Surface collapse risk toughness distributed acoustic sensing system and method

Abstract

The invention discloses a surface collapse risk toughness distributed acoustic sensing system and method, and the method is characterized in that the method comprises the steps: carrying out the normalization processing of a collected acoustic vibration signal, dividing the signal into a plurality of signal segments according to a preset time window, and forming a signal segment sequence; inputting a pre-training deep learning model, performing event identification, and determining an event type corresponding to each signal segment; calculating a time domain feature and a frequency domain feature of a signal segment corresponding to each event type; calculating a surface collapse risk value according to the time domain feature and the frequency domain feature corresponding to each event type; and comparing and judging a surface collapse risk level according to the surface collapse risk value and a set risk threshold value, and generating corresponding early warning information according to the surface collapse risk level. The system adopts a polymer measuring pipe to realize acoustic sensing, solves the problem that a directly-buried optical fiber is easy to break in the prior art, supports the sensing method, and realizes early warning of surface collapse.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring, and more specifically to a ground collapse risk resilience distributed acoustic sensing system and method. Background Art

[0002] Ground collapse (also known as ground subsidence or ground settlement) is a geological phenomenon caused by natural or human factors. It may be caused by underground cavities, mine collapse, groundwater level drop, loose soil, etc. Ground collapse not only causes damage to buildings and roads, but may also cause secondary disasters such as pipeline rupture, power outages, and even casualties. Therefore, effective prediction and monitoring of ground collapse is crucial to ensure public safety.

[0003] At present, the technologies used for ground collapse monitoring mainly include seismographs, global positioning systems (GPS), geological radars (GPR), etc. Although these methods can provide a certain degree of early warning information, there are still some shortcomings in practical applications: 1. High cost: the purchase and maintenance costs of high-end equipment are high, and it is not suitable for large-scale promotion and use; 2. Limited coverage: a single device can usually only monitor changes within a certain range around it, and it is difficult to achieve continuous monitoring of a large area; 3. Long response time: some equipment requires manual intervention to perform data analysis, which may lead to delayed early warning; 4. Poor adaptability: the effectiveness of some technologies is reduced in specific environments (such as complex geological conditions).

[0004] Therefore, how to improve the reliability of ground collapse risk monitoring and reduce the monitoring cost while increasing the monitoring range is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0005] In view of this, the present invention provides a ground subsidence risk resilience distributed acoustic perception system and method. The system combines resilience theory and uses advanced acoustic sensing technology to solve the problem of easy breakage of horizontal directly buried optical fibers in collapse target areas such as goafs, improve monitoring reliability, and achieve early warning of ground subsidence, so as to reduce the losses caused by geological disasters to the social economy and people's lives, and provide a beneficial solution for urban geological disaster risk monitoring, early warning and emergency management.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A distributed acoustic sensing method for ground collapse risk resilience includes the following steps:

[0008] Step 1: using a distributed acoustic sensor demodulator to collect acoustic vibration signals, normalizing the acoustic vibration signals and dividing them into a plurality of signal segments according to a preset time window to form a signal segment sequence;

[0009] Step 2: Input the signal segment sequence into a pre-trained deep learning model to perform event recognition and determine the event type corresponding to each signal segment;

[0010] Step 3: Calculate the time domain features and frequency domain features of the signal segment corresponding to each event type;

[0011] Step 4: Calculate the ground collapse risk value R according to the time domain characteristics and frequency domain characteristics corresponding to each event type;

[0012] Step 5: According to the ground collapse risk value and the set risk threshold, compare and judge the ground collapse risk level, and generate corresponding warning information according to the ground collapse risk level.

[0013] Preferably, the event types include groundwater pneumatic action events and activation events of goaf and soil cave development; the time domain characteristics calculated in step 3 include amplitude and duration, and the calculation expression of the amplitude is: Where A represents the amplitude, T represents the length of the signal segment, m represents the sequence number of the signal segment in the signal segment sequence, and s n represents the nth data point in the signal segment; sets an amplitude threshold, starts when the amplitude of the data point in the signal segment is higher than the threshold, and ends when the amplitude is lower than the threshold, and takes the duration between the start and the end as the duration D;

[0014] The frequency domain features calculated in step 3 include spectrum and dominant frequency. The signal segment is subjected to fast Fourier transform to obtain spectrum S(f)=FFT(s), where s represents the signal segment and f represents the frequency. The dominant frequency analysis is performed on the frequency point with the largest energy in the spectrum to obtain the dominant frequency.

