While traditional methods for modelling the thermal and electrical behaviour of photovoltaic (PV) modules rely on analytical and empirical techniques, machine learning is gaining interest as a way to reduce the time, expertise, and tools required by designers or experts while maintaining high accuracy and reliability. This research presents a data-driven machine learning tool based on artificial neural networks (ANNs) that can forecast yearly PV electricity directly at the optimal PV inclination angle without geographic restrictions and is valid for a wide range of electrical characteristics of PV modules. Additionally, empirical correlations were developed to easily determine the optimal PV inclination angle worldwide. The ANN algorithm, developed in Matlab, systematically and quantitatively summarizes the behaviour of eight PV modules in 48 worldwide climatic conditions. The algorithm's applicability and robustness were proven by considering two different PV modules in the same 48 locations. Yearly climatic variables and electrical/thermal PV module parameters serve as input training data. The yearly PV electricity is derived using dynamic simulations in the TRNSYS environment, which is a simulation program primarily and extensively used in the fields of renewable energy engineering and building simulation for passive as well as active solar design. Multiple performance metrics validate that the ANN-based machine learning tool demonstrates high reliability and accuracy in the PV energy production forecasting for all weather conditions and PV module characteristics. In particular, by using 20 neurons, the highest value of R-square of 0.9797 and the lowest values of the root mean square error and coefficient of variance of 14.67 kWh and 3.8%, respectively, were obtained in the training phase. This high accuracy was confirmed in the ANN validation phase considering other PV modules. An R-square of 0.9218 and values of the root mean square error and coefficient of variance of 31.95 kWh and 7.8%, respectively, were obtained. The results demonstrate the algorithm's vast potential to enhance the worldwide diffusion and economic growth of solar energy, aligned with the seventh sustainable development goal.

A user-friendly and accurate machine learning tool for the evaluation of the worldwide yearly photovoltaic electricity production

Mazzeo D.;Leva S.;Matera N.;
2023-01-01

Abstract

While traditional methods for modelling the thermal and electrical behaviour of photovoltaic (PV) modules rely on analytical and empirical techniques, machine learning is gaining interest as a way to reduce the time, expertise, and tools required by designers or experts while maintaining high accuracy and reliability. This research presents a data-driven machine learning tool based on artificial neural networks (ANNs) that can forecast yearly PV electricity directly at the optimal PV inclination angle without geographic restrictions and is valid for a wide range of electrical characteristics of PV modules. Additionally, empirical correlations were developed to easily determine the optimal PV inclination angle worldwide. The ANN algorithm, developed in Matlab, systematically and quantitatively summarizes the behaviour of eight PV modules in 48 worldwide climatic conditions. The algorithm's applicability and robustness were proven by considering two different PV modules in the same 48 locations. Yearly climatic variables and electrical/thermal PV module parameters serve as input training data. The yearly PV electricity is derived using dynamic simulations in the TRNSYS environment, which is a simulation program primarily and extensively used in the fields of renewable energy engineering and building simulation for passive as well as active solar design. Multiple performance metrics validate that the ANN-based machine learning tool demonstrates high reliability and accuracy in the PV energy production forecasting for all weather conditions and PV module characteristics. In particular, by using 20 neurons, the highest value of R-square of 0.9797 and the lowest values of the root mean square error and coefficient of variance of 14.67 kWh and 3.8%, respectively, were obtained in the training phase. This high accuracy was confirmed in the ANN validation phase considering other PV modules. An R-square of 0.9218 and values of the root mean square error and coefficient of variance of 31.95 kWh and 7.8%, respectively, were obtained. The results demonstrate the algorithm's vast potential to enhance the worldwide diffusion and economic growth of solar energy, aligned with the seventh sustainable development goal.
2023
Artificial neural network
Electricity
Machine learning
Photovoltaic module
PV forecasting
Validation
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S2352484723009654-main.pdf

accesso aperto

Descrizione: Articolo finale
: Publisher’s version
Dimensione 7.58 MB
Formato Adobe PDF
7.58 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1243737
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
social impact