Subway tunnel intrusion monitoring and early warning system and method

Abstract

The invention discloses a subway tunnel intrusion monitoring and early warning system and method, and relates to the technical field of subway safety monitoring. The system comprises a vibration monitoring module, a stress monitoring module, a signal analysis unit, an alarm unit and a database module, wherein the vibration monitoring module adopts a distributed vibration sensing optical cable to collect tunnel wall surface vibration signals in real time; the stress monitoring module monitors the stress change of the tunnel structure through a spirally wound and pre-embedded fiber grating array; the signal analysis unit decomposes vibration frequency band energy through wavelet packet transformation, counts a stress standard deviation in combination with a sliding time window, and performs multi-modal feature matching with an abnormal signal database by using a dynamic time warping algorithm; and the alarm unit triggers multi-stage response according to a feature matching degree threshold value or vibration-stress space-time correlation abnormity, and train speed limiting control is linked. The problems that the single-mode false alarm rate is high, multi-source interference cannot be decoupled and stress monitoring coverage is insufficient are solved, recognition and early warning of construction intrusion events are achieved, and tunnel structure safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of subway tunnel safety monitoring, and specifically to a subway tunnel intrusion monitoring and early warning system and method. Background Art

[0002] With the rapid increase in the density of the urban rail transit network, the structural safety of subway tunnels faces increasingly severe external construction threats. In recent years, unauthorized underground projects around tunnels, such as pipe jacking and foundation pit excavation, have frequently caused accidents such as structural deformation and water leakage, posing significant hidden dangers to subway operation safety. Existing monitoring technologies mainly rely on single-dimensional perception of physical quantities such as vibration and strain. For example, a distributed fiber optic vibration sensing system (DVS) is used to detect vibration abnormal events, or a resistive strain gauge is used to monitor structural deformation. However, such technologies face significant challenges in practical applications:

[0003] On the one hand, the internal environment of the tunnel is complex, and the vibration signals generated by normal operations such as train operation and equipment start-stop overlap highly with the disturbance signals of external construction activities such as mechanical tunneling and blasting in the frequency domain and time domain, making it difficult for traditional single-modal sensing systems to effectively distinguish features; on the other hand, in the development of urban underground space, miniaturized and non-blasting new construction technologies such as static pile drivers and directional drills have emerged, and the signals they generate are of low intensity and have a special frequency band distribution, posing higher requirements for the adaptability and sensitivity of existing monitoring algorithms. In addition, tunnel structure damage often has a spatio-temporal cumulative effect, and early warning can only be achieved through dynamic correlation analysis of multi-physical field data, while traditional systems have obvious shortcomings in data fusion and decision-making mechanisms.

[0004] Therefore, there are the following technical bottlenecks:

[0005] 1) Traditional vibration sensing networks are difficult to effectively distinguish the basic vibration caused by train operation from the high-frequency impact signals of external construction activities, resulting in feature confusion;

[0006] 2) Discrete stress monitoring nodes cannot accurately capture local stress mutations and propagation paths of tunnel linings due to insufficient spatial coverage density;

[0007] 3) Existing signal analysis models mostly adopt a static threshold discrimination mechanism, lacking the ability to dynamically decouple the coupling effect of multi-source disturbances and being difficult to adapt to the non-linear signal attenuation characteristics under complex geological conditions. Summary of the Invention

[0008] In view of the above problems, the present invention proposes a subway tunnel intrusion monitoring and early warning system and method to solve the problems of vibration feature confusion, difficulty in capturing stress propagation, and limited adaptability to complex working conditions; to achieve the above object, a subway tunnel intrusion monitoring and early warning system of the present invention includes:

[0009] A vibration monitoring module, which consists of a vibration sensing optical cable laid along the inner wall of the tunnel and a vibration signal demodulation unit, is used to collect vibration signals on the tunnel wall in real time and demodulate them into vibration data;

[0010] A stress monitoring module, which consists of a stress sensing optical cable embedded in the tunnel concrete structure and a stress signal demodulation unit, is used to monitor the stress change of the tunnel structure in real time and demodulate it into stress data;

[0011] A signal analysis unit, which is connected to the vibration signal demodulation unit and the stress signal demodulation unit, is used to extract multi-modal features from the vibration data and stress data, and perform similarity matching with the feature data in the abnormal signal database based on the dynamic time warping algorithm, and output the feature matching degree;

