A laser vibration measurement vibration information noise processing method, system and device for a complex tunnel environment

Abstract

The present application relates to the technical field of tunnel vibration signal noise reduction, and particularly relates to a laser vibration information noise processing method, system and equipment for tunnel complex environment, and proposes a two-stage progressive noise reduction scheme: the initial noise reduction utilizes the ant lion optimization algorithm to adaptively optimize the modal number and penalty parameter of the variational mode decomposition, so as to obtain the optimal parameters with the minimum weighted sum of the reconstruction error and modal sparsity as the target; the original signal is decomposed accordingly, the noise-dominant IMF component is removed based on the Pearson correlation coefficient, the effective component is retained for reconstruction, the coarse noise reduction is realized; the secondary noise reduction takes the initial noise reduction signal as a training sample, adopts the L1 regularization improved K-SVD dictionary learning to train an over-complete dictionary; the initial noise reduction signal is sparsely represented and reconstructed by using the dictionary, the residual noise is further stripped, and the final pure vibration signal is output. The present application can improve the signal-to-noise ratio and fidelity of the tunnel weak vibration signal through adaptive parameter matching and complementary enhancement of sparse dictionary learning.

Description

Technical Field

[0001] This invention relates to the field of tunnel vibration signal noise reduction technology, and in particular to a method, system and equipment for processing laser vibration information noise in complex tunnel environments. Background Technology

[0002] In the complex environment of tunnel construction, the surrounding rock surface is often unstable, making it difficult to reliably install traditional contact sensors (such as accelerometers), which are prone to detachment or damage. Although scanning laser vibration meters have advantages such as non-contact operation, high precision, and full-field monitoring, the use of non-contact vibration monitoring equipment in tunnels is subject to many limitations due to the special nature of the complex construction environment of tunnels and the inherent limitations of laser vibration measurement technology.

[0003] Vibration signals acquired by laser vibrometers contain high-frequency noise (optical speckle, electronic noise), low-frequency environmental interference (ground vibration), and aliasing of signals of the same frequency. While single-signal noise processing methods can filter and denoise vibration signals to some extent, each has its limitations: Fourier transform is suitable for filtering and denoising periodic signals, but its effect is poor when dealing with highly random and non-stationary signals; wavelet transform can provide analytical resolution for different frequency bands, but its denoising effect depends on the choice of threshold function; Empirical Mode Decomposition (EMD) lacks systematic mathematical theoretical support, resulting in poor algorithm stability when dealing with signal sampling rate and noise fluctuations; Traditional Variational Mode Decomposition (VMD) requires manual setting of the number of modes K and bandwidth parameters, which can easily lead to incomplete mode separation or the introduction of redundant modes; Traditional K-SVD algorithm uses OMP to solve sparse coefficients based on L0 norm constraints, which suffers from problems such as L0 norm nonconvexity, easy getting trapped in local optima, and overfitting.

[0004] Therefore, this invention proposes a method, system, and device for processing vibration information noise from laser vibration measurement in complex tunnel environments to solve the above-mentioned problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method, system, and device for noise processing of laser vibration measurement information in complex tunnel environments. This invention integrates the advantages of adaptive variational mode decomposition and sparse representation learning dictionary analysis. Through complementary enhancement, multi-dimensional analysis, and robustness improvement, it achieves noise separation and signal enhancement, significantly improving the signal-to-noise ratio and accuracy of micro-vibration measurement.

[0006] The technical solution of this invention is a noise processing method for laser vibration measurement information in complex tunnel environments. It involves acquiring the original vibration signal from a laser vibration measurement in the complex tunnel environment and includes a two-stage noise reduction process: initial noise reduction and secondary noise reduction. The specific steps are as follows: (1) Initial noise reduction processing: Typical vibration signals were selected from the original vibration signals as parameter optimization samples. Using these samples as input, the Antlion Algorithm (ALO) was applied to optimize the mode number of Variational Mode Decomposition (VMD). and penalty parameters Perform adaptive optimization and output the optimal parameter combination. , To be the optimal number of modes, The optimal penalty parameter; Use the optimal parameter combination Variational mode decomposition was performed on the original vibration signal to obtain Each intrinsic mode function (IMF) modal component, and the center frequency corresponding to each IMF modal component; Calculate the Pearson correlation coefficient between each IMF modal component and the original vibration signal, retain the effective information components whose absolute value of the correlation coefficient is greater than or equal to a preset threshold, and reconstruct all effective information components by superimposing them in the time domain to obtain the initial noise-reduced signal. (2) Secondary noise reduction processing: Typical vibration signals are selected from the initial noise reduction signal as dictionary training samples. The dictionary training samples are input into the improved K-SVD dictionary learning model, and an optimized overcomplete dictionary is obtained by training based on L1 regularized sparse representation. The initial noise reduction signal is input into an optimized and complete dictionary. The second noise reduction is completed through sparse representation solution and signal reconstruction, and the final noise-filtered pure vibration signal is output.

