Powertrain-Aware Explainable Reinforcement Learning Energy Management in Off-Road Vehicles

In this work, the authors introduce and evaluate a series-hybrid range-extender powertrain architecture for high-power off-highway vehicles, addressing key challenges in electrification. This architecture offers precise control of engine operating points but introduces multiple energy conversions that can lead to significant power losses. An effective energy management strategy (EMS) is essential to overcome the losses while minimizing fuel consumption, maintaining battery state of charge (SOC), and ensuring high engine efficiency. However, the highly variable, stochastic, and uncertain operating conditions typical of off-highway applications limit the effectiveness of threshold-rule-based control strategies and reduce the optimality of model-based optimal control strategies. To address these challenges, this work proposes an explainable, interpretable reinforcement-learning-based EMS that adapts to high-variance environments, handles uncertainty, and generalizes to unseen operating conditions. A powertrain constraints-aware, explainable, and interpretable Double Deep Q-Network (XI-DDQN) controller is devel-oped, trained on actual experimental duty cycles, and evaluated on unseen actual real-world scenarios to assess its adaptability and generalization. The proposed EMS is benchmarked against dynamic programming (DP) and compared to a rule-based controller. The results, based on the validated model, demonstrate that the proposed XI-DDQN successfully captures the optimal control behavior of DP while maintaining robust adaptability across uncertain and generalized operating conditions. The XI-DDQN-based EMS achieves up to a 23% reduction in fuel consumption compared with a conventional powertrain and delivers an average 11.7% reduction relative to the rule-based strategy under unseen operating conditions. These findings highlight the effectiveness of the proposed approach in achieving fuel-efficient, robust, and interpretable energy management for next-generation hybrid off-highway machinery. • Electrifying high-power off-road vehicles is challenging. • Series-hybrid with adaptive EMS addresses operating variability for maximum efficiency • Explainable and interpretable DDQN trained on real off-highway duty cycles • Robust control under uncertainty, unseen, and general operating conditions • Up to 23% fuel savings vs conventional and 11.3% vs rule-based