Parametricandnonlinearfeatureextractionofnuclearexplosions andearthquakesbasedonVolterraseries
-
摘要: 地震时间序列的模型参数可以作为核爆地震信号识别的特征。现有基于ARMA(autoregressive movingaverage)模型参数的地震信号特征提取方法是一种线性方法,且只利用了信号的二阶统计信息,识别 精度不高。为此,利用地震波的混沌特性,提出了一种核爆地震非线性特征提取方法:首先对地震波信号进行 相空间重构,然后利用Volterra级数在重构的相空间内建立自适应预测模型,最后提取模型参数作为特征。 在核爆地震分类实验中,非线性特征与线性特征相比,表现出更好的分类性能。研究结果表明:综合利用地震 波信号的线性、非线性以及高阶统计信息对于核爆地震识别是非常重要的。Abstract: ARMAparameterscanbeusedasfeaturesfordiscriminationbetweennuclearexplosions andearthquakes.Currentmethodsforfeatureextractionarelinearandonlytakesadvantageofthe second-orderstatistic,sotheclassificationaccuracyofisnothigh.Tosolvethisproblem,amethod forextractingnonlinearfeaturesofnuclearexplosionsandearthquakeswasproposedbasedonthechaoticfeatureofseismicwaves. Firstly,thephasespaceofseismicwaveswasreconstructed.Secondly, theadaptivepredictionmodelbasedonVolterraserieswasestablishedinthephasespace.Finally,the modelparametersweretakenasthefeaturesofseismicsamples.Intheclassificationexperimentofof nuclearexplosionsandearthquakes,nonlinearfeaturesobtainedbetterperformancethanlinearfeatures. Investigatedresultsshowthatthecombinationoflinear,nonlinearandhigher-orderstatistical informationiscriticalfortheclassificationofnuclearexplosionsandearthquakes.
计量
- 文章访问数: 2319
- HTML全文浏览量: 74
- PDF下载量: 209
- 被引次数: 0