Automatic identification of P-phase based on wavelet packet and Kurtosis-AIC method
-
摘要: 基于小波包变换和峰度赤池信息量准则(AIC)， 提出了一种新的自动识别P波震相的综合方法， 即小波包-峰度AIC方法． 首先对由加权长短时窗平均比(STA/LTA)法粗略确定的P波到时前后3 s的记录进行小波包三尺度的分解与重构， 分别计算每个尺度重构信号的峰度AIC曲线并将其叠加， 叠加曲线的最小值则为P波震相到时; 然后对原始地震记录进行有限冲激响应自适应滤波以提高信噪比和识别精度; 最后将小波包-峰度AIC方法应用到合成理论地震图及实际地震记录的P波初至自动识别中. 结果表明： 初至清晰度对识别精度的影响比信噪比对其影响更大； 与单独使用加权STA/LTA方法和峰度AIC法相比， 小波包-峰度AIC法具有更强的抗噪能力， 识别精度更高； 当初至清晰时， 小波包-峰度AIC法自动识别与人工识别的P波到时平均绝对差值为(0.077±0.075) s．
-
关键词:
- P波震相 /
- 有限冲激响应(FIR)数字滤波 /
- 震相自动识别 /
- 小波包-峰度AIC方法
Abstract: Automatic identification of P-phase is of significance to the study on earthquake location, earthquake warning and structure of deep earth. Combining wavelet packet transform with Kurtosis-AIC (Akaike information criterion) technology, this paper puts forward a new synthetic method named wavelet packet and Kurtosis-AIC method for automatic recognition of first P-phase. Three scales of discrete wavelet packet transforms are applied to decompose and reconstructure the original recordings three seconds before and after the rough P-wave arrival time, which is picked up by weighted STA/LTA (short term average/long term average) method, then the Kurtosis-AIC values of the three-scale reconstruction signal are calculated respectively and superposed together, finally the minimum value of the superposed AIC curve is taken as the first P-wave arrival time. In order to test the new method, it is applied to theoretically synthetic seismograms and real seismic recording for automatic P-phase arrival time detection. Adding white Gaussian noise and real seismic noise to synthetic seismograms with different SNR, the optimal frequency band of adaptive FIR (finite impulse response) digital filtering is used to improve the SNR and P-wave recognition accuracy of the original signals. The results show that, with respect to the impact of SNR, the accuracy of P-wave identification is more affected by the clarity of first break; our method has greater noise immunity and higher P-wave recognition accuracy as compared to the weighted STA/LTA algorithm and Kurtosis-AIC method. When the first break of P-wave is clear, average absolute error of P-phase arrival time between automatic identification based on our method and manual identification is (0.077±0.075) seconds. -
-
![]()
图 2 利用小波包-峰度AIC方法拾取P波到时 (a) 垂直向地震波形； (b) 尺度1； (c) 尺度2； (d) 尺度3； (e) 3个尺度叠加
Figure 2. Automatic identification of P-wave onset time based on wavelet packet and Kurtosis-AIC method (a) The vertical seismic waveform; (b) The first scale; (c) The second scale; (d) The third scale; (e) Three-scale superposition
![]()
图 3 实际地震记录(a)与合成理论地震图(b)
Figure 3. Real seismic recording (a) and theoretically synthetic seismogram (b)
![]()
图 4 合成地震记录的P波到时自动拾取 (a) 地震垂直向记录； (b) STA/LTA方法； (c) 峰度AIC方法； (d) 小波包-峰度AIC方法
Figure 4. Automatic identification of P-wave arrival time for synthesis seismic recording (a) Vertical seismic recording; (b) STA/LTA method; (c) Kurtosis-AIC method; (d) Wavelet packet and Kurtosis-AIC method
![]()
图 5 滤波前、 后信噪比(a)及P波初至识别差值ΔT′(b)的对比
Figure 5. Comparison of signal-to-noise ratio (a) and deviation ΔT′ of first P-wave recognition (b) before (denoted by open circles) and after (denoted by red dots) filtering
![]()
图 6 P波初至识别差值ΔT′、 信噪比及初至清晰度的关系 (a) 初至差与信噪比的关系； (b) 信噪比与初至清晰度的关系； (c) 初至差与初至清晰度的关系
Figure 6. Relationship between deviation ΔT′ of first P-wave recognition， signal-to-noise ratio and clarity of first break where black and blue dots represent the first break is clear and unclear respectively (a) Relationship between deviation of first break and SNR; (b) Relationship between SNR and clarity of first break; (c) Relationship between deviation of first break and its clarity
![]()
图 7 初至清晰时P波震相自动识别与人机交互识别结果对比图中数值分别为3种方法得到的平均绝对差值. (a) STA/LTA方法；(b) 峰度AIC方法； (c) 小波包-峰度AIC方法
