Distributed optical fiber large-range sensing positioning method based on time delay reserve pool calculation regression model

Abstract

The invention discloses a distributed optical fiber large-range sensing positioning method based on a time delay reserve pool calculation regression model, and the method comprises the steps: dividing a sensing optical fiber into Q segments, and selecting a certain number of disturbance positions in each segment for data collection; the virtual node state of the reserve pool is obtained through data post-processing; respectively configuring P linear regression devices used for positioning each section of sensing optical fiber and a regression device 0, P = Q used for positioning the whole sensing optical fiber on an output layer calculated by a reserve pool, training a regression model by using a linear regression algorithm and training set data of different sensing ranges, and performing parameter optimization by using verification set data; and judging whether the test outputs of the P regression devices are within a threshold range, comparing the test outputs within the threshold range with the test output of the regression device 0, and selecting the test output closest to the test output of the regression device 0 as the disturbance position output. The method is stable and accurate in positioning result, and can be used for disturbance positioning of any distributed optical fiber large-range sensing system based on time delay reserve pool calculation.

Description

Technical Field

[0001] The present invention relates to a distributed optical fiber sensing positioning method, in particular to a distributed optical fiber large-scale sensing positioning method based on a time delay reserve pool calculation regression model, and belongs to the field of optical fiber sensing. Background Art

[0002] Distributed fiber-optic disturbance sensing holds significant application potential in pipeline leak detection, perimeter security, and structural safety monitoring. Achieving early warning requires not only sensing and identifying external disturbance signals but also, more importantly, determining the location of the disturbance. Different sensing systems employ different positioning methods, but these results are often affected by environmental and system noise. Therefore, achieving rapid and accurate disturbance location in large-scale sensing environments with low signal-to-noise ratios presents a challenge.

[0003] Thanks to the high noise tolerance of machine learning algorithms for disturbance signal localization, in recent years, machine learning models such as convolutional neural networks, gated recurrent units, and random forests have been used to achieve disturbance localization and have achieved good localization results. However, the training of machine learning models, especially deep learning models, is relatively complex, and feature extraction is time-consuming, making it difficult to meet the immediacy requirements of distributed fiber optic sensing systems. Therefore, the applicant has proposed a distributed fiber optic sensing system and disturbance localization method based on delay reservoir calculation. The distributed fiber optic sensor and delay optical reservoir are combined into one, and the sensing process is transformed into a test process for the reservoir calculation. Disturbance localization is achieved through the training and testing process of the delay optical reservoir calculation. However, this localization method is based on the concept of classification. That is, the sensing fiber is divided into multiple segments according to the required spatial resolution, and the classification model calculated by the reservoir is used to determine which segment the disturbance occurs in. This classification localization method is conducive to the localization of multiple disturbance events, but the localization resolution is low, which loses the advantage of distributed fiber optic sensing for continuous monitoring. When the localization resolution and sensing distance are improved, the number of classifiable segments and the localization complexity will inevitably increase. The above shortcomings can be overcome by using a regression model calculated using a time delay reserve pool to locate disturbances. However, in actual applications, it has been found that simply using regression positioning for the entire sensing optical fiber will fail when sensing over a large range, and the positioning error will increase sharply with the increase in sensing distance. Using time delay reserve pool calculation to achieve distributed optical fiber large-scale sensing positioning has become a technical problem that needs to be solved urgently. Summary of the Invention

[0004] The present invention aims to overcome the shortcomings of existing technologies and provide a distributed fiber optic large-scale sensing and positioning method based on a time-delay pool calculation regression model. The sensing fiber is divided into multiple segments. Based on the concept of ensemble learning, the output layer uses multiple linear regressors, one corresponding to the positioning range of the entire sensing fiber, and the remaining linear regressors corresponding to each segment of the sensing fiber. The segmented positioning result closest to the overall positioning result is selected as the final positioning result. With segmented positioning, the position range involved in training is known. Based on this, a suitable threshold range can be set for the positioning results of each segmented positioning regressor. If the perturbation position of the test sample is not within the position range of the regressor, the positioning error will generally be large, and the positioning result will be outside the threshold range and discarded. Conversely, the error is small, and the positioning result will be within the threshold range. This result will be retained for subsequent comparison with the positioning result of the overall positioning regressor to reduce positioning error. The size of each threshold range can be adjusted through the training and verification process.

