Correlating trends in operational data from wind turbines


The vision is to visualize and analyse the variations in energy production across a park of wind turbines by combining different sources of data. The aim is to give the wind turbine owner an insight into the historic and current performance of his wind turbine park, and thereby identify opportunities for future performance improvements.

Keywords: Big Data, Data mining, Real Time Data, Preventive maintenance, Wind Turbines, Renewable Energy, Economic performance.

Branche: Renewable Energy, Software development

Data set: Yes

Focus Area

When planning a new wind farm, the position of each wind turbine is optimized to provide a maximum energy production of the entire park. Factors like turbine design, weather conditions, relative position of the individual turbines and additional environmental conditions have a large impact on the performance of the individual turbines and therefore on the overall park performance driving the financial business case.

Verification and optimization of the realized wind farm energy production have significant economic impact and will drive the operation and maintenance of the wind farm over the lifetime (20 years). Data driven identification of trends in the park performance and / or identification of individual turbines deviating from the calculated energy production, will enable preventive maintenance decision. It can be repair scheduling, decisions on upgrading selected turbines, decisions on development specific performance improvements etc. all in all focusing on increasing the energy production from the site over the lifetime.


Based on historical turbine data, combined with other data sources, the CHALLENGE is to answer one or more of the following opportunities for improving the wind farm performance:

  • Wind farm power curve vs Individual power curve(s): Calculate and visualize the spatial variations in energy production across the park together with the wind speed and direction reported from publicly available weather data. Do the turbines influence each other?
  • Calibration of turbine direction vs wind direction: Compare and calibrate the reported yaw direction (the direction in which the turbine is pointing) relative to the wind direction reported from publicly available weather data. Visualize deviations.
  • Alarmlog patterning: Use the alarmlogs for the turbines to explain and visualize dropouts in the production (distinguish for example between wind farm faults and individual turbine faults)
  • Power curve benchmarking within the wind farm: Analyse the changes in energy production over time for all turbines and spot possible outliers, indicating turbine degradation. Visualize the trends.
  • Power curve benchmarking with similar wind farms: Compare the park performance to other parks in Denmark and visualize the variations


The operational data quality will not always be optimal. Data can thus contain noise from failing sensors, dropouts in the data stream, resetting of parameters due to software updates.

Furthermore, the additional data sources are not always reliable, might need calibration and/or synchronization of timestamps, coordinate systems etc.


Some possible data sources are:

From the turbines:

  • Historical scada data from a single wind farm consisting of 125 individual turbines (10-min sampled data). Available data tags are
    o Wind speed and wind direction measured by each turbine
    o Yaw direction
    o Ambient temperature measured by each turbine
    o Power production from each turbine
  • Transition log of alarms for each turbine
  • GPS position of each turbine in the windfarm

Other sources: