A knowledge-guided adaptive tunnel time series data enhancement method, system, terminal and storage medium

Abstract

The application relates to the field of data processing and discloses a knowledge-guided adaptive tunnel time series data enhancement method, system, terminal and storage medium.The method comprises the following steps: analyzing original time series data in a tunnel construction process and constructing a knowledge-guided target function; then performing variational mode decomposition and controllable reconstruction on the original sequence to obtain a preliminary enhanced sequence; on this basis, explicit constraints of distortion disturbance and prediction delay are introduced, and a multi-objective optimization strategy is adopted to minimize sequence reconstruction error and lag order, and the final enhanced sequence is obtained through iterative solution; multi-source time series data of a tunnel site are collected, the enhanced sequence is input into a prediction model to obtain a next-time prediction value; then, the real value at this time is obtained and spliced with a historical sequence, and the updated sequence is subjected to enhancement and prediction again, so that rolling enhancement and rolling prediction of the sequence are realized. The application provides an efficient and reliable data enhancement method for a time series prediction task in the tunnel field.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a knowledge-guided adaptive tunneling time-series data augmentation method, system, terminal, and computer-readable storage medium. Background Technology

[0002] With their advantages of high efficiency and continuous construction, mechanized tunnel excavation methods, represented by full-face hard rock tunnel boring machines (TBMs) and various types of shield tunneling machines, have become the core technology of modern tunnel construction. Real-time and accurate prediction of tunnel time-series data helps to dynamically perceive the excavation process and achieve forward-looking control, which is of great significance for ensuring construction safety and improving construction efficiency.

[0003] However, during tunnel excavation, the raw data collected often contains significant random errors due to external environmental factors such as mechanical vibration, resulting in obvious non-stationarity and extremely short-term time dependence in the sequence. In such cases, models often struggle to effectively extract useful information from the sequence, easily leading to prediction lag, feature failure, and other problems, severely impacting the effectiveness of time-series-based predictions. Therefore, enhancing tunnel time-series data is crucial to improving its usability.

[0004] Currently, mainstream time-series data augmentation methods include time-domain transformation and frequency-domain transformation. While these methods show promise in improving sequence usability, their application in tunnel time-series data remains limited. Firstly, existing methods are largely derived from general domains, often ignoring the non-stationary and strongly nonlinear characteristics of tunnel data. Simple noise reduction / de-noiseing, scaling, and time transformations can disrupt the physical coupling between parameters, reducing data fidelity. Furthermore, these methods typically have limited effectiveness when applied to tunnel sequence data. Secondly, many data augmentation methods heavily rely on preset hyperparameters, and their effectiveness largely depends on the accuracy of parameter selection. However, as tunnel excavation progresses, these hyperparameters often require frequent adjustments, significantly reducing the reliability and practicality of the methods.

[0005] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0006] The main objective of this invention is to provide a knowledge-guided adaptive tunnel time-series data augmentation method, system, terminal, and computer-readable storage medium, aiming to solve the problems in existing tunnel time-series data augmentation methods, such as lack of domain knowledge guidance, difficulty in handling noise and prediction delays, and non-adaptive parameter selection, which lead to insufficient accuracy of tunnel construction time-series data.

[0007] To achieve the above objectives, the present invention provides a knowledge-guided adaptive tunneling time-series data augmentation method, which includes the following steps: The original sequence data collected during mechanized tunnel excavation is analyzed to construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term. Variational mode decomposition is performed on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure, and reconstruction processing is performed based on the multiple subsequence mode components to obtain preliminary enhanced sequence data; Based on the preliminary augmented sequence data, a lightweight time series model is trained and the predicted values ​​of the test set are obtained. Based on the actual values ​​and predicted values ​​of the test set, the lag order is calculated. The L2 norm of the reconstruction error and the hysteresis order are used as optimization objectives to optimize the parameters obtained by the variational mode decomposition process, thereby obtaining the optimal sequence enhancement parameters. Based on the optimal sequence enhancement parameters, variational mode decomposition and reconstruction are performed on the original sequence again to obtain the target enhanced sequence; When performing real-time prediction of tunnel time series data, the target enhancement sequence is input into a deep learning or time series prediction model for training, and the target value at the next moment is predicted. The target value is then output as the prediction result at the current moment. Once the true observation values ​​of all variables at the next time step are obtained, the true observation values ​​are concatenated with the original sequence, the sequence data augmentation operation is performed again, and the target value prediction for the next time step is continued based on the updated sequence until the preset termination condition is reached, so as to achieve rolling prediction.

[0008] Optionally, the knowledge-guided adaptive tunnel time-series data augmentation method, wherein parsing the original sequence data collected during mechanized tunnel excavation and constructing a knowledge-driven objective function specifically includes: Based on the parsed original sequence data, Norm construction and reconstruction error term: ; in, This represents the reconstruction error term. This refers to the original sequence data. express Norm, This indicates preliminary augmented sequence data. Indicates time, Represents the ordinal number of the subsequence. Indicates the first Subsequence modal components; Based on the parsed original sequence data, and according to the dynamic drift pattern caused by the prediction delay, a lag order term characterizing the lag of the sequence response is constructed: ; ; in, This represents the lag order term. For cross-correlation coefficients, Indicates a time delay. Indicates the total length of the sequence. Indicates in The true value of a lightweight time series model at any given time. Indicates in Predictions from a lightweight time series model; By fusing the reconstruction error term and the lag order term, a knowledge-driven objective function is obtained to guide the sequence decomposition and reconstruction process: ; in, This represents the knowledge-driven objective function. This indicates a strong penalty.

[0009] Optionally, the knowledge-guided adaptive tunneling time-series data augmentation method, wherein performing variational mode decomposition on the original sequence to obtain multiple sub-sequence mode components describing the dependency structure at different time scales specifically includes: Based on the variational objective minimization form, an optimization problem is constructed to decompose the original sequence: ; in, Represents the modal function. Indicates the center frequency. express Norm, This represents the convolution operation. Indicates to Find the partial derivative. Let be the Dirac distribution function. Indicates time, Represents the ordinal number of the subsequence. Represents the imaginary unit. Represents the natural constant; An augmented Lagrange function is constructed by introducing a penalty factor and a Lagrange operator. Based on the augmented Lagrange function, the optimization problem is iteratively solved using the alternating direction multiplier method to obtain multiple narrowband mode components. These multiple narrowband mode components are then used as multiple subsequence mode components. The parameters of the variational mode decomposition include the number of modes and the penalty factor.

[0010] Optionally, the knowledge-guided adaptive tunneling time-series data augmentation method, wherein the step of constructing an augmented Lagrangian function by introducing a penalty factor and a Lagrangian operator, and iteratively solving the optimization problem based on the augmented Lagrangian function using the alternating direction multiplier method to obtain multiple narrowband modal components, specifically includes: Initialize the modal function set, center frequency set, and Lagrange operator, and set the initial number of iterations; In each iteration, each modal function, each center frequency, and the Lagrange operator are updated sequentially. After the update is completed, the change in modal function between two adjacent iterations is calculated. The update is based on minimizing the subproblems of the augmented Lagrange function. When the change in the modal function is less than a preset threshold or the number of iterations reaches a preset maximum value, the iteration loop is terminated, and the set of modal functions obtained from the final iteration is taken as the multiple narrowband modal components.

