Seismic Response Prediction of One-Story Bidirectionally Eccentric Structures Based on the BP Neural Network

Predicting a structure’s seismic responses is of great importance during seismic design, especially in the preliminary stage. Hence, in this paper, a model for rapid prediction of the seismic responses of one-story bidirectionally eccentric structures based on the BP neural network is proposed. First, the equations of motion of one-story two-way eccentric structures under unidirectional seismic excitation are derived and solved through the modal response spectrum method. Factors, including eccentricity ratios and frequency ratios, are analyzed in detail to investigate their effects on these structures’ seismic responses. The results show that the eccentricity ratio perpendicular to the seismic direction has a greater impact on translational responses along the seismic direction and rotational responses than the eccentricity ratio along the seismic direction. In addition, the torsional–translational frequency ratio exhibits a larger impact on translational responses along the seismic direction and rotational responses than the translational frequency ratio. Following this, the back propagation (BP) neural network model is developed, fully considering the effects of various factors. A SHAP analysis is also conducted to compare the relative importance of different factors. The outcomes show that the seismic prediction model developed in this study achieves high accuracy while adequately considering the factors influencing the seismic responses of one-story two-way eccentric structures. Finally, an example is presented to demonstrate the applicability of the proposed prediction model.