Abstract

The Atacama Desert, located in the Antofagasta Region of Chile, has the highest levels of solar radiation and albedo worldwide. These conditions are ideal for bifacial photovoltaic (PV) technologies, as they maximize efficiency and energy generation. To assess the potential installable PV capacity in this region under extreme desert conditions, a machine learning-based estimation model is proposed. This model will provide approximate energy generation values based on climatic variables such as solar radiation, ambient temperature, and wind speed. With this approach, it will be possible to more accurately plan the optimal locations for bifacial PV plant installations in the Antofagasta region, reducing investment uncertainty and improving energy planning. For this study, monofacial and bifacial PV modules were installed at the Atacama Desert Solar Platform (PSDA) in Chile to compare different technologies. Two mounting configurations were evaluated: a fixed system with a 20° tilt (TFE20°) and a single-axis horizontal tracking system (HSAT). The modules used were of the PERC (Passivated Emitter Rear Contact) type, and data was collected between June 2022 and September 2023. For the analysis, two estimation models were trained: an Artificial Neural Network (ANN) and a Multiple Linear Regression (MLR) model, aiming to compare different statistical approaches. The results showed that, in all cases, the models achieved RMSE% values below 6% and MBE values close to 0%. Notably, the ANN demonstrated slightly better performance compared to the MLR, for both monofacial and bifacial modules in both mounting configurations. In conclusion, ANN models are an effective tool for estimating photovoltaic power and optimizing mounting configurations, especially in regions with extreme environmental conditions such as the Atacama Desert.