Long-Term Forecasting Potential of Photo-Voltaic Electricity Generation and Demand Using R

• Main Authors: Karina Vink, Eriko Ankyu, Yasunori Kikuchi
• Format: Article
• Language: English
• Published: MDPI AG 2020-06-01
• Series: Applied Sciences
• Subjects: energy demand; long-term forecasting; machine learning; R programming; solar power generation; support vector regression
• Online Access: https://www.mdpi.com/2076-3417/10/13/4462

For micro-grid cost-benefit analyses, both energy production and demand must be estimated on the long-term of one year. However, there remains a scarcity of studies predicting energy production and demand simultaneously and in the long-term. By means of programming in R and applying linear, non-linear, and support vector regression, we show the in depth analysis of the data of a micro-grid on solar power generation and building energy demand and its potential to be modeled simultaneously on the term of one year, in relation to electricity costs. We found solar power generation is linearly related to solar irradiance, but the effect of temperature on total output was less pronounced than anticipated. Building energy demand was found to be related to multiple parameters of both time and weather, and could be estimated through a quadratic function in relation to temperature. Models for both solar power generation and building energy demand could predict electricity costs within 8% of actual costs, which is not yet the ideal accuracy, but shows potential for future studies. These results provide important statistics for future studies where building energy consumption of any building type is correlated in detail to various time and weather parameters.