Abstract
There is an increasing tendency towards application of bifacial photovoltaic (BFPV) technologies to overcome the current energy crises all around the world [1, 2]. The 14th International Technology Roadmap for Photovoltaic (ITRPV) [3] has estimated that the current shares of monfacial photovoltaic (MFPV) and BFPV in the world market are ~65% and ~35%, respectively. In 2027, BFPV is forecasted to take the lead with the contribution of ~55%. Three years later, i.e., in 2030, BFPV share will be ~60%. Accurate estimation of power plays an important role in designing and monitoring of a BFPV system. The way to obtain BFPV by PVlib, as the common simulation approach, is identical with MFPV, where using bifaciality concept as a simplifying assumption, an effective irradiance is employed to consider both front and back radiation together. Taking this point into consideration, this work is done with the aim of providing a novel data-driven physical model to estimate BFPV power output. As a big advantage, the proposed model does not need rear tilted irradiance (RTI) or albedo, and the input parameters consist of only the angle of incidence (AOI), sun elevation and azimuth angles, and clearness index. The model has been developed for different albedo conditions and various tracking strategies, including fixed-tilt (FT) and single-axis tracking (SAT). The accuracy of the developed model for each case is compared with the best combination of electrical and thermal models in PVlib for each case, which has indicated the superiority of the proposed data-driven approach. According to the results, for DTU plant, in which FT BFPV is used, the normalized root mean square error (NRMSE) has the values of 1.7% and 1.4% for the PVlib and the proposed models, respectively. For Bolzano (BZ) plant with the black ground and SAT, the improvement gets better, where NRMSE reaches from 2.2% for the PVlib to 0.9% for our data-driven model. The corresponding NRMSE values for BZ plant with white ground and SAT are 4.3% and 1.8%, respectively. The findings show the more significant improvement of the proposed model in comparison to PVlib simulation for high albedo and SAT, which is usually the preferred working condition of BFPV systems.