Explore the vast range of titles available.

Turbine power and yaw set points can be adjusted across a wind farm to minimise the overall power losses and the additional fatigue loads caused by wake interactions. Detailed modelling is required to understand the complex flows in sufficient detail to allow a realistic practical control design. High-fidelity computational fluid dynamics requires enormous computational resources, so simpler engineering models are needed which capture the most important effects while running fast enough to allow sufficient testing. This paper describes a steady-state optimisation tool which has been extended to optimise all the power reduction set-points and yaw offsets simultaneously for different wind conditions. It also describes a fast time-domain simulation model which captures turbine and wake dynamic effects, so that wind farm controllers of all kinds can be tested in realistic and time-varying conditions. To demonstrate its application for controller testing, the performance of the combined power and yaw controller is tested during changing conditions of wind speed, direction and turbulence derived from measured site data. Finally, the need for validation is discussed, as many uncertainties still need to be resolved in order to obtain sufficient confidence that the potential benefits of such wind farm control schemes can be realised in practice.

A surrogate model for turbine loading is presented which has been developed for predicting the fatigue loads on each turbine in a wind farm, taking full account of wake effects, for detailed site-specific load assessment and/or to predict the effects of wind farm control on the loads. The model has been implemented in the wind farm simulation code LongSim , where it can be used both for steady-state setpoint optimisation against a merit function which can include both power and turbine loads, and for dynamic time-domain simulations, for example to evaluate the effects of wind farm control on turbine loading. For this application, fatigue loads need to be evaluated much more rapidly than is possible with full aeroelastic turbine simulations. Surrogate models based on fatigue load interpolation from a database of pre-computed aeroelastic turbine simulations are unwieldy and likely to miss important wake effects. The key simplifying approximation of the new model is that the load contributions from different causes can be calculated separately and then combined. This makes it straightforward to account for all relevant effects, at the same time vastly reducing the number of aeroelastic simulations required to populate the model.

The wind turbines in a wind farm must turn to face the wind to achieve maximum power production. Until now, each turbine achieves this individually: the nacelle-mounted wind vane measures the yaw misalignment, from which the turbine controller decides when to make a yaw correction by using a combination of heavy filtering and counters or dead-band hysteresis. In a wind farm context, each turbine could make use of information from its neighbours, and therefore, in principle, it should be possible to achieve better yaw control by using a centralised farm-level yaw control algorithm which uses input measurements from all the turbines, and then tells each turbine how to yaw. \"Better\" yaw control will always be a compromise between maximising energy production and minimising yaw actuator duty. Using a realistic dynamic simulation model of a wind farm which uses actual measured wind data as input, different possible yaw strategies are tested, and compared in terms of energy production and yaw actuator duty. The results indicate that centralised yaw control is likely to achieve a better compromise. There also is much current interest in the use of wake steering, where yaw setpoints are optimised to steer the wakes of some turbines away from others. Centralised yaw control may offer a better way to implement these setpoints, rather than sending yaw offsets to be acted on by the normal turbine yaw controllers with their own filtering, dead-band logic, etc.

Wake losses significantly reduce wind farm output, but wind farm flow control (WFFC) can substantially reduce these losses, using wake steering (yawing upstream turbines) and/or axial induction control (reducing turbine power and thrust to weaken their wakes). Previous work shows that ignoring wind farm blockage in traditional wake models represents a prediction bias of similar order to the gains achievable with WFFC. Axial induction control works by changing turbine thrust, which is also the cause of blockage; this raises the question of how the two effects interact. Induction control can more than compensate for any loss due to blockage, but here we investigate the relationship further. Induction control reduces turbine thrust coefficients to reduce wake losses, but this should also reduce blockage, suggesting that induction control might achieve higher gains in practice than predicted with blockage effects ignored. An engineering model for blockage effects was added to the wind farm code LongSim, and steady-state gains calculated for a well-known offshore wind farm, with and without blockage. The results confirm that the power gains are indeed higher if blockage is modelled. These results are corroborated by comparisons against RANS (Reynolds-Averaged Navier Stokes) simulations, in which blockage effects are implicitly modelled.

An engineering wake model based on the Ainslie model is proposed. The eddy viscosity term associated with momentum diffusivity is modified to take into account the effects of atmospheric stability. The parameters used are typically available from high-quality on-site measurement campaigns and the effects of atmospheric stability are based on empirical models for the estimation of the Monin-Obukhov length. The dependence on physical quantities only is particularly advantageous for fast wake modelling, since no further parametrical tuning is needed for each specific case. The proposed wake model is initially compared to wind tunnel data and CFD simulations to test the chosen Obukhov lengths for specific flow conditions. Wind farm production data and concurrent meteorological data at one onshore site are then used to validate the model for specific on-site flow conditions, obtaining good results. Two offshore wind farms are also used to assess the model in a large-scale wind farm scenario: results look promising although some reservations are expressed on the effect of the wake superposition model. Models for the prediction of wake centreline deflection due to yaw are compared at different yaw angles using wind tunnels data and CFD simulations. Although the EPFL model showed some advantage, especially in non-neutral conditions, both models give satisfactory results. Furthermore, it was showed that the wake deflection caused by the rotating wake for non-yawed turbines can have a large impact on the predictions of the centreline deflection.

