A Physics-Guided Machine Learning Approach for Accurate Prediction of Hysteretic Response in Reinforced Concrete Shear Walls

Accurately predicting the hysteretic behavior of reinforced concrete (RC) shear walls is essential for seismic performance assessment and resilient structural design. Traditional empirical formulas are limited in capturing nonlinear force–deformation interactions, while conventional machine learning (ML) models often operate as “black boxes,” offering little interpretability. To address these challenges, this study proposes an interpretable expression-guided machine learning (IEG-ML) framework that integrates backpropagation neural networks with swarm intelligence optimization. A comprehensive database of 210 experimentally tested rectangular RC shear walls was developed, covering diverse geometric, material, and reinforcement conditions. Empirical proportionality constraints were incorporated into the optimization process to ensure physical consistency and enhance interpretability. Comparative analysis shows that while all three optimization-based models significantly outperform empirical regression formulas, the DBO-enhanced model achieves superior performance, with average R² = 0.937 and substantial reductions in RMSE and MAE relative to PSO and GA. The predicted hysteretic feature points enable the direct construction of restoring force models, allowing automatic generation of hysteretic curves under varying design conditions. The proposed IEG-ML framework provides an accurate, interpretable, and computationally efficient tool, bridging mechanics-based models and advanced data-driven methods for seismic design of RC shear walls.