[0007] The specific process of adaptive optimization in the initial noise reduction process is as follows: Define the optimization objective, in terms of modal number. and penalty parameters Let the variable to be optimized be an objective function. The optimization objective is to minimize the weighted sum of signal reconstruction error and modal sparsity. The expression of the objective function is: , in, express The corresponding fitness value, This indicates that the input parameters optimize the sample signal. The decomposition yields the first... indivual Modal components, Indices representing modal components; Represents the L2 norm; It is an L1 norm; Indicates the balancing weight coefficient; Set the population parameters for the ALO algorithm, including the ant population size. Antlion population size Maximum number of iterations Set the boundary values ​​for the optimization variables, and the number of modes. It is a positive integer, and its value range is 1. Penalty parameters It is a positive real number, and its range is 1. Set the initial search step size factor Convergence criterion threshold , , Representing the number of modes respectively The minimum and maximum values ​​are taken. , These represent the penalty parameters. The minimum and maximum values ​​can be obtained. Randomly generated within the range of variable values. Group of ant positions, each group of ant positions corresponds to a group Parameter combination; selecting from all ant locations The position with the best fitness in the group is taken as the initial antlion position; fitness is determined by the objective function value, the smaller the objective function value, the better the fitness; by For the current iteration number, execute The loop iteration, Indicates the maximum number of iterations; After the iteration terminates, the parameter combination corresponding to the globally optimal antlion is taken as the optimal parameter combination. Output.

[0008] The iterative optimization process is as follows: Each ant updates its position by randomly circling its corresponding antlion, using the following formula: , in, For the first The ant's position in the next iteration. For the first The globally optimal antlion position in the next iteration. For the first The location of the antlion is randomly selected in the next iteration. , A uniformly random number within the interval [0,1]. For the first The search step size factor for the next iteration; The search step size decreases linearly with the number of iterations, gradually shifting from a global search to a local fine-grained search. The step size update formula is: ; Parameters that exceed the range after the update are truncated and corrected, and the modal number is adjusted accordingly. Limited to Within the range and rounded down, the penalty parameter is... Limited to Within the range; For each ant The parameter combination is used to perform VMD decomposition, and the corresponding fitness value is calculated by substituting it into the objective function. Replace the ant position with the corresponding antlion position with the better fitness, while retaining the globally optimal antlion position as the historical optimal solution. If the improvement in the objective function value for a consecutive preset number of iterations is less than the convergence threshold If the iteration is terminated prematurely, then the iteration will be terminated early.

[0009] The process of solving the initial noise reduction signal is as follows: Using the original vibration signal as input, the optimal parameter combination is employed. Perform variational mode decomposition to decompose the original signal into... A finite bandwidth IMF modal component And simultaneously obtain the center frequency corresponding to each IMF component. ; Calculate the first One IMF modal component Compared with the original vibration signal Pearson correlation coefficient The calculation formula is: , in, Represents the mathematical expectation operator. Represents the variance operator; Set the threshold for Pearson correlation coefficient ,like If the condition is met, the corresponding IMF component is a valid information component. IMF components that do not meet this condition are judged as noise-dominant components and are directly removed. By superimposing all effective information components point-by-point in the time domain, the initial denoised signal can be reconstructed. The first noise mode removal was completed.

[0010] In the secondary noise reduction process, the K-SVD dictionary learning model is improved by replacing the L0 norm sparsity constraint in the traditional K-SVD algorithm with L1 regularization Lasso. The improved K-SVD global optimization objective function is: , in, For the training sample matrix, The sampling length of a single vibration signal. This represents the total number of training samples; For the purpose of learning a complete dictionary, The number of atoms in the dictionary. ; It is a sparse matrix. For the first The sparse coefficient vector corresponding to each training sample The index for the training samples; It is the Frobenius norm; It is an L1 norm; The regularization coefficient is used. This is an operation to find the minimum value.

[0011] The training steps for the improved K-SVD dictionary learning model are as follows: Step 1: Construct the dictionary training sample matrix Y, and randomly select samples from the training samples. 1 sample as the initial dictionary atom For each atom, L2 norm normalization is performed to obtain an initial dictionary. The normalization formula is: , ; in, Represents the first in the dictionary One atom; Step 2: Fix the current dictionary and apply it to each training sample. Solve the L1-regularized Lasso optimization problem to obtain the corresponding sparse coefficients. The single-sample optimization problem is expressed as: , in, For the first training samples, Let be the sparse coefficient vector to be solved; The coordinate descent method is used to solve the above convex optimization problem. After traversing all training samples, the complete sparse coefficient matrix is ​​obtained. ; Step 3: Fix the sparse coefficient matrix Update the dictionary atoms column by column; for the first column... dictionary atoms Perform the following operations: 1) Fixed use of the first A set of sample indices of atoms , For the first The first sample sparse coefficient vector of the th sample One element; 2) Calculate the global error matrix and remove the first error. The contribution of each atom yields the corresponding residual matrix. : , in, For atomic indexing, Represents the first in the dictionary One atom, Represents the sparse coefficient matrix The Middle Transpose of a row vector; 3) From the residual matrix Extract the index set The corresponding columns yield a submatrix. ,right Perform Singular Value Decomposition (SVD): , in, It is a left singular matrix. It is a singular value diagonal matrix. It is a right singular matrix. Indicates the transpose operation; 4) Take The first column is the updated number dictionary atoms ,Pick The first singular value and The product of the first column is used as the corresponding sparse coefficient row to complete the optimal update of a single atom; Step 4: Perform L2 norm normalization on the updated dictionary atoms and adjust the magnitude of the sparse coefficients accordingly; Step 5: Repeat steps 2 to 4 until the change in global reconstruction error before and after dictionary update is less than the preset convergence threshold, or the preset maximum number of iterations is reached, and output the optimized overcomplete dictionary D after training.

[0012] The method also includes evaluating the denoised signal, with evaluation metrics including signal-to-noise ratio (SNR) and root mean square error (RMSE). The formula for calculating the signal-to-noise ratio (SNR) is as follows: ; The formula for calculating the root mean square error (RMSE) is as follows: ; in, The first denoised signal is the denoised signal. One sampling point, The first reference pure signal One sampling point, This represents the total number of sampling points for the signal. For sampling point index; A higher signal-to-noise ratio and a smaller root mean square error indicate a better noise reduction effect.