[0015]

[0016] Preferably, in step 4, the ground subsidence risk value is calculated according to the time domain characteristics and frequency domain characteristics of the groundwater pneumatic action event and the goaf and soil cave development activation event, and the expression is:

[0017]

[0018] Among them, A gw,i , D gw,i , S gw,i and f dom,gw,i are the amplitude, duration, spectrum and dominant frequency of the signal segment belonging to the groundwater aerodynamic event; A cav,j , D cav,j , S cav,j and f dom,cav,jare the amplitude, duration, spectrum and dominant frequency of the jth signal segment belonging to the activation event of the goaf and soil cave development; w A,gw 、w D,gw 、w S,gw and w fdom,gw are the weights of the characteristics of groundwater aerodynamic events; w A,cav 、w D,cav 、w S,cav and w fdom,cav are the weights of the activation event characteristics of goaf and soil cave development; N gw and N cav They are the number of signal fragments of groundwater pneumatic action events, goaf and soil cave development activation events.

[0019] Preferably, a one-dimensional convolutional neural network model is used to construct a deep learning model, and the model is trained and optimized using the Adam algorithm with a cross entropy loss function to obtain the pre-trained deep learning model.

[0020] Preferably, the deep learning model includes an input layer, a convolution layer, a ReLU function, a pooling layer and a fully connected layer connected in sequence; wherein the input of the input layer is a fiber acoustic vibration multi-channel signal, expressed in the form of {batch size, M, C}; the convolution layer has C convolution kernels, and the input feature map of the convolution layer is x∈R N×C , N represents the number of time steps of the input signal, and the convolution kernel is w∈R K , K is the length of the convolution kernel; the ReLU function introduces nonlinear features, enhances the model's expressiveness, and learns more complex signal patterns and feature relationships; the pooling window of the pooling layer is k, and the step size is s, then the output feature map is y∈R M×C , The i-th output feature of the c-th channel i=1,2,3,…,M,c=1,2,3,…,C;the fully connected layer integrates the extracted output features and outputs the event category prediction value through the weighted sum of neurons using the activation function.

[0021] Preferably, the cross entropy loss function Where y is the true label 0 or 1, Predict the probability value for the model.

[0022] Preferably, the process of iteratively training the deep learning model in batch mode is:

[0023] Step 21: Collect acoustic vibration signal samples, normalize them, and segment them according to the set time window, and mark the corresponding event category for each segment; the event categories included in the acoustic vibration signal samples include normal state, groundwater pneumatic action events, and goaf and soil cave development activation events;

[0024] Step 22: Input the labeled acoustic vibration signal samples into the input layer of the deep learning model;

[0025] Step 23: Obtain output features through the convolution layer and the pooling layer;

[0026] Step 24: transmitting the output feature to the fully connected layer to obtain the event category prediction value;

[0027] Step 25: Calculate the loss value according to the event category prediction value and the cross entropy loss function, calculate the gradient of the loss value to the model parameters through the back propagation algorithm, and use the Adam algorithm to update the model parameters using the gradient. When the accuracy, precision and recall rate of the deep learning model meet the set thresholds, stop updating the model parameters to obtain the pre-trained deep learning model.

[0028] Preferably, the expressions of accuracy, precision and recall are:

[0029]

[0030] Among them, TP represents a true positive example; TN represents a true negative example; FP represents a false positive example; FN represents a false negative example; Set the F1 threshold to achieve a balance between Precision and Recall.

[0031] Preferably, the deep learning model can select a decision tree model and use a recursive elimination algorithm (RFE) to optimize the model. The contribution of the feature in the model is measured by the importance of the splitting of the feature in the tree structure (such as the reduction in the Gini index, etc.). The feature with the lowest contribution is deleted each time, and then the model is retrained using the remaining features. This process is repeated until the preset number of features is reached or the model performance is no longer improved.

