Electromagnetic method signal-to-noise discrimination method and system based on extreme learning machine

Abstract

The application relates to the technical field of controllable source electromagnetic methods, and particularly discloses an electromagnetic method signal-noise identification method and system based on an extreme learning machine, which divides electromagnetic method data into equal-interval segments, extracts multi-domain multi-feature parameters of each segment of electromagnetic method data respectively, carries out cluster analysis on the multi-domain multi-feature parameters in a sample library, performs priority sorting on the feature parameters and a known sample library category, takes an inter-class distance as a judgment index, determines optimal and suboptimal feature parameter combinations, trains the extreme learning machine by using the optimal and suboptimal feature parameters of each segment of electromagnetic method data, and then intelligently identifies electromagnetic data signal-noise; the data identified as noise is directly eliminated, effective signals are reserved and combined, and high-quality controllable source electromagnetic data is obtained. The application improves the intelligent identification effect of a machine learning algorithm on electromagnetic methods and improves electromagnetic method data quality.

Description

Technical Field

[0001] This application relates to the field of controllable source electromagnetic method technology, and specifically discloses an electromagnetic signal-to-noise identification method and system based on extreme learning machine. Background Technology

[0002] The national strategic need for efficient and low-cost mineral resource development poses a severe challenge to geophysical exploration technology. Electromagnetic exploration is an important method in geophysical exploration. Electromagnetic methods 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, various resources and energy sources are explored, and various hydrological, environmental, and engineering geological problems are solved, making a significant contribution to socio-economic development.

[0003] Among them, the controlled-source electromagnetic method is simpler to use than the natural-source electromagnetic method because its emitted signal is known; however, under normal circumstances, the collected electromagnetic data cannot avoid abnormal interference such as noise. These interferences are significantly affected in both the time and frequency domains, exhibiting uncertainty and randomness. In particular, the noise distribution in the frequency spectrum is intricate and complex, and even co-frequency interference occurs, often without any discernible pattern, posing a great challenge to noise suppression.

[0004] Since the introduction of the controlled-source electromagnetic method, scholars both domestically and internationally have studied various denoising methods, which have suppressed noise and improved data quality to a certain extent. Addressing the issue of noise affecting wide-area electromagnetic signals in the controlled-source electromagnetic method, domestic scholars have also continuously proposed new signal-to-noise separation methods. These methods all process data from different perspectives. If they can start from the essential characteristics of signal and noise, they will play a crucial role in achieving high-precision signal-to-noise separation in subsequent electromagnetic methods, improving the shortcomings of time-domain processing techniques in noise suppression and preserving a richer amount of effective signal to the maximum extent.

[0005] Therefore, in view of this, the inventors have provided an electromagnetic signal-to-noise identification method and system based on extreme learning machine to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to solve the problem that traditional machine learning algorithms cannot achieve high-precision signal-to-noise separation for controllable source electromagnetic data.

[0007] To achieve the above objectives, the basic solution of this invention provides an electromagnetic signal-to-noise identification method based on extreme learning machine, comprising the following steps:

[0008] Step S1: Divide the electromagnetic method data into equally spaced segments, extract the multi-domain multi-feature parameters of each segment of electromagnetic method data and store them in the sample database;

[0009] Step S2: Perform cluster analysis on the multi-domain and multi-feature parameters in the sample database data, sort the feature parameters according to their priority with the known sample database categories, use the inter-class distance as the evaluation index, and determine the optimal and second-best combination of feature parameters.

[0010] Step S3: Train the extreme learning machine using the optimal and suboptimal combination of feature parameters of each electromagnetic data segment, and then perform signal-to-noise intelligent identification on the electromagnetic data. In the process of signal-to-noise intelligent identification, pseudo-random signals are identified as valid signals, and obviously abnormal waveforms are identified as noise.

[0011] Step S4: Directly remove the data identified as noise in step S3, retain the valid signals and merge them to obtain high-quality controllable source electromagnetic data.

