Abstract: In order to perform a comprehensive and real-time objective evaluation of the status quo of enterprise work safety and to avoid deviation caused by manual intervention in traditional risk evaluation methods, firstly, a barrier-consequence hierarchical key indicator system was proposed according to accident cause theory, and indicators were quantified through mining structured data such as accidents and unstructured data such as checking text. Then, PCA and EW were adopted for data dimensionality reduction and risk evaluation, and exponential smoothing method was used to predict risk level of enterprises in the next cycle before an risk evaluation and early warning model was established. Finally, the model was applied to two oil and gas production enterprises. The results show that this risk evaluation model can accomplish assessment of current risks and prediction of risks in future cycles through acquisition, processing and learning of barrier-consequence historical data, and provide guidance for enterprises' safety management (risk reduction, risk elimination, and that of chemical hazard barriers, etc.).

Key words: enterprise word safety, key index system, risk evaluation, text matching, principal component analysis (PCA), entropy weight method (EW)