Machine Learning Techniques for Photovoltaic Power Prediction in the UAE

Abstract

Due to the rapid increase of energy demand and the continuous decrease of renewable energy cost, photovoltaic (PV) installed capacity has increased significantly. The PV power output depends on the available solar irradiance and other meteorological data such as air temperature, wind speed, and relative humidity. The performance of PV panels also depends on the cleaning frequency and maintenance of these panels. Soiling is considered to be a key factor on PV performance in desert areas. The Middle East has one of the highest dust intensities in the world which results in dramatic PV power losses. Therefore, forecasting the power output of PV panels is essential for the development of smart grids and smart metering techniques. The correlation coefficients for the concentration of air pollutants and the dust accumulation was calculated using two different datasets. The first data set was obtained by Dubai Electricity and Water Authority (DEWA) and the second data set was obtained through an experimental setup on the roof of SAN campus – Khalifa University. A shading analysis was conducted to ensure the accuracy of the power readings of the PV panels. Moreover, the study yielded very high correlation coefficient for both the concentration of air pollutants and the dust accumulated. In this study, a comparison between different Machine learning techniques was conducted to forecast the PV power generation. The machine learning techniques consisted of decision trees, neural networks, linear regression and Gaussian Process regression. All ML techniques were trained on the local weather and solar data as well as different cleaning frequencies. The results are presented in terms of different statistical indices such as the R-squared and R squared adjusted values, root mean square error (RMSE) and the mean bias error (MBE).

Date of Award	Jul 2021
Original language	American English