Landslide Displacement Prediction Based On Optimized Extreme Learning Machine

The prediction of landslide is essentially based on the prediction of landslide deformation,because the deformation measurement has the advantages of high precision and convenient operation,which can directly reflect the state of landslide.At present,the influence of high frequency and low frequency factors in the induced factors on the prediction results of the prediction model has not been considered in the prediction process of the landslide displacement series,and the further optimization of the induced factors of the landslide has not been considered.At the same time,the prediction accuracy and stability of the current landslide prediction model have not achieved satisfactory results.In view of the above problems,this paper will improve the landslide prediction model and optimize the induced factors.The main contents of this paper are as follows:(1)A landslide prediction model based on Grey-ELM was constructed.In this paper,the gray relational degree analysis method is introduced to analyze the gray relational degree of the output of the hidden layer node and the expected output of the output layer node.In order to improve the quality of model input,a combination model of average influence value and extreme learning machine(MIV-ELM)was proposed,through which the relationship between the displacement sequence of base station and the change of landslide inducers was analyzed,and the inducers selected by experience were further screened.The experimental results show that the root-mean-square error of the model is 13.53 mm,compared with the E-ELM model,the error is reduced by 7.8%,and the goodness of fit reaches 0.997.(2)A neural network displacement prediction model based on genetic algorithm and integrated learning is constructed.The trend term and period term of base station displacement were extracted by the high-pass Filter(HP Filter)method,and the two components were predicted respectively.Because of the periodicity of the induced factors,the change of the periodic term in the base station sequence is regarded as the result of the influence of the inducing factors in this paper.In order to improve the correlation between the induced factors and the displacement sequence,the high frequency and low frequency components of the induced factors sequence are obtained by using the complete empirical mode decomposition method.The gray correlation degree method is used to analyze the correlation degree between the decomposed induced factors sequence component and the periodic term of the base station displacement,and to find the component combination with the largest correlation degree with the periodic term in the displacement sequence,Then the induced factors were reconstructed.In this paper,genetic algorithm is used to initialize the weight of elm,and a GA-ELM model is constructed.Then on the basis of GA-ELM model,the GA-ELMs model is constructed by using the integrated learning method.Finally,this paper integrates the optimization process of landslide induced factors and the integrated learning prediction model,and integrates a complete set of landslide displacement sequence prediction model with high stability and good prediction effect.The experimental results show that the root mean square errors of the model are 12.68 mm and 8.46 mm respectively in the prediction of XD-01 and ZG118 base station displacement series.Compared with the limit learning machine model and GA(cpg)-SVM model,the errors are reduced by 57.7% and86.1% respectively,and the goodness of fit is 0.999.