Soft sensors for dust estimation with high accuracy: AI-driven approach for CSP solar mirror soiling classification

Concentrating solar power plants convert solar energy into electricity by using mirrors to concentrate sunlight, but dust, particularly in arid and semi-arid regions, reduces their efficiency. This study proposes an innovative AI-based approach for monitoring and predicting soiling. For real-time analysis of CSP mirror images, a ResNet-50 convolutional neural network (CNN) is used, while a recurrent neural network (RNN) with a long, short term memory (LSTM) architecture predicts future soiling rates using the temporary variation in this parameter. By integrating mirror imagery and meteorological data, our hybrid AI model effectively combines CNN for image analysis and RNN temporal predictions, with high accuracy ranging from 89.5% to 95.2%, depending on the application. The application of advanced AI techniques can significantly improve the management of concentrated solar power plants and lead to significant breakthroughs in the solar energy industry.