基于生理信号的消防员体力疲劳识别  

Physical fatigue recognition of firefighters based on physiological signals

作　　者：付文卓 郑欣[1] 张放 FU Wenzhuo;ZHENG Xin;ZHANG Fang(School of Resources and Civil Engineering,Northeastern University,Shenyang Liaoning 110819,China;CNPC Research Institute of Safety&Environment Technology,Beijing 102200,China)

机构地区：[1]东北大学资源与土木工程学院,辽宁沈阳110819 [2]中国石油安全环保技术研究院有限公司,北京102200

出　　处：《中国安全生产科学技术》2025年第3期218-226,共9页Journal of Safety Science and Technology

基　　金：国家重点研发计划项目(2021YFC3001300)。

摘　　要：为研究消防队员体力疲劳状态的评测指标并进行疲劳识别,以体能训练诱导消防员体力疲劳,并以时间知觉分析方法确定体力疲劳是否产生,采集18名被试消防员在日常训练诱导疲劳前后的脑电、肌电和眼动信号进行疲劳识别研究。首先采用配对t检验、秩和检验对采集到的脑电、肌电和眼动信号进行统计分析,7种肌电指标和1种眼动指标在体力疲劳状态下发生显著变化。其次,使用最小冗余最大相关性(mRMR)算法和ReliefF算法对初选的生理指标进行特征优选,AEMG、Median F、Mean F、Mean P和Total P这5项肌电指标和眼动指标中的瞳孔直径为指标降维优选结果。最后,基于特征优选后的生理指标,使用Logistic Regression、Random Forest、Support Vector Machine、XG-Boost和K-Nearest Neighbors机器学习方法开展疲劳识别对比分析。研究结果表明:基于ReliefF算法优选后的指标采用Random Forest机器学习方法对消防队员的体力疲劳识别性能最好(ACC=0.943,SN=1.000,SP=0.882,PR=0.900,F1=0.947,AUC=0.971)。研究结果可为有效识别消防队员体力疲劳和制定合理的日常训练计划提供参考。In order to study the evaluation indexes for the physical fatigue status of firefighters and carry out the fatigue recognition,the physical training was used to induce the physical fatigue of firefighters,and the time perception analysis method was used to determine whether the physical fatigue occurred.The EEG,EMG,and eye movement signals of 18 firefighters before and after the daily training induced fatigue were collected for fatigue recognition research.Firstly,the paired t-test and rank sum test were used to statistically analyze the collected EEG,EMG,and eye movement signals,and seven EMG indexes and one eye movement index presented significant change under the physical fatigue status.Secondly,the minimum redundancy maximum relevance(mRMR)algorithm and ReliefF algorithm were used to optimize the features of the primary selected physiological indexes.The five EMG indexes of AEMG,Median F,Mean F,Mean P,and Total P,as well as the pupil diameter in the eye movement indexes,were used as the optimization results for index dimensionality reduction.Finally,based on the physiological indexes selected through feature optimization,a comparative analysis of fatigue recognition was conducted using the machine learning methods of Logistic Regression,Random Forest,Support Vector Machine,XG Boost,and K-Nearest Neighbors.The results show that the indexes optimized based on the ReliefF algorithm have the best performance in the physical fatigue recognition of firefighters using the Random Forest machine learning method(ACC=0.943,SN=1.000,SP=0.882,PR=0.900,F1=0.947,AUC=0.971).The research results can provide a reference for effectively recognizing the physical fatigue of firefighters and formulating reasonable daily training plans.

关 键 词：疲劳识别 特征选择 机器学习 生理信号 消防队员 

分 类 号：X912[环境科学与工程—安全科学]