A DAS multi-point joint vibration source positioning method based on PINO-EPFES-TDOA

Abstract

The application discloses a kind of DAS multi-point joint vibration source positioning method based on PINO-EPFES-TDOA, belong to signal processing and machine learning field;Specifically: first, the DAS original sensing signal under the input time slot position is denoised;Then, the position information of sampling point is obtained using sensing DAS, and the optimal propagation speed of soil medium is calculated;Based on the propagation speed, the loss function of PINO medium propagation model is calculated in combination with the denoised signal, and the sound wave attenuation coefficient in soil is obtained;Then, the soil medium propagation model is established, and the optimal feature interval boundary of soil medium propagation model energy main peak is calculated using EPFES;In the optimal feature interval boundary, a TDOA sampling point time delay estimation model is established, and finally a multi-point joint positioning error model is established, and the vibration source position is obtained by minimizing solution.The application provides a method with high positioning accuracy for the field of vibration source positioning.

Description

Technical Field

[0001] This invention belongs to the field of signal processing and machine learning, specifically a DAS (distributed acoustic sensing) multi-point joint vibration source localization method based on PINO-EPFES-TDOA (Physics-Informed Neural Network-Energy Peak Feature Extraction Strategy-Time Difference of Arrival). Background Technology

[0002] Currently, the main approach to locating DAS (Diverterless Orbital Signal) sources is to calculate the angle of arrival (AOA) and received signal strength (RSS). The general process involves first fitting the signal to establish a propagation model in the medium, and then using an energy propagation estimation algorithm to solve for the location.

[0003] For the problem of media propagation modeling of DAS signals, commonly used methods include ray tracing based on the assumption of a homogeneous medium and simplified modeling based on empirical Green's function. However, the former has low positioning accuracy, and the latter simplified modeling method is poor. In addition, due to the strong non-homogeneity and anisotropy of the underground medium, there may be situations such as curved propagation paths and abnormal velocity structures, which may affect the positioning accuracy of the vibration source.

[0004] While traditional beamforming-based methods offer excellent spatial resolution for vibration source localization, they suffer from limitations in angular resolution, high array calibration requirements, and significant computational complexity. In particular, for some close-range vibration sources, angular ambiguity and decreased positioning accuracy may occur.

[0005] The application of DAS (Digital Angle Sensor) for vibration source localization is relatively mature in many fields. For example, CNN, KNN, and various neural network algorithms have achieved good vibration source localization results in areas such as external threats to pipelines and perimeter security positioning.

[0006] Due to increasingly stringent safety requirements and the widespread application of distributed fiber optic sensing technology in fields such as aerospace, geological monitoring, and industrial safety, relatively high demands have been placed on the accuracy and precision of DAS source positioning.

[0007] Therefore, to achieve real-time and accurate vibration source positioning and meet the requirements of vibration source positioning, it is necessary to establish an efficient and accurate DAS vibration source positioning method, effectively reduce positioning errors, improve the positioning accuracy of DAS vibration source positioning, and provide real-time and accurate vibration source positioning results for many application fields of distributed fiber optic sensing technology, so that staff can promptly detect problems and make decisions in advance. Summary of the Invention

[0008] This invention provides a DAS multi-point joint vibration source localization method based on PINO-EPFES-TDOA, which realizes high-precision and real-time localization of vibration sources and solves the problems of low localization accuracy and poor fitting degree of vibration sources in complex non-uniform media environments.

[0009] The specific steps of the DAS multi-point joint vibration source localization method based on PINO-EPFES-TDOA are as follows:

[0010] Step 1: Input Time Slot Next position The original DAS sensor signal is denoised to obtain the denoised signal. :

[0011]

[0012] In the formula, Represents wavelet operators; Represents the threshold function; Indicates time slot Next position The original DAS sensor signal.

[0013] Step 2: Using the sensor DAS to acquire the first The and the first The location information of each sampling point is used to calculate the optimal propagation velocity in the soil medium. :

[0014]

[0015] In the formula, This indicates the speed at which sound waves travel through the soil. Indicates the first The and the first Spacing between sampling points; Indicates the number of experimentally measured values. The and the first The delay of each sampling point; , This represents the total number of sampling points in the DAS sensor.

