Intelligent Prediction of Stability Bearing Capacity of New-Type Modular Assembled Latticed Shells with Flange Connections

Modular assembled reticulated shell structures have the advantages of high construction efficiency and good join quality, and have a wide application prospect. Based on machine learning method, an intelligent prediction model for the bearing capacity of modular prefabricated reticulated shell structures was established. Firstly, 864 finite element models of modular assembled reticulated shells were analyzed for a series of parameters that affect the stable ultimate bearing capacity, thus generating the database required for the machine learning algorithm. Secondly, six machine learning algorithm models were established based on the open source platform Scikit-learn, and all algorithm models were trained and tested by using the generated database. In addition, the artificial neural network model (ANN), XGBoost and gradient enhancement algorithm models were overfitted, and the reliability of ANN model was tested. The results showed that the determination coefficients (R²) of ANN, XGBoost and gradient enhancement algorithm models in the test set were all greater than 0.95, and the prediction accuracy of bearing capacity was very high. The ANN model had the best robustness and accuracy in predicting the stable ultimate bearing capacity, with an average absolute percentage error (MAPE) of 7.1% and an R² of 0.982. It showed high prediction accuracy and generalization capacity, and could well capture the complex mapping relations between the ultimate bearing capacity and input parameters.