神 经 网 络 模 型 在 压 气 机 通 流 特 性分 析 中 的 应 用

To solve the problem of insufficient prediction accuracy of the compressor performance by the original model in the through⁃flow analysis program，and to improve the reliability of the compressor through flow analysis process， a compressor cascade performance database was established based on the numerical simulation results of a large number of multiple circular arc cascades. Based on this database，neural network modeling method was used to establish the baseline loss coefficient and baseline deviation angle models of compressor cascade. Results showed that，the prediction accuracy of the two models for the baseline loss coefficient and baseline deviation angle of cascade met the requirements of engineering applications，with the accuracy of ±0. 002 and ±1° ， respectively. During the verification process， it could be found that the neural network models significantly improved the prediction accuracy of both compressor's overall performance and the flow details， especially at the core flow region. Moreover， the improvement of the accuracy of baseline loss coefficient and baseline deviation angle had a positive effect on the prediction accuracy of loss coefficient and deviation angle at off⁃design conditions.