A novel quantitative risk assessment method for storage tank fires based on numerical modelling and machine learning

Tank fires are common hazardous scenarios in storage facilities that present serious risks to both industrial safety and environmental protection in liquid energy storage systems. In this study, a novel method was developed combining numerical modelling and machine learning (ML) for prediction of radiative heat flux and quantitative risk assessment of tank fires. Tank fires with different tank diameters were simulated (11 sets of numerical simulations, covering diameters from 0.4 m to 75 m) and the predicted radiative heat flux was analyzed and compared with that calculated from traditional analytical models. The simulation data (approximately 3000 indivadual data points) were then used as the inputs into the extreme gradient boosting (XGBoost) model to predict the radiative heat flux. Six key hyperparameters in the XGBoost model were identified and optimized. A quantitative risk assessment method was proposed to predict the thermal hazards of tank fire based on the threshold method and probabilistic method. This developed method was validated in a case study involving one or two tank fires to assess its thermal impact on the surrounding equipment and subsequently the minimum spacing requirements by different national standards. Calculated yielded the safe rescue distance corresponding to the radiation threshold (single tank fire, 4kW/m 2 : 133.2 m, two tank fires, 4 kW/m 2 : 180 m) and probability of accident escalation. This research proposes a computationally efficient approach for predicting radiative heat flux and quantitatively assessing risks associated with tank fire accidents.