Machine learning assisted Differential Evolution for the Dynamic Resource Constrained Multi-project Scheduling Problem with Static project Schedules

In large modular construction projects, such as shipbuilding, multiple similar projects arrive stochastically. At project arrival, a schedule has to be created, in which future modifications are difficult and/or undesirable. Since all projects use the same set of shared resources, current scheduling decisions influence future scheduling possibilities. To model this problem, we introduce the Dynamic Resource Constrained Multi-project Scheduling Problem with Static project Schedules. To find schedules, both a greedy approach and simulation-based approach with varying scenarios are introduced. Although the simulation-based approach schedules projects proactively, the computing times are long, even for small instances. Therefore, a method is introduced that learns from schedules obtained in the simulation-based method and uses a neural network to estimate the objective function value. It is shown that this method achieves a significant improvement in objective function value over the greedy algorithm, while only requiring a fraction of the computation time of the simulation-based method.