论文部分内容阅读
本文提出了一种基于神经网络的显式自校正控制方案:用二级神经网络NNR(神经网络调节器)和NNI(神经网络辨识器)构成自校正控制器.NNR与受控对象构成一种反馈控制结构.NNI建立受控对象的模型,并预报对象的输出,以此为基础修改NNR的参数. 调节器NNR采用面向实时控制的,具有传统的PID控制机理的二层神经网络.其参数采用随机逼近法或梯度法实时修改.NNI采用单隐层神经网络,对受控对象进行实时辨识,其辨识算法为递推预报误差法(RPE).该算法克服了BP算法收敛速度慢、鲁棒性差的缺点,具有收敛速度快、预报精度高等优点.本文将RPE算法引入自校正控制,并对相关问题作了深入的研究.运用本文提出的控制方案,对一典型的工业过程进行了数字仿真研究.结果证实了本文所提出的基于神经网络的显式自校正控制在克服对象参数时变及较大的随机干扰方面的有效性.
In this paper, an explicit self-tuning control scheme based on neural network is proposed: a self-tuning controller is constructed by two-level neural network NNR (neural network regulator) and NNI (neural network identifier) Feedback control structure.NNI establishes a model of the controlled object and predicts the output of the object and modifies the parameters of the NNR.Regulator NNR adopts a two-layer neural network with the traditional PID control mechanism for real-time control.The parameters Which is modified in real time by using random approximation method or gradient method.NII uses single hidden layer neural network to recognize the controlled object in real time and the recognition algorithm is Recursive Prediction Error Method (RPE) The advantages of poor convergence, fast convergence and high forecast accuracy.In this paper, the RPE algorithm is introduced into the self-tuning control and the related problems are deeply studied.Using the control scheme proposed in this paper, a typical industrial process Simulation results show that the proposed self-tuning neural network-based explicit self-tuning control is effective in overcoming the time-varying and random disturbances of object parameters.