Yash Vijay Amonkar , David J Farnham, Upmanu Lall

Spatially distributed renewable energy generation poses unique risks to power systems since the aggregate amount of energy produced in any hour depends on the spatial correlation structure of the sources. Moreover, the spatial correlation structure can vary with the time of day and season and depend on the state of the large-scale climate. These features pose a challenge for resource adequacy risk assessment using traditional statistical or machine learning methods. A new algorithm based on spatially clustered k-nearest neighbors to capture the spatio-temporal dynamics of wind and solar fields is presented and applied to data from ERCOT, Texas. The algorithm skill is analyzed both at the aggregated field level and also at the individual site level. The algorithm's utility in assessing temporally varying risks of lower-than-expected target wind and solar energy production across ERCOT is demonstrated.