Drainage tunnel environment intrusion event identification method based on distributed optical fiber monitoring
The method addresses low efficiency and accuracy in drainage hole intrusion event identification by deploying a distributed optical fiber system, calibrating sensors, constructing an array signal library, and using time-frequency conversion and a neural network model to achieve high accuracy in identifying and locating intrusion events.
Patent Information
- Application Number
- CN202411196836.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-06
AI Technical Summary
Existing methods for identifying intrusion events in drainage holes using distributed optical fiber monitoring suffer from low efficiency and accuracy due to the overlap of frequency bands and the lack of utilization of multiple sensor features, especially in environments with seepage water and varying propagation distances.
A method involving distributed optical fiber deployment, sensor position calibration, qualitative identification of water flow rate, construction of an array signal library, time-frequency conversion using continuous wavelet transform, and an intrusion event recognition model based on LeNet convolutional neural network and random forest to accurately identify and locate intrusion events.
Achieves an accurate recognition rate of about 90% for various intrusion events, with rock fall and walking intrusion events reaching 94.29% and 91.76% training accuracy, respectively, and seepage water release exceeding 85%, effectively distinguishing between seepage water and background noise.
Smart Images

Figure CN119091552A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of distributed optical fiber monitoring, and in particular to a method for identifying drainage hole environment intrusion events based on distributed optical fiber monitoring. Background Art
[0002] Distributed fiber acoustic sensing (DAS) technology, as an emerging monitoring method, has been increasingly used in the field of engineering safety. Among them, intrusion event recognition, as one of the important research directions of distributed fiber acoustic sensing, is of great significance for clarifying the types of intrusion events and taking targeted prevention and control measures.
[0003] Initially, DAS monitoring intrusion event identification was determined by artificial observation of signals, which was greatly affected by the subjective judgment of personnel. In response to this, people proposed to use fast Fourier transform to convert DAS signals from time domain to frequency for identification, which provided a new perspective for intrusion event identification. However, due to the influence of propagation distance, the frequency bands of different types of intrusion events often overlap. Furthermore, DAS signal amplitude threshold, short-time energy, amplitude zero-crossing rate, Mel-frequency cepstrum coefficients, etc. were also used for intrusion event identification. At the same time, people introduced wavelet decomposition, empirical mode decomposition and other means to realize multi-scale extraction of DAS signal features, enriching the DAS signal feature database. Based on the above features, machine learning methods such as support vector machine, random forest, artificial neural network, hidden Markov model, etc. were used for DAS signal classification, which enhanced the intrusion event identification effect. In recent years, people have begun to use time-frequency technology to identify the type of intrusion event. For example, Xu et al. (2018) used short-time Fourier transform to convert DAS signals to time-frequency domain, and used this as the input of convolutional neural network to carry out intrusion event identification, achieving good recognition results.
[0004] The above methods all use a single signal to identify intrusion events. In fact, an intrusion event usually triggers multiple sensors. Ignoring a large number of features of multiple signals will reduce the accuracy of intrusion event identification. At the same time, during the distributed optical fiber monitoring of drainage holes, optical cables are laid inside and outside the drainage holes, and distributed optical fiber sensors need to be effectively screened. Furthermore, the release of seepage water in drainage holes usually exists for a long time, and there may be overlaps with intrusion events such as rock falling and people walking, so targeted identification methods need to be proposed. Therefore, it is necessary to fully explore the characteristics of multiple signals and develop an intrusion event identification method that is suitable for distributed optical fiber monitoring in drainage hole environments. Summary of the invention
[0005] The purpose of the present invention is to provide a method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring in order to address the above-mentioned shortcomings, thereby solving the problems of low efficiency and low accuracy in identifying each signal of an intrusion event in distributed optical fiber monitoring in the prior art.
[0006] The present invention is achieved through the following scheme:
[0007] A method for identifying drainage hole environment intrusion events based on distributed optical fiber monitoring comprises the following steps:
[0008] Step 1: Deploy a distributed optical fiber monitoring system adapted to the drainage hole environment;
[0009] Step 2: Distributed optical fiber sensor position calibration;
[0010] Step 3: Qualitative identification of water flow rate of drainage pipe in the drainage hole;
[0011] Step 4: Construct an array signal library suitable for drainage hole monitoring;
[0012] Step 5: Time-frequency conversion of sensor array signal continuous wavelet transform;
[0013] Step 6: Build an intrusion event recognition model based on LeNet convolutional neural network and random forest;
[0014] Step 7: Output the type and location of the intrusion event.
