Investigation On Tunnel Deformation Monitoring Methods Based On The EOF And Neural Network

With the rapid development of tunnel technology and operational requirements, the speed of tunnel construction is accelerated increasingly. Higher requirements are requested to correspond to the construction scale, moulding and difficulty of tunnel. The safe operation of tunnels will influence the national economy immediately. Therefore, it is particular important to monitor deformation of large tunnels and to analyze the monitoring data. In this thesis, a new model is proposed to analyze the tunnel subsidence and global deformation.First, significance of deformation monitoring, current situation of monitoring techniques at domestic and overseas and analysis methods commonly used in deformation monitoring are introduced and summarized. According to the characteristics of tunnel data, a number of suitable analytical methods are summarized. Based on the advantages and disadvantages of data processing methods, a neural network algorithm which is suitable for trend analysis of tunnel data is selected to simulate the future deformation trends.Second, the basic principle of empirical orthogonal function (EOF) and its application in the tunnel’s spatial-temporal deformation are systematically elaborated. Compared with other methods, EOF decomposition method has its own specialty in the spatial distribution and time variation characteristics of the deformation field in a sensitive way. The deformation field can be decomposed to two independent vectors which are spatial response vectors and time functions (principal components). Spatial vectors are independent with time, and are determined by the main features of deformation field. Because some leading principal components can present the main characteristics of the deformation field sufficiently, only they should be took in account. The processing theory, method and modeling of a tunnel monitoring data using EOF are elaborated, and a trend forecasting method combining empirical orthogonal function (EOF) and neural network is proposed.Then, the instance data of Yanshuigou tunnel in West-East Gas Pipeline are used to monitoring analysis and early-warning project. Tolerance analysis of deformation monitoring data is conducted, and the spatial-temporal patterns of the tunnel deformation are obtained by using EOF. The actuary of time interpolation, spatial interpolation and spatial extrapolation are tested, and time extrapolation and trend prediction of deformation pattern of tunnels are performed by using neural network algorithm.Finally, the method combining EOF and neural network is used to analyze the deformation trend of the tunnel. EOF is applied to obtain the space distribution and time variation characteristics of Yanshuigou tunnel, while the neural network method is used to predict the future time trend function and restore deformation tendency of the tunnel.Through the study of the application of the empirical orthogonal decomposition and BP neural network method in deformation analysis of tunnels, preliminary conclusions are obtained as follows:1. EOF can be well applied to the monitoring data processing of tunnels. The temporal deformation characteristics and spatial distribution characteristics of tunnels can be obtained by analyzing cumulative displacement and time-state function of Yanshuigou tunnel; this method has a preferable practical and guiding significance in data analysis of tunnel.2. The neural network prediction model of tunnel is developed by the means of BP neural network model and combined with the monitoring data of Yanshuigou Tunnel, and conducted time interpolating, spatial interpolation and spatial extrapolation experiments on multiple sets of data. By comparing actual value and predicted value, it is proved that this method can offer a feasible solution on the problems of irregular monitoring period and inaccurate monitoring on some points. The results of tunnel deformation trend analysis show-that the tunnel deformation tends are to be stable and deformation anomalies will not arise.3. By integrating the ideologies of neural network and EOF, a trend predicting model on the basis of EOF and neural network are established. The analysis of the actual data of Yanshuigou tunnel proves the method’s feasibility. The results show that the deformation trend of tunnel can be well predicted based on the combined model.4. Research results indicate that the deformation characteristics of the Yanshuigou tunnel are opposite before and after the earthquake. Before the earthquake, the tunnel shows a trend of gradual uplift from the import to the export in the vertical direction; after the earthquake, the deformation is mainly subsiding and the subsiding trend increases from inlet to outlet. In the vertical direction towards the tunnel, the deformation turns to left side from the outlet to the600m inside the tunnel while the rest part of the tunnel shows deformation to the right side, but it. shows opposite situation with that after the earthquake. Along the direction of the tunnel, the deformation situation shows that the both ends of the tunnel extrude to the middle part, while is in the contrary with the situation after earthquake.