适用于降阶模型的新型多步预测控制算法

A New Long-Range Predictive Control Algorithm Based on a Reduced Model

中文关键词  预测控制  自适应控制  降阶模型  鲁棒性

英文关键词  predictive control  adaptive control  reduced model  robustness

中文摘要

      基于参数自校正机理在线修正预测模型的广义预测控制（GPC）[1]具有自校正控制器的共同问题—对降阶模型存在鲁棒性问题。针对这一问题，本文在文[2~3]的基础上，用频域建模和时域控制相结合的方法，提出了一种对降阶模型鲁棒的新型多步预测控制算法，并得到该算法的稳定裕度的定量结果，再用主要频率特性拟合的方法得到对象的降阶模型，以保证建模误差在控制算法的稳定裕度之内。二者结合便形成了本文的算法。理论分析及仿真均表明，该算法对降阶模型时鲁棒的。

英文摘要

      Aim at the problem that GPC has a poor robustness to the reduced order model, we propose a new long-range predictive control algorithm (NLRPC) based on a new idea that is to combine the modelling in frequency domain with the controller design in time domain in this paper. The main design idea is as follows: by introducing a prefilter in the identification part we have a low order model that is reliable in the selected frequency interval, while it is uncertain in other frequency ranges. The uncertainty could then be fed into the robustness boundary of the weighted long-range predictive controller, and obtain the robust adaptive NLRPC. Both theoretical analysis and simulation show that NLRPC has a good robustness to the reduced model.