氢燃料电池拖拉机双模自适应在线能量管理策略研究

Hydrogen Fuel Cell Tractor Dual-mode Adaptive Online Energy Management Strategy

摘要:

针对氢燃料电池拖拉机能量管理策略在线应用优化效果欠佳、对先验工况信息依赖强等问题,本研究提出一种双模自适应在线能量管理策略（Dual-mode adaptive online, DAO）。确立工况识别-优化决策-执行输出3层协同的总体架构,在上层工况识别层,通过构建基于随机森林的工况识别模块,实现犁耕与运输工况的精准判别;在中层优化决策层,以离线生成的全局最优控制序列为训练样本,以设计的改进金枪鱼群算法优化传统反向传播（Back propagation, BP）神经网络为优化求解器,分别针对犁耕与运输工况,构建中层神经网络控制器,形成DAO策略;在下层执行输出层,开展模型在环仿真与实车田间试验。由模型在环仿真结果可知,相较于离线动态规划策略,所提策略氢罐总氢耗在犁耕工况下降低0.002%,在运输工况下降低0.07%,表明所提策略具备良好的预测、决策能力。实车试验结果表明,DAO策略下等效氢耗率降低5.05%,验证了所提策略具有实时应用与性能提升能力。研究结果可为氢燃料电池拖拉机控制系统设计提供理论支撑。

Abstract:

Aiming at the problems of suboptimal optimization performance for online application and strong dependence on prior driving cycle information in energy management strategies for hydrogen fuel cell tractors, a dual-mode adaptive online (DAO) energy management strategy was proposed. A three-layer collaborative architecture comprising driving cycle recognition, optimization decision-making, and execution output was established. In the upper layer of driving cycle recognition, a random forest-based recognition module was constructed to accurately distinguish between plowing and transportation operating conditions. In the middle optimization decision-making layer, globally optimal control sequences generated offline were used as training samples, and the designed improved tuna swarm algorithm that optimized the traditional back propagation (BP) neural network served as the optimization solver. Neural network controllers were constructed for plowing and transportation conditions respectively, thereby forming the DAO strategy. In the lower execution output layer, model-in-the-loop simulations and real-vehicle field tests were carried out. Model-in-the-loop simulation results showed that, compared with the offline dynamic programming strategy, the proposed strategy reduced the total hydrogen consumption of the hydrogen tank by 0.002% under plowing conditions and by 0.07% under transportation conditions. This indicated that the proposed strategy possessed good prediction and decision-making capabilities under both plowing and transportation conditions. Real-vehicle test results showed that the equivalent hydrogen consumption rate was reduced by 5.05% under the DAO strategy, validating the real-time applicability and performance improvement capability of the proposed strategy. The research result can provide theoretical support for the design of control systems for hydrogen fuel cell tractors.