摘要
针对经典BP神经网络运行中存在的缺陷,提出了改进的BP神经网络,不仅解决了经典BP网络易陷入局部最小的弊端,而且应用的0．618分割选取法能使网络快速找到较优隐含层节点数,初始权值的自相关修正进一步提高了网络的稳定性。实际应用证明,改进的BP神经网络有效提高了网络质量,适合大型网络的构建与训练。
Aiming at the demerits of the classical BP neural network, an improved BP neural network is put forward, by which the problem of easy relapse into local minimum of the classical BP network is solved. Besides, a good node number of the hidden layer can be found quickly by use of the 0. 618 section selection, and the stability of the network can be enhanced with self-correlation correction for the initial weights. The application shows that the improved BP neural network can improve the network quality effectively, and fits for the construction and training of big networks.
出处
《水电自动化与大坝监测》
2006年第4期62-65,共4页
HYDROPOWER AUTOMATION AND DAM MONITORING
关键词
大坝安全监控
BP神经网络
数值优化
0.618分割选取
自相关修正
dam safety monitoring
BP neural network
numerical value optimization
0. 618 section selection
self-correlation correction
