Edge Machine Learning for Solar Power Forecasting

The integration of edge computing and machine learning (ML) in energy forecasting marks a transformative shift in optimizing energy systems. As energy demands fluctuate and renewable adoption grows, traditional forecasting methods struggle to adapt. By deploying ML algorithms for data streams on edge devices, it becomes possible to analyse large datasets in real time, uncovering complex patterns that improve forecast accuracy and reliability. The emergence of energy-edge orchestration, which supports continuous and efficient edge operation, further drives the need for edge-based forecasting, particularly in industrial processes powered by renewables like solar and wind. Local data processing reduces latency, lowers energy use, and enables real-time decisions for smart grids and predictive maintenance. This paper evaluates data stream ML models optimized for cloud and IoT settings, tackling challenges like concept drift, computational cost, and performance penalty. Unlike many deep learning approaches, our models maintain accuracy with reduced complexity, making them suitable for resource-constrained devices. We validate this on real-world solar power data from South Wales and energy market pricing from New Zealand, demonstrating improved renewable energy integration and sustainability through edge-based intelligence.