基于改进K-means算法的大跨屋盖结构表面风荷载分区研究

Study on Wind Load Zoning of Largespan Roofs Based on Improved Kmeans Algorithm

针对K-means算法进行大跨屋盖结构表面风荷载分区中存在的分类数k值需凭经验事先给定以及所有初始聚类中心均需随机选取带来的分类情况数过多、从中寻找最优分类结果工作量大且效率低的问题,提出基于改进K-means算法的大跨屋盖结构表面风荷载分区方法。首先,建立分类数k与其相应测点风荷载的误差平方和(Sum of the Squared Errors:SSE)关系曲线,引入手肘法基本思想,实现最优分类数kst值的精准识别;其次,在首个初始聚类中心随机选取基础上,引入轮盘法基本思想,完成对剩余初始聚类中心的高效选取;然后,根据类内紧凑、类间分散的原则,通过类内紧凑性判定指标S(k)和类间分散性判定指标D(k),构造并借助SD(k)值有效性检验,得到最优的风荷载分区结果;最后,以北京奥林匹克网球中心大跨悬挑屋盖结构为例,针对风洞试验所得风荷载测试结果,采用所提方法对其表面最不利风压系数进行分区计算,并与传统K-means算法进行对比,结果表明,所提方法能够高效实现大跨屋盖结构表面风压分区计算,具有较好的工程应用价值。

In the application of the K-means algorithm for the wind load zoning on the surface of largespan roof structures,the classification number k values are given in advance by experience,and all initial clustering centers are randomly selected.This often results in an excessive number of classification,leading to increased workload and low efficiency in identifying the optimal classification results.To address these issues,this study proposed a wind load zoning method for large-span roof structures based on improved K-means algorithm.First,a relationship curve between the classification number k and the sum of the squared errors(SSE)of wind loads at the corresponding pressure taps was established,incorporating the Elbow Method to accurately determine the optimal classification number kst values.Next,after randomly selecting the first initial clustering center,the Roulette Wheel method was introduced to efficiently select the remaining initial cluster centers.Following this,based on the principles of intra-cluster compactness and inter-cluster dispersion,compactness criterion S(k)and dispersion criterion D(k)were employed to construct and validate the zoning effectiveness using the SD(k)value,which resulted in the optimal wind load zoning scheme.Finally,taking the large-span cantilevered roof structure of Beijing Olympic Tennis Center as an example,wind tunnel test results were employed to calculate the most unfavorable wind pressure coefficients on the surface of the structure.A comparison with the traditional K-means algorithm demonstrated that the proposed method efficiently achieved wind pressure zoning for large-span roof structures and holds significant engineering application value.