Automated multi-dimensional dynamic planning algorithm for solving energy management problems in fuel cell electric vehicles

Dynamic programming (DP) is widely recognized as an optimal method for addressing energy management challenges in fuel cell electric vehicles (FCEVs). However, traditional DP struggles with history-dependent and future-dependent constraints that violate Markov properties, such as state of power (SOP) and gas purge duration. Additionally, it faces issues related to penalty function tuning and discretization errors. To overcome these challenges, this study proposes a multi-dimensional automated DP (AuDP) algorithm. The algorithm addresses these critical time-dependent constraints by converting SOP and gas purge duration into Markov states. Moreover, it eliminates the need for penalty functions by using the fuel cell power rate as the control variable, defining state of charge (SOC) grid boundaries using the Brent iteration method, and integrating gas purge mode into the FCEV operating mode. The algorithm also introduces adaptive SOC grid density and SOC grid reconfiguration methods to minimize discretization errors, while employing parallel computing to expedite computation. AuDP is validated under CLTC and WLTC driving cycles, demonstrating its effectiveness in managing time-dependent constraints and co-optimizing the battery and fuel cell. Compared to conventional DP, AuDP shows significant improvements, achieving at least a 2.33 % increase in accuracy and a 57.63 % reduction in computational time.