基于深度学习的不安全因素识别和交互分析

发布时间：2018-02-25 16:25

  本文关键词： 深度学习 行为安全 人-机-环 不安全因素 交互分析　出处：《中国安全科学学报》2017年04期 　论文类型：期刊论文

【摘要】：为解决行为安全领域不安全因素识别和交互分析困难的问题,构建基于深度学习的不安全因素识别和交互分析模型。首先,从"人-机-环"3方面构建不安全因素识别层,分别采用不同的深度学习结构识别作业人员行为属性、工作环境场景和操作设备工作状态的不安全因素;然后,通过因素交互层,采用关联和回归多值算法完成对不安全因素的交互分析;最后,通过输出显示层实现分析结果的表征。以某煤矿综采、掘进、通风3个生产活动类别的视频音频数据为例,通过Matlab操作平台选取最优深度学习结构,进行模型交互分析。结果表明,用该模型能实现对采煤面空顶作业、喷浆机故障仍然加料、主要通风机异常响动未停机检查等不安全因素的识别和交互分析,完成不安全行为的描述以及风险分级、行为痕迹的分类。
[Abstract]:In order to solve the problem of identification and interaction analysis of unsafe factors in the field of behavioral security, a model of identification and interaction analysis of unsafe factors based on in-depth learning is constructed. Firstly, the identification layer of unsafe factors is constructed from the three aspects of "man-machine-ring". Different depth learning structures are used to identify the unsafe factors of operator behavior attributes, work environment scene and operating equipment working state, and then, through the factor interaction layer, The interaction analysis of unsafe factors is accomplished by using association and regression multi-valued algorithms. Finally, the analysis results are represented by output display layer. Taking the video and audio data of three production activity categories of a coal mine as an example, such as fully mechanized mining, tunneling and ventilation, The optimal depth learning structure is selected through Matlab operating platform, and the model interaction analysis is carried out. The results show that the model can be used to realize the work of coal mining surface with empty roof, and the ejector fault can still be fed. The identification and interactive analysis of unsafe factors such as abnormal noise and unshutdown check of main ventilator are carried out to describe unsafe behavior and to classify risk classification and behavior trace.
【作者单位】： 中国矿业大学(北京)资源与安全工程学院;
【基金】：国家自然科学基金资助(51674268)
【分类号】：TD771
，

本文编号：1534288