다단계 기계학습을 활용한 철근콘크리트 기둥 포락선 예측

초록

Piloti-type buildings are vulnerable to earthquakes because the soft story formed on the first floor concentrates structural damage in the lower story during seismic events. This highlights the need for a methodology that can rapidly and accurately predict the backbone curve, a key indicator of seismic performance. Accordingly, this study developed a code-based combined model for predicting the backbone curve of piloti-type RC buildings using regression-based machine learning. The model used nine input variables, one of which was the failure mode, derived from a previously developed prediction model. Optimal models for predicting displacement and strength at the three key points of the backbone curve—yield, ultimate, and residual—were selected based on regression performance metrics and combined in code to develop the final prediction model. To verify the proposed methodology, a comparative analysis with experimental results of piloti-type buildings was conducted based on key indicators of lateral resistance capacity: effective stiffness, strength ratio, and ductility. The results confirmed that the developed machine learning model reliably predicts the backbone curve, demonstrating its potential as a rapid and efficient alternative to conventional numerical analysis methods.

키워드