CN115790890A - Signal processing method and terminal of a Brillouin optical time domain analysis system - Google Patents

Abstract

The invention discloses a signal processing method of a Brillouin optical time domain analysis system, which comprises the following steps of: s1, acquiring a Brillouin gain spectrum, and performing two-dimensional variational modal decomposition and reconstruction; s2, performing spectrum type correction through time slot difference, and correcting the distorted Brillouin gain spectrum into a Lorentz spectrum; and S3, extracting temperature information from the corrected Brillouin gain spectrum based on the data model. The three functions of noise reduction, spectrum type correction and temperature information extraction are realized, and the defects of poor measurement precision and long processing time of the traditional fitting algorithm are overcome.

Description

Signal processing method and terminal of a Brillouin optical time domain analysis system

technical field

The invention relates to the technical field of signal processing, in particular to a signal processing method of a Brillouin optical time domain analysis system and a terminal thereof.

Background technique

The distributed optical fiber sensing system based on Brillouin scattering has many advantages such as large sensing dynamic range, long monitoring distance and continuous distributed measurement. It is widely used in the health status diagnosis of bridges, oil pipelines and transmission lines. The Brillouin optical time domain analysis system (BOTDA) realizes the monitoring along the optical fiber according to the functional relationship between the Brillouin frequency shift and temperature/strain. Therefore, it is very important to obtain the Brillouin frequency shift accurately. Long-distance Brillouin optical time-domain sensing systems are susceptible to non-local effects, and their Brillouin gain spectrum will be distorted. When the sensing signal is distorted and the signal-to-noise ratio is low, conventional signal processing methods have problems such as poor measurement accuracy, insufficient denoising performance, loss of spectral details, and long processing time.

Contents of the invention

The technical problem to be solved by the present invention is to provide a signal processing method and a terminal of the Brillouin optical time domain sensing system for the situation where the sensing signal is distorted and the signal-to-noise ratio is low, so as to improve the accuracy of the temperature information along the optical fiber. measurement accuracy.

In order to solve the above-mentioned technical problems, a kind of technical scheme that the present invention adopts is:

A signal processing method of a Brillouin optical time domain analysis system, comprising the following steps:

S1. Collect Brillouin gain spectrum, perform two-dimensional variational mode decomposition and reconstruction;

S2. Perform spectral type correction through time slot difference, and correct the distorted Brillouin gain spectrum to Lorentzian spectral type;

S3. Extract temperature information from the corrected Brillouin gain spectrum based on the data model.

In order to solve the above-mentioned technical problems, another kind of technical scheme that the present invention adopts is:

A signal processing terminal of a Brillouin optical time-domain analysis system, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, and the processor implements the above-mentioned computer program when executing the computer program Various steps in a signal processing method of a Brillouin optical time domain analysis system.

The beneficial effects of the present invention are: to provide a signal processing method of a Brillouin optical time-domain sensing system and its terminal, to realize three functions of noise reduction, spectrum type correction, and temperature information extraction, so as to improve the signal-to-noise ratio while maintaining The edge characteristics of the Brillouin gain spectrum image signal; after the distorted Brillouin gain spectrum is corrected by the spectral type of the time slot difference, it presents a good Lorentzian shape, which helps to improve the accuracy of temperature extraction; and based on the data model The temperature information along the optical fiber is directly extracted, which overcomes the disadvantages of poor measurement accuracy and long processing time in the traditional fitting algorithm.

Description of drawings

1 is a schematic diagram of a single-mode optical fiber stimulated Brillouin scattering of a signal processing method of a Brillouin optical time domain analysis system according to an embodiment of the present invention;

Fig. 2 is a noise reduction flowchart of a signal processing method of a Brillouin optical time domain analysis system according to an embodiment of the present invention;

Fig. 3 is an optimization flowchart of a signal processing method of a Brillouin optical time domain analysis system according to an embodiment of the present invention;

4 is a general flowchart of a signal processing method of a Brillouin optical time domain analysis system according to an embodiment of the present invention;

FIG. 5 is a structural diagram of a signal processing terminal of a Brillouin optical time domain analysis system according to an embodiment of the present invention.

