Tunnel Deformation Prediction Based On Spatio-temporal Data Fusion Model

A tunnel is an important infrastructure in a city.Its structure and function are easily affected by the natural environmental factors,such as temperature,humidity,geological structure.The overall performance is constantly deteriorating.Its safety will affect the security of passing vehicles and pedestrians,and the deformation of the tunnel can partly reflect the health of its structure.Therefore,prediction of tunnel deformation can help early warn the security of tunnel.Meanwhile,a large number of sensors are deployed in the tunnel to collect and accumulate a large amount of data to provide sufficient historical data for tunnel deformation prediction.However,there are few tunnel deformation prediction models based on multi-source data in the past.Therefore,the thesis mainly studies the tunnel deformation prediction models based on fusion of spatio-temporal data.The paper is completed as follows:1.Focusing on the characteristics such as insufficient samples of tunnel deformation and related features data,long time span,multiple monitoring points,strong correlation with time and space,use double-layer sliding windows to sample historical data so as to build a sufficient sample set,and design a tunnel deformation prediction model based on graph convolution neural network(Graph Convolution Neural Network,GCNN)and convolutional gated recurrent unit(Convolution Gated Recurrent Unit,Conv GRU)to fuse spatio-temporal data.In addition,the model merges the extracted features through the Conv GRU network and the fully connected neural network.Finally,compare the model with Conv GRU network,ASTGCN,Cheb Net and other models.The experimental results show that the model based on spatio-temporal data fusion has a smaller error of prediction value and a higher degree of fit.2.Deformation and related features data of tunnel include low-frequency and highfrequency information,while the model of work 1 using GCNN to extract spatio-temporal features does not distinguish and selectively extract the two parts of information of the features.For this problem,the model of work 1 needs to be further optimized.So,construct a tunnel deformation prediction model based on Cheb Net,Graph Heat,and Conv GRU to fuse spatio-temporal data.The model mainly extracts information from the relationship between features and space,and features and time,and fuses the extracted features by weighting.At the same time,the experimental comparison with work 1 model,ASTGCN,GCNN shows that the prediction error of the optimized model is lower than that of the work1 model and other models as a whole,and the model fits data well.In summary,the proposed tunnel deformation prediction models based on spatiotemporal data fusion in the thesis can fuse multiple features,make full use of the spatiotemporal characteristics of features,solve the problem of historical data with long time span and insufficient samples,and can simultaneously predict different numbers of tunnel monitoring points.At last,the predicted values have low error,and the models have good fit.The models can provide a useful help for the tunnel deformation prediction and have good practical significance.