Abstract: In order to find the best welding process parameters for resistance spot welding of ultra-high strength steel, a three-factor and five-level flat plate lap spot welding experiment designed by orthogonal test method was carried out. With welding time, welding current and electrode pressure as adjustable process parameters, the nugget diameter, indentation depth, the tension-shear strength and spatter were used as the quality evaluation indicators of welded joints. Based on Gaussian process regression and BP neural network, a proxy model of the relationship between the process parameters and the quality evaluation indicators of welded joints was established. The training results showed that the accuracy of the model was very high. Finally, the multi-objective optimization was realized by using the genetic algorithm NSGA-Ⅱ with elite strategy and non-dominated sequencing, and the optimal pareto solution set between the evaluation indicators was obtained. The relative error value of each evaluation model was very small, which indicated that the optimization method had good prediction effect and stability. By using less experimental data, the method of establishing the optimization model had important guiding value for the selection of the best welding process parameters in resistance spot welding and other welding fields.