Detailed ways

In order to describe the technical content, achieved goals and effects of the present invention in detail, the following descriptions will be made in conjunction with the embodiments and accompanying drawings.

Please refer to Fig. 1 to Fig. 4, an embodiment of the present invention provides a signal processing method of a Brillouin optical time domain analysis system, including the following steps:

S1. Collect Brillouin gain spectrum, perform two-dimensional variational mode decomposition and reconstruction;

S2. Perform spectral type correction through time slot difference, and correct the distorted Brillouin gain spectrum to Lorentzian spectral type;

S3. Extract temperature information from the corrected Brillouin gain spectrum based on the data model.

It can be seen from the above description that the beneficial effect of the present invention is to provide a signal processing method for a Brillouin optical time-domain sensing system, which realizes three functions of noise reduction, spectrum type correction, and temperature information extraction, and improves the signal-to-noise ratio. At the same time, the edge characteristics of the Brillouin gain spectrum image signal are maintained; after the distorted Brillouin gain spectrum is corrected by the spectral type of time slot difference, it presents a good Lorentzian shape, which helps to improve the accuracy of temperature extraction; and based on The data model directly extracts the temperature information along the optical fiber, which overcomes the shortcomings of poor measurement accuracy and long processing time in the traditional fitting algorithm.

Further, the acquisition of the Brillouin gain spectrum described in step S1 is specifically:

Obtain the BOTDA system to obtain m sets of one-dimensional Brillouin gain spectrum signals along the optical fiber through frequency sweeping. The one-dimensional Brillouin gain spectrum signal is a two-dimensional matrix M with m columns and n rows, and the matrix M includes each position of the optical fiber The local Brillouin gain spectrum at :

In the formula, f(v _i , z _j )(1≤i≤m, 1≤j≤n) represents the Brillouin gain spectrum at the i-th sweep frequency point and the j-th sampling point within the fiber length range.

From the above description, it can be seen that corresponding to different scanning frequencies, the BOTDA system can obtain the one-dimensional Brillouin gain curve of each spatial position point along the optical fiber, and decompose the two-dimensional Brillouin gain spectrum signal obtained by frequency scanning into multiple sub-modes, then The center frequency of each mode is different, that is, the collected Brillouin gain spectrum can be regarded as a two-dimensional signal distributed with frequency and distance.

Further, the two-dimensional variational mode decomposition and reconstruction described in step S1 are specifically:

Set f(v _i ,z _j )(1≤i≤m,1≤j≤n) as f(x), and set its constrained variational model as:

In the formula, u _k ={u ₁ ,u ₂ ,…,u _k } are the decomposed K modal functions, and w _k ={w ₁ ,w ₂ ,…,w _k } are the corresponding center frequencies;

Introduce the quadratic penalty factor and the Lagrangian multiplier λ, and use the alternating direction multiplier method to optimize the constrained variational model to obtain the unconstrained problem-extended Lagrangian function:

直到满足

时，迭代停止并输出各个IMF分量：alternate update

until satisfied

When , the iteration stops and the individual IMF components are output:

和

分别代表u_k和w_k的频域性质；In the formula, τ is the residual error,

and

represent the frequency domain properties of u _k and w _k respectively;

The reconstructed signal is obtained by combining the IMF components obtained from the decomposition.

It can be seen from the above description that the reconstructed signal has removed the noise component of the original signal and has a good signal-to-noise ratio.

Further, step S1 also includes performing transform domain noise reduction processing for signals with a low signal-to-noise ratio, specifically:

S101, input the noisy two-dimensional Brillouin gain spectrum signal collected by the BOTDA system, and set the initial value of the number of decomposition layers to 2;

S102. Set the correlation coefficient screening threshold to 0.3, calculate the correlation coefficient between IMF1 and the original signal under the current value of the modal decomposition, and judge whether the correlation coefficient between IMF1 and the original signal is greater than the threshold, and when it is greater than the threshold, then K=K-1 is is the optimal number of decompositions, otherwise K=K+1 and continue to execute step S102;

S103. Perform a decomposition process using the values determined in step S102 to obtain a modal signal and reconstruct it.

