Research on Energy Management Strategy Based on Deep Q Learning

Hybrid electric vehicles offer a solution to traditional fuel vehicle emissions and range anxiety in pure electric vehicles, serving as a crucial transitional product from internal combustion engine models to pure electric models. However, the Q-learning algorithm is limited to low-dimensional state problems due to the need to discretize both state and action, resulting in dimension explosion in the Q-value table, leading to low computational efficiency and convergence difficulties with few samples. To address this, the study proposes combining Q-learning with deep learning to form deep Q-learning for energy management problems. By utilizing a deep neural network as the Q-value approximation function, continuous or high-dimensional states can be effectively handled. Simulation results demonstrate that the proposed strategy differs by 7.36% from optimal fuel economy under NEDC offline training conditions, with the DQN-EMS effectively maintaining state of charge (SOC) and reducing equivalent fuel consumption. This approach presents a promising method for optimizing energy management in hybrid electric vehicles.