水电工程施工安全隐患多标签文本智能分类方法

Multi-label text intelligent classification method for construction safety hazards of hydropower projects

水利水电工程施工安全隐患体量大、形式多元、类型多样,同一隐患可能涉及多个类型,且隐患类型的界定存在模糊不清的现象。隐患分类多以人工经验为主导,极易导致隐患管理混淆,增加了隐患管理的难度。针对上述问题,本文提出了一种水电工程施工安全隐患文本多标签智能分类方法。首先,利用ALBERT模型对文本信息进行编码,实现非结构化安全隐患文本的高精度量化;然后,以文本量化结果为基础,考虑安全隐患中文本内容权重,利用Attention机制改进的双向门控循环单元(Bi-GRU),构建安全隐患文本多标签智能分类模型,提升施工安全隐患识别效率;最后,利用水电工程施工安全隐患文本,测试方法性能,得到方法的F1值达到了92.11%,证明了该方法的适用性,有望为水电工程施工的安全管理、事故隐患排查和分析提供信息支撑。

In water conservancy and hydropower projects,construction safety hazards come large in volume and are usually diversified in forms and types.The same hidden danger may involve multiple types,but the definition of its types is ambiguous;previous classification of hazard types is mostly dominated by manual experience,easily leading to confusion and difficulty in hazard management.To address such issues,this paper presents a multi-label intelligent identification method for construction safety hazards in hydropower projects.First,the ALBERT model is used to encode text information to achieve a high-precision quantification of the unstructured risk texts.Then,we construct a multi-label intelligent classification model of safety hazards text to improve identification efficiency,considering the content weight of the Chinese text for safety hazards and using the Bidirectional Gated Recurrent Unit(Bi-GRU)improved by the Attention mechanism.Finally,the performance of this method is tested using the texts of hydropower engineering construction safety hazards,verifying its F1 value reaches 92.11%compared with previous text classification methods.It is proved applicable,as a useful support to safety management of hydropower construction and analysis of its safety hazards.