[0032] A ground collapse risk resilience distributed acoustic sensing system, comprising a polymer measuring tube, an optical fiber, a distributed acoustic sensor demodulator and a ground monitoring workstation;

[0033] The optical fiber passes through the polymer measuring tube and is connected to the distributed acoustic sensor demodulator; the distributed acoustic sensor demodulator is connected to the ground monitoring workstation; the polymer measuring tube is horizontally laid in the stratum above the collapse hazard target area; the ground monitoring workstation is loaded with a pre-trained deep learning model to process the acoustic vibration signal transmitted by the distributed acoustic sensor demodulator.

[0034] Preferably, guide grooves are axially provided on the neutral layer on both sides of the outer wall of the polymer measuring tube, and a group of plugs are respectively provided at both ends; the plugs include a left plug and a right plug with a semicircular protrusion structure, wherein the left plug is provided with a left positioning hole on the semi-cylindrical side surface and a male buckle is provided on the semicircular top surface, and the right plug is provided with a right positioning hole on the semi-cylindrical side surface and a female buckle is provided on the semicircular top surface, and the left plug and the right plug are fixed to the two ends of the polymer measuring tube by the matching male buckles and the female buckles; the two groups of optical fibers are respectively placed in the guide grooves on both sides, and respectively pass through the corresponding left positioning hole and the right positioning hole. The semi-cylindrical sides of the left and right plugs are sleeved on the outer wall of the polymer measuring tube, and the semi-circular top surfaces seal the outlets at both ends of the polymer measuring tube. The plugs after snapping are similar to bottle cap structures; the plugs can be installed at any position of the optical fiber without having to insert it into one end of the optical fiber first; the plug positioning holes and guide grooves can ensure the straight layout of the optical fiber and reduce the scattering effect caused by bending; the plugs can also prevent foreign matter from entering the measuring tube and generating interference signals.

[0035] Preferably, epoxy resin is poured into the guide groove to fix the optical fiber. Isocyanate curing agent can also be added into the epoxy resin to speed up the curing time and maintain good flexibility.

[0036] Preferably, the end of the optical fiber extends to the ground through a protective sleeve and is connected to the distributed acoustic sensor demodulator, which is used to receive the acoustic vibration signal collected by the optical fiber encapsulated in the guide groove of the polymer measuring tube, and the acoustic vibration signal includes the acoustic vibration information generated by the normal state, groundwater and gas dynamic events and / or goaf and soil cave development activation events.

[0037] Preferably, the ground monitoring workstation receives the acoustic vibration signal demodulated by the distributed acoustic sensor demodulator, and uses the pre-trained deep learning model to perform event recognition, obtains the ground collapse risk level after feature analysis, and generates early warning information.

[0038] Preferably, the process of using the pre-trained deep learning model to perform event recognition on the acoustic vibration signal, and performing feature analysis to obtain the ground collapse risk level, and generating early warning information is:

[0039] Step 1: normalize the acoustic vibration signal and segment it according to a set time window to obtain a signal segment sequence;

[0040] Step 2: Input the signal segment sequence into the pre-trained deep learning model, obtain the event category prediction value, and determine the event category corresponding to each signal segment; for example, if the model predicts that a certain signal segment has the highest probability of belonging to a groundwater aerodynamic event, the signal segment can be marked as related to the groundwater aerodynamic event;

[0041] Step 3: respectively record the number of signal segments corresponding to the groundwater pneumatic action event, goaf and soil cave development activation event, and perform feature analysis to calculate the time domain features and frequency domain features corresponding to the two types of events, wherein the time domain features include amplitude, duration, etc., and the frequency domain features include spectrum, dominant frequency, etc.;

[0042] The time domain features are calculated as follows:

[0043]

[0044] Among them, A rms represents the amplitude, T is the length of the signal segment, m represents the sequence number of the signal segment in the signal segment sequence, s n represents the nth data point in the signal segment; sets an amplitude threshold, starts when the amplitude of the data point in the signal segment is higher than the threshold, and ends when the amplitude is lower than the threshold, and takes the duration between the start and the end as the duration D m ;

[0045] When calculating the frequency domain features, the signal fragment is transformed using the fast Fourier transform (FFT) to obtain the spectrum S m (f) = FFT(s m ), S m represents the spectrum obtained after FFT of the mth signal segment, s m represents the mth signal segment, f is the frequency, and the dominant frequency analysis is performed at the frequency point with the largest energy in the spectrum to obtain the dominant frequency

[0046] Step 4: Calculate the ground collapse risk value R based on the time domain characteristics and frequency domain characteristics corresponding to the two types of events. The expression is:

[0047]

[0048] Among them, A gw,i , D gw,i , S gw,i and f dom,gw,i are the amplitude, duration, spectrum and dominant frequency of the signal segment belonging to the groundwater aerodynamic event; A cav,j , D cav,j , S cav,j and f dom,cav,j are the amplitude, duration, spectrum and dominant frequency of the jth signal segment belonging to the activation event of the goaf and soil cave development; w A,gw 、w D,gw 、w S,gw and w fdom,gw are the weights of the characteristics of groundwater aerodynamic events; w A,cav、w D,cav 、w S,cav and w fdom,cav are the weights of the activation event characteristics of goaf and soil cave development; N gw and N cav They are the number of signal segments of groundwater pneumatic action events, goaf and soil cave development activation events;

[0049] Step 5: According to the ground collapse risk value R and the set risk threshold, the ground collapse risk level is compared and determined, and corresponding warning information is generated according to the ground collapse risk level.

[0050] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a distributed acoustic perception system and method for ground collapse risk resilience, which makes full use of the advantages of acoustic perception and involves an acoustic monitoring system based on a distributed architecture. The system has high resilience and reliability, and can maintain overall functions in the case of failure of a single node by deploying multiple polymer measuring tubes; in addition, by analyzing massive data through intelligent algorithms, potential precursors of collapse can also be discovered, thereby providing more valuable information support for decision makers. The present invention uses acoustic perception technology to capture subtle changes underground and issue alarms in a timely manner. Compared with traditional monitoring methods, it has the following advantages: low cost, easy deployment, real-time monitoring and wide coverage. The present invention is particularly suitable for areas with complex geological environments such as urban infrastructure construction, mining areas, water conservancy projects, and occasions where long-term stable monitoring of ground stability is required. By distributing acoustic sensors, the system can collect sound wave signals generated by underground structural changes in real time, and identify possible signs of ground collapse through data analysis and processing, thereby providing timely and reliable early warning information for relevant departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0052] Figure 1 A schematic diagram of a process flow of a ground collapse risk resilience distributed acoustic sensing method provided by the present invention;

[0053] Figure 2 A schematic diagram of the structure of the polymer measuring tube provided by the present invention;

[0054] Figure 3 A schematic top view of the plug structure provided by the present invention;

[0055] Figure 4 A schematic front view of the plug structure provided by the present invention;

[0056] Figure 5 This is a left side schematic diagram of the plug structure provided by the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] The embodiment of the present invention discloses a ground collapse risk resilience distributed acoustic perception method, comprising the following steps:

[0059] Step 1: Use a distributed acoustic sensor demodulator to collect acoustic vibration signals, normalize the acoustic vibration signals, and divide them into several signal segments according to a preset time window to form a signal segment sequence;

[0060] Step 2: Input the signal segment sequence into the pre-trained deep learning model for event recognition to determine the event type corresponding to each signal segment;

[0061] Step 3: Calculate the time domain features and frequency domain features of the signal segment corresponding to each event type;

[0062] Step 4: Calculate the ground collapse risk value based on the time domain characteristics and frequency domain characteristics corresponding to each event type;

[0063] Step 5: Compare and judge the ground collapse risk level based on the ground collapse risk value and the set risk threshold, and generate corresponding warning information based on the ground collapse risk level.

[0064] Furthermore, the event types include groundwater pneumatic action events and activation events of goaf and soil cave development; the time domain characteristics calculated in step 3 include amplitude and duration, and the calculation expression of amplitude is Where A represents the amplitude, T represents the length of the signal segment, m represents the sequence number of the signal segment in the signal segment sequence, and s n Indicates the nth data point in the signal segment; sets the amplitude threshold, starting when the amplitude of the data point in the signal segment is higher than the threshold and ending when the amplitude is lower than the threshold, and takes the duration between the start and the end as the duration D;

[0065] The frequency domain features calculated in step 3 include spectrum and dominant frequency. The signal segment is subjected to fast Fourier transform to obtain spectrum S(f)=FFT(s), where s represents the signal segment and f represents the frequency. The dominant frequency analysis is performed at the frequency point with the largest energy in the spectrum to obtain the dominant frequency

