With the deepening of oil exploration and development, wellbore conditions have become increasingly complex, and casing wear has become one of the major threats to well integrity. Wear leads to a reduction in the residual strength of the casing, directly affecting the safety and development efficiency of oil and gas wells. Existing methods for evaluating casing residual strength based on physical models or finite element numerical simulations struggle to balance accuracy and efficiency. To overcome this bottleneck, a data-driven approach, combined with artificial intelligence theory, an intelligent prediction model integrating BPNN-Optuna-XGBoost was proposed. This model first constructed a multi-size, multi-defect casing residual strength database through numerical simulation. It then introduced a back propagation neural network (BPNN) to achieve precise mapping of the residual strength distribution across the entire well section of a single well. Furthermore, by combining Optuna hyperparameter optimization and the XGBoost(extreme gradient boosting)algorithm, an integrated learning framework incorporating multi-well data was constructed to enable transfer prediction of residual strength for target wells in the same block. Experimental results show that the BPNN model has a prediction error of less than 5%, and the Optuna-XGBoost model achieves a goodness-of-fit of up to 0.996 4. The result provides a more efficient and accurate theoretical model and tool for evaluating casing residual strength under complex working conditions.