Comparison of three machine learning models for the prediction of hourly PV output power in Saudi Arabia

The optimum integration of photovoltaic (PV) technologies into existing power systems necessitates accurate PV performance planning, which is critical for both plant operators and the grid. This study investigates the application of different machine learning (ML) models to predict the PV power output at Jubail Industrial City, Kingdom of Saudi Arabia. Specifically, three techniques have been explored which include k nearest neighbour (kNN), Multiple regression and decision tree regression each with its own set of hyper-parameters & functions. Using the Ostwald's technique, large dataset comprising the hourly solar irradiance and temperature covering a three-year period, i.e. 2016 to 2019, has been initially used to estimate the PV power for Jubail. Then, the PV power was predicted using the aforementioned ML models. The kNN outperformed the other models, with root mean square error, mean absolute error and normalized root mean square error of 18.68%, 80.6%, and 13.2%, respectively. The other two ML techniques i.e. MLR and the DTR performed reasonably well. As a further contribution, the kNN was applied to forecast the day ahead PV output power for Jubail and the results shows a good agreement between the predicted the actual values. The results imply that using ML techniques, it is possible to predict PV output power across Saudi Arabia, and that this data may be used as a reference for predicting PV output power in different regions of the country. (c) 2021 THE AUTHOR. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-ncnd/4.0/).