Abstract: To tackle the problems of low automation, difficult quality assurance, and insufficient construction efficiency in nuclear power pipeline welding, this paper proposes an intelligent pipeline welding system based on data-driven decision-making. This system is developed on the basis of the 23HY-Arc400-P full range pipeline automatic welding equipment of CNNC-23. It integrates a synergistic system of active and passive vision sensors and constructs a typical groove welding process database. The development of an automatic planning system for the layer arrangement of the welding process is completed: The precise recognition of weld bead width is achieved by combining passive vision with U-Net neural network.The 3D morphology parameters of the weld bead are extracted based on active vision, and the collaborative fusion of active and passive vision coordinates is realized through camera calibration.The "equal area filling algorithm" and the mathematical model of weld bead arrangement are adopted to complete the pre-arrangement planning of multi-layer and multi-pass welding. Furthermore, trajectory design and differential compensation strategies for weld width and height deviations are implemented, integrated with a position-distance dual closed-loop control system to enable real-time deviation correction during welding. Experimental results demonstrate that the system can accurately extract key weld parameters such as bead width and height, effectively plan weld bead deposition, and perform real-time correction of welding deviations. These capabilities contribute to enhanced automation and improved weld quality in nuclear power pipeline construction, offering technical support for large-scale nuclear energy projects.