Research On Intelligent Fault Diagnosis Method Of Power Electronic Circuits

With the development of power electronic device and the improving complexity of power electronic equipment, how to design a real-time, reliable and intelligent system of fault diagnosis has become more and more important.Firstly, this paper introduces the meaning, background and current condition of FDPEC, illustrates its key technologies, development prospect and fault modes. Then it introduces the principle, structure and algorithm of back propagation neural networks (BPNN) and ellipsoidal unit neural networks and compares their similarities and differences also. Secondly, this paper emphatically researches hierarchical fault diagnosis method based on fuzzy clustering and neural networks, the method based on improved ellipsoidal unit networks with initial multi-weights (IEUNIM) and the method based on composite and improved ellipsoidal unit networks (CIEUN). The numbers of neural networks are reduced and their convergence and generalization ability are all improved by applying Fuzzy C-means Clustering to training samples in hierarchical fault diagnosis method based on fuzzy clustering and neural networks. Using the diagnosis method of IEUNIM, the initial multi-weights are gained by classifying the space of fault character and the identification of network is improved greatly by using K-means Clustering. A large network is composed of many small IEUN in diagnosis method of CIEUN, and many ellipsoidal units are used to approximate a fault pattern. By this way, better diagnosis precision is gained.Finally, all those methods diagnosing the typical power electronic circuits are realized by matlab software and they are also compared with single BP network. Their simulation results show that the methods presented in this paper are effective. At the end of this paper, the research is summarized and its further direction is pointed out.