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

[0006] A distributed optical fiber large-scale sensing positioning method based on a time delay reserve pool calculation regression model includes the following steps:

[0007] 1) Data Acquisition: First, the sensing fiber is divided into Q segments, and a certain number of perturbation locations are selected on each segment. Perturbations are applied at each perturbation location, and the output signal of the sensing system is collected multiple times. A number of perturbation locations are unevenly selected on different segments of the sensing fiber as training locations. The data collected at these locations are used as training data for segmented and overall positioning, respectively. The validation data for both segmented and overall positioning include signal data collected at some training locations but not involved in training, as well as data from some non-training locations, to improve the generalization ability of the regression positioning model. The test data, i.e., the output signal of the sensing system to be located, can be data collected at any perturbation location on the sensing fiber.

[0008] 2) Data post-processing: Post-process the collected data to obtain the virtual node state of the reserve pool at time n x(n) = [x1(n), x2(n), ..., x N (n)] T , where N is the number of virtual nodes, x j (n) is the state of the jth virtual node, j = 1, 2, ..., N;

[0009] 3) Training process: In the output layer of the delay reserve pool calculation, P linear regressors are configured for segmented positioning of disturbances within each sensing fiber range and one regressor is configured for positioning disturbances within the entire sensing fiber range, where P = Q. The regression model of the delay reserve pool calculation is trained using the linear regression algorithm and training set data of different sensing ranges to obtain the optimal output connection weights W of different linear regressors. outi , where i is the regressor number, i = 0, 1, 2, 3, ..., P, and parameter optimization is performed using the validation set. An appropriate threshold range is set for the output of regressors 1-P, and only outputs within the threshold range are retained. The size of each threshold range is determined by training and validating the model using all training and validation set data.

[0010] 4) Test process: W outi Multiply it with the virtual node state extracted from the test set data to obtain the test output y of different regressors i (n) = W outi *x(n), after judgment and comparison, disturbance positioning is achieved.

[0011] Preferably, the number Q of sensing fiber segments in step 1) or the number P of segmented positioning regressors in step 3) is set according to the length of the sensing fiber and the training effect.

[0012] Preferably, the data post-processing in step 2) includes: low-pass filtering, clipping and normalizing the output signal of the sensing system, and extracting the virtual node state of the reserve pool on the normalized output signal segment; wherein the data clipping length is the optimal length to retain the length of the disturbance position characteristics.

[0013] Preferably, the linear regression algorithm in step 3) can be one of the linear regression algorithms such as pseudo-algorithm, ridge regression, and lasso regression.

[0014] Preferably, the parameters for parameter optimization in step 3) include parameters in the linear regression algorithm and the number of virtual nodes in the reserve pool.

[0015] Preferably, the judgment and comparison process in step 4) is as follows: determine whether the test outputs of regressors 1-P are within their respective threshold ranges, discard test outputs outside the threshold ranges, compare the test outputs of regressors 1-P within the threshold ranges with the test outputs of regressor 0, and select the one closest to the test output of regressor 0 as the perturbation position.

[0016] Preferably, the closest measurement standard in the judgment and comparison process in step 4) is the minimum absolute value of the difference between the two or other standards for measuring the proximity of two values.

[0017] Compared with the prior art, the present invention has the following obvious outstanding substantial features and significant advantages:

[0018] 1. This invention uses a time-delayed reservoir to calculate a regression model to locate disturbance signals. Compared to classification model positioning methods, this invention can achieve higher spatial resolution and can locate disturbances at unknown locations without collecting data at all disturbance locations. This maintains the continuous detection advantage of distributed fiber optic sensing while reducing the amount of data required for model training.

[0019] 2. This invention achieves large-scale disturbance location by segmenting the sensing fiber and configuring a linear regressor corresponding to each sensing range at the output layer. Unlike the classification approach of segmenting the sensing fiber to determine which segment the disturbance occurred, this invention segments the fiber longer, on the order of kilometers, rather than the tens of meters or one or two hundred meters of the classification approach. The number of segments is smaller, and within each segment, the disturbance location is achieved using regression. This means that increasing the sensing range does not significantly increase the complexity of model training. Therefore, the distributed optical fiber large-scale sensing location method based on the time delay pool calculation regression model is simple and efficient.