[0013] This invention also provides a laser vibration measurement vibration information noise processing system for complex tunnel environments, which performs a laser vibration measurement vibration information noise processing method for complex tunnel environments, including: The signal acquisition module is configured to acquire the original vibration signal from the laser vibration measurement in the complex environment of the tunnel, and select typical vibration signals from the original vibration signal as parameter optimization samples and dictionary training samples, respectively. The parameter optimization module is configured to receive parameter optimization samples, use the antlion optimization algorithm to adaptively optimize the number of modes and penalty parameters of variational mode decomposition, and output the optimal parameter combination. The decomposition and reconstruction module is configured to receive the original vibration signal and the optimal parameter combination, perform adaptive variational mode decomposition, screen the effective IMF components through the correlation coefficient and reconstruct the signal, and output the initial noise-reduced signal. The secondary noise reduction module is configured to receive dictionary training samples and the initial noise reduction signal, first train and optimize the overcomplete dictionary based on the improved K-SVD model, and then perform secondary noise reduction on the initial noise reduction signal based on L1 regularized sparse representation to output the final clean vibration signal. The evaluation output module is configured to calculate the signal-to-noise ratio and root mean square error of the noise reduction effect, and output the final clean vibration signal and quantitative evaluation results.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements steps such as those of a laser vibration measurement vibration information noise processing method for complex tunnel environments.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements steps such as those of a laser vibration measurement vibration information noise processing method for complex tunnel environments.

[0016] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. The above technical solutions have the following advantages or beneficial effects: This invention discloses a method, system, and device for noise processing of laser vibration information in complex tunnel environments. It utilizes the Antlion Algorithm (ALO) to globally optimize the mode number and penalty parameters of the Virtual Vibration Decomposition (VMD). Using a weighted sum of reconstruction error and modal sparsity as the fitness function, it automatically matches the frequency domain distribution characteristics of the tunnel vibration signal, fundamentally solving the problem of incomplete or redundant mode separation caused by the empirical setting of traditional VMD parameters, thereby improving decomposition accuracy. The Pearson correlation coefficient is used to quantitatively evaluate the correlation between each IMF component and the original signal. A unified threshold is set to remove noise-dominant modes while retaining effective feature components, achieving rapid filtering of frequency domain noise with controllable computational complexity, suitable for real-time on-site processing. L1 regularization (Lasso) is introduced to replace the traditional L0 norm sparsity constraint, transforming the non-convex optimization problem into a convex one. A coordinate descent method is used for stable solution, avoiding local optima and overfitting. Simultaneously, the dictionary is updated atom-by-atom through singular value decomposition, making the learned overcomplete dictionary more closely match the inherent shape of the weak vibration signal in the tunnel. The initial denoising removes most of the frequency domain noise, while the secondary denoising uses dictionary sparse representation to remove residual time domain noise. The two work together to significantly improve the signal-to-noise ratio in simulation experiments. Moreover, this method does not rely on prior signal models and has good robustness to complex working conditions such as high dust, low illumination, and strong vibration in tunnels. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0018] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0019] Figure 2 The images are simulated original vibration signals with noise, where (a) is the time domain signal, (b) is the frequency domain signal, and (c) is the time-frequency signal.

[0020] Figure 3 Figure 1 shows the signal images after the initial denoising of the noisy signal. Figure 2 shows the time domain signal, Figure 3 shows the frequency domain signal, and Figure 4 shows the time-frequency signal.

[0021] Figure 4 Figure 1 shows the signal image after the noisy signal has been denoised twice by the traditional K-SVD dictionary learning model. Figure 2 shows the time domain signal, Figure 3 shows the frequency domain signal, and Figure 4 shows the time-frequency signal.

[0022] Figure 5 Figure 1 shows the signal image after secondary denoising of the noisy signal using the improved K-SVD dictionary learning model. Figure 2 shows the time domain signal, Figure 3 shows the frequency domain signal, and Figure 4 shows the time-frequency signal.

[0023] Figure 6 The images show the original vibration signals under high dust concentrations. In Figure (a), the signal is in the time domain; in Figure (b), the signal is in the frequency domain; and in Figure (c), the signal is in the time-frequency domain.

[0024] Figure 7 The images are the signal images after initial noise reduction under high dust concentration. In Figure (a), the signal is in the time domain; in Figure (b), the signal is in the frequency domain; and in Figure (c), the signal is in the time-frequency domain.

[0025] Figure 8 The images are signal images after secondary noise reduction using the traditional K-SVD dictionary learning model under high dust concentration. In Figure (a), the signal is in the time domain; in Figure (b), the signal is in the frequency domain; and in Figure (c), the signal is in the time-frequency domain.

