Abstract: Objective In the strategic context of automotive lightweighting, steel/aluminum hybrid structures are increasingly being employed in vehicles, yet the significant differences in physical properties between steel and aluminum make it extremely challenging to fuse them seamlessly. Therefore, it is crucial to judiciously select welding process parameters to enhance the performance of welded joints. Methods A CNN regression model is eatablished based on orthogonal experimental design and simulation results obtained using Visual environment. The model takes laser power, welding speed, and laser spot diameter as inputs, with residual stress and the amount of deformation in the z-direction of the base material as outputs. Two such regression models are then utilized as objective functions within a MOGWO algorithm to explore the optimal welding parameters. Results The errors between the predicted values from the CNN regression model and the simulated values are 1.76% and 3.30%, both falling below 4%. The MOGWO algorithm yields the optimal result when the laser power is 706.61 W, the welding speed is 15.00 mm/s, and the spot diameter is 0.91 mm. Compared to before optimization, the residual stress is reduced by 7.80%, and the deformation in the z-direction decreases by 24.17%. Conclusion The CNN model can accurately describe the nonlinear relationship between welding parameters and performance, and its integration with the MOGWO algorithm enables the identification of optimal welding parameters, thereby enhancing the performance of steel/aluminum dissimilar metal laser welded joints.