Optical fiber current sensor vibration measurement error compensation method and device based on embedded sensing array

Abstract

The invention discloses a fiber current sensor vibration measurement error compensation method and device based on an embedded sensing array, and belongs to the technical field of fiber current sensor vibration measurement. The device comprises a sensing optical fiber, an optical fiber sheath layer, an MEMS sensing array and an optical fiber ring shell, the method comprises the following steps: constructing a prior knowledge base of multi-azimuth degrees of freedom and vibration errors measured when an optical fiber is subjected to external vibration and impact on the basis of a deep learning theory, obtaining error data under different vibration conditions, and training a deep learning model; the vibration error corresponding to the prior knowledge base is sought by measuring the acceleration induced by external vibration and sensed by the MEMS sensors which are arranged in the sensing optical fiber in an embedded array manner, so that the vibration measurement error compensation of the optical fiber current transformer is realized. The method is simple and easy to implement, the measurement method is accurate and efficient, and a feasible scheme is provided for compensation of measurement errors caused when the optical fiber current transformer suffers from external vibration and impact.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical fiber current sensor vibration measurement, and particularly relates to a vibration measurement error compensation method and device for an optical fiber current sensor based on an embedded sensing array. BACKGROUND

[0002] Optical fiber current sensors (FOCS) are widely used in smart grids and rail transit due to their anti-electromagnetic interference, large dynamic range, excellent insulation performance, and other characteristics.

[0003] However, in a complex mechanical vibration environment (such as a transformer near field, high-voltage switch operation impact, etc.), the sensing optical fiber is affected by factors such as polarization state disturbance caused by vibration, stress birefringence effect superposition, etc., which can cause significant deviation (error up to 1%) in current measurement, severely restricting its application in high-precision measurement scenarios.

[0004] Traditional optical fiber current transformers are easily affected by multi-degree-of-freedom external impact in a vibration environment, resulting in abnormal optical signal phase or polarization state, and thus causing current measurement error. Existing compensation methods rely on a single sensor or a linear calibration model, making it difficult to accurately capture the nonlinear relationship between complex vibration modes and errors.

[0005] In addition, the high sensitivity of the optical fiber sensing system makes error compensation in a dynamic vibration scenario a technical difficulty. SUMMARY

[0006] The present application proposes a vibration error compensation method and device based on an embedded multi-modal sensing array and deep learning to address the measurement error problem caused by polarization state drift of existing optical fiber current sensors in a complex vibration environment. Through multi-dimensional vibration feature extraction and a nonlinear mapping model, high-precision real-time compensation is achieved.

[0007] The present application adopts the following technical solution: a vibration measurement error compensation device for an optical fiber current sensor based on an embedded sensing array, comprising: a sensing optical fiber, an optical fiber sheath layer, a MEMS sensing array, and an optical fiber ring shell.

[0008] The sensing optical fiber is wound into a closed ring-shaped sensing head using a polarization maintaining optical fiber, transmits polarized light, and senses the Faraday phase shift caused by the current magnetic field.

[0009] The optical fiber sheath layer is wrapped outside the sensing optical fiber and is wound in a ring shape to form an optical fiber current sensing ring.

[0010] The MEMS sensing array includes a plurality of MEMS sensors that are distributed at equal intervals in a ring shape and are embedded in a composite buffer material in the optical fiber sheath layer.

[0011] The optical fiber ring shell constitutes the outermost shell structure of the optical fiber current sensor ring.

[0012] Preferably, the optical fiber sheath layer comprises: an anti-kinking layer, a buffer layer for realizing basic anti-vibration of the optical fiber current sensor ring.

[0013] The buffer layer adopts polyurethane-silica gel composite buffer material for inhibiting external stress from transversely extruding the sensing optical fiber.

[0014] The anti-kinking layer adopts aramid fiber woven mesh, which is covered outside the buffer layer, for preventing deformation of the sheath structure when the bending radius of the sensing optical fiber is less than 50 mm.

[0015] Preferably, the composite buffer material of the buffer layer is internally pre-engraved with a spiral groove, and the MEMS sensing array is embedded and packaged in the spiral groove in a spiral structure.