[0011] Optionally, the knowledge-guided adaptive tunneling time series data augmentation method, wherein the step of training a lightweight time series model and obtaining predicted values ​​for a test set based on the preliminary augmented sequence data, and calculating the lag order based on the actual and predicted values ​​of the test set, specifically includes: The preliminary enhanced sequence data is normalized to obtain a normalized sequence: ; in, This refers to the preliminary enhanced sequence data; The normalized sequence is reconstructed using a sliding window approach to generate a feature matrix and a corresponding target vector. ; in, Represents the feature matrix, Represents the target vector. This indicates a data reconstruction operation. Indicates the length of the sliding window; The feature matrix and the target vector are divided into a training set and a test set. A lightweight time series model is trained using the training set, and the trained model is used to predict the test set to obtain normalized prediction values. Perform an inverse normalization operation on the normalized predicted values ​​to restore them to the same data volume as the initial augmented sequence, resulting in inverse normalized predicted values: ; in, This represents the normalized predicted value of the inverse regression. This represents the normalized predicted value; Based on the true values ​​of the test set and the predicted values ​​of the inverse normalization, the cross-correlation coefficients at different time offsets are calculated, and the time offset corresponding to the maximum value of the cross-correlation coefficient is taken as the lag order.

[0012] Optionally, the knowledge-guided adaptive tunneling time-series data augmentation method, wherein optimizing the parameters obtained from the variational mode decomposition process using the L2 norm of the reconstruction error and the lag order as optimization objectives to obtain optimal sequence augmentation parameters specifically includes: An initial population is randomly generated within a preset parameter space, which includes the range of values ​​for the number of modes and the range of values ​​for the penalty factor. Each individual in the initial population corresponds to a set of parameters for variational mode decomposition. Based on the parameters of the variational mode decomposition, the original sequence is subjected to variational mode decomposition and preliminary reconstruction, and the corresponding reconstruction error and lag order are calculated. When the hysteresis order is not zero, the reconstruction error is corrected to a maximum value; Based on the performance of the variational mode decomposition parameters on the two optimization objectives of reconstruction error and lag order, non-dominated sorting is performed on the individuals in the population to obtain the set of individuals at each non-dominated level. Based on the set of individuals at each non-dominated level, and combining ideal point sorting, normalization operation, and reference point mechanism, individuals that meet the optimization requirements are selected from the current population as retained solutions. Crossover and mutation operations are performed on the retained solutions to generate the next generation of candidate parameter population; the iterative process is repeated until the preset termination condition is met to obtain the target number of modes and the target penalty factor that achieves the optimal balance between reconstruction error and lag order, and the target number of modes and the target penalty factor are used as the optimal sequence enhancement parameters.

[0013] Optionally, the knowledge-guided adaptive tunneling time-series data augmentation method, when applied to target variable prediction in real-time tunneling, further includes: The system acquires the continuous time-related parameters corresponding to the target variable in real time, performs data augmentation processing on the target variable to obtain the target augmentation sequence, and concatenates the target augmentation sequence with the continuous time-related parameter continuous time series data to obtain multidimensional time series data. The multidimensional time series data is normalized and segmented using a sliding window to obtain training samples. The training samples are then input into a deep learning or time series prediction model for training, and the predicted value of the target variable at the next time step is output. Once the actual observation values ​​of the target variable and other variables are obtained at the next time step, the actual observation values ​​are concatenated with the original dataset to form an updated sequence. Based on the updated sequence, data augmentation is performed again, and target value prediction continues for the next time step to achieve rolling prediction.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a knowledge-guided adaptive tunneling time-series data augmentation system, wherein the knowledge-guided adaptive tunneling time-series data augmentation system includes: The objective function construction module is used to parse the raw sequence data collected during the mechanized tunnel excavation process and construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term. The decomposition and reconstruction module is used to perform variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the dependency structure at different time scales, and to perform reconstruction processing based on the multiple subsequence mode components to obtain preliminary enhanced sequence data; The lag order calculation module is used to train a lightweight time series model based on the preliminary augmented sequence data and obtain the predicted values ​​of the test set, and calculate the lag order based on the actual values ​​and predicted values ​​of the test set. The multi-objective parameter optimization module is used to optimize the parameters obtained by the variational mode decomposition process by taking the L2 norm of the reconstruction error and the hysteresis order as optimization objectives to obtain the optimal sequence enhancement parameters. The target augmented sequence generation module is used to re-perform variational mode decomposition and reconstruction on the original sequence based on the optimal sequence augmentation parameters to obtain the target augmented sequence; The real-time prediction module is used to input the target enhancement sequence into a deep learning or time series prediction model for training when making real-time predictions of tunnel time series data, predict the target value at the next time step, and output the target value as the prediction result at the current time step. The rolling update module is used to concatenate the real observation values ​​of all variables with the original sequence after obtaining the real observation values ​​of the next time step, perform the sequence data augmentation operation again, and continue to predict the target value of the next time step based on the updated sequence until the preset termination condition is reached, so as to realize rolling prediction.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a knowledge-guided adaptive tunneling time-series data augmentation program stored in the memory and executable on the processor, wherein when the knowledge-guided adaptive tunneling time-series data augmentation program is executed by the processor, it implements the steps of the knowledge-guided adaptive tunneling time-series data augmentation method as described above.

[0016] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a knowledge-guided adaptive tunneling time-series data augmentation program, which, when executed by a processor, implements the steps of the knowledge-guided adaptive tunneling time-series data augmentation method as described above.

[0017] In this invention, raw time-series data collected during tunnel construction is analyzed to construct a knowledge-guided objective function that quantitatively characterizes prediction delay and feature failure mechanisms. Subsequently, variational mode decomposition and controllable reconstruction are performed on the original sequence to obtain a preliminary enhanced sequence. Based on this, explicit constraints on distortion disturbances and prediction delays are introduced, and a multi-objective optimization strategy is adopted to minimize the sequence reconstruction error and lag order. The final enhanced sequence is obtained through iterative solution. Multi-source time-series data from the tunnel site are collected, and the enhanced sequence is input into the prediction model to obtain the predicted value for the next time step. Then, the actual value at that time step is obtained and concatenated with historical sequences. Enhancement and prediction are performed again on the updated sequence, and this process is repeated to achieve rolling enhancement and rolling prediction of the sequence. This invention provides an efficient and reliable data augmentation method for time-series prediction tasks in the tunnel field. Attached Figure Description

[0018] Figure 1 This is a flowchart of a preferred embodiment of the knowledge-guided adaptive tunneling temporal data augmentation method of the present invention; Figure 2 This is a technical principle diagram of the knowledge-guided adaptive tunneling time-series data augmentation method of this invention; Figure 3 This is a schematic diagram of the original TBM penetration sequence data collected during the mechanized tunnel excavation process, in a preferred embodiment of the knowledge-guided adaptive tunnel time-series data enhancement method of the present invention. Figure 4 This is a comparison diagram of the enhanced sequence and the original sequence in a preferred embodiment of the knowledge-guided adaptive tunneling time-series data augmentation method of the present invention. Figure 5 This is a structural diagram of a preferred embodiment of the knowledge-guided adaptive tunneling temporal data augmentation system of the present invention; Figure 6 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0019] This application provides a knowledge-guided adaptive tunneling time-series data augmentation method, system, and terminal. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0020] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0021] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0022] The knowledge-guided adaptive tunneling time-series data augmentation method described in the preferred embodiment of the present invention, such as... Figure 1 and Figure 2 As shown, the knowledge-guided adaptive tunneling time-series data augmentation method includes the following steps: Step S10: Analyze the original sequence data collected during the mechanized tunnel excavation process and construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term.

