A tunnel deformation monitoring and prediction system

Abstract

The present application belongs to the technical field of tunnel engineering, and particularly relates to a new type of tunnel deformation monitoring and prediction system. The system comprises a numerical simulation tunnel module, a tunnel three-dimensional real scene modeling module, an LSTM prediction module and an actual tunnel deformation monitoring module. The method comprises the following steps: step 1: obtaining the tunnel deformation in the model through the numerical simulation tunnel module; step 2: generating a tunnel three-dimensional real scene model by using the tunnel three-dimensional real scene modeling module; step 3: monitoring and collecting the actual deformation data of the tunnel monitoring section by using the actual tunnel deformation monitoring module, and performing pretreatment; step 4: establishing a prediction model based on a long short-term memory (LSTM) neural network; and step 5: inputting the measured data into the prediction model to generate predicted deformation, then performing fitting analysis, and formulating an emergency plan and countermeasures. The method of the present application can prevent data time lag by updating data in real time, and can reduce the collapse accident and save unnecessary losses.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of tunnel engineering, and particularly relates to a novel tunnel deformation monitoring and predicting system. BACKGROUND

[0002] With the continuous updating of tunnel engineering technology in China, tunnel collapse can be predicted to prevent major safety accidents, but tunnel collapse still affects the progress of tunnel construction and more resources are invested in tunnel reinforcement after collapse.

[0003] The existing tunnel deformation detection system includes a monitoring process using a soil pressure cell, a prism-total station instrument and the like. Generally, manual operation is adopted, and the accuracy of the data cannot be guaranteed. Better ones will use automatic collection, that is, the collection device is placed at the monitoring section, and then the data is transmitted through the network by automatic machine collection. The safety of the tunnel only plays a monitoring role, and does not predict the next deformation of the tunnel. Prediction can timely discover potential dangers, increase support operation space, have more time for judgment and processing, and is more beneficial to personal safety and property safety. If no prediction is made, the reaction time may be insufficient when potential dangers are discovered. SUMMARY

[0004] In order to comprehensively solve the above problems, the present application proposes a novel tunnel deformation monitoring and predicting system to deduce the tunnel deformation trend from the actual tunnel deformation amount, predict the tunnel deformation amount, and achieve prediction before the tunnel appears a deformation amount mutation trend, timely take measures, develop a turning measure, and prevent economic losses caused by tunnel collapse.

[0005] In order to achieve the above purpose, the technical solution adopted by the present application is as follows:

[0006] A novel tunnel deformation monitoring and predicting system, comprising a numerical simulation tunnel module, a tunnel three-dimensional real scene modeling module, an LSTM predicting module, and an actual tunnel deformation amount monitoring module.

[0007] The numerical simulation tunnel module: a finite element numerical simulation software is used to establish a numerical simulation tunnel model according to the tunnel working condition and the tunnel design scheme, and the numerical simulation tunnel deformation amount and stress condition are obtained.

[0008] The tunnel three-dimensional real scene modeling module: the actual tunnel deformation amount monitoring arrangement points are arranged according to the tunnel excavation, and the tunnel three-dimensional real scene modeling is created and updated.

[0009] The LSTM predicting module: a time series predicting model is created based on the tunnel three-dimensional real scene modeling.

[0010] The actual tunnel deformation amount monitoring module: a total station instrument is used for manual fixed-point monitoring.