[0012] Furthermore, in step S1, when dividing the electromagnetic data into equally spaced segments, the sampling rate is used as the number of segments.

[0013] Furthermore, in step S1, the multiple domains are time domain, frequency domain, and time-frequency domain, and the multiple feature parameters are pulse factor, peak factor, sample entropy, fuzzy entropy, average frequency, waveform factor, wavelet energy entropy, and wavelet singular entropy.

[0014] Furthermore, in step S2, fuzzy C-means clustering is used to perform cluster analysis, prioritizing the known sample database data categories and the signal-to-noise categories represented by multi-domain multi-feature parameters, and determining the optimal and second-best combination of feature parameters.

[0015] The basic solution of the present invention also provides an electromagnetic signal-to-noise identification system based on extreme learning machine, which further includes at least one module, wherein the module is used to perform any of the methods described above.

[0016] Furthermore, the module is an Extreme Learning Machine module.

[0017] The basic solution of this invention also provides an Extreme Learning Machine (ELM), which is used to implement step S3 in the above method. The ELM includes an input layer, a hidden layer, and an output layer. The input layer and the output layer are provided with multiple neuron nodes. The input layer is used to receive data from the combination of feature parameters as training samples and pass the training samples to the hidden layer. The hidden layer assigns a center to each neuron node, and the number of neurons in the hidden layer is equal to the number of training samples. The output layer calculates the distance between the input and the center and outputs a scalar value, which is the discrimination result of signal-to-noise identification.

[0018] Furthermore, the connection weights between the input layer and the hidden layer, and the threshold of the hidden layer can be set randomly, and once set, they do not need to be adjusted. The connection weights between the hidden layer and the output layer do not need to be iteratively adjusted, but are determined directly by solving the data.

[0019] Furthermore, when there are N arbitrary samples (X) i ,t i It has n input layer neurons and m output layer neurons;

[0020] Among them, input data X i =[x i1 ,x i2 ,...,x in ] T ∈R n Target output data t i =[t i1 ,t i2 ,...,t im ] T ∈R m The output expression for a single-hidden-layer neural network with L hidden layer nodes is:

[0021]

[0022] In the formula, g(·) is the activation function, and W i =[w i,1 ,w i,2 ,...,w i,n ] T For the input weights, β i For the output weights, b i W is the bias of the i-th hidden layer unit. i ·X j W i and X j The inner product;

[0023] Among them, X j and X i Both represent input samples, where j and i are both references to the quantity, and O j This represents the model output of a single hidden layer neural network.

[0024] The goal of learning a single-hidden-layer neural network is to minimize the error of the output, which can be expressed as:

[0025]

[0026] Random initialization W i By combining these, we can obtain the hidden layer output matrix H:

[0027] Hβ=T

[0028] In the formula, H represents the output of the hidden layer node, β is the output weight, and T is the expected output, which is solved using the generalized inverse of the matrix:

[0029]

[0030] In the formula, Let H be the generalized inverse matrix.

[0031] The principle and effect of this solution are as follows:

[0032] This invention divides electromagnetic method data into equally spaced segments and extracts multi-domain, multi-feature parameters from each segment. It then performs cluster analysis on these multi-domain, multi-feature parameters from the sample database, prioritizing them according to their relationship to known sample database categories. The inter-class distance is used as the evaluation metric to determine the optimal and second-best combinations of feature parameters. An extreme learning machine is trained using the optimal and second-best feature parameters from each segment to perform intelligent signal-to-noise identification on the electromagnetic data. Data identified as noise is directly removed, while valid signals are retained and merged to obtain high-quality, controllable-source electromagnetic data. This application improves the effectiveness of machine learning algorithms for intelligent electromagnetic method identification and enhances the quality of electromagnetic method data. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 The flowchart illustrates the steps of an electromagnetic signal-to-noise identification method and system based on extreme learning machine proposed in an embodiment of this application.

[0035] Figure 2 A schematic diagram of the sample library of an electromagnetic signal-to-noise identification method and system based on extreme learning machine proposed in an embodiment of this application is shown.

