Clustering Data-driven Local Control Schemes in Active Distribution Grids

Abstract

Controllable Distributed Energy Resources (DERs) in Active Distribution Grids (ADGs) provide operational flexibility to system operators thereby offering the means to address various challenges. Existing local controllers for these resources are communication-free, robust, and cheap, but with sub-optimal performance compared to centralized approaches that heavily rely on monitoring and communication. Data-driven local controls can bridge the gap by providing customized local controllers designed from historical data, off-line optimization, and machine learning methods. These local controllers emulate the optimal behavior, under expected operating conditions, without the use of communication. However, they exhibit high implementation overhead with the need of individual programming of DER controllers, especially when there are many DERs or when new units are installed at a later stage. In this paper, we propose a clustering method to decrease the implementation overhead by reducing the individual DER controls into a smaller set while still achieving high performance. We show the performance of the method on a three-phase, unbalanced, low-voltage, distribution network.