Power forecasting of three silicon-based PV technologies using actual field measurements. (February 2021)

• Record Type: Journal Article
• Authors: Omar Nour-eddine, Id; Lahcen, Boukhattem; Hassani Fahd, Oudrhiri; Amin, Bennouna; aziz, Oukennou
• Is Part Of: Sustainable energy technologies and assessments. Volume 43(2021)
• Publication Date: 2021-02
• Subjects: Grid connection -- Linear and nonlinear models -- PV power forecast -- Safi -- Morocco; Renewable energy sources -- Periodicals; Energy development -- Technological innovations -- Periodicals; Electric power production -- Periodicals; Energy storage -- Periodicals; 333.79
• DOI: 10.1016/j.seta.2020.100915
• Languages: English
• ISSNs: 2213-1388

Abstract

Highlights: Use of widely common variables in such weather station as predictors. Choice of fitting statistical method to eliminate outliers and influential points from raw data. 5 min ahead PV power forecasting with simple linear and nonlinear models. Comparison of the developed models with a wide variety of forecasting models. New models provide better accuracy with MAE and RMSE less than 2.107% and 2.674% respectively. Abstract: This work focuses on short-term photovoltaic (PV) power forecasting each 5 min of the following day of a 5.94 kWp grid-connected PV plant located in Safi, Morocco. The PV system is composed of three silicon technologies: mono-crystalline (m-Si), poly-crystalline (p-Si) and amorphous (a-Si). The field measurements from June 18, 2016 to July 15, 2018 were used to build both linear and nonlinear models. These suggested models have been compared with the ones presented in the literature, including the persistence and an Artificial Neural Network (ANN) models widely used for short-term PV output forecasting. The comparison has been performed on statistical scores: MBE, MAE, MAPE, RMSE, and nRMSE. The two proposed models maintain simple mathematical expressions and involve only a small number of predictors. The latter are plane of array solar irradiance and module or ambient temperatures. The models are also able to follow the time-varying of solar irradiance. The findings showed that the suggested models outperform all the tested ones and achieved better performances in terms of prediction accuracy. In fact, MAE and RMSE of our models do not exceed 2.107% and 2.674% respectively, whereas they may achieve respectively 2.488% and 5.796% for persistence and 4.430% and 5.650% for ANN.