This study implemented and validated View Factor (VF) and Ray Tracing (RT) simulation models against four-season field-measured data to evaluate the accuracy of energy yield prediction in bifacial PV systems under three installation configurations: (1) Single-row tilted, (2) Multi-row tilted, and (3) Vertical East–West facing. Front-side and rear-side irradiance and electrical energy yield were evaluated using nRMSE and nMBE metrics, and the relationship between irradiance component ratios (direct, diffuse, reflected) derived from RT results and error trends was analyzed. For front-side irradiance prediction, VF and RT methods showed similar performance in Single-row tilted and Multi-row tilted systems (nRMSE difference within 1 percentage point), while the RT method generally showed lower error than the VF method for rear-side irradiance prediction across the evaluated systems. Notably, in the Multi-row tilted system with high structural complexity, RT achieved nRMSE of 13.3%, which was 14.4 percentage points lower than VF (27.7%). Critically, the performance difference between the two methods was maximized under diffuse-dominant conditions. During shaded periods of the Vertical East–West facing system (diffuse ratio 76–81%), VF method’s nRMSE increased to 14.1–33.5%, while RT method maintained a stable level of 8.8–14.3%. This difference is likely related to differences in diffuse-radiation treatment between the two approaches, including the anisotropic sky representation used in the RT workflow and simplified assumptions in the VF-based rear-side model. In the temporal resolution analysis, the 1 h interval showed the lowest nRMSE for cumulative energy prediction, while the 5 min interval more accurately reproduced peak timing and rapid power fluctuations. This study suggests that the RT method can improve rear-side irradiance prediction accuracy, particularly under conditions of increased structural complexity and higher diffuse radiation ratios. It also indicates that simulation temporal resolution should be selected according to research objectives, such as long-term energy yield estimation or short-term power fluctuation analysis.
Yoo et al. (Tue,) studied this question.