Research On Data Assimilation Method Of Large Landslide Based On Improved Particle Filter

Landslides are one of the most frequent and destructive natural disasters on the planet,posing a huge threat to human life and property security.In recent years,the research on the evolution process,geomechanics,structural features and mechanism models of landslides has developed rapidly.With the increasing maturity of geostationary observation technology,long-term and large-scale landslide surface deformation monitoring technology has also been promoted and applied.Data assimilation is a technique that combines the mechanism model with the observation data to update the model parameters and correct the model prediction trajectory.It has been deeply studied and widely applied in the fields of atmosphere,ocean and land surface.Particle filtering is an emerging data assimilation algorithm since the 21 st century.It has become a research hotspot in data assimilation because it is not subject to the constraints of model linear conditions and is not subject to the constraints of Gaussian distribution assumptions.However,there are still some difficulties in the application of landslide data assimilation and particle filter algorithms.Firstly,standard particle filter has the problem of particle degradation and low particle efficiency.It is usually necessary to sample a large number of particles to get better results in large landslide data.When applied in assimilation,it will undoubtedly increase the computational burden.Secondly,due to the complexity of its type and geological structure,landslides also have some difficulties in applying data assimilation techniques.Again,most of the existing landslide mechanism models are difficult to directly apply to landslides.Surface deformation monitoring data is also difficult to achieve continuous parameter update and feedback with the increase of observation data.It is difficult to realize the assimilation process.Finally,although there are existing data assimilation techniques in geological disaster monitoring at home and abroad,there are no reports of complete and systematic large-scale landslide data assimilation research cases.The above problems bring great difficulties to the application of landslide data assimilation.This paper has carried out a series of research work on the above problems and proposed corresponding solutions.The main research contents are as follows:(1)Two problems have been proposed for the problem of particle starvation and low particle computation efficiency,which are difficult to avoid by particle filtering: mixed Gaussian distribution sequence resampling and posterior particle distribution adjustment.The former makes particle filtering avoid the repetition of particles in the process of resampling,and ensures the diversity of particles,thus overcoming the phenomenon of particle starvation;the latter adds a “gain” term in the process of particle updating to make the particle update distribution.Translating to the interval with high observation probability density greatly increases the number of effective particles and reduces the inefficient particles,which can greatly reduce the number of particle filter samples and improve the efficiency of particle filtering.(2)Discuss the spatial difference and projection relationship between the GPS monitoring 3D deformation data and the In SAR radar line-of-sight deformation data,and analyze the proportional relationship between the radar line-of-sight shape variable and the actual total shape variable.Multiply the In SAR observation data by the proportional coefficient is used as the observation data of the assimilation experiment.The spatial difference between the In SAR observation data and the TRIGRS grid model data is studied.According to the correlation characteristics between the observation data and the TRIGRS grid model,the interpolation method of observation data is studied.The feature construction of the particle filter algorithm is combined.A data assimilation framework for particle filtering.(3)Aiming at the key parameters that need to update feedback in the process of data assimilation,a method of nonlinear parameter estimation with particle filter is proposed.A set of parameter particles corresponding to the state particles is established by TRIGRS model,according to Monte Carlo.The idea uses the weighted average of the parameter particles as the parameter estimation value to realize the parameter update;the updated parameter is used as the initial parameter input model of the next assimilation time to realize the parameter feedback.(4)A landslide data assimilation platform based on TRIGRS model is designed.The data assimilation software based on particle filter algorithm is developed.The functions of data format conversion,sensitivity analysis of key parameters,interpolation of observation data,data assimilation and result analysis are realized.(5)The small-scale slope stability assimilation simulation experiment and large-scale landslide surface deformation monitoring data and TRIGRS model assimilation experiments were carried out.The internal friction angle is taken as the parameter to verify the feasibility and effectiveness of the improved particle filter algorithm in landslide data assimilation.The safety factor,internal friction angle,groundwater pressure head and deformation of the sloped and mesh elements after assimilation are analyzed.The rate changes with the assimilation time;and the research prospects for establishing the landslide data assimilation system are discussed.