Research On Denoising Analysis And Modeling Of High Slope Deformation Data Based On Improved CEEMD-SVD

In the field of high slope deformation monitoring,due to the complexity of the actual project and the uncertainty of the external environment,the monitoring instruments are susceptible to various factors(environmental incentives,human operations,etc.),resulting in the original monitoring data often containing noise.Because of the existence of noise,the volatility of the monitoring data is enhanced,and the actual deformation trend of the project is largely obscured.In severe cases,the monitoring data may even be completely opposite to the real deformation,resulting in the risk of engineering safety.In the process of judgment and later predictive modeling,wrong conclusions are drawn,which in turn causes casualties and property losses.Therefore,it is urgent to carry out reasonable de-noising processing on the original monitoring data,so that the de-noised data can fully describe the actual deformation characteristics of the high slope and reflect its inherent regularity,so as to establish an accurate forecasting model to prevent it from happening.Traditional de-noising methods and prediction models have been unable to meet today’s engineering needs due to their own shortcomings(easy to lose detailed information in the original data,cumbersome calculations,etc.).In view of this,this project aims to construct a prediction model with strong self-adaptation,simple structure and few pre-adjusted parameters,which will be explained from two aspects of denoising and prediction:(1)The improved CEEMD-SVD denoising method for high slope The de-noising processing of deformed data is effective and has high adaptability;(2)The ARIMA-PSO-GRNN model with residual error correction has the advantages of simple structure,few pre-adjusted parameters,and high modeling efficiency,and can better predict high The future deformation trend of the cumulative horizontal displacement of the slope.The main research contents and results of this project are as follows:1.A denoising method based on improved CEEMD-SVD is proposed.Through the CEEMD decomposition of the observation data,a series of IMF components with frequency from high to low are obtained,and the randomness of each frequency component is tested by permutation entropy.For the components greater than the set threshold,singular value decomposition is used to filter out each component.High-frequency white noise;components less than the threshold value are analyzed by correlation function to eliminate the false components generated by CEEMD decomposition,so as to achieve the purpose of noise elimination.Experiments show that the de-noised data can not only clearly reflect the detailed information in the observation data,but also improve the modeling efficiency of the ARIMAPSO-GRNN model,and provide a certain reference for the study of high slope deformation and stability in accordance with.2.Use the denoising method proposed in this subject to denoise the simulation signal and the measured deformation data of the high slope,and remove the first high frequency component generated by the CEEMD decomposition,remove the component whose permutation entropy is greater than the set threshold and be based on different The threshold and wavelet denoising methods of different threshold functions are compared.In the process of wavelet threshold denoising,select heursure,rigrsure,sqtwlolg,minimaxi threshold methods to denoise soft and hard threshold functions respectively.In order to more intuitively reflect the denoising ability of various denoising methods,entropy is used the weighting method empowers multiple traditional evaluation indicators to verify the de-noising effect of the method proposed in this topic.3.Establish an ARIMA-PSO-GRNN model with residual correction.Apply the ARIMA model to the improved CEEMD-SVD denoising data to make linear predictions;use GRNN neural network to regress and correct the prediction errors and make nonlinear predictions.In terms of parameter optimization,select particle swarm optimization(PSO)to optimize GRNN The smoothing factor of the neural network.By comparing with a variety of prediction models,the model proposed in this topic does improve the prediction accuracy and prediction effect to a certain extent.