As wind energy becomes increasingly more prevalent, having accurate wind-power forecasts becomes important in order to help energy operators make better grid integration and economic decisions. Since a majority of energy trading is done 24 hours in advance, most of the wind-power forecasts that are utilized by operators today are aimed at reducing the uncertainty in wind-power production 24-48 hours in advance. Although this is an important component, these forecasts do not address short-term issues such as sudden wind-power ramps.

Unforeseen wind-power ramps caused by a sudden increase in wind-speed over a farm can produce excess energy that cannot be used because energy has already been allocated from other sources. If operators have some indication that a wind ramp is approaching in the next few hours, they can turn off other sources of energy in order to fully utilize the wind-power produced by the ramp. Therefore, short-term forecasts (0-6 hours out) specifically designed to forecast wind ramps are becoming increasingly more important.

Forecasting wind-power ramps at a specific location and time is a very challenging problem. Currently, most approaches rely on high-resolution models such as the Weather Research and Forecasting (WRF) model to address this issue. High resolution modeling has shown some success at forecasting ramps, but often features are misplaced in time and space. Therefore other techniques may be needed to better forecast wind ramps. In conjunction with the Xcel Energy wind-power forecasting project at NCAR, research and development efforts have been performed to determine if publicly available upstream observations could be used to predict when a ramp is likely to impact a wind farm. This paper discusses an experimental observation-based wind ramp forecasting expert system configured at NCAR for one of the Xcel wind farms.

The application uses nearby observations to predict when a wind ramp (or strong wind event) will occur at the farm within a 1-hour time window out to 6 hours in the future. The algorithm uses publicly available observing sites (METARs and mesonet sites) and searches for wind ramp signatures in upstream observations. A preliminary application has been configured to predict wind-power ramps at a wind farm in northeastern Colorado for events originating from the north, northwest and northeast. This paper discusses the techniques used to utilize upstream surface observations to predict wind-power ramps at the farm. The algorithm's performance is examined for a number of cases. Lastly, recommendations are made about future work.