A Study on the Decision Tree Approach for Predicting EO/EG Facility Leakage and Diffusion Hazards

Ethylene oxide (EO) and ethylene glycol (EG) are important materials in the petrochemical industry. However, EO/EG equipment is prone to fire and explosion accidents, which can have severe consequences. Based on the decision tree theory, a prediction model was developed to assess the consequences of wide-area accidents. 90 leakage scenarios of EO/EG facilities were defined, and these scenarios were simulated using FLACS software. This study utilized 80 sets of gas-phase concentration distribution data generated by FLACS under various leakage scenarios. The data set was divided into 60 samples for training the prediction model and 20 samples for validation. The decision tree prediction model accurately and rapidly predicted EO/EG leakage dispersion over an area exceeding 1 km², as confirmed by simulation comparisons. The model's prediction accuracy exceeded 90%. This study successfully achieved rapid prediction of EO/EG leakage dispersion patterns and concentration distributions using limited simulation data, providing critical support for risk prevention and control in EO/EG facilities.