Enhanced accuracy estimation model energy import in on-grid rooftop solar photovoltaic

Abstract

Installing rooftop solar photovoltaic (PV) with an on-grid system benefits consumers because it can reduce imports of electrical energy from the grid. This study aims to model the estimation of energy imports generated from on-grid rooftop solar PV systems. This estimation model was carried out in 20 provincial capitals in Indonesia. The parameters used are weather conditions, orientation angle, and energy generated from the on-grid rooftop solar PV system. The value of imported energy is divided into three combinations based on the azimuth angle direction, which describes the type and shape of the roof of the building (one-direction, two-directions, and three-directions). Modeling was done using machine learning with neural network (NN), linear regression, and support vector machine. A comparison of the machine learning algorithm results is NN produces the smallest root mean square error (RMSE) value of the three. Model enhancement uses a grid search cross-validation (GSCV) to become the GSCV-NN model. The RMSE results were enhanced from 53.184 to 44.389 in the one-direction combination, 145.562 to 141.286 in the two-direction combination, and 81.442 to 76.313 in the three-direction combination. The imported energy estimation model on the on-grid rooftop solar PV system with GSCV-NN produces a more optimal and accurate model.