It can be known from the above description that the low-frequency and high-frequency parts are separated by using the correlation coefficient, and the low-frequency part is reconstructed to obtain a two-dimensional Brillouin gain spectrum signal after noise reduction.

Further, step S2 is specifically:

Let the Brillouin gain spectrum after noise reduction be M':

Randomly select a line from the time-slotted spectral data that does not carry stimulated Brillouin scattering information, denoted as C:

C=[F(v ₁ )F(v ₂ )...F(v _m )]

The difference between each row of data in matrix M' and C is obtained to obtain the corrected Lorentzian Brillouin gain spectrum data.

It can be seen from the above description that, for the distorted Brillouin gain spectrum caused by non-local effects, the spectral type correction scheme of time slot difference is used to correct the distorted spectrum to the Lorentzian spectral type. Because the pump pulse repetition time of the system is longer than the time for the pulse to traverse the length of the fiber and return to the scattered signal, there is a time slot during each frequency sweep measurement, during which the photodetector collects the light that has not been combined with the pump pulse Encountered continuous optical signals do not carry SBS information. The purpose of spectral type correction can be achieved by making a difference between the Brillouin gain spectrum at each position along the fiber and the continuous light spectrum without SBS.

Further, the temperature information directly extracted from the Brillouin gain spectrum based on the data model in step S3 is specifically:

S301. Establish a data model, and conduct continuous learning and training to obtain a mapping function relationship between Brillouin frequency shift and temperature;

S302. Through simulation and laboratory calibration, generate L training samples (A _i , T _i ), i=1,2,...,L, where A _i =[a _i (v ₁ ),a _i (v ₂ ) ,...,a _i (v _m )] is the i-th sample, that is, the Brillouin gain spectrum under different frequency sweeps; T _i is the corresponding output vector, that is, the temperature value; the number of hidden layer nodes is N, and the output vector as follows:

In the formula, w _j is the input weight between the input node and the hidden layer node, β _j is the output weight between the hidden layer node and the output node, b _j is the bias of the jth hidden layer node, g(·) is the activation function, let β=[β ₁ β ₂ …β _N ] ^T , T=[T ₁ T ₂ …T _L ], then the output of hidden layer nodes is:

The output vector is expressed in the form of matrix: Hβ=T;

S303, there are n sampling points along the optical fiber, select the Brillouin gain spectrum at each sampling point of the optical fiber, input it into the trained model for testing, and obtain the corresponding temperature, that is, the temperature information along the optical fiber.

From the above description, it can be seen that at the beginning of measurement or in the laboratory calibration stage, by learning and training the Brillouin spectrum data collected under different temperature conditions, the mapping function relationship between the Brillouin frequency shift spectrum and temperature is established, and the actual There is no need to determine the Brillouin frequency shift during the monitoring process, and the temperature information along the optical fiber can be obtained directly through the trained functional relationship. While improving the measurement accuracy, the data processing time is effectively reduced.

Further, step S3 also includes optimizing the input weight w and bias b: taking the root mean square error of the prediction result of the training set as the fitness function, setting the goal of parameter optimization to minimize the root mean square error of the prediction result, Jump out of the loop when the fitness value is the minimum or the maximum number of iterations is met.

It can be seen from the above description that by optimizing the input weight w and bias b, the prediction accuracy of the model can be improved.

Please refer to Fig. 5, another embodiment of the present invention provides a signal processing terminal of a Brillouin optical time domain analysis system, including a memory, a processor, and a computer program stored in the memory and operable on the processor When the processor executes the computer program, each step in the signal processing method of the above-mentioned Brillouin optical time domain analysis system is realized.

