CS229: Machine Learning Models for Inverse Power Flow in the Grid

The electricity distribution system is undergoing major changes today with the addition of distributed energy resources (DERs) such as solar and wind, electric vehicles, and energy storage. These new features are challenging traditional control methods, straining grid infrastructure, and adding variability which makes voltage regulation and planning difficult using conventional modeling tools. Without measurements at every node in the system and good estimates of all line parameters, utilities cannot accurately model the flow of electricity to meet this new need. Consequently, new modeling methods have been proposed to address this challenge [1]. The forward power flow mappings represent the physical flow of electricity, giving the real and reactive net power injections, p and q, from the voltage phasor, v = |v|e, at each bus and the line parameters of the system, Gkj and Bkj [2]: