Evolution of scheduling heuristics for the resource constrained scheduling problem

Pronalazak mogućih rasporeda dobre kvalitete postaje velik izazov u projektnom menadžmentu s porastom kompleksnosti, veličine i zahtjeva projekata. Iz tog razloga je problem raspoređivanja s ograničenim sredstvima iscrpno istraživan kroz prošla desetljeća, rezultirajući u velikom rasponu egzaktnih i heurističkih metoda rješavanja. U tom kontekstu se ovaj rad fokusira na primjenu genetskog programiranja za evoluciju heuristika raspoređivanja sposobnih za efikasno savladavanje generičkih projekata u statičkim i dinamičkim uvjetima. Za tu potrebu implementiran je hiper-heuristički model s mogućnošću nadogradnje. Izvršen je niz iscprnih pokusa s ciljem optimizacije parametara evolucije te generiranja što boljih rješenja. Analizirana je efikasnost predloženog postupka uzimajući u obzir postojeće heuristike raspoređivanja i uvjete okruženja.With the growth of projects in complexity, size and resource demands, finding feasible schedules of good quality becomes a very challenging task within project management. Therefore the resource constrained project scheduling problem (RCPSP) has been studied extensively during the last decades yielding a variety of exact and heuristic solving approaches. In this context, the main issue described within this thesis is the application of genetic programming for the evolution of scheduling heuristics capable of efficiently handling generic project instances in both static and dynamic conditions. For this purpose a hyper-heuristic model was devised and implemented with generic extension capabilities. An extensive round of experiments was conducted for parameter optimization and the generation of best possible solutions. The effectiveness of the proposed approach is analysed considering existing scheduling heuristics and environmental conditions