[0016] Preferably, the MEMS sensing array comprises eight triaxial MEMS sensors; each triaxial MEMS sensor comprises a three-dimensional acceleration and a three-dimensional angular acceleration measurement unit, and the spacing ΔL between adjacent MEMS sensors satisfies: ΔL = v / 2f res ;

[0017] Wherein, v is the mechanical vibration wave propagation speed in the optical fiber, f res is the threshold value of the optical fiber resonance frequency.

[0018] The technical scheme of the present application also provides: a vibration measurement error compensation method of a fiber current sensor based on an embedded sensing array, which is applied to any of the vibration measurement error compensation devices of the fiber current sensor and comprises the following specific steps:

[0019] Step 1: In the vibration isolation table environment, the current output amplitude of the vibration measurement error compensation device of the fiber current sensor is measured under the condition of no vibration interference by means of the signal solving unit.

[0020] Step 2: The vibration measurement error compensation device of the fiber current sensor is placed on the vibration excitation platform, constant acceleration vibration and time-varying acceleration vibration in different directions are applied, three-dimensional acceleration, three-dimensional angular velocity and six-channel time-varying vibration signals are collected by each MEMS sensor in the MEMS sensing array, real current and measured current under different vibration environments are obtained, and a constant acceleration vibration measurement error priori knowledge base and a time-varying acceleration vibration measurement error priori knowledge base are formed.

[0021] Step 3: The constant acceleration vibration current measurement compensation amount is calculated according to the real current and the measured current under the constant acceleration vibration environment in different directions; and the time-varying acceleration vibration current measurement compensation amount is calculated according to the real current and the measured current under the time-varying acceleration vibration environment in different directions.

[0022] Step 4, based on the collected six-channel time-varying vibration signals, time-frequency analysis based on Hilbert-Huang transform is carried out to obtain six-channel vibration signal time-frequency diagram;

[0023] Step 5, a current measurement error neural network learning model based on constant acceleration vibration is constructed, the input variable is the three-dimensional acceleration and three-dimensional angular velocity collected in the constant acceleration vibration environment, and the output variable is the constant acceleration vibration current measurement compensation;

[0024] Step 6, a current measurement error deep learning model based on time-varying acceleration vibration is constructed, the input variable is the six-channel vibration signal time-frequency diagram, and the output variable is the time-varying acceleration vibration current measurement compensation;

[0025] Step 7, based on the constant acceleration vibration measurement error priori knowledge base and the time-varying acceleration vibration measurement error priori knowledge base, the current measurement error neural network learning model training and the current measurement error deep learning model training are carried out respectively until the preset convergence condition is reached;

[0026] Step 8: in actual application, according to the acceleration information collected by each MEMS sensor in the MEMS sensing array, it is judged whether the current vibration state is in constant acceleration state or time-varying acceleration state;

[0027] If it is in constant acceleration state, the trained current measurement error neural network learning model is used for error prediction; if it is in time-varying acceleration state, the trained current measurement error deep learning model is used for error prediction;

[0028] Step 9: the predicted error value is fed back to the closed-loop demodulation circuit of the optical current transformer through the FPGA hardware module in real time, the phase compensation parameter is dynamically corrected, and the polarization state drift error caused by vibration is suppressed.

[0029] Preferably, in step 2, different direction constant acceleration vibration is applied to the excitation platform, and the current output amplitude under the action of different direction acceleration and angular acceleration is tested respectively, and the three-dimensional acceleration (a x , a y , a z ) and three-dimensional angular acceleration (α x , α y , α z ) of each MEMS sensor measuring point in the MEMS sensing array are collected; the real current size I (a x , a y , a z , α x , α y , α z ) in the vibration environment is changed, and the corresponding measured current size I m (a xa y ,a z ,α x ,α y ,α z ), forming a constant acceleration vibration measurement error prior knowledge base;

[0030] Applying time-varying acceleration vibration in different directions to the excitation platform, respectively testing the current output amplitude under the action of time-varying vibration signals in different directions, collecting the three-dimensional acceleration and three-dimensional angular velocity of each MEMS sensor measuring point in the MEMS sensing array, obtaining six-channel time-varying vibration signals (a x (t),a y (t),a z (t),α x (t),α y (t),α z (t)); changing the real current size in the vibration environment, obtaining the corresponding measured current size, and forming a time-varying acceleration vibration measurement error prior knowledge base.

