CE BESS Arbitrage tells a battery operator what to do in the next half-hour, and why. Day-ahead and intraday schedules are co-optimised across day-ahead, intraday auction, balancing mechanism, and frequency response — over the same 27-zone European dispatch model that drives CEGridSight, so the price paths the optimiser bids into are the price paths we'd publish to a forecaster. One model, no decoupling drift.
The legacy stack is a forecast service stapled to a deterministic dispatch heuristic, often run by a PE-owned vendor on a yearly licence with no view into how the prices were made. CE BESS Arbitrage is different in three places that matter:
Energy, balancing, and FFR/DC price paths are produced by the same 27-zone unit-commitment MILP that drives CEGridSight — not a separate ML model trained on lagged prices. The optimiser bids into prices that are physically self-consistent with the fleet, fuel, and weather assumptions.
Decisions are taken under a scenario tree of price paths, not a single forecast. The bid you submit hedges across the realisations the dispatch model thinks are plausible — quantified by a published P10–P90 band, not a black box.
Every dispatch decision links back to the price scenario, the binding state-of-charge constraint, and the degradation model run that produced it. Bring a regulator or an LP a year of operations, replay any half-hour, see what the model believed.
Hourly (or half-hourly) charge/discharge schedule for the day-ahead auction, co-optimised against the forecast distribution of intraday and balancing prices. Produces explicit bid curves, not a single number — so you can submit price-quantity pairs that survive volatile clears.
Half-hourly re-optimisation against the latest intraday auction prints and the live BM imbalance signal. State-of-charge is carried forward, FFR/DC commitments are honoured, and the model decides whether to defend a stack position or rebuild it.
Co-optimised participation in Dynamic Containment, Dynamic Moderation, Dynamic Regulation, and the FFR family — including the operating-envelope constraints each product places on adjacent half-hour windows.
Plain-English first; this section for anyone vetting the maths. Skip if you're not bidding the asset yourself.
Day-ahead, intraday, balancing, and FFR price paths are sampled from the production CEGridSight scenario tree — a 27-zone, 480-period MILP with HiGHS LP backend, per-hour markup calibration, and a LightGBM residual correction on 51 dispatch features, regenerated 6-hourly on production shadows. CE BESS Arbitrage consumes the scenario tree directly; there is no second model and no decoupled forecast drift.
First stage: day-ahead bid curves, fixed before clearing. Second stage: intraday and balancing recourse, indexed by scenario. Solved as a scenario-tree LP for the linear case (price-taker) and as a MIP when state-of-charge cycling caps or non-convex degradation costs bind. Currently HiGHS; CEMeridian (in build) is the in-house LP/MIP/QP backend that will replace it.
Round-trip efficiency, throughput-based degradation, depth-of-discharge weighting, and calendar fade — all configurable per cell chemistry (LFP, NMC, sodium-ion). Degradation enters the objective as a marginal cost per MWh-cycled, not a hard cap, so the optimiser is free to cycle harder when the spread justifies the wear.
Each day-ahead bid carries a published P10–P90 P&L band and a CVaR constraint configurable per portfolio. Daily ops report includes realised vs scenario-mean P&L, attribution by product (DA / ID / BM / FFR), and the binding constraint for every half-hour the asset was held back.
CE BESS Arbitrage launches Q3 2026. Early-access partners are paired with the engineering team during build for spec and validation review against their real assets — including a 30-day shadow-bid run before any capital is committed.
GB and 27 European bidding zones at launch
All four revenue stacks co-optimised in one model
Run alongside your incumbent before any switch
Drop-in replacement for an in-house or vendor optimiser. Your asset, your tenant, your scenario assumptions — we operate the model, you keep the data.
Multi-asset portfolio mode with per-asset attribution. Aggregator splits, manager fees, and asset-level reporting handled in the same workflow.
Read-only audit access to scenario assumptions, P10–P90 bands, and binding-constraint history. Make a debt or equity decision against a model you can interrogate, not a quarterly PDF.
Cohort one is sized to the bandwidth of the engineering team. Early-access partners get co-design on the bid format, free shadow operation through Q4 2026, and grandfathered pricing for the first contract year.