The signal processing method and terminal of the above-mentioned Brillouin optical time domain sensing system of the present invention can effectively improve the measurement accuracy of the Brillouin optical time domain analysis system. The following will be described through specific implementation methods:

Embodiment one

Please refer to Figure 4, a signal processing method of a Brillouin optical time domain analysis system, including the following steps:

S1. Collect Brillouin gain spectrum, perform two-dimensional variational mode decomposition and reconstruction;

Wherein, the Brillouin gain spectrum collected in step S1 is specifically:

Obtain the BOTDA system to obtain m groups of one-dimensional Brillouin gain spectrum signals along the optical fiber through frequency scanning. Specifically, corresponding to different scanning frequencies, the BOTDA system can obtain the one-dimensional Brillouin gain curve of each spatial position point along the optical fiber:

The one-dimensional signals of m groups of sweep frequency points are stored in columns to obtain a two-dimensional matrix M with m columns and n rows. The matrix M contains the local Brillouin gain spectrum at each position of the fiber, then:

v ₁ ,v ₂ ,…,v _m is the scanning frequency, z ₁ ,z ₂ ,…,z _n is the fiber position, S ₁ ,S ₂ ,…,S _m is the scanning frequency v ₁ ,v ₂ ,…,v m sets of one-dimensional signals obtained under _m .

The two-dimensional Brillouin gain spectrum signal obtained by frequency sweep is decomposed into multiple sub-modes, and the center frequency of each mode is different, that is, the collected Brillouin gain spectrum can be regarded as a two-dimensional signal distributed with frequency and distance.

The one-dimensional Brillouin gain spectrum signal is a two-dimensional matrix M with m columns and n rows, and the matrix M contains the local Brillouin gain spectrum at each position of the fiber:

In the formula, f(v _i , z _j )(1≤i≤m, 1≤j≤n) represents the Brillouin gain spectrum at the i-th sweep frequency point and the j-th sampling point within the fiber length range.

Wherein, the two-dimensional variational mode decomposition and reconstruction in step S1 are specifically as follows:

To simplify the description, f(v _i ,z _j )(1≤i≤m,1≤j≤n) is set as f(x), and its constrained variational model is:

In the formula, u _k ={u ₁ ,u ₂ ,…,u _k } are the decomposed K modal functions, and w _k ={w ₁ ,w ₂ ,…,w _k } are the corresponding center frequencies;

Introduce the quadratic penalty factor and the Lagrangian multiplier λ, and use the alternating direction multiplier method to optimize the constrained variational model, and get the unconstrained problem-extended Lagrangian function:

直到满足

时，迭代停止并输出各个IMF分量：alternate update

until satisfied

When , the iteration stops and the individual IMF components are output:

和

分别代表u_k和w_k的频域性质；In the formula, τ is the residual error,

and

represent the frequency domain properties of u _k and w _k respectively;

The reconstructed signal is obtained by combining the IMF components obtained from the decomposition. The reconstructed signal has removed the noise component of the original signal and has a good signal-to-noise ratio.

Wherein, referring to FIG. 2 , step S1 also includes noise reduction processing for signals with a low signal-to-noise ratio through transform domain, specifically:

S101, input the noisy two-dimensional Brillouin gain spectrum signal collected by the BOTDA system, and set the initial value of the number of decomposition layers to 2;

S102. Set the correlation coefficient screening threshold to 0.3, calculate the correlation coefficient between IMF1 and the original signal under the current value of the modal decomposition, and judge whether the correlation coefficient between IMF1 and the original signal is greater than the threshold, and when it is greater than the threshold, then K=K-1 is is the optimal number of decompositions, otherwise K=K+1 and continue to execute step S102;

S103. Perform a decomposition process using the values determined in step S102 to obtain a modal signal and reconstruct it.

Specifically, the frequency of sub-modes decreases successively, so the center frequency of IMF1 is the highest. When over-decomposition occurs, IMF1 corresponds to the decomposed pseudo component. The effective information and the edge part are kept in the low-frequency mode, while the noise to be removed is concentrated in the high-frequency mode. The low-frequency and high-frequency parts are separated by the correlation coefficient, and the two-dimensional Brillouin gain spectrum signal after noise reduction can be obtained by reconstructing the low-frequency part.

