Research And Application Of Reactor Fault Warning Based On Neural Network

As a key electrical equipment to prevent overvoltage and suppress harmonic currents in the power system,the health status of reactors is the foundation of power system stability.Therefore,studying reactor fault warning technology can detect reactor faults in advance,which is of crucial practical significance for ensuring the stable operation of the power system.At present,the research on reactor fault warning methods is mainly based on traditional machine learning methods such as random forest,linear regression,Bayes,but these methods have obvious shortcomings: in the actual reactor fault data of the power grid,there are few samples of a single model,and the reactor fault warning will appear overfitting phenomenon.Moreover,the current research methods are customized and lack of general modeling methods,Constraints in sample size and modeling applicability have resulted in low accuracy and reliability of reactor fault warning results.In view of the above problems,this paper proposes a long short-term memory neural network algorithm to achieve reactor fault early warning.The state of the reactor changes with time,and the state order is dependent.The long short-term memory neural network algorithm can process the sequence data well,capture the dependencies between sequences,and learn the dynamic change behavior of the reactor in a healthy state.However,the long-term and short-term memory neural network algorithm is not ideal for small sample heterogeneous reactor fault early warning.Therefore,the reactor fault early warning method based on the combination of short-term memory neural network and twin neural network algorithm is studied.The twin network structure can map the Information mapping of low dimensional reactors to high-dimensional space for feature learning,which is used for fault early warning of heterogeneous reactors.The method proposed in this article can effectively predict reactor models that do not appear in the training set,and can also effectively predict data sets with uneven distribution.This article constructs a reactor fault warning model,and then proposes a reactor inspection system.The reactor inspection system is established through object-oriented methods,and information technology is used to reduce the workload of reactor inspection and improve the efficiency of reactor inspection,which has important practical significance.