Abstract
In recent years, wind turbine yaw misalignment that tends to degrade the
turbine power production and impact the blade fatigue loads raises more
attention along with the rapid development of large-scale wind turbines. The
state-of-the-art correction methods require additional instruments such as
LiDAR to provide the ground truths and are not suitable for long-term operation
and large-scale implementation due to the high costs. In the present study, we
propose a framework that enables the effective and efficient detection and
correction of static and dynamic yaw errors by using only turbine SCADA data,
suitable for a low-cost regular inspection for large-scale wind farms in
onshore, coastal, and offshore sites. This framework includes a short-period
data collection of the turbine operating under multiple static yaw errors, a
data mining correction for the static yaw error, and ultra-short-term dynamic
yaw error forecasts with machine learning algorithms. Three regression
algorithms, i.e., linear, support vector machine, and random forest, and a
hybrid model based on the average prediction of the three, have been tested for
dynamic yaw error prediction and compared using the field measurement data from
a 2.5 MW turbine. For the data collected in the present study, the hybrid
method shows the best performance and can reduce total yaw error by up to 85%
(on average of 71%) compared to the cases without static and dynamic yaw error
corrections. In addition, we have tested the transferability of the proposed
method in the application of detecting other static and dynamic yaw errors.