[0066] Furthermore, in step 4, the ground subsidence risk value is calculated based on the time domain characteristics and frequency domain characteristics of the groundwater pneumatic action events and the activation events of the goaf and soil cave development. The expression is:

[0067]

[0068] Among them, A gw,i , D gw,i , S gw,i and f dom,gw,i are the amplitude, duration, spectrum and dominant frequency of the signal segment belonging to the groundwater aerodynamic event, respectively; A cav,j , D cav,j , S cav,j and f dom,cav,j are the amplitude, duration, spectrum and dominant frequency of the jth signal segment belonging to the activation event of the goaf and soil cave development; w A,gw 、w D,gw 、w S,gw and w fdom,gw are the weights of the characteristics of groundwater aerodynamic events; w A,cav 、w D,cav 、w S,cav and w fdom,cav are the weights of the activation event characteristics of goaf and soil cave development; N gw and N cav They are the number of signal fragments of groundwater pneumatic action events, goaf and soil cave development activation events.

[0069] Furthermore, a one-dimensional convolutional neural network model is used to construct a deep learning model, and the Adam algorithm is used to train and optimize the model using the cross entropy loss function to obtain a pre-trained deep learning model. The Adam algorithm is used to update the model parameters. The algorithm combines the advantages of momentum and adaptive learning rate, can automatically adjust the learning rate during the training process, speed up the convergence of the model, and to a certain extent avoid falling into the local optimal solution. It is easy to use and concise, and only relevant hyperparameters (such as learning rate, momentum parameters, etc.) need to be set.

[0070] Furthermore, the deep learning model includes an input layer, a convolution layer, a ReLU function, a pooling layer, and a fully connected layer connected in sequence; wherein the input of the input layer is a fiber acoustic vibration multi-channel signal, expressed in the form of {batch size, M, C}; the convolution layer has C convolution kernels, and the input feature map of the convolution layer is x∈R N×C , N represents the number of time steps of the input signal, and the convolution kernel is w∈R K , K is the length of the convolution kernel; the ReLU function introduces nonlinear features, enhances the model's expressiveness, and learns more complex signal patterns and feature relationships; the pooling window of the pooling layer is k, and the step size is s, then the output feature map is y∈R M×C , The i-th output feature of the c-th channel i=1,2,3,…,M,c=1,2,3,…,C;the fully connected layer integrates the extracted output features and outputs the event category prediction value through the weighted sum of neurons using the activation function.

[0071] Furthermore, the cross entropy loss function Where y is the true label 0 or 1, Predict the probability value for the model.

[0072] Furthermore, the process of iteratively training the deep learning model in batch mode is as follows:

[0073] Step 21: Collect acoustic vibration signal samples, normalize them, and segment them according to the set time window, and label the corresponding event category for each segment; the event categories included in the acoustic vibration signal samples include normal state, groundwater pneumatic action events, and goaf and soil cave development activation events;

[0074] Step 22: Input the labeled acoustic vibration signal samples into the input layer of the deep learning model;

[0075] Step 23: Obtain output features through convolutional layer and pooling layer;

[0076] Step 24: Transfer the output features to the fully connected layer to obtain the event category prediction value;

[0077] Step 25: Calculate the loss value based on the event category prediction value and the cross entropy loss function, calculate the gradient of the loss value to the model parameters through the back propagation algorithm, and use the Adam algorithm to update the model parameters using the gradient. When the accuracy, precision, and recall rate of the deep learning model meet the set thresholds, stop updating the model parameters and obtain the pre-trained deep learning model.

[0078] Furthermore, the expressions of accuracy, precision and recall are:

[0079]

[0080] Among them, TP represents a true positive example; TN represents a true negative example; FP represents a false positive example; FN represents a false negative example; Set the F1 threshold to achieve a balance between Precision and Recall

[0081] On the other hand, the deep learning model can also be used to analyze multi-dimensional features and perform sensitivity training on small signals to make the distinction between the two events more sensitive. The deep learning model can use a decision tree model and use a recursive elimination algorithm (RFE) to optimize the model. The contribution of the feature in the model is measured by the importance of the split in the tree structure (such as the reduction in the Gini index, etc.). The feature with the lowest contribution is deleted each time, and then the remaining features are used to retrain the model. This process is repeated until the preset number of features is reached or the model performance is no longer improved.

