A unified hyper-heuristic framework for solving bin packing problems

• Our evolutionary model produces hyper-heuristics that are condition-action rules.
• The hyper-heuristics combine up to six different heuristics in the solution process.
• One and two-dimensional (regular and irregular) bin packing problems are considered.
• We present an algorithm to create random bin packing instances with concave shapes.
• We include several analyses about how heuristics are combined solving problems.

One- and two-dimensional packing and cutting problems occur in many commercial contexts, and it is often important to be able to get good-quality solutions quickly. Fairly simple deterministic heuristics are often used for this purpose, but such heuristics typically find excellent solutions for some problems and only mediocre ones for others. Trying several different heuristics on a problem adds to the cost. This paper describes a hyper-heuristic methodology that can generate a fast, deterministic algorithm capable of producing results comparable to that of using the best problem-specific heuristic, and sometimes even better, but without the cost of trying all the heuristics. The generated algorithm handles both one- and two-dimensional problems, including two-dimensional problems that involve irregular concave polygons. The approach is validated using a large set of 1417 such problems, including a new benchmark set of 480 problems that include concave polygons.