CN112541480A - Online identification method and system for tunnel foreign matter invasion event - Google Patents
Online identification method and system for tunnel foreign matter invasion event Download PDFInfo
- Publication number
- CN112541480A CN112541480A CN202011560037.1A CN202011560037A CN112541480A CN 112541480 A CN112541480 A CN 112541480A CN 202011560037 A CN202011560037 A CN 202011560037A CN 112541480 A CN112541480 A CN 112541480A
- Authority
- CN
- China
- Prior art keywords
- signal
- event
- value
- time
- tunnel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/04—Architectural design, interior design
Abstract
The invention discloses a tunnel foreign matter invasion event online identification method and a system, belonging to the technical field of security monitoring, wherein the method comprises the following steps: obtaining sound field distribution data in the whole tunnel based on an optical fiber sound wave detection technology and combined with a three-dimensional mapping imaging technology; extracting a time domain bottom noise value, a frequency domain bottom noise value, a time domain maximum amplitude value and a frequency domain maximum energy peak value of the sound wave signal, and distinguishing a bad channel signal, a background noise signal and an abnormal signal; and learning and training the characteristics of the abnormal signals through a neural network model based on the time-space characteristic information of the abnormal signals and the characteristic information of the main frequency band energy distributed along with time so as to identify destructive intrusion events in the abnormal signals. The method is suitable for the online identification of the foreign matter invasion in various underground infrastructures such as tunnels, comprehensive pipe galleries and the like, and has the characteristics of no blind area, no blind time, high real-time performance, high identification accuracy, intelligence, low risk, low operation cost and the like.
Description
Technical Field
The invention belongs to the technical field of security monitoring, and particularly relates to a tunnel foreign matter invasion event online identification method and system.
Background
With the acceleration of the urbanization process of China, the pressure brought to urban traffic by the increase of urban population becomes increasingly obvious. However, the development of urbanization can never be constrained by traffic pressure. Therefore, underground traffic corresponding to the traditional above-ground traffic becomes a new channel for relieving urban traffic pressure. The subway traffic is green engineering, has the advantages of environmental protection, energy conservation, small occupied area and the like, and conforms to the sustainable development strategy of China. In recent years, with the rapid development of urban subway tunnel facilities construction in China, great convenience is brought to people in urban traffic travel. However, the third-party rough construction on the ground is often prohibited, and the safety of lives and property of people is seriously threatened. For example, in 2017, the event that the tunnel is punctured and the subway is forced to stop in the process of piling construction together in Shenzhen of 12 months, so that people can be more aware of the importance of forbidding illegal construction in the subway protection area, however, most construction units have less awareness of the rail transit line protection area and weak awareness of protecting the rail transit structure. Therefore, an intelligent online monitoring technology for foreign matter invasion is urgently needed, and early warning is carried out on events such as damage to a subway tunnel structure in rough construction and the like.
At present, the technical means for monitoring the invasion of the foreign matters in the tunnel mainly comprise: manual inspection, camera video monitoring and optical fiber distributed sound wave sensing technology. The manual inspection technology can only realize the intrusion monitoring in a small range, and has low efficiency and wastes time and labor. And camera video monitoring can realize whole tunnel invasion monitoring along the line, but its is with high costs, easily receives weather environment's influence moreover, especially is difficult to realize the on-line monitoring to destructive invasion event in heavy fog weather. The optical fiber distributed acoustic wave sensing technology is taken as a new intelligent sensing technology, has a series of outstanding advantages of wide sensing coverage, high anti-interference performance, high sensitivity, good tolerance to severe environment and the like, is favored by more and more researchers in recent years, and simultaneously, the rapid development of the artificial intelligence technology also injects new vitality into the distributed acoustic wave sensing technology, so that the monitoring technology based on the optical fiber is more intelligent. In recent years, optical fiber sensing and artificial intelligence technology are widely applied to the fields of structural safety monitoring, pipeline detection, foreign body intrusion monitoring, perimeter security and the like. Although these methods have certain advantages in each aspect, there are problems of low recognition accuracy, poor disturbance resistance, poor real-time performance, and the like.
Disclosure of Invention
Aiming at the defects and improvement requirements of the prior art, the invention provides a tunnel foreign matter invasion event online identification method and a system, and aims to solve the technical problems of low identification precision, poor disturbance resistance, poor real-time performance and the like of the conventional tunnel foreign matter invasion event online identification method.
In order to achieve the above object, one aspect of the present invention provides an online identification method for a tunnel foreign object intrusion event, including the following steps:
s1, acquiring sound wave signals in the whole tunnel;
s2, calculating the time domain noise floor value D of the sound wave signaltSum frequency domain background noise value DfIf the time domain floor noise value DtIf the threshold value TH is larger than the threshold value TH, the signal is judged to be a bad channel signal or a first class abnormal signal, and is distinguished through a classifier; otherwise, go to S3;
s3, calculating the time domain maximum amplitude Y of the rest signalssMaximum energy peak Y of sum frequency domainfIf the maximum amplitude Y of the time domain issGreater than the time domain amplitude threshold T1 and the frequency domain maximum energy peak YfIf the frequency domain energy threshold is larger than the frequency domain energy threshold T2, the signal is judged to be a second type abnormal signal; otherwise, judging the signal as a background noise signal; wherein, T1 ═ α Dt，T2＝βDfAlpha and beta are multiple factors;
and S4, acquiring the space-time characteristic information of the first-class abnormal signals and the second-class abnormal signals and the characteristic information of the main frequency band energy distributed along with time so as to identify destructive intrusion events in the first-class abnormal signals and the second-class abnormal signals.
