NARMA-L2模型的改进及其神经网络自校正控制器

Modified NARMA-L2 model and its neural network implicit self-tuning controller

带预测误差补偿的NARMA-L2模型是由NARMA模型在零工作点处由一阶泰勒展开逼近的,其误差项取值较大。通过分析NARMA-L2模型存在误差项值较大的问题,利用自适应滤波动态工作点处由一阶泰勒展开逼近NARMA模型,构建改进的NARMA-L2模型,采用BP神经网络辨识改进NARMA-L2模型的参数,基于广义目标函数与改进的NARMA-L2模型给出了非线性系统的隐式自校正控制器算法,以直接极小化指标函数的自适应优化算法寻优BP神经网络的连接权重值,获得了一种新的在线学习算法。研究表明,改进模型误差值较传统NARMA-L2模型小,控制算法使系统具有优良的控制效果。

This paper proposes a modified NARMA-L2 model as a viable alternative to the current NARMA-L2 model with prediction error compensation which suffers from larger value of error term due to the approximation by NARMA model by first-order Taylor expansion at zero working point.The modified NARMA-L2 model is enabled by the following steps:analyzing the occurrence of the problem of large error term in NARMA-L2 model;developing a modified NARMA-L2 model using the NARMA model approximated by its first-order Taylor expansion at adaptive filtering dynamic working point;identifying the parameters of the improved NARMA-L2 model using BP neural network;obtaining the modified NARMA-L2 model,an implicit self-tuning controller algorithm for nonlinear system,based on the generalized objective function and the improved NARMA-L2 model;optimizing connection weight values of BP neural network using adaptive optimization algorithm for direct minimization of index objective function;and thereby developing a novel online learning algorithm.The results show that the modified NARMA-L2 model exhibits smaller error term value than old NARMA-L2 model and the controller algorithm gives the system an excellent controlling performance.