改进的求解非线性方程组的迭代神经网络算法  

Improved iterative algorithm of neural network for solving nonlinear equation groups

作　　者：赵庆兰[1] 李文[2] 董晓丽[1] 

机构地区：[1]西安邮电大学,陕西西安710121 [2]西安工业大学,陕西西安710032

出　　处：《现代电子技术》2013年第8期20-22,26,共4页Modern Electronics Technique

基　　金：陕西省教育厅自然科学基金项目(11JK0897);陕西省教育厅科学研究计划基金资助项目(12JK0551)

摘　　要：利用BP神经网络可以逼近多变元非线性方程组的函数的反函数,并利用迭代算法进行求解方程组的根。在实验中发现迭代算法在迭代过程中由于误差过早收敛出现无穷迭代,且由于误差并不一直随迭代次数的增加而减少导致最后收敛的值并不一定是最小值。针对这两个问题对其进行了改进,并利用算例进行了实验分析。在输入区域偏离解区域很远的情况下也能求得方程组的近似解。Neural network can precisely approach the function and inverse function of multivariate nonlinear equation groups, and solve the groups by using iterative algorithm. In the experiment, the endless iterations caused by premature conver- gence of error was found in the iteration process, and it is also found that the value of convergence not always is the minimum. The iterative algorithm is improved for the two problems, and the experimental analysis is performed by using examples. The ap- proximate solution of the equation can be got even if input sample interval deviates from the right solution very far.

关 键 词：BP神经网络 非线性方程组 迭代算法 误差收敛 

分 类 号：TN911.34[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]