[0023] The process of parsing the raw sequence data collected during mechanized tunnel excavation and constructing a knowledge-driven objective function specifically includes: Based on the parsed original sequence data, Norm construction and reconstruction error term: ; in, This represents the reconstruction error term. This refers to the original sequence data. express Norm, This indicates preliminary augmented sequence data. Indicates time, Represents the ordinal number of the subsequence. Indicates the first Subsequence modal components; Based on the parsed original sequence data, and according to the dynamic drift pattern caused by the prediction delay, a lag order term characterizing the lag of the sequence response is constructed: ; ; in, This represents the lag order term. For cross-correlation coefficients, Indicates a time delay. Indicates the total length of the sequence. Indicates in The true value of a lightweight time series model at any given time. Indicates in Predictions from a lightweight time series model; By fusing the reconstruction error term and the lag order term, a knowledge-driven objective function is obtained to guide the sequence decomposition and reconstruction process: ; in, This represents the knowledge-driven objective function. This indicates a strong penalty.

[0024] like Figure 3 As shown, in this embodiment, the original sequence data is minimized. and reconstructed sequence The differences between them, building based on Norm reconstruction error term (white noise residual constraint term) ,in This is the result of superposition and reconstruction of the subsequence modal components.

[0025] Furthermore, based on the dynamic drift characteristics caused by prediction delays in tunnel construction systems, constraint terms characterizing the lag in sequence prediction response are constructed. By training a lightweight time series model, preliminary predictions are made on the reconstructed sequence, and the cross-correlation coefficients between the model predictions and actual values ​​under different lag orders are calculated. ,in and They are respectively and Real and predicted values ​​of a lightweight time series model at any given time. Let be the length of the sequence. Take the sequence with the maximum cross-correlation coefficient. As the lag order .

[0026] Understandably, when At that time, a stronger penalty term is applied to the objective function. This is used to enforce a lag order of 0, ensuring the temporal consistency of the enhanced sequence. In this embodiment... Set to 1×10 6 When applied to other embodiments, It can take any maximum value. The value will only be available in To play a role, The optimization process will not cause interference.

[0027] In this embodiment, a knowledge-driven objective function is constructed by parsing the raw sequence data collected during mechanized tunnel excavation. This objective function couples a reconstruction error term and a lag order term. The reconstruction error term constrains the difference between the raw and reconstructed sequences, while the lag order term quantifies the response lag of the sequence in subsequent prediction tasks. This step directly injects prior knowledge from the tunnel construction field (i.e., prediction delays can damage the usability of time-series data) into the mathematical constraints of data augmentation, making the subsequent decomposition and reconstruction processes no longer blind noise reduction but a clearly guided optimization process. By incorporating both the reconstruction error and the lag order into the objective function, dual guidance for noise suppression and delay correction is achieved. This solves the problem that traditional augmentation methods easily lead to over-smoothing or feature distortion when processing non-stationary tunnel time-series data, providing high-quality constraint criteria for subsequent steps.

[0028] Step S20: Perform variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure, and perform reconstruction processing based on the multiple subsequence mode components to obtain preliminary enhanced sequence data.

[0029] The step of performing variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure specifically includes: Based on the variational objective minimization form, an optimization problem is constructed to decompose the original sequence: ; in, Represents the modal function. Indicates the center frequency. express Norm, This represents the convolution operation. Indicates to Find the partial derivative. Let be the Dirac distribution function. Indicates time, Represents the ordinal number of the subsequence. Represents the imaginary unit. Represents the natural constant; An augmented Lagrange function is constructed by introducing a penalty factor and a Lagrange operator. Based on the augmented Lagrange function, the optimization problem is iteratively solved using the alternating direction multiplier method to obtain multiple narrowband mode components. These multiple narrowband mode components are then used as multiple subsequence mode components. The parameters of the variational mode decomposition include the number of modes and the penalty factor.

[0030] It is understood that the present invention performs variational mode decomposition on the original sequence, decomposing the original sequence into a set of components with the narrowest possible bandwidth and the clearest possible spectrum.

[0031] The decomposition process achieves optimal partitioning of modal components by minimizing the following variational objective: ; and by introducing a penalty factor With Lagrange operators An augmented Lagrangian form is constructed to obtain narrowband modal components. The variational optimization problem is solved using the alternating direction multiplier method, ultimately obtaining narrowband modal components with concentrated and non-overlapping spectra.

[0032] Furthermore, the process involves constructing an augmented Lagrangian function by introducing a penalty factor and a Lagrangian operator. Based on this augmented Lagrangian function, the optimization problem is iteratively solved using the alternating direction multiplier method to obtain multiple narrowband modal components. Specifically, this includes: Initialize the modal function set, center frequency set, and Lagrange operator, and set the initial number of iterations; In each iteration, each modal function, each center frequency, and the Lagrange operator are updated sequentially. After the update is completed, the change in modal function between two adjacent iterations is calculated. The update is based on minimizing the subproblems of the augmented Lagrange function. When the change in the modal function is less than a preset threshold or the number of iterations reaches a preset maximum value, the iteration loop is terminated, and the set of modal functions obtained from the final iteration is taken as the multiple narrowband modal components.

[0033] In this embodiment, the core solution process of variational mode decomposition is concretized into an executable iterative process: first, the set of modal functions, the set of center frequencies, and the Lagrange operator are initialized; then, an iterative loop is entered, in which each modal function, each center frequency, and the Lagrange operator are updated sequentially in each iteration, and the change in modal function between two adjacent iterations is calculated; when the change is less than a preset threshold or the maximum number of iterations is reached, the loop is terminated, and the final set of modal functions is taken as multiple narrowband modal components.

[0034] This step in the invention transforms the abstract variational optimization problem into an engineering-feasible iterative solution algorithm, ensuring that the spectra of each modal component are concentrated and non-overlapping, thus achieving effective separation of features at different time scales in the original sequence. By introducing a convergence threshold and a maximum number of iterations as termination conditions, the algorithm avoids unnecessary computational consumption while ensuring solution accuracy, thereby enhancing its robustness.

[0035] Furthermore, a reconstruction process is performed based on multiple sub-sequence modal components to obtain preliminary enhanced sequence data.

[0036] Specifically, in this embodiment, the raw data collected during the mechanized excavation process is decomposed into six different components. Five of these are core modal components, corresponding to the long-term trend characteristics, medium-period fluctuation characteristics, and short-period disturbance characteristics in the raw data, respectively. The remaining component is a residual component (mainly containing high-frequency noise and unpredictable interference). All six components have the same time length as the raw data.

[0037] Selective reconstruction processing was performed on the six subsequences obtained from variational mode decomposition. The reconstruction strategy of "discarding residuals and superimposing modes" was adopted to directly retain the effective mode components and superimpose and fuse them to obtain preliminary enhanced sequence data.

