基于小波分析和支持向量机的变形监测数据处理

摘要

变形监测数据分析与预报结果是进行测量决策的主要依据.针对如何使用已知有限的线性关系复杂的变形信息进行分析建模的难题,该文利用基于小波分析和支持向量机的方法对变形测量数据进行处理,解决了小样本数据处理以及数据中存在噪声的问题.先通过对变形监测数据时间序列小波分解,使用阈值消噪处理后进行重构.另外同时对消噪后的时间序列分量进行SVM和LS-SVM预测,然后重构.对比SVM预测、LS-SVM预测以及消噪后的原始序列的各自重构序列,结果经评价得出,采用此方法SVM与LS-SVM均能得到较好预测结果,且LS-SVM预测结果更优.%The deformation monitoring data analysis and prediction results are the main basis for measure decision-making.But due to the problems that how to analyse and build model by the existing,limited and complex linear relation deformation information,this paper processes the deformation measuring data by the methods of wavelet analysis and SVM,and it sloves the problems of small sample data processing and the noise existing in the data.First,the time series of deformation monitoring data is decomposed by wavelet.After the threshold denoising,the reconstruction is made.At the same time,the components of time series after denoising are predicted by the SVM and LS-SVM,and then reconstructed.Comparing the the SVM prediction、LS-SVM prediction and the recombinant sequences by original sequence after denoising,the evaluated results show that,by these methods both of SVM and LS-SVM all can get better prediction results,and the prediction results by LS-SVM are better.