[0012] An alarm unit, which is communicatively connected to the signal analysis unit. If the feature matching degree of the vibration data or stress data exceeds a preset threshold, or the spatio-temporal correlation between the two is abnormal, a multi-level alarm is triggered;

[0013] A database module, which stores an abnormal signal database and a historical alarm database. The abnormal signal database includes: an abnormal vibration feature library and an abnormal stress feature library. The abnormal vibration feature library and the abnormal stress feature library are constructed based on the medium and high frequency vibration features and low frequency stress features extracted by the wavelet packet transform and the short-time Fourier transform algorithm.

[0014] Preferably, the vibration sensing optical cable is fixed in segments along the axial direction of the tunnel, with an interval of 5-10 meters for each segment, and the vibration sensing optical cables of adjacent segments are connected in series through optical fiber jumpers to form a distributed vibration sensing network.

[0015] Preferably, the stress sensing optical cable is embedded in the steel bar framework of the tunnel concrete structure in a spiral winding manner, and a stress monitoring node is set every 20-30 meters along the longitudinal direction of the tunnel. Each stress monitoring node is equipped with an independent fiber Bragg grating sensor.

[0016] Preferably, the signal analysis unit includes:

[0017] An interference filtering module, which is used to separate the low-frequency vibration signals generated by the normal passage of vehicles through the empirical mode decomposition algorithm;

[0018] A vibration feature extraction module, which performs time-frequency analysis on the medium and high frequency vibration signals by using the Hilbert-Huang transform, extracts the instantaneous frequency variance as the vibration feature value, and performs dynamic comparison with the feature data in the abnormal vibration feature library;

[0019] A stress feature extraction module, which is used to calculate the standard deviation and kurtosis of the stress data by using a sliding time window, and perform dynamic comparison with the feature data in the abnormal stress feature library;

[0020] The abnormality judgment module is used to set a preset threshold. The preset threshold is when the vibration feature matching degree exceeds 85%, the duration is greater than 10 seconds, or the stress standard deviation exceeds 2 times the historical baseline value, it is judged as a risk intrusion event and the feature matching degree is output.

[0021] Preferably, the alarm unit adopts a multi-level response mechanism:

[0022] The secondary alarm trigger condition is that the vibration data or stress data exceeds the preset threshold alone, triggering the local sound and light alarm;

[0023] The first-level alarm triggering condition is that the vibration data and stress data are abnormally correlated in the time and space dimensions, the distance is less than 50 meters and the time difference is less than 30 seconds. At the same time, the alarm information is pushed to the monitoring center and the train speed limit instruction is linked.

[0024] Preferably, the database module further includes a self-learning unit, specifically:

[0025] Based on the data in the historical alarm database, the matching threshold of the abnormal vibration feature library is dynamically optimized through the neural network model;

[0026] Perform cluster analysis on unidentified vibration signals, generate new abnormal signal categories and update the abnormal vibration feature library.

[0027] Preferably, the vibration signal demodulation unit and the stress signal demodulation unit both use a phase-sensitive optical time domain reflectometer with a sampling frequency not less than 1 kHz and a spatial resolution less than 1 meter.

[0028] Preferably, an early warning method of a subway tunnel intrusion monitoring and early warning system comprises the following steps:

[0029] Step S100: pre-burying stress sensing optical cables during the tunnel construction phase, and fixing vibration sensing optical cables along the inner wall during the operation phase to form a dual-mode monitoring network;

[0030] Step S200: collecting vibration data and stress data in real time through the vibration signal demodulation unit and the stress signal demodulation unit, and transmitting them to the signal analysis unit;

[0031] Step S300: The signal analysis unit uses wavelet packet transform to perform frequency band energy decomposition on the vibration signal, extracts the energy proportion of the 8-16kHz frequency band as a preliminary abnormality judgment, and uses the dynamic time warping algorithm to perform similarity matching between the decomposed frequency band energy distribution and the feature data in the abnormal vibration feature library. When the energy proportion exceeds 3 times the historical baseline value and the similarity matching degree is ≥90%, it is marked as a vibration abnormality event and the abnormal vibration location is confirmed;

