基于K-means算法的隧道机电分项工程设备设施评定标度的研究

Study on Evaluation Scale of Equipment and Facilities of Tunnel Electromechanical Subdivisional Project Based on K-means Algorithm

在依据《公路隧道养护技术规范》(JTG H12—2015)标准,对分项工程设备设施的设备完好率和技术状况评分的基础上,为了满足各省运营公司分类养护,提出了基于K-means算法聚类的基于标度定义的设备设施的养护措施。首先,依据标准H12的内容,在某省75座隧道机电设施定检数据的基础上,得到分项工程的设备完好率和技术状况评分2个参数,共计4 770个数据,将每座隧道分项工程设备设施的设备完好率和技术状况评分组合形成2种结构的2个数组数据集合。然后,进行了K-means算法分类,从分第k类(k为自然数)开始,选取设备完好率和技术状况评分距离最远的2个参数(通过数值均一化处理)进行迭代,直至选取第k类结果结束。采用迭代的方法把数据集合划分为k类(采用基于距离的聚类算法),使相同类的内部数据之间的具有极大的相同性,而类之间的内部数据相同性尽量小,达到最优的分组聚类的效果。结果表明:分项工程设备设施的技术状况标度分为4类是合理的,并且提出了标度为0,1,2,3的分类并分为不同的养护措施。该研究成果在全国部分省份的隧道机电设施养护中得到了积极推广应用,在资金使用、养护队伍人员建设、日常养护、养护科学决策和预防性养护方面获得了良好的效果。

On the basis of scoring the equipment intactness rate and technical condition of equipment and facilities of subdivisional projects in accordance with the Technical Specification of Maintenance for Highway Tunnel(JTG H12—2015), in order to meet the classification and maintenance of operating companies in each province, the maintenance measures of equipment and facilities based on scale definition and K-means algorithm clustering are proposed. First, according to the content of standard H12, on the basis of the regular inspection data of electromechanical facilities of 75 tunnels in a province, the of equipment readiness rate and technical condition score of subdivisional projects are obtained, with a total of 4 770 data. The equipment readiness rate and technical condition scores of equipment and facilities of subdivisional project of each tunnel are combined to form 2 data sets of 2 structures. Then, the K-means algorithm is divided into several classes. Starting from the k-class(k is a natural number) division, the 2 parameters with the farthest distance between the equipment intactness rate and the technical condition score(through numerical homogenization) are selected for iteration, until the k-class result is selected. The data sets are divided into k classes by using the iterative method(the distance based clustering algorithm), so that the internal data of the same class have great similarity while the internal data similarity between classes is as small as possible to achieve the optimal grouping and clustering effect. The result shows that it is reasonable to divide the technical status scales of equipment and facilities of subdivisional projects into 4 classes, and the classification with scales of 0, 1, 2 and 3 are proposed, and different maintenance measures are divided as well. The research result has been actively promoted and applied in the maintenance of tunnel electromechanical facilities in some provinces of China, and good effect have been achieved in terms of fund use, maintenance team construction, daily maintenance, scientific maintenance decision-making and preventive maintenance.