This is a set of multiple presentations, held throuout 2022. The series has been organised by B. Doekemeijer and S. Mulders as part of the IEA Wind Task 44. In the talks multiple researchers present their current work which is in some way connected to the topic on wind farm flow control.
This series of 4 talks was presented on December 11th 2023 and was recorded. The talk was organized by IEA Wind Task 44, Work Package 4. The 4 projects presented were:
The presentation was held on the 4th of February 2022 and was recorded.
As wind energy becomes a more relevant part of the current and future energy mix, we have to investigate how we can use wind farms most efficiently. The complex aerodynamic nature of the interaction between the turbines makes real-time decisions to increase the power generated difficult. Steady state engineering models have proven to be a viable option: These approximate the flow behavior with a set of analytical equations which allows a fast evaluation. This way, look up tables can be created for different wind directions, which provide improved operational setpoints for the wind farm in steady state. This presentation will focus on the FLORIDyn (FLOw Redirection and Induction Dynamics) model. It utilizes the low computational cost of steady state engineering models and creates a framework which adds the lost dynamics. Since its release FLORIDyn has undergone many changes which lead to the dynamic and flexible framework it is today. In its current state the simulation of large wind farms in heterogeneous and changing wind conditions is possible in real-time. With its features and the still very low computational cost it has become a viable option for model-based dynamic wind farm control. The presentation will highlight key ideas of the versions, followed by a showcase of the most recent implementation.
The presentation was held on the 11th of February 2022 and was recorded.
The reference open-source controller (ROSCO) has been developed to provide the wind energy research and development community with a wind turbine controller that can be applied in numerous applications. The ROSCO framework provides methods of automated tuning and implementation of a controller. This framework can be leveraged to rapidly and easily tune a controller for any standard wind turbine model, and the controller can be implemented in many popular wind turbine simulators. This talk provides a high-level overview of what the ROSCO framework contains, how it can be used by the research and development community, and what the future development of ROSCO entails.
The presentation was held on the 18th of February 2022 and was recorded.
In the last years, Wind Farm Control has started to demonstrate its important potential up to the field implementation and in the industry, mostly for power maximization. But uncertainties remain on the effect of wind farm control on turbine loads, which the scientific community has identified as a further research need. Predicting turbine fatigues, using Damage Equivalent Loads (DELs), require computationally expensive aero-servo-elastic simulations. It is thus not so straight-forward to include in real-time "live" control and/or in offline iterative optimizations (for instance in FLORIS optimal-configuration finder). For this reason, the development and the use of load surrogate models is of current interest as it could allow to tackle these challenges. This talk will present some methodology that has been recently explored to develop load surrogate models based solely on Rotor-Equivalent Inflow Quantities (REIQs). If it can be shown to be accurate enough, it is believed that using a load surrogate model based only on the individual turbine REIQs would be quite general and universal (not needing to know the turbine position in the wind farm, the presence of undetected atmospheric effect, or the local terrain specificities....). After generating a reference dataset, Artificial Neural Networks are trained to predict the DELs based on the REIQs, and the prediction capabilities are assessed. Some (promising) preliminary results of the application of this methodology in both a simulation environment (FAST.Farm) and in a wind tunnel experimental environment (using TUM scaled wind turbines) will be presented and discussed.
This presentation was held on the 25th of February 2022 and was recorded.
Historically, control protocols have optimized the performance of individual wind turbines resulting in aerodynamic wakes which typically reduce total wind farm power production 10-20%. Wake steering, the intentional yaw misalignment of turbines in a wind farm to deflect energy deficit wake regions, has demonstrated potential as a wind farm control approach to increase collective power production. Leveraging aerodynamic wind farm models, we designed a physics-based, data-assisted wake steering control method to increase the power production of wind farms. Parameters in aerodynamic wake models are inherently uncertain. We develop approaches for the efficient calibration and uncertainty quantification of wake model parameters and we perform optimization under model parameter uncertainty. The method was tested in a multi-turbine array at a utility-scale wind farm, where it statistically significantly increased the power production over standard operation. The analytical gradient-based power optimization methodology we developed can optimize the yaw misalignment angles for large wind farms on the order of seconds, enabling online real-time control. To improve wake steering control in transient ABL conditions, we developed a closed-loop wake steering control strategy, which is tested in large eddy simulations of the terrestrial diurnal cycle, altogether, the results indicate that closed-loop wake steering control can significantly increase wind farm power production over greedy operation provided that site-specific wind farm data is assimilated into the aerodynamic model.
This presentation was held on the 4th of March, 2022 and was recorded.
In recent years, industrial controllers for modern wind turbines have been designed as a combined wind speed estimator and tip-speed ratio (WSE-TSR) tracking control scheme. In contrast to the conventional and widely used K-omega-squared torque control strategy, the WSE-TSR scheme provides flexibility in terms of controller responsiveness and potentially improves power extraction performance. However, both control schemes heavily rely on prior information about the aerodynamic properties of the turbine rotor. Using a control-oriented linear analysis framework, it is shown that the WSE-TSR scheme is inherently ill-conditioned. The ill-conditioning is defined as the inability of the scheme to uniquely determine the wind speed from the product with other model parameters in the power balance equation. Uncertainty of the power coefficient contribution in the latter mentioned product inevitably leads to a biased effective wind speed estimate. As a consequence, in the presence of uncertainty, the real-world wind turbine deviates from the intended optimal operating point, while the controller believes that the turbine operates at the desired set-point. Simulation results confirm that inaccurate model parameters lead to biased estimates of the actual turbine operating point, causing sub-optimal power extraction efficiency.
This presentation will be held on the 11th of March, 2022.