[0032] Step S400: The signal analysis unit calculates the standard deviation of the stress data using a sliding time window, and performs similarity matching on the stress data within the window and the feature data in the abnormal stress feature library through the dynamic time warping algorithm. When the standard deviation of three consecutive time windows exceeds twice the historical baseline value and the similarity matching degree exceeds the preset threshold ≥ 85%, it is marked as a stress abnormal event, and the stress abnormal position is confirmed;

[0033] Step S500: If the vibration abnormal event and the stress abnormal event occur in association in the spatio-temporal dimension, the alarm unit is triggered and a comprehensive risk assessment report is generated, where:

[0034] When the data of a single sensor exceeds the threshold, a secondary alarm is triggered, and only the local audible and visual alarm is activated;

[0035] When the vibration data and the stress data are abnormally associated in the spatio-temporal dimension, a primary alarm is triggered, and the train speed limit control and the emergency response plan are linked.

[0036] Preferably, the step S300 further includes:

[0037] S301. During the train passing gap, start the high-sensitivity monitoring mode and increase the vibration signal sampling frequency to 5 kHz;

[0038] S302. Separate the environmental noise through the independent component analysis algorithm and retain the construction feature signal in the 0.52 kHz frequency band.

[0039] Preferably, the criterion for the associated occurrence in the spatio-temporal dimension in the step S500 includes:

[0040] The straight-line distance between the vibration abnormal position and the stress abnormal position is less than 50 meters;

[0041] The time interval between the vibration abnormal event and the stress abnormal event is less than 30 seconds;

[0042] If the above conditions are met simultaneously, it is determined as a high-risk intrusion construction, a primary alarm is triggered, and the emergency response plan is activated.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] The present invention achieves breakthroughs through the following technologies:

[0045] 1) Collaborative analysis of vibration and stress dual-modal data, effectively separating environmental interference and intrusion feature signals;

[0046] 2) High-density distributed sensing layout combined with a spiral-wound grating array, significantly improving the capture sensitivity of micro-deformations and stress gradients;

[0047] 3) The self - learning algorithm realizes the dynamic optimization of multi - physical - field correlation features, enhancing the adaptability of the system to new construction modes and the reliability of early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the principle of the present invention.

[0049] Figure 2 It is a schematic diagram of the device of the subway tunnel intrusion monitoring and early warning system. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0051] Please refer to Figure 1 - Figure 2 , the system of the present invention breaks through the limitations of traditional single - sensor monitoring through the collaborative monitoring of the vibration and stress dual - mode sensing network.

[0052] The vibration monitoring module consists of a vibration - sensing optical cable laid along the inner wall of the tunnel and a vibration signal demodulation unit, and is used to collect the vibration signals on the tunnel wall surface in real time and demodulate them into vibration data;

[0053] The stress monitoring module consists of a stress - sensing optical cable embedded in the tunnel concrete structure and a stress signal demodulation unit, and is used to monitor the stress changes of the tunnel structure in real time and demodulate them into stress data;

[0054] The signal analysis unit is connected to the vibration signal demodulation unit and the stress signal demodulation unit, and is used to extract multi - modal features from the vibration data and stress data, and perform similarity matching based on the dynamic time warping algorithm and the feature data in the abnormal signal database, and output the feature matching degree;

[0055] The alarm unit is communicatively connected to the signal analysis unit. If the feature matching degree of the vibration data or stress data exceeds a preset threshold, or the spatio - temporal correlation between the two is abnormal, a multi - level alarm is triggered;

[0056] The database module stores an abnormal signal database and a historical alarm database. The abnormal signal database includes: an abnormal vibration feature library and an abnormal stress feature library. The abnormal vibration feature library and the abnormal stress feature library are constructed based on the medium - high - frequency vibration features and low - frequency stress features extracted by the wavelet packet transform and short - time Fourier transform algorithms.

