Solving zero-one mixed integer programming problems using tabu search

We describe a tabu search (TS) approach for solving general zero-one mixed integer programming (MIP) problems that exploits the extreme point property of zero-one solutions. Specialized choice rules and aspiration criteria are identified for the problems, expressed as functions of integer infeasibility measures and objective function values. The first-level TS mechanisms are then extended with advanced level strategies and learning. We also look at probabilistic measures in this framework, and examine how the learning tool Target Analysis (TA) can be applied to identify better control structures and decision rules. Computational results are reported on a portfolio of multiconstraint knapsack problems. Our approach is designed to solve thoroughly general O/l MIP problems and thus contains no problem domain specific knowledge, yet it obtains solutions for the multiconstraint knapsack problem whose quality rivals, and in some cases surpasses, the best solutions obtained by special purpose methods that have been created to exploit the special structure of these problems.