Strength reduction factor prediction model for non-structural components based on ensemble algorithm with Bayesian optimization

doi:10.1016/j.istruc.2025.110901

Return

Strength reduction factor prediction model for non-structural components based on ensemble algorithm with Bayesian optimization

2025-12-01

PRE

Jia, Jia

Subramaniyan, Bharathiraja *

Zhanxuan Zuo

Liu, Bo

DOI：10.1016/j.istruc.2025.110901

Original

Original request for help

Save

Abstract

En 中文

Non-structural components may be subjected to strong seismic actions, leading to nonlinear deformation. Quantitative investigations into the influence of NSCs' nonlinearity on seismic response remain limited. This study calculates the strength reduction factor using structural floor seismic response records from instrumented buildings of Center for Engineering Strong Motion Data (CESMD) during various earthquakes, followed by a comprehensive statistical analysis of the influencing factors. Predictive models for the strength reduction factor are developed using ensemble learning algorithms, namely extreme gradient boosting (XGBoost), random forest (RF), light gradient boosting machine (LightGBM), Gradient Boosting Decision Tree (GBDT), and Adaptive Boosting (AdaBoost). The hyperparameters are optimized using the Bayesian optimization algorithm. Validation results on the test set show that the XGBoost model achieves the best prediction performance. Interpretability techniques are employed to assess the influence of each input parameter on model outputs. A numerical model and an actual structure are used to validate the performance of the proposed model. The small prediction errors demonstrate that the models provide excellent approximations. The research findings and conclusions can be applied to the design and evaluation of non-structural components.

Keywords:

Non-structural component

Strength reduction factor

Nonlinearity

Ensemble learning

Prediction model