小波包神经网络与数据降维的移相全桥变换器的故障诊断

Phase-shift Full Bridge Converter Fault Diagnosis Based on Wavelet Packet and Neural Network and Data Dimensionality Reduction

移相全桥变换器作为机车控制电源的核心电路其故障特征类型极其丰富,故障信息量大,为了彻底全面地挖掘故障信息,提出了小波包神经网络和数据降维的新型故障诊断模式,主要利用流形学习来对高维的故障特征量进行降维,提取其本质特征解决了由小波包多层分解带来的"维数灾难",减轻了模式识别的压力。利用Matlab仿真软件分析,此方法可以使模式识别的时间缩短,准确率提高,从而验证了该方法的有效性。

As a core circuit of locomotive control power supply, phase-shift full-bridge converter fault type is extremely rich, and it has large amount of fault information. In order to dig fault information thoroughly and comprehensively, there is a new model that wavelet packet and neural network and data dimensionality reduction for fauh diagnosis. The main idea is taking use of manifold learning to reduce the dimension of high dimensional fault characteristic quantity and to extract its essential characteristics, so as to solve＂curse of dimensionality＂ from the multi- wavelet packet decomposition, and reduce the pressure on pattern recognition. Under the analysis of Matlab simulation software, this method can shorten the time of pattern recognition and improve the accuracy rate. Finally demonstrate that this method is effective.