A method and device for predicting shield tunneling parameters throughout the entire life cycle

Abstract

The present invention discloses a method and device for predicting shield tunneling parameters throughout its lifecycle. The method comprises obtaining historical tunneling task data and predicted steps, preprocessing the historical data, and constructing a sample dataset using the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters; using the sample dataset to train a deep convolutional network based on an attention mechanism; the deep convolutional network takes geological parameters and historical tunneling parameters as input, outputs future tunneling parameters, and has an output dimension of predicted steps; and using the trained deep convolutional network to predict the tunneling parameters to obtain prediction results. The present invention proposes a standardized preprocessing method and uses a deep convolutional network based on an attention mechanism to dynamically distribute the weights of input features. This method fully considers the ease of use of different devices, while also taking into account the relationship between geological information and tunneling parameters, and enables the possibility of multi-step prediction.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology of shield technology, and in particular to a method and device for predicting shield tunneling parameters throughout its life cycle. Background Art

[0002] The shield method is a fully mechanized construction method commonly used in tunnel excavation today. It offers high construction safety and minimal impact on the ground surface, weather, buildings, and underground pipelines. During shield construction, operators adjust a series of tunneling parameters. However, in current practice, these adjustments often rely on subjective experience. The accuracy of these adjustments depends on factors such as the current state, historical experience, and subjective perception, leading to uncertainties in construction quality. Furthermore, becoming a qualified operator is time-consuming. Therefore, there is an urgent need for an intelligent tunneling parameter adjustment method to improve the quality of shield construction.

[0003] Existing intelligent tunneling parameter adjustment methods can be divided into three categories: 1. Methods based on traditional machine learning algorithms, such as least squares support vector machines, adaptive neuro-fuzzy inference algorithms, random forests, and Bayesian optimization algorithms. 2. Methods based on traditional neural networks and deep network algorithms, such as BP neural networks and their optimization algorithms, including particle swarm optimization, which dynamically adjust neural network hyperparameters such as neurons. 3. Methods based on time series networks. The more common ones are algorithms based on long short-term memory networks (LSTMs) and their optimization algorithms, and the more similar ones are algorithms using convolutional neural networks (CNNs). Here, the optimization algorithm adjusts the model's hyperparameters, similar to automatic parameter optimization.

[0004] For category 1, traditional machine learning methods often perform poorly in terms of applicability and reliability when faced with datasets with such a large number of features due to algorithmic limitations. Good values can only be obtained for individual problems, and different problems need to be analyzed separately, which makes them less practical.

[0005] For category 2, traditional neural networks, such as the back propagation (BP) algorithm, are prone to falling into local minima and are highly dependent on the initial network parameter settings. While some optimization algorithms have addressed this issue, they have not been able to fundamentally address these shortcomings. Furthermore, the accompanying issues of overfitting and slow convergence have become bottlenecks in algorithm development. While solely using deep learning methods like CNNs can address some of these issues, they require careful feature selection and extensive experimentation by professionals, and the inability to determine the impact of different features on the output is a key factor limiting the models' applicability. The aforementioned models in categories 1 and 2 serve only as validation of existing data and do not offer predictions for the future, making them of limited practical value.

[0006] For category 3, while time series network-based methods achieve future predictions, they place excessive emphasis on the temporal relationships between tunneling parameters, which in turn places extremely high demands on dataset accuracy. This overemphasis on the changing trends of individual parameters also neglects the influence of internal system parameters and changes in external geological parameters. However, fitting future parameter changes solely based on internal changes is incomplete for shield machines operating in complex geological conditions. Furthermore, the high number of parameters in the time series model increases the computational burden.

[0007] Furthermore, existing intelligent tunneling parameter adjustment methods lack a standardized, unified description of preprocessing methods, which are highly correlated with the final output and impose strict restrictions on input characteristics, limiting the ease of use of different shield equipment models. Furthermore, existing technologies fail to fully consider the relationship between geological information and the various sub-devices of the shield equipment. Besides time series models, none consider multi-step predictions into the future, relying instead on validation of existing data. Summary of the Invention

[0008] The present invention provides a method and device for predicting shield tunneling parameters throughout the entire life cycle, so as to at least to some extent solve the technical problems existing in the prior art proposed in the above background technology.

