Illuminating Solar Potential: A Real-Time Deep Learning, IoT, and Image Processing Approach to Ground-Based Cloud Movement Forecasting for Enhanced Solar Energy Management

Cloud movement impacts the performance of photovoltaic (PV) power plants by causing sudden fluctuations in output power, leading to voltage instability in connected electricity networks. This paper introduces a novel and cost-effective deep learning solution for forecasting shortterm cloud movement to regulate PV power output. The methodology involves capturing raw image data through a PyCamera connected to a Raspberry Pi, followed by rigorous pre-processing and cloud detection techniques. Using colourbased segmentation, morphological operations, and contour analysis, the developed pipeline accurately identifies cloud regions in images and predicts their movement for the subsequent time interval. The proposed algorithm is designed to be efficient, lightweight, and suitable for deployment on low-cost processors. This study successfully implements and optimizes a cloud detection system for resource-constrained environments using the Raspberry Pi 4 Model B. The method significantly improves inference speed and model efficiency while addressing the high latency constraints associated with embedded devices with execution periods measured in seconds. The model performance metrics validate the approach, with an accuracy of 89.84%, indicating the practical potential of the model for real-time applications, offering a promising solution towards the integration of effective cloud forecasting into cost-sensitive energy management systems.