基于可变尺寸滑动窗口的框架结构地震响应预测研究

王飞; 徐国越; 刘琦璇; 张晨; 吕潇然; 陈宏宇

doi:10.11899/zzfy20260073

基于可变尺寸滑动窗口的框架结构地震响应预测研究以北京市某实验楼为例

doi: 10.11899/zzfy20260073

王飞^{1, 2,},
徐国越^1, ,,
刘琦璇²,
张晨²,
吕潇然²,
陈宏宇²

1.
应急管理大学, 地震工程与建筑安全学院, 河北三河 065201
2.
北京市地震局, 北京 100080

基金项目: 北京市地震局震灾风险防治中心技术微创新项目（BGWC-2026009）

详细信息

作者简介:
王飞，男，生于1979年。正高级工程师。主要从事结构监测与抗震研究。E-mail：wangfei@bjseis.gov.cn

通讯作者:
徐国越，男，生于2000年。硕士研究生。主要从事结构抗震研究。E-mail：24661404@st.cidp.edu.cn

中图分类号: P315.9；TU375.4
计量
- 文章访问数: 28
- HTML全文浏览量: 11
- PDF下载量: 30
- 被引次数: 0
出版历程
- 收稿日期: 2026-04-27
- 录用日期: 2026-06-08
- 修回日期: 2026-06-05
- 网络出版日期: 2026-08-31

Research on Seismic Response Prediction of Frame Structures Based on Variable-Size Sliding Window—A Case Study of an Experimental Building in Beijing

Wang Fei^{1, 2
,},
Xu Guoyue^{1
, ,},
Liu Qixuan²,
Zhang Chen²,
Lv Xiaoran²,
Chen Hongyu²

1.
School of Earthquake Engineering and Building Safety, University of Emergency Management, Sanhe 065201, Hebei, China
2.
Beijing Earthquake Agency, Beijing 100080, China

摘要

摘要: 在使用固定尺寸滑动窗口处理数据的过程中，无法根据地震动强度差异和结构地震响应幅值变化动态提取时频特征，易造成关键特征丢失、预测精度下降等。本文提出一种使用可变尺寸滑动窗口处理数据的结构响应预测方法。该方法以一栋框架结构为研究对象，以100条地震动加速度时程记录及有限元模型输出的层间位移角作为数据集，基于提出的算法预测结构层间位移角值。预测结果与基于固定尺寸滑动窗口方法对比发现：本文提出的算法可有效实现框架结构的响应预测，且相较于固定尺寸滑动窗口方法，平均绝对百分比误差降低了12.27%，峰值层间位移角误差降低了7.87%。
- 卷积神经网络 /
- 层间位移角 /
- 框架结构 /
- 可变尺寸滑动窗口
Abstract: In the process of processing data using a fixed-size sliding window, it is difficult to dynamically extract time-frequency features based on differences in ground motion intensity and amplitude variations of structural seismic responses, which can easily lead to the loss of key features and a reduction in prediction accuracy. This paper proposes a structural response prediction method that uses a variable-size sliding window for data processing. Taking a frame structure as the research object, and using 100 ground motion acceleration time-history records and the inter-story drift ratios output by a finite element model as the dataset, the proposed algorithm predicts the inter-story drift ratios. Comparison with prediction results based on a fixed-size sliding window shows that the proposed algorithm can effectively predict the response of the frame structure. Compared with the fixed-size sliding window method, the mean absolute percentage error is reduced by 12.27%, and the peak inter-story drift ratio error is reduced by 7.87%.
- Convolutional neural network /
- Inter-story drift angle /
- Frame structure /
- Variable-size sliding window

