【作者】 岳俊红；

【导师】 刘吉臻；

【作者基本信息】 华北电力大学（北京） ， 热能工程， 2008， 博士

【摘要】 预测控制来源于工程实践,具有模型精度要求低、在线计算方便、控制综合效果好的特点,在实践中得到十分成功的应用。随着工业过程控制要求的不断提高,基于单一模型的预测控制器已经难以满足复杂工业过程的控制要求,对多模型预测控制的研究能够拓宽预测控制的应用范围,提高复杂工业过程的控制性能,无疑具有理论意义和实际价值。文章首先总结已有的研究成果,将多模型预测控制分为加权多模型预测控制和切换多模型预测控制,根据控制律求取方式的不同,每一类又可分为基于控制器的和基于模型的两种形式。多模型预测控制的关键不在于预测控制算法本身,而在于如何选择加权或者切换策略,文中总结了已有的切换和加权方式,提出多步预测控制律加权平均是多模型预测控制特有的平滑切换方式。文章主要对以下几个方面进行了研究:在建立多模型集方面,针对一类工业过程,采用可以描述系统动态特性且容易获得的主要参数如主导时间常数、增益和纯滞后时间的极大极小值,通过等模型距离的方法建立系统的多模型集表示,以在线递推贝叶斯概率加权方法组合子模型作为预测模型设计预测控制器,仿真结果验证了此方法的有效性。在控制算法方面,为了减少计算量,对具有较少计算量的预测函数控制进行了研究。针对具有时滞可测扰动的一阶惯性典型工业过程,基于Smith预估补偿思想,考虑扰动通道和控制通道纯滞后时间的相对大小,提出一种改进的具有解析解的预测函数控制算法,并将其推广到了多变量系统;另外,针对基于T-S模糊模型的预测控制存在计算量大的缺点,从预测控制的误差补偿环节入手,提出一种简化的控制算法,相对于多步线性化方法,减少了计算量,相对于单步线性化方法,提高了控制精度。通过仿真验证了上述结论的正确性。在应用研究方面,将模糊增益调度多模型预测控制应用于ALSTOM气化炉控制标准问题,以负荷作为调度变量,选取3个工况点的线性模型为基准模型,仿真结果表明,按照标准问题的测试要求,本方法具有最好的控制性能,证明了多模型预测控制解决复杂工业过程控制问题的有效性。更多还原

【Abstract】 Model predictive control (MPC) coming from engineering practice has many merits such as lower demand for model matching, convenience to calculate on-line and higher control quality, so it was applied successfully in the industrial application. With more and more rigorous demands of control quality in industrial processes, MPC based on one model can’t already satisfy the control requirements of the complex industrial processes. Research on multi-model predictive control(MMPC) is significant in theory and valuable in application because it can not only widen the applying range of MPC but of improve the control quality for complex industrial processes.Firstly, the dissertation summarizes the present researches and classifies MMPC into two groups: weighting MMPC and switching MMPC, witch each one can be sorted as two forms: by controllers and by models. The key of MMPC is not the MPC algorithm itself but how to select the weighting strategies and switching strategies. it is proposed that the weighting average to the multi-step moves is impactful method to switch controllers smoothly. Afterward, the study addresses the following topics:At the aspect of building multi-model bank, a simple effective method is presented for a class of industrial process. The maximum and minimum values of the parameters describing system dynamic behavior such as time-constant, model-gain and dead-time can be acquired from experiential knowledge and testing data, then the multi-model bank was set up by the means of dividing its extreme models composed of the maximum and minimum parameters via equidistance between sub-models. The predictive model is obtained by weighting the sub-models with the recursive Bayesian scheme. Simulation results demonstrate the efficiency of the method.At the aspect of control algorithm, the predictive functional control (PFC) was studied so as to reduce the calculation load of MMPC. Focusing on the first order plus dead time system with measurable disturbance, an improved PFC algorithm was proposed based on the ideal of Smith predictor, which emphasize the difference between control channel dead-time and disturbance channel dead-time. The algorithm was applied further to multiple variable systems. In addition, focusing on the MPC based on T-S fuzzy model, a new error compensating means is stated to simplify the calculation of the control moves. It has less computing load comparing with multi-step linearization method and better control quality comparing with one-step linearization method. They are all proved correct by simulations.In the sixth chapter, fuzzy gain scheduled MMPC was applied to the ALSTOM gasifier benchmark problem, in which the load is selected as scheduled variable and three models lying respectively at three operation conditions are select as sub-models. Simulation results, accordingly to the requirement of the benchmark problem, show that the method has best control performances. The ability to control the complex industrial processes for MMPC was qualified once more.更多还原