| 摘要: |
| 传统的谱比法在地震场地分类中存在一定的主观性和不确定性,引入机器学习算法可以通过大量的样本数据和特征提取方法,提高分类方法的客观性和准确性,为地震工程设计和地震危险性评估提供更可靠的依据。本研究利用川滇地区历史地震的强震台事件波形数据,计算不同场地类别的强震动记录的水平和垂直速度反应谱值获取谱比曲线,分析了其在卓越周期、峰值和整体幅值上的特征。引入深度学习算法,整理并生成了川滇地区强震动台站谱比曲线样本集,并将其用作VGG16卷积神经网络模型的输入数据,训练得到了三类NEHRP标准场地的概率分布模型。通过特征提取和交叉验证算法,提高了模型的精度和泛化能力。最终得到的场地分类模型在验证结果中表现良好,对于谱比法的场地分类方法的改进和应用具有重要的参考价值。 |
| 关键词: 地震学 场地分类 谱比法 深度学习 |
| DOI： |
| 分类号: |
| 基金项目:地震科技星火计划项目（XH23052C） |
|
| Research on Spectral Ratio Method Site Classification in Sichuan-Yunnan Region Based on VGG16 |
|
zhangyifan1, xinan2, yangtianqing2, jianglixin3
|
|
1.Institute of Earthquake Forecasting (IEF) of the China Earthquake Administration (CEA);2.China Earthquake Networks Center;3.北京市西城区三里河南横街5号
|
| Abstract: |
| The traditional spectral ratio method has certain subjectivity and uncertainty in seismic site classification. By introducing machine learning algorithms and utilizing a large amount of sample data and feature extraction methods, the objectivity and accuracy of the classification method can be improved, providing more reliable basis for seismic engineering design and seismic hazard assessment. In this study, the strong-motion waveform data of historical earthquakes in the Sichuan-Yunnan region were used to obtain spectral ratio curves by calculating the horizontal and vertical velocity response spectrum values of different site categories. The characteristics of the curves in terms of predominant period, peak, and overall values were analyzed. By introducing deep learning algorithms, a sample dataset of spectral ratio curves of strong-motion stations in the Sichuan-Yunnan region was organized and generated, and it was used as the input data for the VGG16 convolutional neural network model to train a probability distribution model for the three NEHRP standard site categories. The accuracy and generalization ability of the model were improved through feature extraction and cross-validation algorithms. The resulting site classification model performed well in the validation results, which is of great reference value for the improvement and application of the spectral ratio method in site classification. |
| Key words: Seismology Site classification Spectral ratio method Deep learning |
