Passenger-aware reinforcement learning for efficient and robust energy management of fuel cell buses

Abstract

Energy management strategies (EMSs) are essential for enhancing the efficiency, durability, and economic viability of fuel cell buses (FCBs). However, existing EMSs typically rely on fixed vehicle loads or idealized passenger assumptions, while neglecting the dynamic variations in passenger number and composition. This simplification introduces biased power demand distributions, underestimates the impact of human-occupancy heat loads under hot-weather conditions on air-conditioning system (ACS) energy use, and ultimately hinders the reproducibility of reported energy savings in real-world operation. To address these limitations, this study proposes a passenger-aware collaborative EMS aimed at enhancing the driving economy of FCBs under hot-weather conditions. Distinct from prior approaches, this study leverages a dual-source passenger perception framework that fuses video recognition with electronic card swiping data to obtain reliable real-time estimates of both passenger count and gender distribution. Gender-dependent body mass differences and heterogeneous metabolic heat generation are systematically integrated into the EMS framework, ensuring accurate modeling of passenger-induced variations in vehicle mass and cabin thermal load. Within this framework, the twin delayed deep deterministic policy gradient algorithm achieves the coordinated control of the fuel cell output power and the ACS cooling capacity. Extensive evaluations under real-world driving cycles and surveyed passenger datasets demonstrate the superiority of the proposed EMS. Compared with state-of-the-art baselines, the proposed method achieves at least a 0.62 % reduction in ACS energy consumption and a 2.11 % reduction in overall operational costs, without compromising cabin comfort. Importantly, in a representative scenario with 40 passengers, this method improves driving economy by 0.92–1.87 % over a gender-agnostic baseline at male passenger proportions of 0 %, 50 %, or 100 %, confirming the practical significance of incorporating passenger information. Given that urban buses operate continuously and costs scale near-linearly with energy and degradation, even modest percentage improvements over fleet-scale deployments and vehicle lifetimes can yield meaningful economic benefits.