Denoising method and system for controllable source electromagnetic method data
Through fractal features and optimization learning methods, deep neural networks are trained using fractal analysis and improved hippo optimization algorithm to eliminate noise data, solving the problem of electromagnetic noise in controllable source electromagnetic method, and improving data quality and deep detection effect.
Patent Information
- Application Number
- CN202510367232.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2025-07-04
AI Technical Summary
Electromagnetic noise in the existing controllable source electromagnetic method data seriously affects the deep detection effect, resulting in low signal-to-noise ratio, making it difficult to achieve high-quality underground deep resource and energy exploration.
The method of fractal features and optimization learning is adopted to process the controllable source electromagnetic data in equal-space segmentation, and the features are extracted using fractal analysis, combined with the improved hippo optimization algorithm and deep neural network for hyperparameter optimization, train the optimization model, and perform signal-to-noise identification to eliminate noise data and retain effective signals.
The medium and low frequency band quality of controllable source electromagnetic method data has been improved, the deep detection effect has been improved, and the signal-to-noise ratio and data quality have been improved.
Smart Images
Figure CN120256931A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular, to a method and system for denoising controlled-source electromagnetic method data. Background Art
[0002] With the increasing development of computer technology, artificial intelligence technology has been applied to practical engineering applications. Among them, artificial intelligence technology represented by machine learning has been widely used in signal processing. Machine learning is a method based on statistical learning, and deep learning is a method of representation learning based on data in machine learning. It is usually used to build a neural network model for data processing and analysis, especially suitable for the case of large amounts of data, and the application effect is better than that of small-sample data processing such as machine learning.
[0003] Controlled-source electromagnetic method is one of the important means for exploring underground deep resources and energy. To achieve deep exploration by the controlled-source electromagnetic method, high-quality medium- and low-frequency data must be obtained. Due to frequent human activities, electromagnetic noise is becoming increasingly serious, the quality and signal-to-noise ratio of the observed data are low, and the deep exploration effect is becoming more and more unsatisfactory. At the same time, complex noise greatly affects important areas of underground deep exploration, which also brings difficulties and challenges to the research of geophysical exploration. Therefore, in view of the fractal characteristics and optimized learning model, a method for denoising controlled-source electromagnetic method data based on fractal fusion and optimized learning is provided to solve the influence of electromagnetic noise on deep exploration in existing controlled-source electromagnetic method data. Summary of the Invention
[0004] The present invention provides a method and system for denoising controlled-source electromagnetic method data to solve the influence of electromagnetic noise on deep exploration in existing controlled-source electromagnetic method data.
[0005] To achieve the above object, the present invention is realized by the following technical solutions: In the first aspect, the present invention provides a method for denoising controlled-source electromagnetic method data, including: S1. Perform equally spaced segmentation on the noisy controlled-source electromagnetic method data, and use the time-domain data of each segment of the controlled-source electromagnetic method after segmentation for fractal analysis to extract three fractal characteristics obtained after fractal analysis; S2. Use multiple strategies to optimize the best strategy to improve the hippopotamus optimization algorithm for hyperparameter optimization of the deep neural network learning model; S3. Input the three fractal characteristics into the optimized deep neural network for training to obtain an optimized deep neural network learning model; S4. Use the optimized deep neural network learning model to perform simulation test of simulated electromagnetic data, and apply the model to denoising processing of the noisy controlled-source electromagnetic method data; S5. Perform signal-noise identification on the denoised controlled source electromagnetic method data, remove the noise data segments, retain the valid signal segments, and integrate and reconstruct them in the original sampling order to obtain high-quality controlled source electromagnetic method data.
[0006] Optionally, the three fractal features include: Hurst exponential feature, box count feature and Higuchi fractal feature; The Hurst index feature is obtained by calculating the double logarithmic regression of the average rescaled range of each segment of data and the data length; The box counting feature is obtained by dividing boxes of different sizes and counting the relationship between the number of data points in each box and the size; The Higuchi fractal feature is obtained by calculating the slope of the linear relationship between the logarithm of the curve length and the data length under different segment lengths.
[0007] Optionally, in step S2, the multiple strategies preferably include: Different strategies are adopted in the population initialization phase, exploration and development phase, iterative update and stop phase of the Hippo optimization algorithm; Through the standard test function test, the lens imaging reverse learning strategy in the middle stage and the adaptive t-distribution variation perturbation strategy in the end stage are selected to improve the Hippo optimization algorithm; The hyperparameters include the number of layers, learning rate and weight values of the deep neural network.
[0008] Optionally, in step S3, the optimized deep neural network model includes an input layer, a hidden layer and an output layer; The input layer receives three fractal feature parameters, the hidden layer performs nonlinear mapping through an activation function, and the output layer generates a noise identification result.