[0011] A method for using a new tunnel deformation monitoring and prediction system, comprising:

[0012] Step 1: Before tunnel excavation, based on the geological exploration report and the tunnel design scheme, a numerical simulation tunnel model is created through the finite element numerical simulation software of the tunnel module, and the tunnel deformation in the model is obtained;

[0013] Step 2: A tunnel three-dimensional real scene modeling module is used to generate a tunnel three-dimensional real scene model;

[0014] Step 3: The tunnel three-dimensional real scene model established in step 2 is segmented according to the tunneling direction, a monitoring section is selected, an actual tunnel deformation monitoring module is used to monitor and collect actual deformation data of the tunnel monitoring section, and the obtained actual tunnel deformation data is preprocessed;

[0015] Step 4: A surface subsidence prediction model, a vault subsidence prediction model and a horizontal convergence prediction model based on a long short-term memory (LSTM) neural network are established, and the preprocessed data is input into the model for training;

[0016] Step 5: The measured data collected in step 3 is input into the prediction model created in step 4, the measured deformation is input, the predicted deformation is generated, and the fitting analysis is performed with the deformation at the corresponding position in the numerical simulation tunnel deformation in step 1, and an emergency plan and countermeasures are formulated.

[0017] Preferably, step 2 comprises:

[0018] Step 2.1: A tunnel three-dimensional real scene modeling coordinate origin is arranged in the tunnel open cut section, a laser tractor is arranged to shoot inside the tunnel, a UAV with high-precision 360° photographing function is controlled to fly in a U-shaped route at double heights inside the tunnel and take photos for scanning, and tunnel photos containing position coordinates are obtained;

[0019] Step 2.2: The photos obtained in step 2.1 are processed by using the SIFT method.

[0020] Preferably, step 4 is specifically:

[0021] Step 4.1: The hyperparameters of the surface subsidence prediction model are defined;

[0022] Step 4.2: The actual tunnel deformation data preprocessed in step 3 is input into the surface subsidence prediction model;

[0023] Step 4.3: The same method as step 4.2 is used to establish the vault subsidence prediction model and the horizontal convergence prediction model and to train them.

[0024] A computer device comprises a processor and a memory, the memory stores a computer program, the computer program is loaded and executed by the processor to realize the use method of the novel tunnel deformation monitoring and prediction system.

[0025] A computer readable storage medium, the storage medium stores a computer program, the computer program is loaded and executed by the processor to realize the use method of the novel tunnel deformation monitoring and prediction system.

[0026] Compared with the prior art, the beneficial effects of the present application are:

[0027] 1. Based on the traditional monitoring system, the present application adds a prediction system, which can verify the accuracy of numerical simulation, correct numerical simulation data in time and discuss and evaluate the tunnel situation when a warning occurs, prevent data from lagging behind through real-time data updating, and reduce the collapse accident and unnecessary loss.

[0028] 2. The three-dimensional real scene modeling is introduced into the tunnel detection and prediction system, which can record the actual deformation of the tunnel.

[0029] 3. The combination of numerical simulation, three-dimensional real scene modeling and time series prediction can verify the accuracy of advanced geological prediction and the feasibility of surrounding rock support design, and the contents of the three can be corrected in time according to the actual surrounding rock condition, forming a closed loop to ensure the timeliness and accuracy of data. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, used to explain the present application, and do not constitute a limitation of the present application.

[0031] In the drawings:

[0032] Fig. 1 The method flowchart of the present application;

[0033] Fig. 2 The structure diagram of the long short-term memory (LSTM) neural network. DETAILED DESCRIPTION

[0034] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings Figs. 1-2 The preferred embodiments described herein are used to illustrate and explain the present application, and do not limit the present application.

[0035] A novel tunnel deformation monitoring and prediction system comprises a numerical simulation tunnel module, a tunnel three-dimensional real scene modeling module, an LSTM prediction module and an actual tunnel deformation monitoring module.

[0036] Numerical simulation tunnel module: a finite element numerical simulation software is used to establish a numerical simulation tunnel model according to the tunnel working condition and the tunnel design scheme, and the numerical simulation tunnel deformation and stress condition are obtained;

[0037] Tunnel three-dimensional real scene modeling module: according to the tunnel excavation arrangement actual tunnel deformation monitoring arrangement point, the tunnel three-dimensional real scene modeling is created and updated;

[0038] LSTM prediction module: a time series prediction model is created based on the tunnel three-dimensional real scene modeling;

[0039] Actual tunnel deformation monitoring module: a total station is used for artificial fixed-point monitoring.

