Method for analyzing construction time of secondary lining of tunnel

Abstract

The invention provides a tunnel secondary lining construction time analysis method which is suitable for the field of tunnel construction. The method comprises the steps of data collection and preprocessing, initial model training, model prediction effect evaluation, knowledge playback training, model application and feedback and iteration, and adopts data of previous projects to train a secondary lining construction time basic prediction model in a machine learning mode. For projects with similar project profiles, the model can be directly adopted for prediction, a model knowledge base is further optimized according to deformation and crack feedback of a later secondary lining structure, data of a new project is directly added into the knowledge base, the prediction model is optimized through a knowledge playback mechanism, the generalization ability of the model is improved, and the prediction efficiency is improved. According to the method, the model can be reused in different tunnel projects, the efficiency of secondary lining construction decision making is remarkably improved, and the method has the advantages of being high in adaptability and analysis efficiency and can be widely applied to the field of tunnel secondary lining construction.

Description

Technical Field

[0001] The invention relates to the field of tunnel secondary lining construction, and in particular to a method for analyzing the construction time of tunnel secondary lining. Background Art

[0002] In tunnel construction projects, secondary lining is an important safety reserve to ensure the structure. In order to ensure that the secondary lining does not bear large loads other than its own gravity during construction and to leave enough safety margin for the structure, the construction personnel will ensure that the tunnel has been deformed for a long enough time before the secondary lining construction. Therefore, the reasonable determination of the construction time of the secondary lining is the focus of each tunnel construction project.

[0003] The most traditional approach is to directly determine whether the current tunnel has reached a stable state based on long-term deformation monitoring data. Although this method is reliable, it is time-consuming and requires continuous on-site monitoring, which increases project costs and construction cycles.

[0004] Another common approach is to construct a mathematical model based on partial deformation monitoring data and predict the time required for deformation to reach a stable state through curve fitting. This method can reduce the monitoring time, but in actual applications, it is affected by geological conditions, construction methods, environmental factors, etc., and repeated monitoring and modeling work is required for different projects and different tunnel sections. In other words, the scope of application of the established mathematical model is very limited, and the results are difficult to reuse.

[0005] In addition, finite element analysis, as an advanced numerical simulation method, also plays an important role in predicting tunnel deformation and determining the time for secondary lining construction. However, this technology also faces the problem of difficulty in selecting model parameters and poor adaptability to new projects. Each application requires recalibration of the model to match specific engineering conditions.

[0006] In summary, although the existing technologies can provide relatively accurate tunnel secondary lining construction time, they have limitations in adaptability and versatility. Therefore, it is necessary to propose a tunnel secondary lining construction time analysis method with strong applicability and versatility. Summary of the invention

[0007] The purpose of the present invention is to solve the problems of long deformation monitoring time and poor versatility of the existing tunnel secondary lining construction time analysis method, and to provide a method for predicting the tunnel secondary lining construction time through a lifelong learning mode of knowledge playback, which can be widely used in the field of tunnel secondary lining construction.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] S101, data collection and preprocessing;

[0010] The data collection and preprocessing include collecting the stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of completed length of secondary lining to total tunnel length, partial vault deformation monitoring value, and secondary lining construction time after initial support completion of each segment of each previous tunnel construction project as the original data set, converting the stratum type into numerical data, and then using a clustering method to identify abnormal values ​​in the original data set, replacing and filling abnormal values ​​and missing values ​​with the middle value of adjacent data, each segmented data is used as a sample data in the original data set, and the original data set containing N sample data is randomly divided into a training set and a validation set at a ratio of 8:2, and the training set data is placed in an experience playback buffer;

[0011] S102, initial model training;

[0012] The initial model training includes defining the model structure, setting the training parameters, and then using the training set to train the GRU model. After each epoch, the performance of the model on the validation set is evaluated. When the performance of the validation set no longer improves, the training is stopped in advance. During the training process, a grid search is performed on the model's hyperparameters, and k-fold cross validation is used to evaluate the model performance under different hyperparameter settings to find the optimal hyperparameter combination;

[0013] S103, model prediction effect evaluation;

