Improved Intelligent Algorithm-based Electric Locomotive Traction Transformer Fault Diagnosis Technology

Railway construction is undergoing a high-speed development period in China. At present, our country has build the world's biggest and fastest high-speed rail network, railway operational mileage also leaps into the second place of the world. With the rapid development of railway construction, the work of guaranteeing security of railway transportation appears to be more and more important. The train traction transformer, as a device that takes electricity from traction power grid directly, is no doubt the whole train power sources, or that is to say the heart of the train. Its operation statues could directly affect the safety and working efficiency of electrified railway system, once it breaks down, the railway transportation would be influenced seriously. Therefore, this paper proposed an improved intelligent algorithm in order to increase the efficiency of traction transformer fault diagnosis.Firstly, this paper states the significance of developing electric locomotive traction transformer fault diagnosis technology as well as the development situation at home and abroad. According to the structure and special work circumstance of electric locomotive traction transformer, this paper analyzes the possible fault types and the way to extract characteristic value under these faults in detail.After balancing the advantages and disadvantages of BP neural network, RBF wavelet neural network, wavelet neural network, genetic algorithm and particle swarm optimization algorithm, this paper choose the combination of multi-resolution analysis wavelet neural network and quantum improved particle swarm optimization algorithm to further enhance the accuracy of the traction transformer faults diagnosis.In the experimental tests, the traction transformer 600 sample datum are extracted from the electric locomotive main transformer comprehensive test and fault diagnosis system which was develop by institute in 2006, as well as the Rogowski Coil that is installed at place of the transformer outlet terminal, bushing tap grounding wire and core grounding wire and so on. Then these datum are used to train and test four neural networks including BP neural network, two kinds of wavelet neural network and the proposed multi-resolution analysis wavelet neural network. In the end, the experimental results proved that this paper proposed diagnosis algorithm is not only steadier and converges faster, but also achieves higher accuracy (reached 95%).