In order to realize the intelligence and flexibility of V- groove robot welding of medium-thick plates, a multi-layer welding process control strategy for welding with a weaving robot is proposed based on three-wire structural light vision. For the optical localization of the welding start point, a “rough first, accurate later” guidance approach is proposed. A technique utilizing Gaussian-weighted plane fitting is employed to model the groove and extract information about the weld shape. The welding torch’s orientation planning is based on the angular bisector of the groove, and the GA-BP neural network is utilized to forecast welding parameters for various welding techniques. It is shown that the strategy in this paper can guide the robot to complete the welding job. The error between the guide start point and the actual welding start point does not exceed 2 mm, and the predicted welding parameter results align with the welding technology standards.