基于深度强化学习的新能源汽车动力电池组降温自适应控制系统设计

Design of Adaptive Control System for Cooling of New Energy Vehicle Power Battery Pack Based on Deep Reinforcement Learning

摘要:

在新能源汽车高负荷运行场景下,动力电池组面临复杂电热耦合环境,其内部电化学反应剧烈、热积累效应显著,易引发温场能效失衡问题,难以实现热状态的精准感知与实时控制。为了解决该问题,设计了基于深度强化学习的动力电池组降温自适应控制系统。以数据驱动决策、闭环保障精准为核心,构建了三层协同的全流程闭环总体架构,围绕深度强化学习算法决策中枢,联动数据采集、信息交互、执行控制三大模块,实现动力电池热特性与冷却控制的自适应匹配。硬件上,采用“测温+测流”双组件协同设计的热特性多维度采集模块,确保数据精准量化；通过CAN总线实现整车工况通信模块的实时数据交互；自适应冷却指令执行模块则将控制指令转化为实际冷却动作,形成闭环控制。软件上,通过计算动力电池温度偏差变化率,量化产热、温度偏差与趋势递进关系,结合深度强化学习算法,构建“状态输入-行为输出-奖励函数-策略更新”闭环控制框架,实现温场能效平衡,完成动力电池热特性的自适应降温。在高温爬坡极端工况下进行系统测试,结果表明,该方法在温度分布均匀性(温差≤1℃)与响应延迟(≤5ms)方面均显著优于现有主流控制方法,展现出良好的控制性能。

Abstract:

In the high load operation scenario of new energy vehicles, the power battery pack faces a complex electric thermal coupling environment, with intense internal electrochemical reactions and significant thermal accumulation effects, which can easily lead to temperature field energy efficiency imbalance problems, making it difficult to achieve accurate perception and real-time control of thermal state. To address this issue, a deep reinforcement learning based adaptive control system for cooling power battery packs was designed. With data-driven decision-making and precise closed-loop assurance as the core, a three-layer collaborative full process closed-loop overall architecture has been constructed. Around the deep reinforcement learning algorithm decision-making center, the three modules of data acquisition, information exchange, and execution control are linked to achieve adaptive matching between the thermal characteristics of power batteries and cooling control. On the hardware side, a multi-dimensional acquisition module for thermal characteristics is designed with the collaboration of "temperature measurement+flow measurement" dual components to ensure accurate data quantification; Real time data exchange of the vehicle operating condition communication module is achieved through the CAN bus; The adaptive cooling command execution module converts the control command into actual cooling action, forming a closed-loop control. On the software side, by calculating the rate of temperature deviation change of the power battery, quantifying the relationship between heat generation, temperature deviation, and trend progression, and combining deep reinforcement learning algorithms, a closed-loop control framework of "state input behavior output reward function strategy update" is constructed to achieve energy efficiency balance in the temperature field and achieve adaptive cooling of the thermal characteristics of the power battery. System testing was conducted under extreme conditions of high temperature climbing, and the results showed that this method significantly outperformed existing mainstream control methods in terms of temperature distribution uniformity (temperature difference ≤ 1 ℃) and response delay (≤ 5ms), demonstrating good control performance.