As the lift of turbine increases, the secondary flow in the endwall region seriously restricts improvement of the aerodynamic performance of low-pressure turbine. The particle swarm optimization is improved to pursue stronger optimization ability. An optimization platform is built that combines Bayesian optimization (BO), the eXtreme Gradient Boosting (XGBoost) algorithm and improved chaos particle swarm optimization. The test results show that the surrogate model optimization method proposed in this paper has more prominent application potential in the field of aerodynamic optimization. The above optimization method is used for the optimization design of the nonuniform height endwall fence (NHEF) of a high-lift low-pressure turbine cascade to achieve an efficient secondary flow control effect. The NURBS curve is used to realize the parameterization of the NHEF. The secondary flow control mechanism of traditional fences and NHEF has been compared and analysed in detail. The numerical results show that the NHEF can effectively prevent the cross migration of the boundary layer, and the fence vortex induced by the fence can suppress the development of the passage vortex. The total pressure loss at the outlet of the cascade installed NHEF is reduced by 14.82%. The secondary flow control effect of the NHEF is better than that of a traditional fence. The feature importance analysis results of input variables show that the pitchwise position of the fence has the most obvious influence on the loss of the cascade. In addition, based on the results of feature importance, a 53% length reduction fence is designed to reduce the extra weight brought by the fence. The design achieves the effect of comprehensively considering the weight and aerodynamic performance.

随着涡轮升力的增加，端壁区域的二次流严重制约了低压涡轮气动性能的提高。对粒子群优化进行改进，追求更强的优化能力。构建了一个结合贝叶斯优化（BO）、极端梯度提升（XGBoost）算法和改进的混沌粒子群优化的优化平台。试验结果表明，本文提出的代理模型优化方法在气动优化领域具有更加突出的应用潜力。将上述优化方法用于高扬程低压水轮机叶栅非均匀高度端壁围栏（NHEF）的优化设计，以达到高效的二次流控制效果。NURBS曲线用于实现NHEF的参数化。对传统fences和NHEF的二次流控机制进行了详细的对比分析。数值结果表明，NHEF可以有效阻止边界层的交叉迁移，栅栏诱发的栅栏涡可以抑制通道涡的发展。安装NHEF的叶栅出口总压损降低14.82%。NHEF的二次控流效果优于传统围栏。输入变量的特征重要性分析结果表明，栅栏的俯仰位置对级联损失的影响最为明显。另外，根据特征重要性的结果，减少53%长度的围栏旨在减少围栏带来的额外重量。该设计达到了综合考虑重量和气动性能的效果。