Decentralized Optimal Power Flow for time-varying network topologies using machine learning

Abstract

Distributed Energy Resources (DERs) have been continuously increasing over time. When DERs are uncontrolled, technical challenges arise such as over-voltage and congestion. Therefore, Distribution System Operators (DSO) would benefit from controlling PV inverters and batteries. However, the communication infrastructure required to control DERs in a centralized manner by solving Optimal Power Flow in real time is not developed and is intractable. Furthermore, as new DERs are deployed, a new question arises on how to manage these agents given the time-varying environment (new nodes and branches in the graph). This work addresses this gap by proposing a machine learning framework to design controllers that are robust to topological changes for each new inverter-connected DER. We show that the “Learned controller” strategy always enables a more economical operation in comparison to using utilities’ standards. Results show savings ranging from 113% up to 487%. The strengths of the controller we propose are its interpretability, ease of implementation and decentralized nature. From these results we propose guidelines for the DSO to design controllers for new DERs deployed in time-varying networks.