Prediction of global solar irradiation using hybridized k-means and support vector regression algorithms

By T. Ayodele, A. Ogunjuyigbe, A. Amedu, J. Munda. Renewable Energy Focus, 2019

64 citations recorded by the source corpus.

An algorithm of k-means hybridized with support vector regression (SVR) has been developed and utilised for the prediction of daily global solar radiation. The k-means was used to cluster the seasonal data into separate clusters independent of the seasons. Two observations, each from the group of wet (rainy) seasons and dry seasons, were used as the initial centroids for the clustering process. Subsequently, the clustered data were subdivided into four sky conditions as each serve as input to seven corresponding SVR models developed. The proficiency of this model was verified using six years of meteorological data of Ibadan (2010–2015) to predict daily global solar radiation for the future years (2016–2017). The predicted values were also compared with the forecasted values obtained from three established models i.e. ANN, Angstrom–Prescott and ARMA models. The performance of the model was investigated using different statistical test of fits (i.e. RMSE, RRMSE, MAPE and R2). Some of the key results revealed that the proposed k-means-SVR has the capability to predict the global solar irradiation for the year 2016 with overall accuracy of R2 of 0.9816, RMSE of 0.4275 MJ/m2/day, RRMSE 2.6515% and MAPE of 1.7928%. Similarly, the model was able to predict the 2017 global solar irradiation with the overall values of R2, RMSE, RRMSE and MAPE of 0.9842, 0.434 MJ/m2/day, 2.7498% and 1.795% respectively. The proposed model was also found to possess better predictive capability compared to the existing models, thus making it applicable in energy planning.

Subfields: Artificial Intelligence, Electrical and Electronic Engineering, Renewable Energy, Sustainability and the Environment

View Paper: https://www.semanticscholar.org/paper/bfd85dc50f487badf355cbfdf95d1fe299fabd0a