| Jointless track circuit plays an important role in China Train Control System in use, the low frequency information contained in its signal is very significant for train control to ensure the transportation safety. Recently, the existing demodulation method of track circuit siganl has not completely satisfied rapid development of railway. So, based on the analysis of the existing demodulation method, data fusion theory based demodulation method of jointless track circuit is of much significance of reality. The main research contents of this dissertation are as follows:Firstly, according to the frequency spectrum characteristics, the feature vectors of track circuit signals with different low frequency are constructed and their extraction method is given.Secondly, based on the feature vectors constructed, the demodulation method of track circuit is converted to pattern recognition. Three pattern classifiers including K-nearest neighbor, adaptive probabilistic neural network and gray relational analysis are designed. The demodulation of track circuit is realized through computing manhatan distance, similar probability and gray relational grade between the frequency spectrum characteristic of signal to be analyzed and each corresponding feature vector. Experiment results shows that the three mehods have high classification accuracy.Lastly, D-S evidence theory based integration demodulation method is proposed to improve further the demodulation performance.Basic probability assignment of classification basises of the three classifiers is accomplished by fuzzy membership function, then the combination rule without conflict is adopted to combin if the three mehods all support the same focus element, or the combination rule with conflict is adopted, and then the final classification results are given. Experiment results shows that the integration demodulation method has strong anti-interference ability and higher classification accuracy.
|
PDF Full Text Request