[0026] Figure 9 The images are of the signals after secondary noise reduction using the improved K-SVD dictionary learning model under high dust concentration. In Figure (a), the signal is in the time domain; in Figure (b), the signal is in the frequency domain; and in Figure (c), the signal is in the time-frequency domain. Detailed Implementation

[0027] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0028] Example 1 like Figure 1 As shown, a noise processing method for laser vibration measurement information in complex tunnel environments is described. The method involves acquiring the original vibration signal from a laser vibration measurement in the complex tunnel environment. The method includes a two-stage noise reduction process: initial noise reduction and secondary noise reduction. The specific steps are as follows: (1) Initial noise reduction processing: From the original vibration signal Typical vibration signals were selected as parameter optimization samples. Using these samples as input, the Antlion Algorithm (ALO) was applied to the variational mode decomposition (VMD) modal number. and penalty parameters Perform adaptive optimization and output the optimal parameter combination. , To be the optimal number of modes, The optimal penalty parameters are used to adaptively match the frequency domain distribution characteristics of tunnel vibration signals, solving the problem of under-decomposition / over-decomposition of modes caused by manually setting VMD parameters, and improving the mode separation accuracy. Use the optimal parameter combination Variational mode decomposition was performed on the original vibration signal to obtain The original signal with mixed noise is decomposed into multiple groups of single-frequency narrowband modes, realizing the separation of effective vibration characteristics from noise modes in the frequency domain. Calculate the Pearson correlation coefficient between each IMF modal component and the original vibration signal, retain the effective information components whose absolute correlation coefficient is greater than or equal to a preset threshold, and reconstruct the signal by superimposing all effective information components in the time domain to obtain the initial denoised signal. Quantitatively distinguish between noise-dominant modes and effective characteristic modes, eliminate tunnel-irrelevant interference components, and complete the first layer of coarse noise reduction; (2) Secondary noise reduction processing: Typical vibration signals are selected from the initial noise reduction signal as dictionary training samples. The dictionary training samples are input into the improved K-SVD dictionary learning model. An optimized overcomplete dictionary is obtained by training based on L1 regularized sparse representation. The inherent characteristic basis functions of tunnel vibration signals are learned to construct a sparse expression dictionary that adapts to the weak surrounding rock vibration characteristics. The initial noise-reducing signal is input into an optimized and complete dictionary. A second noise reduction is then performed through sparse representation solving and signal reconstruction, outputting a clean vibration signal after final noise filtering. The sparse characterization method is used to remove the weak noise remaining after the initial noise reduction, and to accurately restore the true vibration waveform of the tunnel under low signal-to-noise ratio.

[0029] In a specific implementation, the adaptive optimization process in the initial noise reduction process is as follows: Define the optimization objective, in terms of modal number. and penalty parameters Let the variable to be optimized be an objective function. The optimization objective is to minimize the weighted sum of signal reconstruction error and modal sparsity. The expression of the objective function is: , in, express The corresponding fitness value, This indicates that the input parameters optimize the sample signal. The decomposition yields the first... indivual Modal components, Indices representing modal components; Represents the L2 norm; It is an L1 norm; Indicates the balance weighting coefficient; Synchronization constraint signal distortion and modal redundancy are used to quantitatively evaluate the decomposition quality of each set of VMD parameters; Set the population parameters for the ALO algorithm, including the ant population size. Antlion population size Maximum number of iterations Set the boundary values ​​for the optimization variables, and the number of modes. It is a positive integer, and its value range is 1. Penalty parameters It is a positive real number, and its range is 1. Set the initial search step size factor Convergence criterion threshold , , Representing the number of modes respectively The minimum and maximum values ​​are typically determined by the modal number setting. , , , These represent the penalty parameters. The minimum and maximum values ​​are typically used to set the penalty parameter. , ; Limit the search range of parameters to a reasonable level, control the computational load of the algorithm, and avoid meaningless global traversal; Randomly generated within the range of variable values. Group of ant positions, each group of ant positions corresponds to a group Parameter combination; selecting from all ant locations The position with the best fitness in a given set is used as the initial antlion position; fitness is determined by the objective function value, and the smaller the objective function value, the better the fitness; multiple sets of candidate parameter solutions are initialized, and high-quality initial solutions are selected to improve the convergence speed of the algorithm; by For the current iteration number, execute The loop iteration, Indicates the maximum number of iterations; iteratively searches for a better combination of VMD parameters. After the iteration terminates, the parameter combination corresponding to the globally optimal antlion is taken as the optimal parameter combination. Output: Output the unique and optimal VMD parameters that fit the current tunnel vibration signal for subsequent mode decomposition.

[0030] In a specific implementation, the iterative optimization process is as follows: Each ant updates its position by randomly circling its corresponding antlion, using the following formula: , in, For the first The ant's position in the next iteration. For the first The globally optimal antlion position in the next iteration. For the first The location of the antlion is randomly selected in the next iteration. , A uniformly random number within the interval [0,1]. For the first The search step size factor for the next iteration; Simulates the random walk mechanism of antlions hunting, and searches for potential optimal parameters globally to prevent local optima; The search step size decreases linearly with the number of iterations, gradually shifting from a global search to a local fine-grained search. The step size update formula is: ; Iterate through global exploration in the early stage and local in-depth exploration in the later stage, taking into account both the breadth of optimization and the accuracy of parameters; Parameters that exceed the range after the update are truncated and corrected, and the modal number is adjusted accordingly. Limited to Within the range and rounded down, the penalty parameter is... Limited to Within the specified range; ensure the validity of constraint parameters to avoid invalid parameters causing VMD decomposition failure; For each ant The parameter combination is used to perform VMD decomposition, and the corresponding fitness value is calculated by substituting it into the objective function; the decomposition effect of each group of candidate parameters is evaluated in real time, and the fitness is quantified for the survival of the fittest; Replace the ant position with the corresponding antlion position with the ant position with the better fitness, while retaining the globally optimal antlion position as the historical best solution; the elite retention strategy continuously retains the best parameters to prevent the optimal solution from being lost in iteration. If the improvement in the objective function value for a consecutive preset number of iterations is less than the convergence threshold If the accuracy requirement is met, the iteration will terminate early; if the accuracy requirement is met, the loop will exit early to reduce the computation time of the algorithm and improve the efficiency of real-time processing.

