| With the rapid development of the wind power industry,wind turbines gradually tend to be large megawatts,large in size,and complicated in structure,and the noise radiated by the wind turbines also increases.The noise generated by wind turbines not only affects the lives of residents near the wind farm,but also the noise radiation level has become an important indicator for unit type certification and evaluation of unit quality.Studying the acoustic characteristics of the noise signal of wind turbines can provide a reference for the evaluation of the operation status of the unit,as well as the research on vibration reduction,noise reduction and fault diagnosis of the unit.Considering the problems such as long sampling period,difficult test conditions,a nd large interference from environmental factors during the noise test of wind turbines,the use of non-acoustic signals that are easily measured by the unit to predict the noise signal of the unit can not only save the tester’s repeatability test The econ omic cost also overcomes the problem of difficult testing caused by the harsh environment.Therefore,this paper focuses on the two issues of analyzing the acoustic characteristics of the wind turbine’s noise signal and predicting the wind turbine’s noise through non-acoustic signals.1.According to the IEC61400-11 wind turbine unit noise test standard,Our research group built a complete machine noise test platform at a wind farm in Xilin hot,Inner Mongolia,and Collect the noise signal of the whole machine of the wind farm with 2.5MW,3.0MW and 3.2MW models,which provides reliable experimental data for the research content of subsequent chapters.2.Analyzing the acoustic characteristics of the wind turbine noise signal,and combining wavelet packet theory to extract the wavelet packet energy feature of the noise signal of the wind turbine at different speeds.In order to identify the subtle structure of the frequency spectrum with a large energy proportion in the noise signal,the low-frequency part is decomposed by a second wavelet packet,using digital downsampling to process the reconstructed signal,Combining Fourier transform to perform spectrum analysis on the down-sampled reconstructed signal,through simulation analysis,found that there is a spectrum peak in the frequency spectrum of the noise signal of the wind turbine at different speeds,which is linearly related to the speed.3.For the question of wind turbine noise prediction,first using multiple linear regression to predict the noise of the wind turbine,and combining regression analysis to analyze the significance of the variables of the wind turbine,and optimiz ing the regression model through Collinear analysis and outlier diagnosis.According to the characteristics of unit signal volatility and discontinuity,the wavelet packet transform is used to decompose and reconstruct the diagnosed samples to obtain subsequences in different frequency bands,and multiple linear regression is used to predict the subsequences in these frequency bands respectively,and the subsequence prediction results The actual prediction model is obtained by superimposing the combined prediction.Then consider that when performing linear regression prediction,the difference in sample diagnosis methods will cause errors in the prediction results of the established model,so the integrated learning XGBoost algorithm is applied to wind turbine noise prediction.Finally,in order to verify the feasibility of the above wind turbine noise prediction model,it is verified by the wind field measured data,and the prediction effect of the model is compared and analyzed.4.Based on the content of this article,based on MATLAB GUI design and development of analysis software for the above simulation calculation and function realization,the software includes user management,machine noise signal feature analysis,model prediction effect evaluation,non-acoustic signal pair Acoustic signal prediction,analysis results export and save functions. |
