Abstract:Deformation prediction of concrete dams is an important method for evaluating the dam operation status and predicting the dam behavior.However,the application of the conventional multiple linear regression model has been limited because of random loading and strong nonlinear deformation of concrete dams.Conversely,the artifi-cial neural network model shows good adaptability to complex and highly nonlinear behavior.A dam deformation prediction model based on the extreme learning machine with optimal parameters selected by the improved particle swarm optimization(ELM-IPSO) algorithm is proposed to solve the problems of slow convergence and overfitting of the conventional neural network model based on the gradient descent method.The conventional PSO algorithm, which easily reaches the local optimum,is improved by dynamically adjusting the adaptive inertia weight and learn-ing factor.Research shows that the IPSO algorithm improves the global search capability of particle swarm optimiza-tion and improves the computational efficiency.The IPSO is applied to optimize the initial weight and threshold of the ELM model.The prediction performance of the ELM-IPSO model was verified using the measured data of the Dongjiang concrete arch dam.The results are compared with those of backpropagation neural network(BPNN), ELM,and conventional ELM-PSO models.The values of the squared correlation coefficient R2 of the BPNN, ELM,ELM-PSO,and ELM-IPSO models are 89.15%,91.13%,93.87%,and 94.36%,respectively.The R2 value of the ELM model is larger than that of the BPNN model,indicating that the ELM model has higher prediction accuracy and better generalization performance than the BPNN model.The prediction accuracy of the ELM-PSO model is higher than that of the ELM model,which indicates that the optimization of the ELM by PSO has a good effect on improving the prediction accuracy.Among the four models,the ELM-IPSO model has the largest R2 value and highest prediction accuracy,which indicates that the proposed ELM-IPSO model can effectively improve the prediction capability of the deformation for concrete dams.

Nonlinear Agent Combined Forecasting Model for Dam Deformation

Combined forecasting model for slope stability based on support vector machines with particle swarm optimization

Modified Hybrid Forecast Model Considering Chaotic Residual Errors for Dam Deformation

Combination Forecast Model for Concrete Dam Displacement Considering Residual Correction

APPLICATION OF THE OPTIMAL NON-NEGATIVE VARIABLE WEIGHT COMBINATION MODEL FOR MONITORING DAM DEFORMATION

Optimized Prediction Model for Concrete Dam Displacement Based on Signal Residual Amendment

Dam Deformation Forecasting Using SVM-DEGWO Algorithm Based on Phase Space Reconstruction

Time series forecast model of foundation pit deformation based on DE-SVM

A Multipoint Prediction Model for Nonlinear Displacement of Concrete Dam.

Hybrid Model for Concrete Dam Deformation in Consideration of Residual Correction by Frequency Division

Optimized Combined Forecasting Model for Hybrid Signals in the Displacement Monitoring Data of Concrete Dams

A Deformation Prediction Model of High Arch Dams in the Initial Operation Period Based on PSR-SVM-IGWO

Multi-point Hybrid Model Based on PSO-SVM for Concrete Arch Dam Deformation Monitoring

Dynamic early-warning model of dam deformation based on deep learning and fusion of spatiotemporal features

Spatiotemporal Hybrid Model for Concrete Arch Dam Deformation Monitoring Considering Chaotic Effect of Residual Series

Multi-scale deformation combination forecast model for concrete dam based on BP-ARIMA

Nonlinear Combination Predicting Based on Support Vector Machines for Landslide Deformation

Deformation Prediction of Concrete Dam Based on Improved Particle Swarm Optimization Algorithm and Extreme Learning Machine

Research on Safety Monitoring Forecasting Model for Dam Based on Chaos Optimization Support Vector Machine Algorithm

LS-SVM Prediction Model Based on Phase Space Reconstruction for Dam Deformation

A Novel Deformation Forecasting Method Utilizing Comprehensive Observation Data