Day-Ahead Prediction of PV Generation Using Weather Forecast Data: a Case Study in the UK

With the ever-increasing capacity of PV installations, the requirement for accurate and computationally efficient day-ahead forecasts is becoming more evident, while the solution remains a real challenge. Different techniques are being employed to transform a combination of weather forecasts and historical measurements into PV generation predictions. In this paper, the weather forecast data, provided by the UK Met Office, and the historical measurements are used to construct three different prediction models, based on linear least square regression (LSR), artificial neural network (ANN), and fuzzy. All the models can learn from new available data while running the forecasts. The results of an almost one-year study show that the best permutation (for the under-study case) is achieved by averaging the forecasts from LSR and ANN.