Abstract: This paper proposes an intelligent optimization method based on Gaussian process regression (GPR) surrogate model and particle swarm optimization (PSO) algorithm for the heat dissipation optimization of finned liquid cooling plates. First, a sample dataset of fin angles, temperatures, and pressures is constructed through COMSOL simulations, and the GPR surrogate model is trained to establish a high-precision response surface. On this basis, the PSO algorithm is used to perform global optimization within the feasible range of fin angles, with the optimization objectives of simultaneously reducing the bottom wall temperature of the liquid cooling plate and the internal flow pressure. This method can efficiently determine the optimal fin angle. The entire process uses only coarse mesh data to train the model, and the average relative errors of the peak temperature and pressure are 0.07% and 5.36%, respectively. At a flow rate of 500 mL/min, the peak temperature of the liquid cooling plate after fin angle optimization is reduced by approximately 8.94 K. This method significantly reduces the dependence of the optimization process on computing resources and provides an effective approach for the rapid design optimization of finned liquid cooling plates.

Key words: finned liquid cooling plate, Gaussian process regression, surrogate model, particle swarm optimization algorithm, thermal management

摘要： 该文针对翅片式液冷板散热性能优化问题,提出一种基于高斯过程回归代理模型与粒子群优化算法的智能优化方法。首先通过COMSOL仿真构建不同翅片角度下温度与压力的样本数据集,训练高斯过程回归代理模型以建立高精度响应面。在此基础上,采用粒子群优化算法在翅片角度可行域内进行全局寻优,优化目标设为同时降低液冷板底部壁面温度与内部流动压力。该方法能高效确定翅片最佳角度,且全过程仅使用粗网格数据训练模型,峰值温度与峰值压力的平均相对误差分别为0.07%与5.36%。在500 mL/min流速下,通过翅片角度优化后的液冷板峰值温度降低约8.94 K。该方法显著降低了优化过程对计算资源的依赖,为翅片式液冷板的快速设计优化提供了有效途径。

关键词: 翅片式液冷板, 高斯过程回归, 代理模型, 粒子群优化算法, 热管理