Tunnel structure safety dynamic evaluation method and system based on data collaborative awareness, computer equipment and storage medium

Abstract

The invention relates to a tunnel structure safety dynamic evaluation method based on data collaborative awareness, and the method comprises the steps: synchronously collecting a vibration signal and a strain signal through a vibration sensor and a strain gauge which are disposed on a tunnel structure; performing cooperative denoising on the vibration signal and the strain signal according to the physical relevance of the vibration signal and the strain signal in structural mechanical response to obtain the denoised vibration signal and the denoised strain signal; and extracting features of the denoised vibration signals and strain signals, performing feature fusion, and inputting the fused features into a pre-trained safety evaluation model to obtain a comprehensive safety state of the tunnel structure. Cooperative denoising is carried out by using the physical relevance between vibration and strain signals, random noise irrelevant to structural response can be extremely accurately stripped, the signal-to-noise ratio and reliability of the signals are remarkably improved, and thus the accuracy of safety evaluation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a tunnel structure safety dynamic evaluation method and device based on data collaborative perception, a computer device and a storage medium. BACKGROUND

[0002] With the continuous growth of the construction scale and operation mileage of tunnel projects, the long-term service safety of tunnels faces severe challenges. Factors such as complex geological conditions, continuously increasing traffic loads, and material aging lead to frequent structural diseases such as cracking, water leakage, and spalling of tunnel linings. Traditional manual inspection methods are highly subjective, inefficient, and difficult to detect internal defects, and cannot meet the needs of modern tunnel intelligentization and preventive maintenance.

[0003] Currently, automatic monitoring technology based on the Internet of Things has become an important means to ensure the safety of tunnel operation. By deploying various sensors such as vibration, strain, and displacement in tunnels, real-time collection of structural response data is achieved. However, the existing monitoring scheme still has the following problems: 1. The internal environment of the tunnel is complex, and sensor signals are easily affected by environmental noise such as vehicle vibration, electromagnetic interference, and temperature and humidity changes. Existing technologies mostly use independent filtering and denoising of single sensor signals (such as traditional wavelet threshold method), lacking a mechanism for collaborative denoising using the internal correlation between different physical quantity signals. This isolated processing method cannot effectively distinguish between the true response of the structure and environmental noise, resulting in a low signal-to-noise ratio and hidden dangers for subsequent accurate analysis.

[0004] 2. Vibration and strain monitoring data have different physical meanings, dimensions, and sampling frequencies, and are typical multi-source heterogeneous data. Existing systems often simply display the data side by side or use a simple weighted average method for fusion, failing to deeply process the uncertainty, contradiction, and even conflict between the data from the perspective of evidence theory.

[0005] 3. Tunnel structures are in a healthy state most of the time, and abnormal data samples representing structural damage or danger are extremely rare, which leads to a serious small sample problem for deep learning-based early warning models. Traditional supervised learning models are prone to overfitting to normal conditions when the training data is imbalanced, and are not sensitive to real damage, resulting in insufficient generalization ability and reliability of the early warning model. SUMMARY

[0006] The present application aims to provide a tunnel structure safety dynamic evaluation method based on data collaborative perception to solve the above problems.

[0007] The first aspect of the present application provides a tunnel structure safety dynamic evaluation method based on data collaborative perception, comprising: The vibration signal and the strain signal are synchronously collected through vibration sensors and strain gauges arranged on the tunnel structure. According to the physical correlation of the vibration signal and the strain signal in structural mechanics response, the vibration signal and the strain signal are cooperatively denoised to obtain denoised vibration signal and strain signal. The cooperative denoising includes: analyzing the correlation of each frequency band component of the strain signal and the vibration signal, determining the vibration signal frequency band with a correlation higher than a preset threshold as a frequency band representing effective structural response, reconstructing the effective vibration signal, and calibrating and denoising the strain signal using the reconstructed effective vibration signal. The features of the denoised vibration signal and strain signal are extracted, and the fused features are input into a pre-trained safety evaluation model to obtain the comprehensive safety state of the tunnel structure.

