Part 2:
Enabling decentralized control through Edge AI

By switching to distributed intelligence, with Edge computing, wind can finally realize the promised benefits of these many AI/ML models, not just individually, but the compounding returns that come from combining them at scale.

Wind is all about decentralized energy generation. It should also consider decentralizing its intelligence – enabling the turbines themselves to make real time decisions, instead of building big centralized ‘brains’ to supervise everything from data centers.

Edge AI enables this shift. It puts computational power and AI/ML models on the turbine, allowing them to collect data from their local environment, run models, and dynamically adjust to optimize generation. This all happens with minimal cost, whilst removing risks from network and connectivity issues.

The wind turbine might not be able to run the largest and most sophisticated models at the Edge, but usually there is no need. By training Edge models to identify normal operation, most day-to-day optimizations can be performed, whilst notably unusual patterns can be flagged to the cloud.
Wind turbines are also particularly suited to Edge technology as, although there are many types of turbine, they have fairly similar architectures, working principles, and issues.

Cloud will not disappear, of course, but wind will shift towards hybrid computing models. Data will still be backhauled to centralized computing units for non-urgent tasks, like developing and training new models. And a cloud based management system will still be needed to oversee and control the wind turbines, enabling virtual power plants (VPPs) which can also optimize operations at a portfolio level. But AI models themselves will mostly run at the Edge, delivering real time optimization.

Edge AI has already found many use cases in consumer electronics and high frequency manufacturing processes. Now it could usher in a new era of wind energy, resolving the complexity of managing ever-growing numbers of turbines in diverse settings, whilst maximizing each individual asset’s profitability. And the cherry on the cake is that Edge AI also reduces carbon emissions, by removing a tremendous amount of unnecessary data transfer and processing.

The major breakthrough will come from switching from a group of wind turbines to a network of intelligent wind power generators.

The benefits of Edge AI

At the turbine level:

Near real-time issue prediction and detection: pre-trained algorithms will offer real-time detection of ‘normal’ vs ‘unusual’ operation, allowing improved responsiveness to turbine underperformance. Edge algorithms could also integrate data from local IoT to monitor other signals (e.g. component fatigue, corrosion, etc.)
Dynamic contextualization and adjustments: using local sensor data (e.g. local weather, data, or grid signals), the turbine can dynamically adapt its parameters to optimize its production.
Asset security/data privacy: by processing most of the raw data inside the turbine, the amount of data flowing outside the asset (and beyond the wind farm) is limited and hence more secure.
Robust against network/connectivity issues: processing at the Edge allows the turbine to reduce its dependence on network quality and communication with control centers, reducing downtime from network issues.

At the global fleet level:

The industrialization of data analytics and AI: unlocking value from 10+ years of data science research and investments from asset owners and OEMs.
Turbine interference and interdependence management: minimizing interference (i.e. wake effect) between turbines.
Sensor redundancy and data consistency: integrating data from the surrounding turbines to check consistency with the turbine’s own sensor data, or ensure continuity of operation in the case of sensor failure (e.g. working with the closest turbine’s anemometer data if needed).

Self-optimized VPPs and services to the grid: turning a wind farm into a generation unit, able to couple with storage systems and act as a controllable asset to optimize revenues, manage load balancing, avoid penalties, and provide services to the grid.
Weather anticipation: generating better forecasts of wind speed and direction by using upstream wind turbine sensor data (i.e. wind direction, speed, temperature).
Contained infrastructure size and cost: by reducing raw data flows to/from assets to servers, Edge AI constrains infrastructure costs and size, reducing required network bandwidth and storage.

Federated learning: enabling turbines to integrate new anomaly detection capabilities (developed by other wind turbines) with their own sensors through machine learning. This is one of the most advanced features enabled by Edge computing and IoT networks (‘IoT swarms’) that would give turbines autonomous and collective ‘learning’ capabilities. Capgemini has developed a specific demonstrator of this technology, dedicated to wind farms, in the framework of the OASEES project.
Smaller carbon footprint: removing the daily upload and storage of hundreds of terabytes of raw data and time series would immediately offset data storage and data management carbon emissions.

How Edge AI will change business models

The Edge is not just a technological improvement, it will have dramatic implications for wind industry business models. Over the last decade, turbine manufacturers have developed a wider range of products and scaled designs to create powerful offshore turbines (15MW and beyond). This has led to increased complexity in design and manufacturing, just as demand for wind power is surging. As a result, wind OEMs face reliability issues and backlogs, and suffer from inflated warranty and service maintenance costs. This is leading to a focus on fewer product variants, and better fleet availability and reliability^[12]. Edge AI will support this strategy switch. It sets new standards for availability and reliability, and introduces features that enhance energy output by intelligently optimizing it based on environmental conditions, rather than simply adding larger blades to turbines.Turbines may soon differentiate based on CPU power and intelligent software packages, similar to the smartphone industry. Manufacturers could release a single turbine model with varying CPU power and software packages, offering different levels of intelligence, autonomy, and business insights.

We may soon see turbines differentiate based on CPU power and intelligent software packages (rather than just physical design), as in the smartphone industry.

This will have further implications for OEMs. They will need to evolve to manage new activities, like CPU obsolescence, and software lifecycle management – as other digitizing industries have. Software maintenance services will become critical, as a turbine's integrity and performance will strongly depend on its CPU condition monitoring and software updates.Energy developers will also benefit from these changes and see their business models evolve – as Edge AI enables them to manage clusters created by asset aggregation and hybridization, managing energy production through VPPs, maximizing power generation and monetization at the portfolio level. They might also give up developing algorithms to improve turbine performance (leaving this to OEMs). Instead, they might focus more on models that maximize the whole portfolio value, by improving energy production from various interdependent assets that collaborate on grid management and responding to market signals.