7. Conclusion
This paper showshow genetic programming canbeused to build scheduling algorithms for the parallel unrelated machines scheduling environment with arbitrary scheduling criteria. The proposed heuristic is composed of two parts: a meta-algorithm and a priority function. The meta-algorithm we propose is defined manually, while the priority function is evolved using GP. This allows the users to specify an arbitrary criterion, and evolve the appropriate priority function for it. The experiments have shown that the proposed algorithm achieved results which were in most cases better than the results achieved by the existing scheduling heuristics. The GP was still unable tofindsolutionsbetter thanthose foundby the search-based methods. However, the goal of this approach is not to provide optimal or near optimal solutions, but to find solutions with acceptable quality in a small amount of time. Additionally, several different GP approaches like dimensionally aware GP, GEP and GP with iterative dispatching rules were tried out. GP with iterative dispatching rules achieved the best results when compared to any of the other GP approaches, but is applicable only in off-line scheduling. Dimensionally aware GP and GEP achieved results which were mostly comparable to the standard GP, but offer some benefits which could make them more appropriate for certain situations.
