贝壳形屋盖风压系数密度峰值聚类分区研究

Study on peak density clustering partition of wind pressure coefficient of shell-shaped roofs

鉴于贝壳形屋盖表面风压分布的特殊性,提出基于快速搜索技术的密度峰值聚类风压系数分区方法。以某贝壳形屋盖表面风压分布试验结果为基础,进行密度峰值聚类风压系数分区,采用SD有效性指标验证分区数的合理性,并与改进k-means聚类分区结果进行对比。结果表明:密度峰值聚类风压系数分区以风压系数相对距离、局部密度和综合指数为特征参数,能较好反映屋面风压分布特性,有效保证类聚合性和类类分离性;相较于改进k-means分区法,不同风向角下密度峰值聚类得到的分区数与SD最优聚类数接近;密度峰值聚类分区结果能更准确反映贝壳形屋盖表面风压分布特性,充分体现测点风压系数局部密度和相对距离特征值较大的基本原则,对于贝壳形屋盖的风压系数分区具有更好的适用性;贝壳形屋盖密度峰值聚类分区最大负风压系数在-0.59~-1.74之间。

Because of the particularity of the distribution of wind pressure on the surface of the shell-shaped roof,a partition method of wind pressure coefficient based on the peak density clustering of the fast search technology was proposed.Based on the test results of surface wind pressure distribution on a shell-shaped roof,the wind pressure coefficient partition of peak density clustering was carried out.The SD validity index was used to verify the rationality of the number of divisions,and compared with the improved k-means clustering partition results.The results show that the peak density clustering wind pressure coefficient partition can better reflect the wind pressure distribution characteristics of the roof with the relative distance of wind pressure coefficient,local density and comprehensive index as the characteristic parameters,and effectively ensure the class aggregation and class separation.Compared with the improved k-means partition method,the number of partitions obtained by peak density clustering under different wind angles is close to the SD optimal clustering number.The peak density clustering partition results can more accurately reflecting the characteristics of wind pressure distribution on the surface of shell-shaped roofs,fully reflecting the fundamental principle that the local density and relative distance of wind pressure coefficients at measurement points have larger characteristic values,which has better applicability for wind pressure coefficient partitioning of shell-shaped roofs.The maximum negative wind pressure coefficient of peak density clustering partition of shell-shaped roofs ranges from-0.59 to-1.74.