Evaluation of total solar radiation with daily breakdown based on regression models

The study aims to propose new regression models using available weather data by analyzing the data published on the development of regression models for evaluating the flux of total solar radiation. Following an analysis of literature sources, primary stages in developing regression models and approaches to their implementation are described. Models are developed and compared for accuracy based on weather data (maximum and minimal temperature, air humidity, overall and lower cloudiness) in Irkutsk over 2007–2019. For calibration and validation of the models, open databases of ground measurements of weather stations were used. Ten known and seven new regression models were calibrated and validated, including three models based on the support vector method. The new models based on air temperature and humidity, atmospheric pressure, as well as overall and lower cloudiness, showed the highest accuracy in evaluating the total solar radiation with daily breakdown. The maximum mean absolute error in evaluating daily total solar radiation over 2016–2019 comprised 627.52 W·h/m2·day for the analyzed known models, 504.7 W·h/m2·day for the newly proposed regression models, and 463.2 W·h/m2·day for the regression models based on the support vector method. The conducted analysis of the mean bias error revealed models having the highest accuracy in evaluating monthly and annual sums of total solar radiation were determined. These include a known regression model using air humidity data and a regression model based on the support vector method.

solar radiation, regression model, support vector machine, validation, calibration, meteorological parameters