Wind Turbines: The Yaw Misalignment Challenge

Participants are expected to develop algorithms capable of reliably detecting yaw misalignment–related underperformance using 15-second high-frequency SCADA data. The submitted approaches should demonstrate the ability to distinguish yaw misalignment effects from normal operational variability and other non-yaw-related performance losses. In addition, participants should provide consistent and physically meaningful estimates of yaw misalignment angles. 

The proposed methods are expected to be robust across different turbines, operating conditions, and wind regimes, and to generalize beyond turbine-specific tuning. The results of the challenge should illustrate the added value of high-frequency SCADA data compared to traditional 10-minute SCADA analysis, highlighting improvements in detection sensitivity and angle estimation. Alongside quantitative performance, clarity of methodology, interpretability, and reproducibility are considered important components of the expected contributions.

Yaw misalignment occurs when a wind turbine rotor is not correctly aligned with the wind direction. Even small misalignment angles can lead to significant power losses and increased mechanical loads, making yaw misalignment one of the most impactful sources of underperformance in wind farms. 

Traditionally, wind turbine performance monitoring relies on standard SCADA data aggregated over 10-minute intervals. While suitable for high-level performance tracking, 10-minute SCADA often lacks the temporal resolution needed to reliably identify yaw misalignment, as control actions, wind direction variability, and transient effects are averaged out. 

In contrast, high-frequency SCADA data sampled at 15-second intervals preserves dynamic turbine behaviour, yaw controller response, and short-term power fluctuations. This higher temporal resolution enables more accurate identification of yaw-related underperformance patterns and allows for direct estimation of yaw misalignment angles. Nadara aims to leverage 15-second high-frequency SCADA data to develop robust, data-driven methods for detecting yaw misalignment that go beyond the limitations of traditional 10-minute analysis.

The dataset comprises approximately two years of 15-second SCADA data from four wind turbines at two wind farms operated by Nadara. Ground-truth yaw misalignment angles (per day basis), obtained from periodic LiDAR measurement campaigns are provided for a subset of turbines and time periods. 

The data is split into: 

• Unsupervised learning set (all four turbines, no labels provided)
• Labelled training set (one turbine, with yaw offset labels for known misalignment periods). Data is split into 30-min sessions.
• Held-out sets of 30-min sessions for validation (continuous leader board evaluation) and final test (to avoid hyper-parameter fitting)
• SCADA signals include generator speed, nacelle direction, wind speed, wind direction, pitch angle, rotor speed, and active power, all at 15-second resolution.

Working with nature, backed by shareholders, and with the support of local communities, Nadara is a NextGen Independent Power Producer (IPP+) that develops, owns and operates renewable energy sites across Europe and the US. Our technologies include over 4GW of installed onshore wind, solar photovoltaic, biomass and energy storage, and our IPP+ services encompass energy optimisation, trading, and flexibility solutions that maximise value across the energy chain, supporting energy security and creating long-term value that powers homes, businesses, and change.

• Time-series data analysis experience
• Optional: knowledge of wind energy systems