The Research Of Baishuihe Landslide State Transition Based On Dempster-Shafer Theory

Landslide is one of the most harmful geological disasters. It makes a serious threat and loss to the public facilities and the security of people’s lives and properties. Bedding landslide is one of the most common type of landslide in the three gorges reservoir area. Once the landslide instability, it causes great damage to reservoir safety. Study shows that the probability of landslide is low when bedding landslide is at the stage of slow deformation. But when bedding landslide is at the stage of accelerative stage, the probability of landslide is high. The research of landslide transition probability has tremendous practical value to landslide prediction.Taking "ZG93" monitoring point of Baishuihe landslide as study object, this paper deeply studies the probability of landslide transition. The main research contents are as follows:(1) Expand sample data. Use engineering analog method and artificial rainfall experiment to expand as prior data. (2) Preprocess the prior data. Analyze the data of cumulative displacement, rainfall and reservoir level respectively. Divide landslide states and extract the rainfall feature vector. (3) Use Dempster-Shafer evidence theory in prior data fusion. Based on the geological and hydrological conditions of the Baishuihe landslide, set whether the landslide state is transited to the stage of accelerative as the problem determination, and determine the evaluation index and recognition framework. Adopt statistical method to calculate prior data to get the basic probability assignment of each evaluation index, and form evidence’s basic reliability. Use D-S fusion theory to calculate the probability of landslide transition.The main innovations are as the following:(1) D-S evidence theory is used to study landslide state transition objectively. This paper not only retains the advantages of expert knowledge, but also introduces the artificial intelligence and the objective validity of math, which provides a new method for landslide state transition study.(2) Through analyzing and extracting the rainfall data and combining with the geologic characters of Baishuihe landslide, this paper retains the inner link and rules of rainfall data to calculate the weighted evaporation, infiltration, runoff [α,β,γ] and takes this as the rainfall feature vector for each month.Through the above research, the paper makes some conclusions. Taking Baishuihe landslide as a practical engineering example, this paper uses the statistical method to calculate prior data to get the basic probability assignment for each evaluation index.There are three specific situations. First, the transition probability is large when the early landslide state is at the uniform deformation stage. Second, the transition probability is large when the rainfall feature vector belongs to the first type. Third, the transition probability is large when the reservoir level change belongs to the first type. D-S evidence theory is used in the fusion of three evaluation indexes. The transition probability is the largest when the three evaluation indexes are all in their respective specific situation.