摘要
根据梯度算法中网络权值的演化规律,并基于终端吸引子,提出一种能全局寻优自适应的快速BP算法,该算法的基本思想是最小二乘算法,采用梯度搜索技术,以期使网络的实际输出值与期望值的误差的均方值为最小.同时,进行BP学习算法的稳定性和快速收敛问题分析研究.并进一步给出改善BP算法学习率修正、假饱和现象消除等训练结果的措施.
In order to optimize the BP online algorithm,globally convert adaptive quick back propagation algorithm was proposed based on termind attractors according to the law of evolution of the Neural Networks's authority,and the measures for improving the performance of this algorithm were discussed.
出处
《应用科技》
CAS
2004年第6期46-47,50,共3页
Applied Science and Technology
