Electromagnetic data processing method and system based on gated recurrent unit

Abstract

The application relates to the technical field of electromagnetic data processing, and particularly discloses an electromagnetic data processing method and system based on a gated recurrent unit, which comprises the following steps: step S1: segmenting electromagnetic data according to a sampling rate, and constructing an electromagnetic data sample library; step S2: inputting the sample library into a gated recurrent unit for data training, and generating a data-driven model; step S3: testing electromagnetic actual measurement data by using the data-driven model, dividing effective signals into a first type with a label of 1, and dividing noises into a second type with a label of 2; and step S4: removing data with the label of 2, and retaining data with the label of 1, thereby reconstructing high-quality electromagnetic data, and solving the problems that traditional electromagnetic data is susceptible to noise interference, collected signals are fuzzy with noises, and effective identification cannot be achieved.

Description

Technical Field

[0001] This application relates to the field of electromagnetic data processing technology, and specifically discloses an electromagnetic data processing method and system based on a gated loop unit. Background Technology

[0002] Electromagnetic methods are an effective means of Earth exploration. They mainly include natural source electromagnetic methods (MT, AMT, etc.) and controlled source electromagnetic methods (CSAMT, TEM, etc.). By studying the electrical properties of underground media, they help to explore various resources and energy sources, and solve various hydrological, environmental, and engineering geological problems, making a significant contribution to socio-economic development. However, electromagnetic methods are facing increasing noise. Noise and other abnormal interferences are unavoidable during data acquisition, severely impacting data quality. In particular, the complex distribution of noise in the time and frequency domains often blurs the signal and noise, making effective identification difficult. This presents a technical bottleneck for electromagnetic data processing.

[0003] Researchers both domestically and internationally have studied numerous denoising methods, but some problems remain to be solved regarding the extraction of effective signals in electromagnetic methods. To address the issue of noise interference in wide-area electromagnetic signal processing, a number of new data processing methods have emerged. These methods can improve data quality from different perspectives. If we can start with the time-domain sequence of signal and noise, we can further enhance the high-precision Earth detection capabilities of subsequent electromagnetic methods.

[0004] Therefore, in view of this, the inventors have provided an electromagnetic data processing method and system based on a gated loop unit to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that traditional electromagnetic data is easily affected by noise, resulting in blurred signals and noise that cannot be effectively identified.

[0006] To achieve the above objectives, the basic solution of the present invention provides an electromagnetic data processing method based on a gated loop unit, comprising the following steps:

[0007] Step S1: Divide the electromagnetic method data into segments according to the sampling rate and construct an electromagnetic method data sample library;

[0008] Step S2: Input the electromagnetic method data sample library into the gated loop unit for data training to generate a data-driven model;

[0009] Step S3: Use the data-driven model to test the electromagnetic method measured data, classify the effective signal and noise, and classify the effective signal as the first category and the noise as the second category;

[0010] Step S4: the data of the noise classified as the second type is removed, and the data of the effective signal classified as the first type is reserved, so as to reconstruct the electromagnetic method data with high quality.

[0011] Further, in step S1, the electromagnetic method data sample library includes effective signals, attenuated noises, impulse noises and trend noises.

[0012] In step S3, when the classification of the effective signal and the noise is performed, the data is also labeled, wherein the effective signal labels are all 1, and the labels of the remaining electromagnetic method data samples are all 2.

[0013] Further, the data-driven model contains the sample library signal-to-noise categories and data labels.

[0014] The basic scheme of the present application also provides an electromagnetic method data processing system based on the loop unit, which comprises at least one module, wherein the module is used for executing any one of the methods described above.

[0015] Further, the module is a gated loop unit module.

[0016] The basic scheme of the present application also provides a gated loop unit, which is used for implementing step S2 in the above method, and comprises a reset gate, an update gate, a candidate hidden state and a hidden state, and the internal activation function comprises a sigmoid function and a tanh function.

[0017] The reset gate controls the neuron state vector of the previous moment at this selected moment, and determines the usage amount of the historical information in the hidden state vector at this selected moment.

[0018] The update gate controls the cell state of the hidden state vector at this selected moment.

[0019] Further, at a moment t, the external input X t At t-1, the neuron state h t-1 is inputted, and the combination forms (h t-1 ,x t ), (h t-1 ,x t ) is the input vector X t of the tth time step, that is, the tth component of the input sequence X, h t-1 saves the information of the previous time step t-1, and the external input X t refers to the signal in the electromagnetic method data sample library.

