For the selection of the double ellipsoid heat source parameters, the thermal cycle of different process was measured and compared with the simulation result. The experimental results were according to the actual conditions and the simulation results were obtained by finite element method. The sensitivity of heat source parameters was discussed. Pattern search method was used to obtain the most accurate parameters corresponding to the specific process. By summarizing all of the processes of experiment, the relationship between the experimental process parameters and the corresponding double ellipsoid heat source parameters was obtained. The artificial neural network algorithm was applied to predict the relationship between all possible process and the double ellipsoid heat source parameters. The verification experiment showed that the prediction model was accurate.