建筑行业是我国国民经济的四大支柱之一，近十年来建筑行业的死亡人数不降反升，事故总量持续保持在高位，严重影响社会的和谐安定。通过统计分析发现，不安全行为是导致建筑行业安全事故发生的重要原因。为了深入研究疲劳状态对建筑工人不安全行为的影响，本文基于脑电实验开展对建筑工人的疲劳状态特征的研究，对一天内不同时间段人的脑电数据、行为数据和主观数据进行分析，并基于分析结果进行疲劳状态分类，在此基础上构建一维卷积神经网络进行建筑工人疲劳状态识别，达到减少因疲劳状态导致的工人在施工过程中的不安全行为的目的，为建筑行业的安全管理提出一种新方法。

本实验邀请了18名志愿者，仿照真实的建筑作业场景构建运动方案，设置不同时间段，让志愿者进行运动以模拟建筑工人作业情况；完成工作状态模拟后，依据Oddball范式诱发疲劳并采集志愿者脑电数据，同时使用E-prime软件收集行为数据，如反应时间、正确率等；在实验前后要求被试填写KSS量表以获得被试的主观疲劳数据，发现在整天三个不同时间段的实验过程中被试的疲劳度呈上升趋势。用KSS量表、平均反应时间、平均正确率三个指标来表征疲劳状态并使用层次聚类法进行分类，最终将疲劳状态分为活跃、清醒、正常、轻微疲劳、严重疲劳。通过对60个通道的功率谱密度值进行配对样本t检验，分析结果发现P5、P3、P6、PO5、PO6、PO8六个通道具有显著差异，因此选择上述六个通道的α/β、θ/β、(θ+α)/β、(θ+α)/(α+β)值作为脑电信号中疲劳状态的表征指标，同时结合平均反应时间、平均正确率、KSS量表得分等指标作为输入数据，选择五个疲劳等级作为输出结果，构建了一维疲劳状态识别卷积神经网络。之后将卷积神经网络、BP神经网络以及支持向量机三种分类方式进行对比，结果表明本文所构建的一维疲劳状态识别卷积神经网络模型的正确率更高，能够有效对建筑工人疲劳状态进行分类。

  本文构建了疲劳状态识别神经网络模型，结合脑电信号、行为数据、主观指标对人的疲劳状态进行识别，能够较为全面准确地反映人的疲劳状态，为建筑工人的疲劳检测和状态管理提供了依据，最终达到减少不安全行为发生的目的。

The construction industry is one of the four pillars of my country's national economy. In the past ten years, the death toll in the construction industry has not decreased but increased, and the total number of accidents has remained at a high level, which has seriously affected the harmony and stability of society. Through statistical analysis, it is found that unsafe behavior is an important cause of safety accidents in the construction industry. In order to deeply study the influence of fatigue state on unsafe behavior of construction workers, this paper studies the characteristics of fatigue state of construction workers based on EEG experiments, and analyzes the EEG data, behavior data and subjective data of people at different time periods in a day. And based on the analysis results, the fatigue state is classified, and on this basis, a one-dimensional convolutional neural network is constructed to identify the fatigue state of construction workers, so as to reduce the unsafe behavior of workers during the construction process caused by the fatigue state. A new approach to safety management is proposed.

In this experiment, 18 volunteers were invited to build a movement scheme based on the real construction operation scene, and set the movement modes in three different periods of the day to let the volunteers exercise to simulate the operation of construction workers; After completing the work state simulation, induce fatigue according to oddball paradigm and collect EEG data of volunteers. At the same time, E-Prime software is used to collect behavior data, such as reaction time, accuracy, etc; Before and after the experiment, the subjects were asked to fill in the KSS scale to obtain the subjective fatigue data of the subjects. It was found that the fatigue degree of the subjects showed an upward trend during the experiment in three different time periods of the whole day. Through KSS scale, average reaction time, average accuracy and other three indicators to characterize people's fatigue state, the hierarchical clustering method is used to classify, and finally the fatigue level is divided into five levels: active state, awake state, normal state, slight fatigue state and severe fatigue state. Through paired sample t-test on the power spectral density values of 60 channels, the analysis results show that there are significant differences among P5, P3, P6, PO5, PO6 and PO8. Therefore, the power spectral density values of the above six channels are selected α/β、θ/β、( θ+α)/β、(θ+α)/(α+β) Value is used as the characterization index of fatigue state in EEG signal. At the same time, combined with the indexes such as average reaction time, average accuracy and KSS scale score as the input data, five fatigue grades are selected as the output results, and a one-dimensional fatigue state recognition convolution neural network is constructed. Then the convolution neural network, BP neural network and support vector machine are compared. The results show that the one-dimensional fatigue state recognition convolution neural network model constructed in this paper has a higher accuracy and can effectively classify the fatigue state of construction workers.

This paper constructs a neural network model for fatigue state identification, which can identify people's fatigue state combined with EEG, behavior data and subjective indicators, which can comprehensively and accurately reflect people's fatigue state, provide a basis for construction workers' fatigue detection and state management, effectively carry out state early warning for workers in fatigue state, and finally achieve the purpose of reducing unsafe behavior.