神经网络非线性多步预测逆控制方法研究*

提出了基于多步预测控制方法的多变量非线性神经网络逆控制方案。利用预测模型对系统动态特性进行预测,使用一个带有时延因子的前馈神经网络作为控制器,利用多步预测性能指标对其在线训练,实现神经网络逆系统;在多步预测过程中还对每一步的预测误差进行预测,以实现预测误差补偿。将所提出的控制算法用于锅炉这种大滞后非线性对象的控制,仿真实验证明,该控制策略具有良好的解耦和动态跟踪性能。

Multi-step predictive inverse control based on neural networks

This paper developed a neural-network inverse control scheme for nonlinear MIMO discrete systems based on multi-step prediction. Used a predictive model for predicting dynamic characteristics of the system. A neural network with time-delay modules acted as the inverse controller, which was trained with predictive optimization algorithm to implement the inverse system. Used another neural network to model the prediction error to compensate for the predictive output. Simulation results illustrate that the control strategy has a favorable decouple and dynamic tracking performance to nonlinear systems.