Machine Learning-Based Prediction of Photovoltaic Panel Soiling Using Satellite Images and Spectral Index

Velásquez R.A.; Donaires L.C.; Lara J.M. · Proceedings of the 2025 IEEE 32nd International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2025 (2025) · DOI: 10.1109/intercon67304.2025.11244682

Resumen

This paper developed a robust, data-driven methodology for predicting soiling levels on photovoltaic (PV) solar panels using high-resolution satellite data and a combination of machine learning techniques. The approach integrates multi-temporal Sentinel-2 Surface Reflectance (S2SR) data with advanced spectral indices to quantify particulate accumulation and atmospheric conditions that impact PV performance. The study area, located in northern Colombia in 2.16 km2 and Peru in Moquegua 1.00 km2, Satellite images were extracted from the Google Earth Engine within a defined time range and filtered using day-of-year constraints to ensure the inclusion of only high-quality scenes. This methodological framework supports the scalable analysis of PV soiling across large areas and seasonal cycles, facilitating proactive maintenance strategies in tropical solar environments.

Instituciones: Universidad Nacional de San Agustín de Arequipa; Universidad Tecnológica del Perú