[0031] In a specific implementation, the process of solving the initial noise reduction signal is as follows: Using the original vibration signal as input, the optimal parameter combination is employed. Perform variational mode decomposition to decompose the original signal into... A finite bandwidth IMF modal component And simultaneously obtain the center frequency corresponding to each IMF component. Precisely separate multi-frequency coupled tunnel hybrid vibration signals, with each mode independently corresponding to a single frequency band vibration component; Calculate the first One IMF modal component Compared with the original vibration signal Pearson correlation coefficient The calculation formula is: , in, Represents the mathematical expectation operator. It represents the variance operator; it quantifies the correlation between a single mode and the original signal features, and objectively distinguishes between noise modes and effective modes; Set the threshold for Pearson correlation coefficient ,like If the corresponding IMF component is valid, then the IMF component that does not meet this condition is determined to be a noise-dominant component and is directly removed; establish a unified screening standard to remove tunnel environment-irrelevant noise modes in batches. By superimposing all effective information components point-by-point in the time domain, the initial denoised signal can be reconstructed. The first noise mode removal is completed; all effective vibration components are fused to output a coarse noise-reduced signal that filters out most of the frequency domain noise.

[0032] In a specific implementation, the K-SVD dictionary learning model is improved in the secondary noise reduction process, and L1 regularization Lasso is used to replace the L0 norm sparsity constraint in the traditional K-SVD algorithm; the improved K-SVD global optimization objective function is: , in, For the training sample matrix, The sampling length of a single vibration signal. This represents the total number of training samples; For the purpose of learning a complete dictionary, The number of atoms in the dictionary. ; It is a sparse matrix. For the first The sparse coefficient vector corresponding to each training sample The index for the training samples; It is the Frobenius norm; It is an L1 norm; The regularization coefficient is used. This is an operation to retrieve the minimum value; By replacing the non-convex L0 constraint with L1 convex relaxation, the defects of local optima and overfitting in traditional algorithms are avoided, and the weak vibration characteristics of tunnels are extracted stably.

[0033] In a specific implementation, the training execution steps of the improved K-SVD dictionary learning model are as follows: Step 1: Construct the dictionary training sample matrix Y, and randomly select samples from the training samples. 1 sample as the initial dictionary atom For each atom, L2 norm normalization is performed to obtain an initial dictionary. The normalization formula is: , ; in, Represents the first in the dictionary One atom; Generate initial characteristic basis functions, unify atomic amplitude scales, and eliminate the interference of numerical bias on iterative convergence; Step 2: Fix the current dictionary and apply it to each training sample. Solve the L1-regularized Lasso optimization problem to obtain the corresponding sparse coefficients. The single-sample optimization problem is expressed as: , in, For the first training samples, Let be the sparse coefficient vector to be solved; The coordinate descent method is used to solve the above convex optimization problem. After traversing all training samples, the complete sparse coefficient matrix is ​​obtained. Solve for the sparse weights of the samples in the dictionary, retain only a small number of key atoms to represent the vibrational signal, and suppress residual noise expression; Step 3: Fix the sparse coefficient matrix Update the dictionary atoms column by column; for the first column... dictionary atoms Perform the following operations: 1) Fixed use of the first A set of sample indices of atoms , For the first The first sample sparse coefficient vector of the th sample Each element; positioning depends on all samples of the current atom, updating the atom only for valid samples, reducing computational load; 2) Calculate the global error matrix and remove the first error. The contribution of each atom yields the corresponding residual matrix. : , in, For atomic indexing, Represents the first in the dictionary One atom, Represents the sparse coefficient matrix The Middle Transpose each row vector; remove interference from other atoms, calculate the fitting residual corresponding to the current atom separately, and accurately optimize the single-atom features; 3) From the residual matrix Extract the index set The corresponding columns yield a submatrix. ,right Perform Singular Value Decomposition (SVD): , in, It is a left singular matrix. It is a singular value diagonal matrix. It is a right singular matrix. This indicates a transpose operation; the most significant vibrational features in the residuals are extracted through SVD decomposition to generate the optimal update atom; 4) Take The first column is the updated number dictionary atoms ,Pick The first singular value and The product of the first column is used as the corresponding sparse coefficient row to complete the optimal update of a single atom; the feature atoms and their corresponding sparse weights are updated simultaneously to minimize the global signal reconstruction error. Step 4: Perform L2 norm normalization on the updated dictionary atoms and adjust the magnitude of the sparse coefficients accordingly; unify the magnitude standard of all atoms to avoid numerical divergence during the iteration process and ensure the stability of the dictionary; Step 5: Repeat steps 2 to 4 until the change in global reconstruction error before and after dictionary update is less than the preset convergence threshold, or the preset maximum number of iterations is reached, and output the optimized overcomplete dictionary D after training.

[0034] In a specific implementation, the method further includes evaluating the denoised signal, and the evaluation metrics include signal-to-noise ratio (SNR) and root mean square error (RMSE). The formula for calculating the signal-to-noise ratio (SNR) is as follows: ; The formula for calculating the root mean square error (RMSE) is as follows: ; in, The first denoised signal is the denoised signal. One sampling point, The first reference pure signal One sampling point, This represents the total number of sampling points for the signal. For sampling point index; A higher signal-to-noise ratio and a smaller root mean square error indicate a better denoising effect; quantifying the degree of signal distortion and noise suppression before and after denoising objectively verifies the effectiveness of the two-stage denoising method.

