Wheel-Rail Force Identification Method Based on CNN-BiLSTM Hybrid Model

Abstract

layer is designed as the output wheel-rail force identification result. Taking the C80 vehicle as an example for analysis, the performance of the proposed method is evaluated from three aspects: model identification accuracy, generalization, and robustness. The results show that compared to traditional algorithms and single network models, the proposed method reduces the MSE value of wheel-rail lateral force identification by 44.4%~78.5%, and increases the R2 value by 1.3%~132.4%; the MSE value of wheel rail vertical force identification by 36%~75.9%, and the R2 value by 4.4%~87.9%. The proposed method can be applied to data of different working conditions and different noise levels.