重庆理工大学学报(自然科学) ›› 2023, Vol. 37 ›› Issue (8): 203-211.

• 智能技术 • 上一篇    下一篇

采用因子分析与改进 GMM的施工安全评价方法

於三大,朱 浪,苏 立   

  1. (1.中国三峡建工(集团)有限公司,成都 610041; 2.重庆大学 微电子与通信工程学院,重庆 400044)
  • 出版日期:2023-09-15 发布日期:2023-09-15
  • 作者简介:於三大,男,博士,正高级工程师,主要从事企业管理信息化、工程数字化、智慧化研究,Email:yu_sanda@ctg. com.cn;通信作者 廖勇,男,博士,副研究员,主要从事下一代无线通信技术、人工智能及其在行业中的应用研 究,Email:liaoy@cqu.edu.cn。

Construction safety evaluation method using factor analysis and improved GMM

  • Online:2023-09-15 Published:2023-09-15

摘要: 随着施工安全问题日益复杂,为进一步减少施工安全事故的发生,针对传统安全评 价方法无法有效挖掘各安全指标之间的内在联系,并且现有聚类方法存在紧凑性不足、结果解 释性差的问题,提出一种采用因子分析与变分贝叶斯高斯混合聚类的安全评价方法。该方法利 用因子分析将复杂的施工安全评价指标转换为有内在联系的因子变量,作为变分贝叶斯高斯混 合方法的输入,并使用 T分布随机相邻嵌入法 (tdistributedstochasticneighborembedding, TSNE)对多维聚类结果进行可视化,充分挖掘各施工安全指标之间的内在关联性并对施工安 全做出评价。案例分析表明,与层次聚类分析、Kmeans以及高斯混合模型(gaussianmixture model,GMM)方法相比,所提方法具有更好的聚类效果和全局寻优性能,不仅验证了所提方法 的可行性和有效性,还通过可视化的方法增强了多维聚类问题的可解释性。

关键词: 施工安全评价, 因子分析, 变分贝叶斯高斯混合模型, 可视化, 聚类

Abstract: With the continuous increase of construction needs, the problem of construction safety is becoming more and more complex. In order to further decrease the occurrence of construction safety accidents, the traditional safety evaluation methods cannot effectively excavate the internal connection between the safety indicators. Besides, the existing clustering methods have the problems of insufficient compactness and poor interpretation. Therefore, this paper presents a safety evaluation method using factor analysis and variational Bayesian Gaussians mixture model, which adopts factor analysis to convert complex construction safety evaluation indicators into internally related factor variables as the inputs of variational Bayesian Gaussians mixture method. The multidimensional clustering results are visualized using the T-distributed stochastic neighbor embedding (T-SNE) method, so as to fully explore the intrinsic correlation between various construction safety indicators and evaluate construction safety. Case analysis shows that compared with hierarchical clustering analysis, K-means and Gaussian mixture model (GMM) method, the proposed method has better clustering effect and global optimization performance. It not only verifies the feasibility and effectiveness of the proposed method, but also improves the interpretability of multidimensional clustering problems by visualization method.

中图分类号: 

  • TP391