Study On Slope Displacement Prediction Based On Nuclear Extreme Learning Machine

In recent years,slope displacement and deformation accidents have increased sharply,which has caused a great impact on people’s life and property safety.It is particularly important to monitor slope deformation and predict its future deformation trend.However,the slope displacement data curve is highly nonlinear and complex,and the traditional prediction methods have many shortcomings.Therefore,the machine learning algorithm with strong nonlinear learning ability is introduced into the slope deformation prediction.Among many machine learning algorithms,the kernel extreme learning machine has been widely used in slope deformation prediction with its advantages of fast running speed and strong generalization ability,and has achieved good results.Therefore,this paper mainly uses KELM algorithm to predict the slope displacement sequence.The main research contents include:(1)First analysis summarizes the knowledge of the deformation of the slope,the slope monitoring system is introduced in this paper,on the basis of selecting the June2019 to July 2020,the actual slope displacement data as the specific research object,the anomalous point correction data collected,missing data supplement,such as pretreatment,build slope displacement prediction model to provide data support for the follow-up.(2)In view of the chaotic characteristics of slope displacement sequence,the phase space reconstruction method is used to reconstruct the slope displacement sequence,so that it can be established and predicted in the phase space.Because the embedded dimension in phase space reconstruction and the penalty function and kernel function of the kernel extreme learning machine are difficult to be selected,the bird flock algorithm is adopted to optimize the parameters,and finally the BSA-KELM prediction model is established.(3)In fact,the original slope displacement data is a non-stationary and nonlinear signal,and the mutagenicity and randomness of the original data will reduce the accuracy of the prediction.The original displacement sequence was decomposed into a series of sub-components by the variational mode decomposition technique,and the BSA-KELM prediction model was built for each sub-component.This not only improves the smoothness of the data but also extracts the detailed features of the data.Combined with the actual data,it is proved that this method can improve the prediction accuracy.(4)Aiming at the lack of information provided by the existing single-step prediction models,two prediction models,recursive multi-step and multi-output multistep,were established.The results show that the accuracy of the recursive multi-step model is better than that of the multi-output model when the training data samples are small,and the accuracy of the multi-output model is higher when the training data samples are sufficient.