-source intelligent causal graph is queried through the pre-trained graph attention network in the main inference model, and the attention weights of the initial features and the node features in the multi-source intelligent causal graph are calculated. The graph attention network is trained through a cross-modal contrastive learning strategy. The initial features are fused and corrected based on the attention weights to obtain an enhanced text feature vector. Based on the enhanced text feature vectors and the multi-source intelligent causal graph, candidate training fault causal chains are generated through the master inference model.

[0032] In one specific embodiment, a graph attention network (GAT) is used to modify and enhance the text features output by a pre-trained language model (such as BERT). A cross-modal contrastive learning strategy is used to align the semantic description of the text with the causal paths in the graph. For example, the description of "insulation breakdown" in a fault report is mapped and associated with the dynamic causal chain of "insulation degradation → partial discharge → arc development" in the graph, enabling the model to understand the causal logic behind the text.

[0033] It should be noted that the process of generating the candidate fault inference chain in step 1 above is similar to the process of generating the candidate training fault inference chain described above, and will not be repeated here.

[0034] In this embodiment, after generating the candidate fault inference chain in step 1, the physical compliance verification can be performed based on the candidate fault inference chain using a pre-built dynamic equation discriminator, and the power grid fault analysis inference results can be output. Specifically, this includes: The event sequence, timestamps, and electrical quantity parameters of the candidate fault causal chain are received through a differentiable computation graph in a dynamic equation discriminator constructed based on the dynamic equations of the power system. The physical compliance score of the candidate fault causal chain is calculated based on the event sequence, timestamp, and electrical quantity parameters. Based on the physical compliance score, a result filtering operation is performed, and the power grid fault analysis reasoning result with the highest physical compliance score is output.

[0035] Before obtaining the power grid fault analysis and inference results through physical compliance verification, a dynamic equation discriminator needs to be pre-built, which specifically includes: The implicit trapezoidal integration method is used to discretize the differential algebraic equations of the power system into difference equations; Based on the difference equation, a differentiable computational graph is constructed using an automatic differentiation tool.

[0036] In one specific embodiment, the dynamic equation discriminator specifically includes the following: Implicit trapezoidal rule is used to discretize the transiently stable differential-algebraic equations (DAEs) over time, constructing a differentiable computational graph consisting of residual nodes and algebraic constraint nodes at each time step. This graph serves as a discriminator to verify the physical compliance of the fault sequences predicted by the model. Through an adversarial training mechanism, the discriminator provides gradient feedback to the generator (i.e., the main inference model), thereby prompting the generator's output to satisfy the constraints of the dynamic equations (e.g., the time interval between events must meet the requirements of the protection action equations). This constraint mechanism enables the model to automatically identify fault sequences that do not conform to the laws of physics, thus improving the reliability and stability of the inference results.

[0037] Example 2: The following example, a complete power grid system event analysis process, illustrates the implementation of this solution. The system input is the natural language fault alarm text received by the dispatch center (e.g., "35kV bus voltage abnormal fluctuation, followed by line protection tripping"). First, the text is encoded by a pre-trained language model to obtain an initial text feature vector, which is then input to the graph attention interaction layer. This layer queries a pre-built multi-source intelligent causal graph (combining GIS, FMEA, regulations, and other multi-source knowledge) to locate the causal path related to the input text, such as "voltage fluctuation → protection device measurement element sensing → protection logic judgment → tripping signal output -> circuit breaker tripping". The model integrates text features and graph information to generate possible fault development chains. The dynamic equation discriminator performs physical compliance checks on the generated results: for example, verifying whether the time interval from the detection of "voltage fluctuation" to the issuance of "protection action" meets the inherent action time limit equation requirements of the relay protection equipment, or whether the electrical quantity changes comply with the constraints of the transient stability equation. Predictions that comply with physical constraints are reinforced; those that do not are corrected through gradient feedback. This process... Figure 2 The overall flowchart shown is complete.

[0038] 2.1 Explanation of core data processing steps: (1) Construction of causal graph: A multi-source data fusion strategy is adopted. The basic topology information comes from the power grid GIS system, which includes the physical connection relationships of equipment such as substations, lines, and switches.

