Innovative dust detection and efficient cleaning of PV Panels: A CNN‑RF approach using I–V curve data transformed into RGB mosaics

• RGB mosaic images derived from I to V curve parameters enhance operational state detection and dust level classification without requiring external imaging devices. • A novel CNN-RF approach enables automated anomaly detection and classification for PV panels. • The CNN-RF approach accurately distinguishes dust levels, optimizing maintenance strategies. • The proposed method outperforms conventional techniques in accuracy and resource efficiency. • Intelligent cleaning framework minimizes water consumption while ensuring energy efficiency. Photovoltaic (PV) panels are vital for renewable energy generation, yet their efficiency is critically hindered by environmental challenges such as dust accumulation, especially in arid regions like the UAE. Dust buildup can reduce efficiency by up to 30% within a month, threatening the sustainability of solar power, which is projected to supply 10% of global energy by 2030. Existing cleaning methods are unsustainable, consuming an estimated 10 billion gallons of water annually, enough to meet the drinking needs of 2 million people, necessitating the development of a cost-effective, resource-efficient alternative. This research presents a novel machine learning-based system to automate dust detection and optimize cleaning, significantly reducing water consumption while improving power generation efficiency. The methodology transforms I-V curve electrical parameters into RGB mosaic images, enabling precise classification of operational states such as normal operation, dust accumulation, shading, and faults without relying on external imaging devices. The system is built on a hybrid model combining Convolutional Neural Networks (CNN) and Random Forest (RF) classifiers (CNN-RF), where the CNN extracts high-level features from RGB mosaic images, and the RF classifier accurately categorizes operational states. Upon detecting dust accumulation, a secondary CNN-RF model classifies the severity into low, moderate, or heavy, guiding an optimized cleaning process that minimizes water usage while maintaining cleaning effectiveness. The primary CNN-RF model achieved 100% accuracy in classifying operational states using RGB mosaic images, surpassing the 97% accuracy achieved by I–V curve-based methods. Furthermore, the secondary CNN-RF model for dust severity classification attained an accuracy of 98% using RGB mosaic images, compared to only 68% when using traditional I–V curves, highlighting the superior performance of RGB mosaic images in detecting fine-grained dust levels. This optimized classification approach guides an automated cleaning system that minimizes water usage while maintaining PV panel efficiency.