Electric power fault event extraction method combining deep learning and concept map

Abstract

The invention provides an electric power fault event extraction method combining deep learning and a concept map. In a feature selection stage, complex feature design is abandoned, only basic distributed semantic word vector features, dependency syntax structure features and position features are selected, and on this basis, concept expansion of a power failure text is achieved through a concept map based on a Chinese knowledge map. A long-term and short-term memory recurrent neural network is used for automatically carrying out feature learning, and a model training result is used for replacing original features and serves as the basis of trigger word recognition and event element recognition. In an element identification stage, an event element identification task is converted into a trigger word-entity and trigger word-trigger word relationship extraction task, training is performed in combination with a dynamic multi-pooling convolutional neural network, and event elements of simple events and complex events are identified at the same time. Rules are formulated according to the characteristics of a power field, and further an identification result is optimized. The method is simple and high in execution efficiency and accuracy.

Description

technical field [0001] The invention relates to electric power and computer applications, in particular to a method for extracting power failure events combining deep learning and concept maps. Background technique [0002] With the development of new energy, distributed power, and the increasingly abundant downstream applications of power, the uncertainty of power grid operation has increased significantly, and the traditional dispatching mode based on online security analysis functions based on mechanism and physical modeling has gradually been unable to meet the needs of power grids. It is required to fully excavate power grid operation texts such as dispatching procedures, fault plans, and dispatching logs, and use natural language processing technology, knowledge graph technology and corresponding big data analysis technology to mine empirical rules in the above data, which will have a clear understanding of the situation of large-scale hybrid power grids. Perception, i...

AI Technical Summary

Problems solved by technology

However, although the current natural language processing technology and knowledge graph technology have made some research progress in the identification of entities and relationships, they have not achieved great results for the most important domain events in the formation of empirical rules, especially grid domain events. breakthrough

[0003] The natural language processing in the field of electric power and power grid has not yet carried out systematic and in-depth research and development. The current research and development mainly focuses on the entity recognition of electric power text. For the recognition of power fault events, it is still blank and limited in terms of research and application. The application support of electric power text to intelligent decision-making and auxiliary dispatching of power grid

Images