Fiber and semiconductor composite laser welding has been widely used in the field of laser precision joining due to its ability to give full play to the unique advantages of fiber laser and semiconductor laser heat source. However, the laser process multi-parameters coupled with each other, the impact on the actual welding quality has not been clear, how to accurately predict the weld formation and process optimization is the key to improve the welding quality. Therefore, for the power battery 3003 aluminum alloy, this paper carries out fiber and semiconductor composite laser welding process test, systematic research on the effect of different energy ratios on the weld morphology of the law, and to elaborate the formation mechanism. The results show that the fiber laser mainly changes the keyhole absorption efficiency of laser energy, directly affecting the depth of fusion, while the semiconductor laser mainly determines the surface fusion width and internal effective fusion width. Meanwhile, a weld shape prediction model based on gradient boosted regression tree GBRT is constructed using laser process parameters as multidimensional input vectors, and its root mean square error (RMSE) is lower than 20%, with high prediction accuracy and strong generalization ability. Finally, the constructed prediction model realizes the optimization of laser welding process, and good welding quality can be obtained when the fiber laser power is 0.61.0 kW and the semiconductor laser power is 1.62.0 kW. This paper provides key theoretical support and guidance basis for fiber and semiconductor composite laser welding quality prediction and process optimization.