HTML全文

图 1 结构有限元模型

Figure 1. Structural finite element model

下载: 全尺寸图片幻灯片

图 2 传递函数识别自振频率图

Figure 2. Transfer function diagram for natural frequency identification

下载: 全尺寸图片幻灯片

图 3 100条地震动加速度反应谱

Figure 3. Acceleration response spectra of 100 ground motions

下载: 全尺寸图片幻灯片

图 4 加速度时程

Figure 4. Acceleration time history

下载: 全尺寸图片幻灯片

图 5 固定尺寸窗口处理地震动

Figure 5. Ground motion processing with fixed-size window

下载: 全尺寸图片幻灯片

图 6 可变尺寸窗口处理地震动

Figure 6. Ground motion processing with variable-size sliding window

下载: 全尺寸图片幻灯片

图 7 贝叶斯搜索热力图

Figure 7. Bayesian optimization heat map

下载: 全尺寸图片幻灯片

图 8 最优幅值阈值比例

Figure 8. Optimal amplitude threshold ratio

下载: 全尺寸图片幻灯片

图 9 最优超参数点状图

Figure 9. Scatter plot of optimal hyperparameters

下载: 全尺寸图片幻灯片

图 10 可变尺寸窗口网络收敛过程

Figure 10. Convergence process of the network with varia ble-size window

下载: 全尺寸图片幻灯片

图 11 固定尺寸窗口网络收敛过程

Figure 11. Convergence process of the network with fixed-size window

下载: 全尺寸图片幻灯片

图 12 预测结果对比图

Figure 12. Comparison chart of prediction results

下载: 全尺寸图片幻灯片

表 1 结构自振周期与振动性质

Table 1. Natural periods and vibration modes of the structure

振型编号	振型特性	模拟周期/s	识别周期/s	周期误差
1	东西向	0.707	0.753	6.1%
2	南北向	0.699	0.744	6.0%
3	东西向	0.267	0.241	10.8%
4	南北向	0.257	0.238	8.0%

下载: 导出CSV

表 2 层间位移角

Table 2. Interstory drift angle

地震动名称	层间位移角
地震动名称	第1层	第2层	第3层	第4层	第5层
H-PG5090	0.007 551	0.012066	0.011781	0.007283	0.003124
EIL-EW	0.005 049	0.009330	0.011090	0.007285	0.003119
NSD180	0.005 677	0.009879	0.010576	0.007624	0.003424
ICC090	0.005 595	0.007814	0.007572	0.005468	0.002957
A-ELC180	0.005 814	0.007749	0.006156	0.003859	0.001819
2053b360	0.005 139	0.007650	0.006857	0.004534	0.002044
3914a235	0.004 907	0.007241	0.007151	0.005256	0.002572
H-SC2000	0.005 742	0.007143	0.005743	0.004065	0.002302
2037a090	0.004 225	0.006745	0.006181	0.004038	0.001836
0194 a090	0.004 500	0.006646	0.006321	0.004082	0.001724

下载: 导出CSV

表 3 CNN网络架构

Table 3. CNN architecture

层类型	激活函数	输出形状
InputLayer	—	（None，379，1）
ResidualBlock×6 + Pooling×6	ReLU	Downsampling
GlobalAveragePooling1 D	—	（None，256）
Dense	ReLU	（None， 36）
Dropout	—	（None， 36）
Dense	Linear	（None， 5）

下载: 导出CSV

表 4 训练环境

Table 4. Training environment

项目	配置
CPU	Intel Xeon E5-1620 v4 @ 3.50 GHz
GPU	NVIDIA GeForce GTX 1080 Ti
内存	8 GB DDR4 @ 2400 MHz
硬盘	1.8 TB HDD
操作系统	Windows（DirectX 12）
软件	MATLAB R2025 a

下载: 导出CSV

表 5 训练集0514c090预测值

Table 5. Predicted values of training set 0514c090

楼层	真实值/m	固定窗口预测值/m	固定窗口误差	可动窗口预测值/m	可动窗口误差
1	0.000901	0.000698	22.49%	0.000898	0.37%
2	0.001409	0.001051	25.44%	0.001410	0.04%
3	0.001509	0.001162	22.97%	0.001367	9.44%
4	0.001091	0.000887	18.65%	0.001055	3.32%
5	0.000649	0.000465	28.33%	0.000559	13.82%

下载: 导出CSV

表 6 训练集0534a180预测值

Table 6. Predicted values of training set 0534a180

楼层	真实值/m	固定窗口预测值/m	固定窗口误差	可动窗口预测值/m	可动窗口误差
1	0.002016	0.001355	32.79%	0.002183	8.28%
2	0.003393	0.002269	33.14%	0.003296	2.84%
3	0.003614	0.002441	32.46%	0.003152	12.78%
4	0.002760	0.001919	30.48%	0.002417	12.43%
5	0.001395	0.000872	37.47%	0.001214	12.94%

