来源：塑料托盘 发布时间：2010年7月28日

【Abstract】 Process optimization is becoming one of the most valuable techniques for system designing, analysis and operating. And the rapid trends of building model based on first principles, optimizing dynamic process, optimizing in the whole plants and on-line accelerate the need of optimizing large-scale systems consist of many equations. Generally, these process optimization problems have small degrees of freedom and relatively sparse structures. To make full use of these characteristics, reduced SQP algorithm, which is based on space decomposition technique, has been developed to solve these kinds of problems. But great efforts and improvements are still going to be done.This dissertation focuses on the investigation, development and implementation of efficient algorithms and techniques for large-scale optimization of process systems. The main contributions include the following aspects:1). As the improved algorithm, a new rule of basis selection is adopted. According to this rule, basis is checked and even adjusted every iteration, thus the stability can be improved greatly. Also an integrated line search of filter method, with both advantages of normal line search method and filter method, is incorporated into the algorithm to obtain better steplength. Numerical results of some benchmark examples and other three large examples with variable dimensions demonstrate that, the proposed algorithm can increase stability as well as reduce the number of iterations and function evaluations. It is quite More effective than standard Sequential Quadratic Programming (SQP) algorithm.2). An extended Reduced Space Sequential Quadratic Programming algorithm based on Limited Memory method is presented to solve problems with relatively larger degrees of freedom. With Limited Memory method used, the largest matrixes of reduced Hessian and cross item aren't saved directly, but are created and used in the process of calculation. Thus, the memory can be reduced greatly. Besides, some special procedures are considered to apply this method to RSQP and improve its efficiency. Numerical results demonstrate that the presented method is More efficient than either standard SQP or the original RSQP in solving large-scale problems with relatively larger degrees of freedom.3). The simulation and optimization strategies of distillation column operations are studied based on first principles and open equations. According to the problems' characters of small degrees of freedom, special structure, strong sparsely and so on, an approach based on reduced SQP algorithm and hybrid derivative method is presented to solve this kind of optimization problems. In the approach, the first order gradient information for the algorithm is obtained by analytical method and preconditioned automatic differentiation.The computing results prove that the RSQP algorithm presented in this paper is More efficient than standard SQP algorithm. With optimal operation achieved at different conditions by this approach, the profits can be significantly improved. Apart from the efficiency and profits, the operation andproduct requirements should also be taken into the consideration. Therefore, an intelligent operation optimization method, with intelligent rules added, is addressed to make the balance between them.4). Simulation and optimization of the large-scale ethylene process consists of depropanizer and debutanizer are researched in this paper. The research is very significative because of the difficulty in dealing with multi-columns and the strong coupling between them. With model numerical disposal, hybrid derivative method and advanced algorithm used, the simulation and optimization are successfully accomplished. Calculation results show that the improved RSQP algorithm is More efficient than the existing popular optimization software SNOPT. All the work has been done is significant and crucial to the future study of the whole ethylene process.5). SQP algorithms can be classified into full-space methods and reduced-space methods. Based on these two classes, the large-scale nonlinear optimization software named UniOptima has been developed in Matlab. In UniOptima there are two nonlinear solvers named RSQP and THSQP. RSQP is designed to solve large-scale problems with large number of equality constraints and relatively smaller degrees of freedom, while THSQP is designed to solve the general nonlinear problems with True Hessian matrix obtained easily. The optimization software has many options for users to select, and is compatible with the original optimization tools in Matlab, which can largely increase the convenience for users.6). Dynamic optimization problems are generally formulated by differential algebraic equations. With optimization methods of dynamical problems reviewed, simultaneous approach of full discretization by collocation method on finite element, with obvious advantages in efficient and accurate, is addressed and analyzed in detail. Simultaneous approach transfers dynamic optimization problems into NLPs, these NLPs often have relatively small degrees of freedom, complicated objective, constraints function, and fixed sparse structure. Therefore, RSQP algorithm can be used to solve this kind of NLP problems. Also, automatic differntaion and symbol automatic differentiation can be introduced to quickly obtain accurate gradient information. In this dissertation, a full solving framework is given to solve these dynamic optimization problems