Research On Fault Traveling Wave Feature Extraction And Recognition Algorithm Of Railway Traction Network

With the rapid development of China’s comprehensive national strength and improving the railway transportation, the railway system in our country has entered the electrification, high speed, intelligent age, while the contact network is an important part of railway electric traction system, it undertakes the task of power supply. Catenary fault will be on the stable operation of the railway system, the people’s daily travel and the construction of national economy brought serious harm and influence. Therefore, research of railway traction power supply line rapid positioning and fault type identification has important significance. In this paper, the fault traveling wave signal of railway traction network is studied, and the method of extracting the characteristic of the traveling wave signal and the classification and recognition algorithm are studied. In this paper, the problem of fault data is difficult to distinguish between the fault and the traveling wave, and the waveform feature extraction algorithm is studied. The following work is mainly carried out:(1) the wavelet packet energy spectrum method is used to extract the feature vector of the traveling wave signal. Using wavelet packet decomposition signal of each frequency band are independent of each other and the characteristics of energy conservation, the signal of three layers of wavelet packet decomposition and reconstruction, and the frequency band energy spectrum is calculated, finally normalized form of wavelet packet energy spectrum feature vectors.(2) using the fractal theory, the fractal dimension algorithm is extended to the feature extraction of the traveling wave signal waveform. By using the fractal theory of self similarity and fractal dimension, the fractal dimension is extended to the one-dimensional traveling wave signal, and the fractal dimension of the traveling wave signal is extracted as the feature vector.(3) a new method of feature extraction is proposed, which combines wavelet packet energy spectrum and fractal dimension feature vector to form a new feature vector. Wavelet packet energy spectrum can reflect the signal intrinsic features and local characteristics, but overall is poor, and the fractal dimension can well reflect the overall properties of the object, but the local feature and does not have the advantage, combination of the two, both reflect the overall signal characteristics, and can well reflect the signal of local features in order to get a more comprehensive feature vector.The neural network method to classify and identify the type of fault, to extract the feature vector as the sample for neural network training, realize fault type recognition, from the point of view of the recognition effect, classification and recognition of the overall accuracy rate reaches above 97%, and the wavelet packet energy spectrum feature weight alone as input compared, wavelet packets fractal feature combination of traveling wave signal recognition and classification accuracy and reliability were improved.