Distributed optical fiber sensing system based on delay reservoir computing and its signal identification and positioning method

Abstract

The application discloses a distributed optical fiber sensing system based on delay reservoir computing and a signal identification and positioning method thereof. The system is composed of an input layer, a reservoir and an output layer. The input layer has two nodes, one of which inputs a non-essential reservoir excitation signal and the other of which inputs an external vibration signal; the reservoir is composed of a delay nonlinear system, and a delay optical fiber thereof is a sensing optical fiber; the output layer has two nodes, one of which outputs a vibration type and the other of which outputs a vibration position. The signal identification and positioning method comprises the following steps: before the system is started, network training is performed, and output connection weights are determined through target outputs and virtual node states of training signals; after the system is started, the output connection weights and virtual node states of a vibration signal to be measured are used to obtain an output of the reservoir calculation, i.e. a vibration signal type and a vibration position. The application realizes accurate and real-time identification and positioning of external vibrations by using the powerful learning ability, modeling ability and fast processing ability of optical reservoir calculation.

Description

TECHNICAL FIELD

[0001] The present application relates to a distributed optical fiber sensing system and a signal recognition and positioning method thereof, in particular to a distributed optical fiber sensing system and a signal recognition and positioning method thereof based on delay reservoir computing, and belongs to the field of optical fiber sensing. BACKGROUND

[0002] The distributed optical fiber sensing system uses optical fiber as a sensitive element and transmission medium, which can continuously monitor external vibration signals over a long distance, and has been rapidly developed in the fields of pipeline safety detection and perimeter defense. At present, the distributed optical fiber vibration sensing technology generally has the problems of high false alarm rate, low positioning accuracy and slow processing speed.

[0003] In recent years, the application of machine learning in distributed optical fiber vibration sensing has gradually become a research hotspot. There are a large number of studies on using neural networks or other machine learning models to identify different types of sensing event signals, and there are also reports on improving positioning accuracy. However, whether it is signal recognition or vibration positioning, it is a combination of hardware sensing system and software learning model. Due to the slow processing speed of software, the real-time performance of detection is limited. In addition, signal recognition generally requires prior extraction of signal features. Feature extraction is a time-consuming task and may be inaccurate due to noise interference, so the speed and accuracy of recognition need to be improved. More importantly, most of the studies on signal recognition do not consider the location of vibration, and the same type of vibration at different positions may be misjudged as different types. Therefore, how to quickly and correctly identify different types of external vibrations combined with the sensing position is a problem to be solved in this field.

[0004] On the other hand, as a simplified recurrent neural network, reservoir computing has received extensive attention in the academic community. Because its training process is simple, only the output connection weight needs to be trained, and the input connection weight and internal connection weight are randomly generated and fixed, it is easy to implement with hardware. In recent years, optical or optoelectronic reservoir computing implemented with optoelectronic and all-optical devices, especially delay nonlinear systems, has important significance for faster and lower power consumption information processing, and has been rapidly developed. SUMMARY

[0005] The present application aims to overcome the deficiencies of the prior art, and provides a distributed optical fiber sensing system based on delay reservoir computing and a signal identification and positioning method thereof. The present application combines optical reservoir computing and distributed optical fiber sensing system, and uses the sensitivity response of reservoir computing to the external environment to detect external vibration signals. The hardware-based delay nonlinear system is used to realize optical reservoir computing, and the reservoir pool is used as a sensor to realize distributed optical fiber vibration sensing. The machine learning method is used to identify and locate the external vibration. The powerful learning and modeling capabilities of reservoir computing are fully utilized to improve the accuracy of detection, and the advantages of fast optical processing and simple reservoir computing test process are used to improve the real-time performance of detection.

[0006] To achieve the above-mentioned purposes, the present application adopts the following technical solutions:

[0007] A distributed optical fiber sensing system based on delay reservoir computing, which is composed of an input layer, a reservoir pool and an output layer. The input layer has two nodes, one of which inputs a reservoir pool excitation signal, and the other inputs an external vibration signal; the reservoir pool is composed of a delay nonlinear system, and the delay optical fiber is the sensing optical fiber; the output layer has two nodes, one of which outputs the vibration type, and the other outputs the vibration position.

