Data-driven calibration for cup anemometer at different measurement locations around buildings using transfer learning based on domain adaptation

Abstract

In recent years, high-precision sensors, e.g., ultrasonic anemometers, have been widely used for wind measurement. However, conventional sensors, e.g., cup anemometers, are yet to be replaced owing to their low-cost advantages and high robustness in an uncertain environment. Considering that data-driven calibration methods are used to improve the measurement accuracy of cup anemometers, this study proposed a transfer learning method based on domain adaptation, so that existing measurement data can be used for model training in new measurement scenarios, thus reducing the cost of secondary data collection. In summary, at the corner and side of a building in a new measurement site considered for experiments, the results of the proposed method indicated that the relative errors of pulsation parameters, e.g., the standard deviation wind speed, turbulence intensity, and gust factor, more significantly reduced compared to the conventional machine learning method.