[0015] In step 1, the specific steps for deploying a distributed optical fiber monitoring system adapted to the drainage hole environment are as follows:
[0016] Distributed optical fiber is laid along the slope and drainage holes. The optical fiber on the slope passes through the PVC pipe fixed by steel nails and pipe clamps. When the inner wall of the drainage hole is flat, it is fixed with saddle clips. When the inner wall of the drainage hole is uneven, it is fixed with steel nails and saddle clips. A-DAS distributed optical fiber modulator and demodulator is used to monitor the acoustic signals of intrusion events.
[0017] In step 2, the specific steps of position calibration and calibration of distributed optical fiber sensor are as follows:
[0018] The turning points of the slope and drainage hole are used as important position marking points. A small hammer is used to knock on the turning points at a certain distance to generate sound wave signals. The sensor number corresponding to the maximum amplitude of the distributed optical fiber monitoring signal during the calibration period is the knocking point, which realizes the calibration of the distributed optical fiber sensor position.
[0019] In step 3, the specific steps for qualitatively identifying the water flow rate of the drainage pipe in the drainage hole are as follows:
[0020] The sensor position at the drain pipe is determined by knocking, and the template matching method is used to determine whether the sensor signal is a seepage water release event. The seepage water flow rate is then qualitatively determined based on the maximum amplitude of the signal. Among them, the template matching uses indoor experimental data of different water flow rates as the template signal, and calculates the mutual correlation coefficient between the monitoring signal and the template signal. If the mutual correlation coefficient of the signal pair is greater than 0.9, the signal is recorded as a seepage water release event.
[0021] In step 4, the specific steps for constructing an array signal library suitable for drainage hole monitoring are as follows:
[0022] According to the positional relationship between the inside and outside of the drainage hole, the number of the distributed optical fiber sensor located in the drainage hole is determined, and the signals are arranged from bottom to top in the order of the sensor numbers, and the horizontal axis time value of the signal is unified; among them, the signal length is 5s, the signal sliding window length is 1s during identification, and the signal uses a Butterworth filter to remove high-frequency noise greater than 20Hz; on this basis, the signal is augmented by using the disorder characteristics between signal channels: the data is expanded by randomly exchanging the channel order and superimposing the data, and this process does not change the category label of the signal.
[0023] In step 5, the specific steps of time-frequency conversion of sensor array signal continuous wavelet transform are as follows:
[0024] Use continuous wavelet transform to convert each signal from a time series into a time-frequency diagram to ensure that the time, frequency range and amplitude range of the time-frequency diagram are consistent, and then splice the time-frequency diagram into an array signal time-frequency diagram according to the order of the sensors.
[0025] In step six, the steps for constructing an intrusion event recognition model based on LeNet convolutional neural network and random forest are as follows: using continuous wavelet transform to realize the time-frequency conversion of sensor array signals, and then constructing a LeNet convolutional neural network model to extract the time-frequency graph features of the array signal to achieve image feature dimensionality reduction.
[0026] Specifically, the LeNet convolutional neural network for intrusion event recognition is constructed with the array signal time-frequency diagram as input and the intrusion event type as output. The training set is the array signal after monitoring and data augmentation. The input layer size is set to 64*64*3, the convolution kernel size is 5×5, the pooling layer sampling kernel size is 2×2, and the output layer is the 10 extracted feature parameters. These feature parameters are used as the model parameters for random forest classification and recognition.
[0027] In step 7, the specific steps for outputting the intrusion event type and location are as follows:
[0028] A distributed optical fiber 3D layout model including slopes and drainage holes is constructed. The types and locations of drainage hole intrusion events are displayed in the 3D model, and graded alarms are given for intrusion events such as rock falls.
[0029] The hierarchical alarm rules are as follows: general intrusion, green light alarm; major incidents, yellow light alarm; important incidents, red light alarm.
[0030] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0031] 1. By adopting this solution, the accurate recognition rate of various intrusion events and background noise is about 90%, among which the training accuracy of rock fall is 94.29%, and the prediction accuracy is 92.00%; the training and prediction accuracy of walking intrusion events are also high, 91.76% and 88.90% respectively. The training and prediction accuracy of seepage water release and background noise is slightly worse due to the less obvious waveform characteristics, but it also exceeds 85%. The results show that this patent has good applicability for the identification of drainage hole environment intrusion events.
[0032] 2. This scheme first deploys a distributed fiber optic monitoring system adapted to the drainage hole environment, and uses the tapping method to calibrate the position of the distributed fiber optic sensor; then uses the template matching method to determine whether the signal at the drainage pipe is a seepage water release event, and qualitatively identifies the seepage water flow rate based on the threshold; on this basis, a distributed fiber optic sensor array signal library adapted to the drainage hole environment is constructed, and the DAS array signal time-frequency conversion is realized using continuous wavelet transform, and an intrusion event recognition model based on LeNet convolutional neural network and random forest is constructed, and finally the type and location of the intrusion event are output. This method identifies the release of seepage water as a separate intrusion event, solving the problem of long-term existence of seepage water release events; the separation of sensors inside and outside the drainage hole eliminates the influence of intrusion events outside the drainage hole; the continuous wavelet time-frequency conversion of the sensor array signal realizes the rapid processing of multiple signals; multiple signals are used at the same time for the identification of intrusion event types, and an intrusion event recognition model based on convolutional neural network and random forest is established, which has good robustness. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 A flow chart of the present invention as a whole;
[0034] Figure 2 A schematic diagram of the distributed optical fiber deployment and monitoring of the drainage hole environment in the present invention;
[0035] Figure 3 DAS signal diagrams of different types of intrusion events in the present invention;
[0036] Figure 4 This is the acoustic signal time-frequency conversion and convolutional neural network feature extraction model in the present invention. DETAILED DESCRIPTION
[0037] All features disclosed in this specification, or steps in all methods or processes disclosed, except mutually exclusive features and / or steps, can be combined in any manner.
[0038] Any feature disclosed in this specification (including any additional claims and abstract), unless otherwise stated, may be replaced by other equivalent or alternative features having similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.
[0039] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by the terms "up", "down", "left", "right", etc. are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a predetermined direction, be constructed and operated in a predetermined direction, and therefore cannot be understood as a limitation on the present invention.
[0040] In addition, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first", "second", etc. may explicitly or implicitly include one or more of the features.
[0041] Example 1
[0042] like Figure 1 to Figure 4 As shown, the present invention provides a technical solution:
[0043] A method for identifying drainage hole environment intrusion events based on distributed optical fiber monitoring, which includes at least but not limited to the following steps:
[0044] Step 1: Deploy a distributed optical fiber monitoring system adapted to the drainage hole environment;
[0045] Step 2: Distributed optical fiber sensor position calibration;
[0046] Step 3: Qualitative identification of water flow rate of drainage pipe in the drainage hole;
[0047] Step 4: Construct an array signal library suitable for drainage hole monitoring;
[0048] Step 5: Time-frequency conversion of sensor array signal continuous wavelet transform;
[0049] Step 6: Build an intrusion event recognition model based on LeNet convolutional neural network and random forest;
[0050] Step 7: Output the type and location of the intrusion event;
[0051] In step 1, the specific steps for deploying a distributed optical fiber monitoring system adapted to the drainage hole environment are as follows:
[0052] Distributed optical fiber is laid along the slope and drainage hole. The optical fiber of the slope part passes through the PVC pipe fixed by steel nails and pipe clamps. When the inner wall of the drainage hole is flat, it is fixed with saddle clips. When the inner wall of the drainage hole is uneven, it is fixed with steel nails and saddle clips. The self-developed A-DAS distributed optical fiber modulator and demodulator is used to monitor the acoustic signal of the intrusion event. Among them, the length of each PVC pipe in the slope part is 3 to 4 meters, and the interval between pipe clamps is 1 to 2 meters; the interval between saddle clips in the drainage hole is 0.5 to 1.0 meters; the acoustic signal sampling frequency is set to 100Hz.
[0053] In step 2, the specific steps of position calibration and calibration of distributed optical fiber sensor are as follows:
[0054] The turning points of the slope and drainage hole are used as important position marking points. A small hammer is used to knock at the turning points at a certain distance to generate acoustic signals. The sensor number corresponding to the maximum amplitude of the distributed optical fiber monitoring signal during the calibration period is the knocking point, which realizes the calibration of the position of the distributed optical fiber sensor. Among them, the interval distance of the knocking points on the slope is 10 to 20 meters, and the interval distance of the knocking points in the drainage hole is about 1 meter. When determining the sensor number, it is necessary to eliminate abnormal values and calibrate the sensor position based on the optical fiber propagation distance and the knocking point position.
[0055] In step 3, the specific steps for qualitatively identifying the water flow rate of the drainage pipe in the drainage hole are as follows:
[0056] The sensor position at the drain pipe is determined by knocking, and the template matching method is used to determine whether the sensor signal is a seepage water release event, and then the seepage water flow rate is qualitatively determined based on the maximum amplitude of the signal. Among them, the template matching uses the experimental data of different indoor water flow rates as the template signal, and calculates the mutual correlation coefficient between the monitoring signal and the template signal. If the mutual correlation coefficient of the signal pair is greater than 0.9, the signal is recorded as a seepage water release event. When the amplitude of the acoustic signal is 0-1, the seepage water flow rate is small; when the amplitude is 1-2, the seepage water flow rate is medium; when the amplitude is greater than 2, the seepage water flow rate is large.
[0057] In step 4, the specific steps for constructing an array signal library suitable for drainage hole monitoring are as follows:
[0058] According to the positional relationship between the inside and outside of the drainage hole, the number of the distributed optical fiber sensor located in the drainage hole is determined, and the signals are arranged from bottom to top in the order of the sensor numbers, and the horizontal axis time value of the signal is unified. Among them, the signal length is 5s, the signal sliding window length is 1s during identification, and the signal uses a Butterworth filter to remove high-frequency noise greater than 20Hz. On this basis, the signal is augmented by using the disorder characteristics between signal channels: the data is expanded by randomly exchanging the channel order and superimposing the data. This process does not change the category label of the signal.
[0059] In step 5, the specific steps of time-frequency conversion of sensor array signal continuous wavelet transform are as follows:
[0060] Use continuous wavelet transform to convert each signal from time series into a time-frequency diagram, ensure that the time, frequency range and amplitude range of the time-frequency diagram are consistent, and then splice the time-frequency diagram into an array signal time-frequency diagram according to the order of sensors. Among them, the continuous wavelet transform uses Daubechies 5 wavelet basis function, the decomposition scale is 6 layers, the frequency range of the time-frequency diagram is set to 0~20Hz, and the amplitude is set to 0~2.
[0061] In step six, the steps for constructing an intrusion event recognition model based on LeNet convolutional neural network and random forest are as follows: using continuous wavelet transform to realize the time-frequency conversion of sensor array signals, and then constructing a LeNet convolutional neural network model to extract the time-frequency graph features of the array signal to achieve image feature dimensionality reduction.
[0062] Specifically, the LeNet convolutional neural network for intrusion event recognition is constructed with the array signal time-frequency diagram as input and the intrusion event type as output. The training set is the array signal after monitoring and data augmentation. The input layer size is set to 64*64*3, the convolution kernel size is 5×5, the pooling layer sampling kernel size is 2×2, and the output layer is the 10 extracted feature parameters. These feature parameters are used as the model parameters for random forest classification and recognition. When numbering the intrusion event type, human walking is recorded as 1, rock falling is recorded as 2, background noise is recorded as 3, and seepage water release is recorded as 4.
[0063] In step 7, the specific steps for outputting the intrusion event type and location are as follows:
[0064] A distributed optical fiber 3D layout model including slopes and drainage holes is constructed. The types and locations of drainage hole intrusion events are displayed in the 3D model, and graded alarms are given for intrusion events such as rock falls.
[0065] The hierarchical alarm rules are as follows: general intrusion, green light alarm; major incidents, yellow light alarm; important incidents, red light alarm.
[0066] Data testing shows that the accuracy of recognition of various intrusion events and background noise is around 90%, among which the training accuracy of rock fall reached 94.29% and the prediction accuracy also reached 92.00%; the training and prediction accuracy of walking intrusion events were also high, at 91.76% and 88.90% respectively. The training and prediction accuracy of seepage water release and background noise is slightly lower due to the less obvious waveform characteristics, but it also exceeds 85%. This result shows that this patent has good applicability for the identification of intrusion events in drainage hole environments.
