融合电池温度和寿命的深度强化学习PHEV能量管理策略

Deep reinforcement learning of PHEV energy management strategies incorporating battery temperature and life

针对当前插电式混合动力汽车能量管理策略忽略电池老化成本和电池温度变化过大而导致的热失控问题,制定融合电池寿命和电池温度的深度Q-Learning神经网络(DQN)强化学习能量管理策略.首先,从融入能量管理策略的角度,建立动力电池热模型和老化模型,引入调节目标价值函数的严重因子和量化电池老化程度的安时通量.其次,建立由超温惩罚、等效电池老化成本和燃油消耗组成的目标价值函数,进而构建深度强化学习能量管理策略.最后,通过仿真实验对所制定的控制策略进行验证.结果表明:融合了电池老化和电池温度的能量管理策略能够有效抑制电池老化和温度.在4个随机工况中,DQN策略下的电池有效安时通过量相较于CD-CS最大下降了35.75%;与CD-CS相比,DQN策略下单个驾驶任务的行驶总成本最大降低10.36%,证明了所制定策略的有效性.

Due to plug-in hybrid vehicle energy management strategies that ignore battery aging costs and thermal runaway caused by excessive battery temperature changes,a deep Q-Learning neural network(DQN)reinforcement learning energy management strategy that integrates battery life and battery temperature is developed.First of all,from the perspective of integrating energy management strategies,the power battery thermal model and aging model are established,and the severity factor to adjust the object function and the ampere-hour throughput to quantify the degree of battery aging are introduced.Secondly,an object function composed of over-temperature penalty,equivalent battery aging cost and fuel consumption is established,and then a deep reinforcement learning energy management strategy is constructed.Finally,the established control strategy is verified through simulation.The energy management strategy that integrates battery aging and battery temperature can effectively suppress the increase in battery temperature.In the four random driving cycles,the Ah-through output battery under the DQN strategy is 35.75%lower than that of the CD-CS.Compared with CD-CS,the total driving cost of a single driving task under the DQN strategy is reduced by 10.36%,which proves the effectiveness of the strategy.