[0036] Figure 3 The figure shows the feature distribution of samples of an electromagnetic signal-to-noise identification method and system based on extreme learning machine proposed in an embodiment of this application.

[0037] Figure 4 The diagram illustrates the known categories of the sample library and the prediction results of an electromagnetic signal-to-noise identification method and system based on extreme learning machine, as proposed in an embodiment of this application.

[0038] Figure 5 The diagram shows the analog signal processing effect and spectrum effect of an electromagnetic signal-to-noise identification method and system based on extreme learning machine proposed in this application.

[0039] Figure 6The diagram shows the measured data processing effect of an electromagnetic signal-to-noise identification method and system based on extreme learning machine proposed in this application.

[0040] Figure 7 The diagram shows the effect of electric field curve processing before and after processing of an electromagnetic signal-to-noise identification method and system based on extreme learning machine proposed in this application. Detailed Implementation

[0041] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0042] An electromagnetic signal-to-noise identification method and system based on extreme learning machine, Example 1:

[0043] like Figure 1 As shown:

[0044] The electromagnetic signal-to-noise identification method based on extreme learning machine includes the following steps:

[0045] Step S1: Divide the electromagnetic method data into equally spaced segments, extract the multi-domain multi-feature parameters of each segment of electromagnetic method data and store them in the sample database;

[0046] In this step, the sampling rate is used as the number of segments when dividing the electromagnetic data into equally spaced segments.

[0047] In this step, the extracted multi-domains are time domain, frequency domain, and time-frequency domain; the extracted multi-feature parameters are impulse factor, peak factor, sample entropy, fuzzy entropy, average frequency, waveform factor, wavelet energy entropy, and wavelet singular entropy.

[0048] Step S2: Perform cluster analysis on the multi-domain and multi-feature parameters in the sample database data, sort the feature parameters according to their priority with the known sample database categories, use the inter-class distance as the evaluation index, and determine the optimal and second-best combination of feature parameters.

[0049] In this step, the cluster analysis uses the fuzzy C-means clustering method.

[0050] Step S3: Train the extreme learning machine by combining the optimal and suboptimal feature parameters of each electromagnetic data segment, and then perform signal-to-noise intelligent identification on the electromagnetic data. In the process of signal-to-noise intelligent identification, pseudo-random signals are identified as valid signals, and obviously abnormal waveforms are identified as noise.

[0051] Step S4: Directly remove the data identified as noise in step S3, retain the valid signals and merge them to obtain high-quality controllable source electromagnetic data.

[0052] The above steps solve the problem that existing machine learning algorithms cannot achieve high-precision signal-to-noise separation for controllable source electromagnetic data.

[0053] This method extracts signals and noise from multiple domains and features, uses the inter-class distance comparison feature parameter of fuzzy C-means clustering to characterize the priority of signal-to-noise ratio, and combines the extreme learning machine algorithm to train optimal and suboptimal features. It intelligently identifies and predicts effective signals and noise in controllable source electromagnetic data, making up for the shortcomings of existing electromagnetic signal-to-noise separation techniques. It achieves effective identification of signals and noise in intelligent controllable source electromagnetic data, improving the signal-to-noise separation effect, accuracy, and data quality.

[0054] This embodiment also provides an electromagnetic signal-to-noise identification system based on an extreme learning machine. This system includes at least one module. At least one module is used to execute the signal-to-noise identification method of this embodiment, and at least one module used to execute the signal-to-noise identification method is an extreme learning machine module.

[0055] Specifically, the Extreme Learning Machine used in this embodiment includes an input layer, a hidden layer, and an output layer:

[0056] The input layer receives data from training feature samples and passes the training samples to the hidden layer. The hidden layer assigns a center to each neuron node, and the number of neurons is equal to the number of training samples. The output layer calculates the distance between the input and the center and outputs a scalar value, which is the signal-to-noise discrimination result.

[0057] The connection weights between the input and hidden layers, as well as the threshold of the hidden layers, can be randomly set and do not need to be adjusted afterward. The connection weights between the hidden and output layers do not need to be iteratively adjusted but are determined once by solving a system of equations.

