Research On Fault Diagnosis Method Of Three-phase Rectifier Based On Neural Network

Power electronic equipment is widely used in today's production and life.Power electronic rectifier device is the main part of the power electronic equipment.When the internal circuit appears abnormal status,it can damage the internal circuit system,severely it may threaten the safety of life.So in order to improve the reliability of the power electronic equipment operation,it is very important to study and analysis the circuit fault diagnosis of the rectifying device.This paper mainly studies the fault diagnosis of three-phase bridge rectifier circuit which is widely used.Firstly,the possible faults of the thyristor in the three-phase bridge rectifier circuit are analyzed,and then the fault information of the circuit is extracted.One of the main reasons that affects the normal operation of the circuit is the fault of thyristor.In this paper,the thyristor faults commonly encountered in the operation of three-phase bridge rectifier circuit are analyzed.According to the influence of thyristor fault on the circuit output voltage waveform,the possible fault state of the circuit was simulated in SIMULINK module,and the fault state voltage waveform was obtained at the circuit load end.Fourier analysis is used to extract fault characteristic information from waveforms: DC component,fundamental amplitude,second harmonic amplitude and third harmonic amplitude.Then the fault information is normalized.Secondly,BP neural network is used to diagnose the circuit fault.The fault information is input into BP network,and the network output information is compared with the expected value,so as to obtain the fault location and fault diagnosis rate.It is proved that BP network can diagnose circuit fault.Because of BP network is easy to converge to the local minimum and the fault diagnosis rate is low,so it is necessary to optimize the BP network.Then,because of the Particle Swarm Optimization algorithm has global search performance,it can effectively combine with the local search performance of BP network,and then optimize the weight and threshold of BP network,so the Particle Swarm Optimization algorithm is adopted to optimize BP network.Then the network is trained and verified,and get the diagnosis results,which is better than the results of BP network.However,Particle Swarm Optimization algorithm is easy to fall into the problem of local optimal solution,the fault diagnosis rate is not up to the expected value.Finally,the Stacked Auto-Encoder network in the deep learning network is selected to diagnosis the circuit fault.The network optimizes the structure of the above two networks,and it effectively makes up for the deficiencies in the above networks,and at the same time it carries out in-depth feature extraction of fault data,which greatly shortens the network training time and improves the circuit fault diagnosis rate.The simulation results show that compared with the three neural networks,the third improved network has the advantages of less training time,high fault diagnosis rate,good reliability,and wide applicability,and achieves the expected results.