Synthesis Of Complex Distillation Systems Based On Stochastic Optimization

The purpose of distillation systems synthesis, which is of an optimization problem, is to find the best solution among a great number of feasible alternatives in terms of economic competence. Mathematically, it can be formulated as Mixed Integer Nonlinear Programming (MINLP) problem that includes both structural and parametric variables optimization. When complex column configurations and heat integrations are considered, that is often the case for industrial applications, this problem would become extremely complicated due to its large-scale combinatorial feature and non-linearity, and its solution is very difficult. The work presented in this thesis aims at creating a systematic and stable method based on stochastic optimization strategy to tackle this problem.(1) An improved simulated annealing algorithm that can deal with both discrete and continuous variables simultaneously within a same annealing scheme is developed. This approach is robust and suitable for large-scale and complex process synthesis in the way that it can make direct use of the tools available for distillation process design and evolution, to evaluate the criterion value. As a result, the complexities of formulation and handling of an explicit MINLP model can be avoided. This leads to an easier implementation of the approach and a high accuracy.(2) For implementing the proposed approach, an encoding procedure for representing and manipulating the flowsheet structure of the process is proposed. A set of integer numbers series are defined to represent the distillation network structure, which may contain different separation tasks with sharp or non-sharp splits, simple columns, complex columns with side rectifier and/or side stripper and thermally coupled columns of various types, with or without heat integrations. A binary tree approach from Date Structure Theory is introduced to construct a sorting procedure, by which the interconnections between separation tasks and streams can be identified directly based. The proposed encoding procedure can guarantee the feasibility of any structural modification, enabling the application of conventionally evolutionary procedure for separation sequence transformation, and the use of more rigorous cost model in the optimization.(3) The paper deals with for the first time the synthesis problems that allow coexistence of simple columns, prefractionator columns, complex columns with thermally coupled side-stripper and side-rectifier, fully thermal coupling columns, as well as the heat integrations between these columns. The overall structure of the system is derived from three parts: a separation task networks, the interconnections between tasks (thermal coupling with or without a condenser or a reboilor) and a heat integration map. With this complex system structure represented and manipulated by the codes, a conceptual MINLP model for stochastic optimization is developed, and then can be handle with the proposed approach. With so developed solution procedure, it is possible to give designs for various sub-cases, including conventional column sequence, various thermally coupled distillation configurations with different column sections, fully thermally coupled distillation configurations, and of course with heat integrations in all the cases.(4) Examples of different separation problems are solved and studied to illustrate the effectiveness of the developed approach. These problems could be seen as large-scale as the real ones as the numbers of products more than 10 (N≥10) are solved. The economic performances of different results of the examples are analyzed and compared. It can be concluded that the solution procedure developed in this paper, compared with the available methods previously published, can solve more complex problems, can be implemented easily with less computational cost, and therefore is a promising alternative for the addressed problem solution.