[2J5-OS-36a] OS-36

4:15 PM - 4:30 PM JST(7:15 AM - 7:30 AM UTC)

[2J5-OS-36a-04] A Metaheuristic Solver with Robust Performance against Requirement Complexity through Minimal Domain Knowledge Injection and Flexible Formulation

〇TOMOMITSU MOTOHASHI^1,2, Koutarou Tanahashi¹, Shunji Umetani¹, Hyakka Nakada¹ (1. Recruit Co., Ltd., 2. SUSMED, Inc.)

Keywords:

mathematical optimization,metaheuristics,optimization solver,Rust

This study develops a metaheuristic solver robust against requirement complexity in real-world optimization. The solver achieves robustness through two mechanisms: minimal domain knowledge injection into decision variables, enabling additional requirements to be reflected during evaluation without modifying neighborhood operations, and flexible formulation via if-statements, reduce functions, and intermediate variables, which eliminates auxiliary variables and absorbs requirement additions through changes to formulas alone. Experiments on CVRP, JSP, and their extended variants with realistic requirements, compared against PuLP and OR-Tools, confirmed that existing solvers suffer significant performance degradation upon requirement additions, while the proposed solver maintains stable solution quality.