Further, in the step S2,
calculating the time domain background noise value D of the sound wave signaltThe method comprises the following steps:
s21, setting a time domain sliding window with the width of m, taking an absolute value of the time domain signal amplitude in the time domain sliding window, and sequentially finding out the maximum amplitude in the time domain sliding window by moving the time domain sliding window;
s22, finding out the minimum value in all the maximum amplitude values in S21, dividing the time domain sliding window corresponding to the minimum value into n parts, averaging the maximum amplitude values in each part of slices to obtain a signal time domain background noise value Dt；
Calculating the frequency domain background noise value D of the sound wave signalfThe method comprises the following steps:
s23, performing fast Fourier transform on the time domain sound wave signal to obtain frequency spectrum energy information, and taking the energy average value of all points in the frequency spectrum as a signal frequency domain background noise value Df。
Further, in the step S2,
the method for distinguishing the bad track signal from the first class abnormal signal through the classifier comprises the following steps:
s24, setting a frequency domain sliding window with the width of M, and sequentially carrying out integral summation on the energy in each frequency domain sliding window to obtain the energy after each frequency domain sliding window is integrated;
s25, determining a spectrum energy threshold G based on the maximum value of the energy after all frequency domain sliding window integration;
s26, finding out the number f of energy peaks exceeding the spectrum energy threshold G1And corresponding frequency values, and divideRespectively calculating the average f of frequency values corresponding to all energy peaks exceeding the spectral energy threshold value G2Sum variance f3The feature matrix [ f ]1，f2，f3]As the identification basis of the classifier, to distinguish the bad track signal from the first kind of abnormal signal.
Further, the step S1 includes:
acquiring an external sound wave signal detected by each sensor in the tunnel; carrying out three-dimensional modeling on an actual tunnel structure, and calculating a space coordinate corresponding to each sensor; and mapping the external sound wave signals into the three-dimensional model according to the space coordinate information to finally obtain the sound field distribution data in the whole tunnel.
Further, in the step S4,
acquiring space-time characteristic information of the first abnormal signal and the second abnormal signal, wherein the space-time characteristic information comprises the following steps:
s41, quantizing the sound field distribution data through a time-space reconstruction technology, and forming a multi-frame tunnel sound field space distribution gray map according to a time sequence;
the spatio-temporal feature information σ is expressed as:
wherein N is the number of points of the first-class abnormal signal and the second-class abnormal signal, (X)i,Yi) The spatial region coordinate corresponding to the maximum gray level in the ith frame tunnel sound field spatial distribution gray scale image,
the average value of the spatial region coordinate points corresponding to the maximum gray level for all frames.
Further, in the step S4,
acquiring characteristic information of the distribution of the main frequency band energy of the first-class abnormal signals and the second-class abnormal signals along with time, wherein the characteristic information comprises the following steps:
s42, for the first kind abnormal signal and the second kind abnormal signalCarrying out spectrum analysis to obtain main frequency band distribution delta f ═ f corresponding to the signalH-fLWherein f isHUpper cut-off frequency of main frequency band, fLThe lower cut-off frequency of the main frequency band;
the characteristic information E (t) of the energy distribution of the main frequency band along with the time is represented as:
wherein S (t, f) is a time-frequency two-dimensional matrix obtained by S transformation of the signal, and
x (t) is the acoustic signal, t is the time,τused to control the position of the gaussian window in the S-transform in the time domain.
Further, in the step S4,
identifying a destructive intrusion event in the first and second types of exception signals, comprising:
s43, sending the characteristic parameter matrixes [ sigma, E (t) ] of the first type abnormal signals and the second type abnormal signals into a BP neural network model for training to obtain an optimal model;
and S44, identifying destructive intrusion events in the first type abnormal signals and the second type abnormal signals based on the optimal model.
Further, still include:
adding a Softmax layer at the output end of the BP neural network model, and calculating to obtain the probability value P of the corresponding event to which the sample belongsi：
Wherein K is the total number of event types, z is the output result of the BP neural network, PiProbability value of the sample belonging to the i-th type event;
calculating a risk value R:
Wherein α (i | j) is the risk that the ith type event is judged as the jth type event by mistake, and P (i | j) is the probability that the ith type event is judged as the jth type event by mistake;
when the risk value R is smaller than the risk factor R, outputting the event type corresponding to the maximum probability of the predicted event; otherwise, feeding back to the Softmax layer to re-estimate the probability value of each event prediction until R is smaller than R.
Further, still include:
s5, classifying and storing various predicted event information according to event types, and carrying out deep analysis and extraction on common characteristics of the same type of events; introducing a long-term and short-term memory network to remove redundant and miscellaneous information of various events and continuously updating various event databases; training and optimizing parameters of the BP neural network model by reversely adjusting the training parameters in the BP neural network model.
