The water diversion project in central Yunnan is a large-scale engineering project related to Yunnan people’s livelihood, in which the welding quality of steel pipe directly affects the overall safety of the project. The traditional welding process parameters optimization method is inefficient and complex, and it is difficult to achieve multi-objective optimization. In order to optimize the welding process parameters of steel pipe in Yunnan-China water diversion project and improve the project quality, this study proposes a multi-objective optimization method based on the improved particle swarm optimization algorithm. The multi-objective optimization of steel pipe welding process parameters was realized by establishing the correlation model between welding power, welding speed, defocusing amount, swing amplitude, swing frequency and performance indexes such as melting depth, depth-towidth ratio, porosity, etc., combined with the genetic optimization neural network prediction model. The results show that the optimal combination of process parameters obtained by the improved particle swarm algorithm is 4.75 kW of welding power, 2.63 m/min of welding speed, 1.05 mm of out-of-focus, 1.78 mm of swing, and 90 Hz of swing frequency. The welding melt depth obtained by using this parameter combination reaches 8.915 mm, with a depthto-width ratio of 0.729 and a porosity of only 1.137%, which is better than the traditional optimization algorithm in all performance indicators. In engineering practice, the welding process after applying the improved particle swarm algorithm realizes the qualification rate of 99.88% for steel pipe weld seam at one time, the qualification rate of 99.91% for longitudinal seam, 99.85% for circumferential seam, and the qualification rate of 100% for delivery inspection. The study proves that the improved particle swarm algorithm has the characteristics of high efficiency and high precision in the optimization of steel pipe welding process parameters, which can provide reference for similar projects.