[0057] The vibration monitoring module uses a distributed vibration sensing optical cable, such as a Φ0.9mm single-mode optical fiber, etc., which is continuously laid along the entire length of the tunnel inner wall. Through the phase-sensitive optical time domain reflectometry (Φ-OTDR), the wall vibration signals are captured in real time. The vibration signal demodulation unit is built-in with a high-speed optoelectronic conversion module, such as a 1550nm laser light source, which converts the optical phase change into a time-domain voltage signal. The sampling frequency is not less than 1kHz to ensure the complete capture of high-frequency vibration abnormal events. The stress monitoring module uses a fiber Bragg grating (FBG) array sensor, which is embedded inside the tunnel secondary lining concrete. Through the wavelength demodulation technology, the structural strain change is measured in real time, and the monitoring resolution reaches ±1με, which can effectively identify the structural stress redistribution caused by external construction. The signal analysis unit is equipped with a multi-core DSP processor, which integrates the wavelet packet transform algorithm to decompose the vibration signal into 8 layers, extracts the energy distribution characteristics of 16 frequency bands, and at the same time uses the sliding window statistical method to calculate the standard deviation, kurtosis and other time-domain characteristics of the stress data, realizing the feature-level fusion of vibration data and stress data. The alarm unit adopts a multi-level response mechanism: when the data of a single sensor exceeds the threshold, a secondary alarm (yellow warning) is triggered; when the vibration data and the stress data are abnormally correlated in the space-time dimension, a primary alarm (red warning) is triggered, and command linkage is realized with the subway signal system through the OPC-UA protocol. The database module uses a time-series database, such as InfluxDB to store historical alarm data, and constructs an abnormal feature automatic extraction model based on the convolutional neural network (CNN) to realize the dynamic update of the feature library and the optimization of the matching threshold. The invention effectively solves the technical problems of high false alarm rate of a single sensor and the inability of traditional algorithms to decouple multi-source interference, realizes the accurate identification and real-time warning of construction intrusion events, reduces the false alarm rate and the response delay, and significantly improves the safety guarantee ability of the tunnel structure.

[0058] The vibration sensing optical cable is fixed in segments along the tunnel axis, with an interval of 5-10 meters for each segment. And the vibration sensing optical cables of adjacent segments are connected in series through fiber optic jumpers to form a distributed vibration sensing network. The physical deployment of the vibration sensing optical cable directly determines the monitoring sensitivity and anti-interference ability. This system adopts a segmented fixed strategy, uses epoxy resin structural adhesive to bond the optical cable at a height of 1.2m from the rail surface on the tunnel sidewall at an interval of 5-10 meters, and at the same time avoids the area affected by the train piston wind. FC / APC type fiber optic jumpers are installed at both ends of each segment of the optical cable, and the signals of adjacent sections are connected in series through mechanical splicing to form a distributed sensing network with a total length of up to 20km. This design has two major advantages: First, segmented fixation can avoid the cumulative error of the optical cable deformation caused by temperature change; Second, the jumper connection method is convenient for the quick replacement of the faulty section, improving the operation and maintenance efficiency. The outer layer of the optical cable is wrapped with a polyurethane compressive sheath with a thickness of ≥2mm to prevent signal attenuation caused by the impact of ballast stones. Practical tests show that this deployment scheme can still stably detect impact energy events above 0.1mJ in the vibration environment of train passing.

[0059] The stress sensing optical cable is embedded in the steel bar framework of the tunnel concrete structure in a spiral winding manner, and a stress monitoring node is arranged every 20 - 30 meters along the longitudinal direction of the tunnel. Each of the stress monitoring nodes is equipped with an independent fiber Bragg grating sensor. The reliability of the stress monitoring network depends on the coupling degree between the sensor and the concrete structure. In the concrete pouring stage of the tunnel lining, the stress sensing optical cable is wound in a spiral manner with a pitch of 50 cm and tied to the outside of the main steel bars to ensure that the grating array deforms synchronously with the steel bar framework. Four groups of orthogonally arranged fiber Bragg grating sensors with a central wavelength of 1550 nm ± 0.2 nm are set at each monitoring node to measure the circumferential and longitudinal strain components respectively. The setting of the node spacing of 20 meters is based on the finite element simulation results. When the change in the tunnel diameter caused by external construction exceeds 0.5 mm, the change in the strain gradient between adjacent nodes is sufficient to be detected.

