Transient Identification Of Traction Power Supply System Based On FCN-LSTM Parallel Model

Traction power supply system,as a special power supply system to provide electric energy to electric locomotive.If the traction power supply system fails,it will not only bring serious damage to the equipment,affect the stability of the system,but also bring great loss to the national production.A wealth of data is held within the transient signal of a traction power supply system,providing the basis for an accurate assessment of whether it is malfunctioning or not.The staff can swiftly assess the transient process of traction power supply systems,thereby determining the fault type,by recognizing the transient signal of such systems.The overhaul of the traction power supply system can be expedited,and the level of operation and maintenance can be enhanced.The transient signals of traction power supply system have the characteristics of complexity and diversity,and the traditional identification methods have certain limitations.Due to the rapid development of deep learning technology in recent years,its powerful data mining and processing capabilities provide ideas for this topic.Consequently,this topic will employ a deep learning approach to recognize transient signals in traction power supply systems.This paper firstly summarizes the common types of transient processes in traction power supply system,and provides sample sources for deep learning model training through field testing and construction of traction power supply system transient simulation model in traction substation.Secondly,as field measured data belong to unlabeled samples,they cannot be directly used to build the data set required by the deep learning model.Therefore,based on Cohen class time-frequency distribution,four indexes(instantaneous distortion energy ratio,normalized instantaneous distortion energy ratio,instantaneous frequency and instantaneous K factor)are adopted in this paper to evaluate transient signals.Feature extraction is carried out on the measured data of traction power supply system,and the range of four kinds of transient indexes corresponding to common transient types in traction power supply system is obtained,so as to realize the classification and label of the measured data.Finally,for transient signal identification of traction power supply system,a deep learning model based on full convolutional neural network FCN and short and long time memory network LSTM is proposed to identify transient signals of traction power supply system.Verification of the efficacy of this technique is achieved through comparison experiments.