下载: 导出CSV

表 7 训练集12076324预测值

Table 7. Predicted values of training set 12076324

楼层	真实值/m	固定窗口预测值/m	固定窗口误差	可动窗口预测值/m	可动窗口误差
1	0.002976	0.001743	41.42%	0.002205	25.91%
2	0.003953	0.002320	41.31%	0.003393	14.17%
3	0.003158	0.002475	21.63%	0.002942	6.83%
4	0.002129	0.001813	14.85%	0.002179	2.34%
5	0.001107	0.000864	21.91%	0.001162	4.98%

下载: 导出CSV

表 8 测试集0521a360预测值

Table 8. Predicted values of test sets 0521a360

楼层	真实值/m	固定窗口预测值/m	固定窗口误差	可动窗口预测值/m	可动窗口误差
1	0.001397	0.000858	38.60%	0.001393	0.29%
2	0.002251	0.001499	33.39%	0.002204	2.09%
3	0.002143	0.001387	35.27%	0.001894	11.62%
4	0.001507	0.000963	36.09%	0.001460	3.10%
5	0.000797	0.000552	30.80%	0.000785	1.56%

下载: 导出CSV

表 9 测试集1417c270预测值

Table 9. Predicted values of test sets 1417c270

楼层	真实值/m	固定窗口预测值/m	固定窗口误差	可动窗口预测值/m	可动窗口误差
1	0.000572	0.000430	24.91%	0.000576	0.78%
2	0.000872	0.000678	22.24%	0.000907	4.00%
3	0.000817	0.000742	9.14%	0.000850	4.04%
4	0.000603	0.000561	7.03%	0.000622	3.15%
5	0.000334	0.000312	6.49%	0.000373	11.68%

下载: 导出CSV

表 10 测试集H-Z16000预测值

Table 10. Predicted values of test sets H-Z16000

楼层	真实值/m	固定窗口预测值/m	固定窗口误差	可动窗口预测值/m	可动窗口误差
1	0.000477	0.000368	22.76%	0.000438	8.17%
2	0.000703	0.000527	25.11%	0.000664	5.60%
3	0.000634	0.000605	4.55%	0.000668	4.43%
4	0.000506	0.000467	7.71%	0.000507	0.28%
5	0.000316	0.000264	16.39%	0.000300	4.93%

下载: 导出CSV

参考文献(19)

陈佳, 丁永刚, 许启铿等, 2025. 基于改进附加质量的筒仓结构散-固耦合地震响应研究. 振动与冲击, 44(19): 288−297. doi: 10.13465/j.cnki.jvs.2025.19.032

Chen J., Ding Y. G., Xu Q. K., et al., 2025. Seismic response of silo structure with grain-silo interaction based on improved added mass. Journal of Vibration and Shock, 44(19): 288−297. (in Chinese) doi: 10.13465/j.cnki.jvs.2025.19.032

陈吉龙, 潘汝江, 许成顺等, 2026. 液化侧向扩展场地群桩桥梁体系地震响应与失效模式研究. 震灾防御技术, 21(1): 132−145.

Chen J. L., Pan R. J., Xu C. S., et al., 2026. Research on seismic response and failure modes of pile group bridge system in liquefaction-induced lateral spreading site. Technology for Earthquake Disaster Prevention, 21(1): 132−145. (in Chinese)

郭茂祖, 闫济丞, 张庆宇等, 2025. 长短期频域特征嵌入的建筑结构地震响应预测. 建筑结构学报, 46(11): 223−236, 249.

Guo M. Z., Yan J. C., Zhang Q. Y., et al., 2025. Long-short term frequency domain feature embedding for predicting seismic response of building structures. Journal of Building Structures, 46(11): 223−236,249. (in Chinese)

黄敏沾, 彭玉祥, 蒋镇涛等, 2026. 基于物理信息神经网络的薄板静力响应计算模型. 哈尔滨工程大学学报, 47(1): 96−105.

