6 March 2024 in IEEE Transactions on Intelligent Transportation Systems
Fuel cell vehicles represent a groundbreaking form of new energy vehicles, garnering considerable attention and development in recent years. The energy management of fuel cell hybrid electric vehicles has always been a research difficulty. This paper aims to improve the economy and durability of fuel cell and lithium battery hybrid systems, and proposes an innovative online energy management strategy that incorporates road condition recognition. Key features from multiple driving datasets are extracted to construct a classifier based on ensemble learning and decision tree, which can effectively identify road patterns. Then, a fuzzy Q-learning method is proposed for online power allocation. Different action spaces of Q-learning are designed to cater to different road patterns, enabling more effective adaptation to diverse roadway environments. In addition, the reward function is designed to account for hydrogen costs and degradation expenses. Through a series of simulation and hardware tests, the proposed method has been demonstrated to effectively achieve online energy management of fuel cell/battery system under varying road conditions. And a quantitative comparison validates that the proposed method outperforms traditional reinforcement learning algorithms in reducing both fuel consumption and system degradation rates.
7 July 2025 in IEEE Transactions on Transportation Electrification
Combining fuel cells and lithium-ion batteries (LiBs) in a hybrid vehicle presents a promising automotive energy supply technology. The durability of power components, fuel consumption, and safety all depend on effective energy management strategies (EMSs). This article is the first to use the context-aware typed multipopulation genetic programming (CAT-MPGP) algorithm to extract interpretable rules for the offline near-optimal solution of Pontryagin’s minimum principle (PMP), which are applied to the real-time EMS considering system aging factors. First, this article integrates the thermal model of LiBs into the system model to balance system lifespan and energy consumption. The PMP is employed to find the offline near-optimal power allocation results. Second, Bagging method is used to construct the driving condition recognizer, and the interpretable data relationship between driving information and power distribution is mined by MPGP algorithm, and embedded in real-time EMS. Finally, simulations using a real driving route and two standard periodic datasets are conducted as the test sets to compare the results of different EMSs. Hardware-in-the-loop (HIL) test is adopted to verify the feasibility of the controller in practical application. The proposed method achieves a well-performing real-time EMS, which reaches 94.6% of the offline reference results.