为充分发挥航空推进系统的性能，提高性能寻优控制的实时性，将樽海鞘群算法（SSA）与极限学习机（ELM）相结合，基于进/发一体化部件级模型建立数据集，提出一种基于SSA-ELM的数据驱动模型。将该建模方法与广义回归神经网络（GRNN）、BP神经网络（BPNN）和极限学习机（ELM）比较，结果表明，相比于BPNN，ELM，GRNN，SSA-ELM用于预测可以使安装推力的均方根误差（RMSE）分别降低7.41%，17.01%，72.57%，安装油耗的RMSE分别降低4.32%，19.41%，66.77%，具有更高的预测精度。将基于SSA-ELM的数据驱动模型作为机载模型应用到性能寻优控制，结果表明，该机载模型能够维持理想的寻优效果。针对最大安装推力模式开展实时性分析，该机载模型相比于进/发一体化部件级模型，平均计算时间由184.05 ms缩短至1.357 ms，实时性得到显著改善，大大提高了寻优效率。

To fully exploit the performance of the aero-propulsion system and improve the real-time performance of performance seeking control， a data-driven model based on SSA-ELM is proposed by combining Salp Swarm Algorithm （SSA） and Extreme Learning Machine （ELM） and establishing data sets based on integrated inlet/engine component-level model. The modeling method was compared with General Regression Neural Network （GRNN）， BP Neural Network （BPNN） and Extreme Learning Machine （ELM）. The results show that compared with BPNN， ELM and GRNN， SSA-ELM for prediction can reduce the Root Mean Square Error （RMSE） of installed thrust by 7.41%， 17.01%， and 72.57%， respectively， and the RMSE of installed fuel consumption by 4.32%， 19.41%， and 66.77%， respectively， which has higher prediction accuracy. The data-driven model based on SSA-ELM was applied to the performance seeking control as on-board model. The results show that the on-board model can maintain the ideal seeking effect. In the real-time performance analysis of the maximum installed thrust mode， the average computation time of the on-board model is reduced from 184.05 ms to 1.357 ms compared to the integrated inlet/engine component-level model， which significantly improves the real-time performance and greatly enhances the seeking efficiency.

V233.7

国家自然科学基金（61890921；61890924）；国家科技重大专项（J2019-I-0019-0018）；中央高校基本科研业务费（DUT22QN204）。