Application Of Kalman Filter Method In Landslide Deformation Forecast

The Three Gorges reservoir is reached175-meter-high water level in2009. According to the statistics of built reservoir, nearly50%of the reservoir-induced slope failures take place during the first impoundment and the other mainly during the first3-5years after the dam construction. For early warning of geological disasters, the government launched a comprehensive, continuous monitoring of geological disasters. Thus research on landslide displacement prediction model has a strong practical and theoretical research value.Since2003, the Three Gorges reservoir area has carried out a series of geological disasters monitoring and early warning. With automated, real-time monitoring techniques accumulated massive landslide data, and effective use of the monitoring data is critical to explore landslide forecast. Landslide displacement prediction is a highly complex, non-linear problem. On the one hand the landslide deformation is controlled by its own geological conditions, on the other hand will be influenced by the external environment factors. Thus landslide deformation process is not only showing the macro trends of the displacement, but also showing some random volatility. In the paper, Research on landslide system is based on the monitoring-feedback theory of Engineering geology System Science. Based on the analysis of Landslide cumulative displacement, landslide deformation process is divided into trend item and random item. This part takes landslide displacement and deformation rate as the state vector of landslide system, and takes the instantaneous displacement rate as a first-order Markov random process. In order to research on the dynamics, nonlinear and openness of landslide systems, the paper uses the method of Kalman filter and time series analysis.The paper taking Laoshewo landslide as an example, analyzes the causes of the landslide mechanism by the geological characteristics of landslide monitoring data, concludes the law of landslide deformation, and builds the Kalman-ARIMA model to predict landslide displacement and deformation. Through an analysis of these issues and research achieved the following:1. Laoshewo landslide is a typical pushing type of landslide and its deformation process is mainly affected by rainfall and the decrease of reservoir level. According to the2006to 2010landslide monitoring data statistical analysis, landslide displacement deformation has a trend of decreasing from the trailing edge to the leading edge. Analysis of landslide cumulative displacement, strain rate, rainfall and water level curves, and the landslide displacement deformation curve has a characteristic of step response. The faster rate of landslide deformation stages are mainly between the May to September. Analysis of the reasons, landslide deformation is affected by heavy rainfall, continuous rainfall and the decrease of reservoir level.2. Laoshewo landslide displacement prediction model is based on the concept model approach of landslide movement system, which coupling with Kalman model and ARIMA model. This section is building the displacement prediction model of Kalman and Kalman-ARIMA based on the data of landslide displacement during March2008and October2010. Through comparative analysis of experimental results shows that contrast Kalman model predictions have large deviations, maximum is35.58mm and the average absolute error is10.33mm. While the Kalman-ARIMA model predictions have relatively small deviations, the biggest one is19.75mm and the average absolute error is4.59mm. Thus, Kalman-ARIMA model prediction has a higher accuracy as it can effectively consider the influence of external factors to landslide displacement prediction.