Comprehensive Status Assessment And Remaining Useful Life Prediction Of Overhead Contact System Based On Deep Learning

The overhead contact system(OCS)is the main framework of railway electrification project,and shoulders the important mission to electric locomotive power supply,the OCS along the railway lines over the set up at the same time,in the open-air outdoor environment,need repair in maintenance window and repair time is long,if the OCS malfunction,will be a serious threat to the safe operation of the railway,have incalculable consequences.Therefore,the health status of the OCS needs to be accurately evaluated and monitored.On this basis,it is of great significance to guarantee the healthy operation of OCS if it can predict in advance the expected continuous normal working time from the current to the occurrence of potential faults,namely the remaining useful life of OCS,and realize the condition-based maintenance of catenary before the occurrence of faults.In this paper,starting from the comprehensive status assessment of OCS,the OCS comprehensive status assessment model is built by comprehensively selecting the appropriate OCS parameters like the basic parameters of the assessment model based on the comprehensive detection data,impact the degree of index parameters,defect information data and manual experience,etc.Then according to the relevant regulations of electrified railway and line engineering implementation plan,the parameters are divided into different health grades and given normalized values.The initial weighting model uses the analytic hierarchy process and entropy method to carry out the combined weighting.Then,aiming at the problem of insufficient balance of the combined weighting method,particle swarm optimization is introduced to carry out the secondary weighting of the combined weighting method.Based on the subjective and objective advantages of the weight distribution results in the model,the balance between indicators is optimized at the same time.Then,the prediction effect of the remaining useful life prediction model based on traditional machine learning is explored,and the remaining useful life prediction model based on support vector machine,decision tree and artificial neural network is established.After the model evaluation of the traditional machine learning prediction model,Furthermore,a remaining useful life prediction model based on a deep neural network is established.Aiming at the problem that the overall prediction accuracy of remaining useful life prediction model based on the deep neural network is poor,taking OCS comprehensive status assessment model and remaining useful life prediction model based on deep neural network as the basic model,a new method of CCS-DNN remaining useful life prediction based on OCS comprehensive status and deep neural network is designed.The CCS-DNN remaining useful life prediction model is tested experimentally,and compared with the traditional machine learning prediction model and deep neural network prediction model,it is verified that the CCS-DNN model has higher prediction accuracy.Finally,the hyperparameter optimization of the CCS-DNN model was carried out to optimize the structure of the model network and the parameters of the neural network.The hidden layers of the neural network,the number of neurons in each layer,and the activation function were determined through experiments.The learning rate of the model,the number of local random seeds and the number of iterations of the data set were optimized by the genetic algorithm.The prediction results of the optimized prediction model are analyzed significantly.The anchor point of the inspection data of railway line,for example,the comprehensive status assessment and remaining useful life of remaining useful life interval,and the anchor point guide the maintenance time of the OCS next time.