Abstract:

In order to mine potential risk information in reports of hazardous chemical explosion accidents, and enhance public awareness of risks when obtaining information on social media, an intelligent model for risk classification of hazardous chemical explosion accidents was constructed using text mining technology, risk assessment, and deep learning models. Firstly, the key information was extracted from accident investigation reports (e.g. hazardous chemicals, causes of the accident, consequences of the accident, etc.), and risk levels were assigned to the reports according a risk assessment matrix. Secondly, in response to the long-tail distribution characteristics of hazardous chemical accident data, a fusion strategy driven by imbalanced few-shot data was proposed, and four models such as Bidirectional Encoder Representations from Transformers (BERT), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU) were used for comparative verification to test the effects of enhancement strategies. Finally, an accident risk rating model was constructed based on a BiGRU neural network with a risk-factor perception mechanism. The results show that rating model can perform automatic risk assessment for hazardous chemical explosion incidents while enhancing public risk awareness capabilities.

Key words: hazardous chemical explosion accidents, risk rating model, few-shot data-driven, bidirectional gated recurrent unit (BiGRU) neural network, risk perception, attention mechanism