基于脑电的隧道施工人员危险预测能力分类模型  

EEG-based classification of worker's hazard prediction in tunnel construction

作　　者：陈胜利 冯洋 杨艳群[2,3] CHEN Shengli;FENG Yang;YANG Yanqun(Qinghai Traffic Control Construction Engineering Group Co.,Ltd.,Xining 810000,China;College of Civil Engineering,Fuzhou University,Fuzhou 350116,China;Joint International Research Laboratory on Traffic Psychology&Behaviors,Fuzhou University,Fuzhou 350116,China)

机构地区：[1]青海省交控建设工程集团有限公司,西宁810000 [2]福州大学土木工程学院,福州350116 [3]福州大学交通心理与行为国际联合实验室,福州350116

出　　处：《安全与环境学报》2023年第12期4442-4448,共7页Journal of Safety and Environment

摘　　要：在隧道施工中,许多灾害威胁着工地现场施工人员的生命安全,一旦灾害发生,很难救援。但通过人员的早期干预,可以防止许多灾害的发生,这有赖于施工人员的危险预测能力。基于虚拟现实场景,研究了隧道施工人员的脑电数据与危险预测能力之间的关系,并通过自动机器学习框架比较了16种分类模型之间在危险预测能力上的优劣情况;然后,以获得的脑电数据集为基础,建立了Logistic回归分类模型,以判别施工人员危险预测能力的3个不同阶段。结果表明,该分类模型识别效果较好,F_(1)为0.6028。在分类识别中,最关键的特征是脑电δ波和γ波的功率谱密度,这与传统的危险知觉研究不同。提出的分类模型将有助于隧道施工人员更好地了解施工中危险源的变化,减少不安全行为,实现更大的安全性。To model the hazard prediction of tunnel construction workers,this paper studied the relationship between the Electroencephalogram(EEG) data of tunnel construction workers and hazard prediction.Twenty construction scenarios with hazard sources were simulated.These hazard sources were randomly selected and placed in the experimental scenario to be identified by the participants.Thirty participants(19 males and 11 females) participated in the experiment,aged 20 to 26 years(M = 23.8,S_(D) =1.2).Wearable wireless EEG devices and HTC VIVE VR equipment were used to collect the EEG signals in real-time when participants were conducting the experiments.In the experiment,the participants were asked to judge:the hazard cause,direction,and type.After the participants gave the judgment of the current hazard cause the experimenter recorded the result of the hazard cause judgment and the EEG data of the trial(manually controlled by the NIC 2.0 software).Each participant conducted 20 trials.At the end of one trial,the VR scenario was frozen for 30 s as a buffer.During this period,the participants prepared for the subsequent trial.The time limit for each trial was 2 minutes.Once the participants failed to give the judgment of hazard cause within 2 minutes,the next trial would start,and another hazard cause was randomly placed.After the EEG preprocessing,an EEG dataset with 587 data points and 79 features was obtained.Based on the EEG data,a logistic regression classifier for workers' hazard prediction in tunnel construction was built using machine learning.Furthermore,the performance of 16 classifier algorithms on classifying the three stages of hazard prediction was compared through an auto-machine learning framework.Finally,a logistic regression classifier based on EEG data was built.The results show that the classification model is effective,and the key features in classification and recognition are δ wave and γ wave power spectral density,which is different from traditional hazard perception studies.

关 键 词：安全人体学 事故预防 危险预测 隧道施工人员 自动机器学习 分类模型 脑电图 

分 类 号：X928[环境科学与工程—安全科学]