Research On Dam Deformation Prediction Based On ResNet

Aiming at the problems of low prediction accuracy,poor robustness,and low efficiency in traditional dam deformation prediction models,this paper proposes to introduce ResNet into the field of dam deformation prediction to achieve efficient and accurate prediction of dam deformation.Firstly,the deformation monitoring data of the Shangyou River dam are analyzed.In this paper,statistical characteristic analysis and correlation analysis were conducted on the monitoring data of the dam foundation and crest of the Shangyou River Dam,and then the correlation analysis was conducted on the settlement monitoring measurements and water level and temperature monitoring data at each monitoring point of the dam.The analysis found that there are significant differences between the magnitude of settlement deformation at various parts of the dam,as well as significant differences in the correlation between settlement deformation and temperature at different parts of the dam,and there is also significant multicollinearity at some monitoring points,which is an important reason for the low fitting accuracy,poor feature extraction ability,and insufficient robustness of traditional dam deformation prediction models.Then,a dam deformation prediction model based on ResNet is constructed.Firstly,in order to solve the problem of poor feature extraction ability in traditional models,we proposed to encode the deformation impact factors into two-dimensional tensors,and use convolution and pooling operations to extract the features in the deformation impact factors.We designed an input encoder for this model.Secondly,aiming at the problem of low fitting accuracy of traditional models,a multi-layer residual block design is proposed to increase the fitting ability of the model,and a residual encoder for this model is designed.Thirdly,aiming at the problem of repetitive modeling and low efficiency in traditional models,a multi output fully connected layer is proposed to improve the efficiency of model training and prediction.The fully connected layer dense block of this model is designed.Finally,this paper establishes a dam foundation settlement and deformation prediction model based on ResNet(SD-V ResNet)and a dam crest settlement and deformation prediction model based on ResNet(Top-V ResNet).Finally,SD-V ResNet and Top-V ResNet are applied in engineering.SD-V ResNet converged steadily after five training cycles.The converged model predicted 767 dam foundation settlement monitoring data.The absolute value of prediction error for each period was less than 1mm,with an average prediction error of 0.39 mm,0.09 mm lower than the comparison model SD-V Le Net,and 0.59 mm lower than the SD-V MLP.It is shown that the SD-V ResNet established in this paper can accurately predict the settlement and deformation of dam foundation,and can provide a reference for the engineering application of dam deformation prediction.After convergence,Top-V ResNet predicts the settlement monitoring data of 28 monitoring points on the dam crest.The absolute value of prediction error is generally lower than 1mm,with an average prediction error of 0.53 mm,0.55 mm lower than Top-V Le Net,and 0.57 mm lower than Top-V MLP.This indicates that Top-V ResNet can efficiently and accurately predict dam crest settlement and deformation,providing a reference for dam deformation prediction engineering applications.