Research On Dam Deformation Prediction Based On Deep Learning

With the development of new technologies such as GNSS,dam deformation monitoring has achieved real-time dynamics around the clock,and traditional dam deformation prediction methods cannot handle massive amounts of monitoring data well.Deep learning is an optimization algorithm that deepens the number of network layers on the basis of neural networks.It has been widely used in wind power,air quality,disaster warning and other fields.Compared with traditional machine learning algorithms,deep learning emphasizes learning from massive data,which can solve the problems that traditional machine learning algorithms are difficult to deal with,such as highdimensional,redundant,and high-noise dam deformation data.Therefore,it is very necessary to carry out research on dam deformation prediction based on deep learning.In view of the fact that traditional signal extraction methods are prone to end effect and modal aliasing when extracting dam deformation data features,this paper first proposes a variational modal decomposition(VMD)method for dam deformation feature extraction,and the dam deformation sequence The decomposition method is transformed into a variational problem,and the global optimal solution is obtained by calculating the constrained variational problem.The results prove that the VMD method retains the local features of the dam deformation sequence,removes highfrequency noise,and can perform better dam feature extraction.Secondly,according to the development process of deep learning,this article analyzes the nonlinear autoregressive neural network(NARX),the initial deep belief network(DBN)of the early recurrent neural network,and the long and short-term memory network for current time series prediction(LSTM),combined with variational modal decomposition,respectively constructed dam deformation prediction models based on VMD-NARX,VMD-DBN,and VMD-LSTM.Then,the constructed dam deformation prediction model based on deep learning is applied in practice.The experimental results show that the three models of VMD-NARX,VMD-DBN,and VMD-LSTM are combined with the evaluation indicators,and the mean absolute error(MAE)is1.28 mm,1.15 mm,0.95 mm,respectively;the root mean square error(RMSE)is respectively They are 1.62 mm,1.49 mm and 1.22 mm respectively.At the same time,other traditional dam deformation prediction models are applied and compared.The comparison results show that the deep learning models are all excellent,with small error indicators,showing high accuracy and good stability,thereby providing safety monitoring for dams.The basis for reference.