[0038] Understandably, the five modal components correspond to the long-term trend characteristics, medium-cycle construction parameter fluctuation characteristics, and short-cycle effective response characteristics in the original time-series data. Their spectra are concentrated and strongly correlated with the construction status, making them effective components reflecting the core patterns of the data. The residual component, however, mainly contains unpredictable components such as high-frequency sensor noise and instantaneous environmental interference; therefore, this residual component is directly discarded to avoid noise interference. The five effective modal components are directly superimposed along the original time dimension, and the reconstruction calculation is completed using this formula: ,in, For the first One effective modal component, To initially enhance the sequence data (reconstruct the sequence).

[0039] Step S30: Based on the preliminary enhanced sequence data, train a lightweight time series model and obtain the predicted values ​​of the test set. Calculate the lag order based on the actual and predicted values ​​of the test set.

[0040] Specifically, the preliminary enhanced sequence data is normalized to obtain a normalized sequence: ; in, This refers to the preliminary enhanced sequence data; The normalized sequence is reconstructed using a sliding window approach to generate a feature matrix and a corresponding target vector. ; in, Represents the feature matrix, Represents the target vector. This indicates a data reconstruction operation. Indicates the length of the sliding window; The feature matrix and the target vector are divided into a training set and a test set. A lightweight time series model is trained using the training set, and the trained model is used to predict the test set to obtain normalized prediction values. Perform an inverse normalization operation on the normalized predicted values ​​to restore them to the same data volume as the initial augmented sequence, resulting in inverse normalized predicted values: ; in, This represents the normalized predicted value of the inverse regression. This represents the normalized predicted value; Based on the true values ​​of the test set and the predicted values ​​of the inverse normalization, the cross-correlation coefficients at different time offsets are calculated, and the time offset corresponding to the maximum value of the cross-correlation coefficient is taken as the lag order.

[0041] In this embodiment, the reconstructed sequence after superposition... Perform normalization to eliminate the influence of dimensions: ; Furthermore, a sliding window is used (the length of the sliding window in this embodiment is...) (That is, predicting the value of the next minute based on the sequence data of the previous 0 minutes) The normalized sequence generates a feature matrix. and target vector : ,in The dimension is , The dimension is , This represents the total length of the sequence.

[0042] Among them, the normalized sequence The length is constructed in chronological order. Sliding window ( This represents the first data point in the sequence. This represents the second data point in the sequence. (This represents the Nth data point in the sequence). Each window corresponds to the input features, and the data at the next time step after the end of the window is used as the prediction target. The characteristics of each window are represented as follows: The corresponding target value is ,in, Indicates the first The first data point in the sliding window Indicates the first The second data point in the sliding window. Indicates the first The w-th data point in a sliding window.

[0043] The sliding window moves forward with a step size of 1, generating a total of A set of valid samples is used to form a feature matrix and a target vector: Feature matrix The dimension is ( , ), containing all input sequences; target vector The dimension is ( , ), which corresponds to the next time value of the target vector in all windows.

[0044] In a multidimensional sequence composed of multiple variables, all variables in each window are aligned by time and concatenated to form the complete feature vector of that window, so that the input data simultaneously contains multi-source mechanized excavation time series information such as penetration depth, advance speed, and cutterhead torque.

[0045] Furthermore, the feature matrix and target vector The dataset is divided into training, validation, and test sets in a 6:2:2 ratio to train a lightweight deep learning model, MLP, and predict the sequence after the window. And perform inverse normalization on the prediction results to restore the original data volume: ; Understandably, during model training, training parameters are set, and the optimizer is set to Adam (the optimizer's main function is to update the neural network parameters based on gradient information, thereby minimizing the loss function), the loss function is the mean squared error (MSE), the learning rate is 0.001, and the number of iterations is 100. After pre-training, a lightweight time series prediction model is obtained, which can initially achieve prediction of the target variable. The prediction.

[0046] Furthermore, regarding the aforementioned inverse normalized predicted values... and the true value Calculate different based on cross-correlation function Cross-correlation coefficients under ,Pick The maximum value corresponding to As the lag order .

[0047] Step S40: Using the L2 norm of the reconstruction error and the hysteresis order as optimization objectives, optimize the parameters obtained from the variational mode decomposition process to obtain the optimal sequence enhancement parameters.

[0048] The optimization of the parameters obtained from the variational mode decomposition process, using the L2 norm of the reconstruction error and the hysteresis order as optimization objectives, yields the optimal sequence enhancement parameters. Specifically, this includes: An initial population is randomly generated within a preset parameter space, which includes the range of values ​​for the number of modes and the range of values ​​for the penalty factor. Each individual in the initial population corresponds to a set of parameters for variational mode decomposition. Based on the parameters of the variational mode decomposition, the original sequence is subjected to variational mode decomposition and preliminary reconstruction, and the corresponding reconstruction error and lag order are calculated. When the hysteresis order is not zero, the reconstruction error is corrected to a maximum value; Based on the performance of the variational mode decomposition parameters on the two optimization objectives of reconstruction error and lag order, non-dominated sorting is performed on the individuals in the population to obtain the set of individuals at each non-dominated level. Based on the set of individuals at each non-dominated level, and combining ideal point sorting, normalization operation, and reference point mechanism, individuals that meet the optimization requirements are selected from the current population as retained solutions. Crossover and mutation operations are performed on the retained solutions to generate the next generation of candidate parameter population; the iterative process is repeated until the preset termination condition is met to obtain the target number of modes and the target penalty factor that achieves the optimal balance between reconstruction error and lag order, and the target number of modes and the target penalty factor are used as the optimal sequence enhancement parameters.

[0049] In this embodiment, the parameters of the variational mode decomposition include the number of modes and the penalty factor. A multi-objective optimization method is used to optimize the parameters of the variational mode decomposition. Adaptive optimization is performed, where, The parameters representing variational mode decomposition, The number of modal components. This is a penalty factor for variational mode decomposition, used to ensure that the decomposition parameters adaptively match the data characteristics and engineering requirements.

[0050] Specifically, an initial population is generated within a preset parameter space. , Within, 20 parameters to be optimized are randomly generated. This ensures the uniformity and coverage of the population; Step S52: For each parameter to be optimized Perform the above decomposition, reconstruction and The calculation process yields two optimization objectives: the reconstruction error objective. With delay correction target .when To strengthen the delay correction constraint, the reconstruction error target is reassigned. (In this embodiment) ),make The target is to eliminate them first.

[0051] Furthermore, in the multi-objective optimization process, candidate parameters are ranked according to dominance rules. For any two candidate parameters, if one parameter performs no worse than the other parameter across all optimization objectives, and performs better than the other parameter at least in one objective, then that parameter is considered to dominate the other parameter. Through non-dominated ranking, all candidate parameters are divided into non-dominated solution sets of different levels, where parameters with higher levels and stronger dominance relationships exhibit better overall performance across the optimization objectives.

[0052] Furthermore, based on the ideal point sorting method, the crowding distance of each non-dominated solution is calculated. Combining normalization and a reference point mechanism, 10 optimal candidate parameters are selected and retained for the next generation population. Crossover and mutation operations are then performed on the retained candidate parameters. Use integer crossover. Arithmetic crossover is used. The variable asynchronous length is 1. The variable length is 100. Through crossover and mutation operations, 10 new candidate parameters are generated, forming the next generation of a population with 20 candidate parameters.