[0020] The reset gate r t controls the neuron state vector h t-1, determines how much historical information will be used in the hidden state vector of the neuron at time t

[0021] The activation function [0, 1] has a value range of [0, 1] update gate Z t Control h t-1 And The cell state h t of the output neuron at time t t And is calculated by the following formula:

[0022]

[0023] In the formula, ω r And ω z Respectively represent the weight vector of r z And z t , b r And b z Respectively represent the bias vector of r z And z t , sigma (·) represents the sigmoid function, and * represents matrix multiplication operation;

[0024] The hidden state of the neuron Can be represented as:

[0025]

[0026] In the formula, ω h Is the weight vector of the hidden state , b h Is the hidden state, h t Is the bias vector, Is the matrix element multiplication operation, then the cell state h t Of the neuron at time tCan be represented as:

[0027]

[0028] The principle and effect of the scheme are that:

[0029] The present application is implemented by steps in sequence, realizes effective identification of signals and noises in intelligent controllable source electromagnetic data, improves signal-to-noise separation effect, precision and data quality, solves the problem that the neural network algorithm in the prior art cannot realize accurate identification and separation of electromagnetic data, and improves the application effect of the neural network algorithm on electromagnetic data processing. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0031] Figure 1 A flow chart of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application;

[0032] Figure 2 A schematic diagram of a sample library of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application;

[0033] Figure 3 A schematic diagram of a gated recurrent unit of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application;

[0034] Figure 4 A sample library signal-to-noise category and data label effect diagram of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application;

[0035] Figure 5 A processing effect diagram of an analog signal of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application;

[0036] Figure 6 A processing effect diagram of measured data of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application;

[0037] Figure 7 A comparison effect diagram of electric field curves before and after processing of a method and system for processing electromagnetic data based on a gated recurrent unit is shown in the embodiments of the present application. DETAILED DESCRIPTION

[0038] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purposes, the specific embodiments, structures, features and effects according to the present application will be described in detail below in combination with the drawings and preferred embodiments.

[0039] A method and system for processing electromagnetic data based on a gated recurrent unit, as shown in the embodiments of the present application, Figure 1 as shown:

[0040] Embodiment 1

[0041] A method and system for processing electromagnetic data based on a gated recurrent unit, comprising the following steps:

[0042] Step S1: segmenting the electromagnetic data according to the sampling rate to construct an electromagnetic data sample library;

[0043] In this step, the constructed electromagnetic data sample library contains effective signals, decay noise, impulse noise, and trend noise, and the effective signals, decay noise, impulse noise, and trend noise are labeled as label data, wherein the labels of the effective signals are all 1, and the labels of the decay noise, impulse noise, and trend noise are all 2;

[0044] Step S2: inputting the electromagnetic data sample library into a gated recurrent unit for data training to generate a data-driven model;

[0045] Step S3: testing the actual electromagnetic data using the data-driven model, dividing the effective signals into a first category with a label of 1, and dividing the decay noise, impulse noise, and trend noise into a second category with a label of 2;

[0046] Step S4: discarding the data with a label of 2 and retaining the data with a label of 1 to reconstruct high-quality electromagnetic data.

[0047] Through step S3, intelligent signal-noise processing of electromagnetic data can be achieved.

[0048] Through the above steps, effective identification of signals and noise in intelligent controllable source electromagnetic data is achieved, the signal-noise separation effect, precision, and data quality are improved, the problem of inaccurate identification and separation of electromagnetic data by neural network algorithms in the prior art is solved, and the application effect of neural network algorithms for processing electromagnetic data is improved.

[0049] In steps S2 and S3, the generated data-driven model contains sample library signal-noise categories and data labels.

[0050] Embodiment 2 is a specific implementation process of the method, specifically,

[0051] As shown in Figure 1 , the embodiment discloses a method and system for processing electromagnetic data based on a gated recurrent unit, comprising the following:

[0052] Step S1: segmenting the electromagnetic data according to the sampling rate to construct an electromagnetic data sample library, the sample library in this step contains effective signals, decay noise, impulse noise, trend noise, and label data, wherein the labels of the effective signals are all 1, and the labels of the remaining noise are all 2;

[0053] In this embodiment, Figure 2 is a schematic diagram of the constructed electromagnetic data sample library. Figure 2The sample contains four different types of signals: pseudo-random valid signals, signals with impulse noise, signals with attenuation noise, and signals with noisy trend noise. The pseudo-random valid signals are labeled as 1, and the other three types of noise are labeled as 2.

