Recognition Model for Solar Radiation Time Series based on Random Forest with Feature Selection Approach

Abstract

Solar photovoltaic (PV) energy, one of fast growing energy technologies in the world, provides cheap opportunities for future energy generation. Solar radiation estimation models play an important role in efficient utilization of solar PV energy. In this paper, we propose a high performance recognition model for solar radiation based on random forest (RF) with feature selection approach to cope with nonlinear dynamics in time series. Forty-five features have been extracted from the temperature, pressure, wind speed and humidity data as well as solar radiation data using the moving average indicator. In addition, the total of 50 features has been extracted together with the five features obtained based on the previous values of the five parameters specified. The features that will improve model performance have been determined by the forward selection approach and the model, which has the highest performance, has been proposed for solar radiation estimation.