基于EEMD分形和LS-SVM的次声信号识别泥石流类型

Identification of Debris Flow Types by Infrasound Signals Based on EEMD Fractal and LS-SVM

基于无接触的泥石流次声信号的有效波形特征提取识别不同类型的泥石流及预警泥石流规模、危害是国内外泥石流研究的新方向。本研究利用室内实验采集到的65次稀性、过渡性和粘性泥石流次声信号数据,采用集成经验模式分解(EEMD)对次声信号进行分解,提取本征模态函数(IMF)主分量,对比分析了原始信号和主分量信号STFT分布的时频特性差异,计算了主分量IMF盒维数值,并将其作为特征值输入最小二乘支持向量机(LS-SVM)分类器进行训练和分类,初步实现了基于次声分形特征指标识别泥石流类型。研究表明:(1)通过对EEMD重构的主IMF分量信号进行短时傅立叶变换(STFT)时频分析后,主分量信号具有优良的时频聚焦性能,提升了信号识别的准确性和精度;(2)稀性、过渡性和粘性泥石流的原始次声信号盒维数值分别为1.625、1.578和1.519,利用次声盒维数值可以识别泥石流的类型;(3)通过LS-LSVM模型训练测试,正确识别率达87%,其中稀性和过渡性泥石流为80%,粘性泥石流为100%。本研究利用次声特征指标无接触判识了泥石流类型,为次声自动识别、监测和预警泥石流灾害做了积极探索。

Infrasound signals produced by travelling debris flow are informative in recognizing debris flows.The waveform feature of the signals sampled by non-contact method can provide rich information in determining and pre-warning of debris flow types and magnitudes.In this study,by using signals collected at 65 debris flows with high and medium,as well as low viscosities,it extracted the principal components of the Intrinsic Mode Function(IMF)by the method of the Ensemble Empirical Mode Decomposition(EEMD),and compared the Short-Time Fourier Transform(STFT)distribution between the extracted signals and original signals.Then the box dimension of the principal components was input as eigenvalue in the Least Squares Support Vector Machines(LS-SVM)to train and classify the infrasound data for distinguishing debris flow types.It is found that:(1)The principal IMF component signal reconstructed by EEMD had excellent time-frequency focusing performance under STFT,which improved the accuracy and availability of signal recognition;(2)The original infrasound signals presented different fractal-box dimensions for debris flows with low,medium,and high viscosity,respectively of 1.625,1.578 and 1.519.Through LS-SVM model training,the type recognition achieved accuracy of 87%.In conclusion,this research had realized the identification of debris flow types based on a non-contact infrasound signals and it will provide a novel way for monitoring and early warning of debris flow disasters.