Built · CENovaSage

By Compounding Energy

Built · in benchmarking Julia / JuMP · solver-agnostic · HiGHS today, CEMeridian drop-in.

Capacity expansion that's nodal, secure, and continental — and fast enough to actually run.

CENovaSage co-optimises what to build, when, and where — generation, storage, transmission, HVDC, and demand-side flexibility — across multi-decadal horizons under stochastic uncertainty, with N-1 contingency security and reliability enforced as constraints inside the optimisation, not bolted on as a post-hoc check. It takes the founder's decades of experience in capacity expansion and production-cost modelling and strips out as many of the failure modes of the traditional models as possible — and solves in investment cadence.

Get on the early-access list The solver behind it

Why this exists

Zonal models misprice the three things that decide a plan.

Most capacity-expansion tools aggregate geography into a handful of zones, screen security afterwards, and accredit reliability with a flat planning-reserve margin. That hides exactly what drives a least-cost plan: where congestion actually binds, what storage is really worth, and where to site. CENovaSage keeps the network nodal from the start, screens N-1 inside the solve, and closes an ELCC reliability loop — so the plan it returns is one the dispatch model can run.

What you get from a run

An auditable least-cost plan, with the price signals.

Optimal investment trajectory

Build / retire / refurbish by technology, location and year, per scenario — with unit-commitment-fidelity dispatch under each plan, endogenous retirements, and decoupled storage power/energy sizing.

Security & reliability, co-optimised

N-1 contingency-secure operation via CANOPI lazy security cuts, plus ELCC-based reliability accreditation fed back as gradient cuts — not a post-hoc screen, not a flat PRM.

Nodal LMPs + a full audit trail

Nodal prices with a 3-component decomposition (energy / congestion / marginal-loss), plus duals, optimality gap, invariant report and a seed-pinned manifest for bit-exact replay.

Methodology

For the quants in the room.

Plain-English first; this section for anyone vetting the maths. Skip if you're not building the plan yourself.

Decomposition

Three-level nested Benders (AFN-B).

An Adaptive-Fidelity Nested Benders scheme: an investment master, parallel SCUC dispatch subproblems, and CANOPI N-1 screening. Stabilised with a level-bundle method (κ = 0.6) and Magnanti-Wong Pareto-optimal cuts — the structure that lets nodal + sub-hourly + secure problems scale where a monolithic LP cannot.

N-1 security

LODF fast filter + lazy contingency cuts.

Line-outage distribution factors screen contingencies; violations enter as lazy cuts with a persistent active set and a staleness guard, so security binds where it matters without enumerating every outage every iteration.

Time & resource

Multi-resolution + wake-endogenous wind.

5- or 15-minute resolution on stress-detected windows, hourly elsewhere — so adequacy binds on the days that matter, not a smoothed average. Wind capacity factors are wake-endogenous via pruned-ReLU surrogates (<1% MAPE), so siting decisions see the wake penalty.

Scope & data

HVDC, demand flex, multi-commodity fuels.

HVDC as controlled flow with converter losses and ramp; hierarchical demand-side flexibility with comfort bounds; multi-commodity fuels (H₂ / NH₃ / CO₂); policy instruments (RPS / CES / caps / ITC / PTC). TOML / Parquet / Arrow / NetCDF I/O — JSON-free, enforced in CI — with PyPSA and MPS interoperability.

Versus the incumbents

What zonal monoliths leave on the table.

The founder's decades in capacity expansion and production-cost modelling, rebuilt to remove the failure modes of the traditional tools — and to solve in investment cadence.

Nodal from the start

DC OPF (B-θ) inside the expansion, not zonal aggregation with a nodal post-process — preserving the price formation that drives storage value and siting.

N-1 inside expansion

CANOPI lazy cuts with a persistent active set — incumbents run contingency as a separate, after-the-fact analysis, if at all.

Modern decomposition

Level-bundle + Magnanti-Wong Benders, not a monolithic LT solve — the route to nodal + secure + stochastic at continental scale.

No commercial-solver lock-in

Runs on open HiGHS today, CEMeridian drop-in — no per-customer Gurobi/CPLEX licence required. Air-gapped deployment supported.

Measured

Every claim benchmarked on the same instance and hardware.

Claims discipline is part of the programme: no speed or quality number appears in a deck, paper or page until it's benchmarked against monolithic-solver and stabilised-Benders baselines on the same problem.

~55s

IEEE 24-bus RTS, 5 scenarios × 4 representative weeks, N-1 — target < 5 min ✓

25.5 min

IEEE 118-bus, 10 scenarios × 8 weeks, N-1 — target < 30 min ✓

GB scale

200- / 300-bus GB on the CEAtlas topology — on the roadmap, post-CEMeridian for the largest cases

The nested decomposition turns one intractable monolith into nodal subproblems that solve in parallel — and on exactly that large nodal DC-OPF shape, our in-house CEMeridian solver is measured 1.9–4.4× faster than HiGHS, with decomposed capacity expansion 1.5–1.7× faster — the route to GB continental scale. CENovaSage runs on HiGHS today and is solver-agnostic, so the speedup lands as a drop-in.

How it ships

Library, CLI, container — and a built-in dashboard.

Drive a run from Julia, from the shell (cenovasage run config.toml), or as a container that ships the engine plus a web dashboard for run management, convergence monitoring, maps and result exploration. Consumes CEAtlas candidate registries and topologies; shares the dispatch core, resource pipelines and reproducibility discipline of the rest of the stack.

CEAtlas topology CESentinel reliability

Status

Built — in benchmarking.

The core nested-Benders engine, N-1 CANOPI, storage / HVDC / demand-flex / multi-resolution / wake surrogate / ELCC, the CEAtlas connector, PyPSA-MPS interop and the web UI are built and tested; the IEEE 24- and 118-bus benchmarks are met. GB 200/300-bus scale and the CEMeridian backend are the active roadmap. Early-access partners are paired with the team for spec and validation review against their own scenarios.

Request early access Back to all products