An optimal mode decision method for energy management of series–parallel hybrid electric powertrain systems using Transfer-LSTM

Abstract

Series–parallel hybrid electric powertrain systems have emerged as an effective solution for enhancing energy efficiency. In series–parallel hybrid electric powertrain systems, the decision to switch between series and parallel modes is critical for improving fuel economy. Consequently, mode decision methods constitute a prominent research focus, yet most approaches exhibit limited adaptability to varying driving conditions. To address these limitations, this study proposes an optimal mode decision method for energy management of series–parallel hybrid electric powertrain systems using transfer long short-term memory (Transfer-LSTM) that leverages a pretraining phase in the source domain followed by fine-tuning in the target domain, thereby effectively adapting to different driving conditions. The dynamic programming algorithm is employed to find the mode decision sequence that minimizes fuel consumption. To determine the most influential features impacting mode decisions, a sequence of relevant feature parameters is extracted, and a self-organizing feature map is devised for dimensionality reduction. The features are trained via the LSTM network under a transfer learning technology to yield an optimal decision-making network. A specialized loss function is designed to ensure closeness to the minimum fuel consumption sequence while reducing unnecessary mode switching. The results show that, compared with the rule-based method, the proposed mode decision method reduces fuel consumption by 8.74%. Meanwhile, compared to the conventional network, it significantly reduces the training time from 363 to 302 s, shortening it by ∼16.8% while achieving performance comparable to dynamic programming. Finally, the practical effectiveness of the proposed method is validated in a hardware-in-the-loop test.