|
|
Tom van der Beek (Delft University of Technology) - Using a neural network as an objective function for the Resource Constrained Project Scheduling Problem with stochastic new project arrivals Supervisors: Theresia van Essen Recorded full presentation Abstract To increase efficiency and market competitiveness, shipbuilding companies are exploring the options of modular engineering and production to decrease ship production times. This means that, instead of designing and producing each ship as a one-off product, ships are part of a product family. Each ship will be engineered by using a common base-platform and adding modules to maintain variety in the product portfolio. From a scheduling point of view, this can be seen as a stochastic scheduling problem where multiple similar projects arrive and have to be scheduled. This results in a very computationally expensive problem, where traditional methods like sample-average approximation quickly become unusable. Therefore, we create a simulation environment that generates schedules and evaluate the expected future makespan. With this, we train a neural network that will be used in a population-based optimization algorithm, in order to quickly find good schedules. |
