Intelligent Identification Approach for Accident Causation in Hydraulic and Hydropower Engineering Construction
Jiayi Zhou, Xiazhong Zheng, Fei Wang, Dan Tian · Journal of Construction Engineering and Management · 2025
Analyzing historical accident reports is crucial for improving construction safety in hydraulic and hydropower engineering. Current methodologies for identifying accident causation exhibit two limitations: (1) text mining techniques excel at keyword extraction but fail to capture contextual semantic nuances; and (2) a single deep learning model trained on labeled accident data sets inadequately captures causative information, particularly in analyzing specific terms, phrases, and narrative context. These limitations result in reduced accuracy in identifying accident causation. To address the above limitations, this study proposes a novel intelligent accident causation identification approach tailored to hydraulic and hydropower construction safety management. Firstly, a RoBERTa pretrained language model is adopted to encode accident texts into word-level vector representations. Secondly, a feature mining network (FMNet) is developed by combining convolutional neural networks (CNN) with multilayer long short-term memory (LSTM), enabling the extraction of causative features from both local and global perspectives. Thirdly, a self-attention mechanism is utilized to dynamically allocate weights to the extracted features, enabling multifeature fusion and generating the final accident causation results. Finally, the effectiveness of the proposed approach is validated using evaluation metrics. Through comparative and ablation experiments, the results demonstrate that our approach outperforms existing methods, achieving higher accuracy, precision, recall, and F1 scores. Furthermore, an intelligent identification system for accident causation has been developed based on the proposed approach. Managers can utilize the system to efficiently and accurately identify accident causation and develop prevention measures in a timely manner, providing a sophisticated approach for improving the safety management level in real-world scenarios.