[0014] The evaluation of the prediction effect of the model includes comparing the stratum type, groundwater level, ground stress, rock strength, and relative burial depth of the current new tunnel construction project with the values ​​in the original data set. When there is no data with the same stratum type and similar ground stress, rock strength, and relative burial depth, after the completion of the initial support, the deformation of the tunnel vault is observed for a long time until the daily deformation of the tunnel vault does not exceed 1 mm / d for 7 consecutive days, and then the secondary lining construction is started. The stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of the completed length of the secondary lining to the total length of the tunnel, partial vault deformation monitoring value, and secondary lining construction time after the completion of the initial support of each section of the current project are recorded as a sample data in the new knowledge data set;

[0015] S104, knowledge replay training;

[0016] The knowledge replay training includes, when the new knowledge data set reaches n pieces, using a clustering method to identify abnormal values ​​in the new knowledge data set, replacing and filling abnormal values ​​and missing values ​​with intermediate values ​​of adjacent data, forming a new training set with the new knowledge data set and sample data in the experience replay buffer, adding the new knowledge data set to the experience replay buffer in an appended form, and then further training the GRU model with the new training set;

[0017] S105, model application;

[0018] The model application includes, when the stratum type of the current new tunnel construction project is consistent with that in the original data set, and the ground stress, rock strength and relative burial depth data are similar, the trained model is used to predict the secondary lining construction time after the completion of the initial support according to the stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of the completed length of the secondary lining to the total length of the tunnel, and partial vault deformation monitoring values ​​of the current segment, and the secondary lining construction is carried out according to the prediction results;

[0019] S106, Feedback and iteration;

[0020] The feedback and iteration include paying close attention to the deformation and cracking problems of the secondary lining after the secondary lining is constructed according to the prediction results. In the case of large deformation and cracks in the secondary lining, it is considered that the prediction results of the model are inaccurate. According to the specific circumstances of the large deformation and cracks, consider amplifying the construction time of the secondary lining after the initial support is completed in the data sample of the current segment record, and use it as a sample data in the new knowledge data set to iteratively upgrade in the knowledge playback training.

[0021] As a preferred technical solution of the present invention, in step S102, the defined model structure includes no less than 3 GRU hidden layers in the neural network, the activation function adopts a linear activation function, and the model evaluation method is to use the mean square error and mean absolute error indicators to measure the prediction accuracy of the model.

[0022] As a preferred technical solution of the present invention, in step S104, when the data storage in the experience replay buffer reaches the maximum value, representative sample data is screened out and put into the experience replay buffer according to the distribution of the sample data.

[0023] As a preferred technical solution of the present invention, in step S104, the number of samples of the new knowledge data set is not less than 0.1 times the number of samples of the original data set.

[0024] The beneficial effects of the present invention are as follows: by means of machine learning, a basic prediction model for the construction time of secondary lining is trained using data from previous projects. The model can be directly used for prediction for projects with similar engineering profiles, and the model knowledge base is further optimized based on the deformation and crack feedback of the secondary lining structure in the later stage. The data of new projects are directly added to the knowledge base, and the prediction model is optimized through a knowledge playback mechanism to improve the generalization ability of the model, thereby ensuring that the model can be reused in different tunnel projects, and significantly improving the efficiency of secondary lining construction decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of the method for analyzing the construction time of the secondary lining of a tunnel according to the present invention. DETAILED DESCRIPTION

[0026] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention. It should be noted that many specific details are set forth in the following description to facilitate a full understanding of the present invention, but the present invention may also have other embodiments and variations thereof, and therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0027] This embodiment 1 is based on the 2# Tunnel of Serendah in the eighth section of the East Coast Rail Link project in Selangor, at the beginning of the C2 section. The tunnel has a starting mileage of CH04+641.88 and an ending mileage of CH14+235.88. The tunnel is 9594m long and is a single-hole double-track ballastless track composite lining with a standard line spacing of 4.2m. The secondary lining construction time analysis method for a segmented tunnel is as follows:

