Abstract: This paper addresses the welding deformation control of large aluminum alloy profile structures in high-speed train body floors by proposing a welding deformation prediction and optimization method based on local-global mapping. A stepwise numerical simulation strategy is adopted: First, a three-dimensional finite element model of a typical weld joint is established, and precise simulation of the welding temperature field is achieved through a moving heat source loading method. Guided by the temperature similarity principle, the thermal cycle characteristic parameters of the local model are mapped to the global floor structure, establishing a multi-physics coupling model that considers material nonlinearity and contact nonlinearity, thereby significantly improving simulation efficiency for large-scale structures. The modeling process systematically integrates composite constraints from mechanical clamps, hydraulic clamps, and gravity fixtures, while incorporating process parameters such as structural self-weight and welding sequence to establish boundary conditions consistent with actual working conditions. Numerical results indicate a maximum welding deformation of 3.5 mm in the floor structure, with an average relative error below 15% compared to field measurements, validating the model's reliability. Further parametric analysis reveals the influence mechanism of fixture constraint layouts on deformation distribution, with the optimized fixture configuration reducing deformation in critical regions over 50%. This method overcomes computational efficiency limitations inherent in traditional global modeling approaches, providing a novel technical pathway for deformation prediction and control in large-scale complex welded structures.