Abstract: In order to explore the method of optimization and improve the performance of centrifugal compressor with splitters，an optimization system is developed. Bézier curves are used to realize parametric representation and approximate function between geometry parameters and aerodynamic performance is established by artificial neural network. The process of optimization is realized by genetic algorithm optimization together with numerical calculation. To verify the feasibility of the system，a compressor with splitters is optimized to achieve maximum isentropic efficiency. It is significant that the isentropic efficiency increases by 3.68% and the pressure ratio increases by 0.506 at the optimization point. The efficiency increases significantly at all operating conditions at design speed with an expansion of the surge margin. Numerical simulation shows that the relative Mach number at leading edge decreases at 90% span and 50% span，and the loss caused by intersection of shock and boundary layer is reduced. Furthermore，the secondary flow is inhibited and the low-speed area is largely reduced. In addition，the load of the blades is redistributed and part of the load is transferred from the splitters to the main blades.

Key words: Centrifugal compressor；Splitter；Optimization system；Parameterization；Aerodynamic optimization

摘要： 为探索带分流叶片离心压气机的优化方法，进一步提高气动性能，建立带分流叶片离心压气机优化系统:运用Bézier函数建立叶型的参数化表征方法，并采用人工神经网络建立目标函数和几何优化变量间的对应关系，结合遗传算法求解优化命题。为验证系统可行性，以极大化等熵效率为优化目标，针对带分流叶片的离心压气机进行优化设计。结果表明:该优化系统明显提高带分流叶片离心压气机的气动性能，优化后设计点等熵效率提高3.68%，压比提高0.506，设计转速下全工况范围内效率明显提升，综合稳定裕度增大；优化后90%，50%叶高波前马赫数有明显降低，前缘激波有所减弱，激波与附面层相互作用引起的损失减小，二次流得到抑制，叶尖通道内大范围低速区明显消退；叶片载荷重新分配，分流叶片载荷一部分转移至主叶片。

关键词: 离心压气机；分流叶片；优化系统；参数化法；气动优化