Research Of Settlement Prediction And Data Processing Based On Robust Estimation On High-speed Railway Underline Construction

Settlement prediction model has been widely used in existing projects, but compared with the measured data, there still exists a certain deviation in the predicted result. For possible gross error in the observation data and singular value in the process of data preprocessing, mainly accorded to the data processing worker’s personal experience, deleted it subjectively. This approach not only failed to ensure that the predictions were true, and the process was complicated.In order to solve this kind of problem, this article started from common curve forecast models, then introduced optical combination forecasting method before comparing the combination forecasting models based on four different criteria weight function assignment And robust estimation theory was applied to the curve forecast model. Then, this paper put forward robust hyperbolic prediction model and robust Asaoka prediction model. The applicability and the predictive effect were also analyzed combining practical data.The main contents and issues conducted in this paper were generalized as follows:1) According to the existing settlement prediction theory, advantages and disadvantages and limitations of curve prediction model were summarized after focusing on the analysis of the curve prediction model based on measured data. After that, this paper compared the robustness of multiple combination forecasting models based on the weight constraint criteria.2) Robust hyperbolic prediction model and robust Asaoka prediction model were established. With the MATLAB languages, the relevant calculating program was developed and the measured settlement data were used to verify the procedure. Through the analysis of hyperbolic model this paper found that the later the initial prediction time, the more prone to singular value, and the singular value of the order of magnitude greater than the gross error. However, the Asaoka prediction model in the process of data preprocessing is not easy to produce an order of magnitude larger singular value.3) Curve prediction model based on robust estimation of goodness-of-fit was generally superior to ordinary curve prediction model in the data pretreatment process. Beside, Curve prediction model based on robust estimation could solve the predictive result distortion problems of the gross error mixed in original data and the singular value produced in preprocessing.4) Statistical analysis of the experimental results shows that:the model parameters obtained in the unqualified goodness-of-fit (<0.7) could lead to distortion when transformed the curve prediction model into a linear model for solving the model parameters. Under this condition the prediction value or final settlement value was also inaccurate and unreliable.5) GUI programs of corresponding calculation method were developed by MATLAB language. The new settlement prediction system was proved to be efficiency and convenience.The researches above indicated that, the curve prediction models combined with the robust estimation theory can resist the impact of rough observational data and singular value.