[0035] Example 2 A laser vibration measurement vibration information noise processing system for complex tunnel environments executes a laser vibration measurement vibration information noise processing method for complex tunnel environments, including: The signal acquisition module is configured to acquire the original vibration signal from the laser vibration measurement in the complex environment of the tunnel, and select typical vibration signals from the original vibration signal as parameter optimization samples and dictionary training samples, respectively. The parameter optimization module is configured to receive parameter optimization samples, use the antlion optimization algorithm to adaptively optimize the number of modes and penalty parameters of variational mode decomposition, and output the optimal parameter combination. The decomposition and reconstruction module is configured to receive the original vibration signal and the optimal parameter combination, perform adaptive variational mode decomposition, screen the effective IMF components through the correlation coefficient and reconstruct the signal, and output the initial noise-reduced signal. The secondary noise reduction module is configured to receive dictionary training samples and the initial noise reduction signal, first train and optimize the overcomplete dictionary based on the improved K-SVD model, and then perform secondary noise reduction on the initial noise reduction signal based on L1 regularized sparse representation to output the final clean vibration signal. The evaluation output module is configured to calculate the signal-to-noise ratio and root mean square error of the noise reduction effect, and output the final clean vibration signal and quantitative evaluation results.

[0036] Example 3 A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements steps such as those of a laser vibration measurement vibration information noise processing method for complex tunnel environments.

[0037] Example 4 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements steps such as those of a laser vibration measurement vibration information noise processing method for complex tunnel environments.

[0038] Example 5 This embodiment uses a noisy simulated vibration signal to verify the effectiveness of the method of the present invention. An original signal is constructed by superimposing low-frequency (10Hz), mid-frequency (50Hz), and high-frequency (120Hz) sinusoidal signals with Gaussian white noise. The signal-to-noise ratio is set to 5dB, the sampling frequency is 10kHz, and the sampling duration is 0.5s. The time domain, frequency domain, and time-frequency distribution of the original noisy signal are as follows: Figure 2 As shown, the time-domain waveform fluctuates violently, the noise floor in the frequency domain is raised, and the effective signal in the time-frequency graph is submerged by noise. The ALO algorithm was used to optimize the parameters of VMD, with the population size of both ants and antlions set to 30, the maximum number of iterations to 50, and the number of modalities set to [missing information]. ∈[3,12], penalty parameter ∈[1000,5000]; After iterative convergence, the optimal parameter combination (K=4, α=2778) is obtained; This parameter is used to perform VMD decomposition on the original signal to obtain 4 IMF components. The Pearson correlation coefficient between each component and the original signal is calculated, and a threshold of 0.6 is set to select IMF1, IMF2, and IMF3 as effective components, while IMF4 is a noise component and is removed; The three effective components are superimposed in the time domain to obtain the initial denoised signal, and its time domain, frequency domain, and time-frequency plot are shown below. Figure 3 As shown; it can be seen that the time-domain waveform glitches are significantly reduced, the noise floor in the frequency domain is lowered, the three frequency components in the time-frequency graph gradually become clearer, the signal-to-noise ratio is increased to 13.05dB, and the RMSE is reduced to 0.0195; Subsequently, the initially denoised signal was used as training samples, and a complete dictionary was trained based on improved K-SVD (L1 regularization), with the number of dictionary atoms M=2L (L being the signal length). For comparison, traditional K-SVD (OMP, L0 constraint) was also used for secondary denoising. The signal after secondary denoising using traditional K-SVD is shown below. Figure 4 As shown, the signal-to-noise ratio is 13.51 dB, and the RMSE is 0.0185, indicating limited improvement. Residual noise fringes are still visible in the time-frequency plot. However, the signal after secondary noise reduction using the improved K-SVD method of this invention is as follows: Figure 5 As shown, the time-domain waveform is smooth and continuous, the frequency-domain noise floor is further suppressed, the three frequency components in the time-frequency graph are clearly highlighted, the background is clean, the signal-to-noise ratio reaches 19.60dB, and the RMSE is only 0.0091. The results show that the improved K-SVD, due to its L1 convex optimization characteristics, can extract weak signal components more stably, and its noise reduction effect is significantly better than that of the traditional method.

[0039] Example 6 This embodiment collects raw laser vibration signals from a tunnel construction site in a high-dust-concentration environment. The sampling frequency is 5kHz and the sampling duration is 1s. The raw signal is severely affected by laser scattering and dust particle interference. Its time-domain, frequency-domain, and time-frequency plots are shown below. Figure 6 As shown, the time-domain waveform is chaotic, and the high-frequency noise energy is widely distributed in the frequency domain, making it almost impossible to identify the effective vibration components in the time-frequency diagram. Using the method of this invention, the ALO algorithm is first used to optimize the parameters of VMD (parameter settings are the same as in Example 4) to obtain the optimal combination ( =8, =1976); VMD decomposition was performed using this parameter to obtain 8 IMF components. The correlation coefficient threshold was set to 0.5, and 5 effective components were selected for reconstruction to obtain the initial denoised signal. Its time domain, frequency domain, and time-frequency plot are shown below. Figure 7As shown; it can be seen that the overall trend of the time domain signal appears, some high-frequency noise in the frequency domain is suppressed, and low-frequency modulation components can be faintly observed in the time-frequency graph, but there is still a lot of residual noise. The signal-to-noise ratio is 4.92dB and the RMSE is 0.0275. The initial noise-reduced signal undergoes secondary processing, employing both traditional K-SVD and improved K-SVD methods; after traditional K-SVD processing ( Figure 8 The time-domain waveform is slightly smoothed, the signal-to-noise ratio increases to 6.07 dB, and the RMSE decreases to 0.0241. However, the noise distribution in the time-frequency graph is disordered, resulting in limited improvement. In contrast, after using the improved K-SVD processing of this invention (…), Figure 9 The time-domain waveform regularity is significantly enhanced, the noise floor in the frequency domain is significantly reduced, the effective vibration frequency band (approximately 20–40 Hz) is clearly presented in the time-frequency diagram, background interference is greatly reduced, the signal-to-noise ratio reaches 9.54 dB, and the RMSE drops to 0.0202, representing a 94.0% improvement in signal-to-noise ratio compared to the initial noise reduction. Experiments have shown that this invention can effectively separate laser scattering noise and restore the real tunnel surrounding rock vibration signal even in extremely dusty environments, providing reliable data for engineering safety monitoring.

