基于交叉算法的神经网络模拟电路特性分析

Characteristic Analysis in Analog Circuit Based on Neural Network with Intercross Arithmetic

本文利用基于交叉算法的神经网络训练方法对模拟电路进行性能分析。前馈神经网络的监督学习通常是一种从上到下(top-down)的学习模式,具有单隐层结构的前馈神经网络也可采用从下到上(bottom-up)学习模式的非监督学习算法来进行,基于交叉算法的复值神经网络训练方法突破以往算法的各种局限,其学习过程将从下到上的非监督学习和从上到下的监督学习相结合,网络性能更优。模拟电路特性分析的仿真研究表明该算法行之有效。

The characteristic analysis in analog circuit is realized using neural network method with intercross arithmetic here. The supervised learning arithmetic in forward feedback neural network is always a top-down mode; the forward feedback neural network with single hide layer structure can also be carried out by non-supervised learning arithmetic ; it is a bottom-up mode. The training method of the neural network based on intercross arithmetic is presented in this paper. It is a learning process that combines non-supervised learning with bottom-up mode and the supervised learning with top-down mode together effectively ; it breaks through all kinds of limits of classical arithmetic. A characteristic analysis illustration in analog circuit validates the learning arithmetic.