•A novel FEM-ANN coupling dynamic prediction method is proposed.•FEM calculations of LSP-induced residual stress agree with experiment result.•ANN predictions correlated with the dynamic prediction accuracy are obtained.•Laser power density plays a significant role in LSP-induced residual stress. Laser shock peening (LSP) has been frequently used in the aerospace industry for improving the fatigue performance of the load-bearing components by introducing the beneficial compressive residual stresses into the structural materials. By taking the advantages of finite element method (FEM) and artificial neural network (ANN), the FEM-ANN coupling dynamic prediction method is proposed to evaluate the in-depth residual stresses induced by LSP of TC4 titanium alloy. The Python program-based three-dimensional parametric modellings for the repeated LSP associated with the circular-shape laser spot and the multiple LSP associated with the square-shape laser spot are carried out, respectively. The LSP parameters are randomly generated within the given ranges and are treated as the input layer of ANN model, and the in-depth residual stresses induced by LSP are regarded as the output layer. The raw data resulting from FEM calculations are employed to train, test and validate ANN model. Once the test or validation for ANN model fails, the raw data in the test set or validation set would be transferred into the training set for further training the networks. As a result, the prediction accuracy of ANN model could become increasingly higher with the increase of the raw data in the training set. The FEM-ANN coupling dynamic predictions correlated with the dynamic prediction accuracy are well consistent with the FEM calculations as well as the experimental data, indicating that the FEM-ANN coupling dynamic prediction method is feasible and effective to evaluate the LSP-induced residual stresses. It therefore provides a new way to predict the LSP-induced residual stresses with high efficiency and low cost.