[0040] A use method of a novel tunnel deformation monitoring and prediction system, comprising:

[0041] Step 1: before tunnel excavation, a numerical simulation tunnel model is created by a finite element numerical simulation software of a numerical simulation tunnel module based on a geological exploration report and a tunnel design scheme, and the tunnel deformation in the model is obtained;

[0042] Step 2: tunnel photos containing position coordinates are collected by a tunnel three-dimensional real scene modeling module, and the tunnel photos are processed.

[0043] Step 2.1: a tunnel three-dimensional real scene modeling coordinate origin is arranged in a tunnel open cut section, a laser tractor is arranged to shoot inside the tunnel, a UAV with high-precision 360° photographing function is controlled to fly in a U-shaped route at double heights inside the tunnel and take photos for scanning, and tunnel photos containing position coordinates are obtained;

[0044] Step 2.2: the photos obtained in step 2.1 are processed by using SIFT technology to obtain a tunnel three-dimensional real scene model. Specifically, it includes:

[0045] Step 2.2.1: the tunnel photos in step 2.1 are processed by the methods of image pyramid component, spatial extreme point detection, key point positioning and feature vector matching to extract feature points;

[0046] Step 2.2.2: cloud sparse reconstruction, the tunnel picture feature points extracted in step 2.2.1 are matched and restored to a three-dimensional space created by picture coordinates by a bundle adjustment method to form a tunnel three-dimensional model;

[0047] Step 2.2.3: point cloud dense reconstruction, the tunnel three-dimensional model in step 2.2.2 is processed by the methods of initial feature matching, face sheet generation and face sheet filtering to generate a tunnel three-dimensional point cloud model, and finally a tunnel three-dimensional real scene model is generated;

[0048] Step 3: The three-dimensional real scene model of step 2 is segmented according to the tunneling direction, the monitoring section is selected, the actual tunnel deformation data of the monitoring section is monitored and collected by using the actual tunnel deformation monitoring module, and the obtained tunnel actual deformation data is preprocessed;

[0049] Specifically, the tunnel actual deformation data measured by the total station is collected manually

[0050] The tunnel three-dimensional real scene model is segmented according to the tunneling direction 5m, the monitoring section is selected (the monitoring section can be selected according to actual requirements), and the crown subsidence, horizontal convergence and ground subsidence of each monitoring section are taken as monitoring values.

[0051] The crown subsidence is represented as n represents the tunneling to the nth segment, is the crown subsidence monitoring value of the jth monitoring section when the tunneling is to the ith segment;

[0052] The horizontal convergence is represented as n represents the tunneling to the nth segment, is the horizontal convergence monitoring value of the jth monitoring section when the tunneling is to the ith segment;

[0053] The ground subsidence is represented as n represents the tunneling to the nth segment, is the ground subsidence monitoring value of the jth monitoring section when the tunneling is to the ith segment.

[0054] The missing values in the actual deformation data set are inserted by using the cubic spline interpolation method to avoid the influence of missing data on model training.

[0055] The ground subsidence data of all monitoring sections after preprocessing (J represents the data collected by the Jth monitoring section, and m represents the number of monitoring sections), as the ground subsidence prediction model input information, the ground subsidence prediction model of step 4 is trained;

[0056] Step 4: Establish a ground subsidence prediction model, a crown subsidence prediction model and a horizontal convergence prediction model based on a long short-term memory (LSTM) neural network, and input the preprocessed data into the model for training. Specifically:

[0057] Considering that the crown subsidence, horizontal convergence and ground subsidence are three monitoring items, in order to improve the prediction accuracy, three prediction models should be established for the crown subsidence, horizontal convergence and ground subsidence respectively, and the three prediction models are the same, therefore, the establishment of the ground subsidence prediction model based on the long short-term memory (LSTM) neural network is mainly introduced, and the establishment methods of the crown subsidence and horizontal convergence prediction models are the same as those of the ground subsidence prediction model.

