Regional Solar Irradiance Forecasting Using Multi-Camera Sky Imagery and Machine Learning Models

Abstract

With the increasing integration of photovoltaic (PV) systems into power grids, accurate short-term forecasting of solar irradiance is essential for efficient energy management. This paper presents a machine learning model developed using a synthetic dataset designed to analyze the potential of multi- camera sky imaging for regional solar irradiance forecasting. The dataset, generated in a controlled simulation environment, cap- tures cloud dynamics and solar irradiance at multiple locations within a given region. The proposed model utilizes sky images from multiple virtual cameras strategically positioned to provide spatially distributed observations. By combining image-based features with historical irradiance measurements, the model shows improved forecasting accuracy compared to single-camera approaches. The results indicate that multi-camera systems better capture the spatial variability of cloud cover and allow the model to predict solar irradiance for locations without installed cameras. This research highlights the potential of multi-camera configurations for regional forecasting and provides valuable insights for grid operators and energy planners. The results support the adoption of distributed sky imaging networks as a practical approach to improve solar irradiance predictions and ultimately contribute to the stability and reliability of solar-powered energy systems through improved forecast accuracy.