Linear Sagnac interference distributed optical fiber sensing positioning method based on time delay light reserve pool calculation

Abstract

The invention discloses a linear Sagnac interference distributed optical fiber sensing positioning method based on time delay light reserve pool calculation, which comprises the following steps of: randomly selecting part of sensing positions by taking a positioning resolution as a minimum interval, and respectively collecting interference signals caused by external disturbance at the positions; the collected signals are preprocessed, the occurrence time of four waveform changes in the signals is extracted to form feature vectors, and a training set and a test set are formed; according to the length of forgotten data, the starting data of the test set is repeated once, then the training set and the test set are subjected to mask processing, the time delay light reserve pool is input in sequence, corresponding virtual node states are collected, the time delay light reserve pool is trained and tested, a regression model is calculated, and the optimal output connection weight and the test output are obtained respectively; and taking an average value of four adjacent test output values as a disturbance position prediction result of one corresponding feature vector. According to the invention, any sensing position can be quickly and accurately positioned by using part of sensing position data, and the method has the potential of positioning in an optical domain.

Description

Technical Field

[0001] The present invention relates to a disturbance positioning method for distributed optical fiber sensing, in particular to a linear Sagnac interference distributed optical fiber sensing positioning method based on time-delay optical reserve pool calculation. Background Art

[0002] Distributed fiber optic sensing technology is a key industry technology that has attracted considerable attention, offering wide-range and highly sensitive monitoring capabilities. Linear Sagnac interferometry distributed fiber optic sensing systems have become a research hotspot in recent years due to their high sensitivity, ease of deployment, low light source coherence requirements, and insensitivity to slow environmental changes. However, their positioning accuracy and processing speed still require improvement.

[0003] In recent years, the application of machine learning algorithms has played an important role in solving positioning problems. Using classification models for positioning requires dividing the sensing fiber into multiple segments, collecting interference signals caused by external disturbances on each segment, and then identifying the fiber segment corresponding to the interference signal caused by the disturbance to achieve positioning. The positioning resolution is limited by the length of the fiber segment and is generally not high. Otherwise, data collection will be difficult. Using regression models for positioning, however, only the interference signals caused by disturbances at some sensing locations need to be collected to accurately predict the location of any disturbance on the sensing fiber. This requires less data collection, is easy to implement, and maintains the continuous monitoring advantage of distributed fiber optic sensing. However, existing positioning methods based on machine learning regression models are mostly deep learning models, which face problems such as numerous parameters and difficulty in training. Summary of the Invention

[0004] In order to solve the problems of the existing technology, the present invention provides a linear Sagnac interferometer distributed fiber optic sensing positioning method based on time-delay optical reserve pool calculation, which reduces the difficulty of model training, quickly and accurately predicts the position of any disturbance with a small amount of data, and improves positioning efficiency.

[0005] In order to achieve the above invention purpose, the present invention adopts the following technical solutions:

[0006] A linear Sagnac interferometry distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation includes the following steps:

[0007] 1) Randomly select some disturbance positions on the sensing fiber, with the minimum interval between the disturbance positions being the required positioning resolution, and collect the interference signals caused by the external disturbance signals at these positions;

[0008] 2) Preprocessing the collected interference signal, including filtering out the signal part with smaller amplitude and performing numerical scaling operations on the time and amplitude of waveform changes in the signal, in preparation for extracting data features;

[0009] 3) Extract the occurrence time of each waveform change of the interference signal as the signal feature, calculate the time average value within the duration of each waveform change in the signal, and use it as the occurrence time of each waveform change. Each disturbance position corresponds to the occurrence time of four waveform changes, that is, the eigenvalues of the four time average values constitute a eigenvector, which is arranged in the order of the time when the four waveform changes occur;

[0010] 4) The extracted feature vector data is divided into a training set and a test set. The target output of the training set is the result of sampling and holding each corresponding actual disturbance position into four values to match the length of the training data. The training set and its target output are used to determine the output connection weights of the time-delay optical reservoir calculation regression model. The test set is used to locate the disturbance of the interference signal.

