Special Power System Fault Analysis And Feature Extraction Method

China's population is large, inland breadth and depth to address the large-scale population movements, the safest, most efficient, most economical, most environmentally friendly and most reliable modes of transportation, is the high-speed rail. China's railway decided to develop a high-speed as the main direction of modernization.The rapid development of high-speed railway, as well as the stringent requirements for security, you need roadbed and power of high quality and high stability. As to run one of the most important factors in a high-speed railway,the power supply system must be an absolute security and stability.Therefore,timely to diagnose when the faults occur and why they occur is very important. The issue is to ensure that high-speed railway traction power supply system have a good steady supply of electricity,for the traction power supply to run real-time signal detection and fault analysis per se.Fault detection on the traction power supply system mainly includes three steps:signal acquisition, signal processing and analysis, fault diagnosis. This article describes how to use the dual-channel pairs of memory to run virtual machines on the traction power supply signal data collected and saved to the hard drive, the signal pre-processing, that is, the fault occurrence time of the data interception down in order to observe and analyze; then in this paper, the concept of skewness and kurtosis in Mathematical Statistics are used in of the traction power supply to run the signal feature extraction method. Different from the power supply system of subway signals, high-speed rail system in the short-circuit power supply voltage amplitude is not always significant mutations occur at this time setting out from two aspects of the phase and amplitude analysis, theoretical skewness and kurtosis are very appropriate to this time needs. And calculating the time complexity the feature extraction algorithm based on skewness and kurtosis will determine the necessary hardware configuration.. In this study the process of feature extraction will be made of the user interface mode for multiple testing, so as not to modify the program and lead to unnecessary duplication of errors. Finally the characteristic values of wavelet neural network to detect and reach of the operating system, traction power supply fault detection purposes.