遗传算法结合改进前向神经网络进行叶轮设计

为加速遗传算法的进化过程和缩短其进化周期，探讨了将遗传算法融合神经网络进行离心叶轮形状优化设计的方法，即应用神经网格替代有限元法来完成结构优化设计中的静应力分析任务．同时，提出一种改进的前向反馈神经网络(BP算法)，在训练过程中，学习率和动量项依据输出的均方差自适应调整，来加快网络训练速度和改善收敛性．采用混合神经网络的遗传算法对某离心压缩机叶轮进行优化设计，结果表明优化设计时间可缩短至单纯采用遗传算法的几十分之一，同时也验证了该方法的有效性和可行的。

Combining Genetic Algorithm with Neural Networks for Design of Impeller

Neural networks are incorporated into the process of genetic algorithm to optimize the centrifugal impeller.Static stress analysis of the impeller is executed using neural networks in order to compute individual fitness value, which could speed up the evolution process of genetic algorithms and shorten the evolution time. In addition, in order to accelerate the process of training and improve the convergence, an improved feed-forward neural networks(BP algorithm)is proposed,and the learning rate and momentum term are adjusted according to output mean-square error when the neural network is trained.The genetic algorithm coupled with neural network is used to obtain the optimal shape of the impeller.The results show that the period of optimal design is tens times less than that of the genetic algorithm.So the algorithm is reasonable and reliable.