Future extreme intermittency screening and decision support for high-penetration renewable power systems

Operators of power systems with high renewable penetration need to know not only how large the next deviation will be, but whether the present state already warns of an extreme intermittency interval early enough to act. We formulate this as forward-looking screening and propose AERIS, an alert-ranking framework built on a temporal variational risk representation that learns the future stress score alongside the extreme-event label. On one year of open quarter-hourly data from Elia, the Belgian transmission system operator, and with a screening label validated against Elia’s own imbalance-price signals, AERIS is compared with twenty-six alternatives under one chronological, validation-only protocol. At the \(2\%\) alert budget it gives the highest precision ( \(44.7\%\) ) and recall ( \(69.4\%\) ) of all methods, and it ranks the transitions into the extreme regime best. Its advantage is largest under the conditions of real-time operation: when every input arrives one interval late it leads the tuned boosting regressors by \(5.9\%\) to \(43.8\%\) and the persistence rule by \(22\%\) , and at 45 and 60 minutes ahead it outperforms both tuned boosting regressors and the persistence rule, retaining \(38\%\) of its one-step skill against about \(30\%\) for the regressors. Because it runs on routinely published operator data and scores an interval in 0.09 ms on a CPU, AERIS can serve directly as a screening layer in existing control rooms.