Abstract:

A multi-objective optimization based on the combination of neural network and non-dominated sorting genetic algorithm (NSGA-Ⅱ) was carried out for a battery thermal management system with phase change material-coupled liquid-cooling heat dissipation. Firstly, the effects of design variables such as different cell spacing, coolant flow rate and coolant channel aspect ratio on the thermal dissipation effect of the battery were investigated using numerical simulation. The neural network was trained by simulating the target values obtained from the design variables using Latin Hypercubic Sampling (LHS) to obtain the mapping relationship between the design variables and the maximum battery temperature, T_max, and the maximum temperature difference, ∆T_max. Subsequently, the NSGA-Ⅱ algorithm with an elite retention strategy was adopted to seek the Pareto front for the three-objective optimization of minimizing the volume (V_b), the maximum temperature (T_max), and the maximum temperature difference (∆T_max) of the battery pack, and to determine the optimal design. The ultimate optimization results indicate that the multi-objective optimization approach combining the neural network with the NSGA-Ⅱ algorithm is highly effective. In comparison with the initial design, the V_b is decreased by 7.6%, while the energy density is enhanced by 8.22%. The T_max and ∆T_max of the battery group are 34.89℃ and 4.02℃ respectively. Additionally, the utilization rate of the phase change material (PCM) is increased by 7%. The power consumption of the coolant pump has decreased from 8.79×10^-4W to 4.065×10^-4W, with a reduction of up to 53.75%.

Key words: lithium battery pack thermal management system, phase change cooling, liquid cooling, multi-objective optimization, computational fluid dynamics, neural networks

摘要：

针对一种相变材料耦合液冷散热的电池热管理系统进行了基于神经网络与非支配排序遗传算法（NSGA-Ⅱ）结合的多目标优化。首先使用数值模拟方法研究了不同电池间距、冷却液流速和冷却液通道纵横比等设计变量对电池散热效果的影响，然后利用拉丁超立方抽样（LHS）生成设计变量经数值模拟获得目标值进行神经网络训练，建立起设计变量与电池最高温度和最大温差之间的映射关系。随后采用带精英保留策略的NSGA-Ⅱ算法找到最小化电池包体积（V_b）、电池最高温度（T_max）和电池最大温差（∆T_max）的三目标优化Pareto前沿并确定最佳设计。最终优化结果表明，神经网络与NSGA-Ⅱ算法结合的多目标优化方法十分有效。相较于初始设计，优化后的V_b降低了7.6%，能量密度提升了8.22%；电池组T_max和ΔT_max分别为34.89℃和4.02℃；相变材料（PCM）的利用率提升了7%；冷却液泵功耗从8.79×10^-4W降低到了4.065×10^-4W，降低幅度高达53.75%。

关键词: 锂电池热管理系统, 相变冷却, 液冷, 多目标优化, 计算流体力学, 神经网络