[0031] Preferably, eight MEMS sensors are arranged in the MEMS sensing array, and eight measuring points are arranged respectively.

[0032] Eight groups of three-dimensional acceleration and three-dimensional angular velocity are collected under the constant acceleration vibration environment, and 48 vibration acceleration values are obtained and input into the current measurement error neural network learning model.

[0033] Eight groups of six-channel time-varying vibration signals are collected under the time-varying acceleration vibration environment, and 48 vibration signal time-frequency diagrams are obtained and input into the current measurement error deep learning model.

[0034] Compared with the prior art, the above technical scheme has the following technical effects:

[0035] 1. The present application realizes high-precision real-time compensation through multi-dimensional vibration feature extraction and nonlinear mapping model; the device is simple and easy to operate, the measurement method is accurate and efficient, and the present application provides a feasible scheme for the compensation of the measurement error of the optical fiber current transformer caused by external vibration and impact.

[0036] 2. The optical fiber current sensor vibration measurement error compensation method of the present application realizes the fusion of three-dimensional acceleration and angular velocity space-time features through the cooperative optimization of the embedded sensing array and the deep learning model, reduces the current measurement error in the vibration environment, and significantly improves the engineering applicability of the optical fiber current sensor. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The present application is an optical fiber current sensor vibration measurement error compensation method overall flowchart;

[0038] Figure 2 A top view of the vibration measurement error compensation device for a fiber optic current sensor according to the present invention;

[0039] Figure 3 Schematic diagram of the cross section of the optical fiber current sensing loop error compensation device of the present invention;

[0040] Figure 4 This is a structural diagram of the deep learning model for time-varying acceleration vibration current measurement error in the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the application are further elaborated in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments of other researchers in this field on this embodiment fall within the scope of protection of the present invention. At the same time, the step numbers in the embodiments of the present invention are only set for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0042] Example 1

[0043] Provided is a vibration measurement error compensation device for an optical fiber current sensor based on an embedded sensing array. The device comprises: a sensing optical fiber 1, an optical fiber sheath layer 2, a MEMS sensing array 3, and an optical fiber ring housing 4.

[0044] like Figure 2 As shown, the sensing fiber 1 is wound with polarization-maintaining fiber into a closed ring sensor head for transmitting polarized light and sensing the Faraday phase shift caused by the current magnetic field.

[0045] The optical fiber sheath layer 2 is coated on the outside of the sensing optical fiber 1, and plays an anti-vibration effect on the optical fiber ring foundation.

[0046] The MEMS sensor array 3 is distributed at equal intervals in a circular direction and includes a total of 8 MEMS sensors, all of which are embedded in the compliant buffer material in the optical fiber sheath layer.

[0047] The optical fiber ring housing 4 constitutes the outermost housing structure of the optical fiber ring.

[0048] Specifically, if Figure 3 As shown, the optical fiber jacket layer 2 includes an anti-kink layer 2-1 and a buffer layer 2-2.

[0049] Among them, the buffer layer 2-2 adopts polyurethane-silicone composite buffer material with a thickness of 0.6mm, which is used to suppress the lateral extrusion of the optical fiber by external stress; the anti-torsion layer 2-1 adopts aramid fiber woven mesh, which covers the outside of the polyurethane-silicone layer and is used to prevent the sheath structure from deforming when the optical fiber bending radius is less than 50mm.

[0050] In this embodiment, the diameter of the sensing fiber 1 is 125 um, and the sensing fiber 1 is wrapped by the fiber sheath layer 2 on both sides, and the total thickness of the sheath is 0.8 mm. The fiber sheath layer 2 is divided into an inner layer and an outer layer. The outer anti-kinking layer 2-1 has a thickness of 0.2 mm, and the inner buffer layer 2-2 has a thickness of 0.6 mm, which contains silica gel protective material for wear resistance and insulation. A spiral groove 5 is pre-engraved in the sheath, and the groove depth is 0.6 mm, which is used for embedding the MEMS sensing array 3.

[0051] In this embodiment, the MEMS sensing array 3 has eight three-axis MEMS acceleration sensors, which contain three-dimensional acceleration and three-dimensional angular acceleration measurement units, and are embedded in the buffer layer 2-2 of the fiber sheath layer 2 in a spiral structure. The distance ΔL between adjacent sensors satisfies: ΔL = v / 2f res ; wherein v is the mechanical vibration wave propagation speed in the fiber (1200-1500 m / s), and f res is the fiber resonance frequency threshold (calibrated by vibration sweep experiment).