[0082] On the other hand, in a specific embodiment, a ground collapse risk resilience distributed acoustic sensing system comprises: a polymer measuring tube 1, an optical fiber 3, a distributed acoustic sensor demodulator and a ground monitoring workstation; the polymer measuring tube 1 has a structure such as Figure 2 As shown, guide grooves 2 are opened on both sides of the outer wall along the length direction of the neutral layer, and an optical fiber 3 is embedded in each of them. The optical fiber 3 is a common single-mode communication optical fiber. The polymer measuring tube 1 is a high-ductility polymer measuring tube, which is horizontally laid in a certain depth of the formation above the target area of ​​collapse hazards such as the goaf area; after the optical fiber is placed in the guide groove, plugs are installed on both sides of the polymer measuring tube, such as Figure 3-5 As shown, it includes a left plug 4 and a right plug 5. The left plug 4 and the right plug 5 are semicircular protrusion structures, wherein a left positioning hole 42 is arranged on the semi-cylindrical side surface of the left plug 4, and a male buckle 41 is arranged on the semicircular top surface, and a right positioning hole 52 is arranged on the semi-cylindrical side surface of the right plug 5, and a female buckle 51 is arranged on the semicircular top surface. The left plug 4 and the right plug 5 are fixed to the two ends of the polymer measuring tube 1 by the matching male buckles 41 and female buckles 51; two groups of optical fibers 3 are respectively placed in the guide grooves 2 on both sides, and respectively pass through the corresponding left positioning holes 42 and right positioning holes 52.

[0083] Furthermore, the plug can be installed at any position of the optical fiber 3 without having to insert it into one end of the optical fiber first. The left and right positioning holes lead out the optical fibers on both sides respectively. After fixing them, the left and right plugs are closed using male and female buckles for easy use.

[0084] Furthermore, epoxy resin is poured in the guide groove to achieve epoxy resin encapsulation, and isocyanate curing agent is added to the epoxy resin to accelerate the curing time and maintain good flexibility;

[0085] Furthermore, the end of the optical fiber is extended to the ground monitoring workstation through a protective sleeve and connected to the distributed acoustic sensor demodulator to collect the acoustic vibration signal collected by the packaged optical fiber. The signal contains the acoustic vibration information generated by the water-gas dynamic effect of the underground goaf and the activation event of the goaf and soil cave development;

[0086] Furthermore, the ground monitoring workstation is a base station edge computing workstation, which uses a pre-trained deep learning model to filter the acoustic vibration signals collected in real time, identify groundwater aerodynamic events or goaf and soil cave development activation events, and record the number of two types of events and their respective time domain characteristics such as amplitude and duration, and frequency domain characteristics such as spectrum characteristics and dominant frequency. Based on this, the risk of ground collapse in the target area of ​​hidden danger is evaluated for disaster warning and emergency management.

[0087] Furthermore, the characteristics of groundwater-gas dynamic events and goaf and soil cave development activation events are different. The time domain characteristics of groundwater-gas dynamic events are that the amplitude of the acoustic-vibration signal is relatively small and relatively stable, and the duration may be longer. The frequency domain characteristics are that the spectrum is relatively concentrated in a specific low-frequency range, and the dominant frequency is lower. The time domain characteristics of goaf and soil cave development activation events are that the amplitude of the acoustic-vibration signal is usually larger, the duration is shorter, and it is sudden. The frequency domain characteristics are that the spectrum is relatively dispersed, may contain multiple frequency components, and the dominant frequency is relatively high.

[0088] Furthermore, the pre-trained deep learning model is used to identify the acoustic vibration signal and perform feature analysis to obtain the ground collapse risk level. The process of generating early warning information is as follows:

[0089] S1: After normalizing the acoustic vibration signal, segment it according to the set time window to obtain a signal segment sequence;

[0090] S2: Input the signal segment sequence into the pre-trained deep learning model to obtain the event category prediction value and determine the event category corresponding to each signal segment; for example, if the model predicts that a certain signal segment has the highest probability of belonging to a groundwater aerodynamic event, the signal segment can be marked as related to the groundwater aerodynamic event;

[0091] S3: Record the number of signal segments corresponding to groundwater pneumatic events, goaf and soil cave development activation events, and perform feature analysis to calculate the time domain features and frequency domain features corresponding to the two types of events. The time domain features include amplitude, duration, etc., and the frequency domain features include spectrum, dominant frequency, etc.;

