基于CART回归树模型的变电站施工安全事故分析与预测

Analysis and prediction of substation construction safety accidents based on CART model

在当前的变电站施工过程中,主要通过数据包络分析过程预测安全事故,忽略了表征信息中的不确定性,导致预测结果的选取受试者工作特征曲线下面积(AUC)值较低.针对这一问题,本研究应用分类回归树(CART)模型,设计了一种新的变电站施工安全事故分析与预测方法.首先,利用固定型、移动型采集技术相结合的方式,采集变电站施工现场数据,并通过主成分分析算法进行筛选处理.然后,深入分析变电站施工安全事故发生过程,通过基于概率分布的可分性判据,提取施工安全事故前兆特征.最后,利用CART模型构建施工安全事故根节点,再使用支持向量机(SVM)回归算法建立叶节点,形成可用于施工安全事故预测的最优决策树.通过迭代训练多个串联的CART模型实现梯度提升,应用该模型即可得到准确的事故预测结果.实验结果表明:该预测方法灵敏度更高,能够预测出更多的安全事故,并且该预测方法的AUC值高达0.91,具有更高的预测精度.

In the current substation construction process,the data envelopment analysis process is mainly used to predict the safety accidents,and the uncertainty in the characterization information is ignored,which leads to the low receiver operating characteristic curve area(AUC)value in the selection of the prediction results.In order to solve this problem,a new analysis and prediction method of substation construction safety accidents is designed by using classification and regression tree(CART)model.Firstly,the data of substation construction site are collected by combining fixed and mobile acquisition technologies,and filtered by principal component analysis algorithm.Then,the occurrence process of substation construction safety accidents is deeply analyzed,and the precursor characteristics of construction safety accidents are extracted through the separability criterion based on probability distribution.Finally,the CART model is used to build the root node of construction safety accidents,and then the support vector machine(SVM)regression algorithm is used to build the leaf node,forming the optimal decision tree that can be used for construction safety accident prediction.By iteratively training multiple CART models in series,the gradient can be improved,and accurate accident prediction results can be obtained by applying this model.The experimental results show that the prediction method is more sensitive and can predict more safety accidents,and the AUC value of the prediction method is as high as 0.91,which has higher prediction accuracy.