[0008] Preferably, the node of the input layer for inputting the reservoir pool excitation signal is not necessary, i.e. it is needed or not.

[0009] Preferably, the delay nonlinear system is in the form of all-optical delay or optoelectronic delay, and the specific structure requires that it has the properties of reservoir pool and different responses to different types and positions of external vibrations.

[0010] A signal identification and positioning method for a distributed optical fiber sensing system based on delay reservoir computing, which uses the distributed optical fiber sensing system based on delay reservoir computing of the present application, and includes the following steps:

[0011] Step 1: Before the system is started, the delay reservoir computing is trained. An optimal input signal is input into the optical reservoir pool after being masked to excite rich internal dynamics; various vibration type signals are applied to different positions of the sensing optical fiber, the response signals of the reservoir pool are collected, and the corresponding virtual node states are obtained; the network is trained with the type and position of the external vibration signal as the target output, so that the output of the reservoir computing approximates the target output, and the appropriate output connection weight is determined.

[0012] Step 2: after the system is started, the input signal and the mask signal are used to stimulate the internal dynamics of the reservoir, the response signal of the reservoir is collected, the virtual node state of the vibration signal to be measured is obtained, and the output connection weight obtained in the training stage is combined to obtain the output of the delay reservoir calculation, i.e. the type and position of the vibration signal.

[0013] Preferably, the input signal and the mask signal for stimulating the internal dynamics of the reservoir are not necessarily input, i.e. input or not input as needed.

[0014] Preferably, the positioning problem is trained according to the regression task or according to the classification task.

[0015] Preferably, the output connection weight and the output result of the reservoir calculation are obtained online or offline, and are obtained by electricity or by light.

[0016] Compared with the prior art, the present application has the following obvious and substantial characteristics and advantages:

[0017] 1. The present application uses the reservoir calculation method to train and test the system to realize the distributed optical fiber vibration sensing function, without the need to establish an accurate mechanism model or to demodulate the signal or write the relationship of the vibration position.

[0018] 2. The reservoir calculation is realized by using an optical or optoelectronic delay nonlinear system, and compared with the software implemented reservoir calculation, the virtual node state is obtained quickly, so that the signal recognition and positioning process is also fast.

[0019] 3. Through the training of the delay reservoir calculation, the signal recognition and positioning model can be established at the same time, and the signal recognition and positioning functions are solved at one time.

[0020] 4. The whole signal recognition and positioning process is very simple, because the reservoir calculation training process only needs to calculate the output weight, and the test process only needs to multiply the output connection weight with the virtual node state of the test signal.

[0021] 5. The reservoir calculation has strong learning ability and modeling ability, so that the signal features do not need to be extracted, and the recognition and positioning results are not easily affected by noise, and are fast and accurate. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 It is a schematic diagram of the system structure of the present application.

[0023] Figure 2 It is a schematic diagram of the distributed optical fiber sensing system structure of the preferred embodiment of the present application.

[0024] Figure 3Layout of different vibration points on sensing fiber in preferred embodiment of the present application.

[0025] Figure 4 Time-domain waveforms of system output signals before and after filtering when external vibration types are sine and pulse signals in preferred embodiment of the present application.

[0026] Figure 5 Time-domain waveforms of system output signals at five sensing positions when external vibration types are sine and pulse signals in preferred embodiment of the present application.

[0027] Figure 6 Test results obtained when the number of virtual nodes is optimally set to 30 in preferred embodiment of the present application. DETAILED DESCRIPTION

[0028] The above scheme is further described below in combination with specific examples, and preferred embodiments of the present application are described in detail as follows:

[0029] Embodiment One:

[0030] In this embodiment, referring to Figure 1 A distributed fiber sensing system based on reservoir computing is composed of an input layer, a reservoir and an output layer. The input layer has two nodes, one of which inputs a reservoir excitation signal and the other inputs an external vibration signal; the reservoir is composed of a delay nonlinear system, and the delay fiber thereof is a sensing fiber; and the output layer has two nodes, one of which outputs a vibration type and the other outputs a vibration position.

