A team of researchers at the US National Renewable Energy Laboratory (NREL) has developed an AI-based tool that could be used to significantly boost the efficiency of future wind farms, while also reducing their environmental impact.
The researchers developed an AI-based surrogate model called the Wind Plant Graph Neural Network (WPGNN), which was trained on simulations of more than 250,000 randomly generated wind plant layouts under a variety of atmospheric conditions, wind farm designs and turbine operations. The simulation data were generated by another NREL-developed model, the FLOw Redirection and Induction in Steady State (FLORIS) tool.
Advertisement
The AI then used the information to determine the optimal design of a wind farm. The AI facilitates the calculation of the ideal layout and operation of the wind turbines in order to achieve different outcomes, such as reducing land requirements or increasing revenue.
The research focused on a strategy called wake steering, which optimises the amount of energy that a wind farm can produce by controlling the wake moving from an upstream turbine away from a downstream turbine. The use of AI enabled the researchers to determine the impact that wake steering would have on three different objectives: land use, cost and revenue.
Advertisement
The benefits of wake steering have previously been demonstrated at the wind farm level, but most studies have been limited in spatial scale and in the range of optimisation objectives considered. The WPGNN used by the NREL team efficiently represented wake interactions as a directed graph, which allowed for a comprehensive investigation into the optimal settings for both the turbine location and the yaw of the nacelle across a nationwide wind energy portfolio.
‘Previously, site-specific wake steering optimisation studies were very difficult, but the graph representation in the WPGNN dramatically improved our ability to represent flexible layouts, changing wind directions, and perform gradient-based optimisation,’ said Ryan King a senior scientist in the laboratory’s Computational Science Center.
Advertisement
The use of wind as a renewable energy source is expected to become increasingly important in decarbonising the power sector, but obstacles remain as some communities have restricted where wind turbines can be erected. The AI-guided scenario considered a US nationwide deployment of 6,862 plant buildouts with a cumulative 721 gigawatts of generated power, with the goal of reducing 95 per cent of carbon emissions from the energy sector by 2050.
The adoption of wake steering strategies could reduce the land requirements for future wind plants by 18 per cent on average and by as much as 60 per cent in some instances. Across the USA, the land savings total about 13,000 square kilometres, which is equivalent to 28 per cent of the nation’s wind energy footprint.
Wake steering is valuable because simply spreading out turbines is often not enough to avoid wake losses and some wind farms lack the space required to expand further. Besides, wind farms optimised for wake steering would allow for a larger concentration of turbines, thereby satisfying the desire of some local communities to limit how much land the industry is allowed to use. The installation of more turbines in a smaller footprint would offer increased flexibility from a site-planning perspective, potentially allowing developers to tap into economies of scale for larger projects.
Advertisement
The researchers also found that the use of wake steering consistently reduces the cost of energy for wind deployments. The AI allowed the researchers to uncover regional differences where the strategy would be best put into place.
‘We found that different areas of the country are more or less amenable to the benefits of wake steering and the outcomes of those benefits can be realised in different ways,’ said Andrew Glaws, a researcher in applied mathematics. ‘This can be important for helping to understand how and where we should be investing in this new technology.’
Advertisement
The research has been published in Nature Energy.