TY - JOUR

T1 - Predicting diurnal outdoor performance and degradation of organic photovoltaics via machine learning; relating degradation to outdoor stress conditions

AU - David, Tudur

AU - Soares, Gabriela

AU - Bristow, Noel

AU - Bagnis, Diego

AU - Kettle, Jeff

PY - 2021/12

Y1 - 2021/12

N2 - Accurate prediction of the future performance and remaining useful lifetime of next-generation solar cells such as organic photovoltaics (OPVs) is necessary to drive better designs of materials and ensure reliable system operation. Degradation is multifactorial and difficult to model deterministically; however, with the advent of machine learning, data from outdoor performance monitoring can be used for understanding the relative impact of stress factors and could provide a powerful method to interpret large quantities of outdoor data automatically. Here, we propose the use of artificial neural networks and regression models for forecasting OPV module performance and their degradation as a function of climatic conditions. We demonstrate their predictive capability for short-term energy forecasting of OPV modules, showing that energy yield can be predicted if climatic conditions are known. In addition, the model has been extended so that the impact of climatic conditions on degradation can be predicted. The combined model has been validated on unseen OPV module data and is able to predict energy yield to within 4% accuracy.

AB - Accurate prediction of the future performance and remaining useful lifetime of next-generation solar cells such as organic photovoltaics (OPVs) is necessary to drive better designs of materials and ensure reliable system operation. Degradation is multifactorial and difficult to model deterministically; however, with the advent of machine learning, data from outdoor performance monitoring can be used for understanding the relative impact of stress factors and could provide a powerful method to interpret large quantities of outdoor data automatically. Here, we propose the use of artificial neural networks and regression models for forecasting OPV module performance and their degradation as a function of climatic conditions. We demonstrate their predictive capability for short-term energy forecasting of OPV modules, showing that energy yield can be predicted if climatic conditions are known. In addition, the model has been extended so that the impact of climatic conditions on degradation can be predicted. The combined model has been validated on unseen OPV module data and is able to predict energy yield to within 4% accuracy.

U2 - https://doi.org/10.1002/pip.3453

DO - https://doi.org/10.1002/pip.3453

M3 - Article

VL - 29

SP - 1274

EP - 1284

JO - Progress in Photovoltaics

JF - Progress in Photovoltaics

SN - 1099-159X

IS - 12

ER -