摘要
针对一类多输入多输出非线性被控对象,提出一种基于单神经网络的预测控制算法,应用RBF神经网络对非线性系统进行辨识,并计算被控系统多步预测输出值。该方法通过对传统预测目标函数加以改进,给出一种带微分项的多步预测目标函数,通过迭代寻优实时给出优化控制量。该方法实时性好,简化了传统预测控制算法,加快了滚动寻优的速度,有效地抑制了系统惯性和输入时滞所带来的超调,减小了模型误差、干扰及不确定性对控制器的影响。仿真及应用结果表明了该方法的有效性。
A predictive control algorithm based on single neural network is presented for a kind of MIMO nonlinear systems. RBF network is used to calculate the multi-step-ahead predictive outputs. To improve the traditional objective function, a multi-step differential predictive cost function is constructed and the optimum objective values are calculated by using the iterating algorithm. This strategy can accelerate the process of receding horizon optimization and reduce the influence of model error, disturbance and uncertainty to the controller. Simulations and applications show the effectiveness of the proposed method.
出处
《控制工程》
CSCD
北大核心
2009年第1期39-41,55,共4页
Control Engineering of China
基金
国家自然科学基金资助项目(60575009)
