The Research On Intelligent Fault Diagnosis Method Of Power Electronic Circuits

With the rapid development of power electronics technology and the appearance of high-performance power electronic devices, currently power electronic circuits have been widely used in electrical engineering field of industrial production and military defense, such as HVDC, AC-DC power transmission, high-performance power supply are all related to the power electronic circuits. However, the power electronic circuits in these devices is a high incidence of failure, once the power electronic circuit breaks down, would threat the entire device and even personal safety. So, it is significant to realise diagnosing the fault of power electronic device rapidly and accurately, and it is a prerequisite to study its theory and methods for achieving fault diagnosis.Currently existing fault identification methods in power electronic are: fault dictionary method, fault tree, spectral analysis, the direct detecting voltage across the power device and so on. However, all of these methods have some shortcomings, for example fault tree is intuitive, universal, but its applications are very limited for much workload. In recent years, artificial neural network, support vector machine and other methods have been widely used in various devices in fault identification, mainly because they don't need to build the model, and has strong ability to self-learning and parallel processing,but they are also easy to fall into local minimum points when training process.In this paper, it aims to improve the fault recognition rate by optimizing existing diagnostic techniques and then proposing convenient, accurate fault diagnosis algorithm for existing defects in power electronic fault diagnosis technology. Based on the basis of traditional neural network and support vector machine pattern recognition, proposed a method of power electronic circuit fault diagnosis in combination with other swarm intelligence optimization methods and it is proofed fessible and scientific by simulation examples.First selecting a typical power electronic circuits-twelve pulses controlled rectifier, would modeling it, simulate for various fault, and extract fault waveform at last. Then the extracting signal would be compressed and reduced the dimension to obtain suitable troubleshooting input signal by using PCA signal processing methods.A novel method is presented for diagnosis based on multi-population genetic algorithm combination immigration operator and migration operator optimized BP neural network. It changed single population in traditional genetic algorithm into multi-population for optimizing collaboratively, and the various populations can exchange information by the immigration and migration operator, at last the algorithm has a great extent to avoid the "premature" phenomenon. Then fault types are identified through multi-population genetic algorithm optimized neural network for twelve pulses controlled rectifier. Simulation results show that this method can identify and locate each fault type accurately, and has better robustness and high diagnostic accuracy rate characteristics.A novel method is presented for diagnosis based on double-vision adaptive AFSA optimized SVM. The current AF obtains double-vision by measuring the distance between optimal AF and the recent AF. The AF with double-vision not only can timely escape from local optimum value, but also can successfully find the global optimum value in the vicinity of the global optimum. Finally, fault types are identified through AFSA optimized SVM for twelve pulses controlled rectifier. Simulation results show that the algorithm has fast convergence and high precision optimization features.