On the forecastability of solar energy generation by rooftop panels pointed in different directions

Abstract

By increasing the penetration of small-scale rooftop solar panels, forecasting their output has become important to both homeowners and distribution systems operators. In many areas, the roof of residential houses is not such that all solar panels are installed pointing in one direction; so, they are installed pointing in different directions. In this letter, the effect of this phenomenon on the forecastability of the day-ahead solar panels' power output is experimentally investigated. To perform day-ahead energy forecasting, a feedforward artificial neural network (ANN) is created using historical data and weather conditions of a similar day along with the forecast weather conditions of the day for which the forecast is to be performed. A similar day selection algorithm based on Euclidean distance is used to determine the reference day. Two forecasting approaches have been compared: forecasting each panel output and forecasting the total output. Moreover, Long short-term memory (LSTM) is used to validate the conclusion made by the feedforward ANN. The results evidently show that considering different directions of the solar panels increases the forecastability of the rooftop solar power plant.