[0009] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0010] In one aspect, the present invention provides a method for predicting shield tunneling parameters throughout the entire life cycle, comprising:

[0011] Obtaining historical data and predicted steps of the tunneling task, preprocessing the historical data, and constructing a sample data set using the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters;

[0012] Using the sample data set to train a deep convolutional network based on an attention mechanism, wherein the input of the deep convolutional network is geological parameters and historical excavation parameters, the output is future excavation parameters, and the output dimension is the number of predicted steps;

[0013] The trained deep convolutional network is used to predict the tunneling parameters to obtain the prediction results.

[0014] Furthermore, the acquisition of historical data and predicted number of steps of the tunneling task, preprocessing of the historical data, and construction of a sample data set using the preprocessed historical data include:

[0015] Obtain relevant geological parameters of the excavation stratum, actual excavation parameters and predicted number of steps of the excavation task;

[0016] Mark the geological parameters and put the marking results into the database;

[0017] Match the marked geological parameters with the excavation parameters according to the mileage and ring number to generate geological-excavation data;

[0018] Discretize geological parameters;

[0019] Delete non-excavation section data and irrelevant variables in the discretized geological-excavation data to remove redundant data;

[0020] Complete missing data for geological-tunneling data after removing redundant data;

[0021] Perform data normalization on geological-tunneling data to supplement missing data;

[0022] The normalized geological-excavation data are used to create training sets, validation sets, and test sets according to preset proportions.

[0023] Furthermore, the marking of geological parameters and storing the marking results in a database include:

[0024] Classify geological parameters into textual data and tabular data;

[0025] The BERT model is used to annotate text data. The corresponding historical training text is selected to train the BERT model according to the geological parameters to be annotated. Then the text to be annotated is input into the trained BERT model to obtain the annotation results, and the annotation results are placed in the database. The Key-Value method is used to extract the data in the table, and the extracted data is annotated, and the annotation results are placed in the database.

[0026] Furthermore, missing data is completed for the geological-tunneling data after redundant data removal, including:

[0027] The mean of the first two digits and the last two digits of the missing data is used as the value of the current missing data.

[0028] Furthermore, the geological-tunneling data used to complete the missing data are normalized using the formula:

[0029]

[0030] Among them, X′ represents the normalization result, X represents the original sequence to be normalized, and X mean represents the average value of the original sequence, X std represents the standard deviation of the original series.

[0031] Furthermore, the deep convolutional network includes an input layer, a one-dimensional convolutional layer, an attention layer, and a fully connected layer;

[0032] The sample dataset is used to train a deep convolutional network based on the attention mechanism, including:

[0033] Randomly initialize the one-dimensional convolution layer and attention layer, take the training set as input, and put it into the one-dimensional convolution layer for convolution calculation. The formula is as follows:

[0034] y i =f(W*[x i:i+k ]+b)

[0035] Among them, y i Represents the convolution result, x i Represents input features, k represents the step size of each convolution, W represents the convolution kernel, b represents the bias term, and f() represents the RELU activation function;

[0036] The convolution output Y={y1,y2…y i} Input into the attention layer to calculate the weights of different features. The formula is:

[0037] α i =softmax(e(y i ,q j ))

[0038]

[0039] in, q j is the current input context, α i is the attention distribution, whose value is between [0,1]; C i That is, the importance value of the feature, and N represents the length of the input sequence;

[0040] The output is then converted into a one-dimensional vector, input into the fully connected layer, and the final prediction result is output.

[0041] Furthermore, the back propagation optimization in the deep convolutional network uses the Adam optimizer, and the formula is as follows:

[0042]

[0043]

[0044] Among them, α is the step size, is the exponential decay rate, θ t ,θ t-1 is the parameter vector, m t , v t are the first-order and second-order moment vectors, t is the time step, Epsilon is used to avoid division by 0.

[0045] Furthermore, the loss function in the deep convolutional network adopts the Smooth L1 function, and the formula is as follows:

[0046]

[0047] The measurement function adopts the determination coefficient measurement method, and the formula is:

[0048]

[0049]

[0050] Where T={t1,t2…t n} is a real value sequence, is the average value of the true value sequence, and n is the length of the sequence. n} is the predicted value sequence finally output by the model, and its length is also n, R 2 represents the coefficient of determination.

[0051] Furthermore, after predicting the tunneling parameters to be predicted using the trained deep convolutional network and obtaining the prediction results, the full life cycle shield tunneling parameter prediction method further includes:

[0052] The prediction results and corresponding geological parameters are stored as new samples in the training set of the database. After running for a certain period of time, the deep convolutional network based on the attention mechanism is dynamically updated.

