Schedule batteries in Japan before the day-ahead auction

To schedule a battery for the next day, an operator must commit the plan before the 10:00 auction. The gap between the most expensive and the cheapest four hours averages JPY 9.8/kWh in Tokyo and JPY 12.7/kWh in Hokkaido. The forecast determines how much of this spread the battery earns.

The case covers nine JEPX areas. Each area has its own price, so each battery receives the forecast for its area.

To compare forecasts on equal terms, the backtest schedules one 100 MW / 400 MWh battery on each forecast. The battery has 85% round-trip efficiency, one cycle per day, and a degradation cost of JPY 1.5/kWh.

The comparison includes a naive schedule that repeats the previous day and three rival foundation models that read price history only. The free public forecasts are the demand, wind, and solar forecasts published by OCCTO, which a gradient-boosting model converts into a price forecast.

eomer Expert adds three agents to the eomer forecasting model. Site Agent forecasts the output of the largest solar and wind plants, Grid Agent reads the published system forecasts, and Border Agent reads interconnector capacities and neighbouring markets.

eomer Expert captures 83% of the perfect-foresight profit, compared with 78% for the free public forecasts and 74% for the most profitable rival model. eomer Expert earns the most in all nine areas and in each of the 15 months.

The 80% uncertainty interval of eomer Expert contains 75% of outcomes, compared with 59% for the public-forecast benchmark. Relative to the naive schedule, eomer Expert adds JPY 254M per battery-year.

The schedule settles at the day-ahead price without intraday updates, so the estimate excludes intraday trading and balancing markets. A deployment must add the constraints and the market access of the actual asset.