| Summary: | For decades, optimisation research has investigated methods to find optimal solutions to many problems in the fields of scheduling, timetabling and rostering. A family of abstract methods known as metaheuristics have been developed and applied to many of these problems, but their application to specific problems requires problem-specific coding and parameter adjusting to produce the best results for that problem. Such specialisation makes code difficult to adapt to new problem instances or new problems. One methodology that intended to increase the generality of state of the art algorithms is known as hyperheuristics.
Hyperheuristics are algorithms which construct algorithms: using "building block" heuristics, the higher-level algorithm chooses between heuristics to move around the solution space, learning how to use the heuristics to find better solutions. We introduce a new hyperheuristic based upon the well-known ant algorithm metaheuristic, and apply it towards several real-world problems without parameter tuning, producing results that are competitive with other hyperheuristic methods and established bespoke metaheuristic techniques.
|
|---|