[0039] The dynamic causal logic is integrated from three aspects: first, the Fault Mode and Effects Analysis (FMEA) data explicitly listed in the technical specifications of power equipment manufacturers; second, the standard fault propagation theory in power system transient analysis textbooks; and third, the protection coordination principles and timing requirements stipulated in dispatching and operation procedures.

[0040] The extraction of graph nodes and edge relationships adopts a human-machine collaborative approach: nodes are automatically generated by the system through parsing equipment ledgers and monitoring data, and are reviewed by experts; edge relationships are automatically discovered based on text causal extraction, temporal correlation analysis, and topological matching.

[0041] The extracted results are organized using an attribute graph model. Nodes record device type, ID, and key state variables (such as voltage, current, and switch status). Edge relationships are labeled with causal type (such as "electrical connection," "timing-triggered," and "logical dependency") and weight parameters (such as propagation probability and typical delay). See the core process below. Figure 3 .

[0042] (2) Graph Augmentation Training: This is achieved through a cross-modal contrastive learning strategy. For each training sample's fault description text, the model needs to simultaneously complete three collaborative tasks: first, predict the corresponding node in the graph for the device / state mentioned in the text; second, reconstruct the causal path of the fault process described in the text in the graph; and third, distinguish between real causal chains that conform to power logic and randomly generated false causal chains. These three tasks are jointly optimized in a multi-task learning framework, enabling the model to establish an accurate mapping relationship between text words and graph nodes, while internalizing the causal reasoning rules of the power system, significantly improving the generalization ability in data-scarce scenarios. See the core process below. Figure 4 .

[0043] (3) Discriminator implementation: It includes three key steps: First, the implicit trapezoidal integral method is used to discretize the differential algebraic equation of the power system into a difference equation; then, an automatic differentiation tool (such as PyTorch's Autograd) is used to construct a differentiable computation graph; finally, a physical verification layer is added to the model output to receive the event sequence, timestamp and electrical quantity parameters of the predicted fault chain and calculate whether it conforms to the constraints of the discretized equation.

[0044] The discriminator's output is fed back to the main model as a regularization loss or adversarial signal, forcing the generated results to conform to physical laws during training. See the core process below. Figure 5 .

[0045] This invention achieves the following technical effects through the dual constraints of structured causal knowledge graphs and physical equations: (1) Multi-source power grid cause-effect graph construction module: The generated event and fault propagation chain has fewer causal time reversals or logical contradictions, and the reasoning conclusions are closer to the actual situation of the project in terms of semantics and time sequence, which is convenient for dispatch and maintenance personnel to understand and adopt.

[0046] (2) Enhanced data efficiency and robustness of small samples: The prior and cross-modal alignment enhanced by multi-source intelligent graphs reduce the dependence on large-scale labeled data, enabling the model to maintain stable performance in scenarios with scarce labels and cross-domain migration, and reducing the sample cost of platform deployment.

[0047] (3) Physical compliance guarantee: The dynamic equation discriminator verifies the candidate chain in terms of electrical quantity and time scale, which can effectively filter out analysis conclusions that are physically infeasible and improve the engineering verifiability of the platform output.

[0048] All of the above prediction results can be clearly traced back to the specific reasoning path in the causal graph and the physical verification process of the dynamic equation discriminator, which significantly enhances the transparency and credibility of the model decision and meets the requirements of traceability and verifiability for power fault analysis.

[0049] Example 3: Based on the same inventive concept, this invention also provides a power grid analysis and reasoning system based on multi-source intelligent graph enhancement, applied to a power grid analysis and reasoning platform, such as... Figure 6 As shown, it includes: The generation module is used to input the power grid fault text into the main inference model and generate candidate fault inference chains; The output module is used to perform physical compliance verification on the candidate fault inference chain through a dynamic equation discriminator constructed based on the dynamic equation of the power system, and output the power grid fault analysis inference results. Among them, the main inference model integrates a graph attention network, which is used to perform cross-modal feature enhancement on the text features of power grid fault text based on the power multi-source intelligent causal graph in the main inference model. The edge relationships of the power multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition and temporal dependency.

