On Studying The Application Of Local Mean Decomposition And Gaussian Process In Extraction Signal Analysis And Forecasting Research Of Deformation Signal

Deformation monitoring and analysis is an important component of safety monitoring system, the reliability and precision of the analysis method plays an important role in maintaining the safe operation of the deformation. It is the key problem in deformation monitoring to treatment deformation data, extract characteristic vector and conduct deformation prediction, and it is also one of researching hot-spots for the disciplines of surveying and mapping. Local mean decomposition is a new adaptive non-stationary signal processing method. The characteristics of this method is that it can adaptive to more effective resolve the complex non-stationary signal so that it provided a new research method for forecasting and extracting the deformation feature. Gaussiau Process has a strict theoretical foundations of statistical learning theory, good adaptability to deal with the high dimensionality, sample wood and nonlinear complex. In terms of the scientific prediction accuracy and probability, it is a good nonlinear learning method.In this paper, the prediction model and extract deformation feature based on local mean decomposition was studied, and combined with the simulation signal and the dam data of deformation monitoring to analyzes and discusses the related problems.(1)The author did some research in theory of LMD end effect and modal aliasing aspects to improve the algorithm of LMD efficiency. According to the physical meaning of the LMD product functions and analysis components caused by the deformation of the influencing factors of the deformation which are established each component of deformation and deformation relationship between influencing factors and multiscale deformable component prediction model by using the GP algorithm. The deformation prediction model of superimposed component fusion, then get multi-scale deformation prediction model based on nonlinear LMD-GP.(2) For the problem of deformation monitoring data preprocessing, a new wavelet threshold function denoising method based on LMD is established. It makes full use of the new wavelet threshold function and LMD, which greatly improves the efficiency of denoising. Aiming at the problem of gross error detection in the case of different deformation, the author proposed a method which is based on the traditional method of gross error detection and LMD, which is complementary to each other to improve the accuracy and efficiency of gross error detection.(3) The observed sequence of data is processed by LMD decomposition, and using the correlation coefficient to analyze the deformation and feature extraction. Finally,the deformation components are modeled, then the prediction model is obtained by accumulating all the deformation components. The author established the prediction model named LMD-PSO-GP and ELMD-LSSVM. Experimental shows that the modeling accuracy of combination model is much higher than that the single model.The two methods are suitable for nonlinear nonstationary variable geometry data analysis and forecasting.(4) The last section of the fifth chapter, the author consider the GP single point prediction model for the deformation of the adjacent points. The next step is to establish the multi scale multi point prediction model based on LMD and GP theory.Two experiments show that the influence among the deformation of the adjacent points on the monitoring points can not be ignored, and the deformation factors are added to the modeling process to enhance the effect significantly.