[0060] The signal analysis unit includes: an interference filtering module, which is used to separate the low - frequency vibration signals generated by normal vehicle passage through the empirical mode decomposition algorithm; a vibration feature extraction module, which uses the Hilbert - Huang transform to perform time - frequency analysis on the medium - high - frequency vibration signals, extracts the instantaneous frequency variance as the vibration feature value, and dynamically compares it with the feature data in the abnormal vibration feature library; a stress feature extraction module, which is used to calculate the standard deviation and kurtosis of the stress data using a sliding time window and dynamically compares it with the feature data in the abnormal stress feature library; an abnormal determination module, which is used to set a preset threshold. When the vibration feature matching degree exceeds 85% and the duration is greater than 10 seconds or the stress standard deviation exceeds 2 times the historical baseline value, it is determined as a risk intrusion event, and the feature matching degree is output.

[0061] The core of the signal analysis unit is to distinguish normal operation vibration from intrusion construction signals. The interference filtering module uses an improved empirical mode decomposition (EMD) algorithm: First, the original vibration signal is decomposed into intrinsic mode functions (IMFs), and the IMF components with frequencies lower than 20 Hz are removed, which correspond to train wheel - rail vibrations, and the high - frequency components are retained for subsequent analysis. The vibration feature extraction module uses the Hilbert - Huang transform (HHT) to calculate the marginal spectrum energy entropy of the residual signal, and at the same time extracts the instantaneous frequency variance as an auxiliary criterion and compares it with the feature data in the abnormal vibration feature library; the stress feature extraction module dynamically compares the stress data with the feature data in the abnormal stress feature library. The abnormal determination module uses the dynamic time warping (DTW) algorithm to calculate the similarity between the real - time feature vector and the abnormal signal database. When the similarity exceeds 85% and lasts for more than 10 seconds or the stress standard deviation exceeds 2 times the historical baseline value, it is determined as intrusion behaviors such as drilling and blasting. This algorithm combination improves the recognition accuracy of construction activities.

[0062] The alarm unit includes: a local alarm installed in the tunnel machine room, which alarms through sound and light signals; a remote linkage module communicatively connected to the monitoring center, which real-time pushes the alarm location, signal strength, and risk level to the terminal device; and an emergency control interface linked to the subway signal system to trigger a train speed limit or shutdown command.

[0063] The multi-level response mechanism of the alarm unit takes into account the timeliness of early warning and the safety of disposal. The trigger condition for the secondary alarm is that the vibration data or stress data alone exceeds the preset threshold, triggering the local sound and light alarm. The trigger condition for the primary alarm is that the vibration data and stress data are abnormally correlated in the space-time dimension, the distance is less than 50 meters, and the time difference is less than 30 seconds. At the same time, the alarm information is pushed to the monitoring center and the train speed limit command is linked. The local alarm uses a 120dB sound and light alarm installed on the side wall of the emergency evacuation passage in the tunnel. When triggered, the red LED warning light is synchronously lit, and the visible distance is ≥300m. The remote linkage module transmits the alarm information to the monitoring center through LoRa wireless networking. The data packet includes the abnormal position coordinates, signal strength, and risk level. The risk level can be calculated based on the energy entropy and the stress exceeding multiple. The monitoring software automatically retrieves the video monitoring images of the corresponding section for manual review. The emergency control interface adopts a hard-wired connection method and is directly connected to the interlocking device of the subway signal system. When the primary alarm is triggered, an RS-485 command is sent to limit the speed of the train within 500m ahead to 25km / h. If no manual confirmation is received within 10 seconds, emergency braking is executed. The actual measurement shows that the full-process delay from the occurrence of the abnormality to the issuance of the train speed limit command is less than 3 seconds.

[0064] The database module also includes a self-learning unit, specifically: based on the data in the historical alarm database, the matching threshold of the abnormal vibration feature library is dynamically optimized through a neural network model; the unrecognized vibration signals are subjected to clustering analysis to generate new abnormal signal categories and update the abnormal vibration feature library.

[0065] The self-learning unit improves the system's adaptability by continuously optimizing the feature library. Its work process includes: 1) Collect the true positive samples, that is, the intrusion events confirmed by humans, and false positive samples, that is, false alarm events, in the historical alarm data, and extract their feature vectors to construct a training set; 2) Train a one-dimensional convolutional neural network. The input layer receives a 128-dimensional feature vector, and through 3 convolutional layers, the filter sizes are 5, 3, and 3 respectively to extract abstract features. The output layer uses the Sigmoid function to calculate the false alarm probability under the current threshold; 3) With the constraint condition that the false alarm rate is lower than 5%, the feature matching threshold is adjusted inversely. For the unrecognized new vibration modes, the DBSCAN clustering algorithm is used for unsupervised learning. When the number of samples in a certain cluster exceeds 100, a new feature category is automatically generated and added to the feature library, and its false alarm rate decreases.