In another aspect, the present invention provides an online identification system for tunnel foreign object intrusion events, comprising:
the acoustic signal acquisition module is used for acquiring acoustic signals in the whole tunnel;
a first-class abnormal signal identification module for calculating the time domain noise floor value D of the sound wave signaltSum frequency domain background noise value DfIf the time domain floor noise value DtIf the threshold value TH is larger than the threshold value TH, the signal is judged to be a bad channel signal or a first class abnormal signal, and is distinguished through a classifier; otherwise, executing the operation of the second-class abnormal signal identification module;
a second abnormal signal identification module for calculating the time domain maximum amplitude Y of the rest signalssMaximum energy peak Y of sum frequency domainfIf the maximum amplitude Y of the time domain issGreater than the time domain amplitude threshold T1 and the frequency domain maximum energy peak YfIf the frequency domain energy threshold is larger than the frequency domain energy threshold T2, the signal is judged to be a second type abnormal signal; otherwise, judging the signal as a background noise signal; wherein, T1 ═ α Dt，T2＝βDf，αBeta is a multiple factor;
and the destructive intrusion event identification module is used for acquiring the time-space characteristic information of the first-class abnormal signals and the second-class abnormal signals and the characteristic information of the distribution of the main frequency band energy along with time so as to identify destructive intrusion events in the first-class abnormal signals and the second-class abnormal signals.
Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained:
(1) the method distinguishes the bad channel signal, the background noise signal and the abnormal signal by extracting a time domain bottom noise value, a frequency domain bottom noise value, a time domain maximum amplitude value and a frequency domain maximum energy peak value of the sound wave signal; and identifying a destructive intrusion event in the abnormal signal based on the time-space characteristic information of the abnormal signal and the characteristic information of the energy distribution of the main frequency band along with the time. Therefore, the recognition rate of the recognition system can be improved, and the operation efficiency of the system can be improved.
(2) The invention provides a tunnel sound field three-dimensional mapping technology. Laying a plurality of sensing optical cables along the tunnel direction, connecting the optical cables at different positions to form an acoustic wave sensing network, then carrying out three-dimensional modeling on the actual tunnel structure by using engineering modeling software, guiding the acoustic wave sensing network in the tunnel into a three-dimensional model, calculating a space coordinate corresponding to each sensor, mapping an acoustic wave signal detected by each sensor into the tunnel three-dimensional model according to space coordinate information, and finally obtaining an acoustic field space distribution diagram in the actual tunnel. Compared with the traditional method, the tunnel sound field distribution information obtained by the method is more practical, and the accuracy of intrusion event identification can be further improved.
(3) The invention provides a low-risk intrusion event identification and prediction method. Because different events have different carrying threats, the risks brought by misjudging a non-threat event as a destructive intrusion event and misjudging a destructive intrusion event as a non-threat event are different, in order to reduce the risk of misinformation of the identification system, the risk of each type of intrusion event is analyzed and the risk brought by the misinformation of each type of event is evaluated, and the risk caused by the misjudgment of the system prediction is reduced by feedback regulation of the output of a neural network model. Compared with the traditional method, the predicted event result not only has high identification precision, but also has low risk.
(4) The invention provides a method for regularly arranging various event information databases and regularly optimizing an intelligent identification network. The method has the advantages that the redundant and miscellaneous information of various events is periodically removed, so that the space is saved, the new data information is stored, meanwhile, the training parameters in the neural network model are reversely adjusted according to the identification condition of various events, the parameters of the neural network model are periodically and purposefully trained and optimized, and the intelligence of the subway tunnel intrusion event monitoring system is continuously improved. The accuracy of discernment not only can be improved, the security that can also promote subway operation by a wide margin.
Drawings
FIG. 1 is a flowchart of an online identification method for tunnel foreign object intrusion events according to the present invention;
FIG. 2 is a flow chart of the first type and second type of exception signal identification provided by the present invention;
FIG. 3 is a flow chart of the destructive intrusion event identification provided by the present invention;
fig. 4 is a block diagram of a system for online identification of a tunnel foreign object intrusion event according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Example one
Referring to fig. 1, in combination with fig. 2 and fig. 3, the embodiment provides an online identification method for a tunnel foreign object intrusion event, including the following steps:
s1, acquiring sound wave signals in the whole tunnel;
it should be noted that the sound wave signals in the tunnel include bad track signals, background noise and abnormal signals, and the abnormal signals include tunnel top car running signals, tunnel train running signals and destructive construction intrusion signals.
Specifically, a plurality of sensing optical cables are laid along the tunnel direction, and the optical cables at different positions are connected to form an acoustic wave sensing network; and carrying out three-dimensional modeling on an actual tunnel structure by utilizing engineering modeling software, introducing the sound wave sensing network in the tunnel into a three-dimensional model, calculating a space coordinate corresponding to each sensor, mapping the sound wave signal detected by each sensor into the tunnel three-dimensional model according to space coordinate information, and finally obtaining the sound field space distribution map in the whole tunnel.