[0009] Optionally, the mathematical expression of the deep neural network model satisfies the following relationship: ; In the formula, represents the linear relationship coefficient, Indicates the offset, represents the activation function, Indicates the number of characteristic parameters, the value range is .
[0010] Optionally, in step S5, the signal-to-noise identification is implemented by the following formula: ; ; In the formula, Each segment of the controlled-source electromagnetic method time-domain data representing the fractal feature, indicating the number of segments, indicating the signal-to-noise identification process, I-DNN indicating the optimized deep neural network learning model, indicating the outlier of the noise data segment feature, indicating the feature value of the effective signal data segment.
[0011] In a second aspect, an embodiment of the present application provides a denoising system for controlled-source electromagnetic method data, including a processor and a memory; The memory is used to store a computer program; The processor is configured to implement the method steps described in any one of the first aspects when executing the program stored in the memory.
[0012] Beneficial effects: The denoising method for controlled-source electromagnetic method data provided by the present invention realizes the representation of the signal-to-noise data of the controlled-source electromagnetic method by fractal features, and applies the hippopotamus optimization deep neural network improved by the fusion optimization strategy to the processing of controlled-source electromagnetic data, improves the dilemmas in traditional denoising of the controlled-source electromagnetic method, enhances the denoising effect of the controlled-source electromagnetic method and the quality of electromagnetic data in the medium and low frequency bands, and solves the application prospect in the processing of controlled-source electromagnetic method data with fractal feature fusion optimization learning. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 is a flowchart of the denoising method for controlled-source electromagnetic method data according to a preferred embodiment of the present invention; Figure 2 is a schematic diagram of the convergence comparison of the hippopotamus optimization algorithm improved by the preferred strategy according to a preferred embodiment of the present invention; Figure 3 is a schematic diagram of the structure of the deep neural network provided by a preferred embodiment of the present invention; Figure 4 is an effect diagram of the preferred strategy improved hippopotamus optimization deep neural network provided by a preferred embodiment of the present invention; Figure 5 is an effect diagram of the processed simulation signal by the method according to a preferred embodiment of the present invention; Figure 6 is an identification effect diagram of the measured data by the method according to a preferred embodiment of the present invention; Figure 7 is a comparison effect diagram of the measured point curve before and after processing provided by a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0014] The technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.
[0015] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the art to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Similarly, terms such as "a" or "one" do not denote a quantity limitation, but mean that there is at least one. The terms "connected" or "coupled" and the like are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms "upper", "lower", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship also changes accordingly.
[0016] Please refer to Figure 1 , an embodiment of the present application provides a method for denoising controlled-source electromagnetic method data, including: S1: Perform equally spaced segmentation on the controlled-source electromagnetic method data, use the time-domain data of each segment of the controlled-source electromagnetic method for fractal analysis, and extract three fractal features; In this step, use the time-domain data of each segment of the controlled-source electromagnetic method for fractal analysis, and extract three fractal features. The three fractal features are the Hurst exponent feature, the box-counting feature, and the Higuchi fractal feature; S2: Use multiple strategies to optimize the best strategy to improve the hippopotamus optimization algorithm, and perform hyperparameter optimization for the deep neural network learning model; In this step, the multiple-strategy optimization method is as follows: the initial part of the hippopotamus optimization algorithm is the population initialization optimization, the middle part is the strategy optimization in the exploration and exploitation stage, and the end part is the strategy optimization in the iterative update and stopping stage; compare various existing strategies and then apply them to the test of standard test functions, and the best strategy is selected as the lens imaging reverse learning strategy in the middle part and the adaptive t-distribution mutation perturbation strategy in the end part; use the optimized strategy to improve the hippopotamus optimization algorithm for hyperparameter optimization of the deep learning model, where the hyperparameters are the number of layers, the learning rate, and the weight value.
[0017] S3: Input the three fractal features into the optimized learning for training to obtain an optimized deep neural network learning model; In this step, three fractal features are input into the optimized deep neural network for learning and training to obtain an optimized deep neural network learning model. Among them, the three fractal features input into the optimized deep learning network are , where the deep neural network model consists of an input layer, a hidden layer, and an output layer. The deep neural network DNN model is as follows: ; In the formula represents the linear relationship coefficient, represents the offset, represents the activation function, represents the number of feature parameters, and its value range is within .
[0018] S4: Use the optimized deep neural network learning model to conduct simulation tests on simulated electromagnetic data, and apply the optimized deep neural network learning model to the denoising process of actual controlled-source electromagnetic method data; S5: After identifying the signal-to-noise ratio of the controlled-source electromagnetic method data, eliminate the noise data, retain the effective signals, and integrate and reconstruct them in the original sampling order to obtain high-quality controlled-source electromagnetic method data.
