Ambient temperature and solar irradiance forecasting prediction horizon sensitivity analysis

• Main Authors: Jose Ramirez-Vergara, Lisa B. Bosman, Walter D. Leon-Salas, Ebisa Wollega
• Format: Article
• Language: English
• Published: Elsevier 2021-12-01
• Series: Machine Learning with Applications
• Subjects: Machine learning; Mean absolute error; Photovoltaic systems planning; Sensitivity analysis
• Online Access: http://www.sciencedirect.com/science/article/pii/S2666827021000645

Selecting the correct weather forecasting technique is a crucial task when planning an efficient solar energy generation system. Estimating accurate solar photovoltaic systems power output depends on the correct modeling of solar irradiance and ambient temperature, evidencing the need for a framework to select the correct technique to forecast these parameters. This paper presents a review of the forecasting methods to predict solar irradiance and ambient temperature, considering the sensitivity to the forecasting horizon. The methodology includes estimating an interval for ambient temperature and solar irradiance by using the Mean Absolute Error as the percentage of variation in these parameters. To provide context, the study considers best-case and worst-case scenarios for four cities, estimating the power output for a sample array and analyzing the differences between the cases. The power output estimation of the PV array varied between 36% and 50% (on average) for the short-term prediction, and 54% to 95% for the long-term. The changes in the location produced an average variation of 43% in terms of power production, and up to 187% in economic value (USD) for the short term, and 44.5% and 187% for the long term. The results suggest a marked sensitivity to the variation in the forecasting horizon and significance with regards to location selection (considering the changes in solar irradiation and the cost of electricity).