Abstract: To enhance the design efficiency of centrifugal compressor impellers and reduce computational resource consumption, a method combining improved particle swarm optimization (IPSO) and BP neural networks is proposed. A limited number of computational fluid dynamics (CFD) simulation samples are used to train the BP neural network, creating a mapping between impeller parameters and efficiency. IPSO is applied to optimize the network's parameters, and genetic algorithms (GA) are employ to identify the optimal performance parameters. Results show that the IPSO algorithm improved the network's prediction accuracy and optimization efficiency by enhancing adaptability and global search capability. The optimized impeller achieves a 1.34% increase in isentropic efficiency, a 1.04% increase in polytropic efficiency, and a 10.4% rise in flow

Key words: centrifugal compressor; CFD Simulation; impeller parameter optimization; BP neural network; genetic algorithm

摘要： 为了提高离心式压缩机叶轮设计效率并降低计算资源消耗，针对遗传算法优化中计算 量大、效率低的问题，提出基于改进粒子群优化算法（IPSO）优化 BP 神经网络的方法。通 过少量计算流体动力学（CFD）仿真样本，训练 BP 神经网络建立效率与叶轮参数的映射关系， 结合 IPSO 优化其参数，同时利用遗传算法（GA）确定叶轮的最佳性能参数。研究表明，改 进的 IPSO 算法通过增强粒子群的动态适应性和全局搜索能力，提高了 BP 神经网络的预测精 度和优化效率。优化后的叶轮等熵效率提高 1.34%，多变效率提高 1.04%，流量增加 10.4%。 该方法显著提升了离心式压缩机叶轮的设计效率和性能，为复杂流体机械的优化设计提供了 新思路。

关键词: 离心式压缩机；CFD 仿真；叶轮参数优化；BP 神经网络；遗传算法