Search

An appropriate cooperative yaw control strategy, in which the accurate power prediction and efficient optimization method play an essential role, can mitigate the significant power loss owing to the wake interference. The present study develops a novel double-layer machine learning framework, which combines an artificial neural network (ANN) yawed wake model and a Bayesian machine learning algorithm. The former undertakes the power prediction as the 1st layer, which is fed to the 2nd layer of the optimization system. Using computational fluid dynamics (CFD) simulation result as a reference, the performance of the proposed framework is evaluated compared with the framework based on a Gaussian-based analytical wake model by Qian and Ishihara. Furthermore, parametric studies have been conducted on the inflow conditions to explore its applicability. The results show that the proposed framework can realize more accurate power prediction with a mean error of 1.13% and 0.96% for different inflow velocities and turbulence intensities respectively, in contrast with the one based on the analytical model, 7.61% and 1.66%. Moreover, more notable power improvements can be achieved by the proposed framework, seeing a mean rise of 5.59% and 2.22% under four inflow velocities and turbulence intensities respectively, compared with 3.55% and 1.02% based on the analytical model. Specifically, its superiority becomes more obvious under relatively low inflow velocity and turbulence intensity. [ABSTRACT FROM AUTHOR]

The high power extraction of wind farms necessitates a comprehensive optimization system encompassing precise and efficient layout design and cooperative control. This study applies a novel ANN-based power prediction framework to resolve the layout optimization combined with the genetic algorithm, then composes a double-layer machine learning framework for the control optimization incorporating Bayesian machine learning. Three wind scenarios and control schemes are accounted for to investigate their impacts on the actual power benefits of the layout and control optimizations, respectively. The study shows that the divergence of wind resources will lead to more evenly scattered turbine arrangements for each direction sector in the optimized layout but may weaken the functionality of layout design in annual power rise owing to the compromise between different wind directions. Under the spread wind scenario, the spread optimized layout may narrow the space for the power enhancement of the yaw control, whereas the unidirectional optimized layout may give rise to a more considerable power gain from the yaw control. Row-based control shows desirable performance in the regular aligned layout, while column-based control is recommended for irregularly optimized layouts, particularly for unidirectional optimized ones. The spread optimized layout will widen the gap of power improvement between the column-based and independent controls, thus weakening the advantage of the column-based control. [ABSTRACT FROM AUTHOR]

This paper presents a data-driven wind farm layout optimization framework that uses a machine learning wake model that considers physical control stages. The machine learning wake model is trained using well-validated Computational Fluid Dynamics (CFD) simulation data, and consists of thousands of sub-models, each of which is an artificial neural network (ANN) wake model. The ANN wake models are trained separately for low-speed and high-speed inflows to ensure high accuracy of the predictions, with less than 2% error compared to CFD simulations. The accuracy and efficiency of the framework are validated, and the results show better agreement with CFD simulation than an analytical wake model developed in recent years. A parametric study on the number of wind turbines in the Horns Rev wind farm demonstrates that more wind turbines can be added with a minor decrease in average power, with more even and staggered layouts. Under full-wake uniform inflow, the selected analytical wake model exhibits a power prediction error of 5%–8%, while the differences between optimal layouts searched by different wake models range from 2% to 8%. When introducing a wider range of inflow direction sectors, the discrepancy between optimal layout performances decreases. [ABSTRACT FROM AUTHOR]

The accurate assessment of the dynamic response of the offshore wind turbine is fundamental to its economical and safe design, where the precise characterization of the incident wind and waves is essential. While reliable aerodynamic models can accurately reproduce wind characteristics, the widely-used empirical JONSWAP spectrum in wave field simulation may not capture the multi-modal feature of the mixed wave field near the wind farm. In this study, the stochastic structural responses based on an integrated FE model of the soil-pile-structure system are assessed employing the empirical and actual wave spectra, along with discussions on the impacts of wave theory and wave-structure interaction. The main conclusions are: (1) The empirical wave spectrum can overestimate the dynamic responses compared with the measured swell-dominant one, but the dominant wind load weakens its influence on displacement. (2) When the wave serves as the decisive loading, using the empirical spectrum overestimates the dynamic performance for the swell-dominant wave spectrum while underestimating it for the wind-induced-wave-dominant one. The approximation of the actual wave spectrum to the JONSWAP spectrum is only recommended when the coupling effect of wind and wave is inapparent. (3) The wave theory imposes an evident effect on the dynamic behavior of the offshore wind turbine subject to joint wind and wave loading, where the JONSWAP wave spectrum is more sensitive. (4) The influence of water-monopile interaction is negligible when using the JONSWAP wave spectrum but prominent for the simulation's actual wave spectrum. • An integrated FE model considering soil-pile-structure interaction is developed. • Response discrepancies induced by JONSWAP and actual wave spectra are analyzed. • Influence of spectral disparities induced by inhomogeneous wave field is discussed. • Sensitivities of spectrum influence to wave theory and wave-structure interaction are studied. [ABSTRACT FROM AUTHOR]

