Adaptive co-optimization of artificial neural networks using evolutionary algorithm for global radiation forecasting. (June 2021)

• Record Type: Journal Article
• Authors: Kılıç, Fatih; Yılmaz, İbrahim Halil; Kaya, Özge
• Is Part Of: Renewable energy. Volume 171(2021)
• Page Start: 176
• Page End: 190
• Publication Date: 2021-06
• Subjects: Global radiation -- Artificial neural network -- Hybrid model -- Evolutionary algorithm -- Adaptive optimization; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2021.02.074
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: Global radiation is not a regularly measured parameter in any weather station relative to other meteorological parameters due to measurement costs. This study has proposed hybrid artificial neural network models that predicted monthly radiation using typical weather and geographic data. Two datasets and six artificial neural network models were respectively built for indigenous and widespread regions around the world. The referred models co-optimized the artificial neural network properties and feature selection. For this purpose, an adaptive evolutionary algorithm improving prediction performance was developed to train the neural networks. This novel approach has yielded promising results compared to the developed deep learning models in this study. The results revealed that while the indigenous models had common features of longitude, sunshine durations, precipitation, and wind speed, the widespread models involved those of latitude, sunshine durations, and mean daily maximum air temperature. The proposed hybrid model had respectively the best mean absolute percentage errors of 2.45% and 9.93% for validation dataset and 3.75% and 11.03% for testing dataset of the indigenous and widespread regions, respectively. The present findings showed that the proposed hybrid model could be evaluated as a generic model and could improve the forecasting accuracy with the specified optimization parameters. Highlights: A new approach to predict monthly radiation using machine learning is proposed. Feature selection and ANN properties are co-optimized by the proposed hybrid model. Latitude, mean daily maximum temperature, sunshine durations are global attractors. A generic ANN model trained with typical world weather data is proposed. The prediction accuracy is improved with specified optimization parameters.