A Study Of Time Series Analysis Method Based On Robust Estimation For Deformation Monitoring

The analysis and forecast of data are major contents in the processing of deformation monitoring data. How to use finite deformation monitoring data and choose rational model to accurately forecasts deformation is one of the most important task of the study of the processing of deformation monitoring data.Based on the study of deformation point in high slope, for raising the forecasting accuracy, building of forecasting models with the method of time serial analysis are profoundly studied in this paper. It as follows:1.The issue on how to recognize the model, estimate parameter, test applicability of the model about stationary time series are discussed. Aiming at non-stationarity of most deformation monitoring data, two methods of building models for non-stationary data are discussed, that contain the method of eliminating tendency item and the method of distilling tendency item. Using actual data with two methods, model is built; By comparison, it is found that two methods separately have merits and faults, so we should use different method in practice. For eliminating tendency item, the conventional methods contain linear function, power function, exponential function, periodic function, and so on. Gray model is extensively applied to data processing in some fields in the close years; so by eliminating tendency item with the model, gray time serial model is built.2.Robust estimation theory is profoundly studied, and several weight functions are discussed. By simulation test, it shows that robust estimation, which is applied to parameter estimation, can resist gross errors. Given that it is inevitable that there are gross errors in deformation monitoring data, gray time serial model based on robust estimation is built.3.Given that change of body and structure of deformable body is complex and multivariate, the forecasting precision is gradually reduced with the delayed forecasting time with conventional gray time serial model, because of invariable model parameter. So dynamic gray time serial model based on time-varying parameter is built, at the same time, robust estimation is introduced in the process of building model in order to resist gross errors. The result proves that the method of building dynamic gray time serial model based on robust estimation can forecast longer and precisely.