[0052] The sensing fiber 1 is wrapped in the middle of the sheath, and then wound into a sensing ring. Eight 1x1x0.5 mm MEMS acceleration sensors are placed in the spiral groove 5, and the height direction (0.5 mm) is aligned with the groove depth (0.6 mm) to avoid exceeding the total thickness of the sheath. The sensor plane size (1x1 mm 2 ) matches the spiral groove 5 width (≥1.2 mm), ensuring sufficient lateral fixation space.

[0053] Further, the MEMS acceleration sensors are arranged in an array in the ring-shaped sensing ring to measure the acceleration from each direction of the fiber when the fiber is subjected to external vibration and impact, thereby improving the accuracy of the measurement. Since the sensors are arranged in an array in the ring-shaped sensing ring, the distance between adjacent sensors can be determined by the circumference of the sensing ring and the size of the sensor. Assuming that the circumference of the sensing fiber is L, the angle between the MEMS sensor and the sensing fiber is θ, and the size of the sensor is 1x1x0.5 mm, the distance between adjacent MEMS sensors is represented as:

[0054] ΔL = 1 / 8 (L-8 / cosθ)

[0055] In particular, since the size of the MEMS sensor used is very small compared to the length of the sensing fiber, it can be ignored, so the distance between adjacent MEMS sensors is:

[0056] ΔL' = 1 / 8L.

[0057] Embodiment Two

[0058] A method for compensating vibration measurement errors of optical fiber current sensors based on embedded sensor arrays is provided, which is used in combination with the optical fiber current sensor vibration measurement error compensation device described in Example 1. Figure 1 As shown, the following steps are included:

[0059] Step 1: Use the above-mentioned fiber optic current sensor vibration measurement error compensation device and the supporting signal processing unit in a vibration isolation environment to measure the current output amplitude without vibration interference;

[0060] Step 2: Place the fiber optic current sensor vibration measurement error compensation device on the excitation platform, apply fixed acceleration vibrations in different directions to the excitation platform, and test the corresponding current output amplitudes under the action of acceleration magnitudes in different directions and angular acceleration magnitudes in different directions.

[0061] The MEMS sensor array 3 is used to collect the three-dimensional acceleration (a) of 8 three-axis MEMS acceleration sensor measurement points. x ,a y ,a z ) and three-dimensional angular acceleration (α x ,α y ,α z )α z ; Change the actual current size I(a x ,α y ,α z ,α x ,α y ,α z ), obtain the corresponding measured current size I under the vibration environment m (α x ,a y ,α z ,α x ,α y ,α z ), forming a priori knowledge base of constant acceleration vibration measurement error.

[0062] Furthermore, time-varying vibration signals in different directions are applied to the excitation platform, and the corresponding current output amplitudes under the action of time-varying vibration signals in different directions are tested respectively. The three-dimensional acceleration and three-dimensional angular velocity of eight three-axis MEMS acceleration sensor measuring points are collected synchronously using the MEMS sensor array 3 to obtain a six-channel time-varying vibration signal with a sampling frequency of ≥10kHz, which is expressed as: (a x (t),α y (t),a z (t),α x (t),α y (t),α z (t)).

[0063] By changing the actual current magnitude in the above-mentioned vibration environment, the corresponding measured current magnitude in the vibration environment is obtained, and a priori knowledge base of time-varying acceleration vibration measurement error is formed.

[0064] Step 3: Calculate the constant acceleration vibration current measurement compensation based on the real current and measured current under the time-varying acceleration signals in different directions and the angular velocity in different directions; calculate the time-varying acceleration vibration current measurement compensation based on the real current and measured current under the time-varying vibration signals in different directions.

[0065] Step 4: Perform time-frequency analysis based on the Hilbert-Huang transform on the six-channel time-varying vibration signals measured by the eight MEMS sensor arrays to obtain time-frequency graphs of the six-channel vibration signals of the eight MEMS sensor arrays;

[0066] Step 5: Construct a neural network learning model for current measurement error based on constant acceleration vibration. The model input variables are 48 vibration acceleration values, that is, six acceleration values ​​for each of the eight measuring points. The model output is the constant acceleration vibration current measurement error compensation amount.

