This invention discloses a method and
system for rapid prediction of
battery pack collision intrusion based on multi-fidelity data fusion. First, a training dataset containing high-fidelity and low-fidelity samples is constructed through finite element simulations with different mesh accuracies. Then, a two-
level fusion model is established using
Gaussian process regression. The predictions of the low-fidelity model are used as correction features to guide the high-fidelity model in accurate learning. Furthermore, based on an
ensemble learning and
Bayesian optimization framework, an improvement expectation criterion that incorporates
simulation cost trade-offs is used for adaptive sampling. The location and accuracy level of newly added samples are intelligently determined, and the model is continuously optimized through iterative updates. This invention effectively resolves the contradiction between high-precision prediction and high computational cost, significantly reducing
data acquisition overhead while ensuring prediction accuracy, providing an efficient and reliable technical means for
battery pack collision safety assessment and optimization design.