Application Research On Fault Diagnosis Of Power Electronic Circuits Based On BP Neutral Network Optimized By QGA

At present, the power electronics application in industrial production is very extensive, such as the fields of electricity, transportation, machinery, chemicals, mining and metallurgy, textile, aerospace, lasers, communications, robots, efficient use of energy. Power electronic d-evices often act the roles of power supplying or controller, once the diagnosing is not in time, it will damage equipments, destroy continuous production, and even lead to system failures, casualties, of which will result in great economic losses. Therefore, the application of power electronic circuits fault diagnosis technology has enormous economic significance and necessi-ty.Nowadays, integration and complexity is the major development trend of power electroni-cs, it makes the artificial diagnosis has been difficult to meet the requirements. States, both inside and outside frequently used methods of fault diagnosis of power electronic, such as fault dictionary method can only solve the single-fault, fault tree has the defects of heavy workload and error-prone, directed detection is limited in hardware and difficult to achieve, residual method need to establish accurate mathematical model, expert system method is limited by knowledge-bottleneck, these methods have limited application for their limitations. Artificial neural network fault diagnosis methods well adapted to this trend because of its nonlinear mapping and self-learning ability, and it has become a hot research field of fault diagnosis.BP network is widely used in the neural network fault diagnosis of power electronic circuits, but it has the defects of slow convergence and easily to fall into local minimum, these have seriously affected the performance of its fault diagnosis. Quantum genetic algorithm has the features of genetic algorithm and quantum computing that rapid convergence and well abi-lity of global optimization. However, in the later stage of evolution, it has the defects of slow evolution and low efficiency use of chromosome. To proposed two improvments for the defects, then use the improved quantum genetic algorithm to optimize the BP neural network, of which gives a good initial connection and a smaller searching space that positioning the global optimal solution for the BP neural network, and this accelerates the learning speed of network, it also improved the diagnostic performance.This thesis takes the improvement and development of BP neural network method for fau-It diagnosis of power electronic circuits as the purpose, below is the major works in the thesis:To take two improvements that improving the population evolutionary mechanism and adding the searching process of comparison of population to solve the defects of quantum genetic algorithm of double chains and coding based on Bloch spherical coordinates that slower speed, weak local searching capability and low effective use of chromosome in the late evolutionary. Through the examples of multi-peak function optimization, it verifies the effectiveness of the improved algorithm. At last, takes the more excellent quantum geneti-c algorithm coding by Bloch spherical coordinates to improve BP neural network and com-pletes the programming algorithms.To take the rectifier circuit as an example, uses spectrum analysis technology and th-e MATLAB software to simulate and study power electronic circuit's diagnosis, then deter-mines the fault mode, extracts fault information, designs fault samples.To use the hybrid method to diagnose the fault of power electronic circuit, compares the simulation results with the other algorithms. The results show that the improved mixed-algorithm has the advantages of faster network learning speed, higher diagnostic rate and better generalization ability. It verifies that the fault diagnosis method is more effective and advantageous.