[0053] Furthermore, the above iterative process is repeated until the optimal non-dominated solution of the population does not change significantly for 10 consecutive generations, or until the maximum number of iterations (50) is reached as specified in this embodiment. At this point, the iteration is stopped, and the optimal decomposition parameters are output. Based on optimal parameters Re-perform variational mode decomposition and selective reconstruction, discard residual components, and then superimpose the mode components to obtain the target augmentation sequence. .

[0054] In this embodiment, the optimal sequence enhancement parameters are obtained through multi-objective optimization. The corresponding optimization objective is , The results of comparing the enhanced sequence differences with the original sequence are as follows: Figure 4 As shown.

[0055] Step S50: Based on the optimal sequence enhancement parameters, perform variational mode decomposition and reconstruction on the original sequence again to obtain the target enhanced sequence.

[0056] Specifically, based on the optimal sequence enhancement parameters (including the optimal number of modes and the optimal penalty factor) obtained through multi-objective optimization, variational mode decomposition is re-performed on the original sequence. During the decomposition process, the optimal number of modes is used as the decomposition level, and the optimal penalty factor controls the bandwidth of each mode component. Multiple narrowband mode components with concentrated spectra and no overlap are obtained through iterative solution. Subsequently, selective reconstruction processing is performed on the obtained multiple mode components, discarding residual components containing high-frequency noise and unpredictable components, and superimposing and fusing the remaining effective mode components according to the original time dimension to obtain the target enhanced sequence.

[0057] Understandably, this invention, by introducing optimal decomposition parameters, ensures that the variational mode decomposition can adaptively adjust according to the characteristics of the original sequence, avoiding over-decomposition or under-decomposition caused by improper parameter selection. Based on this, by employing a reconstruction strategy that discards noise components and retains effective modes, it maximally suppresses high-frequency noise and random interference in the original sequence while fully preserving the core changing patterns reflecting the tunnel construction status. The resulting target augmented sequence exhibits a higher signal-to-noise ratio and stronger temporal consistency, thereby effectively improving prediction accuracy and generalization ability.

[0058] Step S60: When performing real-time prediction of tunnel time series data, the target enhancement sequence is input into a deep learning or time series prediction model for training, and the target value at the next time step is predicted. The target value is then output as the prediction result at the current time step.

[0059] Specifically, when performing real-time prediction of tunnel time series data, the target enhancement sequence generated above is first used as training data and input into a deep learning model or a time series prediction model; the model is then trained using the target enhancement sequence so that the model learns the temporal dependencies and change patterns in the sequence.

[0060] Furthermore, after model training is complete, historical data prior to the current moment is input into the trained model for inference, predicting the target variable value for the next moment, and outputting this predicted value as the prediction result for the current moment. By using the enhanced target sequence, which has undergone knowledge-guided augmentation, as the model input, the signal-to-noise ratio and temporal consistency of the training data are significantly improved, avoiding interference from high-frequency noise and prediction delay in the original data on model training.

[0061] Step S70: After obtaining the true observation values ​​of all variables at the next time step, the true observation values ​​are concatenated with the original sequence, the sequence data augmentation operation is performed again, and the target value prediction for the next time step is continued based on the updated sequence until the preset termination condition is reached, so as to achieve rolling prediction.

[0062] In this embodiment, after obtaining the true observation values ​​of all variables at the next time step, the true observation values ​​are first concatenated with the original historical sequence to form an updated sequence containing the latest data. Subsequently, based on the updated sequence, the sequence data augmentation operation is re-executed, that is, variational mode decomposition, lag order calculation, multi-objective parameter optimization, and target augmentation sequence generation are performed again. Finally, based on the updated target augmentation sequence, the prediction model is retrained and the target value is predicted at the next time step.

[0063] As can be seen, this invention dynamically updates the sequence by introducing real observations and iteratively executes the data augmentation and prediction process, enabling the entire system to have adaptive capabilities. As the tunnel construction process continues, newly acquired real data can be promptly integrated into the sequence, allowing the data augmentation operation to dynamically adjust the decomposition parameters based on the latest data characteristics, thereby maintaining the adaptability and effectiveness of the augmented sequence at all times.

[0064] Furthermore, when the knowledge-guided adaptive tunneling time-series data augmentation method is applied to target variable prediction in real-time tunneling, it further includes: The system acquires the continuous time-related parameters corresponding to the target variable in real time, performs data augmentation processing on the target variable to obtain the target augmentation sequence, and concatenates the target augmentation sequence with the continuous time-related parameter continuous time series data to obtain multidimensional time series data. The multidimensional time series data is normalized and segmented using a sliding window to obtain training samples. The training samples are then input into a deep learning or time series prediction model for training, and the predicted value of the target variable at the next time step is output. Once the actual observation values ​​of the target variable and other variables are obtained at the next time step, the actual observation values ​​are concatenated with the original dataset to form an updated sequence. Based on the updated sequence, data augmentation is performed again, and target value prediction continues for the next time step to achieve rolling prediction.

[0065] Understandably, when performing real-time prediction of tunnel time-series data, the optimized target enhancement sequence is fused with other associated time-series data (in this embodiment, the extracted associated time-series data includes TBM (Tunnel Boring Machine) cutterhead thrust, TBM cutterhead torque, TBM cutterhead rotation speed, and TBM advance speed), input into a deep learning / time series prediction model for training, and predicts the target value at the next moment, such as the TBM cutterhead penetration in this embodiment. In this embodiment, a temporal convolutional neural network (TCN) is selected for model training.

[0066] Specifically, the target enhancement sequence is concatenated with other associated time-series data (TBM cutterhead thrust, TBM cutterhead torque, TBM cutterhead rotation speed, TBM propulsion speed) into a multi-dimensional time-series data matrix. The matrix data is then normalized within the range [0,1]. A sliding window is used to divide the matrix data into training samples. The feature dimension of the training samples is (10, 4), and the target dimension is (1, 0).

[0067] Next, the training samples were input into the TCN model. The optimal hyperparameters of the TCN model were determined using a grid search method. The optimal hyperparameters of the optimized TCN model were: kernel size of 3, hidden layer dimension of 64×64×32, and dropout rate of 0.2. During training, there were 100 training epochs, a batch size of 128, and the Adam optimizer was selected with a learning rate of 0.001. After training, the latest 10 minutes of multidimensional time-series data were input, and the TBM cutterhead penetration value for the next time step was output.

[0068] Furthermore, once the true observation values ​​of all variables for the next time step are obtained, they are concatenated with the original sequence, the sequence length is updated, one data point is added for each iteration, and the sequence data augmentation operation described above is re-executed to generate a new target augmented sequence. Based on the updated augmented sequence and the multidimensional sequence, the target value prediction for the next time step continues, and real-time rolling prediction is achieved through iterative loops.

[0069] As can be seen, this invention achieves noise suppression and delay correction by constructing a knowledge-driven objective function that integrates reconstruction error and prediction delay, injecting time series domain knowledge into the data augmentation process. It also solves the problem of traditional methods relying on experience for parameter selection by adaptively solving decomposition parameters through multi-objective optimization. Furthermore, the real-time rolling prediction mechanism ensures that the augmented data dynamically matches the temporal characteristics changes of the construction process. The resulting high-quality augmented data effectively improves the accuracy and generalization ability of downstream prediction models, providing reliable technical support for safety monitoring and risk early warning in tunnel construction.

