Relevance

The energy transition runs on batteries. As coal, gas, and nuclear power plants are switched off, grid-forming battery storage systems are stepping in to do their job — actively holding the grid's voltage and frequency stable, providing the inertia that keeps the lights on, and even restarting the grid after a blackout. They are becoming the backbone of a renewable power grid. But there is a catch: a grid-forming BESS must run for around 25 years under harsh, constantly changing conditions, and today nobody can reliably predict how its batteries will actually age in the field. Closing that gap is one of the most valuable open problems in battery intelligence — and exactly the kind of problem where novel AI can change the game. This is your challenge.

Why does it matter so much? A grid-forming inverter plus a BESS is a long-life, safety-critical asset, and knowing how its batteries age — today and years into the future — drives every important decision: how to design and size the system, how to operate it to balance revenue against degradation, when to schedule maintenance, how to honor warranties, and whether it can still deliver the grid services it was built for. The difference between a good and a poor lifetime model is measured in millions of euros and in grid reliability.

And it is genuinely hard. Multiple aging mechanisms act at once and couple together, calendar and cycling aging interact, and degradation is path-dependent and varies from cell to cell.

Challenge 1 — Cycle lifetime prediction

Your first challenge is to predict how long a cell lasts. Given controlled lab aging data, build a model that predicts the full SOH trajectory over cycles down to end-of-life (70% SOH) — including the knee-point where degradation suddenly accelerates — and make it generalize to operating conditions and cell formats it has never seen during training.

Accurate cycle-lifetime prediction is directly useful in two ways: in the design phase, to select cells, size systems, and estimate warranty exposure under the expected duty; and during operation, to understand how temperature and load shape lifetime — enabling operating strategies that trade revenue against degradation.

 

 

Figure 1 — Predicting the SOH trajectory to end-of-life, including the knee-point, from early-life data.

 

Challenge 2 — Field SOH estimation and prediction

Your second challenge is the real-world test. For BESS packs running under real operating conditions — with no begin-of-life reference, only nominal capacity and operational data — first estimate the current pack SOH, then predict how it will evolve over the next 6 to 12 months.

This is the harder, more realistic problem: noisy and irregularly sampled field data, current-sensor drift, pack heterogeneity and cell-to-cell variation, and weak observability in the LFP voltage plateau. Solving it enables predictive maintenance, warranty evaluation, residual-value assessment, and dispatch confidence across deployed fleets.

 

 

 

Figure 2 — Estimating current pack SOH and forecasting it 6–12 months ahead for a deployed field pack.

Data

Participants work with lab cycling data across multiple LFP cell formats, temperatures, and C-rates; dynamic lab packs with field-representative load profiles; and labeled real EU residential field data will be used for validation. Open-source datasets may also be used.