S2. Perform spectral type correction through time slot difference, and correct the distorted Brillouin gain spectrum to Lorentzian spectral type;

Wherein, step S2 is specifically:

Let the Brillouin gain spectrum after noise reduction be M':

Randomly select a line from the time-slotted spectral data that does not carry stimulated Brillouin scattering information, denoted as C:

C=[F(v ₁ )F(v ₂ )...F(v _m )]

The difference between each row of data in matrix M' and C is obtained to obtain the corrected Lorentzian Brillouin gain spectrum data. The corrected Lorentzian shape Brillouin gain spectrum data I is:

In the formula, f ⁿ (v _i , z _j )=f'(v _i , z _j )-F(v _i ), i=1,2,...,m, j=1,2,...,n.

Specifically, please refer to FIG. 1. In the BOTDA system, the pump pulse light and the continuous light propagate in opposite directions, and stimulated Brillouin scattering effect (SBS) will occur when they meet in the optical fiber. Let the repetition interval of the pump pulse light be t, and the length of the sensing fiber be l. A pump pulse light traverses to the end of the optical fiber, encounters continuous light along the way, and returns to the head end with SBS information to be collected by the photodetector. The time taken is set to t ₁ , and the time left for the next pulse light emission is tt ₁ , this time interval is called a time slot. What the photodetector collects in the time slot is a continuous optical signal that has never encountered the pump pulse light, and does not carry SBS information. The purpose of spectral type correction can be achieved by making a difference between the Brillouin gain spectrum at each position along the fiber and the continuous light spectrum without SBS.

The data collected by the BOTDA system includes two parts: corresponding to each scanning frequency, the time t ₁ in Fig. 1 corresponds to the sampling points z ₁ , z ₂ ,…, z _n on the fiber length, where the SBS effect occurs; at the time slot tt ₁ No SBS effect occurred during this period. The data of each frequency point along the length of the optical fiber where the SBS effect occurs constitutes M; the data of each frequency point at any time during the time slot constitutes C.

Its core idea is to use the time slot to make a difference between the spectral signal with or without the SBS effect. Because the system's pump pulse repetition time is longer than the time it takes for the pulse to traverse the length of the fiber and return the scattered signal, there is a time slot during each sweep measurement. What the photodetector collects in this time slot is the spectral signal of continuous light that has not met the pulsed light, and the SBS effect does not occur, so it is named time slot spectrum. The purpose of spectrum type correction can be achieved by differential calculation of the Brillouin gain spectrum and the time slot spectrum at each position along the fiber.

S3. Extract temperature information from the corrected Brillouin gain spectrum based on the data model.

Wherein, in step S3, the temperature information is directly extracted from the Brillouin gain spectrum based on the data model, specifically:

S301. Establish a data model, and conduct continuous learning and training to obtain the mapping function relationship between Brillouin frequency shift and temperature; that is, in the actual monitoring process, there is no need to determine the Brillouin frequency shift, and it can be obtained directly through the trained function relationship Temperature information along the fiber.

S302. Through simulation and laboratory calibration, generate L training samples (A _i , T _i ), i=1,2,...,L, where A _i =[a _i (v ₁ ),a _i (v ₂ ) ,...,a _i (v _m )] is the i-th sample, that is, the Brillouin gain spectrum under different frequency sweeps; T _i is the corresponding output vector, that is, the temperature value; the number of hidden layer nodes is N, and the output vector as follows:

In the formula, w _j is the input weight between the input node and the hidden layer node, β _j is the output weight between the hidden layer node and the output node, b _j is the bias of the jth hidden layer node, g(·) is the activation function, preferably, the Sigmoid function is selected as the activation function. Let β=[β ₁ β ₂ …β _N ] ^T , T = [T ₁ T ₂ …T _L ], then the output of hidden layer nodes is:

The output vector is expressed in the form of matrix: Hβ=T;

S303, there are n sampling points along the optical fiber, select the Brillouin gain spectrum at each sampling point of the optical fiber, input it into the trained model for testing, and obtain the corresponding temperature, that is, the temperature information along the optical fiber.

Specifically, the method directly extracts temperature information corresponding to the Brillouin gain spectrum, and can directly obtain temperature information along the optical fiber without determining the value of the Brillouin center frequency. The method effectively reduces the time of data processing while improving the measurement accuracy. The core idea is to establish the mapping function relationship between Brillouin frequency shift spectrum and temperature by learning and training the Brillouin spectrum data collected under different temperature conditions at the beginning of measurement or in the laboratory calibration stage. In the actual measurement process, the temperature information can be output based on the data model obtained through training.

