Energy management strategy of a novel parallel electric-hydraulic hybrid electric vehicle based on deep reinforcement learning and entropy evaluation

An excellent energy management strategy is paramount to the new energy vehicle safety, durability, and reliability, which invariably affects the driving experience. This paper proposes a novel parallel electric-hydraulic hybrid electric vehicle (PEHHEV), which has the characteristics of multiple working modes and power sources. Aiming at the defects of outmoded control strategies, this paper applies the long short-term memory (LSTM) neural network to the proximal policy optimization (PPO) algorithm. A PPO-LSTM-based energy management strategy for PEHHEV to achieve optimal working mode switching is established. To avoid paying too much attention to the economy and neglecting other performance parameters, we design a local sample Shannon entropy to realize dynamic evaluation for performance parameters. Through offline training, online testing, and entropy evaluation, the energy consumption rate under WLTC and NEDC increased by 18.51% and 15.74% respectively. It is indicated that PEHHEV can maintain dynamic performance and obtain lower energy consumption under the PPO-LSTM-based energy management strategy, which verifies its feasibility and robustness. The investigation in this paper can improve energy management performance and fill the literature gap, which has more preponderance than other strategies. This is the first of its kind to apply PPO-LSTM and entropy evaluation to developing control strategies for modern vehicles.

Automated summary

This paper proposes a novel parallel electric-hydraulic hybrid electric vehicle (PEHHEV) with multiple working modes and power sources. It applies the long short-term memory (LSTM) neural network to the proximal policy optimization (PPO) algorithm to establish an energy management strategy for optimal working mode switching. Through offline training, online testing, and entropy evaluation, the PEHHEV achieved lower energy consumption and maintained dynamic performance under the PPO-LSTM-based strategy.