FLORIS Version 3.0, released in February 2022, represents a major redesign, rewrite, and enhancement of the open-source software to allow for faster, more accurate, and more varied computations with the several significant improvements for academics, manufacturers, developers, and small businesses. Version 3.0 has improved computation speeds. New algorithms enable improved memory usage and reduce the number of mathematical operations required while the ability to leverage a larger portion of a computer's processor allows multiple mathematical operations to be performed at the same time. A complete redesign of the software architecture makes FLORIS Version 3.0 better modularized to foster collaboration and adoption of more modern software best practices. The new Cumulative Curl wake model, developed in collaboration with the National Offshore Wind Research and Development Consortium, improves the accuracy of FLORIS's wake models for large offshore wind farms.
This presentation was recorded
This presentation was recorded
This talk describes results from a wake steering experiment at a commercial wind plant involving two wind turbines spaced 3.7 rotor diameters apart. During the 3-month experiment period, we estimate that wake steering reduced wake losses by 5.6% for the wind direction sector investigated. After applying a long-term correction based on the site wind rose, the reduction in wake losses increases to 9.3 %. As a function of wind speed, we find large energy improvements near cut-in wind speed, where wake steering can prevent the downstream wind turbine from shutting down. Yet for wind speeds between 6–8 m/s, we observe little change in performance with wake steering. However, wake steering was found to improve energy production significantly for below-rated wind speeds from 8–12 m/s. We attribute much of the improvement in wake steering performance at higher wind speeds to a significant reduction in the power loss of the upstream turbine as wind speed increases. Additionally, we find higher wind direction variability at lower wind speeds, which contributes to poor performance in the 6–8 m/s wind speed bin because of slow yaw controller dynamics. Further, we compare the measured performance of wake steering to predictions using the FLORIS (FLOw Redirection and Induction in Steady State) wind farm control tool coupled with a wind direction variability model. Although the achieved yaw offsets at the upstream wind turbine fall short of the intended yaw offsets, we find that they are predicted well by the wind direction variability model. When incorporating the expected yaw offsets, estimates of the energy improvement from wake steering using FLORIS closely match the experimental results. In addition to presenting the results of the wake steering experiment, we discuss the methods for quantifying the increase in energy gain and reduction in wake losses, including uncertainty quantification.
This presentation was recorded
In this talk, Adam Stock from the University of Strathclyde gives a demonstration of the "StrathFarm" wind farm simulation tool, developed at the University of Strathclyde over the last decade. StrathFarm aims to fill the gap of simulation tools for wind farm controller development by providing a computationally efficient platform that includes all the necessary dynamics to allow users to investigate the impact of wind farm control algorithms on power capture and structural loads on the turbine. Currently, StrathFarm is a MATLAB-based tool, though a key aim for the future is to move outside of the MATLAB environment to create an open-source software package that is free for researchers to use.
Adam gives a run through of how the wind farm is modelled, the key features, and a demonstration simulation and results. He also discusses the future of StrathFarm and how others can help in its development, set up joint projects and, in the future, use StrathFarm themselves.
This presentation was recorded
In recent years dynamic induction control has shown great potential in reducing wake-to-turbine interaction by increasing the mixing in the wake. With these wake mixing methods, the thrust force will vary over time. If applied to a floating wind turbine, it will cause the platform to move. In this presentation, the effect of dynamic induction control and dynamic individual pitch control on a DTU10MW turbine mounted on a triple-spar platform and its wake is evaluated. When wake mixing strategies are enabled, the platform will undergo significant motions which will impact wake stability. To understand the impact of the motion on the wake, the motions for dynamic induction control, covered in this presentation, are simulated using the free wake vortex method as implemented in Qblade. Under laminar inflow, results show that the wind speed at a distance of 5 rotor diameters downstream can be increased by up to 10% compared to a fixed-bottom turbine.
This presentation was recorded
The first part of the talk explores whether LES can be fast enough to be used in real-time, by comparing the solution divergence of fine and course-grid simulations over time. In the second part of the talk, we investigate to what extent three-dimensional turbulent flow fields in LES can be reconstructed from sparse measurements obtained from a scanning lidar. This would constitute the state-estimation step in an LES-based optimal control algorithm. To conclude, the next steps and challenges for LES-based optimal control of wind farms are briefly discussed.
This presentation was recorded
Individual Pitch Control (IPC) of wind turbines has the potential to greatly reduce once-per-revolution loads on large wind turbines, thereby extending their lifespan. H-infinity control is a sophisticated, robust state-space control approach that allows us to directly “shape” the closed-loop frequency response of a complex MIMO system such as IPC. However, great expertise is typically required to tune such a controller, which limits its applicability to new systems (i.e. new turbine designs), particularly in the context of iterative design or optimization procedures. There is a need to develop a robust auto-tuning procedure that can tune a robust H-infinity controller for a given wind turbine design to meet given performance criteria. This talk focuses on the early stage problem formulation and system analysis supporting precisely this proposal.
This presentation was recorded
Wake steering collectively controls turbines in a wind farm to increase the farm’s total power output or provide services to the electricity grid. Real-time wake steering using model-based control requires a fast model, which has thus far limited the use of more accurate, but slower dynamic wake models. Learning-based control methods can avoid the need for a model, but typically require long training times before being deployed. This work describes how a hybrid model- and learning-based control method, differentiable predictive control (DPC), can use a dynamic wake model for real-time wake steering while avoiding the long training times of purely learning-based approaches. I start with the building block of hybrid control using a steady-state wake model, FLORIS. I present a new implementation of FLORIS, which is learning-enabled and differentiable--characteristics required for DPC. I then show results for DPC which highlight the promise and outstanding challenges of hybrid control for dynamic wake steering.