Deformation Prediction Model For Arch Dams Based On Monitoring Data

Deformation is a core monitoring indicator for the safety of arch dam structures.Building a high-precision deformation prediction model has significant practical significance for predicting the safe operation status of a dam.This article focuses on the deformation problem of arch dams.Based on the non-stationary and nonlinear characteristics of deformation monitoring data,an empirical mode decomposition method combined with wavelet threshold denoising model is introduced to preprocess the deformation monitoring data to obtain high-quality dam deformation monitoring data.On this basis,aiming at the defects of traditional dam deformation prediction analysis models such as high quality requirements for monitoring data,difficult parameter function settings,and easy to fall into local optima,two deformation prediction analysis models are proposed based on the advantages of artificial intelligence algorithms: adaptive position algorithm optimized twin support vector machine model(APSO-TWSVM)and sparrow search algorithm optimized long-short-term memory network model(SSALSTM).The construction theory,construction structure,and parameter selection of the two models are systematically studied,and the two models are applied to an actual engineering case to analyze the prediction accuracy and the advantages and disadvantages of the models.The main research work and achievements of this paper are as follows:(1)In order to reduce the impact of noise on the abnormal fluctuation of dam monitoring data,an empirical mode decomposition method combined with a wavelet threshold denoising model is adopted.The dam monitoring data is decomposed into IMF components from high to low through the empirical mode decomposition method and wavelet threshold denoising is applied to the high-frequency IMF components obtained from the EMD decomposition for denoising.The high-frequency components denoised by wavelet are combined with the low-frequency components obtained from the EMD decomposition for reconstruction to obtain high-quality dam monitoring data.The empirical mode decomposition method combined with wavelet threshold denoising preprocesses the dam monitoring data,removes excessive noise and errors in the deformation data,while retaining the characteristics of the original data as much as possible,laying the foundation for establishing high-precision models in the subsequent steps.(2)In order to improve the prediction accuracy of the model for complex nonlinear problems,researchers introduced the Long Short-Term Memory Network(LSTM)model and explored the intrinsic laws and representation levels of monitoring data.Compared with the Recurrent Neural Network(RNN),the LSTM model solves the problem of longterm dependence.Its unique "gate" structure can avoid gradient explosion and disappearance,and it has powerful long-term memory ability.In order to avoid the LSTM model falling into a local optimal state,researchers used the Sparrow Search Algorithm(SSA)to optimize the parameters of the LSTM model.Finally,based on SSA-LSTM,a monitoring model for the deformation of concrete dams was constructed(3)To explore the relationship between dam deformation monitoring data and various influencing factors,this paper considers the problems that traditional Support Vector Machine(SVM)is only suitable for small sample data and has a slow iteration speed.Therefore,the Twin Support Vector Machine algorithm(TWSVM)was proposed for improvement.Compared with SVM,the TWSVM model has a faster training speed,but the parameter selection is more difficult.Therefore,based on the Standard Particle Swarm Optimization algorithm(PSO),this paper introduces velocity and position factors to construct an Adaptive PSO(APSO)with self-adaptive positions,which is applied to optimize the parameters of the TWSVM model.Finally,based on APSO-TWSVM,this paper constructs a deformation prediction model for concrete dams.(4)The APSO-TWSVM model and SSA-LSTM model were applied to an arch dam project,and their advantages over traditional deformation monitoring models were compared and analyzed in terms of operational efficiency and computational accuracy,and verified.