كنفرانس بين المللی مهندسی صنايع و سيستم ها

صفحه اصلی / دوازدهمین كنفرانس بين المللی مهندسی صنايع و سيستم ها

An Event-Driven Generation-Selection Hyper-Heuristic with Large Language Models for Online Bin Packing

نویسندگان :

Amirhosein Tabatabaei¹ Farid Khoshalhan²

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه صنعتی خواجه نصیرالدین طوسی

کلمات کلیدی :

Online Bin Packing Problem،Large Language Models،Hyper-Heuristics،Multi-Criteria Beam Search،Heuristic Generation،Combinatorial Optimization

چکیده :

Large Language Model-driven hyper-heuristics discover powerful heuristics for combinatorial optimization problems such as the Online Bin Packing Problem, but typically require large numbers of queries, rendering them impractical for real-time or budget-constrained deployment. We propose an LLM-driven generation-selection hyper-heuristic that couples diversity-aware beam search with event-triggered LLM evolution. A dynamic behavioral diversity metric and lexicographic selection maintain a robust heuristic portfolio during online processing. Rather than querying the LLM continuously, an event monitor tracks performance degradation, diversity collapse, and distribution shifts, invoking severity-matched LLM evolution and self-healing code repair only when significant anomalies occur. Across eight benchmarks our framework matches or outperforms state-of-the-art classical and LLM-based baselines while using fewer LLM API calls per run. These results demonstrate that selective, feedback-guided LLM intervention achieves top-tier online packing quality at a fraction of the computational cost.