This study presents a reliability analysis of cylindrical shell structures utilizing an improved Backpropagation (BP) neural network algorithm. Traditional BP neural networks often struggle with slow convergence and the risk of falling into local minima, which can undermine the accuracy of structural reliability predictions. To address these issues, an enhanced version of the BP neural network is proposed, which incorporates optimization techniques such as particle swarm optimization (PSO) to fine-tune the network’s parameters. The proposed method is applied to model the structural reliability of cylindrical shell structures under various load conditions. The results demonstrate that the improved BP algorithm significantly enhances the convergence speed and prediction accuracy, offering a more reliable assessment of the structure’s performance under uncertain loading and environmental factors. This method provides a promising tool for the design and safety evaluation of cylindrical shell structures in engineering applications.