The Research Of Fault Diagnosis Mothods In Power Electric Circuits

With rapid development of power electronic technology, makes its application widespread increasingly, else the fault problems of its are more and more prominent. And the electric power electronic devices is usually playing the very important role in the engineering system, once ware breaked down, would cause the engine off and the personnel casualty, else bring huge economic loss if cannot obtain the diagnosis and restoration promptly. Therefore the theoretical and methodological research of fault diagnosis in the power electronic devices has its practical and economical significance.At present, neural network is generally used to diagnoze the fault diagnosis of power electronic circuits. This method has much superiority, But its flaws are displayed gradually along with the development of its application. In order to solve these flaws, many researchers and scholars begin to explore new neural network structure and theory, and the researches which based on neural network and other theories are already appearanced. In this paper, three methods of fault diagnosis in power electronic circuits are presented on combination of neural network and other theories.A fault diagnosis method of power electronic circuits based on the wavelet neural network is proposed. It combines the good properties of time-frequency localization of wavelet transform with the advantages of non-linear mapping and syntheszing of neural network, thus makes the BP neural network whose activation function of neurons in hidden level is wavelet function have more strongly approaching and fault-tolerant ability. Then it is utilized to diagnosis the Twinbridge 12 phase pulse wave rectifier circuit. The simulation results prove that it has a higher fault diagnosis rate compared with the BP neural network, else it is correct and feasible.A fault diagnosis method of power electronic circuits based on the quantum neural network is proposed. This mothod applies the thought of quantum state superimposition of quantum theory in neural network to form a quantum neural network which has multi-layer activation function, gives the quantum neural network a inherent fuzziness and well solves the pattern recognition problem of cross-data in fault modes during the power electronic circuits fault diagnosis. By taking the Twin-bridge 12 phase pulse wave rectifier circuit as an example, the simulation results show that this mothod have very high rate and low error rate of fault diagnosis. A fault diagnosis method of power electronic circuits based on genetic neural network is proposed: First, the network parameters of neural network are defined according to its topology , then the global optimization of the parameters is realized by geting the solutions of its using the genetic algorithm which is good at obtaining the complex globally optimal solution and has the advantages of greatly strengthened stability and overall optimization. Finally the parameters are passesd to the neural network to finish the fault diagnosis. The experiment dates of diagnosising the Twinbridge 12 phase pulse wave rectifier circuit shows that the genetic neural network has higher diagnosis rates and low error diagnosis rates(little bad then quantum neural network), and the time of the fault diagnosis is reduced greatly,so can satisfies the high timely request of project application.