The potential of artificial intelligence (AI) in the renewables industry is immense. By harnessing advanced data analysis, machine learning and predictive modeling techniques, AI can revolutionize wind resource assessment and vibration analysis of wind farms.
While there are many opportunities to apply AI to wind farm development, it is essential to have an enterprise-wide AI strategy in place to protect intellectual property and proprietary data. The AI strategy should include the following components, among others:
- What data will be used to train the AI tool?
- Who has access to the intellectual property and proprietary data?
- Where is the intellectual property and proprietary data stored?
- How is the intellectual property and proprietary data protected?
- Who has access to the output from the AI tool?
Until the answers to these questions are clear, it is wise to implement a “No AI Policy” until a comprehensive strategy has been developed.
Before discussing how AI can help in the planning of a modern wind farm, a best practice is to start with discrete data. Machine learning algorithms are difficult enough without having to clean data and label it as natural language programming (NLP). Creating data sets from pre-existing discrete data (think database tables) already labeled is a huge time savings (cost) and faster return on investment.
Once this foundation has been built, the possibilities for AI to enhance wind farm planning are almost limitless. The wind industry can leverage AI for wind resource assessment in several ways, including:
- Metrological data
- Turbulence and wind shear analysis
- Micro siting optimization
- Wind analysis and forecasting
- Remote sensing
- Data analysis
- Preventive and performance management
For meteorological data analysis, AI systems can aggregate and analyze vast amounts of data from many disparate sources, including weather stations, satellite information, historical climate data and real-time sensor data. How data is collected and stored is crucial to successfully using AI tools.
Careful setup and planning are necessary, as many tools, such as OpenAI, enable proprietary data but may store it in their cloud for learning unless specifically configured, not to be trained to store proprietary data.
Machine learning models can also be applied to wind pattern prediction, forecasting wind speed, direction and turbulence intensity over time. These models learn from historical data and adapt to changing weather patterns, providing more accurate forecasts than traditional methods.
For another example, consider a location under assessment where wind data is being evaluated. There are many off-the-shelf and proprietary software tools available for this purpose. Gathering as much data as possible from public resources, including local weather stations, is advisable. While obtaining such data is common practice, what is often overlooked is the comparison between predictions and actuals. For location assessment analysis, collecting data from the past 10 years on or near the location would be beneficial, organizing it into two tables that display predicted vs. actual values, providing vectors of pre-labeled discrete data.
Next, all meteorological data from anemometers at the actual site should be compiled into a distinct table, ensuring that vectors of pre-labeled data are available for comparison with the predicted and actual models. At this point, the analysis can begin. Traditionally, regression analysis and algorithms were used to predict future outcomes. Still, with this approach, the actual vs. predicted data can help determine the real conditions on-site when the prediction indicates a specific value (X).
Another benefit of using AI is its capability to analyze wind turbulence, which impacts the wear and tear of wind turbine components. By assessing turbulence intensity, AI can aid in selecting sites with smoother wind flows, reducing maintenance costs and extending the lifespan of turbines.
Additionally, AI tools can evaluate wind shear. The use of accurate models aids in determining the optimal turbine hub height to maximize energy capture while minimizing costs. Utilizing the previously mentioned pre-labeled data, the same data set can be leveraged to provide a longer time scale for analyzing wind turbulence and shear. This is possible by developing a new algorithm that combines the current model with long-term predictions and actual data.
When assessing wind resources, it is also critical to examine micro-siting – determining the exact placement of each turbine upon a site. AI can use computational fluid dynamics (CFD) simulation and machine learning to identify locations that maximize wind exposure while minimizing turbulence caused by neighboring turbines (wake effect.) It can also analyze terrain data to determine how landforms (e.g., hills, valleys, etc.) affect wind flow. If implemented strategically, AI can identify the best spots for turbine installation to capture the strongest, most consistent wind.
To support this approach, it would be preferable to use additional anemometers, or those that can be easily relocated, allowing public resource vectors to be compared with local actuals to determine the ideal micro-site location. Before finalizing this placement, the algorithms should be run again against the wind shear and turbulence models. While a location may have strong wind drafts, it could also be prone to extreme shear and turbulence in certain conditions.
As previously mentioned, AI can assist in providing wind forecast expectations, but it can also predict long-term wind patterns, from seasonal variations to decadal climate projections. By integrating historical climate data and global circulation models, AI can forecast how wind resources may shift due to climate variability or long-term climate change. In this case, gathering more public data on long-term forecasts and, when compared against actuals, would be beneficial to create a trend analysis data set with labeled discrete data. Developing a new algorithm for this data set would enable long-term trend analysis. This model would account for climate change impacts when combined with previous data sets.
Additionally, AI can quantify uncertainty in wind resource predictions, providing developers with scenarios to assess site viability and financial risks
Leveraging AI can produce more precise energy yield models by incorporating multiple variables such as wind speed, temperature and air pressure to estimate potential energy outputs. These models help investors and developers better estimate the return on investment for various sites. By integrating the models created earlier with energy price models for the potential site, developers can predict output based on a broader data set, ensuring long-term site viability.
Combining local price prediction services with historical wind pattern forecasts and actual local data allows the creation of simultaneous algorithms that analyze all the data sets for more accurate predictions. Using AI, real-time data from existing turbines can also improve short-term energy production forecasts, aiding in grid management and operational planning and predicting the long-term viability of the wind farm over its 20-30-year operational period.
Remote sensing and data enhancement are another advantage of using AI in wind farm development. AI can analyze remote sensing data from satellites, drones and LiDAR to create accurate, high-resolution wind resource maps, automating the data processing to save time and improve forecast accuracy. AI can also generate synthetic data to fill observational gaps, enhancing the robustness of wind resource assessments.
Prior to investment, these pre-labeled data sources can be combined with previous data sets, adding another layer of historical insight and precision to predictions as complex models grow. AI’s use of simultaneous equations, where one algorithm’s error term feeds into another, enhances this process.
Another benefit of AI is its ability to detect anomalies in wind patterns that traditional analysis might miss. This can help identify unique site characteristics or predict extreme weather events that could affect wind farm operations. Machine learning algorithms, like “k-means clustering,” can categorize sites based on wind characteristics, allowing developers to focus on the most promising locations.
In economics, this concept is known as “the power of E” — the value of unexplained factors. By utilizing multi-vector data sets for a specific location, the error term becomes a key variable, and cross-functional algorithms can reduce “unexplained” factors, refining predictions to account for all variables discussed.
AI can also drive decision-support systems that provide real-time recommendations for site selection, turbine placement, and operational adjustments based on continuously updated wind data. With the right dataset, AI can run multiple “what-if” scenarios, evaluating factors such as turbine types, site layouts, and weather changes and their impact on wind farm performance, aiding in strategic planning.
Additionally, AI accuracy can be improved by combining AI datasets with traditional wind resource assessment methods, like mesoscale (large-scale weather forecasting models), creating hybrid models. AI can refine existing wind atlases (maps of average wind speeds across regions) by analyzing localized data, resulting in more granular and accurate information on wind resources.
By utilizing AI, wind farm developers can make more informed, data-driven decisions, optimizing wind resource use, reducing costs and mitigating risks. This powerful technology can potentially drive the future of wind energy in North America by enhancing the performance and cost-effectiveness of new wind farms.
Dave Hopson, PhD, is the managing partner and founder of Triumphus, a leading IT consulting firm based in Houston, Texas.
Filed Under: Featured