[0020] 3. Wide range of applicability. No matter whether it is a sensing system based on an all-optical delay reserve pool or a sensing system based on a photoelectric delay reserve pool, and no matter what specific structure is adopted, as long as it is a system that combines a delay reserve pool with a distributed optical fiber sensor, the positioning method of the present invention can be used. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the positioning method of the present invention.

[0022] Figure 2 Schematic diagram of the system structure of a preferred embodiment of the present invention.

[0023] Figure 3 Schematic diagram of an experimental device according to a preferred embodiment of the present invention.

[0024] Figure 4 This is the waveform of the system output signal after filtering when the external disturbance position is 981m in the preferred embodiment of the present invention.

[0025] Figure 5 This is the waveform of the system output signal after filtering when the external disturbance position is 3645m in the preferred embodiment of the present invention.

[0026] Figure 6 It is the average absolute error of the positioning results of the training set a under 5 different output connection weights in the preferred embodiment of the present invention.

[0027] Figure 7This is a comparison of the mean absolute errors of the positioning results of the training set a after only comparison and range judgment before comparison in the preferred embodiment of the present invention.

[0028] Figure 8 It is the mean absolute error of the positioning results of the test set when the number of virtual nodes takes the optimal value of 350 in the preferred embodiment of the present invention. DETAILED DESCRIPTION

[0029] 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:

[0030] Example 1:

[0031] In this embodiment, see Figure 1 A distributed optical fiber large-scale sensing positioning method based on a time delay reserve pool calculation regression model includes the following steps:

[0032] 1) Data Acquisition: First, the sensing fiber is divided into Q segments, and a certain number of perturbation locations are selected on each segment. Perturbations are applied at each perturbation location, and the output signal of the sensing system is collected multiple times. A number of perturbation locations are unevenly selected on different segments of the sensing fiber as training locations. The data collected at these locations are used as training data for segmented and overall positioning, respectively. The validation data for both segmented and overall positioning include signal data collected at some training locations but not involved in training, as well as data from some non-training locations, to improve the generalization ability of the regression positioning model. The test data, i.e., the output signal of the sensing system to be located, can be data collected at any perturbation location on the sensing fiber.

[0033] 2) Data post-processing: Post-process the collected data to obtain the virtual node state of the reserve pool at time n x(n) = [x1(n), x2(n), ..., x N (n)] T , where N is the number of virtual nodes, x j (n) is the state of the jth virtual node, j = 1, 2, ..., N;

[0034] 3) Training process: In the output layer of the delay reserve pool calculation, P linear regressors are configured for segmented positioning of disturbances within each sensing fiber range and one regressor is configured for positioning disturbances within the entire sensing fiber range, where P = Q. The regression model of the delay reserve pool calculation is trained using the linear regression algorithm and training set data of different sensing ranges to obtain the optimal output connection weights W of different linear regressors. outi, where i is the regressor number, i = 0, 1, 2, 3, ..., P, and parameter optimization is performed using the validation set. An appropriate threshold range is set for the output of regressors 1-P, and only outputs within the threshold range are retained. The size of each threshold range is determined by training and validating the model using all training and validation set data.

[0035] 4) Test process: W outi Multiply it with the virtual node state extracted from the test set data to obtain the test output y of different regressors i (n) = W outi *x(n), after judgment and comparison, disturbance positioning is achieved.

[0036] The distributed optical fiber large-scale sensing positioning method based on the time delay reserve pool calculation regression model in this embodiment does not require signal demodulation or disturbance position relationship, the data processing method is simple, and the positioning results are stable and accurate.

[0037] Example 2:

[0038] This embodiment is basically the same as the first embodiment, with the following special features:

[0039] Preferably, the sensing fiber segment Q in step 1) or the number of regressors P for segment positioning in step 3) is set according to the sensing fiber length and the training effect.

[0040] In this embodiment, the data post-processing in step 2) includes low-pass filtering, clipping, and normalizing the output signal of the sensor system, and extracting the virtual node states of the reservoir based on the normalized output signal segments. The optimal length of the data clipping is the length that preserves the characteristics of the disturbance position.

[0041] In this embodiment, the linear regression algorithm in step 3) may be a pseudo-algorithm or one of linear regression algorithms such as ridge regression and lasso regression.

[0042] In this embodiment, the parameters for parameter optimization in step 3) include parameters in the linear regression algorithm and the number of virtual nodes in the reserve pool.