[0067] Step 6: Construct a deep learning model of current measurement error based on time-varying acceleration vibration. The model input variables are 48 vibration signal time-frequency diagrams formed by eight vibration MEMS sensor arrays, and the model output is the time-varying acceleration vibration current measurement error compensation amount.

[0068] In this embodiment, Figure 4 As shown in the figure, the current measurement error deep learning model adopts a multi-scale CNN-BiLSTM network architecture: a 1D convolutional layer (kernel size 3×1, number of channels 48) extracts the local spatial features of the acceleration signal; a bidirectional LSTM layer (128 hidden units) captures the temporal correlation of the vibration; after feature concatenation, the features are mapped to the error value through a fully connected layer; a spatiotemporal Transformer model: the embedding layer encodes the 6-channel signal into a high-dimensional vector and adds a position code; a multi-head self-attention mechanism calculates the correlation weights of cross-sensor signals; and a feedforward network outputs the error compensation value.

[0069] Step 7: Based on the constructed constant acceleration vibration measurement error prior knowledge base and time-varying acceleration vibration measurement error prior knowledge base, the current measurement error neural network learning model training and the current measurement error deep learning model training are carried out respectively until the preset convergence conditions are reached.

[0070] In this embodiment, the training and testing steps of the current measurement error neural network learning model and the current measurement error deep learning model are as follows:

[0071] (1) Extract the input features and labels of each sample, the input features are 48-dimensional acceleration vectors composed of eight eight three-axis MEMS acceleration sensors, each sensor measures 3-axis acceleration (a x ,a y ,a z ) and three-axis angular acceleration (α x ,α y ,α z ), that is:

[0072] X const =[a x1 ,a y1 ,a z1 ,α x1 ,α y1 ,α z1 ,...,a x8 ,a y8 ,a z8 ,α x8 ,α y8 ,α z8 ] T

[0073] The label is the corresponding current measurement error compensation amount △I const :

[0074] △I const =I measured -I true

[0075] Where I measured , I true are the measured current and true current values;

[0076] (2) Based on the generated time-varying acceleration vibration measurement error prior knowledge base, extract the input features and labels of each sample, the input features are six channels of time-varying vibration signals (a x (t), a y (t), a z (t), α x (t), α y (t), α z (t)) for each MEMS sensor, and perform Hilbert Huang transform (HHT) to generate a time-frequency graph.

[0077] The energy density of the time-frequency graph is calculated as:

[0078]

[0079] Where IMF i is the intrinsic mode function, and the label is the corresponding time-varying current error compensation value △I time , the calculation method is the same as △Iconst .

[0080] (3) For the current measurement error neural network learning model of constant acceleration vibration, the input layer is the 48-dimensional acceleration vector X const , the hidden layer is 3 layers of full connection, and the activation function is ReLU:

[0081] H (I) =ReLU(W (I) H (I-1) +b (I) )I=1,2,3

[0082] Among them, I represents the number of fully connected layers, W (I) 、H (I-1) 、b (I) They represent the weight matrix of layer I, the output (activation value) of layer I-1, and the bias term of layer I respectively.

[0083] The output layer is a linear regression layer, which outputs the compensation value △I′ const :

[0084] ΔI′ const =w out H (3) +b out

[0085] Among them, b out 、w out They represent the bias term of the output layer and the weight vector of the output layer respectively.

[0086] (4) For the deep learning model of current measurement error of time-varying acceleration vibration, the input layer is the time-frequency graph of eight MEMS sensors (size is 64×64, 6 channels), and the convolution layer extracts the spatial features of the time-frequency graph:

[0087] H conv =MaxPool(ReLU(W conv *X time +b conv ))

[0088] Among them, the MaxPool function represents the maximum pooling operation, which is to downsample the feature map output by ReLU and retain the maximum value of the local area, which plays the role of compressing data, highlighting key features, and enhancing robustness. conv 、b conv 、H conv Represent the convolution kernel weight, convolution layer weight, and convolution layer output features respectively, X time Represents the input data, corresponding to the "six-channel vibration signal time-frequency diagram".

