LSTM-based adaptive energy management of connected hybrid mining trucks for improving fuel efficiency

Vehicle connectivity provides excellent opportunities for collecting valuable information on driving, road, traffic, and environmental conditions, which can then be utilized to improve a wide range of performance characteristics. In this study, we show how such historical information on driving conditions of connected hybrid mining trucks can be leveraged to develop a new adaptive equivalent fuel consumption minimum strategy (A-ECMS) for reducing their fuel consumption. We choose a power-split hybrid system as our baseline powertrain which includes a diesel engine, two motor-generators, three planetary gear sets working as a power-split device, and a battery. We employ the vehicle's traveling information along with the tuned equivalent fuel consumption minimum strategy (ECMS), developed by using the truck powertrain model, to train a long short-term memory (LSTM) network. The trained LSTM is then employed to solve the reference value of the state of charge (SOC) in real-time according to the current state of the vehicle. In the LSTM training process, a random disturbance is added to improve the robustness to the vehicle's speed variations. Adjustments by a fine-tuned proportional-integral (PI) controller eliminates the difference between the current SOC and the reference SOC to achieve an optimal adaptive equivalent coefficient. We apply the genetic algorithm (GA) to fine-tune the gains of the PI controller for different SOC values and road segments. Simulation results in the MATLAB/Simulink environment show that the resulting A-ECMS strategy reduces the vehicle fuel consumption up to 6.5% as compared with a rule-based (RB) control strategy.