In the past, offshore wind farms commonly had a large wind turbine distance to avoid power loss induced by the wake effects. For this reason, the operational offshore wind farms with such a large wind turbine distance are relatively low in capacity density and have a great potential to include more wind turbines. This paper aims to provide a suitable renovation plan for those operational wind farms. Different from traditional wind farm layout optimization, the purpose of the renovation is to install new wind turbines while maintaining the status quo, which is an optimization problem with the constraint of initial wind turbines. Therefore, the renovation plan can be applied flexibly without waiting for the decommissioning of initial wind turbines. In this paper, the renovation plan is provided by a robust wind farm layout optimization framework based on a machine learning wake model, which can evaluate overall power production more accurately than traditional analytical models. Both heuristic and gradient-based optimization algorithms are applied and compared. Horns Rev wind farm is analyzed for demonstration. The differences between the two algorithms are limited to 1%. Furthermore, the renovation and freeform optimizations are successfully applied to the cases with 12.5%∼50% more wind turbines. According to the results, the freeform optimizations without the constraint of initial wind turbines have similar performance to renovation plans with an error smaller than 0.3%. When the number of wind turbines is added from 80 to 120, the normalized AEP (annual energy production) slightly decreases from 96.6% to 93.4%, which means the power loss increases by 3%∼4%. In the meantime, the capacity density increases significantly by 50% as the area remains the same. In conclusion, the renovation plan is recommended for aging wind farms with a low capacity density like Horns Rev wind farm. • Machine learning wake model is efficient enough for layout optimization. • Differences between genetic algorithm and gradient-based algorithm are limited to 1%. • Additional wind turbines can significantly increase the capacity density while the power loss only grows by 3%∼4%. • Renovation and freeform layout optimization have similar optimal solutions in the study cases. • Renovation plans are recommended for wind farms with a low capacity density, like Horns Rev wind farm. [ABSTRACT FROM AUTHOR]

• The double-layer ML framework realizes an accurate and efficient power optimization. • The row-based control scheme can increase the optimization efficiency remarkably. • The divergence of wind direction weakens the advantage of the row-based scheme. • The row-based cooperative control scheme is more applicable to the aligned layout. In the real-time cooperative control of large-scale wind farms, the simultaneous achievement of accuracy and efficiency by the optimization framework plays an indispensable role. This paper presents a new double-layer machine learning (ML) framework comprising an Artificial Neural Networks (ANN) yawed wake model and Bayesian ML algorithm to strike a desirable compromise between accuracy and efficiency. Given the control on the iteration number with the scale-up of the wind farm, a novel row-based control scheme is further put forward to improve the optimization rate by reasonably reducing the optimization parameters. Moreover, parametric analysis has been performed considering the wind distribution and layout configuration to explore its applicability compared with the general independent one. The study shows that the novel framework performs favorably in an accurate and efficient power prediction and optimization of the wind farm. The row-based control scheme can further improve the convergence rate of the double-layer optimization framework remarkably at the expense of a slight decrease in optimal power production. The divergence of the wind distribution can dwindle the power gain of the wake steering strategy and weaken the superiority of the row-based cooperative control scheme. The row-based cooperative control scheme is more applicable to the aligned layout than the staggered layout, and this advantage is enhanced with the increase of wind farm scale. [ABSTRACT FROM AUTHOR]

Over the past decades, the increasing energy demand has accelerated the construction of wind farms, raising higher expectations for precise load and power assessments in wind turbine performance. Traditional methods, which rely on analytical wake models and performance curves, often fail to adapt to complex inflow scenarios, leading to significant inaccuracies in predicting turbine loads and power output. This research addresses these challenges by introducing a novel two-phase framework for various phases of wind farm planning and development, using the NREL 5MW baseline wind turbine as a case study. The first part involves deriving recommended values for simplified thrust modulation factors at the preliminary design phase, enabling swift evaluation of maximum and fatigue thrust loads crucial for wind farm optimization. The second part focuses on designing and training a machine learning model at the detailed design phase. A gradient-boosting-based framework based on LightGBM provides comprehensive methods for assessing wind turbine load and power, enhancing the precision and efficiency of these assessments. The proposed model achieves significant improvements in predictive accuracy, achieving mean R-Squared of 0.995, 0.988, and 0.995 for power, peak load, and damage equivalent load evaluation, respectively. The framework streamlines the assessment process, enhancing both the accuracy and speed of power and load evaluations for wind farm design. This is expected to reduce computational costs and improve the effectiveness of downstream tasks, such as layout optimization and wake steering. • A novel framework is introduced to evaluate power and loads at various design stages. • The modulation factor approach evaluates thrust statistics without complex modeling. • The LGBM-based model provides a comprehensive assessment of wind turbine performance. • The framework addresses challenges in predicting load and power in complex inflow. [ABSTRACT FROM AUTHOR]

Wind energy has become one of the most commercially prospective renewable energies. However, the wake effect of wind turbine can reduce the power generation efficiency and increase the fatigue loading of downstream turbines. Hence, the wake effect study has attracted increasing interests. Compared with the extensive study on the single turbine wake, that on the superposition effect of multiple turbine (multi-turbine) wakes is limited. In this study, the characteristics of the wake velocity and turbulence intensity are studied and the exponential superposition model is proposed for the aligned multi-turbine wakes. Firstly, Simulator for Offshore Wind Farm Applications (SOWFA), a high-fidelity simulator for the interaction between wind turbine dynamics and the flow in a wind farm, is used to analyze the distribution of aligned multi-turbine wakes. It is observed that wakes reach the steady state from second turbine in the aligned turbines. Then influences of different factors on the accuracy of existing superposition models are studied. It is found spacing has significant effect on the performance of superposition models. Furthermore, the exponential superposition model with higher applicability is proposed for the wake velocity and turbulence intensity. Finally, this model is validated by the benchmark data of real wind farms. • Characteristics of the wake velocity and turbulence intensity of aligned multiple turbines are addressed. • Influences of different factors on the accuracy of existing superposition models are addressed. • Exponential superposition model is proposed for the wake velocity and turbulence intensity. • Validations in real wind farms show exponential superposition model has good performance. [ABSTRACT FROM AUTHOR]