International audienceIn recent years, the penetration of photovoltaic (PV) generation in the energy mix of several countries has significantly increased thanks to policies favoring development of renewables and also to the significant cost reduction of this specific technology. The PV power production process is characterized by significant variability, as it depends on meteorological conditions, which brings new challenges to power system operators. To address these challenges it is important to be able to observe and anticipate production levels. Accurate forecasting of the power output of PV plants is recognized today as a prerequisite for large-scale PV penetration on the grid. In this paper, we propose a statistical method to address the problem of stationarity of PV production data, and develop a model to forecast PV plant power output in the very short term (0-6 hours). The proposed model uses distributed power plants as sensors and exploits their spatio-temporal dependencies to improve forecasts. The computational requirements of the method are low, making it appropriate for large-scale application and easy to use when on-line updating of the production data is possible. The improvement of the normalized root mean square error (nRMSE) can reach 20% or more in comparison with state-of-the-art forecasting technique
Photovoltaic (PV) power generation is characterized by significant variability. Accurate PV forecasts are a prerequisite to securely and economically operating electricity networks, especially in the case of large-scale penetration. In this paper, we propose a probabilistic spatio-temporal model for the PV power production that exploits production information from neighboring plants. The model provides the complete future probability density function of PV production for very short-term horizons (0-6 hours). The method is based on quantile regression and a L1 penalization technique for automatic selection of the input variables. The proposed modeling chain is simple, making the model fast and scalable to direct on-line application. The performance of the proposed approach is evaluated using a real-world test case, with a high number of geographically distributed PV installations and by comparison with state-ofthe-art probabilistic methods.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.