[0008] In one possible implementation, the analysis of the correlation of each frequency band component of the strain signal and the vibration signal includes: The vibration signal is decomposed into K intrinsic mode functions using a variational mode decomposition algorithm, where the value of K is determined by observing the center frequency of the vibration signal spectrum. The Pearson correlation coefficient sequence of each intrinsic mode function and the strain signal is calculated. The average value of the absolute value of the Pearson correlation coefficient sequence within a time window is selected as the overall correlation measure of the intrinsic mode function and the strain signal.

[0009] In one possible implementation, the reconstruction of the effective vibration signal includes: According to a preset correlation threshold, all intrinsic mode functions with an overall correlation measure greater than the threshold are screened out. The screened intrinsic mode functions are linearly superimposed to reconstruct the denoised vibration signal. The correlation threshold is determined by the correlation coefficient distribution statistical value of the structure in a normal state in historical data.

[0010] In one possible implementation, the calibration and denoising of the strain signal using the reconstructed effective vibration signal includes: The denoised strain signal is subjected to secondary denoising processing using a wavelet threshold denoising method with the reconstructed effective vibration signal as a reference benchmark.

[0011] In one possible implementation, the feature fusion adopts a D-S evidence theory method, specifically including: Basic probability assignments are assigned to the features from the vibration signal and the strain signal, respectively. According to the combination rule of the D-S evidence theory, the belief interval of each safety state level is calculated. The state level with the highest confidence is taken as the real-time safety state of the tunnel structure.

[0012] In a possible implementation, before the features of the denoised vibration signal and strain signal are extracted, fused, and input into a pre-trained safety evaluation model to obtain the comprehensive safety state of the tunnel structure, the method further includes: A refined finite element model of the tunnel segment lining is constructed, and simulated vibration and strain response data are obtained under the working conditions of local cracks and spalling damage caused by vehicle load and soil pressure, serving as a physical verification data set; A conditional generative adversarial network is constructed, the generator of which takes a normal state monitoring data segment and random noise as input and generates simulated data with specific damage characteristics as a condition. The simulated data generated by the generator is filtered and corrected using the physical verification data set to ensure its physical reasonableness, and the corrected data is added to the training set to train the safety evaluation model.

[0013] In a possible implementation, the safety evaluation model adopts a hybrid model of one-dimensional time series convolution network and attention mechanism. The one-dimensional time series convolution network is configured to extract local features from continuous monitoring data sequences. The attention mechanism is configured to assign different weights to the feature sequences output by the convolution network to focus on key time steps most relevant to structural damage. The final output of the hybrid model is the prediction probability of the safety state of the tunnel structure in a future preset time period.

[0014] The second aspect of the application provides a tunnel structure safety dynamic evaluation system based on data collaborative perception, comprising: A data acquisition module is configured to synchronously acquire vibration signals and strain signals through vibration sensors and strain gauges arranged on the tunnel structure. A collaborative denoising module is configured to collaboratively denoise the vibration signals and strain signals according to their physical correlation in structural mechanical response to obtain denoised vibration signals and strain signals. A fusion evaluation module is configured to extract features of the denoised vibration signals and strain signals, fuse the features, input the fused features into a pre-trained safety evaluation model, and obtain a comprehensive safety state of the tunnel structure.

[0015] The third aspect of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for dynamic evaluation of tunnel structure safety based on data collaborative perception according to the first aspect of the present application.

[0016] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the method for dynamic evaluation of tunnel structure safety based on data collaborative perception according to the first aspect of the present application.