[0011] 5) After masking the training set data, input it into the delay optical reserve pool, collect the corresponding virtual node status, train the delay optical reserve pool calculation regression model, and obtain the optimal output connection weight;

[0012] 6) Based on the selected number of forgotten points P, the first P data of the test set are repeated once to ensure that after forgetting the P point data, there are still enough eigenvalues to locate the position to be tested. The test set data is then masked and input into the delayed optical reserve pool. The corresponding virtual node status is collected, and the delayed optical reserve pool calculation regression model is tested to obtain the test output of each eigenvalue. The average of the 4 adjacent test output values is used as the disturbance position prediction result of the corresponding eigenvector to achieve disturbance positioning.

[0013] As a preferred technical solution of the present invention, the structure of the delayed optical reserve pool in steps 5)-6) is an all-optical form or a photoelectric form.

[0014] As another preferred technical solution of the present invention, the structure of the delay optical reserve pool in steps 5)-6) is in the form of a single physical node or multiple physical nodes.

[0015] As another preferred technical solution of the present invention, the structure of the delay optical reserve pool in steps 5)-6) is a single-delay mode or a multi-delay mode.

[0016] Preferably, the mask processing operation in steps 5)-6) is: after each eigenvalue is expanded into M, it is multiplied by a mask sequence of length M, where M is the number of virtual nodes.

[0017] Preferably, in steps 5) to 6), the operation of inputting data into the delay reserve pool is performed in the form of optical injection or electrical injection.

[0018] Preferably, in step 5), the training of the time-delay optical reservoir calculation regression model adopts traditional pseudo-inverse operation, ridge regression, or other training methods.

[0019] Preferably, in step 6), the test operation of the delayed optical reserve pool calculation regression model is: performing weighted summation on the virtual node states of the test data according to the optimal output connection weight obtained in the training process to obtain the test output.

[0020] Compared with the prior art, the present invention has the following obvious outstanding substantial features and significant advantages:

[0021] 1. Compared with the traditional disturbance positioning method, the present invention does not require signal demodulation and has low signal processing complexity;

[0022] 2. Compared with the positioning method based on the classification model, the present invention only needs to collect the interference signals of some disturbance positions to predict the position of any disturbance on the sensing fiber, which is easy to implement and highly efficient.

[0023] 3. Compared with the positioning method based on deep learning regression model, the training and testing process of the reservoir calculation regression model adopted by the present invention is simple and the positioning process is fast;

[0024] 4. Compared with the traditional reserve pool calculation regression model implemented by software, the time-delay optical reserve pool calculation regression model adopted by the present invention can take advantage of the fast optical information processing speed to further accelerate the positioning process, and has low power consumption. If optical mask processing and online all-optical testing can be realized in the future, the present invention is expected to achieve disturbance positioning in the optical domain.

[0025] 5. The present invention extracts the occurrence times of four waveform changes of the interference signal as feature vectors, replaces the interference signal, performs mask processing, and then inputs it into the delay optical reserve pool. Instead of directly inputting the interference signal after mask processing as conventionally, the amount of data input is greatly reduced, the signal processing speed is improved, and the difficulty of implementing the delay optical reserve pool is further reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of the positioning method of the present invention.

[0027] Figure 2 Schematic diagram of the system structure used in the embodiment.

[0028] Figure 3 This is the time domain waveform of the interference signal after preprocessing when the leakage position is 55 meters.

[0029] Figure 4 This is the time domain waveform of the interference signal after preprocessing when the leakage position is 655 meters.

[0030] Figure 5 The semiconductor laser-based delay reservoir structure and data light injection form used in the embodiment.

[0031] Figure 6 200 eigenvalues obtained in the embodiment are tested.