Figure 7. Comparison of the P-wave onset time result by automatic recognition with that by man-machine interaction recognition on the condition that the first break is clearThe average absolute errors of the three methods are also given. (a) STA/LTA method; (b) Kurtosis-AIC method; (c) Wavelet packet and Kurtosis-AIC method
![]()
图 8 初至清晰时P波震相自动识别与 人机交互识别差值分布(a) STA/LTA方法； (b) 峰度AIC方法；(c) 小波包-峰度AIC方法
Figure 8. Error distribution of the P-wave onset time between automatic recognition and man-machine interaction recognition when first break is clear (a) STA/LTA method; (b) Kurtosis-AIC method; (c) Wavelet packet and Kurtosis-AIC method
![]()
图 9 初至不清晰时P波震相自动识别与人机交互识别结果对比 图中数值分别为3种方法得到的平均绝对差值. (a) STA/LTA方法； (b) 峰度AIC方法； (c) 小波包-峰度AIC方法
Figure 9. Comparison of the P-wave onset time result by automatic recognition with that by man-machine interaction recognition on the condition that the first break is unclear The average absolute errors of the three methods are also given. (a) STA/LTA method; (b) Kurtosis-AIC method; (c) Wavelet packet and Kurtosis-AIC method
![]()
图 10 初至不清晰时P波震相自动识别与人机交互识别差值分布 (a) STA/LTA方法； (b) 峰度AIC方法； (c) 小波包-峰度AIC方法
Figure 10. Error distribution of the P-wave onset time between automatic recognition and man-machine interaction recognition when first break is unclear (a) STA/LTA method; (b) Kurtosis-AIC method; (c) Wavelet packet and Kurtosis-AIC method
表 1 合成理论记录中加入实际地震噪声时3种方法在不同信噪比下拾取P波到时的效果对比
Table 1 Comparison of P-wave recognition effect by three methods with different SNRs when real seismic noise is added to theoretically synthetic seismograms
SNR/dB 自动拾取的P波初至到时/s 自动与理论到时差ΔT/s STA/LTA方法 峰度AIC方法 小波包-峰度AIC方法 STA/LTA方法 峰度AIC方法 小波包-峰度AIC方法 25.00 23.55 23.43 23.38 0.18 0.06 0.01 22.00 23.57 23.44 23.39 0.20 0.07 0.02 21.00 23.58 23.45 23.39 0.21 0.08 0.02 20.00 23.59 23.46 23.39 0.22 0.09 0.02 17.00 23.64 23.51 23.42 0.27 0.14 0.05 15.00 23.67 23.55 23.47 0.30 0.18 0.10 13.61 23.70 23.59 23.51 0.33 0.22 0.14 13.00 23.71 23.61 23.54 0.34 0.24 0.17 12.00 23.78 23.63 23.56 0.41 0.26 0.19 10.00 23.98 23.69 23.60 0.61 0.32 0.23 9.00 24.04 23.71 23.61 0.67 0.34 0.24 8.00 24.08 23.75 23.62 0.71 0.38 0.25 7.00 24.19 23.78 23.64 0.82 0.41 0.27 6.00 24.20 23.81 23.67 0.83 0.44 0.30 5.00 24.33 23.85 23.72 0.96 0.48 0.35 
下载: 导出CSV
表 2 合成理论记录中加入高斯白噪声时3种方法在不同信噪比下拾取P波到时的效果对比
Table 2 Comparison of P-wave recognition effect by three methods with different SNRs when white Gaussian noise is added to theoretically synthetic seismograms
SNR/dB 自动拾取的P波初至到时/s 自动与理论到时差ΔT/s STA/LTA方法 峰度AIC方法 小波包-峰度AIC方法 STA/LTA方法 峰度AIC方法 小波包-峰度AIC方法 25.05 23.53 23.45 23.45 0.16 0.08 0.08 22.05 23.65 23.47 23.46 0.28 0.10 0.09 21.08 23.65 23.48 23.47 0.28 0.11 0.10 20.18 23.72 23.53 23.51 0.35 0.16 0.14 17.21 23.79 23.58 23.53 0.42 0.21 0.16 15.07 23.82 23.61 23.57 0.45 0.24 0.20 13.60 24.03 23.66 23.59 0.66 0.29 0.22 10.24 24.05 23.79 23.69 0.68 0.42 0.32 9.09 24.17 23.81 23.75 0.80 0.44 0.38 8.20 24.25 23.83 23.76 0.88 0.46 0.39 7.50 24.29 23.95 23.82 0.92 0.58 0.45
下载: 导出CSV
-
王喜珍. 2004. 小波变换在地震数据压缩和震相到时拾取中的应用研究[D]. 北京: 中国地震局地球物理研究所: 87-90. Wang X Z. 2004. Study on Application of Wavelet Transform in Compressing Seismic Data and Picking the Onset Time of Seismic Phase[D]. Beijing: Institute of Geophysics, China Earthquake Administration: 87-90 (in Chinese).
武东坡. 2004. 震相识别的实时方法研究[D]. 哈尔滨: 中国地震局工程力学研究所: 11-22. Wu D P. 2004. Research on the Real-Time Processing Method of Seismic Phase Identification[D]. Harbin: Institute of Engineering Mechanics, China Earthquake Administration: 11-22 (in Chinese).
Wickerhauser M V. 1992. Acoustic signal compression with wavelet packets[G]//Wavelets: A Tutorial in Theory and Application. San Diego, California: Academic Press: 679-700.
计量
- 文章访问数: 695
- HTML全文浏览量: 309
- PDF下载量: 34