[0017] Compared with the prior art, the present application has the following beneficial effects: 1. By utilizing the physical correlation between vibration and strain signals for collaborative denoising, random noise irrelevant to structural response can be extremely accurately stripped, and the signal-to-noise ratio and reliability of the signal are significantly improved, thereby improving the accuracy of safety evaluation; 2. By applying D-S evidence theory to feature fusion of the denoised signal, the uncertainty and conflict from data of different sensors can be effectively processed, so that the final comprehensive safety state evaluation result is more objective and reliable, and the false alarm and missed alarm rates of the system are significantly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 FIG. 1 is a flowchart of the method for dynamic evaluation of tunnel structure safety based on data collaborative perception according to the present application; Figure 2 FIG. 2 is an architecture diagram of the system for dynamic evaluation of tunnel structure safety based on data collaborative perception according to the present application; Figure 3 FIG. 3 is a schematic diagram of the computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described below in detail with reference to the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0020] It should be noted that the serial numbers of components in the embodiments of the present application, such as “first”, “second”, etc., are only used to distinguish the described objects, and do not have any sequence or technical meaning.

[0021] The following will be described with reference to the accompanying drawings. Figure 1The application discloses a tunnel structure safety dynamic evaluation method based on data collaborative sensing.

[0022] A tunnel structure safety dynamic evaluation method based on data collaborative sensing, comprising: S1, synchronously collecting vibration signals and strain signals through vibration sensors and strain gauges arranged on a tunnel structure; The vibration sensors and the strain gauges are preferably arranged at key mechanical sections of the tunnel structure: (1) positions with large lining internal forces, such as arch crown, arch waist and middle part of inverted arch.

[0023] (2) positions with changing geological conditions, such as soft-hard rock junction sections and positions near fault fracture zones.

[0024] (3) positions with slight diseases, such as positions with found cracks or water seepage.

[0025] For a monitoring section, at least three collaborative sensing nodes can be arranged at the arch crown, left and right arch waists and the like. For a key monitoring section, a monitoring section can be arranged every 20-50 m.

[0026] In the data collection process, the original signals can be preliminarily preprocessed, such as analog filtering and digital filtering, so that the signals in subsequent analysis can accurately reflect the state of the tunnel structure.

[0027] S2, according to the physical correlation of the vibration signals and the strain signals in structural mechanical response, the vibration signals and the strain signals are collaboratively denoised to obtain denoised vibration signals and strain signals; The collaborative denoising comprises the following steps: analyzing the correlation of each frequency band component of the strain signals and the vibration signals, determining a vibration signal frequency band with a correlation higher than a preset threshold value as a frequency band representing effective structural response, reconstructing effective vibration signals, and calibrating and denoising the strain signals by using the reconstructed effective vibration signals. The specific process of the collaborative denoising is described as follows: 1, when analyzing the correlation of each frequency band component of the strain signals and the vibration signals, the following method can be used: The vibration signals are decomposed into K intrinsic mode functions by using a variational mode decomposition algorithm, wherein the value of K is determined by observing the center frequency of the vibration signal spectrum; The Pearson correlation coefficient sequence of each intrinsic mode function and the strain signals is calculated; The average value of the absolute values of the Pearson correlation coefficient sequence in a time window is selected as the overall correlation measure of the intrinsic mode function and the strain signals.

[0028] The selection of K value directly affects the decomposition effect. The application does not use a simple fixed value or an empirical formula, but uses an adaptive determination method based on spectrum analysis to ensure that the decomposed modal has a clear physical meaning.

[0029] Specifically, a vibration signal of a tunnel structure under a normal operation state (such as a vehicle passing at a constant speed) is collected, and a power spectrum density diagram is obtained by performing a fast Fourier transform on the vibration signal. By observing the spectrum diagram, obvious peak points are identified, which usually correspond to the main vibration modal of the structure, the fixed frequency of equipment operation (such as a fan), or the excitation frequency of a vehicle load. The number of identified significant central frequencies is used as the initial value of K. For example, if the spectrum has obvious peaks at 12 Hz (vehicle excitation), 35 Hz (structure first-order modal), and 80 Hz (structure local vibration), K is set to 3. This method ensures that each intrinsic modal function (IMF) decomposed can capture a main frequency component.