[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring, characterized by: The following steps are involved: Step 1: Deploy a distributed fiber optic monitoring system adapted to the drainage tunnel environment; Step 2: Position calibration of distributed optical fiber sensors; Step 3: Qualitative identification of water flow rate of drainage pipe in drainage hole; Step 4: Construct an array signal library suitable for drainage hole monitoring; Step 5: Time-frequency conversion of sensor array signal continuous wavelet transform; Step 6: Build an intrusion event recognition model based on LeNet convolutional neural network and random forest; Step 7: Output the type and location of the intrusion event.
2. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step 1, the specific steps for deploying a distributed fiber optic monitoring system adapted to the drainage tunnel environment are as follows: Distributed optical fiber is laid along the slope and drainage holes. The optical fiber on the slope passes through a PVC pipe fixed with steel nails and pipe clamps. When the inner wall of the drainage hole is flat, it is fixed with saddle clips. When the inner wall of the drainage hole is uneven, it is fixed with steel nails and saddle clips. An A-DAS distributed optical fiber modulator and demodulator is used to monitor the acoustic signals of intrusion events.
3. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step 2, the specific steps of position calibration and calibration of distributed optical fiber sensors are as follows: The inflection points of the slope and drainage hole are used as important position marking points. A small hammer is used to knock on the inflection points at a certain distance to generate acoustic signals. The sensor number corresponding to the maximum amplitude of the distributed optical fiber monitoring signal during the calibration period is the knocking point, which realizes the calibration of the distributed optical fiber sensor position.
4. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step 3, the specific steps for qualitatively identifying the flow rate of water seepage from the drainage pipe in the drainage hole are as follows: The sensor position at the drain pipe is determined by tapping, and template matching is used to determine whether the sensor signal is a seepage water release event. The seepage water flow rate is then qualitatively determined based on the maximum amplitude of the signal. Template matching uses experimental data of different indoor water flow rates as template signals, and calculates the mutual correlation coefficient between the monitoring signal and the template signal. If the mutual correlation coefficient of the signal pair is greater than 0.9, the signal is recorded as a seepage water release event.
5. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step 4, the specific steps for constructing an array signal library suitable for drainage hole monitoring are as follows: Based on the positional relationship between the inside and outside of the drainage tunnel, the distributed fiber optic sensors located in the drainage tunnel were numbered. The signals were arranged from bottom to top in order of the sensor numbers, and the horizontal axis time values of the signals were unified. The signal length was 5 seconds, and the signal sliding window length during identification was 1 second. A Butterworth filter was used to remove high-frequency noise greater than 20 Hz. On this basis, the disorder characteristics between signal channels were used to augment the signals. The data was expanded by randomly exchanging the channel order and superimposing the data. This process did not change the signal category label.
6. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step five, the specific steps of time-frequency conversion of sensor array signal continuous wavelet transform are as follows: Continuous wavelet transform is used to convert each signal from a time series into a time-frequency diagram to ensure that the time, frequency range and amplitude range of the time-frequency diagram are consistent. The time-frequency diagrams are then spliced into an array signal time-frequency diagram in the order of the sensors.
7. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step six, the steps for constructing an intrusion event recognition model based on the LeNet convolutional neural network and random forest are as follows: using continuous wavelet transform to realize the time-frequency conversion of the sensor array signal, and then constructing a LeNet convolutional neural network model to extract the time-frequency graph features of the array signal to achieve image feature dimensionality reduction.
8. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 7, characterized in that: Specifically, the LeNet convolutional neural network for intrusion event recognition is constructed with the array signal time-frequency diagram as input and the intrusion event type as output. The training set is the array signal after monitoring and data augmentation. The input layer size is set to 64*64*3, the convolution kernel size is 5×5, the pooling layer sampling kernel size is 2×2, and the output layer is the 10 extracted feature parameters. These feature parameters are used as the model parameters for random forest classification and recognition.
9. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: In step 7, the specific steps for outputting the intrusion event type and location are as follows: A distributed fiber optic 3D layout model including slopes and drainage holes is constructed. The types and locations of drainage hole intrusion events are displayed in the 3D model, and graded alarms are given for intrusion events.
10. The method for identifying drainage hole environmental intrusion events based on distributed optical fiber monitoring according to claim 1, characterized in that: The hierarchical alarm rules are as follows: general intrusion, green light alarm; major incident, yellow light alarm; important incident, red light alarm.
Citation Information
Cited By
Pipeline multiphase flow state measurement and control simulation method based on distributed optical fibers and related equipment
CN121351669A
