Abstract: In order to improve the performance of the centrifugal fans in the cooling systems of high-speed trains, the parametric model of the centrifugal impeller structure and the computational fluid dynamics(CFD) simulation model were adopted. Combined with the performance tests, the CFD simulation model of the centrifugal fan performance was established. The reasonable values of blade number and impeller width were identified by single factor sensitivity analysis respectively. The Latin hypercube sampling(LHS) model was used to sample the impeller structure and performance parameters of the centrifugal fans. The RBFNN model was used to study the structure and performance parameters, and the performance analysis model of centrifugal fans was reconstructed. The multi-factor sensitivity of the structure parameters of centrifugal impellers was studied. Through the global sensitivity analysis, the key parameters affecting the performance were identified. Finally, an optimization scheme was proposed to improve the performance of the centrifugal fans by reasonably matching the structural parameters of the centrifugal impellers. The efficiency of the working point is increased from 67.2% to 73.7%, which may significantly improve the efficiency of the centrifugal fans and reduce the flow loss.

Key words:  , high speed train, radial basis function neural network(RBFNN), sensitivity analysis, centrifugal fan, parametric model

摘要： 为提高高速列车散热系统的离心风机性能，采用离心叶轮结构参数化模型和计算流体动力学（CFD）仿真模型，结合气动性能试验，建立了离心风机性能的CFD仿真模型；通过单因素灵敏性分析，识别了叶片数和叶轮宽度的合理值。采用拉丁超立方抽样模型对离心风机的叶轮结构与性能参数进行采样；通过径向基神经网络模型对结构和性能参数进行学习，重构了离心风机气动性能分析模型，并对离心叶轮结构参数进行了多因素灵敏性研究。通过全局灵敏性分析，识别了影响性能的关键参数。合理匹配离心叶轮结构参数，提出了改进离心风机性能的优化方案，工况点的效率由67.2%提高至73.7%，显著提高了离心风机效率，减小了流动损失。

关键词: 高速列车, 径向基神经网络, 灵敏性分析, 离心风机, 参数化模型