The Research And Application Of Settlement Prediction Of High-speed Railway Based On ARIMA And IGWO-SVM Optimization Model

In recent years,high-speed railway has become China's main mode of transportation.High-speed railways in the era of rapid transit must ensure the safety,smoothness and comfort of train operations,which is essentially a severe test of the post-construction settlement of under-construction projects.As an important component of railway engineering,subgrade settlement prediction has become a key link in controlling railway safety.Currently widely used settlement prediction methods in railway engineering include Asaoka method,three-point method and hyperbola method.But the "high smoothness" of the ballastless track used in high-speed railways has led to the nonlinear characteristics of roadbed deformation data and the sensitivity of "small-scale,large fluctuations".The prediction of high-speed rail deformation has evolved into a nonlinear and complex dynamic system.It is difficult to effectively analyze the variation trend of the settlement system using a single conventional prediction method.In response to this problem,this paper starts with the commonly used prediction model for the characteristics of high-speed railway settlement curves.In order to conduct a more comprehensive analysis of data in the field of deformation,the concept of a combination model of data preprocessing and residual correction is introduced,the applicability of the IGWO-SVM deformation prediction combination model based on wavelet analysis and time series analysis to high-speed railway deformation data is explored.This paper combines the characteristics of nonlinear fluctuations in the settlement data of high-speed railway bases,summarizes the commonly used deformation prediction methods and research advances,and explains the phase space reconstruction techniques,wavelet analysis,time series analysis,the theoretical basis of support vector machines and the parameters selection basis of each model.Secondly,one-dimensional discrete wavelet is used to denoise the noisy random sequence;A one-dimensional time series analysis model that can predict the future development trend of the variable based on the past regularity of the predicted variable is used to establish a model;By using the strong tracking and adaptability of the support vector machine for nonlinear data,based on the construction of training and forecasting datasets using the reconstructed phase space method and the optimization of the performance of hyperparameter selection using the improved grey wolf algorithm based on support vector machine,the above method is applied to the combined forecasting.This article contains two key research contents: The first part is to improve the standard gray wolf algorithm to improve the parameter search performance of SVM;The second part is to construct the ARIMA model prediction for the wavelet denoised data,and use the IGWO-SVM model to modify and compensate the prediction residual to construct the core combination model of this paper.In this paper,the application analysis of the settlement data of Guizhou-Guangzhou high-speed railway is used to verify the applicability and effectiveness of the model.The experimental results show that wavelet decomposition is used as data preprocessing to decompose the original wave series into a relatively more stable and regularized trend term and a few random terms that have little or no relevance to the prediction result.On the basis of eliminating some redundant errors,the wavelet analysis can better grasp the fluctuation and deformation laws of the settlement time series,and improve the fitting performance of the time series analysis and prediction model to the local features of various aspects of the data.The C-C method is used to reconstruct the phase space of the prediction residual samples,and the improved GWO-SVM model is used to perform regression correction on the ARIMA model.It is indeed more accurate and predictive than the traditional support vector regression machine.The results of this study make up for the fact that the traditional single model shows a fixed form,strong pertinence and other sensitive prediction defects,and provides a theoretical guidance for the settlement prediction of high-speed railway roadbed.