In the ship hull plate welding process, different welding sequences directly affect the deformation of the current welding procedure, which in turn impacts the overall shipbuilding accuracy. This study takes a typical double T-shaped thin plate structure as an example. Based on welding numerical simulation and experimental validation, a corresponding dataset is obtained. To address the issue of BP neural networks being prone to local optima, which can lead to inaccurate results, a Simulated Annealing-Back Propagation (SA-BP) neural network model is used to analyze the dataset. The research aims to determine the optimal welding sequence that minimizes deformation. The training results show that the Mean Squared Error (MSE) of the SA-BP model decreased from 1.0144 in the BP model to 0.67388. Additionally, the SA-BP model's fitting performance is far superior to that of the BP model. Therefore, the SA-BP neural network model provides more stable and accurate results compared to the traditional BP neural network model. The comparison of the optimal welding sequence results derived from both models shows that welding with the optimized SA-BP neural network results in a 21.07% reduction in welding deformation compared to the traditional BP neural network.