S2, calculating the time domain noise floor value D of the sound wave signaltSum frequency domain background noise value DfIf the time domain floor noise value DtIf the threshold value TH is larger than the threshold value TH, the signal is judged to be a bad channel signal or a first class abnormal signal, and is distinguished through a classifier; otherwise, go to S3;
it should be noted that, because the intensity of the backscattered light signals at some positions in the sensing optical fiber is weak, bad channels may appear at the corresponding positions of the demodulation end, and these bad channels affect the early warning function of the monitoring system, so we need to distinguish the bad channels, the background noise, and the abnormal signals. Meanwhile, because the time domain bottom noise value of the bad channel signal is far larger than that of the background noise signal, respectively counting the time domain bottom noise values of a plurality of groups of bad channel signals and background noise signals, and finding out a proper threshold TH to distinguish the bad channel signals and the background noise signals; time domain noise floor D of signaltIf the signal is greater than TH, the signal may be a bad track signal or a first type abnormal signal.
Specifically, the method comprises the following steps:
(1) method for acquiring time domain noise floor value D of signal by using sliding window slicing technologyt：
S21, firstly setting a time domain sliding window with width m, taking absolute value of time domain signal amplitude in the sliding window, finding out the maximum amplitude of the signal in the sliding window, then moving the time domain sliding window, continuously finding the next maximum amplitude until finding out the corresponding maximum amplitude in each window, then comparing the magnitudes of all the maximum amplitudes, finding out the minimum value in all the amplitudes and indexing to the position of the corresponding time domain sliding window, wherein the formula is as follows:
Smin＝Min(Max(|Si|,...,|Si+m|)) i＝N-m+1
wherein N is the number of time domain signal points, SiThe amplitude of the time domain signal corresponding to the ith data point, SminThe minimum value of the maximum amplitude values in all windows;
s22, finding out the minimum value in all the maximum amplitude values in S21, dividing the time domain sliding window corresponding to the minimum value into n parts, averaging the maximum amplitude values in each part of slices to obtain a signal time domain background noise value Dt；
(2) Calculating the frequency domain background noise value D of the acoustic wave signalfThe method comprises the following steps:
s23, Fast Fourier Transform (FFT) is carried out on the time domain sound wave signal to obtain frequency spectrum energy information of the time domain sound wave signal, and then the energy average value of all points in the frequency spectrum is solved to be the signal frequency domain bottom noise value DfThe formula is as follows:
(3) distinguishing a bad channel signal from a first class of abnormal signals through a classifier:
s24, firstly, Fourier transform is used to obtain the spectrum energy information of the signal, the bad track signal spectrum energy peak is distributed widely and mainly concentrated in the high frequency component, and the first abnormal signal spectrum energy peak is distributed concentratedly and mainly distributed in the low frequency component. And then selecting a sliding integration window with the width of MHz on the frequency domain, wherein the moving step of the sliding window is qHz, and sequentially carrying out integration summation on the energy in each window to obtain the integrated energy of each sliding window. The sliding integration window can not only highlight the difference of the frequency domain characteristics of the bad track signal and the first type of abnormal signal, but also eliminate an energy peak caused by uncertain noise in the sound wave signal acquisition module;
s25, setting a spectrum energy threshold G, wherein G is p% of the maximum value of the integral energy of the sliding window;
s26, finding out the number f of energy peaks exceeding the spectrum energy threshold G1And corresponding frequency values, and respectively calculating the average f of the frequency values corresponding to all energy peaks exceeding the spectral energy threshold value G2Sum variance f3The feature matrix [ f ]1，f2，f3]As the identification basis of the classifier, to distinguish the bad track signal from the first kind of abnormal signal.
S3, calculating the time domain maximum amplitude Y of the rest signalssMaximum energy peak Y of sum frequency domainfIf the maximum amplitude Y of the time domain issGreater than the time domain amplitude threshold T1 and the frequency domain maximum energy peak YfIf the frequency domain energy threshold is larger than the frequency domain energy threshold T2, the signal is judged to be a second type abnormal signal; otherwise, judging the signal as a background noise signal; wherein, T1 ═ α Dt，T2＝βDfAlpha and beta are multiple factors;
and S4, acquiring the space-time characteristic information of the first-class abnormal signals and the second-class abnormal signals and the characteristic information of the main frequency band energy distributed along with time so as to identify destructive intrusion events in the first-class abnormal signals and the second-class abnormal signals.
Specifically, for the automobile running on the ground and the train signal running in the tunnel, the spatial position of the signal changes with time, and the spatial position of the signal carrying the threatened excavation signal and piling signal does not change with time basically. Meanwhile, the distribution of the main frequency bands of the events has certain difference, and the energy of the main frequency bands of the signals of destructive excavation and piling construction events is continuously enhanced along with the time. Therefore, the time-space information of the signal source and the time-frequency energy distribution information of the signal can be used as the identification characteristic basis of the events to identify the destructive intrusion event. The method comprises the following steps:
s41, quantizing the sound field distribution data through a time-space reconstruction technology, and forming a multi-frame tunnel sound field space distribution gray map according to a time sequence;
the spatio-temporal feature information σ is expressed as:
wherein N is the number of points of the first-class abnormal signal and the second-class abnormal signal, (X)i,Yi) The spatial region coordinate corresponding to the maximum gray level in the ith frame tunnel sound field spatial distribution gray scale image,
the average value of the spatial region coordinate points corresponding to the maximum gray level for all frames.
