Êtes-vous un étudiant de l'EPFL à la recherche d'un projet de semestre?
Travaillez avec nous sur des projets en science des données et en visualisation, et déployez votre projet sous forme d'application sur GraphSearch.
Motivated by the shift of the photovoltaic (PV) industry towards bifacial solar modules, desert markets, and ever narrower profit margins, this thesis study is aimed at improving the estimation accuracy of the output of bifacial PV systems that are potentially exposed to soiling. The scope is limited to the estimation of photocurrent, the performance parameter that is the most strongly affected by bifaciality and soiling. The thesis consists of three parts with the following topics: ground surface reflectance, module soiling, and bifacial effective irradiance. As for the first topic, the thesis aims at providing tools for surface reflectance estimation in any use case. The model evaluation results that are based on a global database of reflectance measurements can be used to choose a reflectance model that is appropriate for the surface type of interest. On a global average, the results show that data-driven estimation reduces the mean absolute error of reflectance estimates by 20-40% compared to the literature values. The proposed geographically generalised parametrisations of reflectance models can be used to produce temporally variable reflectance estimates if measuring campaigns for local model calibration cannot be implemented. If they can, the thesis provides detailed guidelines for their optimal timing. The choice of the reflectance model can result in deviations of up to 2.5% in the energy yield estimated for a bifacial PV system. In the second part of the thesis, a new physical model for forecasting soiling losses is proposed. The model incorporates new methods e.g., to simulate dust deposition on tilted surfaces, to physically capture the impact of relative humidity on dust accumulation, and to estimate the hemispherical transmittance of dust for diffuse light sources. The model can be used based on the public data products of numerical weather prediction models without the need for soiling measurements. The model is validated against its alternatives and measurements made at a field experiment in the United Arab Emirates. The impact of the soiling loss model on yield estimation is found to be lower than that of the reflectance model, plus minus 1% at maximum considering usual weekly or biweekly cleaning intervals. Finally, the third part of the thesis proposes a new model for simulating the photocurrent of each solar cell of a bifacial PV power plant using a computationally efficient method. It comprises a new algorithmic design that makes it possible to make all the detailed geometric calculations once and move to the temporal domain only to estimate irradiance. The most important novelty of the proposed model is that it estimates the photocurrent contributions as a distribution of the different light sources. The model is validated against the measurements made in France and Italy. The results of the validation show that the model's current estimates are more accurate than those of the widely used "unlimited sheds" approach. The most important reasons for the better performance are the new model's capability to estimate the spatial variability of effective irradiance on the array surface and that of ground-incident irradiance around the system. As per the impact assessment, the new model was found to overestimate yield and underestimate electricity cost by 1% on average. The "unlimited sheds" approach, in turn, resulted in more variable bias. In the case of large arrays, the approach was found to overestimate yield by 4%.
Michael Graetzel, Shaik Mohammed Zakeeruddin, Yuhang Liu, Dmitry Bogachuk
Sophia Haussener, Silvan Suter, Etienne Boutin, Mahendra Patel