Research On Seismic Vulnerability Analysis Of Wind Turbine Tower Based On Machine Learning Method

China is a major country in wind energy resources and is undergoing an important period of energy structure transformation.Vigorously developing wind power generation is an important means to achieve energy structure transformation.However,in the context of the rapid development of China’s wind power industry,there is a huge safety hazard:on the one hand,China is a country with frequent earthquakes.On the other hand,as the core of wind power generation systems,wind turbines are more susceptible to seismic effects due to their towering and"top heavy"characteristics.As the most important load-bearing structure of wind turbines,tower structures are also the most vulnerable structural components to damage due to external extreme loads.Therefore,conducting seismic vulnerability analysis on them has certain scientific research value and practical engineering significance.However,the traditional structural seismic vulnerability analysis method often requires a lot of finite element calculation costs,which does not match the current situation of China’s rapidly developing wind power industry.Therefore,this article takes a reinforced concrete tower wind turbine as an example to introduce machine learning methods in the study of seismic vulnerability analysis of its tower structure,in order to establish a method for quickly analyzing the seismic vulnerability of the reinforced concrete tower of wind turbines by replacing the complex and time-consuming finite element seismic response analysis process with machine learning model prediction.The main content of this article includes:（1）This article provides an overview of the research status of traditional structural seismic vulnerability analysis and the introduction of machine learning methods in traditional structural seismic vulnerability analysis.At the same time,based on the research content of the paper,the basic theories of the three machine learning models used and corresponding model prediction accuracy evaluation indicators were summarized.The basic theories of structural seismic vulnerability analysis,structural dynamics analysis,and artificial ground motion synthesis were also elaborated;（2）A three-dimensional refined finite element model of the"reinforced concrete tower wind turbine foundation"coupled system has been established,and its structural natural vibration characteristics have been studied.Meanwhile,based on the analysis results of the structural natural vibration characteristics,the specific values of the Rayleigh damping parameters of the model were calculated.In addition,the basic principle of infinite element boundaries was also elaborated and implemented in a coupled system model;（3）Based on the obtained 130 seismic motion records,finite element seismic response analysis was conducted on the coupled system model,and the construction of the seismic response database for the reinforced concrete tower of the wind turbine was achieved.Meanwhile,based on relevant research data,the maximum horizontal displacement angle(θ_max)was selected as the evaluation index for the seismic response of the reinforced concrete tower of the wind turbine in this paper,and a reasonable structural damage level was determined;（4）Based on the traditional Intensity Measure（IM）,Class I and II intensity indicators considering multi-dimensional seismic motion are proposed.Select the maximum horizontal displacement of the top of the tower structure as the model output,and use Class I and II strength indicators as the model input,respectively,to compare the impact of the two types of strength indicators on the accuracy of the three machine learning models.The results indicate that the model using Class I strength indicators has higher accuracy and stronger explanatory power for the maximum horizontal displacement at the top of the tower structure;（5）Research on the impact of the number of training sets on the model’s predictive ability and the identification of key seismic intensity indicators based on a debugged machine learning model.The results indicate that the three machine learning models exhibit high prediction accuracy when the number of training sets is only 52.At the same time,it was also found that compared to other strength indicators,I_PGV,I_PGD,and I_HI have better explanatory power for the maximum horizontal displacement of the top of the tower structure,which is a more critical conclusion in this study;（6）Introduce the cloud graph method to establish seismic vulnerability curves for reinforced concrete towers of wind turbines based on the results of finite element seismic response analysis and machine learning model prediction,and conduct comparative analysis.The results show that the maximum difference between the structural seismic vulnerability curve established based on the prediction results of random forest model considering only 3D inputs and the real vulnerability curve is only 4.87%,which verifies the feasibility of the proposed method of replacing the structural finite element seismic response analysis process by machine learning model prediction,so as to realize the rapid seismic vulnerability analysis method of reinforced concrete tower of wind turbine.