Finding pump values to guarantee a secure and fairly operating Water Distribution Network (WDN) is a daunting or even infeasible task. To tackle this challenge, usually methods based on classic control theory are used. However, some data-driven control strategies such as reinforcement learning were already applied successfully for pump scheduling in recent works, which motivates the investigation of other data-driven control strategies.
This thesis aims to investigate on the potential of blackbox optimization methods, originally designed for Algorithm Configuration/ Hyperparameter Optimization for WDNs. The focus lies on implementing and testing the application of blackbox optimization methods on pump-scheduling.
Literature