[0031] In this embodiment, the fiber delay nonlinear system is used as both a reservoir and a sensor, and the response of reservoir computing is used to detect an external vibration signal; through the idea of machine learning, i.e. training and testing of reservoir computing, the type identification and position determination of the external vibration signal are completed. No accurate mechanism model needs to be established, and no signal demodulation or vibration position relationship formula needs to be written.

[0032] Embodiment Two:

[0033] This embodiment is basically the same as Embodiment One, and the particularity lies in that:

[0034] In this embodiment, the node of the input layer for inputting the reservoir excitation signal is not necessary, i.e. it is present or absent according to needs.

[0035] In this embodiment, the delay nonlinear system is in an all-optical delay form or an optoelectronic delay form, and the specific structure requires that it has both the attribute of a reservoir and different responses to different types and positions of external vibrations.

[0036] Compared with the software implementation of the reserve pool calculation, the virtual node state acquisition speed of the embodiment is fast, and therefore the signal recognition and positioning process is also fast.

[0037] Embodiment three:

[0038] The embodiment is basically the same as the above-mentioned embodiments, and the particularity lies in that:

[0039] In the embodiment, referring to Figure 2 , a delay mutual injection semiconductor laser system is used to constitute a reserve pool and a distributed optical fiber sensor. Laser 1 and laser 2 are both semiconductor lasers without built-in isolators. Two polarization controllers are used to adjust the polarization states of the light injected into the two lasers. A section of optical fiber connecting the two lasers serves as a sensing optical fiber, and a phase modulator is inserted into the sensing optical fiber to simulate the effect of external vibration on the sensing optical fiber, that is, as the only input node for inputting external vibration signals. In order to compensate for the loss of light on the mutual injection channel and ensure sufficient mutual injection intensity of the two lasers, an optical amplifier is placed on the mutual injection channel. Part of the optical signal is output through a 1x2 coupler, and converted into an electrical signal through a photodetector. The electrical signal is collected and processed on a computer to obtain the virtual node state. The mutual injection semiconductor laser reserve pool calculation is trained with vibration type and vibration position as output targets, that is, the output nodes are two, one outputs the vibration type and the other outputs the vibration position.

[0040] Laser 1 is a DFB tunable semiconductor laser DFB200903 from Tianjin Junfengguangli Technology Co., Ltd., with an output power range of 0-13 mW and a central wavelength tuning range of 1548.92-1551.32 nm. Laser 2 is a DFB semiconductor laser module DFB200904 from the same company, with an output power of 12.58 mW and a central wavelength of 1550.004 nm. Both polarization controllers use fiber squeezers (PLC-001) from General Photonics. The phase modulator is a LiNbO3 electro-optic phase modulator (10053) from Covega, with the driving signal provided by a signal generator (33250A) from Agilent. The optical amplifier is a semiconductor optical amplifier BOA1004S from Thorlabs. All optical fibers are G.652 standard single-mode fibers, of which the sensing optical fiber is 600 m long. The coupling ratio of the 1x2 coupler is 50:50. The photodetector is a PIN-TIA detector produced by Shenzhen Feitong Company. The electrical signal is collected using a PicoScope 5203 digital oscilloscope from PICO, and the signal processing is realized using MATLAB software on a general microcomputer with an Intel(R) Core(TM) i5-8250U CPU @ 1.60 GHz, 16 GB RAM, and Windows 10 system.

[0041] Embodiment Four

[0042] In this embodiment, a signal identification and location method of a distributed optical fiber sensing system based on delay reservoir computing is provided. The method comprises the following steps:

[0043] Step 1: Before the system is enabled, the delay reservoir computing is trained. A preferred input signal is input into the optical reservoir after being masked to excite rich internal dynamics; various vibration type signals are applied at different positions of the sensing optical fiber, the response signals of the reservoir are collected, and the corresponding virtual node states are obtained; the network is trained with the type and position of the external vibration signal as the target output, so that the output of the reservoir computing approximates the target output, and the appropriate output connection weight is determined.

