Loading
Skip to content
Miracle Minds AI
  • Try “HIPAA”, “model risk”, “FinOps” or “agentic”.

Renewable integration and trading support

A renewable or storage forecast is worth what it changes in a bid, a battery's state of charge going into the evening ramp, or a curtailment decision. Models tuned for average error lose money in the tails, and once a model shapes your offers it is part of your market conduct record.

Part of AI for Utilities & Energy. 7 sources, checked .

Generation and storage capacity actively seeking U.S. transmission interconnection at the end of 2025 (LBNL Queued Up)
2,061 GW
Generation and storage capacity actively seeking U.S. transmission interconnection at the end of 2025 (LBNL Queued Up)
Lawrence Berkeley National Laboratory, 2026(opens in a new tab)
Share of planned 2026 U.S. utility-scale capacity additions that are solar (51%) or battery storage (28%)
79%
Share of planned 2026 U.S. utility-scale capacity additions that are solar (51%) or battery storage (28%)
U.S. Energy Information Administration, 2026(opens in a new tab)
Utility-scale wind and solar output curtailed by CAISO in 2024, up 29% on 2023
3.4M MWh
Utility-scale wind and solar output curtailed by CAISO in 2024, up 29% on 2023
U.S. Energy Information Administration, 2025(opens in a new tab)

The problem

The resource mix is shifting faster than operating practice. Developers plan 86 GW of new U.S. utility-scale capacity in 2026, a record if realized, with solar at 51% and battery storage at 28% of the total.1 Behind that sits a queue of about 2,061 GW of generation and storage seeking transmission interconnection at the end of 2025, and the median project built in 2025 took 61 months from request to commercial operation.2

NERC notes that most new resources due in the next five years are battery storage and solar, inverter-based and weather-dependent resources that make the grid more complex to plan and operate.3 In ERCOT outside winter, reserve scarcity risk is highest in the early evening, when solar output falls while load stays high, and battery storage there is expected to reach 18.9 GW by summer 2026.3 Timing the battery fleet around that ramp is a forecasting problem.

Oversupply is already costing energy. CAISO curtailed 3.4 million MWh of utility-scale wind and solar output in 2024, 29% more than in 2023, and solar made up 93% of it, with the most solar curtailed in spring.4 DOE's 2024 AI for Energy report says operators need significantly more accurate weather forecasts as variable generation grows and that AI can bring higher-resolution modeling to matching load and supply.5

Trading adds a conduct layer. FERC prohibits any scheme to defraud or any deceptive practice in connection with jurisdictional electricity sales,6 and its market behavior rules require sellers to bid and declare outages under market rules, give accurate information to RTOs and market monitors, and keep five years of the data behind their prices.7 An optimizer that prepares bids inherits those obligations.

What we build

  1. 01

    Probabilistic plant-level forecasts

    Solar and wind forecasts by plant and hour, refreshed intraday and published as quantiles, so day-ahead offers, reserve holdings and hedges can be sized to the risk and not just the median.

  2. 02

    Battery dispatch optimization

    Charge and discharge schedules set against price and net-load forecasts, with state-of-charge limits, cycle counts and warranty terms built in as hard constraints.

  3. 03

    Curtailment and congestion prediction

    Models that flag hours likely to see curtailment or negative prices at your nodes, so you can shift storage charging, maintenance windows and contract positions ahead of time.

  4. 04

    Bid preparation with a full audit trail

    An assistant that drafts offers within limits your traders set, and records the inputs, model version, approver and rationale behind each submitted bid.

  5. 05

    Interconnection study triage

    Extraction and comparison of study results, upgrade cost allocations and milestone dates across queue documents, so development teams see which positions are worth carrying.

Where it goes wrong

The failure modes we design against from the first week, and the control for each.

  • Optimizing the average and losing in the tails

    Mean error rewards a forecast that is right on ordinary days and wrong on ramps and storms, which is where imbalance costs sit. Score forecasts with quantile metrics and on the costliest hours, and report ramp error separately.