[0032] Figure 7 These are the positioning results of 50 test positions obtained in the embodiment. DETAILED DESCRIPTION

[0033] The above solution is further described below with reference to specific implementation examples. The preferred embodiments of the present invention are described in detail as follows:

[0034] Example 1:

[0035] In this embodiment, see Figure 1 A linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation includes the following steps:

[0036] 1) Randomly select some disturbance positions on the sensing fiber, with the minimum interval between the disturbance positions being the required positioning resolution, and collect the interference signals caused by the external disturbance signals at these positions;

[0037] 2) Preprocessing the collected interference signal, including filtering out the signal part with smaller amplitude and performing numerical scaling operations on the time and amplitude of waveform changes in the signal, in preparation for extracting data features;

[0038] 3) Extract the occurrence time of each waveform change of the interference signal as the signal feature, calculate the time average value within the duration of each waveform change in the signal, and use it as the occurrence time of each waveform change. Each disturbance position corresponds to the occurrence time of four waveform changes, that is, the eigenvalues of the four time average values constitute a eigenvector, which is arranged in the order of the time when the four waveform changes occur;

[0039] 4) The extracted feature vector data is divided into a training set and a test set. The target output of the training set is the result of sampling and holding each corresponding actual disturbance position into four values to match the length of the training data. The training set and its target output are used to determine the output connection weights of the time-delay optical reservoir calculation regression model. The test set is used to locate the disturbance of the interference signal.

[0040] 5) After masking the training set data, input it into the delay optical reserve pool, collect the corresponding virtual node status, train the delay optical reserve pool calculation regression model, and obtain the optimal output connection weight;

[0041] 6) Based on the selected number of forgotten points P, the first P data of the test set are repeated once to ensure that after forgetting the P point data, there are still enough eigenvalues to locate the position to be tested. The test set data is then masked and input into the delayed optical reserve pool. The corresponding virtual node status is collected, and the delayed optical reserve pool calculation regression model is tested to obtain the test output of each eigenvalue. The average of the 4 adjacent test output values is used as the disturbance position prediction result of the corresponding eigenvector to achieve disturbance positioning.

[0042] This embodiment is based on a linear Sagnac interferometric distributed fiber optic sensing positioning method calculated using a time-delayed optical reservoir. This method does not require signal demodulation and has low signal processing complexity. It only requires collecting interference signals at some disturbance positions to predict the position of any disturbance on the sensing fiber, making it easy to implement and highly efficient. The training and testing processes for the reservoir calculation are simple, and the positioning process is fast. The time-delayed optical reservoir calculation regression model can utilize the advantage of fast optical information processing to further accelerate the positioning process, and it also has low power consumption and the potential to achieve positioning in the optical domain.

[0043] Example 2:

[0044] This embodiment is basically the same as the first embodiment, with the following special features:

[0045] In this embodiment, the structure of the time-delay optical reserve pool in steps 5) to 6) is an all-optical form or a photoelectric form.

[0046] Alternatively, the structure of the delay optical reserve pool in steps 5) to 6) is in the form of a single physical node or multiple physical nodes.

[0047] Alternatively, the structure of the delay optical reserve pool in steps 5) to 6) is a single-delay mode or a multi-delay mode.

[0048] Alternatively, the mask processing operation in steps 5)-6) is: after each eigenvalue is expanded into M, it is multiplied by a mask sequence of length M, where M is the number of virtual nodes.

[0049] Alternatively, the operation of inputting data into the delay reserve pool in steps 5)-6) is performed in the form of optical injection or electrical injection.

[0050] Alternatively, in step 5), the training of the time-delay optical reservoir calculation regression model adopts traditional pseudo-inverse operation, ridge regression, or other training methods.

[0051] Alternatively, in step 6), the test operation of the delayed optical reserve pool calculation regression model is: performing weighted summation on the virtual node states of the test data according to the optimal output connection weight obtained in the training process to obtain the test output.

