Industry: Energy
Product: ModelRisk
Application: Battery storage technology selection and sizing

Sizing a 200 MW / 800 MWh BESS: Storage Technology Choice with ModelRisk

A vertically-integrated utility holds an interconnection queue position for a 200 MW storage asset alongside a 350 MW solar farm in west Texas. The investment committee has narrowed the technology shortlist to three: a 4-hour lithium-iron-phosphate (LFP) battery at $340/kWh installed, an 8-hour vanadium redox flow battery at $510/kWh, and a 10-hour pumped-hydro site at $1,950/kW with site-specific civil costs. Each will live in the same merchant + capacity-revenue market for 20 years. The deterministic NPV ranking changes depending on which point estimate of LMP volatility, cycle life, and capacity-payment escalator you plug in. The decision needs a distribution, not three numbers.

Run over 50,000 stochastic 20-year price/cycle/degradation paths, the three technologies produce three differently-shaped NPV distributions — and the shapes, not the means, are what the committee actually has to choose between:

LFP has the lowest mean NPV ($71M) but the tightest distribution and the highest 5th-percentile NPV ($32M) — the floor case is still comfortably profitable. Vanadium flow's mean NPV ($83M) edges LFP, but its cycle-life advantage is partly cannibalised by lower round-trip efficiency. Pumped hydro has the highest mean ($94M) and the highest 95th percentile ($187M) but a 9% probability of negative NPV, driven entirely by the civil-cost lognormal tail. The committee's risk appetite — not the point-estimate NPV — chose the technology.

Three technologies, three failure modes

The technologies are not just priced differently — they degrade and earn revenue under different stochastic structures.

LFP lithium-ion. Calendar + cycle degradation modelled jointly. End-of-life is reached at the earlier of 12 calendar years or 6,000 equivalent full cycles (EFC). With merchant arbitrage averaging 1.3 cycles/day, expected EFC at year 12 is 5,700 ± 480 (Normal, fitted to OEM warranty + observed degradation). Replacement of the cell stack at year 12 costs $185/kWh in real terms, modelled as LogNormal (μ_log = ln 185, σ_log = 0.22) to reflect supply-chain tail risk.

Vanadium flow. Effectively no cycle degradation, but round-trip efficiency is the weak point: modelled Beta(α = 28, β = 5) centred on 75% with 95% CI [69%, 81%]. Beta is bounded, which RTE must be — a Normal here would let the simulator draw efficiencies above 100%. Auxiliary load (pumps, heaters) is Triangular(3.5%, 5%, 8%) of throughput.

Pumped hydro. Civil-cost overrun is the killer. Reference-class data from 47 completed pumped-storage projects fit LogNormal with μ_log = ln 1.0 and σ_log = 0.38 — the deterministic estimate is the median, but the mean is 8% higher and the 90th percentile is 65% above plan. The asset itself has a 60-year life with round-trip efficiency Normal(78%, 1.5%).

What the deterministic comparison missed

The deterministic NPV table, using point estimates of $52/MWh average DA-RT spread, 85% LFP RTE, 12-year LFP life and 10% civil-cost contingency, ranked the choices as Pumped > Flow > LFP by $14M, $9M, and a baseline NPV of $76M respectively. The Monte Carlo simulation told a more useful story: the ranking by mean (Pumped > Flow > LFP) survives, but the ranking by downside inverts. LFP — last on the mean — is first on the 5th percentile, and pumped hydro's mean-leading position carries a one-in-eleven chance of losing money outright.

What actually moves the answer

For the pumped-hydro option, civil-cost overrun is over three times the next driver. This is not a price problem; it is a project-execution problem, and it routes the diligence dollar to geotechnical investigation and EPC contract structure rather than to power-price forecasting.

Sizing on tail value, not average

Power capacity (MW) and energy capacity (MWh) trade off differently in arbitrage versus capacity-market revenue. The team swept the duration of an LFP system from 2 hours to 6 hours, holding power at 200 MW.

Mean NPV peaks at 3-hour duration, but the P10 NPV peaks at 4-hour. The longer system survives more of the unfavourable price scenarios because its capacity-market accreditation is higher (ERCOT's ELCC curve credits longer-duration BESS more in summer peaks). The 4-hour system was selected on P10-maximising grounds, even though the mean said 3-hour.

What changed

• LFP 4-hour selected over pumped hydro, primarily on the 9% probability of negative NPV in the pumped case driven by civil-cost lognormal tail — a tail invisible in the deterministic ranking.
• Capex contingency repriced: civil-works contingency lifted from 10% to 25%, matching the P75 of the lognormal overrun fit rather than a culturally-set round number.
• Capacity-market revenue treated as stochastic, with ELCC accreditation drawn from a Beta around the published 2024 ERCOT curve — adding 6% to expected revenue and 14% to its standard deviation.
• Stack-replacement reserve funded from year 1 as an annuity sized to the year-12 LogNormal replacement-cost mean plus 1σ, rather than the original deterministic point estimate.

ModelRisk Functionality Used

• Joint calendar + cycle degradation for LFP — Normal cycle count with hard-cap end-of-life at the earlier of 12 years or 6,000 EFC.
• Beta-bounded round-trip efficiency for flow battery and pumped hydro — Normal would have allowed unphysical >100% draws.
• LogNormal civil-cost overrun for pumped hydro, calibrated to 47-project reference-class dataset (σ_log = 0.38) and exposing a 9% probability of negative NPV.
• NPV-distribution comparison across three technologies on a shared simulated price stack, scoring by mean, P5, P95, and P(NPV < 0) jointly.
• Duration sweep that revealed P10 NPV peaks at a longer duration than the mean — driving the 4-hour vs 3-hour decision.
• Tornado ranking that routed diligence spend to geotechnical risk rather than to power-price forecasting for the pumped option.

Storage is not a single technology decision; it is three different distributions of NPV with three different shapes. Monte Carlo simulation in ModelRisk is what lets the investment committee choose the shape it wants, not just the point it sees.