[0016] Step 3: Utilize the denoised signal Optimal propagation speed in soil media Calculate the loss function of the PINO medium propagation model. The sound wave attenuation coefficient in the soil was obtained. ;

[0017] The specific process is as follows:

[0018] First, regarding time slots Next position Through the PINO neural network operator The output is obtained after layer iteration. :

[0019]

[0020] In the formula, This represents the number of operator layers in a neural network. Indicates linear projection. Indicates Fourier transform; Represents the frequency domain convolution kernel; Initial values ​​are set manually;

[0021] Then, the final iterative output of the PINO neural network operator is used. Combined with denoised signal Calculate the loss function :

[0022]

[0023] In the formula, This represents the sound wave attenuation coefficient in the soil; These are the weighting coefficients.

[0024] Finally, the soil medium propagation speed Substitute into the loss function By minimizing the loss function as an auxiliary solution :

[0025]

[0026] Step 4: Utilizing the sound wave attenuation coefficient in the soil Establish a soil media propagation model :

[0027]

[0028] In the formula, For soil media propagation model; For amplitude parameters, ; To correct the parameters, they are defined empirically.

[0029] Step 5: Calculate the soil media propagation model using the Energy Main Peak Feature Extraction Strategy (EPFES). The optimal characteristic interval boundary of the main energy peak;

[0030] The specific calculation process is as follows:

[0031] First, calculate the soil medium propagation model. Energy main peak :

[0032]

[0033] In the formula, The sensing signals at different locations are at the time t when the energy is at its maximum.

[0034] Then, determine the main energy peak. The optimal feature interval:

[0035]

[0036] In the formula, The point where the energy is at its maximum is taken. Front and back distance The nearest neighbor sample is used as the interval boundary, that is, the optimal feature interval is .

[0037] Step 6: Establish a TDOA sampling point delay estimation model within the optimal feature interval boundary. :

[0038]

[0039] In the formula, This represents the coordinates of the sampling points in a coordinate system established with respect to the direction of fiber optic detection. , , , where m represents the index of a different position within the optimal feature interval; Indicates the main peak point For reference points and Inter-delay estimation, Coordinates are This indicates the main peak point. The coordinates of the reference point correspond to the location of maximum energy. Location of vibration source .

[0040] Step 7: Utilize the time delay estimation model A multi-point joint positioning error model is established, and the vibration source location is obtained by minimizing the model.

[0041]

[0042] In the formula, This is a multi-point joint positioning error model.

[0043]

[0044] The location of the vibration source can be solved by minimizing this model. .

[0045] The advantages of this invention over the prior art are:

[0046] (1) This invention overcomes the problems of poor positioning accuracy of the homogeneous medium assumption and low fitting degree of the empirical Green's function for medium modeling. The established soil medium propagation model is more accurate and can effectively improve the positioning accuracy of the vibration source.

[0047] (2) This invention applies the advantages of PINO neural network combined with physical environment constraints to DAS vibration source localization, and achieves a high localization accuracy when combined with TDOA. This shows that this invention can achieve a good localization effect when targeting DAS vibration sources. Attached Figure Description

[0048] Figure 1 This is a flowchart of a DAS multi-point joint vibration source localization method based on PINO-EPFES-TDOA according to the present invention;

[0049] Figure 2 This is a schematic diagram of the PINO neural network used in this invention;

[0050] Figure 3 This is a comparison of the DAS vibration source positioning error of the present invention with that of three other methods;

[0051] Figure 4 This is an experimental diagram showing the error and iteration count of this invention; Detailed Implementation

[0052] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are merely some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0053] This invention provides a DAS multi-point joint vibration source localization method based on PINO-EPFES-TDOA, the specific process of which is as follows: Figure 1 As shown, it includes the following steps:

[0054] Step 1: Input Time Slot Next position The original DAS sensor signal is denoised to obtain the denoised signal. :

[0055]

[0056] In the formula, Represents wavelet operators; Represents the threshold function; Indicates time slot Next position The original DAS sensor signal.

[0057] Step 2: Using the sensor DAS to acquire the first The and the first The location information of each sampling point is used to calculate the optimal propagation velocity in the soil medium. :

[0058]

[0059] In the formula, This indicates the speed at which sound waves travel through the soil. Indicates the first The and the first Spacing between sampling points; Indicates the number of experimentally measured values. The and the first The delay of each sampling point; , This represents the total number of sampling points in the DAS sensor.

