Can big data and artificial intelligence transform the wind sector?

September 27, 2017

Big data and advances in artificial intelligence hold extraordinary potential for transforming the energy sector. But is change in the wind sector likely to be transformational, or can we expect only incremental performance improvement?


For utility-scale wind farms, the benefits of big data technologies may not be on the scale of the transformations occurring in distributed generation, energy storage and demand management – but there is no doubt that there is potential for incremental improvements in the efficiency and cost of wind turbines, improved energy forecasting, and better fault prediction and detection – and that means opportunities for competitive advantage.

Big data, machine learning and AI

Big data is the result of having more data sources and more storage. We are now flooded with data of all types and we have the ability to store it, but the key challenge is making that data useful so that it can create value. Machine learning deals with computer models that apply algorithms allowing them to learn from experience, modify their internal rules and make better predictions. When you put the two concepts together, big data enables machine learning to act like an artificial intelligence (AI), creating systems that mimic cognitive functions such as learning or problem-solving.

Underpinning all this are computer models, which translate input data into outputs or predictions. For a wind farm, input data can include wind speed, turbine output, climatic conditions, error messages and condition monitoring data. The outputs might be improved turbine siting, predicted component failure, or predicted wind farm generation one hour in the future. Let’s have a look at some examples.

How can big data and AI help wind developers and operators now?

Plenty of research is underway into how big data and AI can optimise wind generation. Increased yield, more accurate forecasting, and more effective operations and maintenance are just some of the possibilities.

Increased yield

Greater yield means more generation and more power into the grid. The improvements in yield can come from improving the estimated power curve, and tailoring the operation of each turbine to suit the conditions.

A number of turbine manufacturers are developing and offering individual turbine tuning. This involves analysing the historical weather conditions and resultant production, and setting up specific control settings for different conditions. A more complex approach could allow a system to autonomously vary the turbine settings during the operation of the wind farm and learn the effect of those adjustments on production, retaining them for future use.

The verdict: Incremental yield improvements in the order of 1–5%.

More accurate forecasting

To balance energy supply and demand in the grid, as well as to maximise returns on the spot market, it’s critical to have accurate forecasts of a wind farm’s production.

Weather forecasting is a very popular area for researchers, and it seems that the entire range of machine learning algorithms has been applied in attempts to demonstrate superiority. Generally, in each case studied, researchers are able to show that machine learning methods improve upon the accuracy of classical prediction models. However, models and assessments all depend on the site, and no clear recommendation of one model can be made across the board. The best-performing approaches incorporate multiple numerical weather prediction forecasts, combined with live site data for wind, power and atmospheric conditions, and commonly use neural network methods. 

The verdict: Incremental improvements in forecast accuracy, as meteorological agencies continue to incorporate machine learning components into weather models.

More effective operations and maintenance

More effective operations and maintenance (O&M) plans can significantly reduce a wind farm’s ongoing costs and increase its output and operational life. Because O&M are major expenses with all industrial machinery, there’s a lot of work being done in predictive maintenance and fault detection.

One opportunity is using SCADA data to detect faults. The approach is to train a model during normal operating conditions to predict a measurement, for example, bearing temperatures. Then input the current operating conditions and flag where the observations don’t match the prediction. This can allow abnormal behaviour to be picked up early, before a major component fails. 

Wear and tear on each turbine’s individual components can be tracked using the specific conditions to which the turbine on site has been subjected. Components that need replacement or can be safely left in service can be identified by collecting the full operating history of individual turbines, and using a representative computer model of the turbine. This can provide savings based on extending conservative estimates of serviceable life, and allowing better scheduling of maintenance through earlier signalling of the need for component replacement.

The verdict: Gradual lowering of costs of O&M and lost yield, as parts can be replaced prior to catastrophic failure.

Where are we going?

There’s no doubt that the future will see widespread take-up of machine learning services, and even more data. The ‘internet of things’, or the idea that everything will be network-enabled, will mean that big data will get even bigger. With that come data storage issues, but also potential for even more fine-grained analysis, if the right tools can be developed.

We are likely to see more tech companies offering to apply data analysis and machine learning services. However, it is also likely that data analytics will be built into systems and supplied as standard, particularly for equipment such as gearboxes or entire wind turbines. 

But the biggest related change we see is the impact on the competitiveness of businesses, their services and products in our industry. For example, the last ten years has seen significant consolidation of wind turbine suppliers. In the United States three wind turbine suppliers accounted for 78% of new capacity in 2016 (according to the US Energy Information Agency). Substantial benefits can be offered by those wind turbine suppliers that are able to analyse the data pulled from a critical mass of installed wind turbines and their hard-won operating experience in a range of locations and conditions.

Being part of a large interconnected worldwide network of wind turbines is proving to be an attractive proposition for wind farm owners. The consequences include a trend towards longer and more comprehensive O&M agreements with the original equipment manufacturers, and a reduction in the perceived risk in providing finance to wind farm projects – just two of many trends that are contributing to a truly transformational change in the energy generation sector.

The verdict: Continuing gradual technical improvement, and significant commercial opportunities for suppliers leveraging data from large install bases.


The improvements from applications of machine learning are fascinating for the technically inclined, and are already improving the performance of wind farms worldwide. However, we don’t consider them transformational when compared with the major changes the application of AI will have in other industries such as transportation (e.g. self-driving cars).

Big data, machine learning and artificial intelligence offer current and future benefits for wind farm developers, manufacturers, owners and electricity market operators and traders, and it’s worth considering the applications, while understanding the limitations.

If you would like to find out more about how Entura can help you use data and machine learning to optimise your wind farm, contact  Akhil Pai on +61 406 874 101 or Silke Schwartz on +61 407 886 872.

About the author

Daniel Bennett is a renewable engineer at Entura. He has near a decade of experience investigating feasibility and due diligence energy yield assessments for renewable projects in Australia and around the world. With a background in mechanical engineering and computer science, and an interest in poking around in data to see what falls out, he assists clients seeking to understand the value in their projects.