Resumen
Project Planning and Control (PPC) problems with stochastic job processing times belong to the problem class of Stochastic Resource-Constrained Multi-Project Scheduling Problems (SRCMPSP). A practical example of this problem class is the industrial domain of customer-specific assembly of complex products. PPC approaches have to compensate stochastic influences and achieve high objective fulfillment. This paper presents an efficient simulation-based optimization approach to generate Combined Priority Rules (CPRs) for determining the next job in short-term production control. The objective is to minimize project-specific objectives such as average and standard deviation of project delay or makespan. For this, we generate project-specific CPRs and evaluate the results with the Pareto dominance concept. However, generating CPRs considering stochastic influences is computationally intensive. To tackle this problem, we developed a 2-phase algorithm by first learning the algorithm with deterministic data and by generating promising starting solutions for the more computationally intensive stochastic phase. Since a good deterministic solution does not always lead to a good stochastic solution, we introduced the parameter Initial Copy Rate (ICR) to generate an initial population of copied and randomized individuals. Evaluating this approach, we conducted various computer-based experiments. Compared to Standard Priority Rules (SPRs) used in practice, the approach shows a higher objective fulfilment. The 2-phase algorithm can reduce the computation effort and increases the efficiency of generating CPRs.