[0060] Step 3: Utilize the denoised signal Optimal propagation speed in soil media Calculate the loss function of the PINO medium propagation model. The sound wave attenuation coefficient in the soil was obtained. ;

[0061] The specific process is as follows:

[0062] First, regarding time slots Next position Through the PINO neural network operator The output is obtained after layer iteration. :

[0063]

[0064] In the formula, This represents the number of operator layers in a neural network. Indicates linear projection. Indicates Fourier transform; Represents the frequency domain convolution kernel; Initial values ​​are set manually;

[0065] Then, the final iterative output of the PINO neural network operator is used. Combined with denoised signal Calculate the loss function :

[0066]

[0067] In the formula, This represents the sound wave attenuation coefficient in the soil; These are the weighting coefficients.

[0068] Finally, the soil medium propagation speed Substitute into the loss function By minimizing the loss function as an auxiliary solution :

[0069]

[0070] Step 4: Utilizing the sound wave attenuation coefficient in the soil Establish a soil media propagation model :

[0071]

[0072] In the formula, For soil media propagation model; For amplitude parameters, ; To correct the parameters, they are defined empirically.

[0073] Step 5: Calculate the soil media propagation model using the Energy Main Peak Feature Extraction Strategy (EPFES). The optimal characteristic interval boundary of the main energy peak;

[0074] The specific calculation process is as follows:

[0075] First, calculate the soil medium propagation model. Energy main peak :

[0076]

[0077] In the formula, The sensing signals at different locations are at the time t when the energy is at its maximum.

[0078] Then, determine the main energy peak. The optimal feature interval:

[0079]

[0080] In the formula, The point where the energy is at its maximum is taken. Front and back distance The nearest neighbor sample is used as the interval boundary, that is, the optimal feature interval is .

[0081] Step 6: Establish a TDOA sampling point delay estimation model within the optimal feature interval boundary. :

[0082]

[0083] In the formula, This represents the coordinates of the sampling points in a coordinate system established with respect to the direction of fiber optic detection. , , , where m represents the index of a different position within the optimal feature interval; Indicates the main peak point For reference points and Inter-delay estimation, Coordinates are This indicates the main peak point. The coordinates of the reference point correspond to the location of maximum energy. Location of vibration source .

[0084] Step 7: Utilize the time delay estimation model A multi-point joint positioning error model is established, and the vibration source location is obtained by minimizing the model.

[0085]

[0086] In the formula, This is a multi-point joint positioning error model.

[0087]

[0088] The vibration source location is solved by minimizing this model. When the error meets a user-defined error threshold, the vibration source location is output. .

[0089] Example:

[0090] In this embodiment, 40 samples were collected at each of the 5 tapping points, with each sample containing 10 sampling points. A random sampling method was used, and the data was trained with a ratio of 90% training set and 10% test set, resulting in a total of 180 samples used for training. A total of 20 samples were used for testing.

[0091] The DAS vibration source localization method provided in this example has the following specific steps:

[0092] (1) Determine the denoised signal of the original signal :

[0093] Select a set of sampled data: Denoising is performed to obtain the denoised signal:

[0094] .

[0095] (2) Determine the soil medium propagation parameters :

[0096] Assumption , Seeking .

[0097] (3) Establish the loss function of the PINO medium propagation model :

[0098]

[0099] In the formula, Represents the loss function; This represents the estimated output at position z at time t; This represents the measured signal at position z at time t; pass control, As an intermediate variable, This represents the sound wave attenuation coefficient in the soil; These are the weighting coefficients. For example... Figure 2 The diagram shown is a schematic of the PINO neural network used in this example.

[0100] Then, the soil medium propagation parameters Substituting into the above equation, we can solve the problem by minimizing the loss function. and :

[0101]

[0102] in Take 0.3, and obtain it through iteration. , .

[0103] (4) Establish a soil medium propagation model :

[0104] From the previous step, we can conclude that:

[0105]

[0106] (5) Determine the boundary of the optimal feature interval for EPFES:

[0107] First, calculate the main energy peak point. :

[0108]

[0109] In the formula, assume that we take Seeking Then the interval boundary is .

