LF Energy Summit Europe 2026

Severe weather is the single largest driver of unplanned outages on distribution networks, and climate change is making it less predictable, just as renewable penetration raises the stakes for supply continuity. This session shares how SP Energy Networks (SPEN) and Sia built P4R, a probabilistic system that forecasts the location and expected volume of network faults up to five days ahead, giving control rooms time to pre-position crews and protect customers.

We go under the hood of an open-source ML stack: weather ensemble forecasts (50 members per feature) stored in Zarr, feature pipelines in the Python scientific stack, and XGBoost quantile-regression models that turn weather uncertainty into calibrated probabilistic fault predictions, orchestrated on Kubernetes. We share a hard-won lesson the whole community faces: how to build consistent historical backtests when your most predictive variables (like CAPE, a key thunderstorm indicator) only became available as ensembles recently, and how we reconstruct synthetic historical ensembles to close the gap across 2017-2024.

We close with an invitation: weather-driven reliability forecasting is a shared challenge, a natural candidate for an open LF Energy building block, complementary to forecasting projects like OpenSTEF.