Data-driven calibration of cup anemometer based on field measurements and artificial neural network for wind measurement around buildings

Abstract

In recent years, high-precision sensors such as ultrasonic anemometer have been commonly used for wind measurement. However, conventional sensors such as cup anemometer is yet to be completely replaced owing to its advantages of low cost and high robustness. Thus, to improve the measurement accuracy of cup anemometer in practical applications, this paper proposes a data-driven calibration strategy based on field measurements. Regarding the data sources, two datasets were acquired at multiple locations in the vicinity of the reference buildings to construct a calibration (regression) model. For model development, we proposed artificial neural network (ANN)-based models with cup anemometer measurements as input and ultrasonic anemometer measurements as output. Compared with the baseline models (Kristensen's models and multiple linear regression models) on several performance metrics, the results demonstrated that ANN models with nonlinear computational ability improved the calibration of the measurements. At a location situated on the open space (building side), the ANN models reduced the mean of the relative errors of mean wind speed, maximum instantaneous wind speed, standard deviation, turbulence intensity, and gust factor by approximately 9% (9%), 5% (3%), 2% (10%), 13% (23%), and 6% (12%), respectively, in comparison to the noncalibrated cup anemometer measurements.