[0058] As Fig. 2 , a land subsidence prediction model based on long short-term memory (LSTM) neural network is established.

[0059] Step 4.1: define the hyperparameters of the land subsidence prediction model;

[0060] The size of the input land subsidence data is set to 1, and the time length is set to 6, representing the use of 5 values each time As input, the output prediction value O5 is output, and The loss function MSE is calculated, and then the weights w and the bias b are automatically adjusted (neural network back propagation). The number of layers of the hidden layer is set to 2, the number of units of the hidden layer is set to 50, the learning rate is set to 0.01, the initial hidden state H0=0, and the initial C0=0.

[0061] Among them, MSE is selected as the loss function;

[0062]

[0063] Among them, O t is the prediction value of the tth segment, M represents that there are M O t , that is, M times of prediction during training, is the measured value of the t+1th segment, and the calculation of the loss function can reflect the training effect of the model. The smaller the loss, the more accurate the prediction result;

[0064] Adam is selected as the optimizer, and after bias correction, the learning rate of each iteration has a certain range, which accelerates the training of the land subsidence prediction model, and the learning rate update is as follows:

[0065]

[0066] Among them: θ represents the updated learning rate, m i and v i respectively represent the first-order moment estimate and the second-order moment estimate of the i-th iteration parameter; ε is a very small constant used to prevent the denominator from being 0; α represents the step factor of weight update;

[0067] Among them, w and b are updated as follows:

[0068]

[0069]

[0070] Step 4.2: input the land subsidence data in the following formula in order Calculate H t and O t ;

[0071]

[0072]

[0073] wherein, is the input ground subsidence data, t represents the tth monitoring value of the monitoring index; H t-1 is the hidden state, I t is the importance of the control input ground subsidence data, F t controls the information retained in the neuron state at the previous moment; w xf , w ht , w xf , w hf are weight coefficients of H H t-1 , H t-1 , b i and b f are bias coefficients;

[0074] All the weights w and the bias b are parameters of the ground subsidence prediction model, which are automatically updated, and the whole model is used to find suitable w and b to minimize the value of the loss function MSE, and the size of each parameter update is determined according to the learning rate.

[0075] O t = σ (x t w xo + H t-1 w ho + b o )

[0076] wherein, O t is the output result of the ground subsidence prediction model at t time, σ is a sigmoid activation function; w xo , w ho are weight coefficients of H H t-1 , and b o is a bias coefficient;

[0077]

[0078]

[0079] is the unit state of the current input gate at t time, C t-1 is the unit state at t-1 time, C t is the updated neuron state information;

[0080] H t = O t ⊙ tanh (C t-1 )

[0081] H t It is calculated using the neuron's memory information and output gate from the previous time step, and includes information from the current time step. tanh is the hyperbolic tangent activation function.

[0082] Step 4.3: Using the same method as in Step 4.2 for establishing the surface subsidence prediction model, establish the crown subsidence prediction model and the horizontal convergence prediction model and train them.

[0083] Step 5: Input the measured data collected in Step 3 into the prediction model trained in Step 4, input the measured deformation, and generate the predicted deformation. For example, using the surface settlement data of the [0, n] segment, calculate O. n+1 O n+2 O n+3 O n+4 O n+5 This refers to the predicted surface settlement data for the [n, n+5] segment, which is then fitted and analyzed with the corresponding deformation data in the numerical simulation of tunnel deformation in step 1. If the difference is less than ±5mm, normal construction can proceed. If it is greater than ±5mm, a level II (yellow) early warning will be activated to strengthen monitoring, closely observe the development, analyze the causes, stabilize the tunnel deformation, and formulate emergency plans and countermeasures.