[0052] The present invention extracts the occurrence times of four waveform changes of the interference signal as feature vectors, replaces the interference signal, performs mask processing and then inputs it into the delay optical reserve pool. Instead of directly inputting the interference signal after mask processing as conventionally, the amount of data input is greatly reduced, the signal processing speed is improved, and the difficulty of implementing the delay optical reserve pool is further reduced.

[0053] Example 3:

[0054] This embodiment is basically the same as the above embodiment, with the following special features:

[0055] In this example, a linear Sagnac interferometer distributed fiber-optic sensing system is used for pipeline leak monitoring, where the external disturbance signal is the pipeline leakage signal. This sensing system is simulated using OptiSystem software to verify the feasibility of this example's linear Sagnac interferometer distributed fiber-optic sensing positioning method based on time-delay optical reservoir calculation.

[0056] like Figure 2 As shown in the figure, the simulated linear Sagnac interferometer distributed fiber optic sensing system includes a light source, coupler 1, a delay fiber, coupler 2, sensing fiber 1, a phase modulator, sensing fiber 2, a reflector, and a photodetector. Coupler 1 and coupler 2 are both 3dB fiber couplers, and the delay fiber is 2 km long. The sensing fiber consists of sensing fiber 1 and sensing fiber 2, with a total length of 1 km. The leakage signal is simulated using a pulse signal, and the effect of the leakage signal on the sensing fiber is simulated by inserting a phase modulator between sensing fiber 1 and sensing fiber 2. While maintaining the total sensing fiber length unchanged, the position where the leakage signal acts on the sensing fiber is adjusted by changing the length of sensing fiber 1 and sensing fiber 2 and the corresponding delay time.

[0057] According to Figure 1 The process shown is to locate the leak.

[0058] Assuming the required positioning resolution is 2m, 450 leakage locations are randomly selected with a minimum interval of 2m, and the interference signals caused by the leakage signals at these 450 locations are collected respectively.

[0059] The collected interference signals are pre-processed, including filtering out the signal parts with smaller amplitudes and performing appropriate numerical scaling operations on the time and amplitude of the waveform changes in the signal, in preparation for extracting data features. The interference signals at different leakage locations after pre-processing are as follows: Figure 3 、 4 shown.

[0060] The occurrence time of each waveform change of the interference signal is extracted as the feature of the signal. The time average value within the duration of each waveform change in the signal is calculated as the occurrence time of each waveform change. Each leakage position corresponds to the occurrence time of four waveform changes, that is, the eigenvalues of the four time average values constitute a feature vector, which is arranged in the order of the time when the four waveform changes occur.

[0061] The extracted feature vector data are divided into a training set and a test set. The training set contains feature vectors of 400 leakage locations, each of which includes 4 eigenvalues. Therefore, the training set data length is 400×4=1600. Its target output is the result of sampling and maintaining each corresponding actual leakage location into 4 values, that is, the target output length is also 400×4=1600, which is consistent with the training data length and is used to determine the output connection weights of the time-delay optical reserve pool calculation regression model; the test set selects feature vectors of another 50 leakage locations, and the test set data length is 50×4=200 for leakage location of interference signals.

[0062] In this embodiment, a binary random mask signal is used to mask the training set and test set data. The mask sequence length is 50, corresponding to the number of virtual nodes M being 50. After relevant tests, it was finally determined that the characteristic value should be uniformly adjusted to 10 before injecting the delayed optical reserve pool. 0 For example, when the leak location is 155m, the adjusted four eigenvalues are 0.428855, 0.5877, 1.427267, and 1.583568.

[0063] The delay optical storage pool uses a delay storage pool based on semiconductor lasers, and its structure is as follows Figure 5 As shown in the figure, it consists of a semiconductor laser called a response laser as a nonlinear node and its fiber delayed feedback loop, where θ represents the virtual node interval, x i Indicates the state of the i-th virtual node, and 1, 2, 3, 4...M in the figure represent M virtual nodes. After the input signal is masked, it is modulated by a phase modulator to the output light of another semiconductor laser called the driving laser, and is input into the delayed light reserve pool in the form of light injection.