[0044] Step 2: After the system is enabled, the same input signal and mask signal as in the training stage are used to excite the internal dynamics of the reservoir, the response signals of the reservoir are collected, the virtual node states of the vibration signal to be measured are obtained, and the output connection weight obtained in the training stage is combined to obtain the output of the delay reservoir computing, i.e. the type and position of the vibration signal.

[0045] In this embodiment, the signal identification and location model can be established through the training of the delay reservoir computing, and the signal identification and location functions can be solved at one time; the training process only needs to calculate the output weight, and the testing process only needs to multiply the output connection weight with the virtual node state of the test signal, so the whole signal identification and location processing process is very simple; the reservoir computing has strong learning and modeling capabilities, so it does not need to extract signal features, and the identification and location results are not easily affected by noise, which is fast and accurate.

[0046] Embodiment Five

[0047] This embodiment is basically the same as the above-mentioned embodiments, and the special feature is that:

[0048] In this embodiment, the input signal and the mask signal used to excite the internal dynamics of the reservoir are not necessarily input, i.e. they are input or not input according to the need.

[0049] In this embodiment, the delay reservoir computing is trained according to the regression task or the classification task.

[0050] In this embodiment, the output connection weight and the output result of the reservoir computing are obtained online or offline; they are obtained by electricity or by light.

[0051] Embodiment Six

[0052] The embodiment is basically the same as the above-described embodiment, and the particularity lies in that:

[0053] In the embodiment, the signal recognition and positioning are performed according to the following steps.

[0054] Step 1, before the system is enabled, the delay reservoir pool is calculated and trained. The positioning is performed by using the classification model, and the sensing optical fiber is divided into 5 segments, and the distances from the midpoint of each segment to the laser 1 are 0 m, 120 m, 240 m, 360 m and 510 m respectively, as shown in Figure 3 The phase modulator is connected at these positions, and the sensing optical fiber at one of the positions is simulated to be affected by the external vibration signal. The phase modulator is driven by a sine and pulse signal respectively, and the response signal of the reservoir pool, that is, the output signal of the mutual injection semiconductor laser system, is collected. In order to obtain better recognition and positioning effect, the collected signal is subjected to Butterworth low-pass filtering processing to eliminate the influence of high-frequency noise. Figure 4 The time-domain waveforms of the response signals of the sine and pulse signals before and after filtering are given respectively taking position 1 as an example. The effective segment of the response signal containing the external vibration signal with a length of 3 times the transmission time of the sensing optical fiber is intercepted and subjected to normalization processing. A total of 1400 signal segments are obtained. The signal waveforms output by the system under the conditions of vibration signals being sine and pulse respectively and 5 sensing positions are shown in Figure 5 .

[0055] Supposing that the number of virtual nodes is N, 3N data are extracted from each signal segment as the virtual node state of the reservoir pool, and the virtual node states respectively corresponding to the 1400 signal segments are combined together as the overall virtual node state data set, and the signal type and position are taken as the corresponding output label. 80% of the virtual node state data set, corresponding to 1120 signal segments, is used for training. In the training process, the 1120 target outputs and virtual node states are known, and the output weight of the reservoir pool calculation is obtained by pseudo-inverse algorithm.

[0056] Step 2, after the system is enabled, the test process of the reservoir pool calculation is entered. The remaining 20% of the virtual node state data set, corresponding to 280 signal segments, is used for testing. The test output is obtained by multiplying the output connection weight obtained in the training stage with the 280 virtual node states, and the test output is subjected to decision processing, so as to obtain the type and position of the signal.

[0057] We found that the change of the number of virtual nodes N plays an important role in signal recognition and positioning during the training and testing process. With the increase of N, the classification accuracy of the training result gradually increases until it becomes 100%, i.e. there is no classification error, but the classification accuracy of the test result has a trend of first increasing and then decreasing. When the number of virtual nodes is equal to 30, the classification accuracy of the test result is the largest, and when the number of virtual nodes is greater than 30, the phenomenon of overfitting occurs. Therefore, we select 30 virtual nodes. The test result at this time is shown in Table 1. Figure 6 The consistency between the test output label obtained after the decision of the vibration type test output and the target output label is very high, only a few test output labels that should be "1" are misjudged as "2", and the classification accuracy of the vibration type, i.e. the signal recognition rate, can reach 99.64%; the consistency between the test output label obtained after the decision of the vibration position test output and the target output label is also good, only a small amount of test output labels are misjudged as the adjacent target output labels, and the classification accuracy of the vibration position, i.e. the positioning accuracy, can reach 91.43%.

