Evaluation of wind ramp forecasts from an initial version of a rapid update dynamical-statistical ramp prediction system

As the fraction of electricity generated from wind turbines on a grid system (i.e. the wind penetration) increases, the potential for large changes in wind power production over short time periods (known as wind ramps) to have a significant adverse impact on electric system stability also increases. This issue may ultimately limit the amount of wind power production on a grid system unless effective strategies to mitigate the occurrence or impact of extreme variability events can be developed and implemented. A number of approaches can be used to address this issue including the reduction in wind power production variability by spreading the location of wind power production facilities on a grid system over a number of wind regimes that have a relatively low correlation of wind variability. This will reduce the number and magnitude of significant wind ramp events since the changes in wind power production at individual sites will tend to offset each other rather than combine to generate large system-wide or regional ramp events. However, this approach requires long-term planning and cooperation from a number of commercial and government entities. Another approach is to develop forecasting tools that provide grid operators with information to anticipate ramp events. This would enable operators to manage other generation resources or even the wind farms themselves (for example by limiting production if the large generation changes were in an upward direction) to compensate for the wind variability in a timely and economic manner to minimize the risk to electric system stability.

The amount of wind generation capacity on the grid system operated by the Electric Reliability Council of Texas (ERCOT) is currently over 9000 MW. At this level of penetration, large wind ramp events can sometimes pose a significant challenge for grid management. AWS Truepower and MESO, Inc, have developed a customized short-term wind ramp forecast system to provide information to help ERCOT anticipate wimp ramp events over a 0 to 6 hour look-ahead period. The system is known as the ERCOT Large Ramp Alert System (ELRAS).

The ELRAS forecasts are based on a high-resolution rapid update numerical weather prediction (NWP) model and a statistical regime-based model output statistics (MOS) module. The rapid update NWP component is based on the Advanced Regional Prediction System (ARPS) model. The ARPS simulations have a 2-hour update frequency and a horizontal resolution of 6 km. These are used to produce deterministic high frequency forecasts of the flow within the layer from 50 m to 100 m above ground level (i.e. the turbine rotor levels). The system employs the ARPS three-dimensional variational (3DVAR) scheme to assimilate data from a variety of sources including surface mesonet data, wind profilers, Doppler radars and meteorological observations from the wind farms themselves. The output from the ARPS rapid update simulations, the latest time series data of power production and meteorological variables and the output of feature detection algorithms that operate on radar reflectivity data and high spatial and temporal resolution wind analyses are input into a regime-based statistical algorithm. The statistical scheme generates several types of wind ramp forecast information.

The prediction system produces three types of forecast products with a time resolution of 15 minutes for a look-ahead period of 6 hours: (1) probabilistic ramp rate forecasts for multiple threshold values for over three ramp-rate time periods; (2) deterministic ramp event forecasts; and (3) deterministic forecasts of the power production time series along with probabilistic confidence bands. The forecasts are updated every 15 minutes and are provided for the aggregated wind power production for several regions of the ERCOT system as well as the system-wide wind power production.

The ELRAS system began operation in an experimental mode in November 2009. Since that time there has been an ongoing performance evaluation of all three forecast types. The probabilistic ramp rate forecasts are being evaluated through the use of the Ranked Probability Skill Score (RPSS) metric relative to the climatological ramp rate probabilities. This provides a combined measure of the three critical performance attributes of a probabilistic forecast: reliability, resolution and sharpness. The deterministic ramp event forecasts are evaluated with the Critical Success Index (CSI) metric and are benchmarked against a random forecast of ramp events from the climatological probability distribution of these events. The deterministic time series forecasts are evaluated with the standard metrics of Bias, Mean Absolute Error (MAE), Root Mean Square Error (RMSE) while the associated confidence bands are evaluated for reliability.

The conference presentation will include an overview of the forecast system architecture, a few case examples of forecast performance and an analysis of the statistical performance of all three types of forecasts for the spring, summer and fall of 2010. The analysis will include a breakdown of performance for seasonal and ramp-related weather-regime (e.g. convectively active, prominent low level jet etc.) subsamples.