Tabu Search-Based Heuristic Solver for General Integer Linear Programming Problems

This paper presents a tabu search-based heuristic solver for general integer linear programming (ILP) problems as a dependable alternative to branch-and-bound (B&B) solvers. It aims to expand the range of ILP instances for which optimization practitioners can obtain reasonable solutions using either B&B or heuristic solvers. The challenge in developing such a heuristic solver is ensuring both good performance and versatility. The proposed solver addresses this difficulty by incorporating several performance-improving techniques for search behavior and computation speed while maintaining the versatility. These techniques include instance size reduction, neighborhood filtering, and accelerating solution evaluation. The proposed algorithm was implemented in C++17 and made available as an open-source solver that accepts ILP instances formulated for B&B solvers without modification. In numerical experiments with a 120-second time limit, the proposed solver found better feasible solutions than existing open-source B&B solvers for 17 out of the 82 feasible pure integer instances from MIPLIB 2017 Benchmark Set, including instances with over 100,000 variables. An ablation study verified that the incorporated techniques were essential for the performance.