A lifelong learning hyper-heuristic method for bin packing.

Area of Science:

• Artificial Intelligence
• Operations Research
• Computer Science

Background:

• Combinatorial optimization problems present significant computational challenges.
• Existing heuristic methods often lack adaptability and continuous learning capabilities.
• Artificial immune systems offer bio-inspired models for adaptive network structures.

Purpose of the Study:

• To develop a novel hyper-heuristic system capable of continuous learning and adaptation.
• To address the limitations of static heuristic approaches in solving combinatorial optimization problems.
• To leverage artificial immune system principles for creating a self-sustaining, plastic problem-solving network.

Main Methods:

• Development of a hyper-heuristic system employing continuous learning.
• Integration of problem instances and heuristics into a self-sustaining network inspired by artificial immune systems.
• Testing on a large corpus of 1D bin-packing problems (3,968 new, 1,370 existing).

Main Results:

• The system demonstrated excellent performance in solution quality across diverse datasets.
• The hyper-heuristic system showed strong adaptability to new and dynamically changing problem instances.
• The network's self-adaptation mechanism led to computational efficiency and scalability.

Conclusions:

• The novel hyper-heuristic system offers a robust and adaptive solution for combinatorial optimization.
• The bio-inspired, self-sustaining network architecture enables efficient generalization and problem-solving.
• The system's continuous learning and adaptation capabilities represent a significant advancement in optimization techniques.