[0058] Among them, the training feature samples are the optimal and suboptimal combination of feature parameters for each segment of electromagnetic data.

[0059] The following is an example of using an Extreme Learning Machine to train a set of samples:

[0060] Suppose there are N arbitrary samples (X) i ,t i It has n input layer neurons and m output layer neurons, where the input data X i =[x i1 ,x i2 ,...,x in ] T ∈R n Target output data t i =[t i1 ,t i2 ,...,t im ]T ∈R m The output expression for a single-hidden-layer neural network with L hidden nodes is:

[0061]

[0062] In the formula, g(·) is the activation function, and W i =[w i,1 ,w i,2 ,...,w i,n ] T For the input weights, β i For the output weights, b i W is the bias of the i-th hidden layer unit. i ·X j W i and X j The inner product of.

[0063] Among them, X j and X i Both represent input samples, where j and i are both references to the quantity. Where O j This represents the output of a single hidden layer neural network model.

[0064] The goal of learning a single-hidden-layer neural network is to minimize the error of the output, which can be expressed as:

[0065]

[0066] Random initialization W i By combining these, we can obtain the hidden layer output matrix H:

[0067] Hβ=T

[0068] In the formula, H represents the output of the hidden layer node, β is the output weight, and T is the desired output. The solution is obtained using the generalized inverse of the matrix:

[0069]

[0070] In the formula, Let H be the generalized inverse matrix.

[0071] Example 2:

[0072] Example 2 uses the signal-to-noise recognition method and system disclosed in Example 1.

[0073] In the specific implementation process, the schematic diagram of the sample library in step S1 is as follows: Figure 2As shown. The sample library includes corresponding pseudo-random signal samples and pseudo-random signal spectrum samples, samples containing impulse interference and spectrum samples containing impulse interference, and samples containing attenuation interference and spectrum samples containing attenuation interference.

[0074] in, Figure 3 The sample library is shown below. Figure 2 Feature distribution diagram of the top 10 samples in each class.

[0075] When performing step S2, for Figure 2 and Figure 3 The samples and sample data are prioritized and sorted to determine the optimal and second-best sets of feature parameters.

[0076] In this embodiment, Figure 4 The image shows the known categories in the sample database and the predicted results. With the predicted sample number on the x-axis and the classification result on the y-axis, the image shows that the predicted results match the actual results.

[0077] To verify the effectiveness of this embodiment, the method of the present invention was also compared and analyzed in terms of time and frequency domain processing in simulated noisy data and measured data. Figure 5 The above describes the processing effect and spectrum of the simulated noisy signal according to Example 2 after being processed by the method of the present invention. Figure 6 The image shown is a diagram illustrating the processing effect of a measured signal according to an embodiment of this application after being processed by the method of the present invention; from Figure 5 and Figure 6 It can be shown that the method of this embodiment can identify noise data with high accuracy, and the noise data and its spectrum can also be effectively removed from the time and frequency domains. The identified effective data is sorted, merged and reconstructed into the original high-quality electromagnetic data. The time domain waveform and spectrum are completely consistent with the characteristics of electromagnetic data, and the data quality can be effectively improved.

[0078] The evaluation is conducted by comparing the effects of the original data and the electric field curves processed by the method of this embodiment. For example... Figure 7 As shown, Figure 7 The images show the electric field curves at the measured points before and after processing. Because the original controllable source electromagnetic data is affected by noise in the time domain, abnormal jumps and rises occur in the low-frequency band of the electric field curve, resulting in a deterioration in the overall curve quality. It is easy to see that after processing using the method in this embodiment, the electric field curves at the measured points are more stable and free from abnormal fluctuations, indicating that the noisy electromagnetic data has been effectively identified and removed.