[0053] On the other hand, the present invention also provides a full life cycle shield tunneling parameter prediction device, comprising:

[0054] A data preprocessing module is used to obtain historical data and predicted steps of the tunneling task, preprocess the historical data, and construct a sample data set with the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters;

[0055] a model training module for training a deep convolutional network based on an attention mechanism using the sample data set obtained by the data preprocessing module; wherein the input of the deep convolutional network is geological parameters and historical excavation parameters, and the output is future excavation parameters, and the output dimension is the number of predicted steps;

[0056] The excavation parameter prediction module is used to predict the excavation parameters to be predicted using the deep convolutional network trained by the model training module to obtain prediction results.

[0057] On the other hand, the present invention further provides an electronic device, comprising a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

[0058] In yet another aspect, the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the instruction is loaded and executed by a processor to implement the above method.

[0059] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0060] 1. The present invention proposes a method for standardizing the processing of raw geological and excavation data, which minimizes the impact of different raw data quality on training results, improves the stability of model input, and thus improves the stability of prediction results.

[0061] 2. The present invention fully considers the relationship between geological information and various sub-equipment of the shield equipment, maps the change of time through the change of geological information, and more comprehensively involves various influencing factors in the operation process of the shield.

[0062] 3. This invention considers the differences in tunneling parameter outputs from different devices and the resulting instability in model results caused by changes in output. It proposes a deep convolutional network based on an attention mechanism. The attention layer continuously learns and automatically adjusts the importance distribution between features, assigning each feature an importance weight, making model learning more targeted. This also ensures that model accuracy does not drop significantly after parameter changes. The CNN layer effectively learns the different features of the data, taking into account the local nature of the data.

[0063] 4. The present invention takes into account the actual operation situation and can realize model training and high-precision prediction with a small amount of computing resources, with high robustness, feasibility and stability.

[0064] 5. The present invention takes into account the possibility of being used for multi-step prediction, because the attention layer can autonomously learn to adjust the weight distribution of features. This also allows the model proposed in the present invention to be easily extended to multi-step prediction at any time, improving the practical usability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0066] Figure 1 This is a schematic diagram of a method for predicting shield tunneling parameters throughout the entire life cycle provided by an embodiment of the present invention;

[0067] Figure 2 This is a schematic diagram of the execution flow of the full life cycle shield tunneling parameter prediction method provided by an embodiment of the present invention in actual engineering applications;

[0068] Figure 3 This is a network diagram of a deep convolutional network based on an attention mechanism provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0070] First embodiment

[0071] This embodiment provides a full life cycle shield tunneling parameter prediction method, which standardizes the preprocessing process and uses a deep convolutional network based on the attention mechanism to predict tunneling parameters. The principle is as follows: Figure 1 As shown in the following example, the actual use in engineering is as follows Figure 2 The method can be implemented by an electronic device, which can be a terminal or a server. Specifically, the execution process of the method includes the following steps:

[0072] S1, obtaining historical data and predicted number of steps of the tunneling task, preprocessing the historical data, and constructing a sample data set with the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters;

[0073] Specifically, in this embodiment, the implementation process of the above S1 is as follows:

[0074] S11, obtaining relevant geological parameters under the excavation stratum, historical data of the excavation task and the predicted number of steps;

[0075] S12, classifying geological parameters into text data and table data; annotating the text data using the BERT model, selecting corresponding historical training texts to train the BERT model based on the geological parameters to be annotated, then inputting the text to be annotated into the trained BERT model to obtain annotation results, and storing the annotation results in a database; extracting data from the table using the Key-Value method for unannotated table data, annotating the extracted data, and storing the annotation results in the database;

[0076] S13, matching geological parameters with tunneling parameters according to the mileage and ring number to generate geological-tunneling data;

[0077] S14, discretizing the geological parameters; deleting non-excavation segment data and irrelevant variables, such as cutterhead left turn, cutterhead right turn, etc., from the discretized geological-excavation data to remove redundant data; completing missing data in the geological-excavation data after removing redundant data; and finally normalizing the data;

[0078] Among them, the missing value q i Use the average value method to complete the formula as follows:

[0079]

[0080] The normalization formula is:

[0081]

[0082] Among them, X′ represents the normalization result, X represents the original sequence to be normalized, and X mean represents the average value of the original sequence, X std represents the standard deviation of the original series;

[0083] In S15, the input tunneling parameters to be predicted are used as the output of a deep convolutional network based on an attention mechanism, with the number of prediction steps as the dimension of each output. The remaining data is used as the input of the deep convolutional network based on the attention mechanism. The above data is used to create training, validation, and test sets in a ratio of 0.7:0.2:0.1.

