Deformation Monitoring Analysis And Forecasting Based On Time Series Analysis Combined Model

The purpose of deformation monitoring is to obtain the changes of the deformation body(the whole earth, and small to a building) in state space and time characteristics. And thenanalyzes the cause of the deformation, forecasting future deformation. The important contentof deformation monitoring data processing is the use of the limited monitoring data to select areasonable prediction model, more accurate forecasting of future deformation in the future thedeformation prediction。The object of this paper is two different sets of deformation data. Respectively for thetwo sets of data forecast model is set up using time series analysis method, and carries on thethorough analysis and research. This paper is completed as follows:1. It is discussed how to identify primarily for stationary time series model, how to carryout parameter estimation, practical test and forecasting process. It implements on the timeseries analysis combined with SAS software in this paper. Two different modeling methodsare discussed for the characteristic that most of the deformation datas were non-stationary.The first is through differential method eliminate trend item many times, the second is usingdifferent models to trend item extracted out.2. First use of traditional gray model to extract and predict the two groups ofnon-stationary date trend term, the precision of prediction is getting lower and lower overtime. We propose a time-varying parameters of gray model for this shortcoming and provedthat dependent parameter model than the traditional gray model is more suitable forlonger-term forecasts through the two sets of forecasts instance.3. Sometimes we can not use a certain mathematical expressions to fit the complexity ofdata due to the complexity of the deformation monitoring data. We use BP neural networkmapping function on two sets of data and combined with the MATLAB language to fit thedata and forecasts. The results show that the prediction precision than gray model oftime-varying parameters improved.4. The generalization ability of the BP neural network is relatively weak and theextrapolation is not strong. Wavelet neural network model based on adaptive learning rate isestablished. This model combined the BP neural network basic algorithm with activationfunction of wavelet function, and use the MATLAB language for programming the model.The prediction precision of wavelet neural network is the highest and has good fault tolerance.It has a good practical value in the prediction of deformation monitoring.