Comparative Analysis of Neural Network Models for Predicting Battery Pack Safety in Frontal Collisions

Amid concerns about environmental degradation and the consumption of non-renewable energy, the development of electric vehicles (EVs) has accelerated, with increasing focus on safety. On the road, battery packs are exposed to potential risks from unforeseen objects that may collide with or scratch the system, which may lead to damage or even explosions, thus endangering the safety of transportation participants. In this study, several predictive models aimed at assessing the safety performances of battery packs are proposed to provide a basis for data-driven structural optimization by numerically simulating the deformation of the battery base plate. Initially, a finite element model of the battery pack was developed, and the accuracy of the model was verified by performing modal analysis with various commercial software tools. Then, representative samples were collected using optimal Latin hypercube sampling, followed by collision simulations to gather data under different collision conditions. Next, the prediction accuracy of three models—PSO-BP neural network, RIME-BP neural network, and RBF neural network—was compared for predicting battery pack bottom shell deformation. Finally, the prediction accuracy of the models was compared based on error functions. The results indicate that these neural network models can accurately predict deformation under frontal collision conditions within the specified limits, with the RIME-BP model yielding the best performance beyond those limits. The developed neural network prediction model is able to accurately assess the mechanical response of battery packs under frontal collision, providing support for data-driven structural optimization. It also provides an important reference for improving the safety and durability of battery pack design.