[0070] The present invention has the following beneficial effects: (1) By injecting domain features into the enhanced process through knowledge-driven objective functions, high-frequency noise is effectively suppressed and prediction delay is corrected, thereby enhancing the temporal consistency and availability of data.

[0071] (2) Adaptive solution of decomposition parameters is achieved by using multi-objective optimization, which can match data characteristics and engineering requirements without manual intervention, thus improving the versatility and robustness of the method.

[0072] (3) It provides a reliable training basis for the prediction model, and combined with the real-time rolling prediction mechanism, it significantly improves the prediction accuracy and generalization ability of key targets in tunnel construction. It can support the adjustment of construction parameters and risk warning, and reduce safety hazards.

[0073] (4) Construct a feature matrix through a sliding window to mine the spatiotemporal correlation between parameters. The method is suitable for the monitoring scenario of mechanized tunnel construction, is simple to operate, highly practical, and can be directly integrated into the real-time monitoring system.

[0074] Furthermore, such as Figure 5 As shown, based on the above-mentioned knowledge-guided adaptive tunneling time-series data augmentation method, the present invention also provides a knowledge-guided adaptive tunneling time-series data augmentation system, wherein the knowledge-guided adaptive tunneling time-series data augmentation system includes: The objective function construction module 51 is used to parse the original sequence data collected during the mechanized tunnel excavation process and construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term. The decomposition and reconstruction module 52 is used to perform variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the dependency structure at different time scales, and to perform reconstruction processing based on the multiple subsequence mode components to obtain preliminary enhanced sequence data; The lag order calculation module 53 is used to train a lightweight time series model and obtain the predicted values ​​of the test set based on the preliminary enhanced sequence data, and to calculate the lag order based on the actual values ​​and predicted values ​​of the test set. The multi-objective parameter optimization module 54 is used to optimize the parameters obtained by the variational mode decomposition process by taking the L2 norm of the reconstruction error and the hysteresis order as optimization objectives to obtain the optimal sequence enhancement parameters. The target enhancement sequence generation module 55 is used to re-perform variational mode decomposition and reconstruction on the original sequence based on the optimal sequence enhancement parameters to obtain the target enhancement sequence; The real-time prediction module 56 is used to input the target enhancement sequence into a deep learning or time series prediction model for training when performing real-time prediction of tunnel time series data, predict the target value at the next moment, and output the target value as the prediction result at the current moment. The rolling update module 57 is used to, after obtaining the true observation values ​​of all variables at the next time step, concatenate the true observation values ​​with the original sequence, perform the sequence data augmentation operation again, and continue to predict the target value at the next time step based on the updated sequence until the preset termination condition is reached, so as to realize rolling prediction.

[0075] Furthermore, such as Figure 6 As shown, based on the above-mentioned knowledge-guided adaptive tunneling time-series data enhancement method and system, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 6 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0076] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a knowledge-guided adaptive tunneling time-series data enhancement program 40, which can be executed by the processor 10 to implement the knowledge-guided adaptive tunneling time-series data enhancement method of this application.

[0077] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the knowledge-guided adaptive tunneling time-series data augmentation method.

[0078] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0079] In one embodiment, when the processor 10 executes the knowledge-guided adaptive tunneling timing data enhancement program 40 in the memory 20, the following steps are performed: The original sequence data collected during mechanized tunnel excavation is analyzed to construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term. Variational mode decomposition is performed on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure, and reconstruction processing is performed based on the multiple subsequence mode components to obtain preliminary enhanced sequence data; Based on the preliminary augmented sequence data, a lightweight time series model is trained and the predicted values ​​of the test set are obtained. Based on the actual values ​​and predicted values ​​of the test set, the lag order is calculated. The L2 norm of the reconstruction error and the hysteresis order are used as optimization objectives to optimize the parameters obtained by the variational mode decomposition process, thereby obtaining the optimal sequence enhancement parameters. Based on the optimal sequence enhancement parameters, variational mode decomposition and reconstruction are performed on the original sequence again to obtain the target enhanced sequence; When performing real-time prediction of tunnel time series data, the target enhancement sequence is input into a deep learning or time series prediction model for training, and the target value at the next moment is predicted. The target value is then output as the prediction result at the current moment. Once the true observation values ​​of all variables at the next time step are obtained, the true observation values ​​are concatenated with the original sequence, the sequence data augmentation operation is performed again, and the target value prediction for the next time step is continued based on the updated sequence until the preset termination condition is reached, so as to achieve rolling prediction.

[0080] Specifically, the process of parsing the raw sequence data collected during mechanized tunnel excavation and constructing a knowledge-driven objective function includes: Based on the parsed original sequence data, Norm construction and reconstruction error term: ; in, This represents the reconstruction error term. This refers to the original sequence data. express Norm, This indicates preliminary augmented sequence data. Indicates time, Represents the ordinal number of the subsequence. Indicates the first Subsequence modal components; Based on the parsed original sequence data, and according to the dynamic drift pattern caused by the prediction delay, a lag order term characterizing the lag of the sequence response is constructed: ; ; in, This represents the lag order term. For cross-correlation coefficients, Indicates a time delay. Indicates the total length of the sequence. Indicates in The true value of a lightweight time series model at any given time. Indicates in Predictions from a lightweight time series model; By fusing the reconstruction error term and the lag order term, a knowledge-driven objective function is obtained to guide the sequence decomposition and reconstruction process: ; in, This represents the knowledge-driven objective function. This indicates a strong penalty.

[0081] Specifically, performing variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure includes: Based on the variational objective minimization form, an optimization problem is constructed to decompose the original sequence: ; in, Represents the modal function. Indicates the center frequency. express Norm, This represents the convolution operation. Indicates to Find the partial derivative. Let be the Dirac distribution function. Indicates time, Represents the ordinal number of the subsequence. Represents the imaginary unit. Represents the natural constant; An augmented Lagrange function is constructed by introducing a penalty factor and a Lagrange operator. Based on the augmented Lagrange function, the optimization problem is iteratively solved using the alternating direction multiplier method to obtain multiple narrowband mode components. These multiple narrowband mode components are then used as multiple subsequence mode components. The parameters of the variational mode decomposition include the number of modes and the penalty factor.

[0082] Specifically, the process involves constructing an augmented Lagrangian function by introducing a penalty factor and a Lagrangian operator, and then iteratively solving the optimization problem using the alternating direction multiplier method based on the augmented Lagrangian function to obtain multiple narrowband modal components. This includes: Initialize the modal function set, center frequency set, and Lagrange operator, and set the initial number of iterations; In each iteration, each modal function, each center frequency, and the Lagrange operator are updated sequentially. After the update is completed, the change in modal function between two adjacent iterations is calculated. The update is based on minimizing the subproblems of the augmented Lagrange function. When the change in the modal function is less than a preset threshold or the number of iterations reaches a preset maximum value, the iteration loop is terminated, and the set of modal functions obtained from the final iteration is taken as the multiple narrowband modal components.