[0040] Although the specific embodiments of the invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the invention. Based on the technical solutions of the invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the invention.

Claims

1. A method for noise processing of laser vibratory measurement vibration information in complex tunnel environments, characterized by: acquiring the original vibration signal from on-site laser vibratory measurement in complex tunnel environments; The method includes a two-stage noise reduction process: initial noise reduction processing and secondary noise reduction processing. The specific steps are as follows: (1) Initial noise reduction processing: Typical vibration signals were selected from the original vibration signals as parameter optimization samples. Using these samples as input, the Antlion Algorithm (ALO) was applied to optimize the mode number of Variational Mode Decomposition (VMD). and penalty parameters Perform adaptive optimization and output the optimal parameter combination. , To be the optimal number of modes, The optimal penalty parameter; Use the optimal parameter combination Variational mode decomposition was performed on the original vibration signal to obtain Each intrinsic mode function (IMF) modal component, and the center frequency corresponding to each IMF modal component; Calculate the Pearson correlation coefficient between each IMF modal component and the original vibration signal, retain the effective information components whose absolute value of the correlation coefficient is greater than or equal to a preset threshold, and reconstruct all effective information components by superimposing them in the time domain to obtain the initial noise-reduced signal. (2) Secondary noise reduction processing: Typical vibration signals are selected from the initial noise reduction signal as dictionary training samples. The dictionary training samples are input into the improved K-SVD dictionary learning model, and an optimized overcomplete dictionary is obtained by training based on L1 regularized sparse representation. The initial noise reduction signal is input into an optimized and complete dictionary. The second noise reduction is completed through sparse representation solution and signal reconstruction, and the final noise-filtered pure vibration signal is output.

2. The method for noise processing of laser vibration measurement information in complex tunnel environments according to claim 1, characterized in that, initially... The specific process of adaptive optimization in noise reduction is as follows: Define the optimization objective, in terms of modal number. and penalty parameters Let the variable to be optimized be an objective function. The optimization objective is to minimize the weighted sum of signal reconstruction error and modal sparsity. The expression of the objective function is: ， in, express The corresponding fitness value, This indicates that the input parameters optimize the sample signal. The decomposition yields the first... indivual Modal components, Indices representing modal components; Represents the L2 norm; It is an L1 norm; Indicates the balance weighting coefficient; Set the population parameters for the ALO algorithm, including the ant population size. Antlion population size Maximum number of iterations Set the boundary values ​​for the optimization variables, and the number of modes. It is a positive integer, and its value range is 1. Penalty parameters It is a positive real number, and its range is 1. Set the initial search step size factor Convergence criterion threshold , , Representing the number of modes respectively The minimum and maximum values ​​are taken. , These represent the penalty parameters. The minimum and maximum values ​​can be obtained. Randomly generated within the range of variable values. Group of ant positions, each group of ant positions corresponds to a group Parameter combination; selecting from all ant locations The position with the best fitness in the group is taken as the initial antlion position; fitness is determined by the objective function value, the smaller the objective function value, the better the fitness; by For the current iteration number, execute The loop iteration, Indicates the maximum number of iterations; After the iteration terminates, the parameter combination corresponding to the globally optimal antlion is taken as the optimal parameter combination. Output.

3. The method for noise processing of laser vibratory information in complex tunnel environments according to claim 2, characterized in that, The iterative optimization process is as follows: Each ant updates its position by randomly circling its corresponding antlion, using the following formula: ， in, For the first The ant's position in the next iteration. For the first The globally optimal antlion position in the next iteration. For the first The location of the antlion is randomly selected in the next iteration. , A uniformly random number within the interval [0,1]. For the first The search step size factor for the next iteration; The search step size decreases linearly with the number of iterations, gradually shifting from a global search to a local fine-grained search. The step size update formula is: ； Parameters that exceed the range after the update are truncated and corrected, and the modal number is adjusted accordingly. Limited to Within the range and rounded down, the penalty parameter is... Limited to Within the range; For each ant The parameter combination is used to perform VMD decomposition, and the corresponding fitness value is calculated by substituting it into the objective function. Replace the ant position with the corresponding antlion position with the better fitness, while retaining the globally optimal antlion position as the historical optimal solution. If the improvement in the objective function value for a consecutive preset number of iterations is less than the convergence threshold If the iteration is terminated prematurely, then the iteration will be terminated early.

