Rapid Estimation Methods for Blast Loading on Typical Urban Building Clusters Considering Uncertainty Quantification

To quickly estimate the blast loading from strong blast waves in complex urban building clusters and quantify the uncertainty of the estimates, this study investigates rapid estimation methods based on Gaussian process regression, neural network ensembles, and Bayesian deep learning. The focus is on key locations for peak overpressure in a 3×3 building cluster, considering variables such as equivalent yield, distance to the blast center, and the dimensions of the streets and buildings. First, training data is obtained through orthogonal experimental design and numerical simulations. Then, Gaussian process regression, neural network ensemble, and Bayesian deep learning models are constructed to quickly predict blast loading while quantifying the uncertainty of the estimates. Finally, the models′ prediction accuracy and uncertainty quantification are evaluated using metrics such as mean absolute percentage error and normalized average prediction interval width. Results show that the average estimation errors for the peak overpressure across 144 sets of test data are 11.84%, 7.04%, and 8.51%, respectively, with confidence intervals covering the true values in over 86.11%, 98.61%, and 75.69% of cases. All three models can estimate the blast loading for a single test sample within 2.3 ms. The proposed methods provide a new approach and perspective for the rapid and reliable estimation of blast loading in complex urban scenarios.