Among them, step S3 also includes optimizing the input weight w and bias b: taking the root mean square error of the prediction result of the training set as the fitness function, setting the goal of parameter optimization to minimize the root mean square error of the prediction result, when Jump out of the loop when the fitness value is the minimum or the maximum number of iterations is met. By optimizing the input weight w and bias b, the prediction accuracy of the model is improved.

Specifically, please refer to Figure 3, the optimization steps are as follows:

Step 1: Generate a training sample data set (A, T) through simulation experiment calibration, A is the Brillouin gain spectrum data as a feature vector; T is the prediction result.

Step 2: Initialize parameters, randomly generate input weight w and bias b, obtain output weight H, and predict the result vector as Hβ=T;

Step 3: Calculate the fitness value and update the parameter value. In this paper, the root mean square error is used as the fitness function;

Step 4: Compare whether the current fitness value is better than the last result, if so, update the parameter value, otherwise keep the parameter value unchanged;

Step 5: If the maximum number of iterations is reached, then end and jump out of the loop to obtain the optimal input weight w and bias b; otherwise, return to step 3.

The optimal input weight w and bias b can be obtained through the above optimization process. Select a certain point z _j of the optical fiber, and the data after noise reduction and spectral shape correction processing are expressed as [f”(v ₁ ,z _j )f”(v ₂ ,z _j )…f”(v _m ,z _j )], input it into a well-trained model for testing, and the corresponding temperature T _j can be obtained. The same is true for other positions, and temperature information monitoring along the optical fiber can be realized.

Embodiment two

Please refer to Fig. 5, a signal processing terminal of a Brillouin optical time domain analysis system, including a memory, a processor, and a computer program stored on the memory and operable on the processor, and the embodiment is realized when the processor executes the computer program Each step in one.

To sum up, the signal processing method and its terminal of a Brillouin optical time domain analysis system provided by the present invention, through two-dimensional variational mode decomposition and reconstruction, and then performing spectral type correction through time slot difference, utilize The mapping function relationship between Brillouin frequency shift and temperature can directly obtain the temperature information along the optical fiber, so as to realize the three functions of noise reduction, spectrum correction and temperature information extraction, and overcome the problems of poor measurement accuracy and long processing time in traditional fitting algorithms. shortcoming.

It should be noted that, for the sake of simplicity of description, the aforementioned method embodiments are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. Because of the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

In the foregoing embodiments, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

The above is only an embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the description of the present invention and the contents of the accompanying drawings, or directly or indirectly used in other related technologies fields, all of which are equally included in the scope of patent protection of the present invention.

Claims (8)

1. A signal processing method of a Brillouin optical time domain analysis system is characterized by comprising the following steps:

s1, acquiring a Brillouin gain spectrum, and performing two-dimensional variational modal decomposition and reconstruction;

s2, performing spectrum type correction through time slot difference, and correcting the distorted Brillouin gain spectrum into a Lorentz spectrum;

and S3, extracting temperature information from the corrected Brillouin gain spectrum based on the data model.

2. The signal processing method of the brillouin optical time domain analysis system according to claim 1, wherein the acquiring the brillouin gain spectrum in step S1 specifically is:

the BOTDA system is obtained through frequency sweeping, M groups of one-dimensional Brillouin gain spectrum signals along the optical fiber are obtained, the one-dimensional Brillouin gain spectrum signals are M columns and n rows of two-dimensional matrix M, and the matrix M comprises local Brillouin gain spectrums at various positions of the optical fiber:

wherein f (v) _i ,z _j ) And (i is more than or equal to 1 and less than or equal to m, and j is more than or equal to 1 and less than or equal to n) represents the Brillouin gain spectrum at the jth sampling point in the ith frequency sweep point and the optical fiber length range.