S42, performing spectrum analysis on the first-type abnormal signal and the second-type abnormal signal to obtain a main frequency band distribution Δ f ═ f corresponding to the signalsH-fLWherein f isHUpper cut-off frequency of main frequency band, fLThe lower cut-off frequency of the main frequency band;
the characteristic information E (t) of the energy distribution of the main frequency band along with the time is represented as:
wherein S (t, f) is a time-frequency two-dimensional matrix obtained by S transformation of the signal, and
x (t) is a signal.
S43, fusing the space-time characteristic information sigma and the time-frequency energy distribution characteristic information E (t) of each type of event to form a characteristic parameter matrix [ sigma, E (t) ], and carrying out event marking on the characteristic parameter matrix of each type of event. Then building a BP neural network model, sending the characteristic parameter matrix [ sigma, E (t) of each event into the BP neural network model for training, performing back propagation in the network model according to an error function of a network model output result and modifying the weight of each layer of neuron, and finally enabling the error output by the network model to be minimum, completing the training of the BP neural network model so as to obtain an optimal model;
further, a trained BP neural network model is used for carrying out destructive intrusion event recognition test, a Softmax layer is added at the output end of the BP neural network model to obtain the probability value of the event corresponding to the sample, and the probability value can be obtained by the following formula:
wherein K is the total number of event types, z is the output result of the BP neural network, PiProbability value of the sample belonging to the i-th type event; because different events have different carrying threats, the risks brought by misjudging a non-threat event as a destructive intrusion event and misjudging a destructive intrusion event as a non-threat event are different, in order to reduce the risk of false alarm of an intelligent identification system of the intrusion event, the risk of each type of intrusion event is analyzed and the risk brought by false alarm of each type of event is evaluated, and finally, a risk value R predicted by an intelligent identification module is calculated, wherein the following formula is shown as follows:
Where α (i | j) is the risk of the i-th event being mistaken for the j-th event, and P (i | j) is the probability of the i-th event being mistaken for the j-th event.
And then setting a risk factor R for the identification system, and when the risk value R is smaller than the risk factor R, indicating that the risk of the prediction result of the identification system is low, and outputting the event type corresponding to the maximum probability of the predicted event. If R is larger than R, the sample is marked as a high-risk uncertain event, and then the sample is fed back to a Softmax layer to re-estimate the probability value of each event prediction until R is smaller than R.
And S44, identifying destructive intrusion events in the first type abnormal signals and the second type abnormal signals based on the optimal model.
Further, still include:
and S5, classifying and storing various predicted event information according to event types, deeply analyzing and extracting common characteristics of the same event, and introducing a long-term and short-term memory network to remove redundant and miscellaneous information of various events as the data information of various events is accumulated continuously so as to save space for storing new data information and continuously update various event databases. The intelligent recognition network model parameters are regularly optimized according to various event data information bases, the recognition accuracy and the recognition time of various events are analyzed, the training parameters such as the learning rate, the iteration times and the like in the BP neural network model are reversely adjusted, the BP neural network model parameters are optimized through regular repeated training, and the intelligence of the tunnel foreign matter invasion event online recognition system is continuously improved.
Example two
Referring to fig. 4, the present embodiment provides an online identification system for a tunnel foreign object intrusion event, including:
the acoustic signal acquisition module is used for acquiring acoustic signals in the whole tunnel;
a first-class abnormal signal identification module for calculating the time domain noise floor value D of the sound wave signaltSum frequency domain background noise value DfIf the time domain floor noise value DtIf the threshold value TH is larger than the threshold value TH, the signal is judged to be a bad channel signal or a first class abnormal signal, and is distinguished through a classifier; otherwise, executing the operation of the second-class abnormal signal identification module;
a second abnormal signal identification module for calculating the time domain maximum amplitude Y of the rest signalssMaximum energy peak Y of sum frequency domainfIf the maximum amplitude Y of the time domain issGreater than the time domain amplitude threshold T1 and the frequency domain maximum energy peak YfIf the frequency domain energy threshold is larger than the frequency domain energy threshold T2, the signal is judged to be a second type abnormal signal; otherwise, judging the signal as a background noise signal; wherein, T1 ═ α Dt，T2＝βDfAlpha and beta are multiple factors;
and the destructive intrusion event identification module is used for acquiring the time-space characteristic information of the first-class abnormal signals and the second-class abnormal signals and the characteristic information of the distribution of the main frequency band energy along with time so as to identify destructive intrusion events in the first-class abnormal signals and the second-class abnormal signals.
The division of each module in the online identification system for tunnel foreign matter intrusion events is only used for illustration, and in other embodiments, the online identification system for tunnel foreign matter intrusion events can be divided into different modules as required to complete all or part of the functions of the system.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A tunnel foreign matter invasion event online identification method is characterized by comprising the following steps:
s1, acquiring sound wave signals in the whole tunnel;
s2, calculating the time domain noise floor value D of the sound wave signaltSum frequency domain background noise value DfIf the time domain floor noise value DtIf the threshold value TH is larger than the threshold value TH, the signal is judged to be a bad channel signal or a first class abnormal signal, and is distinguished through a classifier; otherwise, go to S3;
s3, calculating the time domain maximum amplitude Y of the rest signalssMaximum energy peak Y of sum frequency domainfIf the maximum amplitude Y of the time domain issGreater than the time domain amplitude threshold T1 and the frequency domain maximum energy peak YfIf the frequency domain energy threshold is larger than the frequency domain energy threshold T2, the signal is judged to be a second type abnormal signal; otherwise, judging the signal as a background noise signal; wherein, T1 ═ α Dt，T2＝βDfAlpha and beta are multiple factors;
and S4, acquiring the space-time characteristic information of the first-class abnormal signals and the second-class abnormal signals and the characteristic information of the main frequency band energy distributed along with time so as to identify destructive intrusion events in the first-class abnormal signals and the second-class abnormal signals.
