Eco-Driving of Fuel Cell Hybrid Electric Vehicle in Variable Adhesion Coefficients and Obstacle Avoidance Environments

Eco-driving control holds significant potential for energy savings in clean energy vehicles, but its development in this area has been inhibited by complex traffic scenarios and varying road surface conditions. This study proposes a novel eco-driving approach that takes into account the stochastic distribution of road obstacles and the variable road-tire adhesion coefficient during driving. First, a mathematical model of fuel cell hybrid electric vehicles and their power systems is constructed, and convolutional neural networks are used to classify road surfaces. Next, a Markov decision process that matches the obstacle avoidance decision scenarios is systematically established, and a deep reinforcement learning approach is employed to design the decision threshold for obstacle avoidance decisions. Finally, the quadratic programming algorithm is introduced to optimize a coarse velocity profile based on light decision-making, and the original energy management system cost function is convex optimized. Simulation results show that compared with the 3-second rule method, the proposed eco-driving approach not only ensures vehicle safety in obstacle avoidance but also provides more stable state of charge under varying road conditions. Additionally, hydrogen consumption savings improved by 2.16% to 9.67% across different test scenarios.