[0084] S2, using the sample data set to train a deep convolutional network based on the attention mechanism; wherein the input of the deep convolutional network is the geological parameters and the historical excavation parameters, and the output is the future excavation parameters, and the output dimension is the number of predicted steps;

[0085] Specifically, in this embodiment, the deep convolutional network based on the attention mechanism is as follows Figure 3 As shown in Figure 1, it includes an input layer, a one-dimensional convolutional layer, an attention layer, and a fully connected layer. The implementation process of S2 is as follows:

[0086] The data is input into the deep convolutional network based on the attention mechanism for training and prediction, specifically:

[0087] S21 randomly initializes the one-dimensional convolution layer and attention layer, takes the training set obtained in S1 as input, and puts it into the one-dimensional convolution layer for convolution calculation. The formula is as follows:

[0088] y i =f(W*[x i:i+k ]+b)

[0089] Among them, y i Represents the convolution result, x i Represents input features, k represents the step size of each convolution, W represents the convolution kernel, b represents the bias term, and f() represents the RELU activation function;

[0090] S22, the convolution output Y={y1,y2…y i} Input into the attention layer to calculate the weights of different features. The formula is:

[0091] α i =softmax(e(y i ,q j ))

[0092]

[0093] in, q j is the current input context, α i is the attention distribution, whose value is between [0,1]; C i That is, the importance value of the feature, and N represents the length of the input sequence;

[0094] S23, then enters the flattening layer to convert the output into a one-dimensional vector, the input enters the fully connected layer, and outputs the final prediction result.

[0095] Among them, the back propagation optimization in the deep convolutional network uses the Adam optimizer. This optimizer can adapt to sparse gradients and alleviate gradient oscillation problems. The specific formula is as follows:

[0096]

[0097]

[0098] Among them, α is the step size, is the exponential decay rate, θ t ,θ t-1 is the parameter vector, m t , v t are the first-order and second-order moment vectors, t is the time step, Epsilon is used to avoid division by 0.

[0099] The loss function uses the Smooth L1 function, and the formula is as follows:

[0100]

[0101] The measurement function adopts the determination coefficient measurement method, and the specific formula is:

[0102]

[0103]

[0104] Where T={t1,t2…t n} is a real value sequence, is the average value of the true value sequence, and n is the length of the sequence. n} is the predicted value sequence finally output by the model, and its length is also n, R 2 represents the coefficient of determination.

[0105] The use of the coefficient of determination measurement method can solve the problem of inconsistent final measurement results caused by different dimensions.

[0106] S3, using the trained deep convolutional network to predict the tunneling parameters to be predicted, and obtain the prediction results.

[0107] Furthermore, in this embodiment, after the trained model is obtained by using the above method and the parameters to be predicted are predicted, the full life cycle shield tunneling parameter prediction method further includes:

[0108] The results are stored as new input in the training sample set of the database, and the deep convolutional network based on the attention mechanism is dynamically updated after running for a certain period of time.

[0109] It should be noted that although the method of this embodiment introduces the attention mechanism and increases the amount of calculation to a certain extent, it can significantly reduce the model training batches and obtain higher prediction accuracy.

[0110] In summary, this embodiment proposes a method for standardizing the processing of raw geological and excavation data, minimizing the impact of varying raw data quality on training results, improving the stability of model inputs, and thus, the stability of prediction results. Furthermore, this embodiment considers the differences in excavation parameter outputs from different devices and the instability of model results caused by changes in output, and proposes a deep convolutional network based on an attention mechanism. The attention layer continuously learns, automatically adjusting the importance distribution between features and assigning an importance weight to each feature, making model learning more targeted. This also prevents significant drops in model accuracy after parameter changes. The CNN layer effectively learns the different features of the data, accounting for its local nature. Taking into account actual operational conditions, it can achieve model training and high-precision predictions with minimal computing resources, demonstrating high robustness, feasibility, and stability. Furthermore, this embodiment considers the possibility of multi-step prediction, allowing the proposed model to be easily extended to this application at any time, thereby improving its practical usability.

