Mixed integer- and logic-based optimization techniques for scheduling of multistage chemical processes.

摘要

This thesis deals with the development of mixed integer and logic based optimization techniques for the scheduling of multistage continuous and batch plants that are of relevance to the chemical processing industries. The goal is to develop novel mathematical models and efficient solution techniques. Special significance is given to the modeling of inventory and resource constraints in continuous time.; A survey of the main assignment and sequencing models for scheduling problems within the mathematical programming framework is presented. The purpose is to provide a conceptual classification of the assignment and sequencing models for process scheduling. We mainly concentrate on the allocation and timing constraints, which constitute the basis of the mathematical models. The classification identifies two major groups: single-unit assignment models, in which the assignment of tasks to units is known a priori, and multiple-unit assignment models, in which several machines compete for the processing of products.; The problem of determining a cyclic schedule for multistage continuous multiproduct plants is addressed. A large scale mixed integer nonlinear programming (MINLP) model is developed that can handle intermediate storage between stages as well as sequence dependent changeovers. It is first shown that the handling of nondifferentiabilities for intermediate storage are best handled with 0-1 variables.; In the area of batch processes, we propose a model and a solution method for the short term scheduling of multistage batch plants which may contain equipment in parallel, based on a continuous time representation. The objective is to minimize earliness. The problem is restricted to setups that are sequence independent and no resource constraints are considered, since the main goal is to establish the extent to which a continuous time model can be solved for large scale industrial applications. A mixed integer linear programming (MILP) model is proposed that relies on the use of time-matching constraints given parallel time coordinates for the units and the tasks.; Motivated by the success of the application of preordering constraints, an alternate MILP model is presented, in which the preordering is embedded in the underlying representation. Specifically, instead of using preordering as a separate set of constraints, a different representation of the time slots for the units is used which explicitly incorporates the preordering.; Last, we propose a novel approach for modeling discrete resource constraints (e.g., number of operators, fixed amount of electricity) for continuous time MILP scheduling models. A solution method that combines LP-based branch and bound with disjunctive programming is proposed. The basic idea is to first represent the resource constraints as symbolic disjunctions. Furthermore, instead of handling them explicitly in the solution algorithm, only the violated disjunctions are generated within the branch and bound enumeration at nodes with feasible scheduling assignments. At these nodes a new disjunctive tree enumeration is created that switches from a 0-1 representation to a disjunctive program for the resource constraints. (Abstract shortened by UMI.)