[0079] In the above embodiments, the electromagnetic data from a controllable source is segmented at equal sampling rates. Multi-domain, multi-feature, fuzzy C-means clustering, and extreme learning machine (ELM) are used to achieve intelligent identification of electromagnetic data. The identified valid signal data is then merged to reconstruct high-quality electromagnetic data. Multi-domain, multi-feature characterization of signals and noise is employed, and the priority of feature parameters in signal-to-noise differentiation is compared. The extreme learning machine algorithm enables simple operation and accurate classification. Experimental results show that this invention improves the efficiency of existing machine learning algorithms in feature extraction and signal-to-noise differentiation, demonstrating practical value and innovation.

[0080] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An electromagnetic signal-to-noise identification method based on extreme learning machine, characterized in that: Includes the following steps: Step S1: Divide the electromagnetic method data into equally spaced segments, extract the multi-domain multi-feature parameters of each segment of electromagnetic method data and store them in the sample database; Step S2: Perform cluster analysis on the multi-domain and multi-feature parameters in the sample database data, sort the feature parameters according to their priority with the known sample database categories, use the inter-class distance as the evaluation index, and determine the optimal and second-best combination of feature parameters. Step S3: Train the extreme learning machine using the optimal and suboptimal combination of feature parameters of each electromagnetic data segment, and then perform signal-to-noise intelligent identification on the electromagnetic data. In the process of signal-to-noise intelligent identification, pseudo-random signals are identified as valid signals, and obviously abnormal waveforms are identified as noise. The extreme learning machine includes an input layer, a hidden layer, and an output layer. The input layer and the output layer are equipped with multiple neuron nodes. The input layer is used to receive data from the combination of feature parameters as training samples and pass the training samples to the hidden layer. The hidden layer assigns a center to each neuron node, and the number of neurons in the hidden layer is equal to the number of training samples. The output layer calculates the distance between the input and the center and outputs a scalar value, which is the discrimination result of signal-to-noise identification. The connection weights between the input layer and the hidden layer, and the threshold of the hidden layer are randomly set and do not need to be adjusted after being set. The connection weights between the hidden layer and the output layer do not need to be iteratively adjusted, but are determined directly by solving the data. When there is any sample and has One input layer neuron, One output layer neuron; Among them, input data Target output data For a person with The output expression of a single hidden layer neural network with n hidden layer nodes is: ； In the formula, For activation function, For input weights, To output weights, It is the first The bias of each hidden layer unit, express and The inner product; in, and All represent input samples. and All of these refer to quantity, among which, This represents the model output of a single hidden layer neural network. The goal of a single-hidden-layer neural network is to minimize the error of its output, expressed as: ； Random initialization and This yields the hidden layer output matrix. : ； In the formula, This represents the output of the hidden layer node. To output weights, To obtain the desired output, solve using the generalized inverse of the matrix: ； In the formula, Representation matrix The generalized inverse matrix; Step S4: Directly remove the data identified as noise in step S3, retain the effective signals and merge them to reconstruct high-quality controllable source electromagnetic data; In step S1, the sampling rate is used as the number of segments when the electromagnetic method data is divided into equally spaced segments; In step S1, the multiple domains are time domain, frequency domain, and time-frequency domain, and the multiple feature parameters are pulse factor, peak factor, sample entropy, fuzzy entropy, average frequency, waveform factor, wavelet energy entropy, and wavelet singular entropy. In step S2, fuzzy C-means clustering is used to perform cluster analysis, prioritizing the known sample database data categories and the signal-to-noise categories represented by multi-domain multi-feature parameters, and determining the optimal and second-best combination of feature parameters. The sample library includes corresponding pseudo-random signal samples and pseudo-random signal spectrum samples, samples containing pulse interference and samples containing pulse interference spectrum samples, and samples containing attenuation interference and samples containing attenuation interference spectrum samples.

2. An electromagnetic signal-to-noise identification system based on extreme learning machine, characterized in that, It also includes at least one module, wherein the module is used to perform the method of claim 1.

3. The electromagnetic signal-to-noise identification system based on extreme learning machine according to claim 2, characterized in that, The module in question is an Extreme Learning Machine module.

Images