Abstract:
The application of machine learning in everyday life is growing and extending to every aspect of life. Not only is it becoming more accurate but also is showing great improving results in many different industries. Nowadays the machines have the ability to learn from past experiences just like humans of the real world can, and that’s a relieve, they have changed the way we think about problems and they have changed the solutions by making them easier to implement. In this study we are giving a closer look to renewable energy. Also, we will check what are the meteorological factors that affect the most solar radiation. This thesis will explain in details models to predict the Solar Radiation for Tirana city, the capital of Albania. Such prediction has never been performed for Albania, neither for Tirana nor any other city for that matter. The resources available to have valuable data for this study are very few near to none for Tirana. There is no adequate equipment to measure radiation, and neither any meteorological station for such observations. Part of this paper are also econometric models and statistical considerations. The measured date, temperature, wind speed, pressure relative humidity and also solar radiation are used for measuring the accuracy of the forecasting model.
Data used are past time series of meteorological data. This thesis shows a detailed analyze of forecasting methods using statistical means were we can say that pressure(press) and temperature (Temp) have a positive relation with Solar radiation (SR) while relative humidity (RH) and wind speed (WS) have negative relation and all are statistically significant. The main algorithm that is explained in details is ARIMA (3,1,1). This algorithm performed very well with an R =0.92 and RMSE equal to 71.67(Wh/m2). And the best performance is made by Random Forest with an R equal to 0.93 and RMSE 68.76(Wh/m2). In addition to this and prediction of next 30 days is made.
