The Research On Wind Resource Assessment Of A Wind Farm Using Machine Learning And Meteodyn WT

Wind power is considered to be clean energy and is becoming one of the most promising forms of new energy generation.The mature wind power technology has achieved good economic performance that in turn greatly boosts its development.Replacing polluting traditional coal-and-gas and oil-fired power stations with renewable energy sources(e.g.,wind farms),would dramatically reduce carbon emissions and is critical for carbon peak and carbon neutrality.Thus,the wind industry is growing quickly around the world,especially in China.The thesis introduces machine learning to process wind measurement data and predict short-term wind parameters,including wind speed and wind direction.After that,we estimate the regional wind energy resource reserves of a wind farm,and calculate the wind energy resource distribution and annual available hours of the wind farm using the Meteodyn WT software.This research provides technical supports for site selection,wind turbine selection,and unit layout principles.The main research contents of this thesis are listed as follows:(1)A comparative analysis of the performance of support vector regression machine and deep learning in predicting short-time wind speed and wind direction is presented.Results show that the prediction result of the support vector machine(SVM)model with Gaussian kernel outperforms other kernel functions,and deep learning outperforms SVM in wind speed prediction.However,the wind direction prediction accuracy of both deep learning and SVM is lower than the wind speed prediction accuracy.(2)The wind speed and wind direction of a wind farm are compared between the actual and predicted values by LSSVM using different kernel functions.Experimental results suggest that the LSSVM model with Gaussian kernel achieves more satisfactory for predicting wind speed and wind direction compared to that using other kernel functions.(3)Taking a wind farm as an example,the actual data information of the wind farm is applied to calculate the wind resource distribution and the annual available hours of the wind farm using the Meteodyn WT software.In addition,we compare the predicted power generation using short-term wind speed prediction and the actual power generation.Experimental results show that the short-term wind speed prediction is better and the wind speed prediction has some social significance for the actual engineering application.In addition,experimental results show that the performance of the predicted short-term wind speed is satisfactory.The thesis applies SVM and deep learning algorithms to predict wind speed,and achieves satisfactory results.Thus,the output of this thesis has practical significance when applied to short-term wind speed prediction in wind power generation in China.