摘要
为应对城市地下综合管廊建设中的施工安全问题,并能对施工风险类别进行准确评估,加强预防施工安全事故发生的能力,提出由随机森林和卷积神经网络相结合的施工风险评估模型。按照4M1E的分类方式将指标分为人、物料设备、管理、技术、环境5大类,选择RF随机森林对22种特征指标进行属性约简,选择最相关指标,并为施工现场的安全管理提供依据;使用1D-CNN卷积神经网络对降维后的数据进行风险评估。结果表明,RF-1D-CNN模型对80组样本数据进行分析的均方误差可达0.011 2%,随机选取30个样本作为测试集进行模型对比,RF-1D-CNN的准确率高于BP神经网络,该模型对城市地下综合管廊施工安全风险等级具有高识别精度和高效率。
To deal with the construction safety problems in the construction of an urban underground integrated pipe gallery and accurately evaluate the construction risk categories,the paper intends to propose a construction risk assessment model combining random forest and a convolutional neural network.First,according to the 4MIE classification,the indicators are divided into five categories:human,material equipment,management,technology,and environment,with a total of 22 evaluation indicators.The Delphi method is used to process 80 sets of rating data collected according to construction risk level.RF random forest is used to select the optimal index,and the data set is split into a training set and a test set with a ratio of 9:1.The feature_importances_parameter is called to calculate the importance of the feather and select the feature.According to the importance,22 evaluation indicators are ranked.The cumulative importance index is set to 90%,and the indicators can be reduced to 10,to strengthen the safety management of the construction site.Then the 1D-CNN convolutional neural network is used to evaluate the risk of dimensionality reduction data.The sample values are normalized.The reduceddimensionality reduceddimensionality data indicators are divided into a training set and validation set with a ratio of 8:2.The convolution kernel size of the convolution layer is set to 2×1.The activation function is nonlinear activation function ReLu function.To avoid data overfitting,dropout is added after the maximum pooling layer and the dropout loss rate is set at O.2.The Output layer is fully connected,the output is 4,and the activation function is softmax.The evaluation results are compared with the actual data,and the mean square error of data analysis is 0.0112%.Finally,30 groups of samples are selected randomly as a test set to compare and analyze the risk assessment results of the RF-1D-CNN model,CNN model,BP model,and RF-BP model.The accuracy rates are 100%,96.67%,90%and 76.67%respectively.The results show that the acc
作者
秦华礼
祝艺露
QIN Huali;ZHU Yilu(Collegeof ResourcesandCivil.Engineering,Northeast University,Shenyang 110819,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2023年第7期2184-2190,共7页
Journal of Safety and Environment
关键词
安全工程
综合管廊
施工安全
卷积神经网络
随机森林
safety engineering
integrated pipe gallery
safety in construction
convolutional neural network
random forest
