Probabilistic Estimation of PV Generation at Customer and Distribution Feeder Levels Using Net-Demand Data

Reliable operation of the electricity grid requires the grid operator to know detailed information about the system at any given time. This information usually comes from readings of different devices, or from different estimation algorithms. However, having accurate estimations become more challenging due to increasing amount of photovoltaic (PV) generation, especially widespread integration of rooftop PV into distribution systems with only the net demand being recorded. This paper proposes a novel probabilistic approach that estimates behind-the-meter gross PV generation using only net-metered demand and local weather data. It also provides a way to measure the uncertainty around the PV estimations. In areas with low PV penetration, a second probabilistic approach is proposed to estimate aggregated PV generation using feeder-level aggregated net-demand data. Our proposed approaches combine a physics-based PV model, machine learning based clustering or classification methods, and Bayesian framework to estimate hourly PV generation at both customer and feeder levels. Both approaches are validated using real demand and generation data, where the results show reliable PV estimations with the estimated 95% confidence band covers 92% of the actual PV generation data. The median PV estimation also reduces the mean squared error of the existing model by 77%.