Spatial-temporal solar power forecast through use of Gaussian Conditional Random Fields

This paper introduces an application of the Gaussian Conditional Random Fields (GCRF) model for forecasting the solar power in electricity grids. The introduced forecasting technique is capable of modeling both the spatial and temporal correlations of various solar generation stations. It will be demonstrated in this paper how the suggested solution can significantly improve the forecast accuracy compared to the conventional forecasting models such as the persistent (PSS) model and the autoregressive with exogenous input (ARX) model. Besides, the GCRF model outperforms the other two models under the scenarios with unavailable or missing data. The suggested probabilistic model of the GCRF can also help better managing the existing uncertainties of the solar generations. The numerical experiments are conducted through which the effectiveness of the proposed approach is validated.