支持向量机在地震预测中的应用
|
摘要
利用支持向量机分析了发生在美国加州中部的2次6级以上地震的震前大地脉动.通过对离地震最近的3个地震台站的地震数据进行震前大地脉动分析,结果表明:支持向量机能有效地区分震前大地异常脉动和平静时期的大地脉动,并且随着地震的临近预报准确率逐渐增加;2次地震的震前大地异常脉动分别始于地震前48h和12h.分析了加州CI地震台网内的14个地震台站记录的2003年12月22日发生在加州中部的6.4级地震所观测的震前脉动数据,发现处在震中附近的12个地震台站均观测到震前大地的异常脉动,且距离震中附近的断层越近,监测到震前脉动异常的几率越大.对3个观测站进行连续监测,结果表明:监测到大地震(M≥5)所引发的震前脉动异常的概率大于小地震(M<5).因此,该方法有望发展成为地震预报的一种有效手段.
A model of detecting the abnormal earth pulsations was built by the analysis of the earth pulsations before two earthquakes(M1=6.4 and M2=6.0) took place in the central California of USA via support vector machines.After the analysis and classification of the pre-earthquake earth pulsations recorded by the three nearest earthquake observation stations,it is concluded SVM could differentiate the abnormal earth pulsations from the normal earth pulsations recorded in the quiet phases of earthquake,and the classification accuracy increased with the approach of the two earthquakes.The abnormal earth pulsations appeared 48 and 12 hours before the two earthquakes,respectively.The established model was applied to analysis of the pre-earthquake earth pulsations of the M1 earthquake(broken out on 22th Dec.2003) recorded by 14 observation stations in CI earthquake nets.The results showed the model detected the abnormal earth pulsations in the 12 observation stations,and the shorter the distance between observation station and the fault near the epicenter, the higher the probability of detecting the abnormal earth pulsations.This model was also employed to detect the abnormal earth pulsations recorded by three observation stations before the M2 earthquake.The results revealed the probability of detecting the abnormal earth pulsations(M≥5) was higher than that of the earthquakes(M<5).This method can be developed to be an effective approach for earthquake prediction.
A model of detecting the abnormal earth pulsations was built by the analysis of the earth pulsations before two earthquakes(M1=6.4 and M2=6.0) took place in the central California of USA via support vector machines.After the analysis and classification of the pre-earthquake earth pulsations recorded by the three nearest earthquake observation stations,it is concluded SVM could differentiate the abnormal earth pulsations from the normal earth pulsations recorded in the quiet phases of earthquake,and the classification accuracy increased with the approach of the two earthquakes.The abnormal earth pulsations appeared 48 and 12 hours before the two earthquakes,respectively.The established model was applied to analysis of the pre-earthquake earth pulsations of the M1 earthquake(broken out on 22th Dec.2003) recorded by 14 observation stations in CI earthquake nets.The results showed the model detected the abnormal earth pulsations in the 12 observation stations,and the shorter the distance between observation station and the fault near the epicenter, the higher the probability of detecting the abnormal earth pulsations.This model was also employed to detect the abnormal earth pulsations recorded by three observation stations before the M2 earthquake.The results revealed the probability of detecting the abnormal earth pulsations(M≥5) was higher than that of the earthquakes(M<5).This method can be developed to be an effective approach for earthquake prediction.
引文
[1]GELLER R J,JACKSON D D,KAGAN Y Y,et al.Earth-quakes cannot be pred icted[J].Sc ience,1997,275:1616.
[2]AKINAGA Y,HAYAKAWA M,LIU J Y,et al.A precursoryULF signature for the Chi-Chi earthquake in Taiwan[J].Natu-ralHazards and Earth System Sciences,2001,1:33-36.
[3]GOTOH K,HAYAKAWA M,SM IRNOVA N A,et al.Fractal analysis of se ismogen ic ULF em issions[J].Physicsand Chem istry of the Earth,2004,29:419-424.
[4]IFANTIS A,G IANNAKOPOULOS K.Changes of chaoticbehavior of the long-term geoelectric potential d ifference ob-served during a five-year investigation and its possib le rela-tion to se ism ic activity in W eatern G reece[J].Chaos,Soli-tons and Fractals,2002,14:779-795.
[5]OUILLON G,SORNETTE D.The concept of critical earth-quakes app lied to m ine rockbursts w ith tim e-to-failure analy-sis[J].Geophysical Journal International,2000,143:454-468.
[6]B IAG I P F,ERM INI A,K INGSLEY S P.G roundwater ioncontent precursors of strong earthquakes in Kam chatka(Rus-sia)[J].Pure and App lied Geophysics,2000,157:1359-1377.
