基于多源信息融合和WOA-CNN-LSTM的外脚手架隐患分类预警研究

Classification and early warning research of scaffolding hidden trouble based on multi-source information fusion and WOA-CNN-LSTM

面对施工现场外脚手架隐患信息的多样性,传统的基于传感器监测的单一信号预警研究存在容错力不佳、含有信息有限等问题。针对施工现场外脚手架“图像+监测”数据,提出一种基于数据层和特征层信息融合的脚手架隐患分类预警方法。首先,利用Revit三维建模软件建立外脚手架实体模型,对不同初始隐患下的外脚手架进行有限元分析,划分隐患预警等级;其次,利用无迹卡尔曼滤波算法(Unscented Kalman Filter,UKF)及卷积长短时记忆网络(Convolutional Neural Network-Long Short Term Memory Network,CNN-LSTM)实现脚手架同类信息数据层融合及异类信息特征层融合;最后,通过实时收集西安市某在建项目落地式双排扣件式钢管脚手架隐患信息,对其进行分类预警,并使用鲸鱼优化算法(Whale Optimization Algorithm,WOA)对CNN-LSTM网络进行参数优化,发现隐藏节点个数为30、学习率为0.0072、正则化系数为1×10^(-4)时分类效果最佳,优化后预警精度达到了91.4526%。通过可视化WOA-CNN-LSTM、CNN-LSTM、CNN-SVM(Support Vector Machine,支持向量机)及CNN-GRU(Gate Recurrent Unit,门控循环单元)分类预警结果,证实了优化后的CNN-LSTM网络在脚手架分类预警方面的优越性。

In light of the information diversity of hidden dangers of external scaffold at construction sites,the traditional single-signal early warning research based on sensor monitoring has the problems of poor fault tolerance and limited information.Aiming at the"image+monitoring"data of external scaffold at construction site,this study proposes a classification and early warning method of hidden scaffold dangers based on the information fusion at data layer and feature layer.Firstly,the solid model of the floor type double row fastener steel tubular scaffold was established by using the Revit 3D modeling software,and its safety was reviewed.Then,the finite element analysis of the scaffold under different hidden danger conditions was carried out by using the random sampling method,and the displacement monitoring data and image data of hidden dangers were divided into four types of early warning levels respectively.Secondly,the Unscented Kalman Filter(UKF)algorithm was used to denoise and fuse the same kind of information from multiple sources to form a kind of heterogeneous information.Besides,the Convolutional Neural Network-Long Short Term Memory Network(CNN-LSTM)was used to fuse the"image+monitoring"data at the feature level.Finally,through the real-time collection of the monitoring data of the floor type double row fastener steel tubular scaffold of a project under construction in Xi'an,the hidden danger information of the scaffold was classified for early-warning.The Whale Optimization Algorithm(WOA)was used to optimize the parameters of the CNN-LSTM network,When the number of hidden nodes is 30,the learning rate is 0.0072,the regularization coefficient is 1×10^(-4),and the classification effect is the best.After optimization,the early warning accuracy reaches 91.4526%.Through visualization of WOA-CNN-LSTM,CNN-LSTM,CNN-SVM(Support Vector Machine)and CNN-GRU(Gate Recurrence Unit)classification and early warning results,the superiority of optimized CNN-LSTM network in scaffold classification and early warning is confirmed.