[0083] Specifically, the step of training a lightweight time series model based on the preliminary enhanced sequence data and obtaining predicted values ​​for the test set, and calculating the lag order based on the actual and predicted values ​​of the test set, includes: The preliminary enhanced sequence data is normalized to obtain a normalized sequence: ; in, This refers to the preliminary enhanced sequence data; The normalized sequence is reconstructed using a sliding window approach to generate a feature matrix and a corresponding target vector. ; in, Represents the feature matrix, Represents the target vector. This indicates a data reconstruction operation. Indicates the length of the sliding window; The feature matrix and the target vector are divided into a training set and a test set. A lightweight time series model is trained using the training set, and the trained model is used to predict the test set to obtain normalized prediction values. Perform an inverse normalization operation on the normalized predicted values ​​to restore them to the same data volume as the initial augmented sequence, resulting in inverse normalized predicted values: ; in, This represents the normalized predicted value of the inverse regression. This represents the normalized predicted value; Based on the true values ​​of the test set and the predicted values ​​of the inverse normalization, the cross-correlation coefficients at different time offsets are calculated, and the time offset corresponding to the maximum value of the cross-correlation coefficient is taken as the lag order.

[0084] Specifically, the optimization of the parameters obtained from the variational mode decomposition process by using the L2 norm of the reconstruction error and the hysteresis order as optimization objectives to obtain the optimal sequence enhancement parameters includes: An initial population is randomly generated within a preset parameter space, which includes the range of values ​​for the number of modes and the range of values ​​for the penalty factor. Each individual in the initial population corresponds to a set of parameters for variational mode decomposition. Based on the parameters of the variational mode decomposition, the original sequence is subjected to variational mode decomposition and preliminary reconstruction, and the corresponding reconstruction error and lag order are calculated. When the hysteresis order is not zero, the reconstruction error is corrected to a maximum value; Based on the performance of the variational mode decomposition parameters on the two optimization objectives of reconstruction error and lag order, non-dominated sorting is performed on the individuals in the population to obtain the set of individuals at each non-dominated level. Based on the set of individuals at each non-dominated level, and combining ideal point sorting, normalization operation, and reference point mechanism, individuals that meet the optimization requirements are selected from the current population as retained solutions. Crossover and mutation operations are performed on the retained solutions to generate the next generation of candidate parameter population; the iterative process is repeated until the preset termination condition is met to obtain the target number of modes and the target penalty factor that achieves the optimal balance between reconstruction error and lag order, and the target number of modes and the target penalty factor are used as the optimal sequence enhancement parameters.

[0085] When the knowledge-guided adaptive tunneling time-series data augmentation method is applied to target variable prediction in real-time tunneling, it further includes: The system acquires the continuous time-related parameters corresponding to the target variable in real time, performs data augmentation processing on the target variable to obtain the target augmentation sequence, and concatenates the target augmentation sequence with the continuous time-related parameter continuous time series data to obtain multidimensional time series data. The multidimensional time series data is normalized and segmented using a sliding window to obtain training samples. The training samples are then input into a deep learning or time series prediction model for training, and the predicted value of the target variable at the next time step is output. Once the actual observation values ​​of the target variable and other variables are obtained at the next time step, the actual observation values ​​are concatenated with the original dataset to form an updated sequence. Based on the updated sequence, data augmentation is performed again, and target value prediction continues for the next time step to achieve rolling prediction.

[0086] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a knowledge-guided adaptive tunneling time-series data augmentation program, which, when executed by a processor, implements the steps of the knowledge-guided adaptive tunneling time-series data augmentation method as described above.

[0087] In summary, this invention proposes a knowledge-guided adaptive tunnel time-series data augmentation method, system, terminal, and storage medium. The method includes: parsing the raw sequence data collected during mechanized tunnel excavation to construct a knowledge-driven objective function characterizing the formation mechanism of prediction delay and feature failure; and based on the constructed knowledge-driven objective function, performing variational mode decomposition processing on the raw sequence to obtain data describing the dependency structure at different time scales. The data augmentation process involves reconstructing multiple subsequence modal components to obtain a preliminary augmented sequence. A constraint perturbation term is constructed for this preliminary augmented sequence, and a multi-objective optimization method is used to optimize the augmentation strategy, eliminating white noise residuals and unpredictable components in the original sequence and enhancing its predictability. When making predictions based on the augmented sequence, the augmented target sequence is input into a deep learning or time series prediction model to obtain the prediction result for the next time step. Once the true observation values ​​of all variables for the next time step are obtained, they are concatenated with the original sequence, and data augmentation is re-executed. The target value for the next time step is then predicted based on the updated sequence, and this process is iterated to achieve rolling prediction. This invention constructs a knowledge-driven objective function that integrates reconstruction error and prediction delay, injecting knowledge from the tunnel construction field into the data augmentation process. This achieves explicit constraints on noise suppression and delay correction, solving the problem of traditional methods lacking domain specificity and improving the usability of tunnel time series data.

[0088] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0089] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0090] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A knowledge-guided adaptive tunneling time-series data augmentation method, characterized in that, The knowledge-guided adaptive tunneling time-series data augmentation method includes: The original sequence data collected during mechanized tunnel excavation is analyzed to construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term. Variational mode decomposition is performed on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure, and reconstruction processing is performed based on the multiple subsequence mode components to obtain preliminary enhanced sequence data; Based on the preliminary augmented sequence data, a lightweight time series model is trained and the predicted values ​​of the test set are obtained. Based on the actual values ​​and predicted values ​​of the test set, the lag order is calculated. The L2 norm of the reconstruction error and the hysteresis order are used as optimization objectives to optimize the parameters obtained by the variational mode decomposition process, thereby obtaining the optimal sequence enhancement parameters. Based on the optimal sequence enhancement parameters, variational mode decomposition and reconstruction are performed on the original sequence again to obtain the target enhanced sequence; When performing real-time prediction of tunnel time series data, the target enhancement sequence is input into a deep learning or time series prediction model for training, and the target value at the next moment is predicted. The target value is then output as the prediction result at the current moment. Once the true observation values ​​of all variables at the next time step are obtained, the true observation values ​​are concatenated with the original sequence, the sequence data augmentation operation is performed again, and the target value prediction for the next time step is continued based on the updated sequence until the preset termination condition is reached, so as to achieve rolling prediction. The process of parsing the raw sequence data collected during mechanized tunnel excavation and constructing a knowledge-driven objective function specifically includes: Based on the parsed original sequence data, Norm construction and reconstruction error term: ； in, This represents the reconstruction error term. This refers to the original sequence data. express Norm, This indicates preliminary augmented sequence data. Indicates time, Represents the ordinal number of the subsequence. Indicates the first Subsequence modal components; Based on the parsed original sequence data, and according to the dynamic drift pattern caused by the prediction delay, a lag order term characterizing the lag of the sequence response is constructed: ； ； in, This represents the lag order term. For cross-correlation coefficients, Indicates a time delay. Indicates the total length of the sequence. Indicates in The true value of a lightweight time series model at any given time. Indicates in Predictions from a lightweight time series model; By fusing the reconstruction error term and the lag order term, a knowledge-driven objective function is obtained to guide the sequence decomposition and reconstruction process: ； in, This represents the knowledge-driven objective function. This indicates a strong penalty.

2. The knowledge-guided adaptive tunneling time-series data augmentation method according to claim 1, characterized in that, The step of performing variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the time-scale dependency structure specifically includes: Based on the variational objective minimization form, an optimization problem is constructed to decompose the original sequence: ； in, Represents the modal function. Indicates the center frequency. express Norm, This represents the convolution operation. Indicates to Find the partial derivative. Let be the Dirac distribution function. Indicates time, Represents the ordinal number of the subsequence. Represents the imaginary unit. Represents the natural constant; An augmented Lagrange function is constructed by introducing a penalty factor and a Lagrange operator. Based on the augmented Lagrange function, the optimization problem is iteratively solved using the alternating direction multiplier method to obtain multiple narrowband mode components. These multiple narrowband mode components are then used as multiple subsequence mode components. The parameters of the variational mode decomposition include the number of modes and the penalty factor.