[0058] The above describes the embodiments of the present application in combination with the drawings, but the present application is not limited to the above embodiments, and various changes can be made according to the purpose of the invention of the present application. Any change, modification, replacement, combination or simplification made according to the spirit and principles of the technical solution of the present application shall be an equivalent replacement manner, as long as it meets the purpose of the present application, as long as it does not deviate from the technical principles and inventive concepts of the present application, and it belongs to the protection scope of the present application.

Claims

1. A distributed fiber optic sensing system based on delay reservoir calculation, characterized in that: It consists of an input layer, a reservoir, and an output layer. The input layer has two nodes: one for the excitation signal to the reservoir and the other for the external vibration signal. The reservoir is constructed using a delayed nonlinear system, with the delay fiber serving as the sensing fiber. The output layer has two nodes: one for the vibration type and the other for the vibration position. A delayed mutual-injection semiconductor laser system is used to construct the reservoir and the distributed fiber optic sensor. Both lasers are semiconductor lasers without built-in isolators. Two polarization controllers are used to adjust the polarization state of the light injected into the two lasers. A section of fiber connecting the two lasers serves as the sensing fiber, and a phase modulator is inserted into the sensing fiber to simulate the effect of external vibration on it. To compensate for light loss in the mutual injection channel and ensure sufficient mutual injection intensity between the two lasers, an optical amplifier is placed in the mutual injection channel. A portion of the optical signal is output through a 1×2 coupler and converted into an electrical signal by a photodetector. Electrical signals are collected and processed on a computer to obtain the virtual node state; the calculation of the reservoir of the mutually injected semiconductor laser is trained with vibration type and vibration position as the output target.

2. The distributed fiber optic sensing system based on delay reservoir calculation according to claim 1, characterized in that: The nodes in the input layer used to input the excitation signal to the storage pool are not mandatory; they may or may not be present depending on the needs.

3. The distributed fiber optic sensing system based on delay reservoir calculation according to claim 1, characterized in that: The aforementioned delay nonlinear system is either an all-optical delay system or an optoelectronic delay system. In terms of specific structure, it is required to have both the properties of a reservoir and different responses to different types of external vibrations at different locations.

4. A signal identification and localization method for a distributed fiber optic sensing system based on delay reservoir calculation, utilizing the distributed fiber optic sensing system based on delay reservoir calculation as described in claim 1, characterized in that: Includes the following steps: Step 1: Before the system is activated, the delay reservoir calculation is trained; an input signal is input into the optical reservoir after being masked to excite rich internal dynamics; various vibration types of signals are applied at different locations of the sensing fiber to collect the response signals of the reservoir and obtain the corresponding virtual node states. The network is trained using the type and location of external vibration signals as the target output, so that the output calculated by the reservoir approximates the target output and the appropriate output connection weights are determined. Step 2: After the system is started, the same input signal and mask signal as in the training phase are used to excite the internal dynamics of the reservoir, collect the response signal of the reservoir, obtain the virtual node state of the vibration signal to be measured, and combine it with the output connection weight obtained in the training phase to obtain the output calculated by the delayed reservoir, that is, the type and location of the vibration signal.

5. The signal identification and positioning method for a distributed fiber optic sensing system based on delay reservoir calculation according to claim 4, characterized in that: The input signals and mask signals used to stimulate the dynamics inside the storage pool are not mandatory; that is, they can be input or not input as needed.

6. The signal identification and positioning method for a distributed fiber optic sensing system based on delay reservoir calculation according to claim 4, characterized in that: The localization problem is trained on the delay reserve pool computation according to either regression or classification tasks.

7. The signal identification and positioning method for a distributed fiber optic sensing system based on delay reservoir calculation according to claim 4, characterized in that: The output connection weight and the output result calculated by the storage pool can be obtained online or offline; or obtained by electricity or by light.

Images