Solar Radiation Ramping Events Modeling Using Spatio-Temporal Point Processes.

This article focuses on developing a novel spatio-temporal categorical point process model to accurately capture and predict solar radiation ramping events—abrupt changes in solar power generation influenced by weather and exhibiting strong spatial and temporal correlations. Using bihourly solar radiation data from multiple U.S. locations, the model estimates birthrate and interaction parameters via least squares and maximum likelihood methods, providing interpretable insights into spatial dependencies and temporal dynamics. The study introduces dynamic decision thresholds to improve online prediction accuracy under nonstationarity and demonstrates superior performance compared to deep learning baselines across several cities. The framework also offers theoretical guarantees for parameter recovery and can be extended to multistate ramping events, seasonal modeling, and other spatio-temporal phenomena such as wildfire or disease spread prediction.