[0054] Step S2: Input the sample library into the gated recurrent unit for data training to generate a data-driven model;

[0055] The gated loop unit includes a reset gate, an update gate, a candidate hidden state, and a hidden state. The internal activation functions include the sigmoid function and the tanh function.

[0056] The sigmoid function can be approximated as a 0 / 1 step function, so the real value output of sigmoid is monotonically differentiable and can be updated based on error backpropagation.

[0057] The hidden state has the following advantages:

[0058] Reset gates help capture short-term dependencies in a sequence;

[0059] Update gates help capture long-term dependencies in a sequence.

[0060] Specifically, the following is a set of examples of gated loop units:

[0061] At a certain time t, the external input X t At time t-1, the neuron state h t-1 The input to the neuron is combined to form (h) t-1 ,x t ). (h t-1 ,x t Let X be the input vector at time step t. t That is, the t-th component of the input sequence X undergoes a linear transformation. t-1 It stores information from the previous time step t-1, which also undergoes a linear transformation. Furthermore, the external input X of the gated loop unit... t It refers to the signals in the electromagnetic method data sample library.

[0062] Reset door r t The neuron state vector h at time t-1 was controlled t-1 This determines how much historical information will be in the hidden state vector of the neuron at time t. Used in.

[0063] The activation function [0,1] has a range of [0,1], and the update gate Z... t h is controlled by a weighted average. t-1 and The cell state h of the output neuron at time t t . Wherein, r t and are calculated by the following formula:

[0064]

[0065] In the formula, ω r and ω z respectively represent the weight vectors of r z and z t , b r and b z respectively represent the bias vectors of r z and z t , σ(·) represents the sigmoid function, and * represents the matrix multiplication operation.

[0066] The hidden state of the neuron can be represented as:

[0067]

[0068] In the formula, ω h is the weight vector of the hidden state , b h is the hidden state, h t is the bias vector, is the matrix element multiplication operation. Then the cell state h t of the neuron at time t can be represented as:

[0069]

[0070] In this embodiment, Figure 3 a schematic diagram of a gated recurrent unit is shown, wherein ⊕ represents an addition operation.

[0071] The input sample is a new X t formed by external input combination (h t-1 , x t ), and the output is the cell state h t of the neuron. Based on the input sample and the output of the gated recurrent unit, a data-driven model will be automatically generated for identifying signals (label 1) and noises (label 2).

[0072] Step S3: testing the electromagnetic field measured data by using the data-driven model; in this step, the effective signal is divided into the first class with label 1, and the noise is divided into the second class with label 2, so as to realize intelligent signal-noise processing of electromagnetic data;

[0073] The data-driven model contains sample library signal-noise categories and data labels, and the data-driven model is a generated model based on the input sample and the output of the gated recurrent unit.

[0074] In this embodiment, Figure 4 The sample library signal-to-noise category and data label effect diagram are shown.

[0075] Step S4: The data labeled as 2 is removed, the data labeled as 1 is retained, and high-quality electromagnetic method data is obtained by reconstruction.

[0076] In order to verify the effectiveness of the effect of this embodiment, the method of the application is used for time and frequency domain processing comparison and analysis on simulated noise-containing data and measured data. Figure 5 The processing effect diagram of the simulated signal according to the embodiment of the application by the method of the application is shown; analysis Figure 5 It can be seen that the simulated pseudo-random signal is added with pulse noise, attenuation noise and noise-containing trend noise, so that the original effective signal is distorted, and the noise in the spectrum also corresponds to the signal being submerged. Table 1 shows the comparison of the electric field amplitude before and after the simulated signal is processed by the method of the application. Figure 6 The processing effect diagram of the measured data according to the embodiment of the application by the method of the application is shown; analysis Figure 6 It can be seen that the measured data is affected by typical noise, so that the data appears abnormal waveform. Comprehensive analysis Figure 5 and Figure 6 It can be shown that the method of this embodiment can effectively realize the signal-to-noise processing of electromagnetic method data, and effectively identify and predict noise data, obtain high-quality electromagnetic method reconstruction data, and improve the quality of the original data.

