The Use of Machine Learning in Extending Hub-Height Wind Speed

Reliable wind speed data at various heights above sea level are essential in the design, operation, and maintenance of offshore wind farms. While technologies such as Light Detection And Ranging (LiDAR) offer high-resolution vertical wind profiles, they are relatively recent and lack the long-term historical coverage provided by traditional surface buoys. Surface buoys have provided reliable, long-term measurements of oceanic and atmospheric parameters since the early 1970s. This paper investigates the use of advanced LiDAR measurements in conjunction with artificial neural networks (ANN), a machine learning (ML) technique, to develop a model that estimates the power exponent (α), and, thereby, vertical wind profiles. The model is trained using near-surface buoy data and LiDAR-derived wind profiles, enabling it to extrapolate wind speeds to hub-heights beyond the temporal limits of LiDAR observations. Once optimized, the model successfully extends hub-height wind speed time series using only surface buoy data in regions or periods where LiDAR data are unavailable or limited, such as remote areas, or during times when equipment is not operational due to maintenance or adverse weather conditions (for instance, foggy conditions). The resulting ML model, once trained with short-term LiDAR data, can reliably generate long-term hub-height wind speed time series using only buoy measurements, offering critical insights for offshore wind resource assessment and infrastructure planning.