A simulation and genetic algorithm approach to stochastic research constrained project scheduling

Resource constrained project scheduling problems are very difficult to solve to optimality. Because of the computational complexity, scheduling heuristics have been found useful for large deterministic problems. However, these scheduling heuristics have not been applied to problems with stochastic task durations. Heuristics are often combined to try to achieve better performance. When this is done, a search over all possible combinations is generally required. This is again a very computationally intensive task, especially for stochastic problems. We demonstrate how a genetic algorithm can be used to determine the best linear combination of scheduling heuristics. A simulation model is used to evaluate the performance of each combination of the heuristics selected by the genetic algorithm, and this performance information is used by the genetic algorithm to select the next combinations to evaluate. The genetic algorithm and simulation based approach is demonstrated using a multiple resource constrained project scheduling problem with stochastic task durations.