动态输出反馈鲁棒模型预测控制离线算法

An Off-line Approach to Dynamic Output Feedback Robust Model Predictive Control

研究具有多包不确定性和有界噪声系统的动态输出反馈鲁棒模型预测控制(Robust model predictive control,RMPC)的离线方法.先前的在线方法中,在估计状态和估计误差集合已知的情况下,在每一采样时刻通过近似最优算法求解控制器参数.本文采用先前的方法计算离线控制器参数和吸引域.首先,选定一系列估计状态,其中,每个估计状态对应同样一组嵌套的估计误差集合.然后,针对每一估计状态和每一估计误差集合的组合,离线计算唯一的控制器参数和对应的吸引域.这些控制器参数和对应的吸引域存储在表中.如果离线确定的吸引域包含实时的扩展状态,则该离线控制器参数是实时可行的.在线时,根据实时估计状态和选取实时估计误差集合,在表中搜索包含实时扩展状态且优化性能指标最小的吸引域所对应的控制器参数.通过连续搅拌釜式反应器控制系统验证了该方法的有效性.

This paper presents an off-line approach to dynamic output feedback robust model predictive control （RMPC） for a system with both polytopic uncertainty and bounded disturbance. In the previous on-line approach, with the pre- specified estimated state and estimation error set, at each sampling time, a near-optimal optimization algorithm is used to calculate the control parameters. In this paper, the previous approach is invoked to calculate the off-line control parameters and the regions of attraction. First, a sequence of estimated states, each corresponding to the same set of nested estimation error sets, are selected. Then, a unique set of control parameters and the corresponding region of attraction are calculated for each combination of the estimated state and the estimation error set. These control parameters and the corresponding regions of attraction are stored in a table. If an off-line specified region of attraction contains the real-time augmented state, then the corresponding off-line control parameters are real-time feasible. Based on the reM-time estimated state and the selected real-time estimation error set, the real-time control parameters are searched in this table, which correspond to the region of attraction containing the real-time augmented state and having the minimal performance index. A continuously stirred tank reactor control system is utilized to illustrate the effectiveness of the approach.