[0050] Preferably, the system further includes: The construction module is used to input the historical power grid fault text into the main inference model of the power grid analysis and inference platform, and generate initial training features after encoding by the pre-trained language model in the main inference model; Based on the initial training features, cross-modal feature interaction and enhancement are performed through the graph attention network in the main inference model and the pre-constructed multi-source intelligent causal graph to generate candidate training fault causal chains. The physical consistency of the candidate training fault causal chain is verified by using a pre-built dynamic equation discriminator, and the parameters of the main inference model are corrected by adversarial training until the power grid fault analysis inference training result that simultaneously satisfies the constraints of causal logic and physical equation is output, thus obtaining the corrected main inference model. The power grid analysis and reasoning platform is constructed based on the modified master inference model and the dynamic equation discriminator.

[0051] Preferably, the construction of the power multi-source intelligent causal graph in the construction module includes: Obtain the physical connection relationships of equipment from the power grid GIS system as the basic topology; By integrating fault mode and effect analysis data, transient analysis theoretical texts, and protection coordination logic in scheduling procedures, dynamic causal rules are extracted. Based on the aforementioned basic topology and dynamic causal rules, an attribute graph model is used to represent the graph. Nodes record device type, identifier, and state variables, while edge relationships are labeled with causal type and weight parameters.

[0052] Preferably, the dynamic causal rules are extracted from the construction module, including: The system automatically analyzes equipment ledgers and monitoring data to generate nodes, which are then verified by human experience. By using text causal extraction, temporal correlation analysis, and topology matching algorithms, edge relationships are automatically discovered and associated with propagation probability or typical time delay parameters. The nodes and the edge relationships constitute dynamic causal rules.

[0053] Preferably, the generation of candidate training fault causal chains in the construction module includes: Based on the initial training features, the pre-built multi-source intelligent causal graph is queried through the pre-trained graph attention network in the main inference model, and the attention weights of the initial features and the node features in the multi-source intelligent causal graph are calculated. The graph attention network is trained through a cross-modal contrastive learning strategy. The initial features are fused and corrected based on the attention weights to obtain an enhanced text feature vector. Based on the enhanced text feature vectors and the multi-source intelligent causal graph, candidate training fault causal chains are generated through the master inference model.

[0054] Preferably, the construction of the dynamic equation discriminator in the construction module or the output module includes: The implicit trapezoidal integration method is used to discretize the differential algebraic equations of the power system into difference equations; Based on the difference equation, a differentiable computational graph is constructed using an automatic differentiation tool.

[0055] Preferably, the output module is further configured to: The event sequence, timestamps, and electrical quantity parameters of the candidate fault causal chain are received through a differentiable computation graph in a dynamic equation discriminator constructed based on the dynamic equations of the power system. The physical compliance score of the candidate fault causal chain is calculated based on the event sequence, timestamp, and electrical quantity parameters. Based on the physical compliance score, a result filtering operation is performed, and the power grid fault analysis reasoning result with the highest physical compliance score is output.

[0056] Example 4 like Figure 7 As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0057] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to realize the corresponding method flow or corresponding function, so as to realize the steps of the power grid analysis and reasoning method based on multi-source intelligent graph enhancement in the above embodiments.

[0058] Example 5 Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the power grid analysis and reasoning method based on multi-source intelligent graph enhancement in the above embodiments.

[0059] 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.

[0060] 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.

[0061] 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.

[0062] 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.

[0063] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A power grid analysis and reasoning method based on multi-source intelligent graph enhancement, characterized in that, Applications include power grid analysis and inference platforms, including: The input power grid fault text is fed into the main inference model to generate a candidate fault inference chain; The candidate fault reasoning chain is physically verified by a dynamic equation discriminator constructed based on the dynamic equation of the power system, and the power grid fault analysis reasoning results are output. The main inference model integrates a graph attention network, which is used to perform cross-modal feature enhancement on the text features of power grid fault text based on the power multi-source intelligent causal graph in the main inference model. The edge relationships of the power multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition and temporal dependency.

