基于神经网络-遗传算法的液力透平叶片型线优化

提出了一种叶片型线的多工况优化设计方法,该方法包括叶片型线参数化、优化的拉丁超立方试验设计、CFD技术、GA-BP(genetic algorithm-back propagation)神经网络与遗传算法.具体采用三次非均匀B样条曲线参数化叶片型线,优化的拉丁超立方试验设计方法在设计空间内获取训练GA-BP神经网络的样本点,各个样本点性能分析由CFD技术完成,随后开展GA-BP神经网络的学习训练,最后采用GA-BP神经网络和遗传算法相结合的优化技术求解液力透平叶片型线的多工况优化问题.基于上述优化方法对一液力透平进行了叶片型线的优化改进,结果表明,在保证扬程不小于相应初始扬程的约束条件下,优化后的液力透平效率在3个指定工况下分别提高了3.91%,3.59%和3.09%, 证明采用此方法优化叶片型线具有一定的可行性. 

Blade pattern optimization of the hydraulic turbine based onneural network and genetic algorithm

A multi-condition optimization method, including the parameterization of blade pattern, the optimization of Latin hypercube experimental design, the CFD techniques, the genetic algorithm-back propagation (GA-BP) neural network and genetic algorithm, was presented for the blade pattern. Specifically, the non-uniform cubic B-spline curve was used to parameterize the blade pattern, and the optimized Latin hypercube experimental design method was employed for the acquirement of the sample points of GA-BP neural network. The performance analysis of each sample point was accomplished by the CFD techniques. Then, the learning and training of the GA-BP neural network was carried out. Finally, the optimization techniques combining the GA-BP neural network and genetic algorithm were used to solve the multi-condition optimization problems of the blade pattern. Based on the above method, the blade pattern of a hydraulic turbine was optimized and improved. The results show that the efficiency of the optimized hydraulic turbine specified in three conditions is increased by 3.91%, 3.59% and 3.09%, respectively, ensuring the constraints of the head are not less than initial head of the hydraulic turbine. This proves that using the above method to optimize the blade pattern is feasible.