[0066] Both the vibration signal demodulation unit and the stress signal demodulation unit adopt the Φ-OTDR technical solution, and its core parameter configuration is as follows: the output power of the laser is 10 mW, ensuring that the signal-to-noise ratio ≥ 35 dB, the pulse width is 10 ns, corresponding to a spatial resolution of 0.5 m, and the repetition frequency is 1 kHz, meeting the capture requirement of the sampling theorem for the highest vibration frequency of 16 kHz. The gain adjustable range of the photodetector is 60 - 80 dB. The demodulation unit is built-in with a temperature compensation module, which measures the phase drift caused by environmental temperature changes through a reference optical fiber and cancels it during the digital signal processing stage, making the strain measurement accuracy stable within ±2 με.

[0067] The implementation of this method is divided into five key steps:

[0068] Step S100, during the tunnel construction stage, synchronously install the stress sensing optical cable with the steel reinforcement cage to ensure that the sensor array forms a mechanical coupling with the concrete structure, and fix the vibration sensing optical cable along the inner wall during the operation period to form a dual-mode monitoring network. The vibration optical cable during the operation period is installed with detachable fixtures to avoid damage to the existing structure.

[0069] In step S200, the vibration data and stress data are collected in real time through the vibration signal demodulation unit and the stress signal demodulation unit, and transmitted to the signal analysis unit. The vibration signal demodulation unit obtains the time-domain waveform at a sampling rate of 1 kHz, and passes it through a FIR band-pass filter with a passband of 0.5 - 20 kHz to suppress low-frequency interference; the stress signal demodulation unit collects the central wavelength shift of each FBG sensor every 5 seconds and converts it into a micro-strain value.

[0070] Step S300: The signal analysis unit uses wavelet packet transform to decompose the energy of the vibration signal in frequency bands, extracts the energy ratio in the 8 - 16 kHz frequency band as a preliminary abnormal judgment, and performs similarity matching on the decomposed frequency band energy distribution and the characteristic data in the abnormal vibration characteristic library through the dynamic time warping algorithm. When the energy ratio exceeds 3 times the historical baseline value and the similarity matching degree ≥ 90%, it is marked as a vibration abnormal event, and the vibration abnormal position is confirmed;

[0071] Step S400: The signal analysis unit calculates the standard deviation of the stress data using a sliding time window, and performs similarity matching on the stress data within the window and the characteristic data in the abnormal stress characteristic library through the dynamic time warping algorithm. When the standard deviation of three consecutive time windows exceeds 2 times the historical baseline value and the similarity matching degree exceeds the preset threshold ≥ 85%, it is marked as a stress abnormal event, and the stress abnormal position is confirmed;

[0072] Step S500: If a vibration anomaly event and a stress anomaly event occur in a spatio-temporal correlation, trigger the alarm unit and generate a comprehensive risk assessment report, where: when the data of a single sensor exceeds the threshold, a secondary alarm is triggered, and only the local audible and visual alarm is activated; when the vibration data and the stress data are abnormally correlated in space and time, a primary alarm is triggered, and the train speed limit control and the emergency response plan are linked.

[0073] The criteria for spatio-temporal correlation occurrence include: the longitudinal distance difference between the vibration and stress anomaly points along the tunnel is less than 50 meters. Considering the stress wave propagation range, the time difference is less than 30 seconds, and at the same time, considering the upper limit of the delay of the vibration force and the structural response caused by construction activities. If both are satisfied, a comprehensive report including the position coordinates, energy level, and structural damage risk index is generated and transmitted to the emergency command platform through the 5G private network.

[0074] During the train passing interval, which is also the night maintenance window period, the system automatically switches to the high-sensitivity mode: In step S301, the vibration signal sampling rate is increased to 5 kHz, and at the same time, the APD gain adaptive adjustment function is enabled to increase the signal-to-noise ratio of weak signals by 15 dB. In step S302, the FastICA algorithm is used to perform blind source separation on the multi-channel vibration signals, and the number of estimated independent components is 3, corresponding to environmental noise, residual train vibration, and construction signals. Through power spectral density analysis, the components in the frequency band of 0.5 - 2 kHz are retained, including the impact drill characteristic frequency of 1250 Hz ± 200 Hz, and the remaining components are regarded as noise and filtered out. This mode can increase the detection sensitivity of construction signals to 0.05 mJ while ensuring a reduced false alarm rate.

[0075] Spatio-temporal correlation analysis is the core innovation of this method to reduce false alarms, and the specific criteria include:

[0076] 1) Divided by the hundred-meter mark, the vibration anomaly point and the stress anomaly point need to be located in the same tunnel section, and the straight-line distance is less than 50 meters. Considering the influence of the tunnel curvature, the actual path difference is corrected by ±10%.

[0077] 2) The time interval between the start time of the vibration anomaly event and the start time of the stress anomaly event is less than 30 seconds, corresponding to the stress wave propagation speed of 4000 m / s in concrete and the maximum propagation distance of 120 m.

[0078] 3) The energy entropy of the vibration signal is positively correlated with the increase in the stress standard deviation, and the Pearson coefficient > 0.6.

[0079] When the first two points are satisfied or all three are satisfied simultaneously, it is determined as a high-risk intrusion event, triggering a primary alarm and starting the plan: The dispatching center dispatches personnel to conduct on-site verification, and the train runs at a speed limit until the alarm is lifted, reducing the false alarm rate.

[0080] Those skilled in the art can make equivalent modifications without departing from the principle of the present invention. For example, replacing Φ-OTDR with BOTDR technology, optimizing the feature library using other machine learning algorithms, etc. These variations all fall within the protection scope of the present invention. For those skilled in the art, various corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A subway tunnel intrusion monitoring and early warning system, characterized in that: include: The vibration monitoring module consists of a vibration sensing optical cable laid along the inner wall of the tunnel and a vibration signal demodulation unit, which is used to collect the vibration signal of the tunnel wall in real time and demodulate it into vibration data; The stress monitoring module consists of a stress sensing optical cable pre-buried in the tunnel concrete structure and a stress signal demodulation unit, which is used to monitor the stress changes of the tunnel structure in real time and demodulate them into stress data; A signal analysis unit, wherein the signal analysis unit is connected to the vibration signal demodulation unit and the stress signal demodulation unit, and is used to extract multimodal features from the vibration data and the stress data, and perform similarity matching with feature data in an abnormal signal database based on a dynamic time warping algorithm, and output a feature matching degree; An alarm unit, which is in communication with the signal analysis unit and triggers a multi-level alarm if the feature matching degree of the vibration data or the stress data exceeds a preset threshold, or if the temporal and spatial correlation between the two is abnormal; A database module stores an abnormal signal database and a historical alarm database. The abnormal signal database includes an abnormal vibration feature library and an abnormal stress feature library. The abnormal vibration feature library and the abnormal stress feature library are constructed based on medium and high frequency vibration features and low frequency stress features extracted by wavelet packet transform and short-time Fourier transform algorithms.

2. The subway tunnel intrusion monitoring and early warning system according to claim 1 is characterized in that: The vibration sensing optical cable is fixed in sections along the axial direction of the tunnel, with each section spaced 5-10 meters apart, and the vibration sensing optical cables of adjacent sections are connected in series through optical fiber jumpers to form a distributed vibration sensing network.

3. The subway tunnel intrusion monitoring and early warning system according to claim 1 is characterized in that: The stress sensing optical cable is pre-buried in the steel skeleton of the tunnel concrete structure in a spiral winding manner, and a stress monitoring node is set every 20 meters to 30 meters along the longitudinal direction of the tunnel, and each stress monitoring node is equipped with an independent fiber grating sensor.

