COMPARATIVE ANALYSIS OF MACHINE LEARNING MODELS IN PREDICTING THE FUNDAMENTAL PERIOD OF STEEL STRUCTURES CONSIDERING SOIL-STRUCTURE INTERACTION

Abstract

The fundamental period is an important parameter to consider in the design of structures located in earthquake prone regions. The current design formulae available for determining the fundamental period of steel structures cannot accurately predict the fundamental period of real structures since most of them only consider the height of the structure in their formulation. SSI and the orientation of the H-columns may influence the fundamental period and it may be necessary to include these effects in the proposed formula. This research focuses on the comparison of ML models in predicting the fundamental period of steel structures accounting for different geometrical features of the superstructure, where the SSI effect is also accounted for. A dataset of 98 308 numerically obtained fundamental period results were used to train and test the proposed ML models. An HPC called Cyclone was used when analysing the models to minimize the computational time and effort required in developing the dataset. It was found that XGBoost outperformed the other ML methods with a MAPE of 0.5% where the linear regression model performed the worst with a MAPE of 23%. An additional 500 out-of-sample numerical results were numerically obtained and used to test the accuracy of the methods in predicting the fundamental period of steel structures. It was found that all models were able to accurately predict the fundamental period for out of sample geometries and soil properties with the highest resulting a MAPE of 16%. A second validation dataset of 100 numerical results was also developed that consisted of asymmetric structures where it was found that future research work needs to be conducted in accounting for the frames asymmetry within the predictive formulae.