Artificial intelligence application for the performance prediction of a clean energy community. (1st October 2021)

• Record Type: Journal Article
• Authors: Mazzeo, Domenico; Herdem, Münür Sacit; Matera, Nicoletta; Bonini, Matteo; Wen, John Z.; Nathwani, Jatin; Oliveti, Giuseppe
• Is Part Of: Energy. Volume 232(2021)
• Publication Date: 2021-10-01
• Subjects: Machine learning -- Artificial neural network -- Solar PV -- Wind turbines -- Electric vehicle charging -- Battery storage; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2021.120999
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Artificial Neural Networks (ANNs) are proposed for sizing and simulating a clean energy community (CEC) that employs a PV-wind hybrid system, coupled with energy storage systems and electric vehicle charging stations, to meet the building district energy demand. The first ANN is used to forecast the energy performance indicators, which are satisfied load fraction and the utilization factor of the energy generated, while the second ANN is used to estimate the grid energy indication factor. ANNs are trained with a very large database in any climatic conditions and for a flexible power system configuration and varying electrical loads. They directly predict the yearly CEC energy performance without performing any system dynamic simulations using sophisticated models of each CEC component. Only eight dimensionless input parameters are required, such as the fractions of wind and battery power installed, yearly mean and standard deviation values of the total horizontal solar radiation, wind speed, air temperature and load. The Garson algorithm was applied for the evaluation of each input influence on each output. Optimized ANNs are composed of a single hidden layer with twenty neurons, which leads to a very high prediction accuracy of CECs which are different from those used in ANN training. Graphical abstract: Image 1 Highlights: Artificial neural network (ANN) forecasting tool for a clean energy community (CEC). Renewables and energy storage coupled with electric vehicle charging stations. Trained ANNs are able to forecast system performances at worldwide locations. ANNs determine the yearly CEC performance without performing any dynamic simulation. Optimized ANNs are composed of a single hidden layer with twenty neurons.