Huang M. Z., Peng Y. X., Jiang Z. T., et al., 2026. Computational models for static response of thin plates based on physics-informed neural networks. Journal of Harbin Engineering University, 47(1): 96−105. (in Chinese)

李雨润, 王铭, 邵鼎松等, 2025. 液化场地中高层建筑-桩筏基础-地下结构体系地震响应离心振动台试验研究. 震灾防御技术, 20(3): 486−497.

Li Y. R., Wang M., Shao D. S., et al., 2025. Centrifugal shaking table test on seismic response of high-rise building pile raft foundation underground structure system in liquefaction site. Technology for Earthquake Disaster Prevention, 20(3): 486−497. (in Chinese)

李正农, 顾珂泽, 黄斌等, 2024. 地震作用下结构响应实时预测方法研究. 振动与冲击, 43(22): 62−69.

Li Z. N., Gu K. Z., Huang B., et al., 2024. Real-time prediction method of structural responses under earthquake action. Journal of Vibration and Shock, 43(22): 62−69. (in Chinese)

许泽坤, 陈隽, 2021. 非线性结构地震响应的神经网络算法. 工程力学, 38(9): 133−145.

Xu Z. K., Chen J., 2021. Neural network algorithm for nonlinear structural seismic response. Engineering Mechanics, 38(9): 133−145. (in Chinese)

薛建阳, 凌怀泉, 戚亮杰等, 2024. 传统风格建筑塔式结构振动台试验及动力分析. 建筑结构学报, 45(12): 19−30.

Xue J. Y., Ling H. Q., Qi L. J., et al., 2024. Shaking table test and dynamic analysis on a tower structure in traditional style building. Journal of Building Structures, 45(12): 19−30. (in Chinese)

郑史雄, 雷川鹤, 贾宏宇等, 2025. 基于LSTM的近断层桥梁地震响应和易损性快速预测方法. 中国公路学报, 38(7): 18−30.

Zheng S. X., Lei C. H., Jia H. Y., et al., 2025. LSTM-based method for fast prediction of seismic response and vulnerability of near-fault bridges. China Journal of Highway and Transport, 38(7): 18−30. (in Chinese)

Ding Y. J., Wang Z. C., Wang Y., et al., 2026. CNN-BiLSTM hybrid neural network with attention mechanism for structural nonlinear model updating. Journal of Bridge Engineering, 31(5): 04026019.

Du J. J., Wang W., Zhang J. C., et al., 2026. Seismic response of a prefabricated substation equipment cabin with base isolation: full-scale shaking table test and numerical analysis. Structures, 86: 111328.

Guo X., Li W. J., Qiao J. F., 2023. A self-organizing modular neural network based on empirical mode decomposition with sliding window for time series prediction. Applied Soft Computing, 145: 110559.

Liao Y. Y., Tang H. S., Li R. S., et al., 2023. Response prediction for linear and nonlinear structures based on data-driven deep learning. Applied Sciences, 13(10): 5918.

Liu B., Gong M. S., Wang X. M., et al., 2026. Physics-guided neural network for structural seismic response reconstruction based on floor response spectra. Journal of Building Engineering, 118: 114972.

Pamuncak A. P., Salami M. R., Adha A., et al., 2021. Estimation of structural response using convolutional neural network: application to the Suramadu Bridge. Engineering Computations, 38(10): 4047−4065.

Shu J. P., Dong T. F., Li J., et al., 2026. Data-driven convolutional neural network model with multi-channel input framework for predicting structural displacement response. Advances in Structural Engineering, 29(6): 1047−1061.

Wang J. S., Xie Y. Z., Guo T., et al., 2023. Predicting the influence of soil–structure interaction on seismic responses of reinforced concrete frame buildings using convolutional neural network. Buildings, 13(2): 564.

Wang Y., Li X. J., Wang X. H., et al., 2025. Asymmetric structural response characteristics of transmission tower-line systems under cross-fault ground motions revealed by shaking table tests. Symmetry, 17(10): 1646.

Zhou Y., Meng S. Q., Lou Y. J., et al., 2024. Physics-informed deep learning-based real-time structural response prediction method. Engineering, 35: 140−157.

施引文献

资源附件(0)

访问统计