摘要
本文利用增加了匹配点数目的扩展点匹配方法,将广义的电路方程变为一适宜用具有高度集群、并行特性的Hopfield神经网络求解的超定方程。此法如用模拟电路实现,计算过程即可在纳秒量级完成,且计算时间与变量个数无关,因而理论上可用相同时间获得任意的高精度。
In this paper, a generalized circuit equation in moment method is replaced by a super-definiteequation resulted from the extended collocation method which increases matching point numbers, and the resulted super-definite equation is solved by the Hopfield linear programming neural network possessing massive parallism and strong interconnectivity. Once this method is accomplished by an analog circuit, the computation process could stop within a few time constants (i. e. the order of ns). Moreover, the computation time is independent of the variable numbers. As a result, a arbitrarily high accuracy for solutions will be obtained within the same time in theory.
出处
《通信学报》
EI
CSCD
北大核心
1992年第5期80-83,99,共5页
Journal on Communications
关键词
矩量法
神经网络
Moment method, Least-square method, Hopfield neural network, Neural optimization.
