The torque prediction of heavy-duty vehicle driveshaft based on MSGWO-LSTM

The driveshaft is the core component in the driveline system of Heavy-Duty Vehicles (HDV). Its torque transfer performance is related to the operational capability. The torque prediction is of great significance for studying the dynamics of HDV. Therefore, this paper proposes a driveshaft torque prediction model using a Multi-Strategy Grey Wolf Optimizer (MSGWO) to optimize the Long Short-Term Memory (LSTM) network. First, a torque calculation method based on real vehicle sensing data is proposed to provide samples for neural network training. Second, the MSGWO is used to optimize the hyperparameters of LSTM to construct the prediction model based on MSGWO-LSTM. Finally, the model is validated using a dataset composed of real vehicle sensing data and calculated torque. Compared to GWO-LSTM, LSTM, and Backpropagation (BP) models, the MSGWO-LSTM model proposed in this paper reduces the RMSE of real-time shifting torque prediction values by 41.59%, 66.14%, and 86.78%, respectively. The RMSE of short mileage prediction value decreases by 43.63%, 62.22%, and 68.92%, respectively. The results show that the model has high prediction accuracy and can effectively predict the torque of the HDV driveshaft.