[0043] In this embodiment, the judgment and comparison process in step 4) is as follows: determine whether the test outputs of regressors 1-P are within their respective threshold ranges, discard test outputs outside the threshold ranges, compare the test outputs of regressors 1-P within the threshold ranges with the test outputs of regressor 0, and select the one closest to the test output of regressor 0 as the perturbation position.

[0044] In this embodiment, the closest measurement standard in the judgment and comparison process in step 4) is the minimum absolute value of the difference between the two or other standards for measuring the proximity of two values.

[0045] This embodiment ensures the continuous detection advantage of distributed optical fiber sensing while reducing the amount of data required, and can predict the position of any unknown disturbance on the sensing optical fiber. The positioning process is simple and efficient, and can be used for disturbance positioning of distributed optical fiber large-scale sensing systems of any structure based on delay reserve pool calculation.

[0046] Example 3:

[0047] This embodiment is basically the same as the above embodiment, with the following special features:

[0048] In this embodiment, a distributed fiber optic sensing system based on semiconductor optical amplifier (SOA) ring cavity reservoir calculation is selected and used for pipeline leakage monitoring to verify the feasibility of the distributed fiber optic large-scale sensing positioning method based on the time delay reservoir calculation regression model of this embodiment.

[0049] like Figure 2 As shown, a distributed fiber optic sensing system based on SOA ring cavity reservoir calculation consists of an input layer, a reservoir, and an output layer. The input layer has only one node for inputting external disturbance signals. The reservoir is a time-delay nonlinear dynamic system with SOA4 as a nonlinear node. Part of its delay fiber 5 serves as a sensing fiber, so the system is also a distributed fiber optic sensor, which is mainly composed of an SOA ring cavity injected by a broadband light source 1. The SOA ring cavity mainly consists of a 2×2 fiber coupler 2, a circulator 3, an SOA4, a delay fiber 5, a reflector 6, and a photodetector 7. Since actual disturbance signals are mostly low-frequency, wide-spectrum signals with long durations, in order to make the output signal contain complete waveform changes related to the disturbance position, an additional blind fiber is required that does not participate in sensing but only increases the time delay. Therefore, only the part of the delay fiber near the reflector serves as the sensing fiber. The output layer is used to output the disturbance position, that is, the distance between the disturbance point and the reflector 6. The output layer is configured with 5 linear regressors, which correspond to the disturbance positioning of the entire sensing fiber and different segments of the sensing fiber respectively. The disturbance position is determined through judgment and comparison.

[0050] The experimental setup of the distributed optical fiber sensing system based on SOA ring cavity reservoir calculation is as follows: Figure 3 As shown, a polarization controller 8 and an adjustable attenuator 9 are added between the fiber coupler 2 and the circulator 3 to adjust the polarization state and feedback intensity of the light in the ring cavity, respectively. The delay fiber 5 between the SOA 4 and the reflector 6 consists of two parts: the part near the SOA 4 is the blind fiber 5-1, and the part near the reflector 6 is the sensing fiber 5-2.

[0051] A VENUS series desktop superluminescent diode (SLLD) from Shanghai Connet Laser Technology Co., Ltd. was used as a broadband light source, with an output power range of 0 to 20 mW and a bandwidth of 40 nm. The coupling ratio of the 2×2 coupler was 50:50. A General Photonics fiber squeezer (PLC-001) was used as the polarization controller, a Thorlabs variable fiber attenuator (M-VA / 00019881) was used as the variable attenuator, and a CIP semiconductor optical amplifier module (SOA-SC-14-FCA) was used as the semiconductor optical amplifier, with a maximum drive current of 150 mA. Light takes approximately 61.2 μs to circulate once in the ring cavity, resulting in a ring cavity length of approximately 6.203 km. The blind fiber is 2.16 km long, and the sensing fiber is 3.96 km long. All optical fibers are G.652 standard single-mode fibers. An iXblue lithium niobate phase modulator (MPX-LN-0.1) was inserted at various locations within the sensing fiber. A SIGLENT signal generator (SDG1010) was used to apply a 6Vpp sinc signal with a frequency of 60kHz. The sinc signal operated in burst mode with the following parameters: a period of 250μs and a cycle duration. This signal simulated a broad-spectrum disturbance signal generated by concentrated energy release at the moment of a crack or leak, such as a pipeline leak. An OPEAK InGaAs photodetector with a bandwidth of 150MHz was used to detect the output signal. The data acquisition and processing systems consisted of a PicoScope 5203 digital oscilloscope from PICO and a Dell Inspiron 15-7567 laptop, respectively. The oscilloscope transmitted the acquired data to the laptop, where MATLAB R2022b software was used for data post-processing, linear regressor training, and disturbance location determination.

