Day-Ahead PV Power Forecasting for Control Applications

Abstract

The percentage of solar energy among all installed energy sources is increasing each year. Photo-voltaic (PV) power forecasting is key to the application of control methods in systems with PV panels. In this paper, we present a method for day-ahead PV power forecasting at each time step that is easy to train and can be applied to different power data types (e.g data from hot and cold climates, with various sampling times). Predictions made before and after sunrise are handled separately. Exponentially Weighted Moving Average (EWMA) is applied on the normalized daily power data to estimate the shape of the next-day power curve for the predictions before sunrise. Then, the multiplier value which would expectedly produce the best forecast when multiplied with the estimated shape is predicted using a time-series approach. After sunrise, the observed power data is leveraged to improve the previous forecasts. The proposed method is shown to perform well on multiple data sets with varying characteristics. Also, the method is compared with some benchmarks algorithms, and the results are presented.