Models for estimation of solar irradiance in Zimbabwe: A statistical and machine learning approach

Abstract

<p>The present study focused on the statistical development of solar irradiance predictive models for locations with limited solar irradiance measuring equipment. Multiple linear regression models were developed using both measured and satellite corrected meteorological data. The study chose easy to measure and access meteorological data for analysis and modelling. Multicollinearity and correlation analysis were performed to analyse the relationships among the independent and depended variables. Statistical predictive models were developed, and the prediction accuracy of the developed models was analysed using the coefficient of determination (R<sup>2</sup>) and the Mean Absolute Percentage Error (MAPE). The results revealed a higher performance of the developed models compared to generic empirical models. The prediction MAPE for the three models developed were respectively 0.117 kWh/m<sup>2</sup>, 0.132 kWh/m<sup>2</sup> and 0.044 kWh/m<sup>2</sup> for H<sub>g</sub>, H<sub>b</sub> and H<sub>d</sub>. The models also had R<sup>2</sup> values of 0.895, 0.972 and 0.993 respectively for global horizontal irradiance (H<sub>g</sub>), direct normal irradiance (H<sub>b</sub>) and diffuse irradiance (H<sub>d</sub>). The developed models outperformed the generic models by a minimum of 5.74%. The study showed that it is more accurate to predict Global Horizontal Irradiance by summing the predicted component of H<sub>b</sub> and H<sub>d</sub>.</p>