Solar Panel Soiling Classification Using Machine Learning Algorithms

With the widespread adoption of solar panels as a sustainable and cost-effective source of energy, it becomes paramount to ensure that these panels perform well, especially that they are prone to several factors that can affect their performance. One of these factors is soiling and dusting. This research agrees with other related studies confirming that machine learning methods are useful in PV systems monitoring. This paper aims at researching the application of machine learning algorithms to identify the soiled panels. In particular, the paper investigates the ability of the Long Short-Term Memory (LSTM) and Time Large Language Model (LLM) Transformer machine learning algorithms to detect and classify soiling on photovoltaic panels, thus notifying the responsible person to do the needed cleaning. Both models were trained on the time-series data that were collected by the researchers, including: the temperature, humidity, and irradiance. The models’ performances were ana-lyzed with the help of evaluation metrics. The highest accuracy of the LSTM model in the identifications of soiling patterns was up to 96.11%, whereas in the Time LLM Transformer model was 93.6%.