An Improved LSTM-Based Speed Predictor Applied to Energy Management for Fuel Cell Electric Vehicles

Highly accurate speed prediction technology is of great significance for online implementation of energy management strategy (EMS). Due to the complex and variable driving conditions, the accuracy of conventional speed prediction method is yet to be improved by driving pattern adaption. This paper proposes a vehicle speed prediction method based on long- and short-term memory neural network (LSTM) with driving pattern recognition and integrates it in energy management framework based on model predictive control (MPC). First, similar samples belonging to the same driving pattern are selected offline to train a more efficient and targeted LSTM. Then an online speed prediction algorithm based on driving pattern recognition is proposed. The results show that root mean square error (RMSE) of the whole driving cycle is reduced by 39% compared to conventional LSTM. Meanwhile, fuel economy and fuel cell system (FCS) durability are improved, which proves the effectiveness of the proposed method. And the real-time applicability of the proposed predictive EMS is verified.