Renewable forecasts and what they are worth in the market
Wind power, photovoltaic power and solar irradiance forecasting appear on 52 recent papers, almost all in AI journals, and only 8 of them carry a market, bidding, price, imbalance or trading keyword. Forecasts are judged by error metrics, but wind and solar operators pay for forecast errors in imbalance charges. Which forecast improvements pay, and which do not?
What prompted this
Since October 2025, 52 papers carry a wind-power, photovoltaic-power, solar-power or renewable-energy forecasting keyword (46 in AI journals, 5 in economics): "wind power forecasting" on 22, "photovoltaic power forecasting" on 13, "wind speed forecasting" on 11, "solar irradiance forecasting" on 5, "renewable energy forecasting" on 6. Only 8 of the 52 also carry a market, bidding, price, imbalance, trading, economics, cost or dispatch keyword.
Questions
- For a wind or solar plant selling into a day-ahead market with imbalance charges, how much revenue does a given reduction in forecast error add, and does the value depend on the direction of errors?
- Do the deep learning forecasts in recent papers improve revenue relative to standard ensemble weather forecasts, or only improve metrics that do not map to revenue?
- Does better forecasting by many plants reduce system balancing costs, or shift them?
Data that could help
Plant-level production and forecast data, day-ahead and imbalance prices (ENTSO-E, Nord Pool), and the forecast models of published papers.
Next step
Compute the imbalance cost of forecast errors for a set of wind plants under standard and published deep learning forecasts using one year of market prices.