- 无标题文档
查看论文信息

中文题名:

 基于地震动信号特征的滑坡判识定位与动力过程反演分析    

姓名:

 王兴鲁    

一卡通号:

 0000382847    

论文语种:

 中文    

学科名称:

 工学 - 交通运输工程 - 道路与铁道工程    

公开时间:

 公开    

学生类型:

 硕士    

学位:

 工学硕士    

学校:

 西南交通大学    

院系:

 土木工程学院    

专业:

 道路与铁道工程    

第一导师姓名:

 严炎    

第一导师单位:

 西南交通大学    

完成日期:

 2024-05-01    

答辩日期:

 2024-05-24    

外文题名:

 Landslide Identification Localization and Dynamic Process Inversion Analysis Based on Seismic Signal Characteristics    

中文关键词:

 滑坡 ; 地震动信号 ; 定量识别 ; 定位 ; 动力过程反演    

外文关键词:

 Landslides ; Seismic signal ; Quantitative identification ; Localization ; Dynamic process inversion    

中文摘要:

当前所发生的地质灾害中滑坡灾害已成为生产生活中重要的隐患。为了减少滑坡灾害的发生所带来的损失,准确识别滑坡并做出定位,正确认识滑坡运动特征、获得动力学参数变得至关重要。本研究使用能够记录地质灾害全过程的地震动信号作为评估灾害各类属性的数据,并以新磨、白格滑坡为例对其进行识别、定位分析,并应用格林函数反演滑坡运动特征,重构滑坡整个运动过程。本文主要内容和研究成果如下:
(1)为了准确做出滑坡事件的识别,本研究收集了国内外滑坡与地震事件,对其进行时域、频域与时频域的变换处理,总结出两者地震动信号的主要特征:滑坡信号波形为“纺锤体”,无明显的P波与S波到达,频率、振幅的上升与衰减缓慢,频率范围小于5 Hz;地震信号振幅表现为突然增加,存在明显的P波与S波到达,时频谱中形状为“直角三角形”,频率范围是几赫兹到几十赫兹的高频,频率能量为突然的增加且衰减较快。
(2)提取出滑坡地震动信号58个特征的数值,利用机器学习将计算出的信号特征值建立起数据模型,并在Python中采用随机森林法对数据模型进行训练,选择模型参数为树深为15、节点最小样本数为4,并分别采用样本数量的60%-90%为训练集训练模型进行比较,最终选取出训练性能最好的模型用于事件识别分类,结果显示训练模型的测试准确率均在90%以上。
(3)提出双曲线原理用于滑坡事件的定位,并以四川新磨滑坡为例验证了该方法的可靠性。研究中选取了8个台站的位置尽可能的覆盖滑坡事件的每个方位角,基于互相关法得到每个台站信号之间的走时差计算台站距离差;然后应用双曲线原理得到定位结果显示NS方向和垂直方向的信号定位误差仅为3 km,而EW方向的信号定位误差为7.8 km。
(4)本研究中以白格滑坡为例,基于地震动信号反演了滑坡在运动过程中对地表施加的力-时间函数,评估了滑坡体方量为3200×104 m3,与实际方量几乎相同;结合反演的力-时间函数和加速度随时间的变化曲线以及现场调查,重构了滑坡运动的不同阶段:第一个阶段为滑坡开始崩塌加速下滑的过程,第二阶段为滑坡入江;第三阶段为扩散堆积阶段,滑坡已基本滑入江中并堵江形成堰塞湖。

外文摘要:

Landslide disasters have become a significant hidden danger in production and daily life among the current geological hazards. To reduce the losses caused by landslide disasters, it is crucial to accurately identify landslides and determine their locations, understand the characteristics of landslide movements, and obtain dynamic parameters. This study uses seismic signals, which can record the entire process of geological disasters, as data to evaluate various attributes of the disaster. Taking the Xinmo and Baige landslides as examples, we conduct identification and localization analyses, apply Green's function to invert the characteristics of landslide movements, and reconstruct the entire process of the landslide movement. The main content of this study is as follows:
(1)To accurately identify landslide events, this study collected landslide and earthquake events from both domestic and international sources, and performed time-domain, frequency-domain, and time-frequency domain transformations on them. The main characteristics of seismic signals from landslides and earthquakes were summarized as follows: Landslide signals exhibit a "spindle-shaped" waveform with no distinct arrivals of P-waves and S-waves, and show slow increases and decreases in frequency and amplitude, with a frequency range less than 5 Hz. On the other hand, earthquake signals show sudden increases in amplitude, distinct arrivals of P-waves and S-waves, and exhibit a "rectangular triangle" shape in the time-frequency spectrum. The frequency range is in the high-frequency range from several hertz to tens of hertz, with a sudden increase in frequency energy followed by rapid decay. Based on these signal characteristics, a landslide disaster identification method is established using a physical model of seismic signals.
(2)We extracted numerical values of 58 features of seismic signals, and used machine learning to establish a data model based on the calculated signal feature values. In Python, we used the random forest method to train the data model, and selected the model parameters as the tree depth of 15 and the minimum number of samples per node of 4. We trained the model with 60%-90% of the sample size as the training set for comparison, and finally selected the model with the best training performance for event recognition and classification. The results showed that the test accuracy of the trained model was above 90%.
(3)The hyperbolic principle was proposed for the localization of seismic events, and its reliability was verified using the Xinmo landslide in Sichuan as an example. In the study, positions from eight stations were selected to cover as many azimuths of the landslide event as possible. The cross-correlation method was used to obtain the travel time differences between the signals from each station, and the station distance differences were calculated. The hyperbolic principle was then applied to obtain the localization results, showing that the localization error in the NS direction and vertical direction was only 3 km, while the localization error in the EW direction was 7.8 km.
(4)In this study, taking the Baige landslide as an example, the force-time function exerted on the surface by the landslide during its movement was inverted based on seismic signals. The volume of the landslide was estimated to be 32 million cubic meters, which is almost identical to the actual volume. By combining the inverted force-time function with the acceleration-time curve and field investigations, the different stages of the landslide movement were reconstructed: the first stage is the process of the landslide starting to collapse and accelerating downhill; the second stage is the deceleration phase as the landslide enters the river; the third stage is the spreading and accumulation phase, during which the landslide has mostly entered the river, creating a dammed lake.

分类号:

 P642.24    

总页码:

 90    

参考文献总数:

 91    

馆藏位置:

 P642.24 S 2024    

开放日期:

 2024-06-13    

无标题文档

   建议浏览器: 谷歌 火狐 360请用极速模式,双核浏览器请用极速模式