Bringing data science to wind energy, driving the energy transition forward

Data science can be a major asset in making renewable energy sources an environmentally friendly, low-cost and therefore more attractive alternative to fossil energy sources. Data science and data analysis are becoming increasingly relevant for improving day-to-day operations in the renewable energy sector. For example, they can be a very effective approach in optimizing your daily wind farm operations.

Data science methods can improve decision-making for wind energy applications, including wind field analysis and forecasting, power curve fitting and optimizing wind turbine and wind farm maintenance operations (predictive maintenance).

Data-driven decision-making

Data-driven decision-making covers the entire data analysis process, which begins with collection. After that, it’s necessary to clean the data and select the subset with the relevant information. Once the data are prepared for use, an exploratory analysis, including visualization, can help you decide which methods or models are most effective in obtaining the desired knowledge. After choosing the appropriate method, a model will be built and validated. The final process will lead to a set of results (e.g., predictive model) that guides the decision-making, which again, may rely on visualization. It’s worth noting that data collection and preparation are very important steps that should not be neglected during the data analysis process.

By failing to prepare, you are preparing to fail.” – Benjamin Franklin

In fact, during data collection, you should decide what data may be more valuable to collect.

The type of data to collect depends on what you want to analyze, for example:

• Performance and availability analysis: operational data, measurement data, equipment data
• Root cause analysis: failure data
• Maintenance optimization and degradation monitoring: maintenance and inspection history

During the data pre-processing phase, you can clean, validate and consolidate your raw data to prepare them for the next steps. For instance, you may delete the missing data, remove outliers, remove duplicate data and other abnormal operating condition data points, normalize and format data and separate training from validation sets.

Applying AI techniques in the wind energy sector

Artificial intelligence (AI) mostly refers to machine learning and deep learning. There are at least three areas in which AI algorithms are widely applied to wind power generation:

1. Wind speed and power prediction

Since wind power is highly dependent on wind speed variation, an accurate wind speed prediction model is crucial to having full control over energy production and consumption. The most popular models used in AI-based wind speed and power predictions include artificial neural network (ANN), support vector machines (SVM), and linear and logistic regression. In practice, we can use either supervised learning, where input-labelled and output-labelled data are provided, unsupervised learning, where only input data are provided, or semi-supervised learning, which is a mix of supervised and unsupervised learning (in case we don’t have enough labelled data to train the models).

2. Operation and maintenance (O&M) optimization

Another interesting AI application in the field of wind energy is through operation and asset maintenance (O&M). Thanks to AI and, more specifically, machine learning algorithms, it is possible to accurately assess potential component downtime. Several companies in the wind industry invest considerable amounts of money in maintaining and ensuring the proper functioning of their machines. Unexpected failures in their operations can lead to considerable financial losses. Using historical data to accurately predict component failures before they occur is an effective tool for wind farm operators in reducing the cost and complexity of operation and maintenance. Many machine-learning algorithms could be used to implement a fault-detection system to identify any upcoming faults, for example, normal behaviour models (NBM) that use historical data to learn what the turbine’s normal operating condition should be. By using historical supervisory control and data acquisition (SCADA) data (e.g., wind speed and power, generator temperature, turbulence intensity, rotor speed and wind direction), the model makes a prediction of what the normal behaviour of the wind turbine or its components should be at a given time, and this prediction is compared to the actual measured value. A major difference means that the measured value is outside the normal operating range and that a potential fault may occur.

3. Optimization of wind farm operations

Ensuring wind turbines perform optimally over their lifetime (typically 20–25 years) represents a large portion of the installation cost. Machine learning models can be used to detect blade faults, monitor generator temperature and modeling power curve and much more. It has the potential to help wind farm operators make smarter, faster and more data-driven assessments of how their power generation can meet electricity demand. Moreover, accurate prediction of wind speed, power, and electricity supply and demand can increase the value of wind energy while reducing operating costs.

Conclusion

Data science has the potential to improve green energy production. We can reliably and cost-effectively produce carbon-free energy if we have the right data science tools and datasets.

Our experts at BBA are available to assist you with your project to maximize your profits and reduce your costs. Get in touch with us.

References

1. IRENA (International Renewable Energy Agency), https://www.irena.org/wind