巴东组软岩顺层滑坡预测中多参量流数据的信息融合方法研究

The stream data for landside prediction are difficult to process due to its complexity and large quantity. Currently, the numerical methods of using the mean value and the maximum value overlook many important features contained in the data. In this project, we attempt to conduct information fusion of multiple stream data for landslide prediction. The data that include rainfall data, infiltration data, and slop deformation data are from Badong Formation soft rock bedding landside. Firstly, according to the geologic features of stream data, choosing proper feature analytical methods to analyze the features of monitoring stream data, establishing a general model feature analytical model through analysis and demonstration. Secondly, Based on the geologic features, designing feature extraction requirements and purpose, and in this framework aiming at different conditions and different stream data types, to study proper stream data feature extraction algorithm and models, to extract feature of stream data and construct a series of feature subsets. Finally, aiming at feature subsets with the characteristics of time non-synchronization and dimension disunity, to design a sort of heterogeneous space model to realize time match and spatial match, using Bayes theory and D-S theory for feature information integration and the completion of feature class information fusion, to realize an accurate description of the information model for landside state, nature and feature. This study is of great significance for a complete description of the landside stream data and will provide advanced feature model for accurate landside prediction.

滑坡预测中的流数据具有数据量大、包含信息多、处理难度大等特点。目前运用的均值、极值等数值方法大大的弱化了其中包含的特征信息和内在规律。本项目中，以巴东组软岩顺层滑坡中一个典型斜坡的降雨流数据、入渗流数据和边坡变形流数据为研究对象，根据流数据的地质特征选取合适的特征分析方法对监测流数据进行特征分析，通过分析和论证建立特征分析的一般模型；以地质特征为依据设计特征提取的要求和目的，并在此框架下针对不同工况和不同流数据类型研究合适的流数据特征提取算法和模型，对流数据进行特征提取，构造一系列的特征子集；针对时间不同步、维度不统一的特征子集设计一种异构空间模型进行时间匹配、空间匹配，运用贝叶斯理论和D-S理论对特征信息进行集成，完成特征级信息融合，实现信息模型对滑坡状态、性质与特征的准确描述。该研究对滑坡流数据信息量的完整获取有重要意义，为滑坡预测的精确量化信息模型的建立奠定基础。

三峡库区巴东组软岩顺层滑坡稳定性受降雨影响较大，针对降雨、入渗、蒸发、径流数据采集的频率高却多用均值极值等简单处理方法的现状，本项目提出了对多源数据运用信息融合的方法进行处理，提取这些数据的特征量，运用特征进行分析和建模，探索了流数据的信息融合处理方法。主要成果如下：（1）降雨流数据和变形流数据可以分解成不同频率分量的集合；（2）根据降雨实际分配作用，提出了一种基于降雨事件的降雨流数据特征提取方法，该方法在大大缩减数据量的同时提高了降雨参与滑坡位移预测的性能；（3）提出了一种变形流数据特征提取方法并在滑坡稳定性时变规律分析中验证了其有效性；（4）提出了一种多监测点相似性评价方法，并将其运用于滑坡表面分区研究，继而开展了滑坡分区融合和运动形态分析；（5）运用证据理论构建了一种综合运用降雨流数据、库水流数据和变形流数据特征的不同时空尺度的多维多参融合模型，探索了滑坡状态跃迁概率。