【作者】 盖俊荣；

【导师】 何熊熊； 邹涛；

【作者基本信息】 浙江工业大学 ， 控制理论与控制工程， 2012， 硕士

【摘要】 几十年的工业经验表明，在实际工业中要获得复杂过程的准确模型是非常困难甚至是不可能的，而不准确的模型将严重影响控制效果。工业界的需求以及控制理论和计算机技术的发展，强有力地推动着先进控制技术的发展，模型预测控制便是其中的典型代表。它具有易于建模、鲁棒性强的显著优点，从而有效地抑制了算法对于模型参数变化的灵敏性。然而，虽然模型预测控制的一个优势是对模型的精度要求不高，通常不要求稳态增益与实际情况完全相等，只要趋势一致一般就能产生很好的控制效果，但这仅是一种定性的描述，在输入输出模型描述下很难给出模型失配与系统性能之间的定量结论。因此，怎样能使在辨识出现误差的情况下仍然得到良好的控制效果便成为一个值得深入探讨的课题。本文通过引入闭环反馈以及改善模型结构的方法，来改善模型失配情况下的控制效果。完成的主要工作和取得的成果如下：1.针对模型辨识误差较大时单独的模型预测控制难以取得满意的控制效果的情况，提出了融合闭环反馈的多变量预测控制策略。首先，利用相对增益原理判断多变量系统的耦合程度。其次，针对耦合程度较低的系统，讨论了多输入多输出系统中输入和输出的配对问题，以及控制算法中的PID控制强度问题。最后，采用在模型预测控制算法的基础上引入PID控制进行反馈补偿，融合了闭环反馈的方法，从而提高了算法的鲁棒性。2.将改善后的算法应用到一个水箱液位控制的试验中，包括对象模型的辨识、液位的控制等，验证了算法的有效性。3.研究了模型结构问题对于多变量预测控制的影响。将奇异值分解、条件数和相对增益等知识应用到模型预测控制中，分析了稳态模型的线性相关度，提出了一个判定模型结构临界不稳定的方法。并通过摄动的方法给出改善模型结构的初步方法。更多还原

【Abstract】 Decades of industrial experience shows that in the actual industry to obtain the complexprocess of accurate model is very difficult or even impossible. And the not accurate model willaffect the control effect seriously. The industry demands and the development of control theoryand computer technology, strongly promote the development of the advanced control technology.Model predictive control is one of the typical representative. It is easy to modeling androbustness is a significant advantage. So as to restrain the sensitivity of the algorithm for themodel parameters of the change effectively.However, although one of the advantages of model predictive control is not demanding theprecision of the model is very high. Usually, it does not require the steady-state gain with actualsituation equal completely, as long as the trend is consistent, it will bring good control effect, butthis is only a qualitative description. Under the description of the input and output model, it isdifficult to give the quantitative conclusions between the model mismatch and systemperformance. So, in the condition of existing the identification error, how can you still get thegood control effects will be a worth further discussing. This article through the methods of theclose-loop feedback and improve the model structure, to improve the control effect even in thecondition of existing model error. The main work and results are as follows:1．Aiming at the problem of the unsatisfactory control requirement while the dynamicmatrix control is merely used when the model is serious mismatched in model predictive control.An integration of closed-loop feedback for multivariable predictive control strategy is proposed.Firstly, we use the relative gain principle to judgment the coupling degree of multivariablesystem. Secondly, for the system of low levels coupling degree, the matching problem of theinput and output is discussed in the multi-input multi-output system, as well as the PID controlstrength in the control algorithm. Finally, we combin the dynamic matrix control algorithm with PID control, putting forward the method of close-loop feedback. It improves the robustness ofthe algorithms.2. The improved algorithm is used in a water level control experiment, including modelidentification, liquid level control, verifying the validity of the proposed algorithm.3. The effect of the model structural problems for multivariable model predictive control isconsidered. The singular value decomposition, condition number and relative gain are applied tomodel predictive control, analyzing the linear dependence of model structure，and coming upwith a method of judging model structure critical unstable. Finally, through the perturbationmethod to improve the structure of the model.更多还原