High-Level Maintenance Plan Prediction for High-Speed EMUs Based on an Optimized Deep Belief Network

A high-level maintenance plan for high-speed electric multiple units (EMUs) is an important part of maintenance decisions, which have great significance for ensuring safe operation and improving resource utilization of high-speed EMUs. In this article, we propose an optimized deep belief network to predict a high-level maintenance plan quickly and accurately for high-speed EMUs. The time model of a high-level maintenance plan is first established based on the operating mileage and time of the EMUs. Secondly, the deep belief network (DBN) algorithm is applied to mine the corresponding relationship between time and mileage in a real historical operating dataset. The parameter combination of the number of hidden layers, the number of hidden nodes, the number of iterations, and the learning rate for the DBN model is globally optimized by an artificial bee colony algorithm (ABC), and a hybrid model is used to predict the mileage of the EMUs. Finally, a sampling period of the real historical data is determined to improve the predictive accuracy by analyzing and comparing the mileage predictive results. The operating data of high-speed EMUs are taken as an example, and the results of simulation show that, compared with other methods, the proposed method has the highest accuracy and provides an important basis for the formulation of high-level maintenance plans for high-speed EMUs.