Offshore wind: independent versus shared-outage exposure#
News trigger: Vestas’ Q3 2026 report published 11 November 2026 and Reuters coverage discuss costs associated with a blade incident at He Dreiht. The incident occurred in July 2026, not November. EnBW’s August 2026 commissioning update reports 64 Vestas V236-15 MW turbines (960 MW). This does not establish a fleet-wide defect.
Business decision#
For an offshore-wind asset owner, can an independent-per-turbine outage model understate extreme temporary energy exposure when a shared shutdown is plausible? How would risk tolerance differ between independent and dependent scenarios at the same expected unavailable MW?
Data provenance#
Input |
Value |
Status |
|---|---|---|
He Dreiht turbine count |
64 |
Published developer project specification |
Per-turbine nameplate |
15 MW |
Published Vestas V236-15 MW; 960 MW total |
July 2026 blade incident, reported in Q3 results |
One confirmed incident |
Not a measured fleet failure rate, downtime or cause |
Independent unavailability |
0.01 / 24-hour window |
Synthetic, not an annual hazard |
Shared-event probability |
0 to 0.06 / window |
Synthetic sensitivity; no calibration |
Shared-event scope |
All 64 turbines |
Synthetic precautionary shutdown stress case |
Counterfactual capacity factor |
0.45 |
Synthetic, not metered He Dreiht output |
Assessment duration |
24 hours |
Synthetic full-window downtime assumption |
No live provider access or downloaded licensed dataset is needed. Static public nameplate facts motivate the example; probability inputs are entirely simulated. The resulting MWh quantify conditional energy exposure, not actual lost or curtailed energy.
Mathematical method and baseline#
Let K denote unavailable turbines in a single window, n total turbines, p the individual unavailability probability conditional on no shared event, q a shared-event probability, and m the number removed by the shared event. Conditional independence of remaining units gives the exact finite mixture:
The second term is zero when k is less than m. One hypothetical full-window unavailable unit corresponds to rated MW times assumed capacity factor times the assessment hours. No random seeds are needed. For a shared event that affects all n turbines, the marginal individual unavailability is p* = q + (1-q)p. We use this same p* in a fully independent binomial baseline to make expected unavailable turbine count exactly equal. The contrast therefore isolates the effects of dependence on the upper tail. Evaluate P(K ≥ 16), mean MWh exposure and P99 exposure at different synthetic q.
Scientific limits and references#
This is a simplified mixture motivated by probabilistic-risk methods, not a calibrated beta-factor or alpha-factor model. Background: Mosleh, A., Rasmuson, D. M. and Marshall, F. M. (1998), Guidelines on Modeling Common-Cause Failures in Probabilistic Risk Assessment, NUREG/CR-5485. That publication concerns nuclear reliability; it does not supply wind probabilities.
The model ignores measured wind/weather, turbine curves, repair durations, inspection logistics, commissioning, grids, common-cause mechanisms and insurance policies. The probability assumptions are not fitted confidence intervals. The incident does not demonstrate structural defects in other turbines or justify a safety action. Do not infer actual energy losses, financial liabilities, future failure rates or developer culpability.
Reproduce offline#
uv sync --locked --all-packages
uv run --no-sync pytest methods/tests/test_wind_shared_outage.py
uv run --no-sync python examples/analysis/plot_vestas_he_dreiht_blade_risk.py
Reusable API: energy_methods.wind_shared_outage.screen_wind_shared_outage. Unit tests verify analytically computed two-turbine distributions, mean-matched independence, probability normalization, end-point cases, input domains and tail calculations. A calibrated engineering study would need qualified per-unit failure/repair records, relevant common-vulnerability cohorts and incident reports, actual production and time-varying availability.