Research On Transformer Magnetizing Inrush Current Identification Based On Empirical Mode Decomposition And Optimal Probabilistic Neural Network

In recent years,due to the sustained economic development,China’s demand for energy is increasing,and the requirements for relay protection are increasingly strict.As one of the most important and expensive electrical equipment in the power system,power transformer is responsible for the transmission of electric energy and the role of voltage transformation.Its safe operation is directly related to whether the whole power system can work continuously,safely and stably.The excitation inrush current will lead to the misoperation of differential protection.Yes,the correct operation rate of transformer protection is lower than the average correct operation rate of AC power network,so the identification of excitation inrush current has become an important topic.On the basis of studying and analyzing the formation mechanism and characteristics of various excitation inrush current of transformer,the simulation model of excitation inrush current,internal fault and airborne internal fault of three-phase three-winding transformer in a practical system is built by using PSCAD/EMTDC software,and the corresponding simulation waveform is obtained.At the same time,the actual waveform of various cases is obtained according to the actual recording software,the similarities and differences of various cases are analyzed and compared,and the traditional characteristic quantity of each sample is obtained.Secondly,it is difficult to identify some special operating conditions by using traditional characteristic quantities,which will lead to misoperation and rejection of differential protection.Aiming at the above defects,this paper innovates the characteristic quantity and introduces the characteristic quantity based on empirical mode decomposition(EMD)and EMD threshold denoising.The characteristic quantity can well describe the sample waveform.At the same time,the mean influence value method(MIV)based on BP neural network is introduced to reduce the redundancy of feature quantity and prevent over-fitting of samples.The simulation results of four kinds of artificial intelligence classification algorithms show that EMD feature can be used to identify inrush current.Finally,in view of the defect that the conventional artificial intelligence algorithm is difficult to accurately identify the air-dropped internal fault transformer samples,the optimization probabilistic neural network algorithm is proposed to identify the excitation inrush current method.The probabilistic neural network has strong ability of sample addition,high fault tolerance and fast training speed.Based on the recognition results of the probabilistic neural network,samples are added pertinently,and the smoothing factor value is optimized by the particle swarm optimization al gorithm with global search ability.The simulation results show that the recognition effect of the algorithm has been greatly improved through optimization,and the method can be used in practice.