Research On Prediction Model Of Deformation Behavior Of Concrete Face Rockfill Dams Based On The Improved Support Vector Machine

Concrete face rockfill dam(CFRD)is widely used in modern dam construction because of its unique advantages.At present,it has become one of the main dam types for dam construction.The prediction,analysis and control of deformation are the key problems that restrict the safe construction of CFRD.The hidden dangers and unfavorable conditions of CFRD,such as insufficient crest height,cracking of concrete face slab and tensile deformation of joints,are closely related to the deformation of dam body.Effective and reasonable prediction and control of deformation is the key factor for the further development of CFRD.In this paper,threshold regression theory,improved support vector machine and ensemble learning method are used to study the deformation prediction model of CFRD.The main research contents are as follows:(1)Based on the existing measured deformation database of CFRDs,a database containing 94 typical deformation parameters and main construction information of CFRDs is established.On the basis of summarizing the typical deformation law of CFRD,an empirical prediction model between three typical deformation parameters and six main influencing factors is established by using threshold regression analysis theory,and the importance of each influencing factor is evaluated.The empirical relationship between the typical deformation parameters of CFRD and its influencing factors is quantitatively studied from the perspective of mathematical statistics.(2)In view of the shortcomings of traditional empirical prediction method,support vector machine(SVM)is used to predict the deformation of face slab dam.An improved support vector machine model for predicting the deformation of CFRD is established by introducing the mixed weight coefficient to construct the adaptive mixed kernel function and adopting the particle swarm optimization algorithm to optimize the main parameters of the model.In order to further improve the accuracy and robustness of the model,the multiple threshold regression theory is used to cluster the case database,and a single improved support vector machine prediction model is established in different dam height intervals.From the perspective of machine learning,the nonlinear relationship between deformation parameters and their key influencing factors is mined.(3)In order to further improve the prediction accuracy and robustness of the single improved support vector machine prediction model,this paper introduces the idea of integrated learning,combines the integrated learning mode with the improved support vector machine with strong learning ability,and establishes the integrated prediction model of Improved Support Vector Adaptive Lifting(improved SVR AdaBoost)to predict the typical deformation parameters of CFRD.The model combines the characteristics of single improved support vector machine and AdaBoost ensemble learning,and gives different weights to historical data samples to highlight the differences of different instance samples in the model.At the same time,multiple improved support vector machine prediction models obtained by iterative optimization are weighted and accumulated to obtain the final prediction value,which improves the shortcomings of single prediction model.The experimental results show that the improved SVR-AdaBoost integrated prediction model can effectively improve the prediction accuracy,and compared with the single improved support vector machine prediction model,the results have been greatly improved.Finally,in order to further verify the deformation prediction models of CFRD constructed in this paper,taking Gongboxia CFRD as an example,the threshold regression empirical prediction model,single improved support vector machine prediction model and improved support vector adaptive lifting integrated prediction model of CFRD are verified,and good results are achieved,It can provide reference for the prediction of typical deformation parameters of CFRD to be built or without measured data.