A novel deep learning architecture and its application in dynamic load monitoring of the vehicle system

Abstract

In a complex vehicle system, the monitoring of dynamic load is difficult work. Therefore, a novel deep learning architecture based on a divide-and-conquer strategy is developed to construct the dynamic load monitoring model. The mathematical model of the vehicle system is established to provide data and theoretical support for the deep learning model. To solve the problems of long modeling time and poor accuracy under full operating conditions, the unsupervised Deep Temporal Clustering (DTC) method and supervised lightweight 3D convolutional neural network (3DCNN) are respectively used to identify operating conditions and construct local models. The test results show that the proposed deep learning model has high reliability, and the shape and amplitude of the predicted results of the dynamic load are basically consistent with the real results. Compared with the traditional model, the MAE index and MSE loss of the proposed model are smaller and have better performance.