Structural Behavior Analysis And Optimization Of High-Core Rockfill Dam Considering Compaction Characteristics

Structural behavior of high-core rockfill dam is directly related to dam construction and service safety.The physical,mechanical and permeable compaction characteristics of dam materials under the influence of compaction quality are the key factors affecting the dam structural behavior.Therefore,it is of great significance to predict the compaction characteristics of dam materials and carry out numerical analysis of the dam structural behavior,as well as to build a surrogate model and optimize dam structural behavior for guiding the quality control during construction process and ensuring dam safety.However,there are still the following problems:（1）The current studies on compaction characteristics prediction of dam materials mainly focuse on the prediction of single parameter such as dam material physics,mechanics and permeability,which have neglected the correlation between these parameters,resulting in insufficient model performance.（2）The Biot consolidation model,used in the traditional analysis of the structural behavior of earth-rock dams,does not consider the effect of compaction quality on the spatial heterogeneity of the compaction characteristics,and ignores the seepage-stress bidirectional coupling relationship.Moreover,the parameter assignment method using the“Dam Grid”fails to directly construct the mapping relationship between the compaction parameters and finite element model parameters,which limits the assignment accuracy and efficiency.（3）The commonly used single-fidelity surrogate models of dam structural behavior do not consider the influence of compaction characteristics on the dam structural behavior and the spatiotemporal attribute of dam structural behavior.Balancing the accuracy and efficiency is an additional challenge.（4）The multi-objective optimization on dam structural state considering compaction parameters has not been carried out.In view of the above problems,this paper conducts a research on the structural behavior analysis and optimization of high-core rockfill dams considering the compaction characteristics,and the main research results are as follows:（1）Proposing the improved multi-output prediction model of compaction characteristics considering the correlation of compaction characteristics parametersFirst,to construct the original input and output space,the compaction parameters and the physical,mechanical and permeable compaction characteristic parameters are temporally and spatially registered according to the spatiotemporal correlations.Then,the prediction model of compaction characteristic with IMO-GPR（Improved Multi-output Gaussian Process Regression）method considering the correlation is proposed,in which DBSCAN（Density-Based Spatial Clustering of Applications with Noise）algorithm is used to extend the original input space to improve the insufficient prediction performance of the original same input space.Besides,the correlation of compaction characteristic parameters,which can’t be considered in the single-output prediction model,is fully considered through the covariance coefficient matrix in the model.The engineering example shows that the a RRMSE（Average Relative Root Mean Squared Error）of the IMO-GPR model is respectively improved by 24%,17%and 20%compared with Gaussian process regression,multi-output Gaussian process regression and multi-output extreme learning machine.Besides,it has stronger robustness to noise interference,abnormal data and small amount of data.（2）Proposing the nonlinear poroelastic seepage-stress coupling analysis method for the dam structural behavior considering the influence of compaction characteristicsFirst,based on the framework of Biot consolidation theory,combining the IMO-GPR compaction characteristics prediction model for considering the spatial variation of compaction characteristics and the Kozeny-Carman model for considering the dynamic evolution law of permeability coefficient,the nonlinear poroelastic seepage-stress coupling model for the dam structural behavior considering the compaction characteristics is deduced.Second,based on the constrained random field method,the mapping relationship between the warehouse mesh and the finite element mesh is directly constructed,so as to simplify the assignment process and improve the assignment accuracy.Finally,through the further development of COMSOL Multiphysics,the strong coupling calculation of dam structural behavior is carried out.The engineering example shows that compared with the situation of neglecting the compaction characteristics and the change of permeability coefficient,the R² between the calculated values and the monitored values of core wall settlement is increased from0.724 and 0.658 to 0.946,respectively.Compared with the"Dam Grid"parameter assignment method,the R² between the calculated values and the monitored values of the total water head in the core wall is increased from 0.888 to 0.908.The distribution of stress and permeability coefficients in the core wall of the proposed method are more in line with the actual engineering situation.（3）Proposing the global multi-fidelity surrogate model with deep spatiotemporal basis of high-core rockfill dam considering the influence of compaction characteristicsFirst,taking advantage of the strong ability of deep learning to extract features from high-dimensional spatiotemporal data,the multi-fidelity VAE-DCGAN（hybrid deep convolutional generative adversarial network with variational auto-encoder）global surrogate model of dam structural behavior with deep spatiotemporal basis is proposed.First,the structure of Encoder（Generator）in Hybrid VAE-GAN（Advanced Hybrid Deep Adversarial Autoencoder）is improved by CAE-CGAN（Conditional Autoencoder-Conditional Generative Adversarial Networks）with learning method of arbitrary latent spatial distribution,which is to solve the difficulty of learning the latent spatial distribution of compaction parameters caused by the shared network structure of VAE（Variational Autoencoder）and GAN（Generative Adversarial Networks）in Hybrid VAE-GAN.Thus,the single-fidelity global VAE-DCGAN surrogate model with deep spatiotemporal basis is established,and the supervised learning relationship between compaction parameters and global dam structural spatiotemporal behavior is built.Second,based on Copula theory,the correlation sampling of compaction parameters is carried out,and the bridge function method is used to construct the multi-fidelity surrogate model with deep spatiotemporal basis integrating high-and low-fidelity models,so as to overcome the problem that the single-fidelity surrogate model is difficult to give consideration to both calculation accuracy and efficiency.The engineering example shows that the mean correlation coefficient and mean root mean square error of the VAE-DCGAN surrogate model reach 0.991 and 1.3×10^-4 m respectively,which have met the accuracy requirements.Under the same model accuracy,the computational efficiency of multi-fidelity surrogate model is 49%higher than that of single-/high-fidelity surrogate model.（4）Proposing the multi-objective optimization method of dam structural behavior coupling multi-fidelity surrogate model under the influence of compaction characteristicsBased on the step-by-step optimization strategy and HMOHHO-SCA（Hybrid Multi-Objective Harris Hawks Optimizer with Sine Cosine Algorithm）,the multi-objective optimization method of dam structural behavior coupling with multi-fidelity surrogate model is proposed.First,the multi-objective optimization mathematical model for dam structural behavior considering the compaction characteristics is established.Besides,the Morris global sensitivity analysis method is used to optimize the decision variables,and then the step-by-step optimization strategy is used to reduce the solution set space to make up for the low efficiency of the overall optimization strategy in the multi-parameter optimization process.Second,the global exploration stage is improved by SCA（Sine Cosine Algorithm）to improve the insufficient search ability caused by the highly random search agency.Moreover,a new nonlinear dynamic update method of prey escape energy is proposed to balance the global exploration and local exploitation.The multi-fidelity surrogate model is used in sensitivity and optimization analysis to quickly obtain structural behavior objectives and to reduce computational resource consumption.The engineering example shows that the sensitivity rankings of compaction parameters in each zone are obtained,and the recommended values of compaction parameters in each zone are put forward.The analysis efficiency based on the multi-fidelity surrogate model is improved by nearly70 times.Compared with the overall optimization strategy,the efficiency of step-by-step optimization strategy is improved by 39.4%.Moreover,the solution sets of proposed HMOHHO-SCA have obvious performance advantages.