[0064] The following dynamic equations of the optically injected optically feedback semiconductor laser are solved by the Simulink platform of Matlab software, instead of building an actual optical system, to realize a semiconductor laser-based delay reserve pool with input signals in the form of optical injection.

[0065]

[0066] Where E represents the complex amplitude of the slowly varying electric field in response to the laser, N represents the carrier concentration, t represents time, and j represents the imaginary unit. d represents the complex amplitude of the slowly varying electric field driving the laser. I d is the output intensity of the driving laser, which is 5.65×10 20 γ is the input scaling factor, and s(t) is the masked input signal. The meanings and values of other parameters in the equation are shown in Table 1.

[0067] Table 1 Description of relevant parameters of response laser

[0068]

[0069] The masked training set data is driven by phase modulation of the output light of the laser and input into the delay reserve pool based on the semiconductor laser in the form of light injection, which corresponds to the third term on the right side of equation (1). The output power of the response laser is calculated from the complex amplitude of the slowly varying electric field of the response laser, and the output power is sampled according to the virtual node interval θ = 40ps, that is, the virtual node state of the training set data is collected and arranged into a 1600×50 state matrix X according to the length of the training data and the number of virtual nodes. In order to remove the influence of the initial state, it is necessary to forget part of the data at the beginning, and the length of the forgotten data is generally selected to be twice the number of virtual nodes. In this embodiment, the first 108 data are selected to be forgotten, which is 8 more than twice the number of virtual nodes to ensure that the influence of the initial state is removed. Therefore, the size of the state matrix X of the training data finally obtained is 1492×50. The training of the delay light reserve pool calculation regression model is carried out according to the pseudo-inverse algorithm to obtain the optimal output connection weight matrix:

[0070] W out =pinv(X)·Y (3)

[0071] Where W out =[w1,w2,…w i ,…,w 50 ] T , T represents transpose, w i represents the output connection weight of the i-th virtual node, pinv represents the pseudo-inverse operation, and Y is the target output of the training data. In order to match the size of the state matrix X of the training data, the data of Y also forgets the first 108 data. The size of Y actually involved in training is 1492×1.

[0072] According to the selected number of forgotten points 108, the first 108 data of the test set are repeated once to ensure that after forgetting 108 data points, there are still enough eigenvalues to locate the 50 positions to be tested. Then the test set data is subjected to the same mask processing and input into the semiconductor laser-based delay reserve pool in the same light injection form. The virtual node state is collected in the same way, and the delay light reserve pool calculation regression model is tested to obtain the test output.

[0073]

[0074] Figure 6 The test output of 200 eigenvalues corresponding to 50 leak locations between 110 and 600 meters (with a resolution of 10 meters) is presented. It can be seen that the predicted location of each eigenvalue is highly consistent with the actual location, with only a few eigenvalues showing significant discrepancies. The average of four adjacent test output values is calculated and used as the leak location prediction result for the corresponding eigenvector, thus achieving leak location. Figure 7 The figure shows the predicted results of these 50 leakage locations. The predicted locations almost completely overlap with the actual locations, indicating that the prediction results are relatively accurate, with an average absolute error of about 0.69m.

[0075] In summary, the above embodiment is based on a linear Sagnac interferometric distributed fiber optic sensing positioning method calculated using a time-delayed optical reservoir. With positioning resolution as the minimum interval, some sensor locations are randomly selected and interference signals caused by external disturbances at these locations are collected. The collected signals are preprocessed to extract the occurrence times of four waveform changes in the signal to form feature vectors, which are then used to form training and test sets. Based on the length of the forgotten data, the data at the beginning of the test set is repeated once. The training and test sets are then masked and then input into the time-delayed optical reservoir to collect the corresponding virtual node states. The time-delayed optical reservoir is trained and tested to calculate regression models, respectively obtaining the optimal output connection weight and test output. The average of the four adjacent test output values is used as the disturbance position prediction result for the corresponding feature vector. This embodiment can use partial sensor position data to quickly and accurately locate any sensor location and has the potential for positioning in the optical domain.

