How ERCOT Forecasts Load
ERCOT uses neural networks and linear regression trained on 49 weather stations across Texas. These models treat each weather zone as a uniform atmospheric environment — they cannot see urban thermal mass, the heat stored in millions of square meters of asphalt, concrete, and rooftops throughout the day and discharged back into the atmosphere at night.
What Happens When They Overforecast
Grid operators commit too many generation resources day-ahead. Peaker plants — almost always natural gas — are dispatched unnecessarily. Every unnecessary peaker dispatch emits CO2, costs ratepayers money, and degrades air quality in communities near those plants. In Houston, those communities are disproportionately low-income and already carry the highest industrial pollution burden in the country.
What Happens When They Underforecast
Grid operators are caught short during peak demand. Real-time prices spike — ERCOT real-time prices have exceeded $1,400/MWh during forecast misses. Operators scramble for emergency reserves. In the worst cases, controlled outages follow. The May 2025 Memorial Day event in MISO South forced 100,000+ Louisiana customers offline. Underforecasting during peak summer afternoon hours — exactly when urban thermal discharge is highest — is the most dangerous failure mode.
Who Gets Affected
- Ratepayers: pay higher electricity bills when forecast error drives inefficient dispatch
- Industrial customers: face real-time price spikes during undersupply events
- Environmental justice communities: bear the health cost of unnecessary peaker emissions
- Energy traders: lose money on positions built around inaccurate day-ahead forecasts
- Grid operators: face reliability risk and regulatory scrutiny after major miss events
What TormentaAI Does Differently
TormentaAI is a physics-based load forecasting platform that models the thermal behavior of urban environments — the processes that weather-station-based models cannot see. In our first validated backtest against ERCOT actuals, TormentaAI achieved 4.32% MAE on 19 holdout hours. Our target accuracy of 0.5% MAPE represents a 264 MW average improvement over ERCOT's current day-ahead performance. The methodology is proprietary.
The Scale of the Problem
ERCOT reports a system-wide day-ahead MAPE of ~2.7%, translating to 300+ MW of average forecast error per hour in the Houston Coast Zone alone. Based on EPA natural gas emissions factors and conservative peaker attribution, systematic forecast error drives an estimated 300,000+ metric tons of excess CO2 annually in the Houston zone.