<p>The traditional monofacial tracking algorithm does not account for the irradiation received on the rear side of bifacial photovoltaic (bPV) modules. Meanwhile, the existing bifacial tracking algorithm overlooks the impact of high-temperature and high-irradiation conditions on the power generation performance of bPV modules, and also neglects spatial variations in ground-reflected irradiation. Additionally, the traditional photovoltaic tracking methods often lead to unnecessary energy consumption due to frequent motor rotations under low-irradiation conditions. To address these issues, we propose a bifacial tracking algorithm suitable for horizontal single-axis tracking brackets, which comprises three sub-models. (1) The optical-electrical-thermal coupling model divides the ground beneath the module into segments and independently evaluates the irradiation incident on each segment. The cross-string method is used to evaluate the view factors from each segment to the rear side of the bPV module, in order to more accurately assess the total rear irradiation. (2) The theoretical optimal tracking angle model calculates the tracking angle that maximizes the estimated power output under current weather conditions. This angle not only optimizes the irradiation received on both sides of the bPV module, but also considers the impact of high-temperature and high-irradiation environments on power generation efficiency. (3) The motor tracking angle model dynamically adjusts the working time of the motor by weighing the power generation gain and motor loss. Thus, the number of motor rotations under low-irradiation conditions can be reduced. The experimental results show that the relative error of total rear irradiation evaluated by the proposed optical model is 23.41% and 20.92% lower than that evaluated by the traditional optical model under cloudy and sunny conditions, respectively. The power generation estimated by the coupling model was validated over one week, and the relative error between the measured and the estimated power generation is 2.66%. At the same time, the annual energy yield simulated by PVsyst software is compared with that estimated by the coupling model. The relative error between the estimation results and the PVsyst simulation results is 2.89%. In experiments on four typical dates, the bifacial tracking algorithm proposed in this paper increased power generation by 9.2% compared to the monofacial tracking algorithm. The frequency of motor start-up has been reduced by 80% compared to traditional tracking methods.</p>