2. The online identification method for tunnel foreign object intrusion events according to claim 1, wherein in the step S2,
calculating the time domain background noise value D of the sound wave signaltThe method comprises the following steps:
s21, setting a time domain sliding window with the width of m, taking an absolute value of the time domain signal amplitude in the time domain sliding window, and sequentially finding out the maximum amplitude in the time domain sliding window by moving the time domain sliding window;
s22, finding out the minimum value in all the maximum amplitude values in S21, dividing the time domain sliding window corresponding to the minimum value into n parts, averaging the maximum amplitude values in each part of slices to obtain a signal time domain background noise value Dt；
Calculating the frequency domain background noise value D of the sound wave signalfThe method comprises the following steps:
s23, performing fast Fourier transform on the time domain sound wave signal to obtain frequency spectrum energy information, and taking the energy average value of all points in the frequency spectrum as a signal frequency domain background noise value Df。
3. The online identification method for tunnel foreign object intrusion events according to claim 1, wherein in the step S2,
the method for distinguishing the bad track signal from the first class abnormal signal through the classifier comprises the following steps:
s24, setting a frequency domain sliding window with the width of M, and sequentially carrying out integral summation on the energy in each frequency domain sliding window to obtain the energy after each frequency domain sliding window is integrated;
s25, determining a spectrum energy threshold G based on the maximum value of the energy after all frequency domain sliding window integration;
s26, finding out the number f of energy peaks exceeding the spectrum energy threshold G1And corresponding frequency values, and respectively calculating the average f of the frequency values corresponding to all energy peaks exceeding the spectral energy threshold value G2Sum variance f3The feature matrix [ f ]1，f2，f3]As the identification basis of the classifier, to distinguish the bad track signal from the first kind of abnormal signal.
4. The online identification method for the intrusion event of the tunnel foreign object according to claim 1, wherein the step S1 includes:
acquiring an external sound wave signal detected by each sensor in the tunnel; carrying out three-dimensional modeling on an actual tunnel structure, and calculating a space coordinate corresponding to each sensor; and mapping the external sound wave signals into the three-dimensional model according to the space coordinate information to finally obtain the sound field distribution data in the whole tunnel.
5. The online identification method for tunnel foreign object intrusion events according to claim 4, wherein in the step S4,
acquiring space-time characteristic information of the first abnormal signal and the second abnormal signal, wherein the space-time characteristic information comprises the following steps:
s41, quantizing the sound field distribution data through a time-space reconstruction technology, and forming a multi-frame tunnel sound field space distribution gray map according to a time sequence;
the spatio-temporal feature information σ is expressed as:
wherein N is the number of points of the first-class abnormal signal and the second-class abnormal signal, (X)i,Yi) The spatial region coordinate corresponding to the maximum gray level in the ith frame tunnel sound field spatial distribution gray scale image,
the average value of the spatial region coordinate points corresponding to the maximum gray level for all frames.
6. The online identification method for tunnel foreign object intrusion events according to claim 5, wherein in the step S4,
acquiring characteristic information of the distribution of the main frequency band energy of the first-class abnormal signals and the second-class abnormal signals along with time, wherein the characteristic information comprises the following steps:
s42, carrying out spectrum analysis on the first type abnormal signals and the second type abnormal signals to obtain the correspondence of the signalsDistribution of main frequency band of (a) fH-fLWherein f isHUpper cut-off frequency of main frequency band, fLThe lower cut-off frequency of the main frequency band;
the characteristic information E (t) of the energy distribution of the main frequency band along with the time is represented as:
wherein S (t, f) is a time-frequency two-dimensional matrix obtained by S transformation of the signal, and
x (t) is a sound wave signal, t is a time, and tau is used for controlling the position of a Gaussian window in S transformation in a time domain.
7. The online identification method for tunnel foreign object intrusion events according to claim 6, wherein in the step S4,
identifying a destructive intrusion event in the first and second types of exception signals, comprising:
s43, sending the characteristic parameter matrixes [ sigma, E (t) ] of the first type abnormal signals and the second type abnormal signals into a BP neural network model for training to obtain an optimal model;
and S44, identifying destructive intrusion events in the first type abnormal signals and the second type abnormal signals based on the optimal model.
8. The method for online identification of tunnel foreign body invasion events according to claim 7, further comprising:
adding a Softmax layer at the output end of the BP neural network model, and calculating to obtain the probability value P of the corresponding event to which the sample belongsi：
Wherein K is the total number of event types, z is the output result of the BP neural network, PiProbability value of the sample belonging to the i-th type event;
calculating a risk value R:
Wherein α (i | j) is the risk that the ith type event is judged as the jth type event by mistake, and P (i | j) is the probability that the ith type event is judged as the jth type event by mistake;
when the risk value R is smaller than the risk factor R, outputting the event type corresponding to the maximum probability of the predicted event; otherwise, feeding back to the Softmax layer to re-estimate the probability value of each event prediction until R is smaller than R.
