摘要
训练BP网时采用拆分组装法,对防止BP网陷入局部极小值很有效.在这一方法基础上,针对其缺点提出了几个改进措施,使用这些措施可得到结构更加紧凑的BP网,且其输出也有一定关联;对其评价函数作了理论分析,使训练BP网时其精度如何控制有了依据;最后介绍了一个BP网专家系统,验证拆分组装法、BP算法的正确性.
Split-assembly solution is valid in preventing BP network from entering the local minimum in training. On the basis of this method, this article puts forward several improvement measures for its disadvantages. With these measures, we can get a better BP network with compact structure and associated output. This article also analyzes the estimate function of BP network, which provides a basis for the control of its precision. An expert system is presented which validates the split-assembly solution and BP arithmetic.
出处
《湖北工业大学学报》
2007年第3期14-17,共4页
Journal of Hubei University of Technology
关键词
BP神经网络
BP算法
拆分组装法
BP neural network
BP arithmetic
Split-assembly solution