Research On Landslide Monitoring Method Based On Multi-Source Sensor Information Fusion

Landslide disaster is one of the worst types of geological disasters in nature.The complex geological conditions in China lead to frequent landslide disasters,which often cause great losses to people’ s life and production.Therefore,the implementation of efficient landslide monitoring and timely warning is always the top priority of landslide disaster prevention.With the constant progress of technology,the latest technologies such as intelligent sensing,Internet of Things,data fusion,and artificial intelligence are continuously introduced into the field of landslide monitoring to improve the actual effect of landslide monitoring and early warning.In particular,the key technologies such as multi-source sensor real-time monitoring of landslide,data fusion method and displacement prediction model are the research hotspots in the field of landslide monitoring in recent years.However,the current landslide monitoring methods still have the following deficiencies: First,the robustness of the landslide multi-source sensor real-time monitoring system is not strong enough to meet the requirements of the complex and harsh environment in the field;Second,the fusion analysis of landslide multi-source sensor data is insufficient,which is not conducive to the comprehensive analysis of landslide monitoring data features and information fusion;Third,the applicability of the existing landslide displacement prediction model is limited,which can’t accurately study the development trend of landslide displacement.To solve the above problems,this dissertation conducts an in-depth study on landslide monitoring methods from the perspective of the actual needs of landslide monitoring.The project adopts low-power Internet of Things technology to build a landslide wireless sensor Ad-Hoc monitoring system to effectively obtain multi-source sensor information of landslides.The multi-source sensor information fusion method is used to analyze the landslide deformation state,and the trend of landslide displacement and deformation is studied and judged through the machine learning coupling prediction model.The main research content and achievements of this dissertation are as follows:1.Constructed a landslide monitoring system based on low-power Lo Ra(Long Range)wireless Ad-Hoc network technology.The hardware structure of a two-layer heterogeneous adaptive hybrid network for landslides is designed using adaptive wireless Mesh technology,and the corresponding communication working mode and power supply optimization scheme are proposed to improve the redundancy and robustness of the existing wireless Ad-Hoc monitoring system for landslides.The low power consumption mode of the landslide Ad-Hoc network system and the adaptive data acquisition method based on threshold triggering are designed,which can automatically and intelligently collect data.The system realizes the monitoring principle of low-risk,low-frequency and high-risk,high-frequency,which improves the pertinence of landslide monitoring.Furthermore,the monitoring method of landslide surface displacement and deformation based on virtual reference station combined with field monitoring station is studied.The data processing method for both service side and monitoring side is designed to solve the problem of high cost caused by the construction of reference station.This method improves the solution capability of the landslide surface displacement and deformation monitoring method without increasing the cost of the system,providing technical support for its application and extension to the field.2.The multi-source sensor information fusion analysis methods combining landslide deformation characteristics and artificial intelligence is studied.To address the problem of missing data in landslide long time series monitoring data,combined with the formation mechanism and deformation characteristics of landslide disaster,a meanbased low rank autoregressive tensor complementation and prediction algorithm for landslide displacement data under two scenarios of regular missing and random missing is proposed,which improves the efficiency and accuracy of the data complementation algorithm.This dissertation analyzes the correlation between landslide multi-source sensor information using graphical,correlation coefficient and grey correlation methods.Multiple linear regression based on partial least square,inverse weighted extended Kalman filter based on spatial distribution of measurement points and neural network model are respectively used to realize landslide multi-source sensor information fusion.The study points out the advantages and disadvantages of different fusion algorithms and effectively analyses the deformation state of the landslide.3.According to the non-linear characteristics of landslide displacement and deformation,a prediction model based on genetic algorithm to optimize the initial weights and thresholds of Elman neural network is proposed for Jianshanying landslide in karst mountainous area,which solves the problem of Elman network falling into local optimization.An ensemble empirical mode decomposition is used to obtain different frequency characteristics of landslide displacements,combined with a support vector regression model to select a Gaussian kernel function to achieve effective prediction of landslide displacements.Considering the more complex geological conditions of Shuizhuyuan landslide in the Three Gorges Reservoir area,the trend term,periodic term and random term components of landslide displacement are effectively extracted by variational mode decomposition.Eight internal and external factors affecting the landslide deformation are demonstrated by using the grey relational degree method.This dissertation constructs a genetic algorithm optimized Elman model and a grey wolf algorithm optimized support vector regression model based on variational mode decomposition respectively,which effectively addresses the problem of reduced predictive performance due to the randomness of the model parameters.The landslide prediction results in two different geological conditions are verified and analyzed.The constructed prediction models have better prediction accuracy and high prediction stability under respective corresponding geological conditions,which provides a reference for the study of displacement prediction of the same type of landslide.