论文部分内容阅读
本文提出了一种用数模混合电路实现小规模前馈神经网络的方案,并用电子线路实现了每层具有10个节点的3层前馈网,连接权采用固定电阻实现,节点非线性函数用数字电路实现。该前馈网经BP算法离线训练之后,可以广泛用于各种智能控制。文中给出了权值、阈值以及非线性单元的实现方法,并给出了用该方案实现的神经网络控制器实例及测试方法。
In this paper, we propose a scheme to realize small-scale feedforward neural network by using digital-analog hybrid circuits. The three-layer feedforward network with 10 nodes per layer is realized by using electronic circuits. The connection weights are realized by fixed resistors. The nonlinear functions of nodes Digital circuit implementation. The feedforward network can be widely used in various intelligent control after being trained offline by BP algorithm. In this paper, the weights, thresholds and implementation methods of nonlinear elements are given. The examples and test methods of neural network controller with this scheme are given.