基于神经网络的跨越地裂缝框架结构地震损伤及预测研究∗

为开展特殊地质环境下结构的损伤分析，以一跨越西安f4地裂缝的五层框架结构为研究对象，基于振动台试验和ABAQUS 有限元分析结果，进行了BP 神经网络模型的模型训练，选取变形和能量组合形式的双参数损伤模型计算结构损伤指标，采用加权系数法，开展了构件、楼层、结构三个层面的损伤预测分析，给出了不同地震作用下结构损伤程度评估。结果表明：地裂缝场地结构表现出明显上下盘效应，结构首层为薄弱层。BP 神经网络损伤预测值与有限元计算值在不同工况下均较为一致，其对于构件、层间、整体结构损伤指数预测最大误差分别为8.86%、5.66%、7.57%，该研究成果为跨越地裂缝结构的性能评估提供一种准确且高效的研究方法。

Seismic Damage Analysis of Frame Structures Spanning GroundFissures Based on Neural Network

To carry out structural damage analysis on structures under special geological conditions, afive-story frame structure spanning the f4 ground fissure in Xi''an was taken as the research object.Based on the results from shaking table tests and ABAQUS finite element analysis, a BP neural networkmodel was trained. A two-parameter damage model combining deformation and energy was em ployed to calculate structural damage index. The weighted coefficient method was used to carry outthe damage prediction analysis of components, floors, and structures, providing evaluations of structuraldamage under different seismic impacts. The results showed that the structure on the ground fissuresite exhibited a significant hanging-wall/footwall effect, with the first layer being the weak layer.The BP neural network''s damage prediction values closely aligned with the finite element calculation values across different conditions. The maximum prediction errors for components, inter-layer, andoverall structural damage indices were 8.86%, 5.66%, and 7.57%, respectively.