Optimization of a multi-level steam distribution system by mixed integer non-linear programming.

Abstract

The objective of this project is to optimize the SAPREF oil refinery steam distribution in which imbalances between the various levels presently require the venting of steam from the lowest level. The overall steam balance shows that the problem originates from an excess of high·pressure (HP) steam production for too few medium pressure steam users and turbines. We proposed to solve this problem by considering the replacement of selected steam turbines with electrical drives. Given a set of demands of electricity, mechanical power and steam at various pressure levels, the objective is to recommend configuration changes to minimize overall cost. This is not a trivial problem, as steam not passed down through turbines to lower levels can create a shortage there, so a combination of replacements is required. The variables of the problem are both decision variables on every steam turbine and continuous variables, such as flows and enthalpies. These decision variables are integer variables, 0 or 1 for every steam turbine. Depending on whether it is kept on steam use or replaced with an electrical drive, these variables are as follows: E = 0: keep the existing steam turbine E - 1: switch it to an electrical drive. A complete and realistic model of this utility section must be constructed in order to represent the actual distribution accurately. This model will include an objective function to minimize, some equality and inequality constraints, and some cost functions. If we want this model to be accurate, we shall have to deal with nonlinearities to avoid simplifications, and these non-linearities could lead to infeasabilities or sub-optimal solutions. So we are facing a typical MTNLP (Mixed Integer Non-Linear Programming) problem to find optimal configuration changes which will maximize the return on investment, meeting the electrical, mechanical and steam demands of the refinery. In order to solve this difficult optimization problem we shall use the user-friendly package GAMS (General Algebraic Modeling System).