【作者】 王志国；

【导师】 刘飞；

【作者基本信息】 江南大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 控制性能评估技术(CPA)兴起于上世纪80年代末,是工业过程自动化领域一项重要的辅助管理技术。控制性能评估的主要目标是在尽量不影响当前控制系统正常运行的情况下,对控制器的性能做出评价,得出其与理论最优的基准控制器的性能差距。进而,在控制性能不满足设计要求时,对造成性能不良的原因进行诊断并给出改进决策。控制性能评估技术在维护自动化系统安全高效运行中发挥了积极作用,其主要应用于石油化工、水处理、发酵及制药等工业过程领域。控制性能评估技术在理论研究和工程应用中都已取得许多成果。但作为一门综合技术,现有研究还远未覆盖控制性能评估技术的全部阶段,即使作为研究最多的性能基准选择和评估方面,分析对象也主要限于线性系统,非线性系统性能评估因其复杂性致使研究成果甚少。针对此问题,在最小方差性能基准框架下,本文结合当前控制性能评估技术的发展现状,分别在以下几个方面展开研究：(1)提出一种切换结构的最小方差控制性能评估策略。考虑到许多实际工业过程运行初期表现为线性系统,但随时间推移可能因外部因素影响使其出现非线性特征。传统性能评估技术实施中一般忽略该类变化而直接使用线性评估方法,评估结果存在偏差。对一类特定结构的非线性系统,在进行控制性能评估之前首先使用基于高阶统计量的方法进行非线性检测分析,如检测结果为线性,采用线性评估方法,否则,采用非线性评估方法。通过使用多项式级数近似非线性环节,将最小方差控制性能下限的估计问题转换为最小二乘辨识问题,在系统辨识的同时实现对非线性系统的控制性能评估。(2)将方差分析技术用于非线性系统的控制性能评估,首先提出一类特定结构的SISO非线性系统的最小方差控制性能评估方法,进而扩展到非线性前馈/反馈控制系统。针对一类结构为非线性过程模型叠加线性(部分非线性)干扰模型的SISO非线性系统,分析了其最小方差控制性能下限的表达形式,并指出此性能下限和非线性系统的b个超前干扰驱动源直接相关。通过正交最小二乘辨识方法获得系统闭环模型,在此基础上由方差分解公式计算得到非线性系统的最小方差控制性能指标。进一步,文章证明了两种不同结构的非线性前馈/反馈控制系统最小方差性能下限的存在性,分析了方差分析技术用于其控制性能评估研究的可行性,并结合迭代正交最小二乘辨识方法实现对非线性前馈/反馈控制系统进行性能评估。最后通过一研究实例验证了该方法的有效性。(3)针对现有非线性系统广义最小方差(NGMV)控制性能评估方法的不足,提出一种改进的性能评估指标。最小方差控制器因其输出过于激烈以及鲁棒性差而难以实际运用,广义最小方差控制器通过误差权和控制权参数的引入可对控制量幅值进行限制。现有对NGMV控制系统进行性能评估直接使用线性方法,其理论前提是NGMV控制器作用下的系统闭环输出可退化为一个线性移动平均过程。但作为动态系统,此理论前提很容易受模型变化或外部干扰等因素影响而被破坏,导致评估结果存在误差或完全不具有指导意义。文章在方差分析技术的基础上,分析广义最小方差控制作用下的非线性系统闭环输出方差的组成结构,提出一种基于方差分解的NGMV性能评估方法,仿真研究证明了其有效性。更多还原

【Abstract】 Control performance assessment (CPA) began to blossom in the late1980s, and it is an important asset-management technology in the field of industrial process automation. The main objective of CPA is to evaluate the performance of controller without disturbing the normal operation of the control systems, and find the lack with respect to the theoretical best optimized contoller. When the performance of a running controller deviates from the desired aim, we must determinate the reasons of bringing poor performance and point out what should be done to improve the controller. The technology of CPA plays a significant part for maintenance industrial automatic systems with safety and high efficiency work. The application of CPA has been registered in the industrial process fields of petrochemicals, water treatment, ferment engineering, pharmacetical engineering, and so on.There are many researches and industrial applications in control performance assessment. However, because CPA is a synthetical technology, the existing studies are far from containing all the stages. Even in the first step of CPA study that central task is to select performance benchmarks and estimate the performance of running controllers, the most existing methods work on the assumption that the control systems are linear processes. The papers for nonlinear systems in performance assessment have been infrequently written due to its complexity. For the above mentioned problems, this thesis will focus on the following research topics within the framework of minimum variance benchmark.(1) A switching strategy of control performance assessment based on minimum variance is proposed. Many practical industrial processes can be seen as linear systems at the beginning of running, but with the passage of time, the control loops possibly include nonlinearites from external influence. The estimates of most traditional CPA methods will bring mistake because of failing to notice the nonlinearites. The nonlinearity detection tests based on higher order statistics are used to quantify the nonlinear property presented in the output signals before selecting performance assessment methods. If nonlinearity index(NLI) shows that the control systems are linear, then traditional linear performance assessment methods are adopted, otherwise the new nonlinear method is used. Farther, the problem of nonlinear minimum variance performance assessment is converted into model identification with the help of Volterra series. Finally, a simulation example indicates that the proposed algorithm gives better minimum variance performance bound (MVPLB) compared with existing methods for linear system performance assessment.(2) The technology of variance analysis is introduced to the study of control performance assessment for nonlinear systems. First,.a performance estimated method for a specific class of SISO nonlinear systems is proposed. Subsequently, the similar method is expanded into the performance assessment of nonlinear feedforward/feedback control systems. For a class of nonlinear SISO processes that can be described by the superposition of a nonlinear dynamic model and additive linear or partially nonlinear disturbance, the expression of the minimum variance performance lower bound is derived and the conclusion that the MVPLB only depends on the driving forces of most recent ahead disturbances is obtained. Then, the model of closed loop system is identified by using orthogonal least square algorithm. Moreover, based on the achieved model, the minimum variance performance index of nonlinear systems is calculated by using variance decomposition formula. In addition, the existence of MVPLB for two nonlinear feedforward/feedback control systems with different structure is proved in this thesis. And, whether it is appropriate that the variance analysis technology can be used for evaluating the performance of nonlinear feedforward/feedback control systems is analysed. Then, the problem of performance assessment of nonlinear feedforward/feedback control systems is solved with the help of iterative orthogonal least squares identification method. Finally, a simulation example indicates that the algorithm proposed in this paper gives better effect compared with the existing methods.(3) To the shortage of current performance assessment methods for nonlinear systems based on generalized minimum variance (GMV) benchmark, a novel performance index is explored and exploited. The actual application of minimum variance controller is difficult because of the violence of control signal changes and the worst robust performance. Due to the introducing of error and control weighting terms, the magnitude of GMV controller can be restricted. For the present study of estimating the performance of nonlinear GMV control systems, the traditional linear methods are directly adopted. These linear methods are effective because of the output signals of nonlinear control systems in effect of theoretical nonlinear GMV controllers can be described by linear moving average processes. However, this precondition may be easily distorted by model changes and external disturbances in practice, so the results of linear performance assessment methods are inaccurate or at all meaningless. On the basis of variance analysis technology, the configuration of variance of closed-loop output signals is analysed, and a nonlinear GMV control performance assessment method of nonlinear systems is proposed. Finally, a simulation example indicates that the algorithm is effective.更多还原