4.1 Impact of Targeted Measurements and Next-Generation Prediction Techniques on Short-Term Wind Ramp Forecasting in the Tehachapi Wind Resource Area

Tuesday, 24 January 2017: 1:30 PM
606 (Washington State Convention Center )
John Zack, AWS Truepower LLC, Albany, NY; and C. P. van Dam, S. H. Chen, C. Y. Chen, and C. P. MacDonald

The management of wind power variability is a key factor in minimizing the cost of integrating wind-based power generation into electric grid systems while maintaining the required very high level of reliability. Short-term wind power forecasting is one of the most cost-effective and easily implemented tools available to system operators.  However, many North American system operators have a need for more accurate intra-day prediction of events characterized by large changes in wind-based power production over short periods (i.e. “wind ramps”). The California Independent System Operator (CAISO) has indicated that the prediction of such events in the Tehachapi Wind Resource Area (TWRA) of California would have great value due to the large amount of wind generation capacity that is concentrated in the TWRA. 

The California Energy Commission (CEC) has provided support to a diverse team lead by the University of California, Davis for a 2-year (2015-2017) project to improve the accuracy of TWRA wind ramp forecasts through improved meteorological measurements and refinements to physics-based and statistical atmospheric prediction models.  The major components of the project are: (1) the deployment of a targeted regional sensor network of remote sensing systems such as wind profiling radars, sodars, and microwave radiometers; (2) a forecast sensitivity analysis to determine which components of the forecast system have the greatest impact on forecast performance; (3) improvements to the atmospheric boundary layer and earth’s surface sub-models of Numerical Weather Prediction (NWP) models to account for processes typically not well-modeled in existing sub-models; (4) improvements to the NWP data assimilation schemes to account for flow dependent variations in the optimal representation of the spatial influence of measurement data in the forecast model initialization; (5) custom configuration of an ensemble of advanced machine learning methods for application to the very short-term (0-3 hours) prediction of ramp events from time series data from forecast sites and the targeted sensor network; (6) a 1-year evaluation of the impact of these improvements on 0-15 hour wind ramp forecast performance relative to a state-of-the-art baseline prediction system. 

This presentation will be a follow-up to the early stage overview of the project given at last year’s conference.  It will provide an overview of (1) the TWRA and the impact of wind variability in this region on CAISO operations, (2) a description of the specific forecast system improvements implemented in this project, (3) results from an analysis of the impact of each improvement on wind forecast performance and a comparison to the performance of reference forecasts based only on output from operational prediction models from the US National Weather Service, and (4) a preliminary estimate of the anticipated benefits that the forecast improvements will provide to system operations based on feedback from CAISO.

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