基于深度强化学习的轮毂电机驱动电动汽车垂向振动控制

Vertical Vibration Control of an In-wheel-Motor DrivenElectric Vehicle Based on Deep Reinforcement Learning

轮毂电机驱动电动车作为分布式驱动的一种理想的解决方案,对于缓解能源问题具有重要意义,然而当轮毂电机引入轮毂时,其平顺性会恶化。为解决轮毂电机电动汽车平顺性问题,建立了考虑座椅、车身和簧下质量振动特性以及主动与半主动悬架时间迟滞因素的三自由度轮毂电机电动汽车的1/4车模型,并基于深度强化学习算法对轮毂电机驱动电动汽车通过主动悬架进行垂向振动控制。在此基础上,对轮毂电机驱动电动车在随机路面与减速带路面下行驶的情况进行训练,进而对其训练案例的控制效果进行测试,并将之与被动悬架和天棚阻尼控制策略的控制效果进行对比,最后对轮毂电机驱动电动车在随机路面的基于深度强化学习主动悬架控制策略进行泛化能力测试。结果表明,对于训练与泛化测试案例,针对轮毂电机驱动电动车的垂向振动控制,基于深度强化学习的主动悬架控制策略所产生的控制效果均优于被动悬架与天棚阻尼控制策略。

In-wheel-motor driven electric vehicles as an ideal solution for distributed drive are important for alleviating energy problems,but their riding comfort deteriorates when in-wheel-motors are introduced to the wheels.To solve the smoothness problem of in-wheel-motor electric vehicle,this paper established a quarter-vehicle model of a 3-degree-of-freedom in-wheel-motor electric vehicle considering seat,body and unsprung mass vibration characteristics as well as active and semi-active suspension time delay factors,and the vertical vibration control of the in-wheel-motor driven electric vehicle through the active suspension was carried out based on the deep reinforcement learning algorithm.Firstly,the in-wheel-motor driven electric vehicle was trained to drive under random road surface and speed bump road surface,after that,the control effect of its training case was tested and compared with the control effect of passive suspension and skyhook damping control strategies,and finally,the generalization ability of the deep reinforcement learning-based active suspension control strategy for in-wheel-motor driven electric vehicle on random road surface was tested.The results show that for both the training and generalization test cases,the deep reinforcement learning-based active suspension control strategy produces better control results than the passive suspension and skyhook damping control strategies for vibration control of in-wheel-motor driven electric vehicles.