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.