Key points are not available for this paper at this time.
Solar energy plays a critical part in lowering CO2 emissions and other greenhouse gases when integrated into the grid. Higher solar energy penetration is hindered by its intermittency leading to reliability issues. To forecast solar energy production, this study suggests a three-step forecasting method that selects weather variables with a moderate to strong positive correlation to solar radiation using Pearson correlation coefficient analysis. Low-level data fusion is used to combine weather inputs from a reliable local weather station and an on-site weather station, significantly improving the forecasting model's accuracy regardless of the machine learning method used. Weather data was obtained from the Kisanhub Weather Station located in Cranfield University, UK and the meteorological station in Bedford, UK. In addition, PV power supply data was obtained from four solar plants. Using the Regression Learner app in MATLAB, the proposed architecture is tested on a utility scale solar plant (1 MW), showing a 6% and 13% prediction accuracy improvement when compared to solely using data from the on-site and local weather station respectively. It is further validated using data from three residential rooftop solar systems (8 kW, 10.5 kW and 15 kW), achieving root-mean square values of 0.0984, 0.0885, and 0.1425 respectively. The data was pre-processed using both rescaling and list-wise deletion methods. Training and testing data from the 1 MW solar plant was divided into 75% and 25% respectively, while 100% of the residential rooftop solar plants was used for validation.
Falope et al. (Fri,) studied this question.