改进半前馈神经网络学习算法

STUDY ON IMPROVING LEARNING ALGORITHM OF HALF FEEDFORWARD NETWORKS

针对半前馈神经网络传统学习方法中存在的收敛速度慢、易陷于局部最优提出了改进方法，将变尺度法（ＢＦＧＳ法）引入样本训练中．用该算法进行水电站Ｈ－Ｎ－Ｑ曲线的拟合，表明改进是有效的．

An improving learning algorithm for half feedforward networks is presented. Common learning algorithm not only has slow restraining velocity, but also is easy to pluge into local optimization. BFGS method is plused into sample training. It has been applied to simulating H-N-Q curve of hydroelectric station and has better results.