[0111] Second embodiment

[0112] This embodiment provides a full life cycle shield tunneling parameter prediction device, including the following modules:

[0113] A data preprocessing module is used to obtain historical data and predicted steps of the tunneling task, preprocess the historical data, and construct a sample data set with the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters;

[0114] A model training module, configured to train a deep convolutional network based on an attention mechanism using the sample data set obtained by the data preprocessing module; wherein the input of the deep convolutional network is geological parameters, the output is tunneling parameters, and the output dimension is the number of predicted steps;

[0115] The excavation parameter prediction module is used to predict the excavation parameters to be predicted using the deep convolutional network trained by the model training module to obtain prediction results.

[0116] The full life cycle shield tunneling parameter prediction device of this embodiment corresponds to the full life cycle shield tunneling parameter prediction method of the above-mentioned first embodiment; wherein, the functions implemented by each functional module in the full life cycle shield tunneling parameter prediction device of this embodiment correspond one-to-one to each process step in the full life cycle shield tunneling parameter prediction method of the above-mentioned first embodiment; therefore, they will not be repeated here.

[0117] Third embodiment

[0118] This embodiment provides an electronic device, which includes a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the method of the first embodiment.

[0119] The electronic device may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) and one or more memories, wherein the memory stores at least one instruction, which is loaded by the processor to execute the above method.

[0120] Fourth embodiment

[0121] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device. The instructions stored therein can be loaded by a processor in a terminal to execute the method described above.

[0122] Furthermore, it should be noted that the present invention may be provided as a method, apparatus, or computer program product. Thus, embodiments of the present invention may take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media containing computer-usable program code.

[0123] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0124] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0125] It should also be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal device comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal device comprising the element.

[0126] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be noted that although the preferred embodiment of the present invention has been described, it is clear that those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles of the present invention. Such improvements and modifications should also be considered as within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A full life cycle shield tunneling parameter prediction method, characterized in that: include: Obtaining historical data and predicted steps of the tunneling task, preprocessing the historical data, and constructing a sample data set using the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters; Using the sample data set to train a deep convolutional network based on an attention mechanism, wherein the input of the deep convolutional network is geological parameters and historical excavation parameters, the output is future excavation parameters, and the output dimension is the number of predicted steps; Use the trained deep convolutional network to predict the tunneling parameters to obtain the prediction results; The process of obtaining historical data and predicted number of steps of the tunneling task, preprocessing the historical data, and constructing a sample data set using the preprocessed historical data includes: Obtain relevant geological parameters of the excavation stratum, actual excavation parameters and predicted number of steps of the excavation task; Mark the geological parameters and put the marking results into the database; Match the marked geological parameters with the excavation parameters according to the mileage and ring number to generate geological-excavation data; Discretize geological parameters; Delete non-excavation section data and irrelevant variables in the discretized geological-excavation data to remove redundant data; Complete missing data for geological-tunneling data after removing redundant data; Perform data normalization on geological-tunneling data to supplement missing data; The normalized geological-excavation data are used to create training sets, validation sets, and test sets according to preset proportions; The marking of geological parameters and storing the marking results in a database include: Classify geological parameters into textual data and tabular data; The text data is annotated using the BERT model. The corresponding historical training text is selected according to the geological parameters to be annotated to train the BERT model. The text to be annotated is then input into the trained BERT model to obtain the annotation results, which are then stored in the database. The table data is extracted using the Key-Value method, and the extracted data is annotated and stored in the database. The deep convolutional network includes an input layer, a one-dimensional convolutional layer, an attention layer and a fully connected layer; The sample dataset is used to train a deep convolutional network based on the attention mechanism, including: Randomly initialize the one-dimensional convolution layer and attention layer, take the training set as input, and put it into the one-dimensional convolution layer for convolution calculation. The formula is as follows: y i =f(W*[x i:i+k ]+b) Among them, y i Represents the convolution result, x i Represents input features, k represents the step size of each convolution, W represents the convolution kernel, b represents the bias term, and f() represents the RELU activation function; The convolution output Y={y1,y2…y i } Input into the attention layer to calculate the weights of different features. The formula is: α i =softmax(e(y i ,q j )) in, q j is the current input context, α i is the attention distribution, whose value is between [0,1]; C i That is, the importance value of the feature, and N represents the length of the input sequence; The output is then converted into a one-dimensional vector, input into the fully connected layer, and the final prediction result is output.

2. The method for predicting shield tunneling parameters throughout the entire life cycle according to claim 1, wherein: The missing data completion of the geological-tunneling data after removing redundant data includes: The mean of the first two digits and the last two digits of the missing data is used as the value of the current missing data.

