This project aims to be an open-source replication of EUPHEMIA, the European day-ahead electricity market-clearing algorithm — validated against published market results.
EUPHEMIA clears the Single Day-ahead Coupling (SDAC) spanning most of Europe: it maximizes economic welfare over all submitted orders subject to network constraints, and the resulting zonal prices settle billions of euros of energy every year. The algorithm is publicly described but its implementation is closed.
OpenEUPHEMIA's aim is a complete, verifiable open-source implementation for the full SDAC region. The approach is incremental and validation-driven: a market region is added only once its published outcomes can be reproduced exactly, so everything in this repository is proven to work.
A market clearing publishes two outcomes: the zonal prices and the flows between zones. A case is validated against both.
| Case | Period | Prices | Flows | Run it |
|---|---|---|---|---|
| Italy (GME MGP) | April 2025 | ✅ 5,040 / 5,040 zone-hours exact — MAE 0.0000 EUR/MWh | ✅ 0.0012 MWh MAE over 4,320 link-hours — max 0.10 MWh | notebook ·  |
Each case builds its market from published data only, clears it, and compares against the outcome the market operator published. Neither result uses the quantity it predicts: the Italian zonal prices and the flows between Italian zones are held back for scoring.
The two halves are different in kind.
Prices follow from welfare maximization alone. Given the published order book and capacities, the optimum has a unique price solution and it is the one GME published — no rule, no tuning. It needs only the neighbours' published prices to close the borders; France and Greece are never modelled, merely priced.
Flows need a selection rule on top. Whenever two zones settle at the same price the exchange between them is genuinely indeterminate: many flow patterns support the identical, maximal welfare, so the solver returns an arbitrary one. A selection rule — a secondary objective over the welfare optimum, which by construction leaves the prices untouched — decides which. Measured over April 2025:
| Flow selection | Flow MAE | Exact link-hours | Reads the answer? |
|---|---|---|---|
| none — arbitrary optimal vertex | 2.89 MWh | 3,818 / 4,320 | no |
| volume maximization (documented) | 2.90 MWh | 3,814 / 4,320 | no |
| pro-rata sharing (learned) | 0.0012 MWh | 4,208 / 4,320 | no |
| anchored to the published schedule | 0.0001 MWh | 4,311 / 4,320 | yes — a reference bound, not a prediction |
Pro-rata — sharing tied same-price acceptance in proportion to submitted volume — removes three orders of magnitude of error while consuming no outcome data, and lands essentially on the anchored floor. Two caveats stated plainly: it is not in EUPHEMIA's public description but reverse-engineered from published outcomes, and it is era-dependent — before the 2025 abolition of the PUN the published splits instead follow the description's documented merit-order priority rule, which needs an order book rather than aggregated curves.
Scoring flows also requires closing the borders differently. A price-taking border is free to trade any volume within its capacity, so the border volume floats and drags the internal flows with it; the flow figures above pin each border at its published exchange instead (boundary="exchanges"). That schedule with the rest of Europe is an input to Italy's problem — it says nothing about how flow splits between Italian zones, which is what is being predicted.
A market is assembled from three tidy tables and cleared as a welfare-maximizing linear program per delivery period:
- Bid curves — every submitted bid and offer, aggregated into one supply and one demand curve per zone and period.
- Transfer capacities — the published limit of each link between zones.
- Boundary conditions — each border with the un-modelled world is closed either as a price taker at the neighbouring zone's published price (Dirichlet) or at a known exchange volume (Neumann).
Zonal prices are then the dual values of the zonal balance constraints, solved with HiGHS — exactly how EUPHEMIA defines them. An optional flow selection rule resolves what welfare leaves undecided about the flows.
Requires Python ≥ 3.11.
git clone https://github.com/rebase-energy/OpenEUPHEMIA.git
cd OpenEUPHEMIA
pip install -e .or with uv: uv sync.
A PowerMarket declares its own zones and interconnectors — no separate topology object — and is built up incrementally: bid curves, transfer capacities, boundary conditions, then cleared:
from openeuphemia import BidCurve, PowerMarket
market = PowerMarket(
name="example",
delivery_day="2025-04-01",
zones=["NORD", "SUD"],
interconnectors=[("NORD", "SUD")],
periods=[1],
)
market.add_bid_curve(
zone="NORD",
period=1,
supply=BidCurve([(100.0, 10.0), (200.0, 80.0)]), # [(volume, price), ...]
demand=BidCurve([(40.0, 4000.0), (120.0, 30.0)]),
)
market.set_ntc("NORD", "SUD", capacity_mwh=500.0)
market.add_fixed_price_boundary( # a price-taking neighbour
id="NORD_FRAN", period=1, zone="NORD", external_zone="FRAN",
price_eur_per_mwh=60.0, import_capacity_mwh=1000.0, export_capacity_mwh=1000.0,
)
result = market.clear(method="per-period-lp")
result.pricesBidCurve's default constructor takes a list of (volume, price) pairs; passing prices/cumulative_volumes as two separate sequences also works, for callers that already have the curve in that shape. BidCurve.from_steps builds one from unsorted per-step (price, quantity) data, and bid_curves_from_table builds a whole market's worth from a dataframe.
Boundary conditions are two separate methods, since they take different arguments: add_fixed_price_boundary for a price-taking neighbour (Dirichlet — free to trade within a capacity, at a fixed price) and add_fixed_flow_boundary to pin an exchange at a known volume (Neumann).
To resolve which of the welfare-equal flow patterns is returned, pass a selection rule — the prices are read before it applies and stay untouched:
market.clear(flow_selection="pro-rata") # or "volume-max", or "anchored"The Italy notebook applies exactly these steps to real data and reproduces a full month of published prices.
The validation inputs are committed under data/italy, so everything runs offline:
python scripts/replicate_italy_prices.py # prices, all 30 days of April 2025
python scripts/replicate_italy_prices.py --boundary exchanges --flow-selection pro-rataTOTAL MAE 0.0000 max 0.000 exact 5040/5040
TOTAL flow MAE 0.0012 max 0.100 exact 4208/4320
This library is about building, closing, and clearing markets — not about collecting data. Each case's inputs are prepared once from the market operator's public publications and committed as tidy tables; data/italy/README.md documents exactly what they are and how they were derived (including how non-convex block orders are handled).
- Italy — prices and flows both replicated ✅. Open: closing the one remaining flow gap without pinning the border exchanges, and the floor-price curtailment hours, where the sharing set appears to be set by non-public configuration rather than by the optimization.
- Nordics / MIBEL / CWE — extend the same curve + boundary-condition methodology, region by region, each gated on exact replication of published outcomes.
- Full SDAC — one coupled clearing of all regions, closing the boundary conditions internally.
pip install -e ".[dev]"
pytestBuilt by rebase.energy. Contributions and replication reports are welcome — please open an issue.