This paper describes the design and testing of an axial induction controller implemented on a row of nine turbines on the Sedini wind farm in Sardinia, Italy. This work was performed as part of the EU Horizon 2020 research project CL-Windcon. An engineering wake model, selected for its good fit to historical SCADA data from the site, was used in the LongSim code to optimise turbine power reduction setpoints for a large matrix of steady-state wind conditions. The setpoints were incorporated into a dynamic control algorithm capable of running on-site using available wind condition estimates from the turbines. The complete algorithm was tested in dynamic time-domain simulations using LongSim, using a time-varying wind field generated from historical met mast data from the site. The control algorithm was implemented on-site, with the wind farm controller toggled on and off at 35 min intervals to allow the performance with and without the controller to be compared in comparable wind conditions. Data were collected between July 2019 and early February 2020. The results have been analysed and indicate a positive increase in energy production resulting from the induction control, in line with LongSim model predictions, although a larger volume of valid data would be necessary to provide statistically robust conclusions. The measurements also provide a validation of the LongSim model, proving its value for both steady-state setpoint optimisation and time-domain simulation of wind farm performance.

Wake steering is a form of Wind Farm Control (WFC) in which upstream turbines are yawed in order to redirect their wakes away from neighbouring downstream turbines. The reduction in power of the upstream turbines can be more than compensated for by the increase in power from the downstream turbines, as they see a faster and less turbulent inflow; over a wind farm, this can give a net increase in farm power. Calculating the required yaw set-points for WFC over a range of wind conditions and for multiple turbines can be computationally intensive for traditional optimisation methods. Reinforcement Learning (RL) is a form of machine learning where agents learn beneficial behaviour patterns through trial-and-error to achieve long-term goals. It has become a topic of interest for WFC, primarily through using time-averaged (steady-state) wind farm models to train the RL agents. In this work, RL agents are trained using a dynamic flow solver, with incoming wind from a wide range of directions. These RL agents are able to yaw the turbines towards set-points that increase the farm power, despite initially losing power while travelling. Using a realistic wind time history, their performance was compared to similar RL agents trained with a steady-state model. Both training approaches gave agents that can successfully control the wind farm to gain 2-3% additional power. However, the results suggest that it is not necessary to expend the additional wall-clock or computational time to train RL agents using dynamic flow over steady-state.

Lillgrund offshore wind farm near Copenhagen has been the subject of a number of research projects. The close turbine spacing makes it an interesting candidate for research into turbine wakes and wind farm control. As part of the EU TotalControl project, an axial induction controller has been designed for field testing at Lillgrund. The controller was designed using steady-state optimisations in DNV’s LongSim code, which predicted a small but significant power gain along with a considerable reduction in fatigue loading. Some simplifications were introduced for ease of practical implementation on this particular site; the controller was initially optimised to improve the power output of one 7-turbine row, with the aim of extending to the whole wind farm if the first test was successful. The expected power gain was confirmed by running realistic, detailed time-domain simulations in LongSim , and independently corroborated using steady-state calculations with DTU’s linear CFD code Fuga . The controller was implemented on site, and a toggle test was initiated on site in late December 2021, with the wind farm controller toggling between controlled and default states every 50 minutes. First results are expected early in 2022.

We present a full-scale validation study of an optimal power open-loop wind farm control schedule (WFCS) using a full row of turbines (Row-A) at the Lillgrund offshore wind farm as the demonstration case. The derived WFCS is conditioned on the ambient wind speed and wind direction. The WFCS’s are developed using the DNV and DTU numerical models LongSim and Fuga. Prior to implementation, the developed optimal controllers are tested by comparing LongSim results, based on Fuga optimized input, with Fuga optimized results and vice versa. These models are very different in their nature — Fuga being a linearized RANS CFD solver, and LongSim being based on a range of engineering wake models and allowing for unsteady wind farm flow modelling — but nevertheless the overall computed Row-A power gains are reasonably similar. This gives confidence to the developed WFCS’s, even though the two models predict significantly different distributions of the power gains over wind speeds and directions. Based on the numerical analysis, simplified WFCS’s for implementation in the Row-A turbines are developed and tested against the more detailed optimized controller. The test predicts only a small reduction of performance enhancement of the simplified controller compared to the detailed controller, and on this basis it is decided to use the simplified controller for the full-scale turbine implementation. The test, toggling between base control and optimal control in 50-minute successive time slots, has been running since 09-05-2022 and is still ongoing. A huge amount of data is needed to obtain robust statistics for the comparison of measured and numerical data.

Validating engineering wake models under real‐world operational conditions is essential for improving wind farm performance predictions. This study utilises a unique dataset from the Lillgrund offshore wind farm, collected during the Horizon 2020 TotalControl project campaign, integrating synchronous Supervisory Control and Data Acquisition (SCADA) and Light Detection and Ranging (LiDAR) measurements under both baseline operation (i.e., no intentional yaw offset) and active wake steering scenarios. We assess four combinations of analytical wake models, each employing distinct formulations for velocity deficit, added turbulence, wake superposition and deflection, implemented in the LongSim modelling software developed by DNV. The analysis focuses on time‐averaged wake velocity deficit profiles and turbine‐ and farm‐wide power output, normalised by reference velocity and power. Model accuracy is quantified using mean absolute error (MAE) metrics. The evaluated models generally reproduce wake deficit trends under systematic variations in wake overlap in baseline cases, as well as wake deflection due to intentional yaw misalignment during the wake steering cases across a range of atmospheric conditions. Normalised velocity deficit MAE values range from 7% to 15%, with discrepancies primarily attributed to inflow heterogeneity, near‐wake complexity and model‐specific parameterisations. Power predictions reveal error accumulation with increasing farm depth. Model combinations incorporating cumulative wake superposition and refined turbulence schemes demonstrate improved agreement with field data; however, all models face challenges capturing localised flow features, with normalised turbine‐level power output MAE ranging from 3% to 23%. Farm‐wide power output errors ranged between −13% and +30%; though accurate farm‐level predictions may mask compensating errors at individual turbines. Future studies should prioritise dynamic inflow characterisation and inclusion of blockage effects to enhance predictive reliability further.