Research On Fault Diagnosis Of Analogy Circuit By Optimized Extreme Learning Machine

With the rapid development of today's intelligent electronic products and the penetration of various aspects of society,the detection and diagnosis of circuit systems has become a strategic issue of great research value.The integration of analog circuits in circuit systems is becoming more and more high,and once a fault occurs,the circuit system will be paralyzed.The nonlinearity and tolerance of analog circuits make traditional diagnostic methods and theoretical studies difficult to meet the current requirements of circuit fault diagnosis.Therefore,it is particularly important and urgent to research the applicable and effective analog circuit fault diagnosis methods.With the rapid development of group algorithms and artificial intelligence,it provides a new and effective way for diagnosis theory and method.In this paper,the general circuit is taken as the object to research on the aspects of fault feature extraction,selection and fault classifier.The main research work and results are as follows:(1)Research on fault feature set extraction and selection.In the selection of fault feature parameter set,this paper adopts Adaptive Genetic Algorithm(AGA),and AGA as a heuristic search algorithm has a good effect in the optimization problem.In order to ensure the correlation between the feature parameters,the fitness function in the algorithm follows the scatter matrix rule,and the original feature parameters extracted for the sampling points of the feature parameters are binary coded.It has been verified by experiments that it has a good effect in processing low-dimensional data.In order not to be limited to the processing ability of low-dimensional fault characteristic parameters,this paper introduces deep learning theory and method.In the aspect of high-dimensional feature parameter extraction,the sparse auto-encoder of the deep-learning network is cascaded into three layers to form a Stack Auto-Encoder(SAE).In order to further improve the stability and accuracy of the high-dimensional data extraction capability of the SAE,the iterative optimization method of the weight and offset of the SAE is studied by Particle Swarm Optimization(PSO).Then the feasibility of this method is demonstrated by experiments.(2)Research on fault classifier.In this paper,Extreme Learning Machine(ELM)is used as the basic network of fault classifier network.Extreme learning machine network has the characteristics of less parameter configuration,weight and offset random assignment,network learning speed and fast convergence.However,the random assignment of weights and offsets has a certain impact on the stability of the network.Therefore,the Adaptive Wolf Pack Algorithm(AWPA)and the network model of extreme learning machine optimized by adaptive wolf pack algorithm(AWPA-ELM)are proposed.The adaptive wolf pack algorithm is aimed at the Wolf Pack Algorithm(WPA)lack of learning ability in the behaviors of wandering,summoning and siege.It can't adjust itself according to the optimization space and the optimization of resources.The adaptive wolf pack algorithm is built by introducing adaptive thinking strategy,then the feasibility of the method is demonstrated by experiments.