Research On Multi-fidelity Simulation Methods For Wind Farm Wake And Output Power

For large wind farms,the power loss due to wind turbine wake effects can be up to 10?40%.To establish an accurate and efficient wind farm wake simulation approach,which is essential for the improvement of the accuracy of wind farm optimal design and control.The prevailing wind farm wake simulation methods can be divided into two categories,which are engineering wake modeling and computational fluid dynamics(CFD)wake modeling.The former is able to do fast simulation,while has a low accuracy;the latter has a high computational accuracy,while at the expenses of large amounts of computational resources.Taking the simulation characteristics of both of the two methods into consideration,an approach combining the computational advantages of these two methods has been explored,which is the multi-fidelity simulation method.The proposed method is able to balance the contradict of computational accuracy and efficiency.Focus on the multi-fidelity simulation of wind farm wake and output power,the following work has been accomplished.(1)A fast-calculation approach for wind farm wake modeling considering wind direction uncertainty has been proposed.Considering the commonly used methods for the calculation of wind farm output power are all based on engineering wake models without taking the error produced by wind direction uncertainty into consideration.A correction method based on Gaussian Averaging(GA)considering the 10-minute wind direction distribution has been proposed.By means of the weighted average of the calculated power within a moving window,to replace the traditional power calculation based on 10-minute average wind direction.The wind farm output power and annual energy production(AEP)calculation considering wind direction uncertainty is studied.Taking the Lillgrund offshore wind farm as a test case,the Jensen,Gaussian,Larsen wake models are all taken to model wind power output power,and GA correction was used to correct the calculation results of wind farm output power or AEP.Via the comparison with measurement data,results show that the accuracy of the calculated wind farm output power was tremendously improved by using GA correction,and the convergence process has also been speeded up.(2)A surrogate model for wind farm output power based on Gaussian Process has been proposed.The wind power calculation based on CFD method takes huge computation resources,which is difficult to be applied to engineering fields.Due to the above reason,and considering about the variation characteristics of wind power against input wind speed and wind direction as well,a surrogate model based on Gaussian Process(GP)has been proposed for the output power calculation under all possible wind conditions.The k-?-fp turbulence model containing the vertical shear of wind farm local area has been adopted as the CFD simulation scheme to calculate the wind farm flow fields and output power.The sampling method and kernel function scheme can be determined according to the characteristics of different variables,thus the surrogate model for wind farm output power considering the wake effects can be established.Taking the Lillgrund offshore wind farm as an example,the mean value and uncertainty of output power can be obtained based on the established surrogate model,and then the effectiveness of GP method on the modeling of wind farm output power under all the input wind conditions can also be validated.(3)A multi-fidelity method for wind farm wake and output power simulation combining engineering wake model with CFD model has been proposed.On account of the high accuracy and time-consuming of CFD methods as well as the high efficiency and low accuracy of engineering wake models,the semi-empirical Larsen model and the BANS method based on the k-?-fp turbulence model are adopted as low fidelity and high fidelity sub-model,respectively.Considering the similarities of calculation results and the computation characteristics of different models,the multi-fidelity model for wind farm wake and output power simulation is established based on the Co-Kriging algorithm.Via the analysis of the accuracy and the computational time required by different surrogate models in Lillgrund offshore wind farm,the results show that to achieve the same accuracy,the multi-fidelity model has an obviously higher computation efficiency than the surrogate model merely using the high-fidelity model results.Thus,the superiority of multi-fidelity surrogate model for wind farm wake and output power simulation is verified.(4)The wind power forecasting model for complex terrain is developed based on the CFD pre-calculated flow fields.In order to realize the efficient and precise use of wind farm flow field characteristics,a wind power forecasting model based on CFD pre-calculated flow fields is proposed and applied to wind farms with complex terrain.Via the construction of flow character database,the time-consuming flow fields calculation for all wind conditions can be done ahead of the wind power forecasting process.And then,the wind power forecasting will be converted to the interpolation process within a look-up table by taking NWP data as input.Therefore,the prediction effectiveness can be improved without reducing the accuracy.Besides,the established forecasting model has a clear modeling process which is convenient for model optimization.(5)The short-term wind power forecasting methods based on NWP wind speed correction and clustering pre-calculated CFD model are proposed.First,for the low accuracy of traditional meso-scale numerical weather prediction(NWP)data,the correction models for NWP wind speed based on multiple linear regression,radial basis function neural network and Elman neural network are developed,respectively,and then the CFD method for short-term wind power forecasting is proposed by taking the corrected NWP wind speed as input.Case study shows that all of the three proposed correction models can significantly improve the accuracy of meso-scale NWP wind speed,so as the precision of the short-term wind power prediction taking corrected NWP as input.Second,for the fact that a single mast is unable to accurately represent the wind speeds of the whole wind farm,the wind turbine clustering models based on K-means,agglomerate hierarchical clustering and spectral clustering algorithms are developed,respectively,and the rationality criteria to evaluate wind turbine clustering models are also established,by using silhouette coefficient,Calinski-Harabaz index and Within-Between index as indicators.Via the comparison and selection of different clustering schemes,the clustered database of wind farm flow field characteristics are established,so as the wind power forecasting model based on clustering CFD flow field database.Case study indicates that the CFD based wind power forecasting method using the optimal clustering scheme significantly improves the prediction accuracy of the single database,both in temporal and spatial scales,which gives strong support to the improvement of short-term wind power forecasting system.