[0028] Data collection and preprocessing: collect the stratum type, groundwater level, geostress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of completed length of secondary lining to total tunnel length, partial vault deformation monitoring value, and secondary lining construction time after initial support completion of each segment of each previous tunnel construction project as the original data set, convert the stratum type into numerical data, and then use the clustering method to identify the outliers in the original data set, replace and fill the outliers and missing values ​​with the middle value of the adjacent data, and use each segment data as a sample data in the original data set. The original data set containing N sample data is randomly divided into a training set and a validation set at a ratio of 8:2, and the training set data is placed in the experience playback buffer;

[0029] Initial model training, define the model structure, set the training parameters, and then use the training set to train the GRU model. After each epoch, evaluate the performance of the model in the validation set. When the performance of the validation set no longer improves, stop training in advance. During the training process, grid search is performed on the hyperparameters of the model, and k-fold cross validation is used to evaluate the model performance under different hyperparameter settings to find the optimal hyperparameter combination. In the neural network of this embodiment, the GRU hidden layer is 3 layers, and the activation function uses a linear activation function. The model evaluation method is to use the mean square error and mean absolute error indicators to measure the prediction accuracy of the model;

[0030] Evaluation of the model prediction effect: comparing the stratum type, groundwater level, geostress, rock strength, and relative burial depth of the current new tunnel construction project with the values ​​in the original data set, if there is no data with the same stratum type and similar geostress, rock strength, and relative burial depth, after the completion of the initial support, long-term observation of the tunnel vault deformation is carried out until the daily deformation of the tunnel vault does not exceed 1 mm / d for 7 consecutive days, and then the secondary lining construction is started. The stratum type, groundwater level, geostress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of the completed length of the secondary lining to the total length of the tunnel, partial vault deformation monitoring value, and secondary lining construction time after the completion of the initial support of each section of the current project are recorded as a sample data in the new knowledge data set;

[0031] Knowledge replay training, the number of samples in the new knowledge data set reaches 0.1 times the number of samples in the original data set, and the clustering method is used to identify the outliers in the new knowledge data set, and the outliers and missing values ​​are replaced and filled with the middle values ​​of the adjacent data. The sample data of the new knowledge data set and the experience replay buffer are combined into a new training set, and the new knowledge data set is added to the experience replay buffer in the form of appending. The data storage in the experience replay buffer reaches the maximum value, and according to the distribution of the sample data, representative sample data is screened out and put into the experience replay buffer, and then the new training set is used to further train the GRU model.

[0032] This embodiment 2 is based on the 2# Tunnel of the 8th Division of the East Coast Rail Link Project of Malaysia, located in Selangor, at the beginning of the C2 section. The tunnel has a starting mileage of CH04+641.88 and an ending mileage of CH14+235.88. The tunnel is 9594m long and is a single-hole double-track ballastless track composite lining with a standard line spacing of 4.2m. The analysis method for the construction time of the secondary lining of a segmented tunnel is as follows:

[0033] Model application: When the stratum type of the current new tunnel construction project is consistent with that in the original data set, and the ground stress, rock strength and relative burial depth data are similar, the trained model is used to predict the construction time of the secondary lining after the completion of the initial support based on the stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of the completed length of the secondary lining to the total length of the tunnel, and partial vault deformation monitoring values ​​of the current segment, and the secondary lining construction is carried out according to the prediction results;

[0034] Feedback and iteration, after the secondary lining construction is carried out according to the prediction results, the deformation and crack problems of the secondary lining are closely monitored. In the case of large deformation and cracks in the secondary lining, the prediction results of the model are considered inaccurate. According to the specific circumstances of the large deformation and cracks, consider amplifying the construction time of the secondary lining after the initial support is completed in the data sample of the current segment record, and use it as a sample data in the new knowledge data set, and perform iterative upgrades in the knowledge playback training. In this embodiment, the secondary lining does not have large deformation and cracks.

[0035] In summary, the tunnel secondary lining construction time analysis method of the present invention has the characteristics of strong adaptability and high analysis efficiency in the field of tunnel secondary lining construction.

[0036] It should be understood that the above-mentioned embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making pioneering innovations all fall within the scope of protection of the present invention.

