Comparison of physical models for bifacial PV power estimation

• A novel power estimation model is proposed by defining Dynamic Bifacial Power Gain (DBPG). • A horizontal single-axis tracking plant with black and white ground is analyzed. • Power NRMSE for black ground decreases from 2.99% (PVlib) to 1.14% (DBPG model). • PVlib power NRMSE is 4.59% for white ground, which drops to 2.02% using DBPG model. • DBPG model halves yield estimation error compared to PVlib for both ground colors. There is a growing tendency towards the application of bifacial photovoltaic (BFPV) systems all around the world, and it is predicted that they will soon take the largest share of the market. For a BFPV system, accurate estimation of power, as the main output, plays an important role as it is a critical need for trader/forecast providers and operations and maintenance companies to detect deviations from expected output. Typically, in the industry practice in the best case scenario, an effective irradiance is calculated, which combines both the global (front) tilted irradiance (GTI) and the rear tilted irradiance (RTI) into a single input parameter. Considering this point, a data-driven model has been developed here to obtain a parameter called dynamic bifacial power gain (DBPG). Adding DBPG to monofacial photovoltaic (MFPV) power output results in determination of BFPV power production. Therefore, the novel proposed data driven model (called DBPG model) is able to estimate the rear side production separately, which helps to analyze and diagnose a BFPV system better. A BFPV plant with horizontal single-axis tracking (HSAT) in Eurac Research, Bolzano, Italy, is considered as the case-study and model development is done for both black and white ground conditions. The results have confirmed much higher accuracy of the DBPG model compared to PVlib, as a tool that uses the effective irradiance. Normalized root mean square error (NRMSE) values of power estimation by PVlib are 2.99% and 4.59% for black and white ground conditions, while these are reduced to 1.14% and 2.02%, respectively, by taking advantage of the DBPG model. Moreover, excluding the near shadows conditions from the computation, using PVlib leads to 3.26% and 8.25% error in total yield estimation for ground with black over a period of almost 3 months and white surfaces over a period of almost 17 months. The corresponding values for DBPG model are much lower, i.e., 0.58% and 1.98%.