[0052] The sensing fiber was divided into four sections. Twenty-eight locations along the four sections were perturbed to simulate leakage. 1,400 measurements were taken at each location, resulting in a total of 39,200 samples. From these 28 locations, one location was selected every 200 to 300 meters as a training location. From the 1,400 samples at each training location, 800 samples were selected as the training set. Non-training samples collected at some training locations and samples from some non-training locations were used to construct a validation set, with 400 samples at each location. Samples from the remaining locations served as a test set, with 200 samples at each location. The composition of the different datasets is shown in Table 1. The numbers of the training and validation sets correspond to the numbers of the corresponding regressors. The training sets for the five regressors cover the ranges of 249–1107 m, 1107–2100 m, 2100–2979 m, 2979–3855 m, and 249–3645 m, respectively. The beginning and end locations of the fiber segments overlap to prevent missing locations during training. Table 1 also includes a training set and a validation set, labeled a. These are used to determine the threshold range for segmented localization results and the optimal number of virtual nodes in the reserve pool after comparing segmented localization results with overall localization results. The locations and samples in training set a are identical to those used in segmented training. The validation set a shares the same locations as training set a but differs in samples.

[0053] After applying the simulated leakage signal, the output response of the sensing system is low-pass filtered as follows: Figure 4 、 Figure 5 As shown in the figure, the closer the disturbance point is to the reflector, the longer the time interval between the two waveform changes, and the output signal contains features related to the disturbance position. To preserve the complete position features, the signal duration is trimmed to 80μs, and the virtual node state is extracted from this signal.

[0054] Table 1 Composition of different datasets

[0055]

[0056] The regression model was trained with different training sets and the output connection weights were further optimized by continuously changing the number of virtual nodes (N = [50:50:1000]). After training and validation, and by comprehensively comparing the MAE of the training and validation sets of the five regressors, it was found that better positioning results could be obtained when the number of virtual nodes was between 300 and 400. When the number of virtual nodes was selected as 350, the MAE of the positioning results obtained by training set a under 5 different output connection weights is as follows: Figure 6 As shown, the horizontal axis is the position of the training set a, arranged in order from near to far. out0 、W out1 、W out2 、W out3 、W out4 They were trained using training sets 0–4, so Figure 6As can be seen, different output connection weights can only obtain better positioning results within their corresponding training position ranges, and the positioning errors outside the training position ranges are relatively large. The output connection weight W of the regressor 0 out0 can achieve a relatively large positioning range, but the positioning errors at some positions are relatively large, that is, it is difficult for a single regressor to achieve large-range disturbance positioning. Through judgment and comparison, the final positioning result of the training set a is as Figure 7 shown. After adding the judgment of the threshold range of the positioning result, for example, setting the threshold range of the regressor 2 as: 1240m < y2(n) < 1980m, which is close to the training position range, the positioning error is significantly reduced compared with direct comparison without the judgment process. This means that judgment and comparison are effective in improving the positioning accuracy.

[0057] By further verifying the influence of the number of virtual nodes on the final positioning result after adding range judgment and positioning result comparison using the training set a and the validation set a, the MAE of the training set a and the validation set a will also gradually decrease as the number of virtual nodes increases. However, after the number of virtual nodes exceeds 350, the decrease in the MAE of the validation set a is not obvious, which means that the model is overfitting at this time. Therefore, 350 is the optimal number of virtual nodes.

[0058] During the test process, select the number of virtual nodes as 350, and the positioning result of the test set is as Figure 8 shown. It can be seen that the MAE distribution of the test positions is around 40m, with a maximum of 52m and a minimum of 16m. This shows that the distributed optical fiber large-range sensing and positioning method based on the time-delay reservoir computing regression model has the feasibility of positioning any disturbed position of the sensing optical fiber, with a relatively small average positioning error and stable and reliable positioning results.