[0084] Table 1 Management Levels and Corresponding Measures

[0085]

[0086] Table 2 Management Level of Total Deformation

[0087]

[0088] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for using a tunnel deformation monitoring and prediction system, characterized in that: The tunnel deformation monitoring and prediction system includes a numerical simulation tunnel module, a tunnel 3D real-scene modeling module, an LSTM prediction module, and an actual tunnel deformation monitoring module. Numerical simulation tunnel module: Using finite element numerical simulation software, a numerical simulation tunnel model is established based on the tunnel area working conditions and tunnel design scheme to obtain the numerical simulation tunnel deformation and stress conditions. Tunnel 3D Reality Modeling Module: Based on the actual tunnel deformation monitoring points arranged during tunnel excavation, create and update the tunnel 3D reality model; LSTM prediction module: Creates a time-series prediction model based on 3D reality modeling of the tunnel; Actual tunnel deformation monitoring module: Manual fixed-point monitoring is performed using a total station; The usage methods of the tunnel deformation monitoring and prediction system include: Step 1: Before tunnel excavation, based on the geological exploration report and tunnel design plan, a numerical simulation tunnel model is created using the finite element numerical simulation software of the numerical simulation tunnel module to obtain the tunnel deformation in the model. Step 2: Generate a 3D reality model of the tunnel using the tunnel 3D reality modeling module; Step 3: Divide the 3D tunnel model established in Step 2 into segments according to the tunneling direction, select monitoring sections, and use the actual tunnel deformation monitoring module to monitor and collect the actual deformation data of the tunnel monitoring sections. Preprocess the acquired actual tunnel deformation data. Divide the 3D tunnel model into 5m segments according to the tunneling direction, select monitoring sections, and take the crown settlement, horizontal convergence, and surface settlement of each monitoring section as monitoring values. Step 4: Establish surface subsidence prediction models, crown subsidence prediction models, and horizontal convergence prediction models based on Long Short-Term Memory (LSTM) neural networks, and input the preprocessed data into the models for training; Step 4 specifically involves: Step 4.1: Define the hyperparameters of the surface subsidence prediction model; Step 4.2: Input the actual tunnel deformation data preprocessed in Step 3 into the surface settlement prediction model; Step 4.3: Using the same method as in Step 4.2 to establish the surface subsidence prediction model, establish the crown subsidence prediction model and the horizontal convergence prediction model and train them; Step 5: Input the measured data collected in Step 3 into the prediction model created in Step 4, input the measured deformation, generate the predicted deformation, and perform a fitting analysis with the corresponding deformation in the numerical simulation of tunnel deformation in Step 1 to formulate emergency plans and countermeasures; specifically: Input the measured data collected in step 3 into the prediction model trained in step 4, input the measured deformation, generate the predicted deformation, and use the surface settlement data of the [0, n] segment to obtain... This refers to the predicted surface settlement data for the [n, n+5] segment, which is then fitted and analyzed with the corresponding deformation data in the numerical simulation of tunnel deformation in step 1. If the difference is less than ±5mm, normal construction can proceed. If it is greater than ±5mm, a level 2 warning (yellow) is activated, requiring enhanced monitoring, close attention to the development, analysis of the causes, stabilization of tunnel deformation, and the formulation of emergency plans and countermeasures.

2. The method of using the tunnel deformation monitoring and prediction system according to claim 1, characterized in that: Step 2 includes: Step 2.1: Establish the coordinate origin of the 3D reality modeling of the tunnel in the open section, and deploy laser traction devices that direct laser beams into the tunnel interior to control high-precision 360° positioning. o The camera-equipped drone flew in a U-shape at two altitudes inside the tunnel and took photos to scan the tunnel, obtaining photos containing the location coordinates. Step 2.2: Process the data of the photos obtained in Step 2.1 using the SIFT method.

3. A computer device, characterized in that: The computer device includes a processor and a memory, the memory storing a computer program, which is loaded and executed by the processor to implement the method of using the tunnel deformation monitoring and prediction system as described in any one of claims 1-2.

4. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which is loaded and executed by a processor to implement the method of using the tunnel deformation monitoring and prediction system as described in any one of claims 1-2.

Images