Research On Power System Restoration Technology

With the development of modern society,the requirement of power supply reliability is higher and higher.After the power grid fails,if it cannot be restored in time,will have a serious impact on social production.At present,the process of power system restoration is mainly through dispatchers manually reviewing the dispatching documents,obtaining corresponding fault recovery information,and combining personal experience to perform power system restoration.It is essentially an empirical recovery process,and the level of intelligence needs to be further improved.This thesis deeply analyzes the advantages of knowledge graph technology in the field of power system restoration,and proposes a method of constructing a power system restoration knowledge graph and transforming unstructured restoration information into structured knowledge for storage.In the process of fault recovery,dispatchers use the knowledge base question answering technology presented in this paper to quickly obtain information related to fault handling,enhance the efficiency of dispatchers’ handling,and improve the intelligence level of power system restoration process.The main contents of this thesis are as follows:(1)Entity extraction is an important part of building a power system restoration knowledge graph.Existing entity extraction methods require a large amount of labeled corpus,which is difficult to obtain in the field of power system.This thesis proposes an entity extraction model of joint transfer and active learning.In the transfer learning module,ELETRIC-BERT is designed and trained,which reduces the model’s dependence on annotated corpus.In the active learning module,it is proposed to use the average logarithmic probability to measure the uncertainty of the sample,which overcomes the negative impact of sentence length and improves the effect of the active learning strategy in reducing dependence on annotated corpus.Finally,a module replacement strategy is given to compress the model,which significantly improves the training and inference speed of the model.(2)Combining with the proposed entity extraction model,after completing the ontology construction,information extraction and knowledge storage,this thesis constructs power system restoration knowledge graph and visualizes the knowledge graph,and presents dispatching knowledge to dispatchers in a friendly way.(3)When dealing with grid fault,dispatchers often retrieve answers to fault-related questions from dispatch files,and the searched questions include complex questions.In order to deal with such questions,this thesis designs a knowledge base question answering method based on path ordering,including three serial modules: topic entity acquisition,candidate path category acquisition and ordering,and candidate path selection.The combination of the three modules effectively improves the answer recall rate of complex questions.(4)This thesis designs and implements an intelligent auxiliary system for power system restoration.It was deployed and tested in a power company in East China Grid.The test results show that the system can effectively enhance the efficiency of dispatchers and improve the intelligent level of the power system restoration process.