[0030] When calculating the Pearson correlation coefficient sequence, in order to avoid that a single correlation coefficient cannot reflect dynamic changes, the application can calculate a continuous correlation coefficient sequence by using a sliding window method.

[0031] 2. When reconstructing the effective vibration signal, the following method can be used: According to a preset correlation threshold, all intrinsic modal functions with a total correlation measure greater than the threshold are screened out; The screened intrinsic modal functions are linearly superimposed to reconstruct a denoised vibration signal; The correlation threshold is determined by a correlation coefficient distribution statistical value in historical data of a structure in a normal state.

[0032] The reconstructed effective vibration signal obtained by the foregoing method will be used as a reference to denoise the strain signal. When denoising, a wavelet threshold denoising method can be used.

[0033] The specific denoising process is as follows: First, the correlation function of the reconstructed effective vibration signal and the original strain signal in the same time period is calculated to calibrate the time asynchronization that may be caused by sensor sampling delay; The correlation function can be calculated as follows: In the formula, τ is a time delay parameter, v(t) is the effective vibration signal, and s(t) is the original strain signal Then, taking the calibrated reconstructed effective vibration signal as a reference signal, the strain signal is denoised by using a threshold denoising method based on wavelet transform, wherein a Db4 wavelet is selected as a wavelet base function, and a soft threshold function is used as a threshold rule.

[0034] The Db4 wavelet has good compact support and smoothness, and can effectively capture the mutation characteristics of the strain signal. After decomposition, approximate coefficients and detail coefficients under different frequency resolutions are obtained. The shrinkage characteristics of the soft threshold function make the denoised signal smoother, avoiding the oscillation of the reconstructed signal that may be caused by the hard threshold function S3, extracting features of the denoised vibration signal and the strain signal, and performing feature fusion, inputting the fused features into a pre-trained safety evaluation model to obtain a comprehensive safety state of the tunnel structure.

[0035] When performing feature fusion, the D-S evidence theory method can be used for fusion, which can be performed in the following manner: (1) extracting the root mean square value, the peak factor and the spectral barycenter from the denoised vibration signal as the first group of evidence; (2) extracting the maximum principal strain value, the strain rate and the strain energy from the denoised strain signal as the second group of evidence; (3) based on the historical database, establishing a basic probability assignment function for the first group of evidence and the second group of evidence respectively for the three state levels of "safe", "caution" and "dangerous"; Based on the historical data, the BPA function is constructed: a large amount of historical monitoring data (including normal, caution, danger and other states) is used for statistical learning, and the membership degree relationship between the value of each feature and the safety state is established. For example, the method based on the probability density function or the fuzzy membership function can be used to construct.

[0036] For the vibration RMS feature (E1a): When RMS < a1, it is considered to strongly support the "safe" state, and a larger probability is assigned: m1({S})=0.7.

[0037] When RMS is between a1 and a2, the uncertainty increases, and it is considered to support the "caution" compound proposition: m1({S,A})=0.6.

[0038] When RMS > a3, it is considered to support the "danger" state: m1({D})=0.75.

[0039] The remaining probability is assigned to Θ, indicating unknown uncertainty: m1(Θ)=0.1.

[0040] Similarly, BPA functions m1b, m2a, m2b are established for the spectral center of gravity FC (E1b), the maximum principal strain (E2a), and the strain rate of change (E2b), respectively. The sum of all BPA values must be 1.

[0041] (4) The basic probability assignments of the two groups of evidence are fused by using the Dempster combination rule to obtain the final state probability distribution of the comprehensive evidence credibility, and the state level with the maximum probability is taken as the current safety state output.

[0042] In the feature fusion, first, the features within the same evidence source are fused: for example, the two feature evidences m1a(RMS) and m1b(FC) of the vibration signal are synthesized by using the Dempster rule to obtain the comprehensive evidence m1 of the vibration signal. Then, different evidence sources are fused: the comprehensive evidence m1 of the vibration signal and the comprehensive evidence m2 of the strain signal are synthesized again by using the Dempster rule to obtain the final BPA m final .