[7]MOLCHANOV O,FEDOROV E,SCHEKOTOV A,et al.L ithosphere-atmosphere-ionosphere coup ling as govern ingm echan ism for prese ism ic short-term events in atmosphereand ionosphere[J].NaturalHazards and Earth System Sc i-ences,2004,4:757-767.
[8]强祖基,徐秀登,赁常恭.卫星热红外异常—临震前兆[J].科学通报,1990,35(17):1324-1327.
[9]TRAMUTOLI V,BELLO G D,PERGOLA N,et al.Robustsatellite techn iques for remote sensing of se ism ically activeareas[J].Annali d i Geofisica,2001,44:295-312.
[10]TRONIN A A,B IAG I P F,MOLCHANOV O A,et al.Temperature variations related to earthquakes from simu lta-neous observation at the ground stations and by satellites inkam chatka area[J].Physics and Chem istry of the Earth,2004,29:501-506.
[11]VAPNIK V.The nature of statistical learn ing theory[M].New York:Springer,1995.
[12]CAI C Z,HAN L Y,JI Z L,et al.SVM-Prot:web-basedsupport vectorm ach ine software for functional c lassificationof a prote in from its prim ary sequence[J].Nuc le ic Ac idsResearch,2003,31(13):3692-3697.
[13]CAI C Z,HAN L Y,JI Z L,et al.Enzym e fam ily c lassifi-cation by support vector m ach ines[J].Prote ins,2004,55(1):66-76.
[14]CAI C Z,WANG W L,SUN L Z,et al.Prote in functionpred iction via support vectorm ach ine approach[J].M ath-em atical B iosc iences,2003,185:111-122.
[15]CAI C Z,WANG W L,CHEN Y Z.Support vector m a-ch ine c lassification of physical and b iological datasets[J].Inter JMod Phys C,2003,14(5):575-585.
[2]AKINAGA Y,HAYAKAWA M,LIU J Y,et al.A precursoryULF signature for the Chi-Chi earthquake in Taiwan[J].Natu-ralHazards and Earth System Sciences,2001,1:33-36.
[3]GOTOH K,HAYAKAWA M,SM IRNOVA N A,et al.Fractal analysis of se ismogen ic ULF em issions[J].Physicsand Chem istry of the Earth,2004,29:419-424.
[4]IFANTIS A,G IANNAKOPOULOS K.Changes of chaoticbehavior of the long-term geoelectric potential d ifference ob-served during a five-year investigation and its possib le rela-tion to se ism ic activity in W eatern G reece[J].Chaos,Soli-tons and Fractals,2002,14:779-795.
[5]OUILLON G,SORNETTE D.The concept of critical earth-quakes app lied to m ine rockbursts w ith tim e-to-failure analy-sis[J].Geophysical Journal International,2000,143:454-468.
[6]B IAG I P F,ERM INI A,K INGSLEY S P.G roundwater ioncontent precursors of strong earthquakes in Kam chatka(Rus-sia)[J].Pure and App lied Geophysics,2000,157:1359-1377.
[7]MOLCHANOV O,FEDOROV E,SCHEKOTOV A,et al.L ithosphere-atmosphere-ionosphere coup ling as govern ingm echan ism for prese ism ic short-term events in atmosphereand ionosphere[J].NaturalHazards and Earth System Sc i-ences,2004,4:757-767.
[8]强祖基,徐秀登,赁常恭.卫星热红外异常—临震前兆[J].科学通报,1990,35(17):1324-1327.
[9]TRAMUTOLI V,BELLO G D,PERGOLA N,et al.Robustsatellite techn iques for remote sensing of se ism ically activeareas[J].Annali d i Geofisica,2001,44:295-312.
[10]TRONIN A A,B IAG I P F,MOLCHANOV O A,et al.Temperature variations related to earthquakes from simu lta-neous observation at the ground stations and by satellites inkam chatka area[J].Physics and Chem istry of the Earth,2004,29:501-506.
[11]VAPNIK V.The nature of statistical learn ing theory[M].New York:Springer,1995.
[12]CAI C Z,HAN L Y,JI Z L,et al.SVM-Prot:web-basedsupport vectorm ach ine software for functional c lassificationof a prote in from its prim ary sequence[J].Nuc le ic Ac idsResearch,2003,31(13):3692-3697.
[13]CAI C Z,HAN L Y,JI Z L,et al.Enzym e fam ily c lassifi-cation by support vector m ach ines[J].Prote ins,2004,55(1):66-76.
[14]CAI C Z,WANG W L,SUN L Z,et al.Prote in functionpred iction via support vectorm ach ine approach[J].M ath-em atical B iosc iences,2003,185:111-122.
[15]CAI C Z,WANG W L,CHEN Y Z.Support vector m a-ch ine c lassification of physical and b iological datasets[J].Inter JMod Phys C,2003,14(5):575-585.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心