应用果蝇优化算法(FOA)对广义回归神经网络(GRNN)平滑参数spread值进行优化,充分利用果蝇优化算法收敛速度快及径向基函数调整参数少的优点,建立厂房结构的振动响应预测模型,同时结合反向传播神经网络(BP)、局部回归神经网络(ELMAN)对某厂顶溢流式水电站的厂房结构振动响应问题展开对比预测研究。通过比较三种神经网络的预测效果,最终得出:基于果蝇算法优化的广义回归神经网络(FOA-GRNN),在预测能力、学习速度上明显优于BP网络和ELMAN网络。说明运用FOA-GRNN神经网络预测厂房结构振动响应是可行的,为增强厂房结构的智能化监测提供了保障。
A flies optimization algorithm (FOA) is used to optimize the spread value of generalized regression neural network (GRNN). This method takes advantages of FOA in fast convergence and GRNN in few parameters, and is compared with neural network prediction models (BP and ELMAN) for a comparative study on prediction features of the vibration responses of overflow structure on the roof of a hydropower station dam. The comparison of three models concludes that the GRNN based on FOA has both prediction ability and learning speed superior to BP or ELMAN. It also shows the feasibility of FOA-GRNN in vibration predictions that enhances intelligence in monitoring hydraulic structure.