Forecasting the aggregated output of a large fleet of small behind-the-meter solar photovoltaic sites

Abstract Significant growth of behind-the-meter solar Photovoltaic (PV) power generation in recent years is changing the shape of the net demand for electricity from electrical grids. In this work, a framework is proposed to forecast the aggregated power generation of a large fleet of small behind-the-meter solar PV sites. The outputs of those sites are not individually measured and thus, the aggregated output is “invisible” to power system operators. The proposed model uses the available historical power generation data of a very limited number of representative sites in the region, along with Numerical Weather Predictions (NWP) inputs. This way, it is not necessary to constantly monitor all the sites in the region. Fuzzy Arithmetic Wavelet Neural Networks (FAWNN) are used to develop the forecasting engine, providing fuzzy confidence intervals for any desired level, so this methodology can handle various shapes of uncertainties in the input data. The proposed model is validated using actual PV generation data from 6673 sites in California. The simulation results have shown that the proposed approach is capable of forecasting BTM solar PV fleet despite using limited data. The root mean squared error for the forecasts was found to be 3 % for the California region.