Skip to Content
BuildingsBuildings
  • Article
  • Open Access

17 March 2026

19 Pages

A Study on AI-Empowered Behavior Risk Identification and Early Warning in Nuclear Power Engineering Construction

,
,
,
,
,
and
1
China Nuclear Power Engineering Co., Ltd., Beijing 100840, China
2
College of Safety and Environmental Engineering, Shandong University of Science and Technology, Qingdao 266590, China
*
Author to whom correspondence should be addressed.
This article belongs to the Special Issue Human Factor on Construction Safety

Abstract

Any risks arising during the construction phase of nuclear power projects become permanently embedded in the power station’s lifecycle, evolving into inherent and difficult-to-alter potential hazards. Consequently, identifying behavioral risks in this phase is critical to the successful delivery of nuclear power engineering projects. This paper proposes a behavior risk identification and early warning methodology for nuclear power construction operations based on artificial intelligence algorithms. The research employs text mining techniques to construct a risk indicator system for nuclear power construction operations; based on the You Only Look Once (YOLOv8) algorithm, it incorporates modules such as Deformable Convolutional Network (DCN), Generalized Lightweight Attention Network (GELAN), Efficient Channel Attention (ECA), and Atrous Spatial Pyramid Pooling (ASPP) to develop the DCN -GELAN-ECA- ASPP-YOLO for Nuclear Power Engineering (DGEAYoLo-NPE) model, and designs and develops a supporting behavior risk identification and early warning methodology. Results show that the precision of nuclear power construction behavioral risk detection reaches 94.3%, with a 2.2% improvement in precision. This study confirms that artificial intelligence technology can effectively enhance the behavior risk prevention and control capabilities of nuclear power construction operations.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.