Proactive Self-Adaptation of a Flexible Simulation Based Control System Using Forecasting

In recent years, intelligent self-adapting control mechanisms have become a center of research due to their many advantages in complex dynamic real world manufacturing and logistics systems. Their ability to modify control behavior based on detected changes in the environment allows for high flexibility and robustness. While existing self-adapting systems typically only react to changes a posteriori - resulting in a temporary performance drop - we propose a forecast of system parameters in order to maintain constantly high performance. The forecasting is used in context of a simulation based approach where the forecast results are used in a parameterized simulation model of the controlled system. This enables the control to proactively anticipate likely changes in the environment and be prepared in case of their realization. We specifically researched different forecasting methods from business applications for use in self-adapting control. We evaluate our approach in a proposed control system for a warehouse scenario using novel evaluation techniques in a custom prototypical software tool. We identified and adapted suitable forecasting methods and developed meaningful evaluation techniques based not only on parameter forecast, but actual resulting system states. The approach is validated using a prototypical Java implementation and integration into the material flow simulator d'fact.