地震研究

选用P波震相附近的地震波作为研究对象,对近震和远震特征信息进行探讨。选取初至P波主周期作为神经网络输入元,P波到达后2～6 s作为地震波时间窗,选择正确的网络结构和参数,搜集大量的地震样本数据进行训练,实现对近震和远震地震事件的非线性系统识别。结果表明:在样本训练区间检验数据的预测结果置信度达到100%; 在非样本区间也能迅速收敛到标识符0或1附近。近震样本信号最大周期为0.25 s,而置信度达到80%以上的预测区间几乎接近0.35 s; 远震样本信号最小周期为0.9 s,而置信度达到80%以上的预测区间达到0.5 s,表明模型建立得当,具有良好的泛化能力。

We studied seismic waves near the P wave phase to discuss the characteristic of near and distant earthquakes.Firstly,we selected main period of initial P wave as the input element of neural network and picked up 2 ～ 6 s of initial P wave arriving time as time window of seismic wave.Secondly,we chose the right network structure and parameters and collected a large number of earthquake training data to realize the nonlinear system identification between near and distant earthquake.The results show that the predictable result of test data whose confidence reaches 100% in sample training interval can converge to identifier 0 or 1 quickly in non-sample training interval.The maximum period of near earthquake sample is 0.25 s and its prediction interval whose confidence reaches more than 80% is almost close to 0.35 s.The minimum period of distant earthquake sample is 0.9 s and its prediction interval whose confidence reaches more than 80% is 0.5 s,which shows that the model we select is proper and has good generalization ability.

引言
1 BP神经网络
2 判别信息探讨
3 近震与远震识别
4 结论与讨论

图1 基本BP网络的拓扑结构<br/>Fig.1 Topology structure of basic BP network

图1 基本BP网络的拓扑结构
Fig.1 Topology structure of basic BP network

图2 郯城台垂直向原始波形(a)及处理后波形(b)<br/>Fig.2 Vertical original waveform(a)recorded in Tancheng Station and its waveform after pretreatment(b)

图2 郯城台垂直向原始波形(a)及处理后波形(b)
Fig.2 Vertical original waveform(a)recorded in Tancheng Station and its waveform after pretreatment(b)

表1 近震与远震视出射角对比<br/>Tab.1 Apparent emergence angle comparison between near and distant earthquakes

表1 近震与远震视出射角对比
Tab.1 Apparent emergence angle comparison between near and distant earthquakes

图3 即墨台初至P波到达后的地震信号<br/>Fig.3 Seismic waveform after initial P wave arriving in Jimo Station

图3 即墨台初至P波到达后的地震信号
Fig.3 Seismic waveform after initial P wave arriving in Jimo Station

图4 主频计算结果图<br/>(a)能量分布图;(b)波振周谱分布图; <br/>(c)不同周期信号的能谱分布图<br/>Fig.4 Results of dominant frequency calculation (a)energy distribution;(b)distribution of wavelet amplitude spectrum;(c)energy distribution of signals in different periods

图4 主频计算结果图
(a)能量分布图;(b)波振周谱分布图;
(c)不同周期信号的能谱分布图
Fig.4 Results of dominant frequency calculation (a)energy distribution;(b)distribution of wavelet amplitude spectrum;(c)energy distribution of signals in different periods

表2 近震远震训练样本<br/>Tab.2 Training samples of near and distant earthquakes

表2 近震远震训练样本
Tab.2 Training samples of near and distant earthquakes

表3 近震远震检验结果<br/>Tab.3 Testing result of near and distant earthquakes

表3 近震远震检验结果
Tab.3 Testing result of near and distant earthquakes

蔡煜东,宫家文,甘俊人,等.1994.地震序列类型预测的人工神经网络方法[J].地震研究,17(1):40-45.
何斌,张元生,李稳.2010.计算地震初至波视出射角方法[J].西北地震学报,32(1):11-15.
刘希强,周惠兰,沈萍,等.2000.用于三分向记录震相识别的小波变换方法[J].地震学报,32(2):125-131.
穆星.2005.利用地震几何属性和自组织神经网络进行地震相得自动识别[J].地质科技情报,24(3):109-112.
王娟,刘俊民,范万春.2004.神经网络在震相识别中的应用[J].网络与通信,(5):35-37.
赵仲和.2005.地震自动定位的综合解决方案[J].地震地磁观测与研究,26(2):50-56.
中国地震局监测预报司.2007.地震学与地震观测[M].北京:地震出版社.
Allen.1978.Automatic earthquake recognition and timing from single traces[J].BSSA,68(5):1521-1532.
Cichowicz.1993.An automatic S-phase picker[J].BSSA,83(1):180-189.
Coforth T,Herin E.1981.An automatic seismic signal detection algorithm based on the walsh transform[J].BSSA,71(4):1 351-1 360.
Jin W,Ta-Liang T.1988.Artificial neutral network based seismic detector[J].BSSA,85(1):308-319.

备注

引言

1 BP神经网络

2 判别信息探讨

3 近震与远震识别

4 结论与讨论

期刊信息