Abstract
<div class="section abstract"><div class="htmlview paragraph">Reinforcement learning is a promising approach to solve the energy management for
hybrid electric vehicles. In this paper, based on the DQN (Deep Q-Network)
reinforcement learning algorithm which is widely used at present, double DQN,
dueling DQN and learning from demonstration are integrated; states, actions,
rewards and the experience pool based on the characteristics of series-parallel
multi-speed hybrid powertrain are designed; the hybrid energy management
strategy based on D4QN (Double Dueling Deep Q-Network with Demonstrations)
algorithm is established. Based on the training results of D4QN algorithm,
multi-parameter analysis under state and action space, HCU (Hybrid control unit)
application and MIL (Model in-loop) test research are conducted. The results
show that the D4QN algorithm can achieve both the approximate global optimal
results, which differs from the result of dynamic programming by 0.05% under the
training cycle, and establish the positive mapping relationship between state
variables and action variables of the hybrid powertrain with excellent
generalization ability, which can be applied to HCU and control the hybrid
electric vehicle effectively in real time. Compared with the original rule-based
energy management strategy for the hybrid electric vehicle, the fuel economy
under WLTC cycle is improved by 9.64% after the application of D4QN energy
management strategy. The proposed strategy can realize a termination of SOC
(State of charge) within 50±10% under different initial SOC states for a battery
of 1.8kWh, achieving the goal of SOC robustness. In addition, the proposed
strategy has strong adaptability to the unknown cycles; the fuel economy of the
vehicle after the application of D4QN energy management strategy in the
untrained NEDC and CLTC-P cycles is improved by more than 10% compared to the
WLTC cycle.</div></div>