An effective genetic algorithm approach to large scale mixed integer programming problems

Abstract

To effectively reduce the search space of GAs on large-scale MIP problems, this paper proposed a new variable grouping method based on structure properties of a problem. Taking the capacity expansion and technology selection problem as a typical example, this method groups problem’s decision variables over time period and machine line. Based on this new variable grouping method, we developed a variable-grouping based genetic algorithm according to problem’s structure properties (VGGA-S). We tested the performance of VGGA-S by applying it on the capacity expansion and technology selection problem. Numerical experiments suggested that, VGGA-S outperforms the standard GA and variable-grouping based GAs without considering problem’s structure properties, both on computation time and solution quality. Although VGGA-S is proposed based on structure properties of a specific MIP problem, it is a general optimization algorithm and theoretically applicable to other large scale MIP problems.