Improving the Prediction of Solar Radiation Using ANFIS Optimization Ensemble

Abstract

A reliable design and monitoring of solar energy-based systems, requires accurate information on the available solar radiation. However, measurement of solar radiation is difficult due to the high cost of measuring devices, together with their calibration and maintenance, particularly in developing countries like Nigeria. Meanwhile, solar radiations in such regions are often predicted using data-driven techniques. Nevertheless, the existing predictive models often produce unsatisfactory results. This study proposed the development of intelligent models for forecasting of solar radiation in Kano state, Nigeria. The model is developed using ensemble machine learning method, by combining two Adaptive neuro-fuzzy inference systems (ANFIS) with sub clustering optimization (ANFIS-SC) and grid partitioning optimization (ANFIS-GP). The models are developed using meteorological data consisting of a maximum temperature, minimum temperature, and mean temperature as predictors. Three different scenarios were considered. The performance of the models is evaluated using the correlation coefficient (R), determination coefficient (DC), mean-squared error (MSE), root-mean-squared error (RMSE) and mean-absolute error (MAE). The simulation results indicated that the ANFIS ensemble (ANFIS-ENS) outperformed the individual ANFIS models. , Furthermore the overall performance of the ANFIS-ENS with three inputs, has the highest accuracy, with training , DC , and testing , , a training MSE=0.0199, RMSE=0.1412, MAE=0.1014 and testing MSE=0.0198, RMSE=0.1408, MAE=0.1038. The developed models can be reliably used as alternative tool for estimation of solar radiation in Kano