[0092] The time domain features are calculated as follows:

[0093]

[0094] Among them, Arms represents the amplitude, T is the length of the signal segment, m represents the sequence number of the signal segment in the signal segment sequence, s n Indicates that it is the nth data point in the signal segment; set an amplitude threshold, start when the amplitude of the data point in the signal segment is higher than the threshold, and end when the amplitude is lower than the threshold, and take the duration between the start and end as the duration D m ;

[0095] When calculating the frequency domain features, the signal fragment is transformed using the fast Fourier transform (FFT) to obtain the spectrum S m (f) = FFT(s m ), S m represents the spectrum obtained after FFT of the mth signal segment, s m represents the mth signal segment, f is the frequency, and the dominant frequency analysis is performed at the frequency point with the largest energy in the spectrum to obtain the dominant frequency

[0096] S4: Calculate the ground collapse risk value R based on the time domain characteristics and frequency domain characteristics corresponding to the two types of events. The expression is:

[0097]

[0098] Among them, A gw,i , D gw,i , S gw,i and f dom,gw,i are the amplitude, duration, spectrum and dominant frequency of the signal segment belonging to the groundwater aerodynamic event; A cav,j , D cav,j , S cav,j and f dom,cav,j are the amplitude, duration, spectrum and dominant frequency of the jth signal segment belonging to the activation event of the goaf and soil cave development; w A,gw 、w D,gw 、w S,gw and w fdom,gw are the weights of the characteristics of groundwater aerodynamic events; w A,cav 、w D,cav 、w S,cav and w fdom,cav are the weights of the activation event characteristics of goaf and soil cave development; N gw and N cav They are the number of signal segments of groundwater pneumatic action events, goaf and soil cave development activation events;

[0099] S5: According to the ground collapse risk value R and the set risk threshold, the ground collapse risk level is compared and judged, and corresponding warning information is generated according to the ground collapse risk level.

[0100] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0101] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A distributed acoustic sensing method for ground collapse risk resilience, characterized in that: The following steps are involved: Step 1: using a distributed acoustic sensor demodulator to collect acoustic vibration signals, normalizing the acoustic vibration signals and dividing them into a plurality of signal segments according to a preset time window to form a signal segment sequence; Step 2: Input the signal segment sequence into a pre-trained deep learning model to perform event recognition and determine the event type corresponding to each signal segment; Step 3: Calculate the time domain features and frequency domain features of the signal segment corresponding to each event type; Step 4: Calculate the ground collapse risk value based on the time domain characteristics and frequency domain characteristics corresponding to each event type; Step 5: According to the ground collapse risk value and the set risk threshold, compare and judge the ground collapse risk level, and generate corresponding warning information according to the ground collapse risk level.

2. A distributed acoustic sensing method for ground collapse risk resilience according to claim 1, characterized in that: The event types include groundwater pneumatic action events and activation events of goaf and soil cave development; the time domain characteristics calculated in step 3 include amplitude and duration, and the calculation expression of the amplitude is: Where A represents the amplitude, T represents the length of the signal segment, m represents the sequence number of the signal segment in the signal segment sequence, and s n represents the nth data point in the signal segment; Set an amplitude threshold, starting when the amplitude of a data point in the signal segment is higher than the threshold and ending when the amplitude is lower than the threshold, and taking the duration between the start and the end as the duration D; The frequency domain features calculated in step 3 include spectrum and dominant frequency. The signal segment is subjected to fast Fourier transform to obtain spectrum S(f)=FFT(s), where s represents the signal segment and f represents the frequency. The dominant frequency analysis is performed on the frequency point with the largest energy in the spectrum to obtain the dominant frequency.

3. A distributed acoustic sensing method for ground collapse risk resilience according to claim 2, characterized in that: In step 4, the ground subsidence risk value is calculated according to the time domain characteristics and frequency domain characteristics of the groundwater pneumatic action event and the goaf and soil cave development activation event, and the expression is: Among them, A gw,i , D gw,i , S gw,i and f dom,gw,i are the amplitude, duration, spectrum and dominant frequency of the signal segment belonging to the groundwater aerodynamic event; A cav,j , D cav,j , S cav,j and f dom,cav,j are the amplitude, duration, spectrum and dominant frequency of the jth signal segment belonging to the activation event of the goaf and soil cave development; w A,gw 、w D,gw 、w S,gw and w fdom,gw are the weights of the characteristics of groundwater aerodynamic events; w A,cav 、w D,cav 、w S,cav and wf dom,cav are the weights of the activation event characteristics of goaf and soil cave development; N gw and N cav They are the number of signal fragments of groundwater pneumatic action events, goaf and soil cave development activation events.