[0019] In this step, use the optimized deep neural network learning model to conduct simulation tests on simulated electromagnetic data, and apply the optimized learning model to the denoising process of actual controlled-source electromagnetic method data; after identifying the signal-to-noise ratio of the controlled-source electromagnetic method data, eliminate the noise data, retain the effective signals, and integrate and reconstruct them in the original sampling order to obtain high-quality controlled-source electromagnetic method data : ; ; In the formula, respectively represent each segment of the controlled-source electromagnetic method time-domain data segment of the fractal feature, represents the number of segments, represents the signal-to-noise ratio identification process, I-DNN represents the optimized deep neural network learning model, represents the characteristic outlier of the noise data segment, represents the characteristic value of the effective signal data segment.
[0020] Through the above steps, the signal-to-noise processing of the controlled-source electromagnetic method data with fractal fusion optimization learning is realized, the accuracy of describing the signal-to-noise difference by the characteristic parameters is improved, and the multi-strategy optimization and improvement of the intelligent optimized deep neural network model can improve the high-precision intelligent parameter optimization, effectively improve the denoising accuracy and effect of the controlled-source electromagnetic method, and improve the data quality in the middle and low frequency bands of the controlled-source electromagnetic method.
[0021] Embodiment 2 is the specific implementation process of this method. Specifically, like Figure 1 As shown, this embodiment discloses a controlled source electromagnetic method data denoising method based on fractal fusion optimization learning, comprising the following steps: S1: The controlled source electromagnetic method data is processed in equal intervals, and fractal analysis is performed using each segment of the controlled source electromagnetic method time domain data to extract three fractal features; the three fractal features are Hurst index feature, box count feature, and Higuchi fractal feature; S2: Use multiple strategies to select the best strategy to improve the Hippo optimization algorithm and optimize the hyperparameters of the deep neural network learning model; the initial part of the Hippo optimization algorithm is the population initialization optimization, the middle part is the strategy optimization in the exploration and development stage, and the end part is the strategy optimization in the iterative update and stop stage; Figure 2 As shown in FIG. 1 , a comparison of various existing strategies is performed and then applied to the test of the standard test function, and the best strategy is selected as the lens imaging reverse learning strategy in the middle part and the adaptive t-distribution variation perturbation strategy in the end part; Figure 4 The figure shows the use of the optimization strategy to improve the Hippo optimization algorithm for hyperparameter optimization of the deep learning model, where the hyperparameters are the number of layers, learning rate and weight value.
[0022] S3: Input the three fractal features into the optimized learning for training to obtain the optimized deep neural network learning model; the three fractal features are input into the optimized deep learning network: , among which, Figure 3 As shown, the deep neural network model consists of an input layer, a hidden layer, and an output layer. The deep neural network DNN model is: ; In the formula represents the linear relationship coefficient, Indicates the offset, represents the activation function, Indicates the number of characteristic parameters, the value range is .
[0023] S4: Use the optimized deep neural network learning model to simulate electromagnetic data for simulation testing, such as Figure 6 The figure shows the application of the optimized deep neural network learning model to the actual controlled source electromagnetic method data denoising process; S5: After the signal-noise identification of the controlled source electromagnetic method data, the noise data is removed, the effective signal is retained, and the original sampling order is integrated and reconstructed to obtain high-quality controlled source electromagnetic method data : ; ; In the formula, Each time-domain data segment of the controlled-source electromagnetic method representing the fractal features, represents the number of segments, represents the signal-to-noise identification process, and I-DNN represents the optimized deep neural network learning model, represents the outlier of the noise data segment feature, represents the feature value of the effective signal data segment.
[0024] To verify the effectiveness of the effect of this embodiment, the method of the present invention is processed in simulated noisy signals and measured data. Figure 5 The figure shows the processing effect diagram of the simulated noisy signal according to the embodiment of the present application by the method of the present invention; analysis Figure 6 It can be seen that the original data contains noise data. After being processed by the method of the present invention, the effective signal can be effectively obtained, and the noise data is completely removed. Comprehensive analysis Figure 5 and Figure 6 It can be shown that the method of this embodiment can effectively realize the intelligent noise processing of the fractal feature fusion of the controlled-source electromagnetic method data, and improve the data quality of the controlled-source electromagnetic method in the middle and low frequency bands.
[0025] By comparing the effects of the electric field curves before and after processing at three measured points. As Figure 7 shown, due to the influence of noise on the original electromagnetic data, the curve fluctuates at some frequency points in the middle and low frequency bands, the amplitude is unstable, and the curve quality is low; it is not difficult to find that after being processed by the method of this embodiment, the electric field curve of the measured point shows a stable trend and there is no mutation of abnormal frequency points, indicating that the processed controlled-source electromagnetic method data has been effectively improved.
