In recent years, Italy and Europe have been engaged in a decarbonization process aimed at achieving the goal of carbon neutrality by 2050. To meet this challenge, it is essential to optimize the generation of energy from photovoltaic (PV) systems, both by using innovative components (such as bifacial modules on single-axis trackers with the axis of rotation in the North-South direction) and digital O&M techniques, including by accurately forecasting the expected output from non-programmable systems, as well as monitoring their performance over time.

Given the strong development of PV systems with bifacial modules on fixed or solar-tracking structures, and the need for adequate production forecasting systems for these types of modules, the present study focused on analysing the production achieved by these types of PV systems compared to those with single-sided modules by analysing the experimental data acquired within the European GOPV project at the Cadarache site. The study was aimed at identifying the significant differences between these different types of PV systems in order to be able to identify the key elements to be taken into account in their production.

The results of the analysis showed a gain, in terms of production, of about 14 % for installations with bifacial modules compared with single-sided ones, allowing an advantage for the same area occupied. In relation to the use of single-sided trackers, the study showed an increase in production during the morning and afternoon, but a reduction when the Sun is high in the sky, leading to the advantage of increasing daily production hours at high power. The study was also aimed at identifying the existing relationships between production and environmental variables, with a view to developing a method for forecasting production from plants with bifacial modules in the continuation of operations.

For this purpose, a machine learning model was developed that estimates the power output from a bifacial module based on different predictors measured in the field or calculated mathematically. The best performing model, with relative Mean Absolute Error (rMAE) values of 3.5 %, relative Root Mean Square Error (rRMSE) values of 4.7 % and relative Bias Error (rBIAS) values of 1.0 %, was obtained by considering, as is usually done, the variables of global and diffuse horizontal irradiance, the cosine of the PV module tilt angle with respect to the Sun, to which albedo values will be added in the remainder of this research (to account for bifaciality). Therefore, these will be the variables to be accurately predicted in order to develop a reliable model for forecasting production from systems with bifacial modules on single-axis trackers.

The Report is available on the Italian site