Optimizing Artificial Neural Networks for the Accurate Prediction of Global Solar Radiation: A Performance Comparison with Conventional Methods

Obtaining precise solar radiation data is the first stage in determining availability of energy. It also regarded as one major inputs for a variety applications. Due to scarcity measurement many locations throughout world, models are utilized predict global radiation. Indeed, most widely used AI technique artificial neural networks (ANNs). Hitherto, while ANNs have been various studies estimate (GSR), limited attention has given architecture ANN. Thus, this study aimed to: first, optimize design faster and machine-learning (ML) algorithms, ANN, forecast GSR more accurately saving computation power; second, number neurons hidden layer obtain significant ANN model accurate estimation, since it still lacking; addition investigating impact varying on proficiency ANN-based with high accuracy; and, finally, conduct comparative between empirical techniques estimating GSR. The results showed that best provided an excellent estimation GSR, Coefficient Determination R2 greater than 0.98%. Additionally, architectures smaller single (1–3 neurons) performance, > Furthermore, performance developed remained approximately stable when layer’s was less ten (R2 0.97%), their very close each other. However, experienced instability exceeded nine neurons. comparison revealed both performed well 0.98%). Moreover, relative error slightly range, ±10% November December, within range even winter months. obtained work were compared recent related work. While had good RMSE value 0.8361 MJ/m2 day−1 ranges previous work, its correlation coefficient (r) one. Therefore, can be forecasting. efficient using valuable designing evaluation different