Enhancing Solar Farm Performance Monitoring Using Attention-Based LSTM Models and Satellite Weather Data

Accurate performance monitoring of utility-scale photovoltaic plants is challenging without on-site weather data. Such data is essential for determining the weather-corrected performance ratio (WCPR), a key parameter in assessing the plants' operational status. However, installing and maintaining onsite weather stations is expensive. Additionally, these stations are prone to faulty sensors and communication errors, which limit data availability.This study utilises satellite-derived weather data and machine learning models to estimate weather parameters. The quality of the models is evaluated by comparing the WCPR calculated using on-site data and the estimated data. A minor error of only 2.2% demonstrates the significant potential of this approach to reduce costs and improve the availability of weather data for the operation and maintenance of utility-scale photovoltaic plants.