9. The method for online identification of the invasion event of the foreign body into the tunnel according to claim 7 or 8, further comprising:
s5, classifying and storing various predicted event information according to event types, and carrying out deep analysis and extraction on common characteristics of the same type of events; introducing a long-term and short-term memory network to periodically clear redundant and miscellaneous information of various events and continuously update various event databases; training and optimizing parameters of the BP neural network model by reversely adjusting the training parameters in the BP neural network model.
10. An online tunnel foreign object intrusion event identification system is characterized by comprising:
the acoustic signal acquisition module is used for acquiring acoustic signals in the whole tunnel;
a first-class abnormal signal identification module for calculating the time domain noise floor value D of the sound wave signaltSum frequency domain background noise value DfIf the time domain floor noise value DtIf the threshold value TH is larger than the threshold value TH, the signal is judged to be a bad channel signal or a first class abnormal signal, and is distinguished through a classifier; otherwise, executing the operation of the second-class abnormal signal identification module;
anomaly of the second typeA signal identification module for calculating the time domain maximum amplitude Y of the rest signalssMaximum energy peak Y of sum frequency domainfIf the maximum amplitude Y of the time domain issGreater than the time domain amplitude threshold T1 and the frequency domain maximum energy peak YfIf the frequency domain energy threshold is larger than the frequency domain energy threshold T2, the signal is judged to be a second type abnormal signal; otherwise, judging the signal as a background noise signal; wherein, T1 ═ α Dt，T2＝βDfAlpha and beta are multiple factors;
and the destructive intrusion event identification module is used for acquiring the time-space characteristic information of the first-class abnormal signals and the second-class abnormal signals and the characteristic information of the distribution of the main frequency band energy along with time so as to identify destructive intrusion events in the first-class abnormal signals and the second-class abnormal signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011560037.1A CN112541480B (en) | 2020-12-25 | 2020-12-25 | Online identification method and system for tunnel foreign matter invasion event |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011560037.1A CN112541480B (en) | 2020-12-25 | 2020-12-25 | Online identification method and system for tunnel foreign matter invasion event |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112541480A true CN112541480A (en) | 2021-03-23 |
| CN112541480B CN112541480B (en) | 2022-06-17 |
Family
ID=75017398
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011560037.1A Active CN112541480B (en) | 2020-12-25 | 2020-12-25 | Online identification method and system for tunnel foreign matter invasion event |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112541480B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114354744A (en) * | 2021-12-03 | 2022-04-15 | 国网河南省电力公司商丘供电公司 | Power cable external damage prevention monitoring method, system and device |
| CN116204784A (en) * | 2022-12-30 | 2023-06-02 | 成都天仁民防科技有限公司 | DAS-based subway tunnel external hazard operation intrusion recognition method |
| CN117147966A (en) * | 2023-08-30 | 2023-12-01 | 中国人民解放军军事科学院系统工程研究院 | Electromagnetic spectrum signal energy anomaly detection method |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101031956A (en) * | 2004-07-22 | 2007-09-05 | 索福特迈克斯有限公司 | Headset for separation of speech signals in a noisy environment |
| US20140200819A1 (en) * | 2013-01-14 | 2014-07-17 | Cgg Services Sa | High-fidelity adaptive curvelet domain primary-multiple separation processing of seismic data |
| CN105654645A (en) * | 2016-01-22 | 2016-06-08 | 北京北邮国安技术股份有限公司 | Optical fiber security and protection signal processing method and system |
| CN106600870A (en) * | 2016-12-31 | 2017-04-26 | 上海华魏光纤传感技术有限公司 | Distributed optical fiber fence vibration invasion identifying system |
| CN107689127A (en) * | 2017-08-17 | 2018-02-13 | 北京北邮国安技术股份有限公司 | With the adaptive dual recognition methods of optical fiber security signal for adjusting ginseng function and system |
| US20180357542A1 (en) * | 2018-06-08 | 2018-12-13 | University Of Electronic Science And Technology Of China | 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method |
| US20190158340A1 (en) * | 2012-12-05 | 2019-05-23 | Origin Wireless, Inc. | Apparatus, systems and methods for fall-down detection based on a wireless signal |
| CN111341049A (en) * | 2020-03-31 | 2020-06-26 | 华中科技大学 | Tunnel foreign matter invasion identification and positioning method and device |
| CN111983563A (en) * | 2020-08-30 | 2020-11-24 | 山东省科学院海洋仪器仪表研究所 | Remote advanced anti-submarine early warning array and system based on distributed optical fiber acoustic wave sensing |
| CN112032575A (en) * | 2020-08-10 | 2020-12-04 | 武汉理工大学 | Pipeline safety monitoring method and system based on weak grating and storage medium |
-
2020
- 2020-12-25 CN CN202011560037.1A patent/CN112541480B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101031956A (en) * | 2004-07-22 | 2007-09-05 | 索福特迈克斯有限公司 | Headset for separation of speech signals in a noisy environment |
