CN-121998076-A - Power grid analysis and reasoning method and system based on multisource intelligent map enhancement

CN121998076ACN 121998076 ACN121998076 ACN 121998076ACN-121998076-A

Abstract

The invention provides a power grid analysis and reasoning method and a system based on multisource intelligent map enhancement, the method is applied to a power grid analysis and reasoning platform, and generating a candidate fault inference chain by inputting the input power grid fault text into the main inference model. And a dynamic equation discriminator constructed based on a dynamic equation of the power system is used for carrying out physical compliance verification on the candidate fault inference chain, outputting a power grid fault analysis inference result, integrating a graph attention network in the main inference model, and carrying out cross-modal feature enhancement on text features of a power grid fault text according to a power multi-source intelligent causal map in the main inference model, wherein the side relationship of the power multi-source intelligent causal map at least comprises the following causal relationship of physical connection, triggering conditions and time sequence dependence, and solves the problems of easy occurrence of causal errors and physical violation of fault inference caused by multi-source data splitting, labeling scarcity and missing dynamic physical constraint of the existing power grid analysis platform.

Inventors

ZHANG WEI
SUI ZHIWEI
ZHAO XINSHUANG
FAN GUOHAO

Assignees

北京国电通网络技术有限公司
国网信息通信产业集团有限公司

Dates

Publication Date: 20260508
Application Date: 20251203

Claims (10)

1. The utility model provides a multisource intelligent map-based enhanced power grid analysis and reasoning method which is characterized by being applied to a power grid analysis and reasoning platform and comprising the following steps: Inputting the input power grid fault text into a main reasoning model to generate a candidate fault reasoning chain; Performing physical compliance verification on the candidate fault reasoning chain through a dynamic equation discriminator constructed based on a dynamic equation of the power system, and outputting a power grid fault analysis reasoning result; The main reasoning model is integrated with a graph attention network and is used for performing cross-modal feature enhancement on text features of a power grid fault text according to the electric power multi-source intelligent causal graph in the main reasoning model, and the side relationship of the electric power multi-source intelligent causal graph at least comprises the following causal relationship of physical connection, triggering conditions and time sequence dependence.
2. The method of claim 1, wherein the building of the grid analysis reasoning platform comprises: inputting the input historical power grid fault text into a main reasoning model of a power grid analysis reasoning platform, and generating initial training characteristics after coding through a pre-trained language model in the main reasoning model; performing cross-modal feature interaction and enhancement through a graph annotation meaning network in a main reasoning model and a pre-constructed multi-source intelligent causal map based on the initial training features, and generating a candidate training fault causal chain; Performing physical consistency verification on the candidate training fault causal link by using a pre-constructed dynamic equation discriminator, and correcting parameters of the main reasoning model through countermeasure training until a power grid fault analysis reasoning training result meeting causal logic and physical equation constraints is output, so as to obtain a corrected main reasoning model; and forming a power grid analysis and reasoning platform based on the corrected main reasoning model and the dynamic equation discriminator.
3. The method according to claim 1 or 2, wherein the construction of the electric power multisource intelligent causal map comprises: Acquiring a physical connection relation of equipment from a power grid GIS system as a basic topology; Integrating fault mode and effect analysis data, transient analysis theory text and protection coordination logic in a scheduling procedure, and extracting a dynamic causal rule; and representing the map by adopting an attribute map model based on the basic topology and the dynamic causal rule, and recording equipment types, identifiers and state variables by the nodes, wherein the side relationship marks causal types and weight parameters.
4. A method according to claim 3, wherein said extracting dynamic causal rules comprises: generating nodes through automatic analysis equipment ledgers and monitoring data, and checking and confirming through manual experience; automatically finding edge relations through text causal extraction, time sequence association analysis and topology matching algorithm, and associating propagation probability or typical delay parameters; and forming a dynamic causal rule by the nodes and the side relation.
5. The method of claim 2, wherein the generating a candidate training fault causal chain comprises: inquiring a pre-constructed multi-source intelligent causal map through a pre-trained graph-annotation meaning network in a main reasoning model based on the initial training feature, and calculating the attention weights of the initial feature and node features in the multi-source intelligent causal map, wherein the graph-annotation meaning network is trained through a cross-modal contrast learning strategy; performing fusion correction on the initial feature based on the attention weight to obtain an enhanced text feature vector; And generating a candidate training fault causal chain through a main reasoning model based on the enhanced text feature vector and the multi-source intelligent causal map.
6. The method according to claim 1 or 2, wherein the construction of the dynamic equation arbiter comprises: discretizing a differential algebraic equation of the power system into a differential equation by adopting an implicit trapezoidal integration method; a differentiable computational graph is constructed using an automatic differentiation tool based on the differential equation.
7. The method according to claim 1, wherein the performing physical compliance verification on the candidate fault inference chain by a dynamic equation discriminator constructed based on a dynamic equation of the power system, and outputting a power grid fault analysis inference result, comprises: Receiving event sequences, time stamps and electrical quantity parameters of the candidate fault causal chain through a differentiable computational graph in a dynamic equation discriminator constructed based on a dynamic equation of the power system; Calculating a physical compliance score for the candidate fault causal link based on the sequence of events, a timestamp, and an electrical quantity parameter; and executing result filtering operation based on the physical compliance score, and outputting a power grid fault analysis reasoning result with the highest physical compliance score.
8. The utility model provides a multisource intelligence map reinforcing-based electric wire netting analysis reasoning system which is characterized in that is applied to electric wire netting analysis reasoning platform, includes: The generation module is used for inputting the input power grid fault text into the main reasoning model to generate a candidate fault reasoning chain; The output module is used for carrying out physical compliance verification on the candidate fault reasoning chain through a dynamic equation discriminator constructed based on a dynamic equation of the power system and outputting a power grid fault analysis reasoning result; The main reasoning model is integrated with a graph attention network and is used for performing cross-modal feature enhancement on text features of a power grid fault text according to the electric power multi-source intelligent causal graph in the main reasoning model, and the side relationship of the electric power multi-source intelligent causal graph at least comprises the following causal relationship of physical connection, triggering conditions and time sequence dependence.
9. The electronic equipment is characterized by comprising at least one processor and a memory, wherein the memory and the processor are connected through a bus; The memory is used for storing one or more programs; The method of multisource smartgraph-based enhanced grid analysis reasoning as claimed in any one of claims 1 to 7 is implemented when the one or more programs are executed by the at least one processor.
10. A readable storage medium, having stored thereon an execution program, which when executed, implements the multisource intelligent profile enhanced based grid analysis reasoning method as claimed in any one of claims 1 to 7.

Description

Power grid analysis and reasoning method and system based on multisource intelligent map enhancement Technical Field The invention relates to the field of computers, in particular to a multisource intelligent map-enhanced power grid analysis and reasoning method and system. Background Traditional power grid analysis mainly relies on numerical simulation (e.g., electromagnetic transient simulation, electromechanical transient simulation) based on differential algebraic equations. Such methods are based on physical law modeling and have high computational reliability given accurate models and parameters, but their computational process is complex and time consuming and essentially incapable of directly processing and understanding unstructured natural language fault report data. In recent years, deep learning technology, in particular to a natural language processing method based on a pre-training language model (such as BERT), is introduced into the field of power grid fault text analysis and is used for tasks such as fault classification, key information extraction and the like. However, such models are essentially data-driven probabilistic models whose optimization aims at maximizing the likelihood probability of a text sequence, rather than capturing the inherent causal mechanisms of the physical system. This results in the model potentially generating results that violate the causal timing and physical constraints of the actual power system (e.g., predicting that a "circuit breaker trip" occurs before a "line overload"). In the existing improvement scheme, the closest to the invention is a Graph Neural Network (GNN) method combined with a power grid topology knowledge graph. According to the method, static topological structures such as equipment connection relations of the power grid are injected into the model in a knowledge graph mode, so that domain knowledge is introduced to a certain extent. The main limitation is that the map content is mainly concentrated in static information such as physical connection of equipment, and causal rules (such as triggering conditions of state transition and time sequence dependence among events) reflecting the dynamic propagation process of the events cannot be effectively embedded. Physical constraint is missing, and explicit modeling and constraint on a power system core dynamic equation (such as a differential algebraic equation describing transient process and a time sequence logic equation of protection action) are absent, so that compliance of model output at a physical level cannot be ensured. Therefore, the method solves the problems that the existing power grid analysis platform is broken, marked with scarcity and lacks of dynamic physical constraint, and cause and effect errors and physical violations easily occur in fault reasoning, and has very important significance. Disclosure of Invention The invention provides a power grid analysis reasoning method and system based on multi-source intelligent map enhancement, aiming at solving the problems that the existing power grid analysis platform is prone to cause causal errors and physical violations due to multi-source data fracture, scarcity labeling and dynamic physical constraint missing. In a first aspect, a power grid analysis and reasoning method based on multi-source intelligent map enhancement is provided, and is applied to a power grid analysis and reasoning platform, and the power grid analysis and reasoning method comprises the following steps: Inputting the input power grid fault text into a main reasoning model to generate a candidate fault reasoning chain; Performing physical compliance verification on the candidate fault reasoning chain through a dynamic equation discriminator constructed based on a dynamic equation of the power system, and outputting a power grid fault analysis reasoning result; The main reasoning model is integrated with a graph attention network and is used for performing cross-modal feature enhancement on text features of a power grid fault text according to the electric power multi-source intelligent causal graph in the main reasoning model, and the side relationship of the electric power multi-source intelligent causal graph at least comprises the following causal relationship of physical connection, triggering conditions and time sequence dependence. In a second aspect, a system for power grid analysis and reasoning based on multisource intelligent map enhancement is provided, and the system is applied to a power grid analysis and reasoning platform and comprises: The generation module is used for inputting the input power grid fault text into the main reasoning model to generate a candidate fault reasoning chain; The output module is used for carrying out physical compliance verification on the candidate fault reasoning chain through a dynamic equation discriminator constructed based on a dynamic equation of the power system and outputting a power grid fault analysis r