军事知识服务平台
基于BiGRU的单层柱面网壳结构地震响应预测
Seismic Response Prediction of the Single-layer Cylindrical Lattice Shell Structure Based on BiGRU
摘要 柱面网壳结构具有特殊的几何特征和受力状态,在强震作用下响应行为十分复杂.基于深度学习循环神经网络架构及其变体,使用 ABAQUS 软件建立单层柱面网壳的有限元模型,开展结构地震响应预测.以不同震源场地关系的地震动加速度作为动力激励,通过自动化脚本提取结构响应数据构建数据集进行训练.使用4 种循环神经网络模型进行对比实验,结果表明:BiGRU 模型能够高效地预测结构响应,在计算时间和预测精度方面均优于其他模型,能够满足实时决策的需求.模型准确预测了用于量化结构破坏状态的损伤因子 D,在多个破坏状态等级上取得了较高准确率.
Abstract:
Cylindrical reticulated shells,characterized by their unique geometric features and complex stress states,exhib-it highly complex response behaviors under strong seismic actions.This study investigates the prediction of structural seismic response using deep-learning-based Recurrent Neural Network(RNN)architectures and their variants.A fi-nite element model of a single-layer cylindrical reticulated shell is established using ABAQUS software.A dataset is then constructed for training by applying ground motion accelerations with various source-site relationships as dynamic excitations and extracting the structural responses via automated scripts.Comparative experiments are conducted using four RNN models.The results demonstrate that the Bidirectional Gated Recurrent Unit(BiGRU)model can efficient-ly predict the structural response,outperforming other models in both computational time and prediction accuracy,thus meeting the requirements for real-time decision-making.Furthermore,the model accurately predicts the damage factor(D),a metric used to quantify the structural damage state,achieving high accuracy across multiple damage levels.
刊 名 地震研究 ISTICPKU
年,卷(期): 2026, 49(4)
分类号: TU393.3
机标分类号: TU311.3 TU435 TP391
在线出版日期: 2026年7月15日
基金项目: 2024YFC3016904:国家重点研发计划52578616:国家自然科学基金,