Publisher: 2017 IEEE Power & Energy Society Innovative Smart Grid Technologies Conference (ISGT)
Published on: 10/30/2017
DOI: https://doi-org.ezproxy.lib.vt.edu/10.1109/ISGT.2017.8086020
Download the PaperAbstract:
The use of solar photovoltaic (PV) in power generation has grown in the last decade. Unlike the traditional power generation methods (i.e. oil and gas), the solar output power is fluctuating and uncertain, mainly due to clouds movement and other weather factors. Therefore, in order to have a stable power grid, the electricity utilities need to forecast the solar output power, so they can prepare ahead adequately. In this work, hour-ahead solar PV power forecasting is performed using Support Vector Regression (SVR), Polynomial Regression and Lasso. The implemented regression models were tested under different feature selection schemes. These features include weather conditions (i.e. sky condition, temperature, etc.), power generated in the last few hours, day and time information. Based on the comparative results obtained, the SVR forecasting model outperforms the other two models in terms of accuracy.
Keywords:
Solar Forecasting , Machine Learning , Support Vector Regression , Lasso