Energy modelling and risk analysis · South Africa

Averages Hide the Years That Hurt

StatForge builds hourly, stochastic models of South African power projects — capture price under the new wholesale market, curtailment and grid constraint, generation yield, storage value and debt cover. Every result is a distribution across weather and market scenarios, not a single assumed number.

Michael Pietersen (PhD) View Google Scholar →

The market is changing what a model has to do

South Africa is moving from a single-buyer system to a traded one. The market operator was licensed in December 2025, the wholesale market opens with Eskom and publicly procured plants, and the phase after that brings mandatory participation for every generator above 10 MW.

At that point revenue stops being a contracted number and becomes a price you capture in the hours you happen to generate — which depends on the weather, on what everyone else is generating at the same time, and on how often the grid refuses your output. A model built for a fixed tariff cannot answer that question.

What we model

Hourly, stochastic modelling across the chain — from weather and price through to debt cover.

Market and capture-price modelling

What a plant actually earns once price varies by hour: capture price against flat-tariff equivalents, price duration curves, and revenue distributions for merchant, hybrid and contracted structures as the wholesale market phases in.

Generation and resource modelling

Hourly production built from decades of measured reanalysis weather rather than a single typical year — P50 and P90 yield, inter-annual variability, and the correlation between sites and technologies that determines whether a portfolio diversifies or concentrates risk.

Curtailment and grid-constraint modelling

Hourly simulation against published connection pipelines, with energy loss reported across scenarios rather than as one assumed percentage — including how reductions are allocated between projects, which is where most of the project-level uncertainty actually sits.

Storage and flexibility valuation

Arbitrage and curtailment-recovery value tested against the actual duration structure of prices and constraints, not their annual averages — because a battery sized against a mean performs very differently against a hundred-hour spell.

Revenue, contract and debt-cover analysis

Contracted, wheeled and merchant structures compared on the same underlying generation, with deemed-energy and compensation exposure priced explicitly, and debt cover reported as a distribution so the years that break a loan are visible.

Independent model review

Assessment of an existing energy, revenue or curtailment model: specification, assumptions, identification, and whether the uncertainty it reports is credible or decorative.

Selected work

One worked example, built end to end from public data. The subject is grid curtailment; the method — measured weather, hourly simulation, graded assumptions, distributions instead of point estimates — is the same one applied to capture price, yield and storage.

Screening assessment · public data · fully reproducible

Curtailment Risk on the Cape Corridors

How much wind energy will the grid refuse by 2028, and what does that do to a project's finances? Hourly simulation over 1,000 resampled runs drawn from 32 years of measured ERA5 weather, reproducible from a fixed seed, with every key assumption graded and sourced.

The plan implies carrying capacity must grow 70–90% in two years

For the official ~4% curtailment design intent to hold once the announced 2028 build-out connects, the corridor's implied carrying-capacity parameter must rise from 3,121 MW to 5,592 MW. That is a model parameter rather than a construction target — operational fixes, storage and demand all count toward it.

Which contract you hold matters more than which weather you get

Within the constrained future, holding weather identical, expected 2028 revenue runs from R171m with no effective cover, through R195m under the actual capped regime, to R293m if fully funded — a R122m span of pure contract mechanics, against roughly R19m of weather variation between a calm year and a windy one.

Three futures, three completely different projects

Corridor-wide wind curtailment in a typical 2028 year: 0.9% on the design track, 5.0% at the grid company's demonstrated delivery pace — 271 km of line built against a 423 km target — and 20.3% if capacity stays where it is while the build-out proceeds anyway.

Batteries are not a universal fix

A 25 MW / 100 MWh battery recovers about 99% of curtailed energy in the design-track future but only about 29% in the constrained one, because curtailment then arrives in spells averaging 9.9 hours and exceeding 100 hours at worst.

Download the full report (PDF, 21 pages)

Free, no registration required.

Scope and limits. This is a screening assessment built from public data. It is not a project-specific curtailment forecast and not a lender-grade independent assessment. The model represents two large, electrically distinct Cape networks as a single aggregate constraint, and a transmission engineer's review is a stated precondition before investment-grade use. Every key assumption is listed and graded in Appendix A.

Who this is for

Independent power producers and developers

Generation, price and curtailment risk quantified before financial close, in a form lenders can interrogate rather than accept on faith.

Lenders and project finance teams

Debt-cover distributions across weather and market scenarios, and independent review of the revenue assumptions in a sponsor's model.

Technical and transaction advisers

Modelling capacity behind diligence when a mandate needs quantitative depth the team cannot staff that month.

Corporate offtakers and traders

What a wheeled or traded position actually delivers hour by hour, and where the exposure sits between the parties to it.

How engagements work

Three ways to bring in modelling capacity, depending on where the project sits.

Scoped modelling project

A defined question, answered against your own parameters.

  • Scope, deliverables and timeline agreed before work starts
  • Your project or portfolio run through the relevant scenario framework
  • The data-request list your diligence should be demanding

Fees quoted per engagement.

Model build and handover

A model you own, documented and reproducible.

  • Project- or portfolio-specific hourly model
  • Graded assumption register and a validation suite included
  • Reproducible from a fixed seed — every figure traceable to source

Fees quoted per engagement.

Independent review

Assessment of a model someone else built.

  • Specification, assumptions and identification assessed
  • Whether the stated uncertainty is credible
  • Written opinion with material risks ranked

Fees quoted per engagement.

Discuss a project or portfolio

Tell us what needs answering. We respond with scope, method, deliverables and timeline.

Your information is used only to scope your request. We do not sell or share your details.

Or email directly: hello@statforge.co.za