Machine Learning Based Turbulence Intensity Retrievals from Buoy Based Observations

Precise wind and turbulence intensity (TI) measurements are critical for the successful development and operation of offshore wind farms. Floating buoy lidars offer a cost-effective alternative to traditional meteorological masts for collecting hub-height measurements offshore. However, buoy motion significantly impacts the reliability of TI data derived from Doppler lidars. This study addresses this challenge by employing advanced machine learning techniques to correct motion-induced errors in TI estimates across the entire rotor swept area. In this study, the resulting neural network model achieved an R2 of approximately 98% for horizontal wind speeds at a temporal frequency of 1 Hz, and an R2 of approximately 90% for 10-min turbulence intensity estimates compared to a reference Doppler lidar near the Air-Sea Interaction Tower. Moreover, we successfully tested the generalizability of the ML-based TI model, by identifying consistent TI trends against buoy motion observed in the training data, at a different geographic location near the U.S. West Coast. The generalizability underscores its potential widespread implementation in the offshore wind industry.