【摘要】 采用机器学习等智能算法预测结构时程响应时，通常将地震动的某项静态指标作为输入来考虑。这种处理方法丢失了地震动记录本身的时程特性。本文提出将地震动记录处理成二维张量来处理其时程信息，以一栋框架剪力墙结构为对象，通过数据集构造、网络模型搭建和参数优化，训练卷积神经网络(CNN)来预测结构的最大层间位移角。此外，还采用人工神经网络(ANN)以及长短时记忆神经网络(LSTM)进行对比分析，研究结果表明：CNN与LSTM能够有效地预测弹性状态下的结构层间位移角，且CNN的预测精度最高，训练耗时最少。更多还原

【Abstract】 When using intelligent algorithms such as machine learning to predict the time-history response of structures, a static indicator of ground motion is usually considered as input. This treatment loses the time-history characteristics of the ground motion recording itself. In this paper, the ground motion record is processed into a two-dimensional tensor to process its time-history information, taking a frame shear wall structure as the research object, and a convolutional neural network(CNN) is trained to predict the maximum interlayer displacement angle of the structure through dataset construction, network model construction, and parameter optimization. In addition, artificial neural networks(ANN) and long-short-term memory neural networks(LSTM) were used for comparative analysis. The results show that CNN and LSTM can effectively predict the interlayer displacement angle of the structure under elastic state, and CNN has the highest prediction accuracy and the least training time.更多还原