An Electrical Model-Free Optimal Power Flow for PV-Rich Low Voltage Distribution Networks

Abstract

The growing amount of residential photovoltaic (PV) systems is pushing distribution companies to adopt different solutions to manage customer voltage issues resulting from changes in net demand. One potential solution is the active management of PV settings, curtailing generation as needed. The conventional AC Optimal Power Flow (OPF) can be used for this purpose. However, OPF-based techniques require detailed three-phase low voltage (LV) network models which are not always available. This paper proposes the optimal calculation of PV settings to mitigate voltage problems in a LV feeder using, instead of power flow equations, a neural network (NN) trained to capture the nonlinear relationships among historical smart meter data (P, Q, V). In other words, an electrical model-free OPF. Results using a realistic Australian LV feeder with 31 single-phase customers are promising as the approach can calculate PV settings with good accuracy, without the need for electrical models.