4. The subway tunnel intrusion monitoring and early warning system according to claim 1 is characterized in that: The signal analysis unit comprises: Interference filtering module, used to separate low-frequency vibration signals generated by normal vehicle traffic through empirical mode decomposition algorithm; The vibration feature extraction module uses Hilbert-Huang transform to perform time-frequency analysis on medium and high frequency vibration signals, extracts the instantaneous frequency variance as the vibration feature value, and dynamically compares it with the feature data of the abnormal vibration feature library; The stress feature extraction module is used to calculate the standard deviation and kurtosis of stress data using a sliding time window, and dynamically compare it with the feature data of the abnormal stress feature library; The abnormality judgment module is used to set a preset threshold. The preset threshold is when the vibration feature matching degree exceeds 85%, the duration is greater than 10 seconds, or the stress standard deviation exceeds 2 times the historical baseline value, it is judged as a risk intrusion event and the feature matching degree is output.

5. The subway tunnel intrusion monitoring and early warning system according to claim 1 is characterized in that: The alarm unit adopts a multi-level response mechanism: The secondary alarm trigger condition is that the vibration data or stress data exceeds the preset threshold alone, triggering the local sound and light alarm; The first-level alarm triggering condition is that the vibration data and stress data are abnormally correlated in the time and space dimensions, the distance is less than 50 meters and the time difference is less than 30 seconds. At the same time, the alarm information is pushed to the monitoring center and the train speed limit instruction is linked.

6. The subway tunnel intrusion monitoring and early warning system according to claim 1 is characterized in that: The database module also includes a self-learning unit, specifically: Based on the data in the historical alarm database, the matching threshold of the abnormal vibration feature library is dynamically optimized through the neural network model; Perform cluster analysis on unidentified vibration signals, generate new abnormal signal categories and update the abnormal vibration feature library.

7. The subway tunnel intrusion monitoring and early warning system according to claim 1 is characterized in that: The vibration signal demodulation unit and the stress signal demodulation unit both use a phase-sensitive optical time domain reflectometer with a sampling frequency of no less than 1 kHz and a spatial resolution of less than 1 meter.

8. An early warning method using the subway tunnel intrusion monitoring and early warning system according to any one of claims 1 to 7, characterized in that: The following steps are involved: Step S100: pre-burying stress sensing optical cables during the tunnel construction phase, and fixing vibration sensing optical cables along the inner wall during the operation phase to form a dual-mode monitoring network; Step S200: collecting vibration data and stress data in real time through the vibration signal demodulation unit and the stress signal demodulation unit, and transmitting them to the signal analysis unit; Step S300: The signal analysis unit uses wavelet packet transform to perform frequency band energy decomposition on the vibration signal, extracts the energy proportion of the 8-16kHz frequency band as a preliminary abnormality judgment, and uses the dynamic time warping algorithm to perform similarity matching between the decomposed frequency band energy distribution and the feature data in the abnormal vibration feature library. When the energy proportion exceeds 3 times the historical baseline value and the similarity matching degree is ≥90%, it is marked as a vibration abnormality event and the abnormal vibration location is confirmed; Step S400: The signal analysis unit calculates the standard deviation of the stress data using a sliding time window, and performs similarity matching between the stress data in the window and the feature data in the abnormal stress feature library through a dynamic time warping algorithm. When the standard deviation of three consecutive time windows exceeds twice the historical baseline value and the similarity matching degree exceeds a preset threshold value ≥ 85%, it is marked as a stress abnormality event and the stress abnormality location is confirmed; Step S500: If the abnormal vibration event and the abnormal stress event occur in association with each other in the time and space dimensions, the alarm unit is triggered and a comprehensive risk assessment report is generated, wherein: When the data of a single sensor exceeds the threshold, a secondary alarm is triggered, which activates only the local sound and light alarm; When the spatiotemporal correlation between vibration data and stress data is abnormal, a first-level alarm is triggered, linking the train speed limit control and emergency response plan.

9. The early warning method according to claim 8, characterized in that: The step S300 also includes: S301, when there is a gap between trains, start the high-sensitivity monitoring mode and increase the vibration signal sampling frequency to 5kHz; S302. Separate the environmental noise through the independent component analysis algorithm and retain the construction characteristic signal in the 0.52kHz frequency band.

10. The early warning method according to claim 8, characterized in that: The criteria for the occurrence of association in the spatiotemporal dimension in step S500 include: The straight-line distance between the abnormal vibration position and the abnormal stress position is less than 50 meters; The time interval between the abnormal vibration event and the abnormal stress event is less than 30 seconds; If all the above conditions are met at the same time, it will be judged as high-risk intrusion construction, triggering a level one alarm and initiating an emergency response plan.

Images