Improving the Performance of Mass-Consistent Numerical Models Using Optimization Techniques

View More View Less
  • a Pacific Northwest Laboratory, Richland, WA 99352
  • | b FloWind Corporation, Pleasanton, CA 94566
© Get Permissions
Full access

Abstract

This paper describes a technique of using a mass-consistent model to derive wind speeds over a microscale region (about 4 km2) of complex terrain. A serious limitation of these numerical models is that the calculated wind field is highly sensitive to certain input parameters, such as that used to simulate the atmospheric stability. Because accurate values for these parameters are not usually known, confidence in the calculated winds is low.

However, values for these parameters can be found by tuning the model to existing wind observations within a microscale area. This tuning is accomplished with an optimization procedure that adjusts the unknown parameters so that the discrepancy between the observed winds and model calculations of these winds is minimized.

The model was verified with eight sets of hourly averaged wind data. These data were obtained from measurements made at 28 sites covering a windfarm development in the Altamont Pass area of California. When the model was tuned to a small subset of the 28 sites, the model showed skill in predicting wind speeds for the remaining sites in six of the eight cases. The two that did not perform as well were low wind cases.

Abstract

This paper describes a technique of using a mass-consistent model to derive wind speeds over a microscale region (about 4 km2) of complex terrain. A serious limitation of these numerical models is that the calculated wind field is highly sensitive to certain input parameters, such as that used to simulate the atmospheric stability. Because accurate values for these parameters are not usually known, confidence in the calculated winds is low.

However, values for these parameters can be found by tuning the model to existing wind observations within a microscale area. This tuning is accomplished with an optimization procedure that adjusts the unknown parameters so that the discrepancy between the observed winds and model calculations of these winds is minimized.

The model was verified with eight sets of hourly averaged wind data. These data were obtained from measurements made at 28 sites covering a windfarm development in the Altamont Pass area of California. When the model was tuned to a small subset of the 28 sites, the model showed skill in predicting wind speeds for the remaining sites in six of the eight cases. The two that did not perform as well were low wind cases.

Save