A mixed integer optimization strategy for a large scale chemical plant in operation

Abstract

Over the last 10 years much effort has been devoted to the study of optimization strategies for models involving both continuous and integer variables (Mixed-Integer Nonlinear Programming (MINLP) formulations). Several algorithms have been proposed and a large number of research applications have been presented. However, only a few applications have been reported for real large scale plants. This paper presents the structural and operational optimization of a real ethylene plant in operation. A procedure is used involving the Outer-Approximation (OA) method for the MINLP problem and an existing ad-hoc simulator. The optimization algorithm interfaces the simulation code at the nonlinear programming step. The model handles four continuous independent variables, 33 integer variables, more than 200 dependent variables and 48 process constraints. As compared with the best continuous solution, the results show that an increase of 296,337 US$/yr in gross profit can be obtained by simultaneous parameter and structural optimization using a MINLP model. This study also shows that the OA method works very well on real problems reaching convergence in few iterations. This characteristic makes the method a practical tool for large applications, and confirms its potential to be introduced as a standard option in commercial process simulators.