具有动态补偿能力的神经网络模型及其在动态系统建模中的应用

A Neural Network Model with Dynamic Compensation and its Applications in Dynamic System Modelling

本文在多层前馈神经网络模型基础上，引入误差动态反馈环节，从而形成一种新的具有动态补偿能力的神经网络模型．新模型的训练利用反向传播原理实现．采用该模型对非线性动态系统进行建模时，能显著提高建模精度，特别是在网络模型工作时，对新出现的输出误差具有动态补偿能力．文中给出了新网络模型的结构和学习算法，最后是仿真实例．

By introducing a dynamic error feedback link in a multi--layer feedforward neural network,this paper proposes a new neural network model which has the dynamically compensating capability. During both training and working of this new network model,we apply the principle of dynamic error backpropagation to make the feedback compensation. Using this model in nonlinear dynamic systems modelling,the dynamic error can be effectively reduced and the modelling accuracy can be significantly raised.The structure and learning algorithm of this new neural network model are given.The application to realdata modelling is included to demonstrate the effectiveness of the new model.