A Stochastic Framework for Solar Irradiance Forecasting Using Condition Random Field

Jin Xu, Shinjae Yoo, Dantong Yu, Hao Huang, Dong Huang, John Heiser, Paul Kalb

Abstract: Solar irradiance volatility is a major concern in integrating solar energy micro-grids to the mainstream energy power grid. Accounting for such fluctuations is challenging even with supplier coordination and smart-grid structure implementation. Short-term solar irradiance forecasting is one of the crucial components for maintaining a constant and reliable power output. We propose a novel stochastic solar prediction framework using Conditional Random Fields. The proposed model utilizes features extracted from both cloud images taken by Total Sky Imagers and historical statistics to synergistically reduce the prediction error by $$25$$ - $$40\%$$ in terms of MAE in $$1$$ - $$5$$ minute forecast experiments over the baseline methods.