[0110] (6) Establish a TDOA sampling point delay estimation model :

[0111] (7) Establish a multi-point joint positioning error model: Substituting the values, we obtain the solution using the gradient descent method. Coordinates are , and actual location The distance is 0.06m, which is less than 0.1m.

[0112] By testing the model using samples, DAS vibration source localization is achieved, and the localization results are obtained, thus completing the DAS multi-point joint vibration source localization method based on PINO-EPFES-TDOA.

[0113] To verify the accuracy of the present invention in locating the DAS vibration source, four sets of vibration source location experiments were conducted. The results are as follows: Figure 3 As shown, the DAS vibration source positioning method established in this invention maintains an error of less than 0.1m in positioning the DAS vibration signal, achieving high positioning accuracy while ensuring stability, and the positioning effect is good. Figure 4 The graph shows the relationship between the number of iterations and the position error. As the number of iterations increases, the error gradually decreases. This indicates that the DAS vibration source localization method established in this invention is effective, providing a better method for establishing accurate vibration source localization models and possessing certain practical value.

Claims

1. A DAS multi-point joint vibration source localization method based on PINO-EPFES-TDOA, characterized in that, The specific steps are as follows: Step 1: Input Time Slot Next position The original DAS sensor signal is denoised to obtain the denoised signal. : Step 2: Using the sensor DAS to acquire the first The and the first The location information of each sampling point is used to calculate the optimal propagation velocity in the soil medium. : Step 3: Utilize the denoised signal Optimal propagation speed in soil media Calculate the loss function of the PINO medium propagation model. The sound wave attenuation coefficient in the soil was obtained. ; loss function The calculation formula is: ； In the formula, This represents the output of the final iteration of the PINO neural network operator. This represents the sound wave attenuation coefficient in the soil; These are the weighting coefficients; To achieve the optimal propagation speed of the soil medium Substitute into the loss function Solving the problem by minimizing the loss function : ； Step 4: Utilizing the sound wave attenuation coefficient in the soil Establish a soil media propagation model : ； In the formula, For soil media propagation model; For amplitude parameters, ; To correct the parameters, they are defined empirically; Step 5: Calculate the soil media propagation model using the Energy Main Peak Feature Extraction Strategy (EPFES). The optimal characteristic interval boundary of the main energy peak; The specific calculation process is as follows: First, calculate the soil medium propagation model. Energy main peak : ； In the formula, The sensing signals at different locations during the time t when the energy is at its maximum; Then, determine the main energy peak. The optimal feature interval: ； In the formula, The point where the energy is at its maximum is taken. Front and back distance The nearest neighbor samples are used as the interval boundaries, i.e., the optimal feature interval is... ; Step 6: Establish a TDOA sampling point delay estimation model within the optimal feature interval boundary. : ； In the formula, This represents the coordinates of the sampling points in a coordinate system established with respect to the direction of fiber optic detection. , , , where m represents the index of a different position within the optimal feature interval; Indicates the main peak point For reference points and Inter-delay estimation, Coordinates are This indicates the main peak point. The coordinates of the reference point correspond to the location of maximum energy. Location of vibration source ; Step 7: Utilize the time delay estimation model A multi-point joint positioning error model is established, and the vibration source location is obtained by minimizing the model. ； In the formula, A multi-point joint positioning error model; ； The location of the vibration source can be solved by minimizing this model. .

2. The method as described in claim 1, characterized in that, The specific calculation for step one is as follows: ； In the formula, Represents wavelet operators; Represents the threshold function; Indicates time slot Next position The original DAS sensor signal.

3. The method as described in claim 1, characterized in that, The specific calculation for step two is as follows: ； In the formula, This indicates the speed at which sound waves travel through the soil. Indicates the first The and the first Spacing between sampling points; Indicates the number of experimentally measured values. The and the first The delay of each sampling point; , This represents the total number of sampling points in the DAS sensor.

4. The method as described in claim 1, characterized in that, In step three, regarding time slots Next position Through the PINO neural network operator The output is obtained after layer iteration. : In the formula, This represents the number of operator layers in a neural network. Indicates linear projection. Indicates Fourier transform; Represents the frequency domain convolution kernel; Initial values ​​are set manually.

Images