针对日益复杂的生产过程及其对控制的超常要求，本论文研究并提出了 几种新型预测控制方法，较好地解决了非线性系统和混合逻辑动态系统中的 难控问题，仿真和实际应用表明本文所提方法的有效性。本文的研究的主要 内容和取得的创新之处体现在以下几个方面： 1．在研究基于尺度函数的小波神经网络的基础上，首次将小波神经网络 应用到复杂青霉素发酵过程的软测量建模与预测控制中，仿真结果和应用结 果表明小波神经网络在生化过程中具有广阔的应用前景。 2．将支持向量机理论引入流程工业预测控制中，针对流程工业提出了基 于支持向量机的非线性动态系统辨识方法和基于支持向量机的非线性预测控 制算法，同时给出了这种非线性预测控制算法稳定性的证明，并在一个复杂 的聚酯生产过程上进行了仿真验证，结果表明所提算法在小样本建模情况下 就能取得良好的控制效果。 3．分析流程工业中存在混杂特性的基础上，研究了混合逻辑动态系统建 模方法，提出了在将逻辑命题转化为混合整数线性不等式时转换方法的选取 准则，并利用混合逻辑动态系统建模方法建立了一个实验装置的MLD模型。 4．针对预测控制中的不可行和优先级问题，利用混合逻辑动态系统的概 念，提出了一种基于输入输出模型的混杂预测控制方法，有效地处理了预测 控制约束不可行与优先级问题。通过将约束优先级表示为命题逻辑并将命题 逻辑转化为整数不等式约束，从而可将约束不可行和优先级问题转化为统一 的预测控制求解问题。在保证高优先级的约束满足的同时能最大化低优先级 约束的满足数目。针对该方法求解过程中遇到的混合整数求解问题，根据流 程工业的连续操作的特点，提出了一种Leas_Low First求解方法来满足控 制实时性的要求。混杂预测控制方法在化工过程中得到了仿真验证，仿真结 果表明该算法使系统控制性能得到了很大的改善。 5．针对一类具有高度非线性和随机特性的对象，提出了变周期单值智能 预测控制方法，引入“过程容量”的概念，对“过程容量”变化率进行估计， 在此基础上对被控变量变化趋势进行预报，并能精确计算出控制作用。基于 专家知识和模糊规则的变周期控制，避免了传统控制算法无论是否得到了正 确的测量信息，都必须盲目地送出控制作用的弊端。该控制策略在常减压装 置上成功地得到了实际应用，并取得了良好的控制效果。; In this dissertation，new nonlinear predictive control apprpoaches are developed for complex process industries and the unordinary control requirements． The new strategies can SUfficiently tackle the difficulties both in nonlinear processes and hybrid control systems．Application and simulation results show that the new nonlinear predictive contr01 approaches are of effectiveness．Main research contributions and achievements in this dissertation are as following： 1．Basing on the research of wavelet neural network using only scale function．the wavelet neural network in developing soft sensors and predictive control of complex penicillin fermentation process iS first successfully implemented．The results of simulation and practical application show that the wavelet neural network’S application in fermentation process is of promising prospect． 2．The support vector machine is introduced into the area ofprocess contro1． A new identification method based on support vector and a new nonlinear predictive control algorithm based on support vector are first developed．Stability for the new predictive control strategy is proved．Simulations are carried out on a complex polymerization process．Even the data forin a sparse distribution in the input space．the excellent nonlinear model can be obtained by the identification method，and the process with desired performance iS shown by using the predictive control strategy． 3．The hybrid property of process industries iS analyzed．Then a modeling framework called MLD(Mixed Logic Dynamic)suitable for modeling the hybrid property in process industries iS studied．A rule for chosing methods when translating propositional logic into inequalities is suggested．Adopting the MLD framework．a model of an experiment process is developed． 4．Using MLD framework，a hybrid predictive contr01 scheme based on input-output model iS presented to handle infeasibility and constraint prioritization issues in MPC．By expressing constraint priority as propositional logic and by translating propositional logic into inequalities，infeasibility and constraint prioritization issues can be unified as the solution of a predictive controller． Satisfaction of constraint on higher priority level is guaranteed together with maximizing the number of satisfied constraints on lower priority level．In order to solve the MIQP problem in the formulation，a Least—Low First strategy based on the continuity property of process industries iS presented to meet the real time requirements．The new method was implemented successfully in the control simulation．Simulation results show that the new method can not only handles infeasibility together with constraint priority,but also improves control performance effectively． 5. A new intelligent predictive control strategy with variable control period and single pr
修改评论