Application Of Swarm Intelligent Optimization Based Machine Learning In Landslide Deformation Prediction

The landslide is a common geological disaster in nature of our country.Its occurrence frequency is high and it has the characteristic of regional distribution.Therefore,it is a necessary and urgent scientific task to carry out high precision monitoring and prediction of slope.With the help of the application of machine learning in the prediction field,this paper takes the Heifangtai landslide in Yongjing County,Linxia Hui Autonomous Prefecture,Gansu Province as the research area,discusses the application of swarm intelligent optimization algorithm and machine learning model algorithm in the prediction of landslide displacement,and analyzes the stability of landslide by establishing a real three-dimensional model.The main achievements are as follows:(1)The Elman neural network model with the combination optimization of genetic algorithm and particle swarm optimization is established,which further improves the convergence speed and prediction accuracy of Elman neural network.Aiming at the complex nonlinear characteristics of loess landslide displacement,an Elman neural network model(GA-PSO-Elman)optimized by genetic algorithm and particle swarm optimization algorithm was established.Considering that Elman neural network may fall into local optimal solution due to the randomness of structural parameters,the combination of genetic algorithm(GA)with strong global search ability and particle swarm optimization(PSO)with strong local search ability is considered to optimize the weight threshold of structural parameters of the prediction model,and improve the prediction accuracy and convergence speed.The example analysis shows that compared with the traditional BP neural network and the single Elman neural network model,the new model has better accuracy and stability.Further considering many factors affecting the landslide and adding humidity,precipitation and other influencing factors into each training model,the learning speed and convergence speed of the multi-source data fusion prediction model are further improved,respectively.which effectively improves the accuracy of deformation prediction results.(2)A combined prediction model of extreme Learning Machine(ELM)optimized by gravity search algorithm(GSA)and variable-dimension Fractal(VDF-Kalman)is established to realize the application of the combined prediction model.In order to reduce the limitation of prediction accuracy caused by the fluctuation of displacement sequence,wavelet decomposition is used to decompose the original displacement sequence into trend term and wave term.Trend item sequence was predicted by GSA-ELM model,and wave item sequence was predicted by VDF-Kalman model.The analysis of the measured data shows that the gravity search algorithm is used to optimize the parameters of the extreme learning machine and decompose the displacement sequence.The error caused by the non-stationary characteristics is reduced,and better prediction results can be obtained.The displacement of the predicted value of the trend item and the fluctuation item is superimposed to obtain the final prediction result of the combined model.(3)A real 3D model that can reflect the landform of the landslide is established,and the landslide stability analysis of the landslide under the action of dead weight and GNSS displacement constraint is experimentally carried out.Based on Arc GIS-Rhinoceros-Griddle,a real 3D model was established based on DEM data obtained by UAV.The internal stress and strain conditions in the process of landslide development were analyzed,the characteristics of stability changes were summarized,and the landslide stability conditions corresponding to different reduction coefficients were analyzed.The stability of loess landslide with additional displacement constraints was experimentally studied to realize the connection between the high-precision GNSS deformation information of landslide external monitoring and the internal deformation mechanism.