Parametric FEM–CLPSO–LightGBM Surrogate Model for Deformation Prediction of Casings in Creep Formations

The conventional analysis of casing deformation relies on engineering logging, analytical models, or repeated finite element modeling and simulation for different wells or field restricts, which may involve high costs, simplifying assumptions, and low computational efficiency. To improve casing deformation and casing configuration analysis in creep formations, a surrogate model is proposed. A dataset is generated through the parametric finite element modeling of the casing–cement sheath–creep formation system, and input features are optimized by feature engineering. Light Gradient Boosting Machine (LightGBM) is adopted as the prediction model, while Comprehensive Learning Particle Swarm Optimization (CLPSO) is employed for hyperparameter optimization. The surrogate model is further used to generate casing deformation contour plots. Algorithm comparisons and finite element simulations are performed to evaluate the prediction accuracy and computational efficiency. The results show that the RMSE for casing nodes in the test set is 0.0825 mm, and the average overlap ratio between the predicted and finite element deformation intervals reaches 98.42%. The predictions show good agreement with the finite element results in terms of the casing deformation magnitude and casing configuration. Compared with conventional analytical models and repeated finite element model construction, the proposed model reduces the dependence on simplifying assumptions, supports the visualization of casing configurations, and improves the computational efficiency.