Enhancing ANN Model Perfomance to be used for Solar Radiation Prediction

A crucial first step in the design of solar energy systems and their efficacy assessment of their diverse applications involves determining solar irradiance. The aim of this work is to create a model developed using an Artificial Neural Network (ANN) to predict the daily global sun irradiance in Tindouf (Algeria). The average temperature, relative humidity, day of the year, duration, pressure, and extraterrestrial solar irradiation are the input data of the used model. A more satisfactory outcome would undoubtedly result from selecting the model's appropriate hyperparameters and handling missing data carefully. In this regard, we have created two models with the contribution of one hyperparameter optimization technique and two different approaches for handling missing data. The correlation coefficient (R) and mean absolute percentage error (MAPE) were used to evaluate the effectiveness of these models. With (R = 0.9026 and MAPE = 7.36%; rRMSE = 11.25%) in the testing phase, the findings over a five-year period showed that the model 1, employing the simple mean imputation approach and the Bayesian optimization method, gives better forecast of daily global sun radiation. Therefore, simple mean imputation may be an effective choice that can offer the best precision for the estimation of the daily solar radiation models when the data that is missing completely at random (MCAR) or missing at random (MAR) and the dataset is huge. In areas with a shortage of radiation monitoring stations, this model is helpful for sizing and building solar energy installations.