Physical and hybrid methods comparison for the day ahead PV output power forecast. (December 2017)

• Record Type: Journal Article
• Authors: Ogliari, Emanuele; Dolara, Alberto; Manzolini, Giampaolo; Leva, Sonia
• Is Part Of: Renewable energy. Volume 113(2017)
• Page Start: 11
• Page End: 21
• Publication Date: 2017-12
• Subjects: PV forecast power production -- Artificial neural network -- PV equivalent electrical circuit -- NMAE -- SolarTechlab; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2017.05.063
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: An accurate forecast of the exploitable energy from Renewable Energy Sources, provided 24 h in advance, is becoming more and more important in the context of the smart grids, both for their stability issues and the reliability of the bidding markets. This work presents a comparison of the PV output power day-ahead forecasts performed by deterministic and stochastic models aiming to find out the best performance conditions. In particular, we have compared the results of two deterministic models, based on three and five parameters electric equivalent circuit, and a hybrid method based on artificial neural network. The forecasts are evaluated against real data measured for one year in an existing PV plant located at SolarTech lab in Milan, Italy. In general, there is no significant difference between the two deterministic models, being the three-parameter approach slightly more accurate (NMAE three-parameter 8.5% vs. NMAE five-parameter 9.0%). The artificial neural network, combined with clear sky solar radiation, generally achieves the best forecasting results (NMAE 5.6%) and only few days of training are necessary to provide accurate forecasts. Highlights: The accuracy of the PV output power forecasted by different models is compared. Different deterministic models and training sizes of hybrid models are examined. Forecasts are performed on the basis of the weather forecasts given by a provider. Results are compared to real measured data of a PV silicon module. Hybrid models differently trained with 50 days provide comparable results.