A Sky Image-Based Hybrid Deep Learning Model for Nonparametric Probabilistic Forecasting of Solar Irradiance

Accurate solar irradiance and photovoltaic power forecasting is critical to ensure the secure and economic operation of power systems with rapidly increasing photovoltaic generation. This paper proposes a sky image-based hybrid deep learning model for nonparametric probabilistic forecasting of solar irradiance. The proposed method utilizes variational autoencoder (VAE) to compress sky images autonomously. Long short-term memory (LSTM) is applied to extract temporal information embedded in images and time series. Quantile regression is adopted to estimate the conditional quantiles. Comprehensive case studies are conducted based on actual dataset, which shows the superiority of the proposed method and the potential for practical applications.