基于小波网络的非线性多变量约束预测控制

为了解决非线性多变量系统的建模、控制和优化问题 ,论文扩展基于小波神经网络的单变量系统辨识到多变量系统辨识 ,并用它实现非线性预测控制。对开环稳定过程 ,引入一个具有输入约束的基于小波神经网络模型的区域预测控制方案 ,它的闭环稳定性能够通过适当选择它的预测水平来保证。基于上述动态控制方案 ,提出了一个稳态状态优化方案。通过对一个聚酯生产过程的仿真研究 ,证实了上述方法的有效性。由于能够通过线性最小二乘 (L S)估计方法来辨识 ,该模型易于实现并可用作通用模型。仿真研究的结果表明了该模型的通用性、辨识和控制方法的简单性 ,所提出方案能够被用于过程工业的非线性系统的建模和控制

Non-linear MIMO predictive control based on wavelet networks

Multi variable non linear system identification, control and optimization problems were addressed using a single variable system identification method based on wavelet neural networks. This method was applied to non linear predictive control. A non linear wavelet neural network model based domain predictive control scheme was introduced for open loop stable systems subjected to input constraints. Its closed loop stability was guaranteed by an appropriate choice of a finite prediction horizon. Further more, an economic objective optimization strategy was developed based on the dynamic control scheme. The theory was applied to a polymerization process. Since the model is identified by a linear least square learning algorithm, it is easily implemented and reliable as a generic model. The generality of the model and the simplified identification and control method are demonstrated through simulation studies which show that the strategy can be considered for industrial non linear systems in process industries.