  • Bids nobody can explain

    An algorithm does not dilute your obligations under FERC's anti-manipulation and market behavior rules.6,7 Set hard position and price limits, require compliance review before strategy changes go live, and keep every input and output for at least the five years the rules require for price data.

  • Battery models that ignore the asset

    An optimizer that cycles a battery against price alone can breach warranty limits and shorten its life. Encode manufacturer cycle and state-of-charge limits as constraints, and reconcile planned against actual state of charge every day.

  • Training on curtailed output

    CAISO alone curtailed 3.4 million MWh of wind and solar in 2024.4 In plant data, curtailed hours look like low wind or sun, so a model trained on raw output learns the wrong resource. Label curtailment and outage periods before training, monitor drift as plants are added, and retrain on a fixed schedule.

Rules that apply

  • 18 CFR 1c.2 (Prohibition of electric energy market manipulation)

    Primary text on ecfr.gov(opens in a new tab)

    Prohibits any device, scheme or artifice to defraud, any material misstatement, and any deceptive practice in connection with FERC-jurisdictional electricity or transmission sales, including bids an algorithm prepares.

  • 18 CFR 35.41 (Market behavior rules)

    Primary text on ecfr.gov(opens in a new tab)

    Requires sellers in organized markets to bid, schedule and declare outages under market rules, give accurate information to RTOs and market monitors, and retain the data behind their prices for five years.

How to measure it

  • Forecast skill against the incumbent forecast by horizon, with quantile (pinball) loss
  • Imbalance or deviation charges per MWh delivered
  • Captured storage spread per MWh discharged, against a perfect-foresight benchmark
  • Curtailed MWh and hours of negative-price exposure at your nodes
  • Share of submitted bids with a complete audit record of inputs, model version and approver

Questions buyers ask

Is AI-generated bidding allowed in wholesale power markets?

Check your market's tariff first, then ask who answers for each bid. FERC's anti-manipulation rule and market behavior rules apply to every offer you submit, whoever or whatever drafted it. Keep traders in control of limits and strategy, log the inputs and logic behind each bid, and retain that record so you can explain any offer to a market monitor.

How do you measure whether a renewable forecast is worth paying for?

Price the errors. Run the new forecast beside your current one, then calculate what each would have cost in imbalance charges, reserve purchases and missed storage spread over the same period. A forecast with slightly worse average error can still be worth more if it is better on ramps and extreme days, which is where most of the cost sits.

Can AI shorten interconnection queue timelines?

Not directly. Transmission providers run the studies and set the schedule. AI helps developers and utilities read and compare study results, track milestones and upgrade cost allocations across many positions, and decide earlier which projects to carry or withdraw. With multi-year waits from request to operation, choosing the right positions sooner is where the time is saved.

Sources

Every figure on this page comes from the sources below. We checked each link on .

  1. 1.New U.S. electric generating capacity expected to reach a record high in 2026(opens in a new tab)U.S. Energy Information Administration, February 20, 2026 · eia.gov
  2. 2.Queued Up: 2026 Edition, Characteristics of Power Plants Seeking Transmission Interconnection As of the End of 2025(opens in a new tab)Lawrence Berkeley National Laboratory, June 2026 · eta-publications.lbl.gov
  3. 3.2025 Long-Term Reliability Assessment(opens in a new tab)North American Electric Reliability Corporation (NERC), January 2026 · nerc.com
  4. 4.Solar and wind power curtailments are increasing in California(opens in a new tab)U.S. Energy Information Administration, May 28, 2025 · eia.gov
  5. 5.AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy(opens in a new tab)U.S. Department of Energy, April 2024 · energy.gov
  6. 6.18 CFR 1c.2: Prohibition of electric energy market manipulation(opens in a new tab)Electronic Code of Federal Regulations · ecfr.gov
  7. 7.18 CFR 35.41: Market behavior rules(opens in a new tab)Electronic Code of Federal Regulations · ecfr.gov