Adaptive Memory Projection Methods for Integer Programming

Abstract

Projection methods, which hold selected variables fixed while manipulating others, have a particularly useful role in metaheuristic procedures, especially in connection with large scale optimization and parallelization approaches. This role is enriched by adaptive memory processes of tabu search, which provide a collection of easily stated strategies to uncover improved solutions during the course of the search. Within the context of pure and mixed integer programming, we show that intensification and diversification processes for adaptive memory projection can be supported in several ways, including the introduction of pseudo-cut inequalities that additionally focus the search. We describe how the resulting procedures can be embedded in constructive multistart methods as well as in progressive improvement methods, and how they can benefit by the application of target analysis.