Long term estimation of global horizontal irradiance using machine learning algorithms

Abstract

The integration of solar power into the smart grid has become quite ubiquitous among scientists and researchers to obviate the need of fossil fuels in the wake of environmental and ecological conservation. Therefore, exact and precisive prediction of global horizontal irradiance (GHI) is indispensable for quantifying the production of solar power in future. The main aim of the present study is to develop the time series forecasting models for next five years by using the past data from the period 2017–2019 of four districts of Rajasthan i.e., Bikaner (D1), Jodhpur (D2), Jaisalmer (D3) and Barmer (D4). This paper compares the efficacy of novel time series estimation models such as auto-seasonal autoregressive integrated moving average (auto-SARIMA), Facebook Prophet (FBP) and Neural Prophet (NP) for predicting the future values of GHI. In order to obtain the best fit between the test data and its prediction outcome, optimized parameters of auto-SARIMA models are selected. The performances of the models are assessed using different error metrics such as root mean squared error (RMSE) and mean absolute error (MAE). The RMSE values calculated for districts D1, D2, D3 and D4 by auto-SARIMA model are 18.69, 21.62, 14.473 and 10.70 W/m2 and their MAE values are 15.00, 19.31, 12.99 and 9.222 W/m2 respectively. Conclusively, it is observed that auto-SARIMA model outperforms the other models as far as reduction in RMSE and MAE values are concerned.