Capacity expansion planning for power systems.

Long-horizon fleet planning — what to build, when, and where. CENovaSage solves a multi-period capacity-expansion MILP across the same 27-zone European network that CEGridSight already models, returning least-cost build schedules that have been verified against the dispatch model rather than asserted to be feasible.

Get on the early-access list CEAtlas Expert pricing

Why this exists

Most capacity-expansion models propose builds the dispatch model can't actually run.

The legacy stack runs expansion and dispatch as separate models against separate datasets, then bolts them together with a feasibility hack at the end. CENovaSage shares fleet, network, and price assumptions with CEGridSight by construction — every plan it proposes can be re-run as an annual hourly dispatch on the same fleet model, and infeasibilities show up before the plan ever leaves the optimiser.

What it does

Three pillars.

Multi-decade horizon

Annual time steps from today through 2050. Resolve build, retire, and refurbish decisions per technology per zone, with capital-cost trajectories from NREL ATB and DDP studies — fully overridable per scenario.

Policy-aware constraints

Toggle CO₂ caps, renewable portfolio standards, capacity-mechanism floors, and interconnector upgrades. See the marginal abatement cost of each policy lever as a shadow price on the binding constraint, not a separate sensitivity run.

Dispatch-feasible output

Every expansion plan is fed back into CEGridSight's annual hourly MILP for verification — no infeasible builds, no hand-waving on flexibility. Plans that fail the check come back with the binding constraint named.

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.

Formulation

Multi-period co-optimised expansion + dispatch.

Decision variables: build, retire, and refurbish capacity by tech × zone × year, plus per-scenario hourly dispatch in a representative-day reduction. Objective: minimise NPV of capex, fuel, carbon, and unserved-energy cost over the planning horizon, subject to reserve margin, RPS, CO₂ cap, and inter-zonal transfer constraints.

Time reduction

k-medoids representative weeks, plus extreme-day padding.

12 representative weeks selected per year via k-medoids on hourly load, wind capacity factor, and solar capacity factor — plus a hard-coded set of 4 extreme-stress days (peak load, low-wind cold snap, summer heat, ramp event) so resource-adequacy constraints bind on the days that actually matter, not the smoothed average.

Network model

Pipe-flow with HVDC topology.

27-zone European topology with explicit HVDC interconnector capacities (NSL, Eleclink, Viking Link, NeuConnect, etc.) and AC tie-line aggregates. Interconnector upgrades enter as binary build decisions with their own capex curves.

Verification loop

Annual hourly back-check.

Each candidate plan is replayed as a full 8,760-hour CEGridSight dispatch on the proposed fleet. Hours of unserved energy, ramp violations, and reserve shortfalls are reported back; the planner re-tightens reserve constraints and re-solves until the plan dispatches cleanly. Solved by HiGHS today; CEMeridian when it ships.