[0077] Table 1 Comparison of simulated signal before and after processing by the method of the application

[0078]

[0079]

[0080] The effect of the electric field curve before and after processing is evaluated by comparing the original data and the electric field curve after processing by the method of this embodiment. As Figure 7 shown, Figure 7 is the effect diagram of the electric field curve of the measured point before and after processing. Due to the influence of the original electromagnetic data in the time domain by the typical noise, the electric field curve appears abnormal fluctuation, and the quality of the electric field curve is poor; it can be found that after the method of this embodiment is processed, the electric field curve of the measured point is more stable and has no abnormal fluctuation, which shows that the quality of the electromagnetic method data is effectively improved.

[0081] In this embodiment, an electromagnetic method data processing system based on a gating recurrent unit is provided, which includes at least one module, wherein the at least one module is used to execute an electromagnetic method data processing method based on a gating recurrent unit in this embodiment, and the at least one module used to implement the method is a gating recurrent unit module.

[0082] In the above embodiment, the equal sampling rate period segmentation of electromagnetic data is utilized, the sample library is constructed, the intelligent recognition of the electromagnetic method data by the gated recurrent unit is realized, the electromagnetic method high-quality data is reconstructed by combining and recognizing the effective signal data. By utilizing the gated recurrent unit algorithm, simple operation and accurate classification are realized. Through the analysis of the experimental results, the complexity and application effect of the existing neural network algorithm in the electromagnetic method data processing are improved, and the application value and innovation are certain.

[0083] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make slight changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the technical solution range of the present application. Any modification, equivalent change and modification of the above embodiments made according to the technical essence of the present application still belong to the scope of the technical solution of the present application.

Claims

1. A method for electromagnetic data processing based on gated loop units, characterized in that: Includes the following steps: Step S1: Divide the electromagnetic method data into segments according to the sampling rate and construct an electromagnetic method data sample library; Step S2: Input the electromagnetic method data sample library into the gated loop unit for data training to generate a data-driven model; Step S3: Use the data-driven model to test the electromagnetic method measured data, classify the effective signal and noise, and classify the effective signal as the first category and the noise as the second category; Step S4: Remove the noise data classified as the second category and retain the effective signal data classified as the first category to reconstruct and obtain high-quality electromagnetic data; In step S1, the electromagnetic method data sample library includes effective signals, attenuation noise, impulse noise, and trend noise; In step S3, when dividing the effective signal and noise, the data is also labeled. The effective signal is labeled with 1, and the other electromagnetic data samples are labeled with 2. The gated loop unit includes a reset gate, an update gate, a candidate hidden state, and a hidden state. The internal activation functions include the sigmoid function and the tanh function. The reset gate controls the neuron state vector of the previous time step at this selected time step, and determines the amount of historical information used in the hidden state vector at this selected time step. The update gate controls the hidden state vector at the cell state selected at this time.

2. The electromagnetic data processing method based on a gated loop unit according to claim 1, characterized in that, The data-driven model includes signal-to-noise categories and data labels in the sample library.

3. The electromagnetic data processing method based on a gated loop unit according to claim 1, characterized in that, At a certain time t, the external input X t At time t-1, the neuron state h t-1 The input to the neuron is combined to form (h) t-1 ,x t ), (h t-1 ,x t Let X be the input vector at time step t. t That is, the t-th component of the input sequence X, h t-1 It stores the information from the previous time step t-1, with external input X. t This refers to the signals in the electromagnetic method data sample library; Reset door r t The neuron state vector h at time t-1 was controlled t-1 This determines how much historical information will be in the hidden state vector of the neuron at time t. Used in; The activation function has a range of [0,1] and the update gate Z is [0,1]. t h is controlled by a weighted average. t-1 and Cell state h of the output neuron at time t t , where r t and Z t The following formula is used to calculate: In the formula, ω r and ω z They represent r respectively z and z t The weight vector, b r and b z They represent r respectively z and z t The bias vector, σ(·) represents the sigmoid function, and * represents matrix multiplication; The hidden state of the neuron Represented as: In the formula, ω h Hidden state The weight vector, b h To be in a hidden state, h t For bias vectors, If we perform element-wise multiplication of matrices, then the cellular state h of the neuron at time t is... t Represented as:

4. An electromagnetic data processing system based on a gated loop unit, characterized in that, It includes at least one module, wherein the module is used to perform the method according to any one of claims 1 to 2.

5. The electromagnetic data processing system based on a gated loop unit according to claim 4, characterized in that, The module is a gated loop unit module.

Images