BP网络的权值诱导与层次训练算法

Weight Value Induction of BP Network and Corresponding Hierarchical Training Algorithm

BP神经网络以其结构简单、工作状态稳定等特点,成为当前应用最广的一种神经网络模型,应用范围包括识别分类、非线性映射、复杂系统仿真等方面。但是,BP网络也存在局部极小、收敛速度慢、网络结构选取困难等缺点。对此。

According to the difficulties of the physical realization of BP network,this paper presents a valuable weight induction training algorithm to induce and adapt the network weight values to the required range of electronical circuits and parts. As an example ,a BP network with two hidden layers on 3 X 3 TC recognition problem is trained with this algorithm, and a brief comment is presented.