[0043] Before the feature fusion and safety evaluation steps, a data enhancement step is further included: A refined finite element model of the tunnel segment lining is constructed, and through simulation of the local crack and spalling damage working conditions under the action of vehicle load, water and soil pressure, simulated vibration and strain response data are obtained as the physical verification data set; A conditional generative adversarial network is constructed, the generator of which takes a normal state monitoring data segment and random noise as input, and takes a damage type label as condition, to generate simulated data with specific damage characteristics; The simulated data generated by the generator is screened and corrected using the physical verification data set to ensure its physical reasonableness, and the corrected data is added to the training set for training the safety evaluation model.

[0044] The safety evaluation model is a hybrid model of one-dimensional time series convolution network and attention mechanism; The one-dimensional time series convolution network is used to extract local features from continuous monitoring data sequences; The attention mechanism is configured to assign different weights to the feature sequences output by the convolution network to focus on the key time steps most relevant to structural damage; The final output of the hybrid model is the prediction probability of the safety state of the tunnel structure in a future preset time period.

[0045] One-dimensional time series convolution network (1D CNN) is mainly used for processing time series data, extracting local features and time series patterns, and suitable for activity recognition, signal analysis, natural language processing and other scenarios. In the present application, the one-dimensional time series convolution network is responsible for automatically learning and extracting local time series patterns related to structural damage from continuous monitoring data sequences, and then obtaining local features. The attention mechanism is used to automatically evaluate and weight the importance of these features.

[0046] As shown in Figure 2 The present application also provides a tunnel structure safety dynamic evaluation system based on data collaborative perception, comprising: A data acquisition module 10 is configured to synchronously acquire vibration signals and strain signals through vibration sensors and strain gauges arranged on the tunnel structure. A collaborative denoising module 20 is configured to collaboratively denoise the vibration signals and strain signals according to their physical correlation in structural mechanical response, to obtain denoised vibration signals and strain signals. A fusion evaluation module 30 is configured to extract features of the denoised vibration signals and strain signals, perform feature fusion, input the fused features into a pre-trained safety evaluation model, and obtain a comprehensive safety state of the tunnel structure.

[0047] In one embodiment, as shown in Figure 3 A computer device 40 is provided, comprising a memory 42, a processor 41, and a computer program 43 stored in the memory 42 and executable on the processor 41. When the processor 41 executes the computer program 43, the steps of the above-mentioned data processing method are implemented. To avoid repetition, details are not repeated here. Alternatively, when the processor 41 executes the computer program 43, the functions of each module in the above-mentioned tunnel structure safety dynamic evaluation system based on data collaborative perception are implemented. To avoid repetition, details are not repeated here.

[0048] In one embodiment, a readable storage medium is provided, which stores a computer program 43. When the processor 41 executes the computer program 43, the steps of the above-mentioned data processing method are implemented. To avoid repetition, details are not repeated here. Alternatively, when the processor 41 executes the computer program 43, the functions of each module in the above-mentioned data processing device are implemented. To avoid repetition, details are not repeated here.

[0049] 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 may include read-only memory (ROM) and programmable ROM. (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of 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 RAMbus dynamic RAM (RDRAM), etc.

[0050] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional modules, sub-modules, and units as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic evaluation of tunnel structure safety based on data collaborative perception, characterized in that, The method comprises the steps of: Synchronously collecting vibration signals and strain signals through vibration sensors and strain gauges arranged on a tunnel structure; According to the physical correlation of the vibration signals and the strain signals in structural mechanics response, the vibration signals and the strain signals are cooperatively denoised to obtain denoised vibration signals and strain signals; The cooperative denoising comprises: analyzing the correlation of each frequency band component of the strain signals and the vibration signals, determining the frequency bands of the vibration signals with a correlation higher than a preset threshold as frequency bands representing effective structural response, reconstructing effective vibration signals, and calibrating and denoising the strain signals by using the reconstructed effective vibration signals; Features of the denoised vibration signals and strain signals are extracted, and feature fusion is performed, and the fused features are input into a pre-trained safety evaluation model to obtain a comprehensive safety state of the tunnel structure.