3. The signal processing method of the brillouin optical time domain analysis system according to claim 2, wherein the performing the two-dimensional variation modal decomposition and reconstruction in the step S1 specifically includes:

f (v) _i ,z _j ) (i is more than or equal to 1 and less than or equal to m, j is more than or equal to 1 and less than or equal to n) is set as f (x), and a constraint variational model is set as follows:

in the formula u _k ＝{u ₁ ,u ₂ ,…,u _k Is the decomposed K mode functions, w _k ＝{w ₁ ,w ₂ ,…,w _k The center frequency is the corresponding center frequency;

introducing a secondary penalty factor and a Lagrange multiplier lambda, and optimizing the constraint variation model by using an alternating direction multiplier method to obtain an unconstrained problem-expanded Lagrange function:

alternate update

Until it is satisfied

Then, the iteration stops and outputs each IMF component:

in the formula, tau is a residual error,

and

each represents u _k And w _k The frequency domain property of (a);

and combining the IMF components obtained by decomposition to obtain a reconstructed signal.

4. The signal processing method of the brillouin optical time domain analysis system according to claim 3, wherein the step S1 further includes performing transform domain noise reduction processing on the signal with low signal-to-noise ratio, specifically:

s101, inputting a two-dimensional Brillouin gain spectrum signal containing noise acquired by a BOTDA system, and setting an initial value of the number of decomposition layers to be 2;

s102, setting a correlation coefficient screening threshold value to be 0.3, calculating a correlation coefficient between IMF1 of modal decomposition and an original signal under a current value, judging whether the correlation coefficient between the IMF1 and the original signal is larger than the threshold value, if so, determining that K = K-1 is the optimal decomposition number, otherwise, K = K +1, and continuing to execute the step S102;

and S103, executing a decomposition process by adopting the value determined in the step S102 to obtain and reconstruct the modal signal.

5. The signal processing method of the brillouin optical time domain analysis system according to claim 1, wherein the step S2 is specifically:

setting the Brillouin gain spectrum after noise reduction as M':

randomly selecting a row from time slot spectrum data which do not carry stimulated Brillouin scattering information, and marking the row as C:

C＝[F(v ₁ ) F(v ₂ ) … F(v _m )]

and (4) subtracting each row of data in the matrix M' from C to obtain corrected Brillouin gain spectrum data in the Lorentz shape.

6. The signal processing method of the brillouin optical time domain analysis system according to claim 1, wherein the step S3 of directly extracting temperature information from the brillouin gain spectrum based on the data model specifically includes:

s301, establishing a data model, and continuously learning and training to obtain a mapping function relation between the Brillouin frequency shift and the temperature;

s302, generating L trainings through simulation and laboratory calibrationExercise sample (A) _i ,T _i ) I =1,2, …, L, where a _i ＝[a _i (v ₁ ),a _i (v ₂ ),…,a _i (v _m )]The ith sample is the Brillouin gain spectrum under different frequency sweeps; t is a unit of _i Is the corresponding output vector, i.e. temperature value; if the number of hidden nodes is N, the output vector is as follows:

in the formula, w _j Is an input weight, beta, between an input node and a hidden node _j As output weights between hidden nodes and output nodes, b _j For the bias of the jth hidden node, g (-) is the activation function, let β = [ β ] ₁ β ₂ … β _N ] ^T ，T＝[T ₁ T ₂ … T _L ]Then the output of the hidden layer node is:

the output vector is expressed in matrix form as: h β = T;

s303, n sampling points are arranged along the optical fiber, a Brillouin gain spectrum at each sampling point of the optical fiber is selected and input into a trained model for testing, and corresponding temperature, namely temperature information along the optical fiber, is obtained.

7. The signal processing method of the brillouin optical time domain analysis system according to claim 6, wherein the step S3 further comprises optimizing the input weight w and the offset b: and taking the root mean square error of the prediction result of the training set as a fitness function, setting the target of parameter optimization to minimize the root mean square error of the prediction result, and jumping out of the loop when the fitness value is minimum or the maximum iteration times are met.

8. A signal processing terminal of a brillouin optical time domain analysis system, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements each step in a signal processing method of a brillouin optical time domain analysis system according to any one of claims 1 to 7 when executing the computer program.

Images