Research Of Power Converter Fault Feature Extraction Technology Based On Learning Network

Over half a century,thanks to the continuous technological breakthroughs and progress in power electronics,high-performance power converters are widely used in various electrical equipments and power systems industry,the important responsibility of energy conversion and power transmission responsible,Become an indispensable part of industrial production and social life.However,due to the complicated working environment of the power converter,it is often in a high-temperature,high-pressure and high-current working environment,and its internal power device is limited in its overload capacity.The power converter is likely to fail due to open circuit failure of the power switch,and this problem can cause the overall electrical equipment and power system to fail to operate properly.It even can cause personal safety accidents.Therefore,carrying out research on power converter fault feature extraction and diagnosis methods is of great research significance and practical application value to ensure the healthy operation of the overall electrical equipment and power system.This paper mainly studies the application of two different learning networks,Deep Belief Networks and Broad Learning System network,in power converter fault feature extraction and diagnosis.The performance and effect of these two methods are analyzed by comparing simulation and actual experimental results.Transplanting the Deep Belief Networks into the Digital Signal Processing realizes online fault diagnosis.The specific research contents are as follows:(1)First,the basic model and principle analysis of the two methods of the deep placement letter network and the extended learning network are introduced.For the Deep Belief Networks,searching random trees is proposed to Propose it.The genetic algorithm method is used to improve the performance of the Broad Learning System.And based on these two algorithms,the power converter fault feature extraction method is designed.(2)Secondly,two power converter simulation circuits with different topologies were constructed by Candence / Pspice simulation software.The primary current and phase current of the transformer were selected as the fault detection signals.The fault mode simulation settings and fault sample collection were carried out.The effectiveness and universality of the two feature extraction methods.On the basis of the first topology power converter simulation experiment,a hardware failure experiment platform consistent with the simulation circuit parameters was independently designed.The two feature extraction methods proposed in the actual physical experiment were verified,and the result of simulation experiments were compared.(3)Finally,Deep Belief Networks algorithm is embedded in the DSP chip,and a set of power converter online fault feature extraction and diagnosis system is designed.The results show that the proposed design method can effectively extract the fault characteristics and classify the fault mode of the power converter in practical applications,so as to realize the efficient diagnosis of the open circuit fault of the power converter.