The purpose of this study is to explore the key risk factors and accident formation mechanisms for different levels and types accidents (DLTAs) in mega construction projects (MCPs). Furthermore, by pinpointing key risk factors and causal chains of DLTAs, the research outcomes provide theoretical foundations for accident prevention and prediction.
This study extracts text semantic features by coupling Word2Vec, latent Dirichlet allocation (LDA) topic model and the Bayesian network model to delve into the intricate couplings among risk factors of DLTAs. First, based on the Word2Vec, LDA topic model to identify safety risk factors. Second, association rule mining is employed to uncover the relationships among the DLTAs risk factors. Finally, the Bayesian network is constructed, leveraging the results from text mining and association rule mining, to delve into the key risk factors and the underlying mechanisms of DLTAs formation.
(1) 28 secondary risk factors, 48 rules associating risk factors with accident levels and 41 rules associating risk factors with accident types were identified. (2) The more severe the accident level, the more risk factors are involved, and timely measures should be taken to prevent the interaction of multiple risk factors. The key causal chains differ among different types of accidents, thus prevention and control measures, as well as emergency response plans, should be formulated specifically for each type of accident. (3) The key risk factors for MCPs include: Low safety awareness, Wildcat operation, Improper safety protection, Improper maintenance of equipment, Linkage fault safety device failure, Poor geological conditions, Poor ventilation conditions, Improper on-site risk identification, Improper management of safety hazards, Improper safety inspection, Improper safety education and training, Safety responsibility is not implemented, Improper special construction scheme.
MCPs are characterized by high complexity, and their construction sites are fraught with numerous risk factors, potentially leading to DLTAs. These accidents, differing in their characteristics, pose significant challenges to the safety management and control of MCPs. Notably, there is a conspicuous lack of focus on the specific characteristics of DLTAs, which represents a significant gap in the field.
The contributions of this study are threefold. First, this study focuses on DLTAs, which are an important perspective on safety management. Second, based on text mining and association rule mining to build a Bayesian network, providing guidance for risk coupling mining. Third, understanding the characteristics of DLTAs plays a crucial role in pre-accident prevention, in-accident rescue and post-accident disposal.
