广义多项式神经网络及其结构自适应确定

General Polynomial Neural Network and Its Structure Adaptive Determination

在前向神经网络模型的基础上,构建了一类新型广义多项式神经网络,依据BP神经网络梯度下降法的思想,给出该神经网络的权值迭代修正公式,并在此基础上建立其结构自适应确定算法,并进行仿真实验.结果表明,该神经网络计算精度高、能自适应地确定网络结构.

In this paper, based on special feed-forward neural network, a novel type of general polynomial neural network is constructed. According to the idea of the declining of gradient of BP neural network, the weight iterative modification formula of this neural network is presented. Accordingly, a strueture-adaptive-determination method is proposed and its simulation experiemnts are conducted. Theoretical analysis and simulation results substantiate that such a neural network could be efficient and accurate and determinate its structure adaptively.