Research On High-Performance Optimization For Large-Scale Complex Process Systems

Optimization is a significant technology for process industries to save energy and reduce emissions. From the perspective of process systems engineering, optimization for process industries is divided into two aspects, optimal process design and optimal process operation. Nowadays, the researches on optimal process design are focused on process synthesis, including heat exchanger network synthesis, separation sequence synthesis and reactor network synthesis. These synthesis problems can commonly be described as mixed integer nonlinear programming problems. The optimization algorithm for nonlinear programming is a core component for solving mixed integer nonlinear programming problems, so its robustness is a prerequisite for successfully solving the mixed integer nonlinear programming problems. If a nonlinear programming problem cannot be solved successfully, the whole procedure cannot go on. On the optimal process operation side, real-time process optimization system is applied presently. Data reconciliation problem, parameter estimation problem and whole operation optimization problem in real-time process optimization system can be depicted as nonlinear programming. The robustness and efficiency of optimization algorithm for solving nonlinear programming is the guarantee of the normal running and executing of the real-time process optimization system. If the optimization algorithm cannot converge to the optimum in time and the optimization results are still implemented, the operation optimization will make no sense. All aforementioned optimization are based on steady-state model. However, in process systems, inevitably, there are dynamic processes which cannot be described by steady-state model. Integrated optimization and control technology are needed to optimize and control these kinds of dynamic processes. Simultaneous approach is the leading way for solving dynamic optimization problem at present. The basic idea is to discretize the original dynamic optimization problem into nonlinear programming problem and then solve it with nonlinear programming solver, so the optimization algorithm for nonlinear programming is an important calculation engine of the simultaneous approach. Meanwhile, the quality of discretization has a significant impact on the accuracy and optimality of the solution of dynamic optimization. Process simulation software is an implementation and application platform for optimization theory and methods. Absolutely, it pushes the development of optimization. Some problems existing in high performance optimization theory and methods and corresponding solutions are discussed in this dissertation. This dissertation makes progresses in the following aspects:1. Convergence depth control for interior point methods is proposed for handling the Pareto phenomenon which appears in solving high nonlinear, multi-scale and large-scale process optimization problems with interior point method. The corresponding properties of convergence depth control for interior point methods are analyzed and proved.2. The leading optimization algorithms still cannot solve many hard problems. Their efficiency also needs to be further improved. The relationship between parameter setting and algorithm performance is analyzed. A combined metaheuristics and direct search method is proposed for the parameter auto tuning problem. On the basis of further analysis of the parameter auto tuning problem, random sampling based parameter tuning algorithm is proposed.3. There are still some problems in simultaneous approach, which need to be further researched. For the determination of number of finite elements and location of finite element nodes under the condition that accuracy of solution is guaranteed and determination of breakpoints in discontinous optimal control trajectory and the guarantee of the optimality of the solution, an adaptive moving finite element procedure is proposed.4. Interior point method based optimization solver is incorporated into Aspen Plus through Aspen Open Solvers interfaces. The first interior point method solver is developed in Aspen Plus. On the basis of Aspen Open Solvers interfaces, compatibility of Aspen Plus is extended for CAPE-OPEN compliant solver. The dissertation is concluded with a summary and the prospect of future work in high performance optimization technology.