About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Energy Technologies and CO2 Management: Resource Efficient Processes
|
| Presentation Title |
Weather-Informed Electricity Price Forecasting for Flexible Metal Production |
| Author(s) |
Mohammad Eissa, Omid Mahdavi, Aref Aasi, Adam Powell |
| On-Site Speaker (Planned) |
Mohammad Eissa |
| Abstract Scope |
Energy-intensive metal production is increasingly exposed to wholesale electricity prices that swing by an order of magnitude within a single day as grids shift toward wind and solar generation. Flexible producers that can schedule, throttle, or pause operations therefore require electricity-price forecasts that inform the timing of operating decisions. This study evaluates a method for predicting hourly ISO prices sixty days ahead using only information available when each forecast is issued, and applies it to ISO New England prices. Weekly FourCastNet 3 (NVIDIA AI) weather forecasts are combined with historical prices, fuel information, published load forecasts, and calendar features. Gradient-boosted models are evaluated chronologically against seasonal climatology and frozen persistence, and the method produces complete forecast horizons while verifying that later observations cannot leak into the inputs. Testing results show promise for capturing broad electricity-price trends over sixty-day horizons, but performance varies across evaluation periods, and reliable prediction of hourly fluctuations requires further validation. Future work will test new seasons, calibrate forecast uncertainty, and evaluate operating schedules under plant-specific constraints. Electricity-cost savings have not yet been demonstrated. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Environmental Effects, Sustainability, Machine Learning |