4. A distributed acoustic sensing method for ground collapse risk resilience according to claim 1, characterized in that: A one-dimensional convolutional neural network model is used to build a deep learning model, and the cross entropy loss function and the Adam algorithm are used to train and optimize the model to obtain the pre-trained deep learning model.

5. A distributed acoustic sensing method for ground collapse risk resilience according to claim 1, characterized in that: The deep learning model consists of an input layer, a convolutional layer, a ReLU function, a pooling layer, and a fully connected layer that are connected in sequence.

6. A distributed acoustic sensing method for ground collapse risk resilience according to claim 5, characterized in that: The process of iteratively training a deep learning model in batch mode is as follows: Step 21: Collect acoustic vibration signal samples, normalize them, and segment them according to the set time window, and mark the corresponding event category for each segment; the event categories included in the acoustic vibration signal samples include normal state, groundwater pneumatic action events, and goaf and soil cave development activation events; Step 22: Input the labeled acoustic vibration signal samples into the input layer of the deep learning model; Step 23: Obtain output features through the convolution layer and the pooling layer; Step 24: transmitting the output feature to the fully connected layer to obtain the event category prediction value; Step 25: Calculate the loss value according to the event category prediction value and the cross entropy loss function, calculate the gradient of the loss value to the model parameters through the back propagation algorithm, and use the Adam algorithm to update the model parameters using the gradient. When the accuracy, precision and recall rate of the deep learning model meet the set thresholds, stop updating the model parameters to obtain the pre-trained deep learning model.

7. A ground collapse risk resilience distributed acoustic perception system, characterized in that: A distributed acoustic sensing method for ground collapse risk resilience according to any one of claims 1 to 6 is applied, comprising a polymer measuring tube, an optical fiber, a distributed acoustic sensor demodulator and a ground monitoring workstation; The optical fiber passes through the polymer measuring tube and is connected to the distributed acoustic sensor demodulator; the distributed acoustic sensor demodulator is connected to the ground monitoring workstation; the polymer measuring tube is horizontally laid in the stratum above the collapse hazard target area; the ground monitoring workstation is loaded with a pre-trained deep learning model to process the acoustic vibration signal transmitted by the distributed acoustic sensor demodulator.

8. A ground collapse risk resilience distributed acoustic sensing system according to claim 7, characterized in that: A guide groove is axially provided on the neutral layer on both sides of the outer wall of the polymer measuring tube, and a group of plugs are respectively provided at both ends; the plug includes a left plug and a right plug with a semicircular protrusion structure, wherein the left plug is provided with a left positioning hole on the semi-cylindrical side surface and a male buckle is provided on the semicircular top surface, and the right plug is provided with a right positioning hole on the semi-cylindrical side surface and a female buckle is provided on the semicircular top surface, and the left plug and the right plug are fixed to the two ends of the polymer measuring tube by the matching male buckle and the female buckle; the two groups of optical fibers are respectively placed in the guide grooves on both sides, and respectively pass through the corresponding left positioning hole and the right positioning hole.

9. A distributed acoustic sensing system for ground collapse risk resilience according to claim 8, characterized in that: Epoxy resin is poured into the guide groove to fix the optical fiber.

10. A distributed acoustic sensing system for ground collapse risk resilience according to claim 7, characterized in that: The end of the optical fiber extends to the ground through a protective sleeve and is connected to the distributed acoustic sensor demodulator, which is used to receive the acoustic vibration signal collected by the optical fiber packaged in the guide groove of the polymer measuring tube, and the acoustic vibration signal includes the acoustic vibration information generated by the normal state, the groundwater gas dynamic action event and / or the activation event of the goaf and soil cave development; The ground monitoring workstation receives the acoustic vibration signal demodulated by the distributed acoustic sensor demodulator, and uses the pre-trained deep learning model to perform event recognition, obtains the ground collapse risk level after feature analysis, and generates early warning information.

Images