[0089] The modeling temporal dependencies of the LSTM layer are as follows:

[0090] h t = LSTM(H conv,t , h t-1 ), t = 1, 2, 3,..., T

[0091] wherein h t-1 , h t represent the hidden state of the previous time LSTM, the hidden state calculated by the current time LSTM, and T represents the total time step of the time series data.

[0092] The full connection layer prediction compensation amount ΔI' is carried out in the output layer time :

[0093] ΔI' time = W FC h T + b FC

[0094] wherein W FC , b FC represent the weight matrix of the full connection layer and the bias term of the full connection layer, and h T represents the feature vector input to the full connection layer.

[0095] (5) Further, the deep learning network model is trained, first, the mean square error (MSE) is combined with L2 regularization:

[0096]

[0097] wherein ΔI, ΔI' represent the real compensation amount (label value) and the model predicted compensation amount, λ represents the regularization coefficient, W represents the learnable parameters of the model, and N represents the number of training samples.

[0098] Then, the Adam algorithm of the optimizer is used for dynamic adjustment:

[0099]

[0100] wherein η is the initial learning rate, ε represents the smoothing term, v t , m represent the second moment of the gradient and the first moment of the gradient, θ t , θ t+1 represent the model parameters at the current time (tth step iteration) and the model parameters at the next time (t+1th step iteration).

[0101] Specifically, in the present embodiment, the initial learning rate is η = 0.001, then the training parameters are defined, the batch size is 32, the training period is 200, and the early stopping method (patience value 20) is adopted.

[0102] (6) Model testing and verification, cross-validation method is adopted, first divide the data set into training set (70%), validation set (15%), test set (15%) three parts. The following two performance indicators are mainly used for verification:

[0103] Root mean square error (RMSE):

[0104] Coefficient of determination (R 2 ):

[0105] Step 8: In practical application, according to the acceleration information collected by the eight measuring points of the MEMS sensor array, it is judged whether the current vibration state is in constant acceleration state or time-varying acceleration state; if it is in constant acceleration state, the trained neural network learning model is used; if it is in time-varying acceleration state, the trained deep learning model is used.

[0106] Step 9: The prediction error value is fed back to the closed-loop demodulation circuit of the optical current transformer in real time through the FPGA hardware module, and the phase compensation parameter is dynamically corrected to suppress the polarization state drift error caused by vibration.

[0107] Specifically, the following functions are realized by hardware description language:

[0108] Vibration state judgment module: real-time monitoring of MEMS sensor signal, constant or time-varying acceleration state is judged.

[0109] Dynamic compensation module: according to the model output △I', adjust the phase compensation parameter φ comp of the closed-loop demodulation circuit:

[0110] φ comp = φ raw +k·△I'

[0111] Wherein, k is the proportional coefficient.

[0112] Finally, it should be pointed out that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A vibration measurement error compensation device for an optical fiber current sensor based on an embedded sensor array, characterized in that: The structure includes: a sensing optical fiber (1), an optical fiber sheath layer (2), a MEMS sensing array (3), and an optical fiber ring shell (4); The sensing optical fiber (1) is wound into a closed ring sensing head using polarization-maintaining optical fiber to transmit polarized light and sense the Faraday phase shift caused by the current magnetic field; The optical fiber sheath layer (2) is wrapped around the outside of the sensing optical fiber (1) and annularly wound to form an optical fiber current sensing ring; The MEMS sensing array (3) includes a plurality of MEMS sensors, which are distributed at equal intervals in a circular direction and embedded in the composite buffer material in the optical fiber sheath layer (2); The optical fiber ring housing (4) constitutes the outermost housing structure of the optical fiber current sensing ring.

2. The optical fiber current sensor vibration measurement error compensation device based on the embedded sensor array according to claim 1, characterized in that: The optical fiber sheath layer (2) comprises: an anti-kink layer (2-1) and a buffer layer (2-2), and is used to achieve basic anti-vibration of the optical fiber current sensing ring; The buffer layer (2-2) is made of a polyurethane-silicone composite buffer material and is used to suppress the lateral extrusion of the sensing optical fiber (1) by external stress; The torsion-resistant layer (2-1) is made of an aramid fiber braided mesh, which covers the outside of the buffer layer (2-2) and is used to prevent the sheath structure of the sensing optical fiber (1) from being deformed when the bending radius is less than 50 mm.