2. The method according to claim 1, characterized in that, The construction of the power grid analysis and inference platform includes: The input historical power grid fault text is input into the main inference model of the power grid analysis and inference platform, and the initial training features are generated after being encoded by the pre-trained language model in the main inference model. Based on the initial training features, cross-modal feature interaction and enhancement are performed through the graph attention network in the main inference model and the pre-constructed multi-source intelligent causal graph to generate candidate training fault causal chains. The physical consistency of the candidate training fault causal chain is verified by using a pre-built dynamic equation discriminator, and the parameters of the main inference model are corrected by adversarial training until the power grid fault analysis inference training result that simultaneously satisfies the constraints of causal logic and physical equation is output, thus obtaining the corrected main inference model. The power grid analysis and reasoning platform is constructed based on the modified master inference model and the dynamic equation discriminator.

3. The method according to claim 1 or 2, characterized in that, The construction of the power multi-source intelligent causal graph includes: Obtain the physical connection relationships of equipment from the power grid GIS system as the basic topology; By integrating fault mode and effect analysis data, transient analysis theoretical texts, and protection coordination logic in scheduling procedures, dynamic causal rules are extracted. Based on the aforementioned basic topology and dynamic causal rules, an attribute graph model is used to represent the graph. Nodes record device type, identifier, and state variables, while edge relationships are labeled with causal type and weight parameters.

4. The method according to claim 3, characterized in that, The extraction of dynamic causal rules includes: The system automatically analyzes equipment ledgers and monitoring data to generate nodes, which are then verified by human experience. By using text causal extraction, temporal correlation analysis, and topology matching algorithms, edge relationships are automatically discovered and associated with propagation probability or typical time delay parameters. The nodes and the edge relationships constitute dynamic causal rules.

5. The method according to claim 2, characterized in that, The generation of candidate training fault causal chains includes: Based on the initial training features, the pre-built multi-source intelligent causal graph is queried through the pre-trained graph attention network in the main inference model, and the attention weights of the initial features and the node features in the multi-source intelligent causal graph are calculated. The graph attention network is trained through a cross-modal contrastive learning strategy. The initial features are fused and corrected based on the attention weights to obtain an enhanced text feature vector. Based on the enhanced text feature vectors and the multi-source intelligent causal graph, candidate training fault causal chains are generated through the master inference model.

6. The method according to claim 1 or 2, characterized in that, The construction of the dynamic equation discriminator includes: The implicit trapezoidal integration method is used to discretize the differential algebraic equations of the power system into difference equations; Based on the difference equation, a differentiable computational graph is constructed using an automatic differentiation tool.

7. The method according to claim 1, characterized in that, The dynamic equation discriminator, constructed based on the dynamic equations of the power system, performs physical compliance verification on the candidate fault inference chain and outputs the power grid fault analysis inference results, including: The event sequence, timestamps, and electrical quantity parameters of the candidate fault causal chain are received through a differentiable computation graph in a dynamic equation discriminator constructed based on the dynamic equations of the power system. The physical compliance score of the candidate fault causal chain is calculated based on the event sequence, timestamp, and electrical quantity parameters. Based on the physical compliance score, a result filtering operation is performed, and the power grid fault analysis reasoning result with the highest physical compliance score is output.

8. A power grid analysis and reasoning system based on multi-source intelligent graph enhancement, characterized in that, Applications include power grid analysis and inference platforms, including: The generation module is used to input the power grid fault text into the main inference model and generate candidate fault inference chains; The output module is used to perform physical compliance verification on the candidate fault inference chain through a dynamic equation discriminator constructed based on the dynamic equation of the power system, and output the power grid fault analysis inference results. The main inference model integrates a graph attention network, which is used to perform cross-modal feature enhancement on the text features of power grid fault text based on the power multi-source intelligent causal graph in the main inference model. The edge relationships of the power multi-source intelligent causal graph include at least the following causal relationships: physical connection, triggering condition and temporal dependency.

9. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the power grid analysis reasoning method based on multi-source intelligent graph enhancement as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that, It contains an execution program, which, when executed, implements the power grid analysis and reasoning method based on multi-source intelligent graph enhancement as described in any one of claims 1 to 7.

Images