| Summary: | Self-driving vehicles has been being popular since it not only improves traffic safety and flow but also the efficiency of energy consumption. Technical advance of self-driving vehicles significantly becomes requiring the energy efficiency of the electric vehicles (EVs) since the most car has very limited power capacity which is based on sole battery source. In this research, a hybrid learning predictive control architecture is proposed to reduce the energy consumption of the EVs. This paper combines the model predictive control (MPC) to the deep reinforcement learning (DRL) networks to solve the optimization problems with constrained environments to maximize energy efficiency of the EVs. Especially, the computed cost of the predictive horizon is transferred into the DRL networks as the state value and the reward is inversely fed into MPC controller to find the optimal control strategy. Resultingly, the proposed architecture can reduce the terminal control cost function using precedent state value. The proposed algorithm was tested in the PGDrive simulation environment and the experimental results show the efficiency of the proposed energy saving architecture.
|
|---|