CENovaSage

Capacity expansion and production cost, solved on the nodal network.

CENovaSage is our flagship capacity-expansion and production-cost model. It decides what to build, where and when (generation, storage, transmission and flexibility) and what the system then costs to run, on a network model with transmission limits, so congestion and prices show up at the buses where they happen. It is the model behind Scenario Studio on CEAtlas. Every day it also runs a production-cost model of each of the nine US regions, about 20,000 buses in total, to forecast hourly day-ahead nodal prices for today and tomorrow on CEAtlas Desk. Those forecasts are scored daily in a free public record across 7 US markets.

One model planning, production cost and prices9 US regions forecast daily~20,000 buses in the US forecast networkLicensed hosted or on your infrastructure

Planning, production cost and daily prices from one nodal model

A nodal model prices each point on the network separately, so congestion shows up where it happens instead of being averaged across a region. CENovaSage uses the same model for planning, for production cost and for the daily forecast.

Capacity expansion

What to build, where and when: generation, storage, transmission and flexibility, with the dispatch and nodal prices under each plan, hourly for every day of a full weather year (Scenario Studio deliverable runs).

Production cost

The same nodal model dispatched hour by hour over a fixed fleet, giving what the system costs to run, where congestion binds and what sets the price.

US day-ahead prices

Each morning, an hourly nodal price forecast for today and tomorrow across all nine US power regions, on the CEAtlas map for Desk and Enterprise subscribers. Accuracy is published every day on a free public scorecard.

Scenario Studio, the daily US forecast and licensed studies

Scenario Studio on CEAtlas (Enterprise) runs on CENovaSage. Interactive runs use 24 representative days and take from minutes to about an hour for a full region. Deliverable runs cover a full weather year at hourly resolution and are solved to a stated optimality gap of 2%, which takes hours rather than minutes. Full-year sweeps run about 3.5× faster, with identical results (measured 30 Sep 2026). It covers Oregon today, a 752-bus model; more regions are on the platform roadmap.

For the US forecast, all nine regions (ISO-NE, NYISO, PJM, MISO, SPP, ERCOT, the West, the Southeast and Florida) are solved every morning and shown together on one map: hourly prices by bus, rolled up to the ISOs’ own pricing nodes; generation by unit and technology; line flows against their limits; and demand. Downloads for each run are on CEAtlas Desk and Enterprise.

Next on the roadmap are long-term price forecasts into the 2050s and continental-scale planning studies. CEForesight, a separate pricing model, is in development.

What is modelled and what is scored

CENovaSage is written in Julia on JuMP and MathOptInterface. Every model is an LP or a MIP, and the solver is a configuration choice.

It is built for multi-decade planning, with yearly investment periods out to the horizon your data covers. Capacity expansion uses nested Benders decomposition: an investment master problem over parallel daily dispatch subproblems, so every day of the weather year is solved hourly at its true weight. The master is stabilised with a level bundle and Magnanti–Wong (Pareto-optimal) cuts, as described in CE-WP-2026-03. For planning studies, N-1 security is built in and benchmarked on IEEE test systems: contingencies are screened with line outage distribution factors and enforced with lazily added cuts. In planning runs the network is a B-θ DC power flow and can include piecewise-linear losses. Full-year Benders runs are live today on Scenario Studio’s deliverable tier (Oregon, 752 buses, to a 2% gap).

Benchmarks: IEEE 118-bus expansion, 10 scenarios × 8 representative weeks with N-1, converged in 22.5 minutes (June 2026). IEEE 24-bus RTS expansion, 5 scenarios × 4 representative weeks with N-1, in about 80 seconds (July 2026).

The US forecast

The forecast network has about 20,000 buses across the nine US regions, with transmission limits, plus published interface limits in MISO and PJM. Each morning’s run forecasts every hour of today and tomorrow, which is the next market day’s day-ahead prices. The model looks one more day ahead so the end of tomorrow isn’t priced at the edge of the horizon; those extra hours are neither published nor scored.

The seven US markets that publish day-ahead prices are scored against them. The West is modelled as the whole Western Interconnection and scored at California’s three trading hubs. The Southeast and Florida have no organised day-ahead market, so they are modelled but have no prices to score.

Outputs and interop

Every run reports nodal prices (the duals), every planning run reports its optimality gap, and results go to the CEAtlas map with CSV downloads. Configuration is TOML; tables are Parquet, Arrow or CSV; networks import from and export to PyPSA (CSV-folder format); and models export as MPS or LP.

The daily US run list and the accuracy scorecard are free on CEAtlas. The forecasts themselves (prices, flows, dispatch and downloads) are on CEAtlas Desk and Enterprise, and Scenario Studio is on Enterprise. Planning studies on your own system are available under licence.