2. The method according to claim 1, wherein, The analysis of the correlation of each frequency band component of the strain signals and the vibration signals comprises: The vibration signals are decomposed into K intrinsic mode functions by using a variational mode decomposition algorithm, wherein the value of K is determined by observing the center frequency of the vibration signal spectrum; A sequence of Pearson correlation coefficients of each intrinsic mode function and the strain signals is calculated; The average value of the absolute values of the Pearson correlation coefficient sequence within a time window is selected as the overall correlation measure of the intrinsic mode function and the strain signals.

3. The method of claim 2, wherein the method further comprises: The reconstruction of the effective vibration signals comprises: According to a preset correlation threshold, all intrinsic mode functions with an overall correlation measure greater than the threshold are screened out; The screened intrinsic mode functions are linearly superimposed to reconstruct the denoised vibration signals; The correlation threshold is determined by the correlation coefficient distribution statistical value of the structure in a normal state in historical data.

4. The method of claim 1, wherein, The calibration and denoising of the strain signals by using the reconstructed effective vibration signals comprises: The strain signals are subjected to secondary denoising processing by using a wavelet threshold denoising method with the reconstructed effective vibration signals as a reference benchmark.

5. The method of claim 1, wherein the method further comprises: The feature fusion adopts a D-S evidence theory method, and specifically comprises: Basic probability assignments are assigned to the features from the vibration signals and the strain signals; According to the combination rule of the D-S evidence theory, the confidence intervals of each safety state grade are calculated; The state grade with the highest confidence is taken as the real-time safety state of the tunnel structure.

6. The method of claim 1, wherein the method further comprises: Before the extraction of the features of the denoised vibration signals and strain signals, the feature fusion, and the input of the fused features into the pre-trained safety evaluation model to obtain the comprehensive safety state of the tunnel structure, the method further comprises the steps of: A refined finite element model of a tunnel segment lining is constructed, and simulated vibration and strain response data are obtained by simulating local crack and spalling damage conditions under the action of vehicle load and water and soil pressure, which are used as a physical verification data set; A conditional generative adversarial network is constructed, wherein the generator takes a normal state monitoring data segment and random noise as input and takes a damage type label as condition to generate simulated data with specific damage characteristics; The simulated data generated by the generator is screened and corrected by using the physical verification data set to ensure its physical reasonableness, and the corrected data is added to the training set for training the safety evaluation model.

7. The method of claim 6, wherein the method further comprises: The safety evaluation model adopts a hybrid model of one-dimensional time sequence convolution network and attention mechanism; The one-dimensional time sequence convolution network is used to extract local features from continuous monitoring data sequences; The attention mechanism is configured to assign different weights to the feature sequence output by the convolution network to focus on the key time steps most relevant to structural damage; The final output of the hybrid model is the prediction probability of the safety state of the tunnel structure in the future preset time period.

8. A data collaborative perception-based dynamic evaluation system for tunnel structure safety, comprising: a data acquisition module configured to synchronously acquire vibration signals and strain signals through vibration sensors and strain gauges arranged on the tunnel structure; a collaborative denoising module configured to collaboratively denoise the vibration signals and strain signals based on their physical correlation in structural mechanics response, and obtain denoised vibration signals and strain signals; a fusion evaluation module configured to extract features of the denoised vibration signals and strain signals, perform feature fusion, input the fused features into a pre-trained safety evaluation model, and obtain a comprehensive safety state of the tunnel structure.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the data collaborative perception-based dynamic evaluation method for tunnel structure safety according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the data collaborative perception-based dynamic evaluation method for tunnel structure safety according to any one of claims 1-7.

Images