3. The optical fiber current sensor vibration measurement error compensation device based on embedded sensor array according to claim 2, characterized in that: A spiral groove (5) is pre-engraved inside the composite buffer material of the buffer layer (2-2), and the MEMS sensor array (3) is embedded and packaged in the spiral groove (5) in a spiral structure.

4. The optical fiber current sensor vibration measurement error compensation device based on embedded sensor array according to claim 1, characterized in that: The MEMS sensing array (3) includes eight three-axis MEMS sensors; Each triaxial MEMS sensor contains a three-dimensional acceleration and a three-dimensional angular acceleration measurement unit. The distance ΔL between adjacent MEMS sensors satisfies: ΔL=v / 2f res ; Where v is the propagation speed of mechanical vibration wave in optical fiber, f res is the fiber resonance frequency threshold.

5. A method for compensating vibration measurement errors of a fiber optic current sensor based on an embedded sensor array, applied to the device for compensating vibration measurement errors of a fiber optic current sensor according to any one of claims 1 to 4, characterized in that: The specific steps are as follows: Step 1: In a vibration isolation platform environment, using a signal processing unit, measure the current output amplitude of the optical fiber current sensor vibration measurement error compensation device without vibration interference; Step 2: Place the fiber optic current sensor vibration measurement error compensation device on a vibration platform, apply constant acceleration vibration and time-varying acceleration vibration in different directions, collect three-dimensional acceleration, three-dimensional angular velocity and six-channel time-varying vibration signals through each MEMS sensor in the MEMS sensor array, obtain the real current and measured current under different vibration environments, and form a priori knowledge base of constant acceleration vibration measurement error and a priori knowledge base of time-varying acceleration vibration measurement error; Step 3: Calculate the constant acceleration vibration current measurement compensation amount based on the real current and the measured current under the constant acceleration vibration environment in different directions; calculate the time-varying acceleration vibration current measurement compensation amount based on the real current and the measured current under the time-varying acceleration vibration environment in different directions; Step 4: Based on the collected six-channel time-varying vibration signal, perform time-frequency analysis based on Hilbert-Huang transform to obtain a time-frequency diagram of the six-channel vibration signal; Step 5: Construct a current measurement error neural network learning model based on constant acceleration vibration, where the input variables are the three-dimensional acceleration and three-dimensional angular velocity collected under the constant acceleration vibration environment, and the output variable is the constant acceleration vibration current measurement compensation amount; Step 6: Construct a deep learning model of current measurement error based on time-varying acceleration vibration, with the input variable being the six-channel vibration signal time-frequency diagram and the output variable being the time-varying acceleration vibration current measurement compensation amount; Step 7: Based on the constant acceleration vibration measurement error prior knowledge base and the time-varying acceleration vibration measurement error prior knowledge base, respectively, perform current measurement error neural network learning model training and current measurement error deep learning model training until a preset convergence condition is reached; Step 8: In actual application, based on the acceleration information collected by each MEMS sensor in the MEMS sensor array, determine whether the current vibration state is in a constant acceleration state or a time-varying acceleration state; If it is in a constant acceleration state, the trained current measurement error neural network learning model is used to predict the error; If it is in a time-varying acceleration state, the trained current measurement error deep learning model is used for error prediction; Step 9: Feedback the predicted error value to the closed-loop demodulation circuit of the optical fiber current transformer in real time through the FPGA hardware module to dynamically correct the phase compensation parameters and suppress the polarization state drift error caused by vibration.

6. The method for compensating vibration measurement errors of an optical fiber current sensor based on an embedded sensor array according to claim 5, characterized in that: In step 2, constant acceleration vibrations in different directions are applied to the excitation platform, and the current output amplitudes under accelerations in different directions and angular accelerations are tested respectively. The three-dimensional accelerations (a x ,a y ,a z ) and three-dimensional angular acceleration (α x ,α y ,α z ); change the actual current size I(a x ,a y ,a z ,α x ,α y ,α z ), obtain the corresponding measured current size I m (a x ,a y ,a z ,α x ,α y ,α z ), forming a priori knowledge base of constant acceleration vibration measurement error; Apply time-varying acceleration vibrations in different directions to the excitation platform, test the current output amplitude under the action of time-varying vibration signals in different directions, collect the three-dimensional acceleration and three-dimensional angular velocity of each MEMS sensor measurement point in the MEMS sensor array, and obtain a six-channel time-varying vibration signal (a x (t),a y (t),a z (t),α x (t),α y (t),α z (t)); changing the actual current size in the vibration environment, obtaining the corresponding measured current size, and forming a priori knowledge base of time-varying acceleration vibration measurement error.