[0026] In the above embodiment, fractal analysis is performed using each time-domain data segment of the controlled-source electromagnetic method to extract three fractal features; the best strategy is selected by multiple strategies to improve the hippopotamus optimization algorithm for hyperparameter optimization of the deep neural network learning model; the three fractal features are input into the optimized deep neural network learning for training to obtain the optimized deep neural network learning model; the optimized deep neural network learning model is used for simulation test of simulated electromagnetic data, and the optimized deep neural network learning model is applied to the denoising process of actual controlled-source electromagnetic method data; after the signal-to-noise of the controlled-source electromagnetic method data is identified, the noise data is removed, the effective signal is retained, and the high-quality controlled-source electromagnetic method data is obtained by integrating and reconstructing according to the original sampling order. Through the analysis of the experimental results, the present invention improves the denoising effect of the controlled-source electromagnetic method, improves the electromagnetic data quality in the middle and low frequency bands, and has certain practical value and innovation.
[0027] The embodiment of the present application also provides a denoising system for controlled-source electromagnetic method data, including a processor and a memory; The memory is used to store computer programs; A processor, when executing a program stored in a memory, implements any of the method steps described in the first aspect.
[0028] The above-mentioned denoising system for controlled-source electromagnetic method data can implement each embodiment of the above-mentioned denoising method for controlled-source electromagnetic method data, and can achieve the same beneficial effects, which will not be elaborated here.
[0029] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the protection scope determined by the claims.
Claims
1. A denoising method for controlled-source electromagnetic method data, characterized in that Including: S1. Perform equally-spaced segmentation processing on the noisy controlled-source electromagnetic method data, use the time-domain data of the controlled-source electromagnetic method in each segment after segmentation for fractal analysis, and extract three fractal features obtained after fractal analysis; S2. Use multiple strategies to optimize the best strategy to improve the hippopotamus optimization algorithm, and use the improved hippopotamus optimization algorithm to perform hyperparameter optimization of the deep neural network learning model to obtain an optimized deep neural network; S3. Input the three fractal features into the optimized deep neural network for training to obtain an optimized deep neural network learning model; S4. Use the optimized deep neural network learning model to perform simulation test of electromagnetic data, and apply the model to the denoising process of the noisy controlled-source electromagnetic method data; S5. Perform signal-to-noise identification on the denoised controlled-source electromagnetic method data, eliminate the noise data segments, retain the effective signal segments, and integrate and reconstruct the retained effective signal segments in the original sampling order to obtain high-quality controlled-source electromagnetic method data.
2. The denoising method for controlled-source electromagnetic method data according to claim 1, characterized in that The three fractal features include: Hurst exponent feature, box-counting feature, and Higuchi fractal feature; Among them, the Hurst exponent feature is obtained by calculating the double logarithmic regression of the average rescaled range of each segment of data and the data length; The box-counting feature is obtained by dividing boxes of different sizes and counting the relationship between the number of data points in each box and the size; The Higuchi fractal feature is obtained by calculating the slope of the linear relationship between the logarithm of the curve length at different segment lengths and the data length.
3. The denoising method for controlled-source electromagnetic method data according to claim 1, characterized in that In the step S2, the multiple-strategy optimization includes: Adopting different strategies in the population initialization stage, exploration and development stage, and iterative update and stop stage of the hippopotamus optimization algorithm; Through the test of standard test functions, the lens imaging reverse learning strategy in the middle stage and the adaptive t-distribution mutation perturbation strategy in the end stage are optimized and used to improve the hippopotamus optimization algorithm; The hyperparameters include the number of layers, learning rate, and weight value of the deep neural network.
4. The denoising method for controlled-source electromagnetic method data according to claim 1, characterized in that In the step S3, the optimized deep neural network model includes an input layer, a hidden layer, and an output layer; The input layer receives three fractal feature parameters, the hidden layer performs non-linear mapping through an activation function, and the output layer generates a noise identification result.
5. The denoising method for controlled-source electromagnetic method data according to claim 4, wherein The mathematical expression of the deep neural network model satisfies the following relationship: ； Wherein, represents the linear relationship coefficient, represents the offset, represents the activation function, represents the number of feature parameters, and the value range is in , represents the set of three fractal features, DNN represents the deep neural network learning model.
6. The denoising method for controlled-source electromagnetic method data according to claim 1, characterized in that In the step S5, the signal-to-noise identification is realized through the following formula: ； ； In the formula, respectively represent each controllable source electromagnetic method time domain data segment of the fractal feature, represents the number of segments, represents the signal-to-noise identification process, I-DNN represents the optimized deep neural network learning model, represents the outlier of the noise data segment feature, represents the feature value of the effective signal data segment.
7. A denoising system for controlled-source electromagnetic method data, characterized in that, Including a processor and a memory; The memory is used to store a computer program; The processor is used to implement the method steps described in any one of claims 1-6 when executing the program stored on the memory.