Regression and ANN-Based Prediction of Fire Temperature Curves for Single-Vehicle Fires in Underground Parking Structures

Vehicle fires in underground parking structures can cause rapid temperature rises, posing significant threats to both structural integrity and occupant safety. This study develops predictive models for quantitatively estimating fire temperatures in single-vehicle fire scenarios within underground parking structures. A total of 25 single-vehicle fire test datasets were collected, and the maximum heat release rate, time to reach maximum heat release rate, and duration of maximum heat release rate were analyzed for each vehicle type. Based on the analysis results, various fire scenarios were constructed, and fire simulations were performed to derive time–temperature curves for each fire scenario. From the simulated curves, regression-based and artificial neural network (ANN)-based prediction models were developed to estimate the peak temperature and the temperature at the end of combustion for each fire scenario, using heat release rate as input variables. The predictive performance of both models was validated against the fire simulation results, with coefficients of variation below 0.2 and coefficients of determination above 0.84, indicating high accuracy. Furthermore, to examine the applicability of the proposed time–temperature curves to structural fire analysis, nonlinear finite element analysis of reinforced concrete beams was performed for fire resistance assessment. The results showed that the structural fire behavior predicted using the regression and ANN-based curves closely matched that obtained using fire simulation-based curves. The proposed temperature prediction models are expected to be effective tools for estimating thermal loads and evaluating the fire safety of structures subjected to single-vehicle fires in underground parking facilities.