J2.5 Lessons Learned Implementing a Real-time Electrical / Gas Load Forecasting System

Tuesday, 12 January 2016: 4:30 PM
Room 354 ( New Orleans Ernest N. Morial Convention Center)
Thomas Brummet, NCAR, Boulder, CO; and J. Williams, S. Dettling, and G. Weiner

Accurate predictions of a utility's electrical and gas load are important for that utility's energy production, distribution, and trading operations. Net load is strongly influenced by weather variables in addition to human usage patterns. The National Center for Atmospheric Research (NCAR) in partnership with Xcel Energy has developed a forecast system that leverages pre-processed weather observations and forecasts along with historical load data to provide short-range net electrical load forecasts across varying targets.

The current load forecast system that NCAR has developed for Xcel Energy utilizes load and transmission data, provided by Xcel, along with observed and forecast weather to predict future load and transmission usage across three separate regions. Historical load observations are combined with enhanced weather data and fed to a data mining package, called cubist, to produce a forecasting model. This model is updated on a weekly basis with recent weather and load observations. Each model is individually tuned to the specific region and target, such as gas load or electric load. Different targets require different input variables in order to maximize performance. The forecasts go out 168 hours, but we have focused on maximizing the performance in the 24-48 hour time frame.

The load forecasting system is unique in that it is driven not only by weather, but also human behavior. This poster describes the forecast system architecture and some of the challenges that we faced when developing and enhancing the load forecast. These challenges include, but are not limited to, cubist parameter tuning, variable selection, efforts to enhance input variables, and most importantly quality control of historic load data. Lastly the poster discusses lessons learned and guidance for future load forecasting.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner