Tunnel flood discharge safety monitoring method and system based on voiceprint recognition

Abstract

The invention discloses a tunnel flood discharge safety monitoring method and system based on voiceprint recognition, and relates to the field of flood discharge tunnel health safety monitoring. The method comprises the following steps: deploying an optical fiber voiceprint sensing network, and obtaining initial voiceprint data; processing the initial voiceprint data to obtain abnormal voiceprint positioning data and feature value data; and performing flood discharge tunnel health assessment according to the positioning data and the characteristic value data. According to the method, optimization algorithm processing is carried out through the voiceprint signals collected by the sensing network, and the efficiency and accuracy of monitoring and judging the abnormal problem of the flood discharge tunnel are improved.

Description

Technical Field

[0001] The present invention relates to the field of health and safety monitoring of flood discharge tunnels, and in particular to a method and system for monitoring the health and safety of flood discharge tunnels based on voiceprint recognition. Background Art

[0002] Floodway tunnels are crucial facilities in water conservancy projects, primarily used to regulate reservoir water levels and discharge floodwaters. Because these tunnels safeguard the safety of these projects, monitoring their safety is crucial. Existing monitoring technologies rely primarily on traditional point sensors, such as pressure gauges and strain gauges. While convenient, these sensors suffer from poor durability and stability, are prone to failure, struggle to cover long tunnels, and offer low accuracy in locating damaged tunnels and limited real-time performance.

[0003] On the other hand, fiber optic distributed detection has the characteristics of technical precision, high efficiency, durability, and strong stability. With the development of fiber optic distributed detection technology, there have been studies on the application of distributed optical cable detection devices to high-speed railway tunnels and high-speed tunnels for data detection, such as the Chinese patent with patent number: CN116519049A. However, when the tunnel is discharging flood water, the detection equipment often has problems of high humidity and high pressure caused by high-speed water flow, as well as special problems such as cavitation, deformation, and cracks. Therefore, there are many differences between the environment, instrument monitoring requirements, data collection methods, and instrument applicability of flood discharge tunnels and traditional tunnels. These differences hinder the development and application of fiber optic detection technology in flood discharge tunnels. Therefore, it is necessary to design a distributed fiber optic safety monitoring technology suitable for the working environment of flood discharge tunnels. Summary of the Invention

[0004] The purpose of the present invention is to provide a tunnel flood discharge safety monitoring method and system based on voiceprint recognition, to achieve full coverage, real-time accurate assessment of flood discharge tunnel health and prediction of flood discharge tunnel structure degradation trend based on long-term voiceprint data prediction.

[0005] In a first aspect, the present invention provides a method for monitoring tunnel flood discharge safety based on voiceprint recognition, comprising the following steps: Deploy a fiber-optic voiceprint sensing network to obtain initial voiceprint data; Process the initial voiceprint data to obtain abnormal voiceprint positioning data and characteristic value data; Conduct health assessment of flood discharge tunnels based on location data and eigenvalue data; The processing of the initial voiceprint data includes the following steps: Perform noise reduction on voiceprint signals and filter out abnormal voiceprints; According to the abnormal voiceprint, the time difference method is used to determine the location information of the abnormal voiceprint; The abnormal voiceprint is subjected to wavelet packet signal analysis according to the position information to obtain a characteristic value of the abnormal voiceprint.

[0006] As a preferred solution of the present invention, the deployment of the optical fiber voiceprint sensing network includes the following steps: The sensing optical fiber is laid along the inner wall of the flood discharge tunnel lining or pre-buried in the lining structure; The sensing optical fiber located in the high-speed water flow area is pre-buried in the groove and reinforced by filling with protective materials or metal sleeves; The sensing optical fiber located in the low flow velocity area or the water flow mutation area is laid on the side wall or top of the tunnel.

[0007] As a preferred solution of the present invention, the optical fiber voiceprint sensing network includes distributed sensing optical fibers and distributed optical fiber system equipment; the distributed optical fiber system equipment includes a laser, a modulator, a detector, a signal processor and a cloud health monitoring system.

[0008] As a preferred solution of the present invention, the noise reduction processing of the voiceprint signal is specifically to eliminate water flow noise by spectral subtraction; the elimination of water flow noise by spectral subtraction includes the steps of: Obtaining a noisy signal spectrum based on the voiceprint signal; Subtract the noise spectrum from the noisy signal spectrum to obtain the pure signal spectrum; Zero out the negative values ​​of the pure signal spectrum: Reconstruct the frequency domain signal based on the phase information of the clean signal spectrum and the noisy signal spectrum; Performing an inverse Fourier transform on the frequency domain signal to obtain a time domain signal; The time domain signals of each frame are overlapped and added to obtain the noise-reduced voiceprint data.

[0009] As a preferred embodiment of the present invention, the pure signal spectrum is expressed as: |X(f)|=|Y(f)|- |N(f)|, Among them, |X(f)| is the pure signal spectrum, |N(f)| is the noise spectrum, |Y(f)| is the noisy signal spectrum, is the over-reduction factor.

[0010] As a preferred embodiment of the present invention, the screening of abnormal voiceprints includes the following steps: Calculate the target frequency band energy and background frequency band energy of the denoised voiceprint data; Obtaining an energy ratio based on the target frequency band energy and the background frequency band energy; Filter abnormal voiceprints based on energy ratio.

[0011] As a preferred solution of the present invention, the method of determining the location information of abnormal voiceprints using the time difference method includes the following steps: Selecting a first sensor and a second sensor that detect an abnormal voiceprint; Acquire a data time difference according to position information of the first sensor and the second sensor; Constructing a first distance relationship based on the position information of the abnormal voiceprint and the position information of the first sensor, and constructing a second distance relationship based on the position information of the abnormal voiceprint and the position information of the second sensor; The location information of the abnormal voiceprint is determined according to the data time difference, the first distance relationship expression, and the second distance relationship expression.

[0012] As a preferred solution of the present invention, the characteristic value of the abnormal voiceprint is wavelet packet energy entropy; and the wavelet packet signal analysis of the abnormal voiceprint according to the position information comprises the following steps: Decompose abnormal voiceprints into several sub-bands; Calculate the energy value of each sub-band, and obtain the normalized energy distribution probability corresponding to each sub-band according to the energy value; The wavelet packet energy entropy is calculated based on the normalized energy distribution probability corresponding to each sub-band.

[0013] As a preferred solution of the present invention, the health assessment of the flood discharge tunnel based on the positioning data and the eigenvalue data is implemented by the ResNet-1D model; The ResNet-1D model is constructed as follows: The positioning data and eigenvalue data are integrated to establish a model dataset: For the positioning data of each optical fiber, set the spacing threshold range to filter out the optical fibers that meet the conditions; Using the positioning data of the screened optical fibers and a selected spatial positioning algorithm, the three-dimensional coordinates of the region are calculated; Performing mean processing on the characteristic value data that matches the positioning data of the screened optical fiber to obtain energy data; Match the three-dimensional coordinates with the energy data to obtain a data sequence, and add a type label to each data sequence according to the energy data to obtain a model data set; The ResNet-1D model is trained based on the input model dataset and the classification results are used as output.

[0014] In a second aspect, the present invention provides a tunnel flood discharge safety monitoring system based on voiceprint recognition, which is applied to the above-mentioned tunnel flood discharge safety monitoring method based on voiceprint recognition, including a distributed sensing optical fiber and a distributed optical fiber system device; the distributed optical fiber system device includes a laser, a modulator, a detector, a signal processor, and a cloud health monitoring system; The distributed sensing optical fiber is used to monitor the flood discharge tunnel in real time; The laser is used to emit continuous light with a narrow line width and is the core light source of the detection system; The modulator is used to modulate continuous light into short pulse light with a high repetition frequency, and to control the width and repetition frequency of the light pulse; The detector is used to convert the post-Rayleigh scattered light signal into an electrical signal; The signal processor includes a pre-screening module, a sound source localization module and a feature extraction module; The pre-screening module is used to perform noise reduction processing on the voiceprint signal and screen abnormal voiceprints; The sound source positioning module is used to determine the location information of the abnormal voiceprint according to the abnormal voiceprint; The feature extraction module is used to perform wavelet packet signal analysis on the abnormal voiceprint according to the position information to obtain the feature value of the abnormal voiceprint; The cloud health monitoring system is used to perform health assessment of the flood discharge tunnel based on positioning data and eigenvalue data.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention abandons traditional point sensors for flood discharge tunnel safety monitoring and adopts a scour-resistant distributed fiber optic voiceprint sensing network. This achieves full tunnel coverage and no blind spots while ensuring the durability and stability of the detection system.

[0016] The present invention designs a signal processor based on distributed optical fiber, which includes a pre-screening module, a sound source localization module, and a feature extraction module. It performs optimization algorithm processing based on the voiceprint signals collected by the sensor network, thereby improving the efficiency and accuracy of monitoring and judging abnormal problems in flood discharge tunnels.

[0017] This paper combines real-time data compression and transmission with the ResNet-1D model to build a cloud-based health monitoring system. By performing interactive verification based on data transmitted by a signal processor, it enables real-time, intelligent, and accurate assessment of flood discharge tunnel health, based on long-term voiceprint data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0020] Figure 1This is a flow chart of a flood discharge tunnel safety monitoring method according to an embodiment of the present invention; Figure 2 This is a structural diagram of a tunnel flood discharge safety monitoring system according to an embodiment of the present invention; Figure 3 FIG. 1 is a structural diagram of a signal processor according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0022] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0023] In addition, the descriptions of "first", "second", etc. in the present invention are for descriptive purposes only and should not be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but this must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0024] Example 1 First, as Figure 1 As shown, the present invention provides a tunnel flood discharge safety monitoring method based on voiceprint recognition, comprising the following steps: S1: Deploy a fiber optic voiceprint sensing network to obtain initial voiceprint data.

[0025] The fiber optic voiceprint sensing network includes distributed sensing fiber (DAS, Distributed Acoustic Sensing) and distributed fiber optic system equipment, wherein the distributed fiber optic system equipment includes a laser, a modulator, a detector, a signal processor and a cloud health monitoring system.

[0026] The laser is used to emit continuous light with a narrow line width and is the core light source of the detection system.

[0027] The modulator is used to modulate continuous light into short pulses of light with a high repetition rate (the pulse width is usually in the nanosecond order), and controls the width and repetition rate of the light pulses, which directly affects the spatial resolution.

[0028] The detector converts the post-Rayleigh scattered light signal into an electrical signal, and has a high bandwidth to capture fast phase changes.

[0029] like Figure 3 As shown, the signal processor includes three data processing submodules, which are a pre-screening module, a sound source localization module, and a feature extraction module in order of data processing.

[0030] The pre-screening module is used to perform noise reduction processing on the voiceprint signal and screen out abnormal voiceprints.

[0031] The sound source positioning module uses the optimized algorithm time difference method based on the screened abnormal voiceprints to accurately determine the location information of the abnormal voiceprints.

[0032] The feature extraction module performs wavelet packet signal analysis on the abnormal voiceprint whose position information has been determined, and extracts the feature value of the abnormal voiceprint so as to identify the fault of the abnormal voiceprint.

[0033] The signal processor is also used to compress the processed signal and transmit it to the cloud health monitoring system through the optical fiber edge node.

[0034] Every point along the optical fiber line can be used as a sensor. There are no blind spots within the optical fiber coverage area, enabling seamless monitoring of the entire line. A single optical fiber can cover tunnels ≥10km, even for long flood discharge tunnels located in certain mountainous areas. Compared with traditional high-speed rail and subway tunnels, the layout of distributed optical fiber in flood discharge tunnels needs to consider the impact of high-speed water flow, high-pressure environment, high humidity environment, and siltation. Therefore, the deployment of optical fiber voiceprint sensing networks in flood discharge tunnels requires anti-scour layout, which specifically includes: The entire sensing optical fiber should be laid along the inner wall of the flood discharge tunnel lining or pre-buried in the lining structure; thereby improving the stability and durability of the optical fiber in high-humidity and high-pressure environments.

[0035] The optical fiber part located in the high-speed water flow area (such as the gradient section and curved section of the flood discharge tunnel) should be pre-buried in the groove: the optical fiber should be embedded in the groove and filled with protective material, or a metal sleeve should be used to protect the optical fiber; thereby improving the optical fiber's ability to resist scour.

[0036] The sensing optical fiber should be kept away from areas with strong vibration (such as trash racks and tunnel energy dissipation facilities) or vibration reduction measures should be taken to reduce noise interference and prevent mechanical fatigue.

[0037] In areas prone to sediment accumulation, such as low-velocity areas or sudden changes in flow, optical fiber sensors should be placed high up on the tunnel walls or roof. This prevents strain measurement distortion caused by sediment deposition and reduces the likelihood of aquatic organisms (such as shells) attaching to the sensor.

[0038] The specific process for constructing the initial voiceprint database is as follows: After the distributed sensing fiber is deployed using a scour-resistant layout, a narrow-linewidth laser is emitted by a laser. The laser is modulated by a modulator and transmitted to the sensing fiber. During the propagation of the incident light, Rayleigh scattering occurs due to microscopic inhomogeneities within the fiber. Some of the scattered light propagates back along the fiber, which is also called post-Rayleigh scattered light. During the flood discharge process, water impact, cavitation, or structural vibrations generate acoustic signals. These acoustic signals act on the optical fiber, affecting the post-Rayleigh scattered light received by the detector, causing changes in the phase and vibration of the received post-Rayleigh scattered light. The detector records the phase and vibration changes of the post-Rayleigh scattered light to obtain the specific conditions of cracks, cavitation, and structural vibration in the flood discharge tunnel. The detector constructs the initial voiceprint database based on the information data on the amplitude, frequency, position, and time of the Rayleigh scattered light.

[0039] S2: Process the initial voiceprint data to obtain abnormal voiceprint positioning data and characteristic value data.

[0040] The signal processor comprises three modules, which respectively include a pre-screening module, a sound source localization module and a feature extraction module according to the data processing order of the initial voiceprint data.

[0041] The pre-screening module is used to perform noise reduction processing on the voiceprint signal and screen out abnormal voiceprints.

[0042] The pre-screening module adopts adaptive noise reduction, i.e., eliminates water flow noise by spectral subtraction, thereby screening abnormal voiceprints. The elimination of water flow noise by spectral subtraction specifically includes: Calculate the spectrum of the noisy signal and obtain the spectrum of the noisy signal |Y(f)|, Subtracting the noise spectrum |N(f)| from the noisy signal spectrum |Y(f)|, the pure signal spectrum |X(f)| is obtained as:

[0043] |X(f)|=|Y(f)|- |N(f)|, In the formula is the over-reduction factor (usually 1≤α≤2).

[0044] Zero negative values: ∣X(f)∣=max(∣X(f)∣,0), Use the phase information of the noisy signal and the spectrum of the clean signal to reconstruct the frequency domain signal X(f): , In the formula is the phase factor.

[0045] Perform an inverse Fourier transform (ifft) on X(f) to obtain the time domain signal x(t): , Finally, overlap and add each frame signal to obtain the noise-reduced voiceprint data. .

[0046] After noise reduction processing, the steady-state background noise in the original signal is removed, leaving the low-frequency residual energy as background noise, providing a cleaner signal base for subsequent analysis. It can significantly quantify voiceprint anomalies.

[0047] The amplitude of a normal voiceprint decays exponentially with increasing frequency and is generally stable (without obvious spikes). The high-frequency energy ratio (i.e., the ratio of the signal energy in the target frequency band to the signal energy in the background noise frequency band) is usually less than 2. Abnormal voiceprint signals are usually accompanied by high-frequency signals, with a sharper spectrum distribution and a high-frequency energy ratio greater than 3. Abnormal voiceprints are judged based on this.

[0048] Specifically expressed as: Calculate the energy of the target frequency band : = , In the formula is the maximum frequency, is the minimum frequency, Calculate the background frequency band energy : = , In the formula is the maximum frequency of background noise, is the minimum frequency of background noise, Get the energy ratio R: R= , In the formula = , If R 3 is judged as an abnormal voiceprint.

[0049] The abnormal voiceprint data that has been judged is transmitted to the sound source positioning module for precise positioning.

[0050] The sound source positioning module accurately determines the location information of the abnormal voiceprint based on the time difference method according to the screened abnormal voiceprint, and the positioning error is .

[0051] The location information confirmed by the sound source localization module includes: The coordinates of the sound source position are ( ), the sensor position coordinates are ( , , the speed of sound is , Get the distance from the sound source to the sensor for: = , The sound waves reach the sensor and sensors The time difference is: Δ = , Solve the above two equations to get the coordinates of the sound source position ( ).

[0052] The abnormal voiceprint data at the determined location is transmitted to the feature extraction module for feature value extraction.

[0053] The feature extraction module decomposes the voiceprint signal based on wavelet packet changes to obtain multiple sub-band signals, thereby calculating the wavelet packet energy entropy (WPEE) and extracting the energy entropy characteristics of each abnormal voiceprint. This allows the cloud-based monitoring system to identify abnormal phenomena such as cavitation, cracks, and deformation based on the energy entropy data, and perform efficient data processing.

[0054] The wavelet packet analysis of the feature extraction module is mathematically described as decomposing the signal into sub-bands, each sub-band contains a specific frequency component.

[0055] Calculate the energy value of each sub-band : = , Where N is the coefficient length (depending on the relationship between the signal length and the number of decomposition layers), Normalized energy distribution probability : = , Where M is the total number of nodes (depending on the number of decomposition layers), Get the value of wavelet packet energy entropy WPEE: WPEE=- , In the formula, the signal processor compresses the processed sound source position and the extracted energy entropy eigenvalue data through the edge node and uploads them to the cloud health monitoring system.

[0056] During the flood discharge process, full-coverage real-time monitoring by sensor optical fibers will generate a large amount of data. Uploading the compressed data to the cloud can reduce bandwidth requirements, further reduce delays, and improve timeliness.

[0057] S3: Conduct health assessment of the flood discharge tunnel based on the positioning data and eigenvalue data.

[0058] The cloud health detection system uses the ResNet-1D model to perform health assessment on the uploaded positioning data and eigenvalue data.

[0059] The ResNet-1D model is constructed as follows: The positioning data and eigenvalue data are integrated to establish a model dataset: For the positioning data of each optical fiber, set the spacing threshold range to filter out the optical fibers that meet the conditions; Using the positioning data of the screened optical fibers and a selected spatial positioning algorithm, the three-dimensional coordinates of the region are calculated; Performing mean processing on the characteristic value data that matches the positioning data of the screened optical fiber to obtain energy data; The three-dimensional coordinates are matched with the energy data to obtain a data sequence, and a type label is added to each data sequence according to the energy data to obtain a model data set.

[0060] The basis for adding type labels based on energy data is: The energy entropy is significantly reduced (sudden energy concentration) and is classified as cavitation; The energy entropy first decreases (energy concentration) and then increases (multi-band resonance superposition) as a crack category; A slight increase in energy entropy (low-frequency energy diffusion) is a deformation category; Build a health assessment model: Use the ResNet-1D model to input the model dataset for model training, and use the classification results as output.

[0061] ResNet-1D is a one-dimensional extension of the classic residual network (ResNet), designed specifically for processing time series data. It has strong applicability and high accuracy in analyzing the timing characteristics of acoustic signals.

[0062] Continuously collect voiceprint data to obtain the positioning data and eigenvalue data of all optical fibers, and then fuse them into the ResNet-1D model to obtain the flood discharge tunnel health assessment category.

[0063] The ResNet-1D model corresponding to the present invention includes an input layer, a backbone network, a global pooling layer, a fully connected layer, and an output layer.

[0064] The input layer receives multi-channel data including energy entropy, position information, etc. The number of channels is set to 18. The global pooling layer uses global average pooling.

[0065] The fully connected layer is set to four categories: cavitation, crack, deformation, and other. No activation function is used, and the output is directly logits.

[0066] The output layer uses the Sofrmas function and cross entropy loss function to output the predicted probability.

[0067] Example 2 like Figure 2 As shown, the present invention provides a tunnel flood discharge safety monitoring system based on voiceprint recognition, which is applied to the above-mentioned tunnel flood discharge safety monitoring method based on voiceprint recognition, including: distributed sensing fiber (DAS) and its distributed fiber system equipment: laser, modulator, detector, signal processor, cloud health monitoring system.

[0068] The distributed sensing optical fiber is arranged in the flood discharge tunnel and is used for all-round real-time monitoring of the flood discharge tunnel.

[0069] The laser is used to emit continuous light with a narrow line width and is the core light source of the detection system.

[0070] The modulator is used to modulate continuous light into short pulses of light with a high repetition rate (the pulse width is usually in the nanosecond order), and controls the width and repetition rate of the light pulses, which directly affects the spatial resolution.

[0071] The detector converts the post-Rayleigh scattered light signal into an electrical signal, and has a high bandwidth to capture fast phase changes.

[0072] The signal processor comprises three data processing submodules, which are a pre-screening module, a sound source localization module and a feature extraction module in order of data processing.

[0073] The pre-screening module is used to perform noise reduction processing on the voiceprint signal and screen out abnormal voiceprints.

[0074] The sound source positioning module uses the optimized algorithm time difference method to accurately determine the location information of the abnormal voiceprint based on the screened abnormal voiceprint.

[0075] The feature extraction module performs wavelet packet signal analysis on the abnormal voiceprint whose position information has been determined, and extracts the feature value of the abnormal voiceprint so as to identify the fault of the abnormal voiceprint.

[0076] The signal processor is also used to compress the processed signal and transmit it to the cloud health monitoring system through the optical fiber edge node.

[0077] The cloud-based health monitoring system is built based on the ResNet-1D model and is used to interactively verify uploaded real-time data, generate real-time flood discharge tunnel health assessment results, and predict the structural degradation trend of the flood discharge tunnel based on long-term voiceprint data.

Claims

1. A tunnel flood discharge safety monitoring method based on voiceprint recognition, characterized by: The steps include: Deploy a fiber-optic voiceprint sensor network to obtain initial voiceprint data; Processing the initial voiceprint data to obtain abnormal voiceprint positioning data and characteristic value data; Conduct health assessment of flood discharge tunnel based on location data and eigenvalue data; The processing of the initial voiceprint data comprises the steps of: Perform noise reduction on voiceprint signals and filter out abnormal voiceprints; According to the abnormal voiceprint, the time difference method is used to determine the location information of the abnormal voiceprint; A wavelet packet signal analysis is performed on the abnormal voiceprint according to the position information to obtain a characteristic value of the abnormal voiceprint.

2. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized in that: The deployment of the optical fiber voiceprint sensing network comprises the following steps: The sensing optical fiber is laid along the inner wall of the flood discharge tunnel lining or pre-buried in the lining structure; The sensing optical fiber located in the high-speed water flow area is pre-buried in the groove and reinforced by filling with protective materials or metal sleeves; The sensing optical fiber located in the low flow velocity area or the water flow mutation area is laid on the side wall or top of the tunnel.

3. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized by: The optical fiber voiceprint sensing network includes distributed sensing optical fiber and distributed optical fiber system equipment; the distributed optical fiber system equipment includes a laser, a modulator, a detector, a signal processor and a cloud health monitoring system.

4. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized by: The noise reduction process of the voiceprint signal is specifically to eliminate water flow noise by spectral subtraction; the elimination of water flow noise by spectral subtraction comprises the following steps: Obtaining a noisy signal spectrum according to the voiceprint signal; Subtract the noise spectrum from the noisy signal spectrum to obtain the pure signal spectrum; Zero the negative values ​​of the pure signal spectrum: Reconstruct the frequency domain signal according to the phase information of the clean signal spectrum and the noisy signal spectrum; Performing an inverse Fourier transform on the frequency domain signal to obtain a time domain signal; Overlap and add each frame of time domain signal to obtain the noise-reduced voiceprint data.

5. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 4 is characterized in that: The pure signal spectrum is expressed as: ∣X(f)∣=∣Y(f)∣- ∣N(f)∣, Among them, |X(f)| is the pure signal spectrum, |N(f)| is the noise spectrum, |Y(f)| is the noisy signal spectrum, is the over-reduction factor.

6. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized by: The method of screening abnormal voiceprints comprises the following steps: Calculate the target frequency band energy and background frequency band energy of the voiceprint data after noise reduction; Obtaining an energy ratio according to the target frequency band energy and the background frequency band energy; Filter abnormal voiceprints based on energy ratio.

7. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized by: The method of using the time difference method to determine the location information of the abnormal voiceprint includes the following steps: Selecting a first sensor and a second sensor that detect an abnormal voiceprint; Acquire a data time difference according to position information of the first sensor and the second sensor; A first distance relationship is constructed based on the position information of the abnormal voiceprint and the position information of the first sensor, and a second distance relationship is constructed based on the position information of the abnormal voiceprint and the position information of the second sensor; The location information of the abnormal voiceprint is determined according to the data time difference, the first distance relationship formula and the second distance relationship formula.

8. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized by: The characteristic value of the abnormal voiceprint is the wavelet packet energy entropy; and the wavelet packet signal analysis of the abnormal voiceprint according to the position information comprises the steps of: Decompose the abnormal voiceprint into several sub-bands; Calculate the energy value of each sub-band, and obtain the normalized energy distribution probability corresponding to each sub-band according to the energy value; The wavelet packet energy entropy is calculated based on the normalized energy distribution probability corresponding to each sub-band.

9. The tunnel flood discharge safety monitoring method based on voiceprint recognition according to claim 1 is characterized by: The flood discharge tunnel health assessment based on the positioning data and the eigenvalue data is implemented by the ResNet-1D model; The ResNet-1D model is constructed as follows: The positioning data and feature value data are integrated to establish the model data set: For the positioning data of each optical fiber, set the spacing threshold range to filter out the optical fibers that meet the conditions; Using the positioning data of the screened optical fibers and a selected spatial positioning algorithm, the three-dimensional coordinates of the region are calculated; Performing mean processing on the characteristic value data matching the positioning data of the screened optical fiber to obtain energy data; The three-dimensional coordinates are matched with the energy data to obtain a data sequence, and a type label is added to each data sequence according to the energy data to obtain a model data set; The ResNet-1D model is trained according to the input model dataset and the classification result is used as output.

10. A tunnel flood discharge safety monitoring system based on voiceprint recognition, characterized in that: Application A tunnel flood discharge safety monitoring method based on voiceprint recognition as described in any one of claims 1 to 9, comprising: a distributed sensing optical fiber and a distributed optical fiber system device; the distributed optical fiber system device comprises a laser, a modulator, a detector, a signal processor and a cloud health monitoring system; The distributed sensing optical fiber is used to monitor the flood discharge tunnel in real time; The laser is used to emit continuous light with narrow line width and is the core light source of the detection system; The modulator is used to modulate continuous light into short pulse light with a high repetition frequency, and control the width and repetition frequency of the light pulse; The detector is used to convert the post-Rayleigh scattered light signal into an electrical signal; The signal processor includes a pre-screening module, a sound source localization module and a feature extraction module; The pre-screening module is used to perform noise reduction processing on the voiceprint signal and screen abnormal voiceprints; The sound source positioning module is used to determine the location information of the abnormal voiceprint according to the abnormal voiceprint; The feature extraction module is used to perform wavelet packet signal analysis on the abnormal voiceprint according to the position information to obtain the feature value of the abnormal voiceprint; The cloud health monitoring system is used to perform health assessment of the flood discharge tunnel based on the positioning data and the characteristic value data.

Images