Case article
From input to implication
01
Decision context
To schedule a battery for the next day, an operator must forecast the real-time price at the battery location. ERCOT clears a separate price at each load zone, so each location needs its own forecast.
At LZ_WEST, prices average USD 17/MWh from 09:00 to 15:00 and USD 62/MWh during the evening peak. Each load zone records 111 to 219 hours above USD 200/MWh, with peaks near USD 3,000/MWh.
02
Method
To compare forecasts on equal terms, the backtest schedules one 100 MW / 400 MWh battery per load zone on each forecast. The battery has 85% round-trip efficiency, one cycle per day, and a degradation cost of USD 10/MWh.
The schedule is fixed at 09:00 Central on the day before delivery and settles at the local real-time price. The comparison includes a naive schedule that repeats the latest complete day and a price-only model.
eomer adds Site Agent, which forecasts the output of the 100 largest solar parks and the 100 largest wind farms in ERCOT.
03
Finding and implication
eomer captures 60% to 70% of perfect-foresight value across the four load zones. At LZ_WEST, revenue exceeds the naive schedule by 94%. Site Agent adds USD 324k to 441k per battery-year relative to eomer alone.
Over a six-day horizon, eomer records 11.8% lower mean absolute error than the most accurate price-only rival model. A pilot must evaluate the battery node, where congestion and losses can change settlement prices.