3. The knowledge-guided adaptive tunneling time-series data augmentation method according to claim 2, characterized in that, The process involves constructing an augmented Lagrangian function by introducing a penalty factor and a Lagrangian operator. Based on this augmented Lagrangian function, the optimization problem is iteratively solved using the alternating direction multiplier method to obtain multiple narrowband modal components, specifically including: Initialize the modal function set, center frequency set, and Lagrange operator, and set the initial number of iterations; In each iteration, each modal function, each center frequency, and the Lagrange operator are updated sequentially. After the update is completed, the change in modal function between two adjacent iterations is calculated. The update is based on minimizing the subproblems of the augmented Lagrange function. When the change in the modal function is less than a preset threshold or the number of iterations reaches a preset maximum value, the iteration loop is terminated, and the set of modal functions obtained from the final iteration is taken as the multiple narrowband modal components.

4. The knowledge-guided adaptive tunneling time-series data augmentation method according to claim 1, characterized in that, The step of training a lightweight time series model based on the preliminary enhanced sequence data and obtaining predicted values ​​for the test set, and calculating the lag order based on the actual and predicted values ​​of the test set, specifically includes: The preliminary enhanced sequence data is normalized to obtain a normalized sequence: ； in, This refers to the preliminary enhanced sequence data; The normalized sequence is reconstructed using a sliding window approach to generate a feature matrix and a corresponding target vector. ； in, Represents the feature matrix, Represents the target vector. This indicates a data reconstruction operation. Indicates the length of the sliding window; The feature matrix and the target vector are divided into a training set and a test set. A lightweight time series model is trained using the training set, and the trained model is used to predict the test set to obtain normalized prediction values. Perform an inverse normalization operation on the normalized predicted values ​​to restore them to the same data volume as the initial augmented sequence, resulting in inverse normalized predicted values: ； in, This represents the predicted value after inversion. This represents the normalized predicted value; Based on the true values ​​of the test set and the predicted values ​​of the inverse normalization, the cross-correlation coefficients at different time offsets are calculated, and the time offset corresponding to the maximum value of the cross-correlation coefficient is taken as the lag order.

5. The knowledge-guided adaptive tunneling time-series data augmentation method according to claim 1, characterized in that, The optimization of the parameters obtained from the variational mode decomposition process, using the L2 norm of the reconstruction error and the hysteresis order as optimization objectives, yields the optimal sequence enhancement parameters. Specifically, this includes: An initial population is randomly generated within a preset parameter space, which includes the range of values ​​for the number of modes and the range of values ​​for the penalty factor. Each individual in the initial population corresponds to a set of parameters for variational mode decomposition. Based on the parameters of the variational mode decomposition, the original sequence is subjected to variational mode decomposition and preliminary reconstruction, and the corresponding reconstruction error and lag order are calculated. When the hysteresis order is not zero, the reconstruction error is corrected to a maximum value; Based on the performance of the variational mode decomposition parameters on the two optimization objectives of reconstruction error and lag order, non-dominated sorting is performed on the individuals in the population to obtain the set of individuals at each non-dominated level. Based on the set of individuals at each non-dominated level, and combining ideal point sorting, normalization operation, and reference point mechanism, individuals that meet the optimization requirements are selected from the current population as retained solutions. Crossover and mutation operations are performed on the retained solutions to generate the next generation of candidate parameter population; the iterative process is repeated until the preset termination condition is met to obtain the target number of modes and the target penalty factor that achieves the optimal balance between reconstruction error and lag order, and the target number of modes and the target penalty factor are used as the optimal sequence enhancement parameters.

6. The knowledge-guided adaptive tunneling time-series data augmentation method according to claim 1, characterized in that, When the knowledge-guided adaptive tunneling time-series data augmentation method is applied to target variable prediction in real-time tunneling, it further includes: The system acquires the continuous time-related parameters corresponding to the target variable in real time, performs data augmentation processing on the target variable to obtain the target augmentation sequence, and concatenates the target augmentation sequence with the continuous time-related parameter continuous time series data to obtain multidimensional time series data. The multidimensional time series data is normalized and segmented using a sliding window to obtain training samples. The training samples are then input into a deep learning or time series prediction model for training, and the predicted value of the target variable at the next time step is output. Once the actual observation values ​​of the target variable and other variables are obtained at the next time step, the actual observation values ​​are concatenated with the original dataset to form an updated sequence. Based on the updated sequence, data augmentation is performed again, and target value prediction continues for the next time step to achieve rolling prediction.

7. A knowledge-guided adaptive tunneling time-series data augmentation system, characterized in that, The knowledge-guided adaptive tunneling time-series data augmentation system is used to implement the knowledge-guided adaptive tunneling time-series data augmentation method according to any one of claims 1-6, wherein the knowledge-guided adaptive tunneling time-series data augmentation system comprises: The objective function construction module is used to parse the raw sequence data collected during the mechanized tunnel excavation process and construct a knowledge-driven objective function. The knowledge-driven objective function guides noise residual suppression and delay correction in the time series enhancement process by coupling the reconstruction error term and the lag order term. The decomposition and reconstruction module is used to perform variational mode decomposition on the original sequence to obtain multiple subsequence mode components describing the dependency structure at different time scales, and to perform reconstruction processing based on the multiple subsequence mode components to obtain preliminary enhanced sequence data; The lag order calculation module is used to train a lightweight time series model based on the preliminary augmented sequence data and obtain the predicted values ​​of the test set, and calculate the lag order based on the actual values ​​and predicted values ​​of the test set. The multi-objective parameter optimization module is used to optimize the parameters obtained by the variational mode decomposition process by taking the L2 norm of the reconstruction error and the hysteresis order as optimization objectives to obtain the optimal sequence enhancement parameters. The target augmented sequence generation module is used to re-perform variational mode decomposition and reconstruction on the original sequence based on the optimal sequence augmentation parameters to obtain the target augmented sequence; The real-time prediction module is used to input the target enhancement sequence into a deep learning or time series prediction model for training when making real-time predictions of tunnel time series data, predict the target value at the next time step, and output the target value as the prediction result at the current time step. The rolling update module is used to concatenate the real observation values ​​of all variables with the original sequence after obtaining the real observation values ​​of the next time step, perform the sequence data augmentation operation again, and continue to predict the target value of the next time step based on the updated sequence until the preset termination condition is reached, so as to realize rolling prediction.

8. A terminal, characterized in that, The terminal includes: a memory, a processor, and a knowledge-guided adaptive tunneling time-series data augmentation program stored in the memory and executable on the processor. When the knowledge-guided adaptive tunneling time-series data augmentation program is executed by the processor, it implements the steps of the knowledge-guided adaptive tunneling time-series data augmentation method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a knowledge-guided adaptive tunneling time-series data augmentation program, which, when executed by a processor, implements the steps of the knowledge-guided adaptive tunneling time-series data augmentation method as described in any one of claims 1-6.

Images