7. The method for compensating vibration measurement errors of an optical fiber current sensor based on an embedded sensor array according to claim 6, wherein: The MEMS sensing array is provided with eight MEMS sensors, each of which has eight measuring points; Collect eight sets of three-dimensional acceleration and three-dimensional angular velocity in a constant acceleration vibration environment to obtain 48 vibration acceleration values, which are input into the current measurement error neural network learning model; Eight groups of six-channel time-varying vibration signals were collected in a time-varying acceleration vibration environment to obtain 48 vibration signal time-frequency graphs, which were input into the current measurement error deep learning model.

8. The method for compensating vibration measurement errors of an optical fiber current sensor based on an embedded sensor array according to claim 7, wherein: The current measurement error neural network learning model in step 5 has an input layer of 48-dimensional acceleration vector X const , the hidden layer H is a 3-layer fully connected layer, and the activation function is ReLU: A (I) =ReLU(W (I) A (I-1) +b (I) ) I=1,2,3 Among them, I represents the number of fully connected layers, W (I) 、H (I-1) 、b (I) Represent the weight matrix of layer I, the output of layer I-1, and the bias term of layer I respectively; The output layer is a linear regression layer, which outputs the compensation value △I' const : ΔI′ const =w out H (3) +b out Among them, b out 、w out They represent the bias term of the output layer and the weight vector of the output layer respectively.

9. The method for compensating vibration measurement errors of an optical fiber current sensor based on an embedded sensor array according to claim 7, wherein: The current measurement error deep learning model described in step 6 adopts a multi-scale CNN-BiLSTM network architecture, including: 1D convolution layer, with a kernel size of 3×1 and 48 channels, extracts the local spatial features of the six-channel vibration signal time-frequency graph: H conv =MaxPool(ReLU(W conv *X time +b conv )) Among them, the MaxPool function represents the maximum pooling operation, W conv 、b conv 、H conv Represent the convolution kernel weight, convolution layer weight, and convolution layer output features respectively, X time Represents the input data, corresponding to the six-channel vibration signal time-frequency diagram; Bidirectional LSTM layer with 128 hidden units to capture the temporal correlation of vibrations: h t =LSTM(H conv,t ,h t-1 ),t=1,2,3…,T Among them, h t-1 、h t They represent the hidden state of LSTM at the previous moment and the hidden state calculated by LSTM at the current moment, respectively. T represents the total number of time steps of the time series data. Output layer, after feature splicing, is mapped to the error value through the fully connected layer to obtain the predicted compensation △I' time : △I' time =W FC h T +b FC Among them, W FC 、b FC Represent the weight matrix of the fully connected layer and the bias term of the fully connected layer, h T Represents the feature vector input to the fully connected layer; The spatiotemporal Transformer model encodes the six-channel signal into a high-dimensional vector through an embedding layer and adds positional encoding; Multi-head self-attention mechanism to calculate the correlation weights of cross-sensor signals; Feedforward network, outputs error compensation value.

10. The method for compensating vibration measurement errors of an optical fiber current sensor based on an embedded sensor array according to claim 7, wherein: In step 7, current measurement error neural network learning model training and current measurement error deep learning model training are performed, and the method includes: Step 7.1: Use the mean square error combined with L2 regularization to train the model: Where △I and △I' represent the actual compensation amount and the compensation amount predicted by the model, respectively, λ represents the regularization coefficient, W represents the learnable parameter of the model, and N represents the number of training samples; Step 7.2: Use the Adam algorithm of the optimizer to make dynamic adjustments: Among them, η is the initial learning rate, ε represents the smoothing term, and v t , m represent the second-order moment of gradient and the first-order moment of gradient respectively, θ t ,θ t+1 Represent the model parameters at the current moment and the next moment respectively; Step 7.3: Define the training parameters, batch size, and training period, use the early stopping method, and iterate until the preset convergence condition.

Images