Risk recognition and prediction of fire and explosion in oil and gas stations based on neural-backed decision trees
To effectively prevent and control fire and explosion incidents at oil and gas facilities,this study proposes utilizing the Neural-Backed Decision Tree algorithm(NBDT)to construct an interpretable predictive model.This approach considers the causal relationships among influencing factors,offering insights into the occurrence of fire and explosion incidents at these facilities.The model facilitates visual analysis of the risks and inducements associated with such incidents by leveraging data.Initially,this method utilizes tokenization techniques and Term Frequency-Inverse Document Frequency(TF-IDF)to perform text analysis on 13 primary risk factors,leading to the identification of 64 secondary risk factors.These 64 secondary risk factors are then binary-encoded to construct a dataset for fire and explosion incidents at oil and gas facilities.Subsequently,the study employs the NBDT algorithm to construct a classification model and further optimizes it.The established model is subsequently utilized to predict and interpret fire and explosion incidents at oil and gas facilities.Based on actual data from fire and explosion incidents at oil and gas facilities,this study conducts comparative experimental analyses,comparing the NBDT method against Artificial Neural Networks(ANN)and CART decision tree models.Subsequently,the study utilizes the NBDT model training results to create a visual decision tree diagram.Additionally,single-factor and multi-factor interpretative analyses are conducted based on the mechanisms of fire and explosion incidents,analyzing the causes,and proposing corresponding measures.The experimental results indicate that,compared to ANN and CART decision trees,the NBDT model exhibits higher accuracy of 0.976,precision of 0.985,recall of 0.971,and AUC of 0.913 values on both the training and testing sets.Through single-factor analysis,the NBDT decision tree identifies seven primary control factors for secondary risks.Through multi-factor analysis,seven combinations of secondary risk factors are identified.Based on the model's decision-making mechanism,six primary control factors for these combinations are determined.In total,single-factor and multi-factor analyses collectively establish 13 primary control factors and 14 risk factors(combinations).The identification of these 13 primary control factors provides theoretical support for existing fire and explosion prediction and prevention mechanisms at oil and gas facilities.
safety engineeringfuel air explosiverisk factorsassociation rulesinterpretabilityNeural-Backed Decision Tree(NBDT)
NETL
NSTL
万方数据