[0059] In summary, the above-described embodiment of the distributed optical fiber large-range sensing and positioning method based on the time-delay reservoir computing regression model divides the sensing optical fiber into Q segments, selects a certain number of disturbed positions in each segment for data acquisition; obtains the virtual node states of the reservoir through data post-processing; respectively configures P linear regressors for positioning each segment of the sensing optical fiber and a regressor 0 for positioning the entire sensing optical fiber at the output layer of the reservoir calculation, where P = Q, uses the linear regression algorithm and the training set data of different sensing ranges to train the regression model, and uses the validation set data for parameter optimization; judges whether the test outputs of the P regressors are within the threshold range, and after comparing the test outputs within the threshold range with the test output of the regressor 0, selects the one closest to the test output of the regressor 0 as the disturbed position output. The positioning result of this embodiment is stable and accurate, and can be used for the disturbance positioning of any distributed optical fiber large-range sensing system based on the time-delay reservoir calculation.

[0060] 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 distributed optical fiber large-scale sensing positioning method based on a time delay reserve pool calculation regression model, characterized in that: The steps include: 1) Data Acquisition: First, the sensing fiber is divided into Q segments, and a certain number of perturbation locations are selected on each segment. Perturbations are applied at each perturbation location, and the output signal of the sensing system is collected multiple times. A number of perturbation locations are unevenly selected on different segments of the sensing fiber as training locations. The data collected at these locations are used as training data for segmented and overall positioning, respectively. The validation data for both segmented and overall positioning include signal data collected at some training locations but not involved in training, as well as data from some non-training locations, to improve the generalization ability of the regression positioning model. The test data, i.e., the output signal of the sensing system to be located, can be data collected at any perturbation location on the sensing fiber. 2) Data post-processing: Post-process the collected data to obtain the virtual node state of the reserve pool at time n x(n) = [x1(n), x2(n), ..., x N (n)] T , where N is the number of virtual nodes, x j (n) is the state of the jth virtual node, j = 1, 2, ..., N; 3) Training process: In the output layer of the delay reserve pool calculation, P linear regressors are configured for segmented positioning of disturbances within each sensing fiber segment and a regressor 0 for positioning disturbances within the entire sensing fiber segment, where P = Q. The regression model of the time delay reservoir calculation is trained using the linear regression algorithm and the training set data of different sensing ranges to obtain the optimal output connection weights W of different linear regressors. outi , where i is the regressor number, i = 0, 1, 2, 3, ..., P, and parameter optimization is performed using the validation set. An appropriate threshold range is set for the output of regressors 1-P, and only outputs within the threshold range are retained. The size of each threshold range is determined by training and validating the model using all training and validation set data. 4) Test process: W outi Multiply it with the virtual node state extracted from the test set data to obtain the test output y of different regressors i (n) = W outi *x(n), after judgment and comparison, disturbance positioning is achieved.

2. The distributed optical fiber large-scale sensing positioning method based on the time delay reserve pool calculation regression model according to claim 1 is characterized in that: The number Q of sensing fiber segments in step 1) or the number P of segmented positioning regressors in step 3) is set according to the length of the sensing fiber and the training effect.

3. The distributed optical fiber large-scale sensing positioning method based on the time delay reserve pool calculation regression model according to claim 1 is characterized in that: The data post-processing in step 2) includes: low-pass filtering, clipping and normalizing the output signal of the sensing system, and extracting the virtual node state of the reserve pool on the normalized output signal segment; wherein the data clipping length is the optimal length to retain the length of the disturbance position feature.

4. The distributed optical fiber large-scale sensing positioning method based on the time delay reserve pool calculation regression model according to claim 1 is characterized in that: The linear regression algorithm in step 3) is one of the linear regression algorithms such as pseudo-algorithm, ridge regression, and lasso regression.

5. The distributed optical fiber large-scale sensing method based on the time delay reserve pool calculation regression model according to claim 1 is characterized in that: The parameters in the parameter optimization in step 3) include the parameters in the linear regression algorithm and the number of virtual nodes in the reserve pool.

6. The distributed optical fiber large-scale sensing positioning method based on the time delay reserve pool calculation regression model according to claim 1 is characterized in that: The judgment and comparison process in step 4) is as follows: determine whether the test outputs of regressors 1-P are within their respective threshold ranges, discard test outputs outside the threshold ranges, compare the test outputs of regressors 1-P within the threshold ranges with the test outputs of regressor 0, and select the one closest to the test output of regressor 0 as the perturbation position.

7. The distributed optical fiber large-scale sensing positioning method based on the time delay reserve pool calculation regression model according to claim 6 is characterized in that: The closest measurement standard is the smallest absolute value of the difference between the two or other standards that measure the proximity of two values.

Images