Prediction and Analysis of the Ultimate Bending Capacity of Circular Steel Tubes Based on Stacking MLP

Circular steel tubes are widely used in marine, chemical, nuclear power and land pipeline engineering. Their stress-deformation behavior under working conditions, especially the ultimate bending capacity (UBC), has been widely studied for a long time. This paper takes four parameters of circular steel tubes: thickness, tube diameter, steel yield stress, and elastic modulus as feature inputs, and the measured final bending moment as output. The dataset is sorted out from previous literatures, and divided into train-set and test-set in the proportion of 80% to 20%. An integrated neural network model StackingMLP is proposed, which is compared with support vector regression (SVR), Gaussian process regression(GPR), Multilayer Perceptron (MLP), XGBoost and LightGBM models. The $\mathbf{R}^{2}$ of the model on the train-set is 99.72%, and on the test-set is 99.32%. The model accuracy is the highest and the generalization performance is good. At the same time, this paper uses SHAP to interpret the model, revealing that the tube diameter has the greatest influence on the final bending moment, followed by the thickness.