[0076] The above describes the embodiments of the present invention in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made according to the purpose of the invention. Any changes, modifications, substitutions, combinations or simplifications made according to the spirit and principles of the technical solution of the present invention should be equivalent replacement methods. As long as they comply with the purpose of the invention and do not deviate from the technical principles and inventive concepts of the present invention, they belong to the scope of protection of the present invention.

Claims

1. A linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation, characterized in that: The steps include: 1) Randomly select some disturbance positions on the sensing fiber, with the minimum interval between the disturbance positions being the required positioning resolution, and collect the interference signals caused by the external disturbance signals at these positions; 2) Preprocessing the collected interference signal, including filtering out the signal part with smaller amplitude and performing numerical scaling operations on the time and amplitude of waveform changes in the signal, in preparation for extracting data features; 3) Extract the occurrence time of each waveform change of the interference signal as the signal feature, calculate the time average value within the duration of each waveform change in the signal, and use it as the occurrence time of each waveform change. Each disturbance position corresponds to the occurrence time of four waveform changes, that is, the eigenvalues of the four time average values constitute a eigenvector, which is arranged in the order of the time when the four waveform changes occur; 4) The extracted feature vector data is divided into a training set and a test set. The target output of the training set is the result of sampling and holding each corresponding actual disturbance position into four values to match the length of the training data. The training set and its target output are used to determine the output connection weights of the time-delay optical reservoir calculation regression model. The test set is used to locate the disturbance of the interference signal. 5) After masking the training set data, input it into the delay optical reserve pool, collect the corresponding virtual node status, train the delay optical reserve pool calculation regression model, and obtain the optimal output connection weight; 6) Based on the selected number of forgotten points P, the first P data of the test set are repeated once to ensure that after forgetting the P point data, there are still enough eigenvalues to locate the position to be tested. The test set data is then masked and input into the delayed optical reserve pool. The corresponding virtual node status is collected, and the delayed optical reserve pool calculation regression model is tested to obtain the test output of each eigenvalue. The average of the 4 adjacent test output values is used as the disturbance position prediction result of the corresponding eigenvector to achieve disturbance positioning.

2. The linear Sagnac interferometry distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation according to claim 1, characterized in that: The structure of the time-delay optical storage pool in steps 5) to 6) is an all-optical or optoelectronic form.

3. The linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation according to claim 1, characterized in that: The structure of the delay optical reserve pool in steps 5) to 6) is in the form of a single physical node or multiple physical nodes.

4. The linear Sagnac interferometric distributed optical fiber sensing positioning method based on delayed optical reserve pool calculation according to claim 1, characterized in that: The structure of the delay optical reserve pool in steps 5) to 6) is a single-delay mode or a multi-delay mode.

5. The linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation according to claim 1, characterized in that: The mask processing operation in steps 5)-6) is: after each eigenvalue is expanded into M, it is multiplied by a mask sequence of length M, where M is the number of virtual nodes.

6. The linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation according to claim 1, characterized in that: In the operations of inputting data into the delay reserve pool in steps 5)-6), optical injection or electrical injection is used.

7. The linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reservoir calculation according to claim 1, characterized in that: In step 5), the training of the time-delay optical reservoir calculation regression model adopts traditional pseudo-inverse operation, ridge regression, or other training methods.

8. The linear Sagnac interferometric distributed optical fiber sensing positioning method based on time-delay optical reserve pool calculation according to claim 1, characterized in that: In step 6), the test operation of the time-delay optical reserve pool calculation regression model is: according to the optimal output connection weight obtained in the training process, the virtual node states of the test data are weighted summed to obtain the test output.

Images