| US20190158340A1 (en) * | 2012-12-05 | 2019-05-23 | Origin Wireless, Inc. | Apparatus, systems and methods for fall-down detection based on a wireless signal |
| US20140200819A1 (en) * | 2013-01-14 | 2014-07-17 | Cgg Services Sa | High-fidelity adaptive curvelet domain primary-multiple separation processing of seismic data |
| CN105654645A (en) * | 2016-01-22 | 2016-06-08 | 北京北邮国安技术股份有限公司 | Optical fiber security and protection signal processing method and system |
| CN106600870A (en) * | 2016-12-31 | 2017-04-26 | 上海华魏光纤传感技术有限公司 | Distributed optical fiber fence vibration invasion identifying system |
| CN107689127A (en) * | 2017-08-17 | 2018-02-13 | 北京北邮国安技术股份有限公司 | With the adaptive dual recognition methods of optical fiber security signal for adjusting ginseng function and system |
| US20180357542A1 (en) * | 2018-06-08 | 2018-12-13 | University Of Electronic Science And Technology Of China | 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method |
| CN111341049A (en) * | 2020-03-31 | 2020-06-26 | 华中科技大学 | Tunnel foreign matter invasion identification and positioning method and device |
| CN112032575A (en) * | 2020-08-10 | 2020-12-04 | 武汉理工大学 | Pipeline safety monitoring method and system based on weak grating and storage medium |
| CN111983563A (en) * | 2020-08-30 | 2020-11-24 | 山东省科学院海洋仪器仪表研究所 | Remote advanced anti-submarine early warning array and system based on distributed optical fiber acoustic wave sensing |
Non-Patent Citations (3)
| Title |
|---|
| GUANDING YU等: "Energy Efficiency Tradeoff in Interference Channels", 《IEEE ACCESS》 * |
| 彭宽等: "基于时/频域综合特征提取的分布式光纤入侵监测系统事件识别方法", 《光学学报》 * |
| 马群凯: "输气管道第三方破坏类型判别及定位技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅰ辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114354744A (en) * | 2021-12-03 | 2022-04-15 | 国网河南省电力公司商丘供电公司 | Power cable external damage prevention monitoring method, system and device |
| CN114354744B (en) * | 2021-12-03 | 2023-08-29 | 国网河南省电力公司商丘供电公司 | Power cable external damage prevention monitoring method, system and device |
| CN116204784A (en) * | 2022-12-30 | 2023-06-02 | 成都天仁民防科技有限公司 | DAS-based subway tunnel external hazard operation intrusion recognition method |
| CN116204784B (en) * | 2022-12-30 | 2023-09-08 | 成都天仁民防科技有限公司 | DAS-based subway tunnel external hazard operation intrusion recognition method |
| CN117147966A (en) * | 2023-08-30 | 2023-12-01 | 中国人民解放军军事科学院系统工程研究院 | Electromagnetic spectrum signal energy anomaly detection method |
| CN117147966B (en) * | 2023-08-30 | 2024-05-07 | 中国人民解放军军事科学院系统工程研究院 | Electromagnetic spectrum signal energy anomaly detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112541480B (en) | 2022-06-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112541480B (en) | Online identification method and system for tunnel foreign matter invasion event | |
| US11562224B2 (en) | 1D-CNN-based distributed optical fiber sensing signal feature learning and classification method | |
| CN113837477B (en) | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data | |
| Ruano et al. | Seismic detection using support vector machines | |
| Scarpetta et al. | Automatic classification of seismic signals at Mt. Vesuvius volcano, Italy, using neural networks | |
| CN110781266B (en) | Urban perception data processing method based on time-space causal relationship | |
| CN102841131B (en) | Intelligent steel cord conveyer belt defect identification method and intelligent steel cord conveyer belt defect identification system | |
| CN108345666A (en) | A kind of vehicle abnormality track-detecting method based on time-space isolated point | |
| CN110111578B (en) | Tensor recovery-based sporadic traffic congestion detection method | |
| CN108009690A (en) | A kind of routine bus system optimized based on modularity steals group's automatic testing method | |
| CN113096397A (en) | Traffic jam analysis method based on millimeter wave radar and video detection | |
| CN104766099A (en) | Distributed optical fiber circumference vibration signal processing and recognizing method based on image | |
| CN109357747A (en) | A kind of identification of online train and speed estimation method based on fiber-optic vibration signal | |
| CN113361825A (en) | Early warning method and system for trampling accident | |
| CN114926825A (en) | Vehicle driving behavior detection method based on space-time feature fusion | |
| CN113947902A (en) | Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model | |
| CN116631186A (en) | Expressway traffic accident risk assessment method and system based on dangerous driving event data | |
| CN114818817A (en) | Weak fault recognition system and method for capacitive voltage transformer | |
| Xie et al. | A novel quality control method of time-series ocean wave observation data combining deep-learning prediction and statistical analysis | |
| CN113156492A (en) | Real-time intelligent early warning method applied to TBM tunnel rockburst disasters | |
| CN110043808B (en) | Water supply network leakage monitoring and early warning method based on time series analysis | |
| CN116838955A (en) | Two-stage oil and gas pipeline line interference identification method | |
| Fathurrahman et al. | Urban network traffic analysis, data imputation, and flow prediction based on probabilistic PCA model of traffic volume data | |
| CN111951505B (en) | Fence vibration intrusion positioning and mode identification method based on distributed optical fiber system | |
| CN110991363B (en) | Method for extracting CO emission characteristics of coal mine safety monitoring system in different coal mining processes |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |