Industry: Energy Product: ModelRisk Application: Grid Capacity Expansion and Reserve-Margin Planning
NERC's reference reserve margin for a thermal-dominated balancing authority is 15%; for a system with growing renewable penetration, the effective load-carrying capability (ELCC) of solar and wind is what really matters, and 1 GW of wind nameplate contributes only 100–250 MW to reserve margin at the right system peak hour. An ISO's 10-year capacity plan that uses nameplate-based reserve math will show 18% reserves on paper and a 22% probability the system falls below 10% reserves on the actual peak hour — a number that triggers emergency procurement, capacity scarcity prices, and load-shed protocols.
A regional system operator with a 38 GW peak rebuilt its capacity-expansion model in ModelRisk to capture three sources of stochastic uncertainty the deterministic Integrated Resource Plan (IRP) was averaging away: load growth, the ELCC of new wind and solar, and the timing of thermal retirements under environmental policy uncertainty. Run across 50,000 ten-year futures, the planning year's reserve margin is not a number on a cover page — it is a distribution with a worryingly long lower tail.
The simulation puts the mean 2032 reserve margin at 14.5%, with a P10 of 5.6% and a P90 of 23.8% — against a deterministic IRP that counted solar and wind at higher placeholder values and reported 19.4%. There is a 55% probability of landing below the 15% target and an 8% probability of falling below 5%, the reliability cliff at which load shed becomes a non-zero hourly probability. The rest of this study is about where that lower tail comes from and which procurement lever cuts it most cheaply.
The deterministic IRP reports reserve margin as:
\[ \text{RM} = \frac{\text{Installed capacity} - \text{Peak demand}}{\text{Peak demand}} \]
with installed capacity counted at nameplate. The Monte Carlo reframing replaces every input with a distribution and counts each resource at its probabilistic ELCC:
It is the ELCC reframing that does most of the work: counting 10.5 GW of solar-plus-wind nameplate at its probabilistic peak-hour contribution — roughly 1.2 GW on average rather than the placeholder values in the deterministic plan — is what pulls the mean reserve margin down from the IRP's 19.4% to 14.5% and opens the lower tail shown above.
The next question is procurement: where to spend the next $800M of capacity additions. The model compared four strategies by re-running the simulation with each strategy's mix layered in:
The CCGT-only strategy delivers the highest mean reserve margin (19.1%) and the highest P10 (10.0%) per dollar of firm nameplate — but it concentrates the entire reliability bet on gas-supply availability during winter coincident peaks, a correlated risk the marginal-MW math does not price. The solar-plus-storage strategy sits lowest of the four on both mean (16.4%) and P10 (7.5%): in low-ELCC-realisation paths the reserve-margin lift is only a fraction of its expected value. The balanced portfolio lifts the P10 reserve margin from 5.6% to 9.1% while spreading the lift across thermal, storage-paired solar, and demand response — the diversification the reliability council actually evaluates against a single-resource concentration limit.
Solar ELCC dominates — a 5%–14% ELCC swing moves P10 reserves by 5.0 percentage points, and the realised ELCC depends on weather correlation patterns the deterministic IRP simply does not model. Load growth and thermal retirement timing follow. The wind ELCC ranks fourth despite the larger uncertainty band, because the nameplate is smaller.
In capacity planning, the reserve margin printed on the IRP cover page is the mean of a distribution. The CFO has to spend capital, the regulator has to approve recovery, and the dispatcher has to keep the lights on — all on the left tail of that distribution. Monte Carlo is how the three conversations stay aligned.