Transformer Switching-in Inrush Current Suppression Based On Deep Forest Regression

Power transformer is a very important electrical equipment in power system.The safe operation of transformer plays an important role in maintaining the stability of the power system.When the circuit breaker is no-load switching-in,the core of the transformer will produce a large peak inrush current under the combined action of steady-state remanence,steady-state periodic component and transient bias.Inrush current may cause misoperation of relay protection device,accelerate aging of transformer winding and further induce a fault.This paper studies the relationship between the characteristics of inrush current and the opening and switching-in angle of the circuit breaker,and puts forward the inrush current suppression scheme based on the precise switching-in control: Based on the physical dynamic simulation test data,the opening angle is predicted by the deep forest regression model,and the ideal switching-in time of the circuit breaker is obtained by looking up the table after the opening angle prediction.Based on the integrated learning algorithm,the switching-in time delay of the circuit breaker is estimated.This paper analyzes the switching-in control principle of the circuit breaker,compensates the switching-in delay of the circuit breaker,calculates the switching-in command time of the circuit breaker,carries out the switching-in control,and realizes the suppression of the switching-in inrush current of the circuit breaker.The scheme overcomes the difficulty of accurate estimation of opening angle and switching-in delay of the circuit breaker.In this paper,the complex transient process of arc in the process of an opening circuit breaker is studied,and a scheme of opening angle prediction based on deep forest regression algorithm is proposed.Through the depth forest model,the residual error between the ideal opening angle and the current predicted value is fitted layer by layer to realize the accurate prediction of the actual opening angle of the circuit breaker.Through the multi granularity scanning,the accurate prediction of the opening angle of the circuit breaker with different opening voltage waveform in the transient process is realized.The structure and transient process of the circuit breaker are studied,and an integrated learning algorithm is proposed to predict the switching-in delay of the circuit breaker.Through the integrated learning algorithm,the advantages of the gating cycle unit(GRU)and the deep forest regression based on LSTM autoencoder(LDRF)model are integrated.The threshold function is used to dynamically change the structure of the integrated model according to the actual input prediction waveform,so as to improve the accuracy of the actual prediction model.The time delay prediction model based on the integrated model can realize the accurate prediction of the inherent switching-in time delay of the circuit breaker under the action of noise signal.