초록 열기/닫기 버튼
The fuel economy of a hybrid electric vehicle(HEV) is vastly influenced by the manner power is distributed. A dynamic, programming-based power distribution strategy can provide a global optimal solution, but it is not applicable to an actual vehicle because it requires further driving information. On the other hand, the reinforcement learning-based power distribution strategy is highly applicable to an actual vehicle because it requires only the current state to construct the policy. Recently, deep Q-networks(DQN) have been developed by applying a deep neural network to reinforcement learning, leading to significant change in the field of reinforcement learning. DQN could solve complex tasks efficiently based on different studies. In this particular study, we developed an energy management strategy for HEVs that is applicable to actual vehicles, and can achieve high efficiency through the DQN.