3. The full life cycle shield tunneling parameter prediction method according to claim 1, characterized in that: The data normalization of the geological-tunneling data that supplements the missing data is performed using the formula: Among them, X′ represents the normalization result, X represents the original sequence to be normalized, and X mean represents the average value of the original sequence, X std represents the standard deviation of the original series.

4. The method for predicting shield tunneling parameters throughout the entire life cycle according to claim 1, wherein: The back propagation optimization in the deep convolutional network adopts the Adam optimizer, and the formula is as follows: Among them, α is the step size, is the exponential decay rate, θ t ,θ t-1 is the parameter vector, m t , v t are the first-order and second-order moment vectors, t is the time step, Epsilon is used to avoid division by 0.

5. The method for predicting shield tunneling parameters throughout the entire life cycle according to claim 4, wherein: The loss function in the deep convolutional network adopts the Smooth L1 function, and the formula is as follows: The measurement function adopts the determination coefficient measurement method, and the formula is: Where T={t1,t2…t n } is a real value sequence, is the average value of the true value sequence, n is the sequence length; P = {p1, p2…p n } is the predicted value sequence finally output by the model, and its length is n; R 2 is the coefficient of determination.

6. The method for predicting shield tunneling parameters throughout the entire life cycle according to claim 1, wherein: After predicting the tunneling parameters to be predicted using the trained deep convolutional network and obtaining the prediction results, the full life cycle shield tunneling parameter prediction method further includes: The prediction results and corresponding geological parameters are stored as new samples in the training set of the database. After running for a certain period of time, the deep convolutional network based on the attention mechanism is dynamically updated.

7. A full life cycle shield tunneling parameter prediction device, characterized in that: include: A data preprocessing module is used to obtain historical data and predicted steps of the tunneling task, preprocess the historical data, and construct a sample data set with the preprocessed historical data; wherein the historical data includes geological parameters and tunneling parameters; a model training module for training a deep convolutional network based on an attention mechanism using the sample data set obtained by the data preprocessing module; wherein the input of the deep convolutional network is geological parameters and historical excavation parameters, and the output is future excavation parameters, and the output dimension is the number of predicted steps; A tunneling parameter prediction module is used to predict the tunneling parameters to be predicted using the deep convolutional network trained by the model training module to obtain a prediction result; The process of obtaining historical data and predicted number of steps of the tunneling task, preprocessing the historical data, and constructing a sample data set using the preprocessed historical data includes: Obtain relevant geological parameters of the excavation stratum, actual excavation parameters and predicted number of steps of the excavation task; Mark the geological parameters and put the marking results into the database; Match the marked geological parameters with the excavation parameters according to the mileage and ring number to generate geological-excavation data; Discretize geological parameters; Delete non-excavation section data and irrelevant variables in the discretized geological-excavation data to remove redundant data; Complete missing data for geological-tunneling data after removing redundant data; Perform data normalization on geological-tunneling data to supplement missing data; The normalized geological-excavation data are used to create training sets, validation sets, and test sets according to preset proportions; The marking of geological parameters and storing the marking results in a database include: Classify geological parameters into textual data and tabular data; The text data is annotated using the BERT model. The corresponding historical training text is selected according to the geological parameters to be annotated to train the BERT model. The text to be annotated is then input into the trained BERT model to obtain the annotation results, which are then stored in the database. The table data is extracted using the Key-Value method, and the extracted data is annotated and stored in the database. The deep convolutional network includes an input layer, a one-dimensional convolutional layer, an attention layer and a fully connected layer; The sample dataset is used to train a deep convolutional network based on the attention mechanism, including: Randomly initialize the one-dimensional convolution layer and attention layer, take the training set as input, and put it into the one-dimensional convolution layer for convolution calculation. The formula is as follows: y i =f(W*[x i:i+k ]+b) Among them, y i Represents the convolution result, x i Represents input features, k represents the step size of each convolution, W represents the convolution kernel, b represents the bias term, and f() represents the RELU activation function; The convolution output Y={y1,y2…y i } Input into the attention layer to calculate the weights of different features. The formula is: α i =softmax(e(y i ,q j )) in, q j is the current input context, α i is the attention distribution, whose value is between [0,1]; C i That is, the importance value of the feature, and N represents the length of the input sequence; The output is then converted into a one-dimensional vector, input into the fully connected layer, and the final prediction result is output.

Images