多变量系统基于回归神经网络的预测控制

以罐式搅拌反应器为例 ,研究多变量系统基于神经网络的预测控制及改善控制性能的方法 .针对复杂多变量系统难以建模的问题 ,采用多层局部回归神经网络离线建立其预测模型 .在反馈校正中 ,考虑到控制准确性和实时性的要求 ,采用偏差补偿和模型修正相结合的方式修正神经网络的预测输出 .实验中 ,研究了改善控制性能的方法 ,得出 :对性能指标中的偏差项负指数加权 ,可大大加快系统的动态响应过程 ,并在一定程度上减少系统超调 .仿真结果表明控制算法有效 .

Predictive Control of Multivariable System Based on Local Recurrent Neural Networks

Taking the stirred tank reactor as example,the predictive control of MIMO system based on multilayer local recurrent neural networks is presented. The methods of improving the control performance are also discussed. Aiming at the difficulties in modeling the complex MIMO system, the multilayer local recurrent neural network is used to build the predictive model of the process off-line. In feedback correction, considering the requirements of the accuracy and practicability, error compensation and model correction are adopted to correct the predictive output of the model. In simulation, the methods to improve the control performance are discussed.It is concluded that negative exponential weighting of future tracking errors can accelerate the dynamical respondance and lower the overshoot of the control system.The results show the effectiveness of the control algorithm.