4. The method for noise processing of laser vibratory information in complex tunnel environments according to claim 1, characterized in that, initially... The process of obtaining the noise-reduced signal is as follows: Using the original vibration signal as input, the optimal parameter combination is employed. Perform variational mode decomposition to decompose the original signal into... A finite bandwidth IMF modal component And simultaneously obtain the center frequency corresponding to each IMF component. ; Calculate the first One IMF modal component Compared with the original vibration signal Pearson correlation coefficient The calculation formula is: ， in, Represents the mathematical expectation operator. Represents the variance operator; Set the threshold for Pearson correlation coefficient ,like If the condition is met, the corresponding IMF component is a valid information component. IMF components that do not meet this condition are judged as noise-dominant components and are directly removed. By superimposing all effective information components point-by-point in the time domain, the initial denoised signal can be reconstructed. The first noise mode removal was completed.

5. The method for noise processing of laser vibratory information in complex tunnel environments according to claim 1, characterized in that, In the secondary noise reduction process, the K-SVD dictionary learning model is improved by replacing the L0 norm sparsity constraint in the traditional K-SVD algorithm with L1 regularization Lasso. The improved K-SVD global optimization objective function is: ， in, For the training sample matrix, The sampling length of a single vibration signal. This represents the total number of training samples; For the purpose of learning a complete dictionary, The number of atoms in the dictionary. ; It is a sparse matrix. For the first The sparse coefficient vector corresponding to each training sample The index for the training samples; It is the Frobenius norm; It is an L1 norm; The regularization coefficient is used. This is an operation to find the minimum value.

6. The method for noise processing of laser vibratory information in complex tunnel environments according to claim 5, characterized in that, The training steps for the improved K-SVD dictionary learning model are as follows: Step 1: Construct the dictionary training sample matrix Y, and randomly select samples from the training samples. 1 sample as the initial dictionary atom For each atom, L2 norm normalization is performed to obtain an initial dictionary. The normalization formula is: ， ； in, Represents the first in the dictionary One atom; Step 2: Fix the current dictionary and apply it to each training sample. Solve the L1-regularized Lasso optimization problem to obtain the corresponding sparse coefficients. The single-sample optimization problem is expressed as: ， in, For the first training samples, Let be the sparse coefficient vector to be solved; The coordinate descent method is used to solve the above convex optimization problem. After traversing all training samples, the complete sparse coefficient matrix is ​​obtained. ; Step 3: Fix the sparse coefficient matrix Update the dictionary atoms column by column; for the first column... dictionary atoms Perform the following operations: 1) Fixed use of the first A set of sample indices of atoms , For the first The first sample sparse coefficient vector of the th sample One element; 2) Calculate the global error matrix and remove the first error. The contribution of each atom yields the corresponding residual matrix. : ， in, For atomic indexing, Represents the first in the dictionary One atom, Represents the sparse coefficient matrix The Middle Transpose of a row vector; 3) From the residual matrix Extract the index set The corresponding columns yield a submatrix. ,right Perform Singular Value Decomposition (SVD): ， in, It is a left singular matrix. It is a singular value diagonal matrix. It is a right singular matrix. Indicates the transpose operation; 4) Take The first column is the updated number dictionary atoms ,Pick The first singular value and The product of the first column is used as the corresponding sparse coefficient row to complete the optimal update of a single atom; Step 4: Perform L2 norm normalization on the updated dictionary atoms and adjust the magnitude of the sparsity coefficients accordingly; Step 5: Repeat steps 2 to 4 until the change in global reconstruction error before and after dictionary update is less than the preset convergence threshold, or the preset maximum number of iterations is reached, and output the optimized overcomplete dictionary D after training.

7. The method for noise processing of laser vibratory information in complex tunnel environments according to claim 1, characterized in that, The method also includes evaluating the denoised signal, with evaluation metrics including signal-to-noise ratio (SNR) and root mean square error (RMSE). The formula for calculating the signal-to-noise ratio (SNR) is as follows: ； The formula for calculating the root mean square error (RMSE) is as follows: ； in, The first denoised signal is the denoised signal. One sampling point, The first reference pure signal One sampling point, This represents the total number of sampling points for the signal. For sampling point index; A higher signal-to-noise ratio and a smaller root mean square error indicate a better noise reduction effect.

8. A laser vibration measurement vibration information noise processing system for complex tunnel environments, comprising executing the laser vibration measurement vibration information noise processing method for complex tunnel environments as described in any one of claims 1-7, characterized in that, include: The signal acquisition module is configured to acquire the original vibration signal from the laser vibration measurement in the complex environment of the tunnel, and select typical vibration signals from the original vibration signal as parameter optimization samples and dictionary training samples, respectively. The parameter optimization module is configured to receive parameter optimization samples, use the antlion optimization algorithm to adaptively optimize the number of modes and penalty parameters of variational mode decomposition, and output the optimal parameter combination. The decomposition and reconstruction module is configured to receive the original vibration signal and the optimal parameter combination, perform adaptive variational mode decomposition, screen the effective IMF components through the correlation coefficient and reconstruct the signal, and output the initial noise-reduced signal. The secondary noise reduction module is configured to receive dictionary training samples and the initial noise reduction signal, first train and optimize the overcomplete dictionary based on the improved K-SVD model, and then perform secondary noise reduction on the initial noise reduction signal based on L1 regularized sparse representation to output the final clean vibration signal. The evaluation output module is configured to calculate the signal-to-noise ratio and root mean square error of the noise reduction effect, and output the final clean vibration signal and quantitative evaluation results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the laser vibration measurement vibration information noise processing method for complex tunnel environments as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the laser vibration measurement vibration information noise processing method for complex tunnel environments as described in any one of claims 1-7.

Images