Claims

1. A method for analyzing the construction time of secondary lining of a tunnel, characterized in that The specific steps include: S101, data collection and preprocessing; The data collection and preprocessing include collecting the stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of completed length of secondary lining to total tunnel length, partial vault deformation monitoring value, and secondary lining construction time after initial support completion of each segment of each previous tunnel construction project as the original data set, converting the stratum type into numerical data, and then using a clustering method to identify abnormal values ​​in the original data set, replacing and filling abnormal values ​​and missing values ​​with the middle value of adjacent data, each segmented data is used as a sample data in the original data set, and the original data set containing N sample data is randomly divided into a training set and a validation set at a ratio of 8:2, and the training set data is placed in an experience playback buffer; S102, initial model training; The initial model training includes defining the model structure, setting the training parameters, and then using the training set to train the GRU model. After each epoch, the performance of the model on the validation set is evaluated. When the performance of the validation set no longer improves, the training is stopped in advance. During the training process, a grid search is performed on the model's hyperparameters, and k-fold cross validation is used to evaluate the model performance under different hyperparameter settings to find the optimal hyperparameter combination; S103, model prediction effect evaluation; The evaluation of the prediction effect of the model includes comparing the stratum type, groundwater level, ground stress, rock strength, and relative burial depth of the current new tunnel construction project with the values ​​in the original data set. When there is no data with the same stratum type and similar ground stress, rock strength, and relative burial depth, after the completion of the initial support, the deformation of the tunnel vault is observed for a long time until the daily deformation of the tunnel vault does not exceed 1 mm / d for 7 consecutive days, and then the secondary lining construction is started. The stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of the completed length of the secondary lining to the total length of the tunnel, partial vault deformation monitoring value, and secondary lining construction time after the completion of the initial support of each section of the current project are recorded as a sample data in the new knowledge data set; S104, knowledge replay training; The knowledge replay training includes, when the new knowledge data set reaches n pieces, using a clustering method to identify abnormal values ​​in the new knowledge data set, replacing and filling abnormal values ​​and missing values ​​with intermediate values ​​of adjacent data, forming a new training set with the new knowledge data set and sample data in the experience replay buffer, adding the new knowledge data set to the experience replay buffer in an appended form, and then further training the GRU model with the new training set; S105, model application; The model application includes, when the stratum type of the current new tunnel construction project is consistent with that in the original data set, and the ground stress, rock strength and relative burial depth data are similar, the trained model is used to predict the secondary lining construction time after the completion of the initial support according to the stratum type, groundwater level, ground stress, rock strength, relative burial depth, excavation speed, initial support completion time, ratio of the completed length of the secondary lining to the total length of the tunnel, and partial vault deformation monitoring values ​​of the current segment, and the secondary lining construction is carried out according to the prediction results; S106, Feedback and Iteration; The feedback and iteration include paying close attention to the deformation and cracking problems of the secondary lining after the secondary lining is constructed according to the prediction results. In the case of large deformation and cracks in the secondary lining, it is considered that the prediction results of the model are inaccurate. According to the specific circumstances of the large deformation and cracks, consider amplifying the construction time of the secondary lining after the initial support is completed in the data sample of the current segment record, and use it as a sample data in the new knowledge data set to iteratively upgrade in the knowledge playback training.

2. The method for analyzing the construction time of secondary lining of a tunnel according to claim 1 is characterized by: In step S102, the defined model structure includes at least 3 GRU hidden layers in the neural network, the activation function adopts a linear activation function, and the model evaluation method uses the mean square error and mean absolute error indicators to measure the prediction accuracy of the model.

3. The method for analyzing the construction time of secondary lining of a tunnel according to claim 1 is characterized by: In step S104, when the data storage in the experience replay buffer reaches the maximum value, representative sample data is screened out according to the distribution of sample data and put into the experience replay buffer.

4. The method for analyzing construction time of secondary lining of a tunnel according to claim 1 is characterized in that: In step S104, the number of samples in the new knowledge data set is not less than 0.1 times the number of samples in the original data set.

Images