基于深度强化学习的锂电池快速充电控制策略

安全高效的锂电池充电控制策略对于电动汽车的发展具有重要推动作用。针对锂电池的快速充电问题,提出一种综合考虑锂电池充电速度、能量损耗、安全约束多目标优化充电控制策略。基于动作-评价网络框架,利用基于近端策略优化的深度强化学习算法,训练出使得充电目标对应的奖励函数最大的充电策略神经网络和策略评估神经网络。然后,利用训练完成的充电策略神经网络根据当前电价和电池SOC智能决策出最优的充电电流。该充电控制策略的优势在于能够在保证快速充电的同时,实现充电花费最小化。同时,充电策略神经网络在线运算量较小,与基于模型的在线优化算法相比更能满足充电控制的实时性要求。最后,仿真结果表明,该充电控制策略与传统恒流-恒压法相比,具有兼顾充电速度与电费支出的优势,满足快速充电任务需求的同时,最高可降低25%的充电成本。

Fast Charging Control for Lithium-ion Batteries Based on Deep Reinforcement Learning

Safe and efficient lithium battery charging control strategy plays an important role in promoting the development of electric vehicles. Aiming at the problem of fast charging of lithium batteries, a multi-objective optimal charging control strategy is proposed that comprehensively considers the charging speed, energy consumption and safety constraints of lithium batteries. Based on the action-evaluation network framework, a deep reinforcement learning algorithm based on proximal policy optimization is used to train a charging policy neural network and a policy evaluation neural network that maximize the reward function corresponding to the charging target. Then, using the trained charging strategy neural network to intelligently decide the optimal charging current according to the current electricity price and battery SOC. The advantage of this charging control strategy is that it can minimize charging costs while ensuring fast charging. At the same time, the online calculation amount of the charging strategy neural network is small, and compared with the model-based online optimization algorithm, it can better meet the real-time requirements of charging control. Finally, the simulation results show that, compared with the traditional constant current-constant voltage method, the charging control strategy has the advantages of taking into account the charging speed and electricity expenditure, and can reduce the charging cost by up to 25% while meeting the requirements of fast charging tasks.