A method and system for suppressing noise of microseismic data, storage medium and seismic information processing equipment

Abstract

A method and system for suppressing noise in microseismic data, a storage medium and a seismic information processing device, which belong to the microseismic data processing technology, are used to solve the problem that the existing methods cannot eliminate all noise and will also lose effective signals. Technical points: Use a detector to collect microseismic data in the target block; divide the obtained microseismic data into training set, verification set and test set in proportion, input the training set data into the time attention mechanism mutual information generation adversarial network model to learn and output the microseismic noise suppression pre-training model; input the verification set data into the trained model, use the signal-to-noise ratio to verify the model parameters, and obtain the optimal microseismic noise suppression model hyperparameter combination; then use the test set data to test the optimal hyperparameter combination and obtain the final microseismic noise suppression model; input the field microseismic data into the final microseismic noise suppression model to obtain the denoised microseismic data, and realize the microseismic data noise suppression.

Description

Technical Field

[0001] The invention relates to a method and system for suppressing microseismic data noise, a storage medium and seismic information processing equipment, and belongs to the technical field of microseismic data processing. Background Art

[0002] In recent years, the potential for conventional oil and gas production in my country has gradually decreased, but the demand for oil and gas brought about by the rapid development of the country's industry has continued to increase. By 2020, my country's dependence on foreign oil has reached 73.5%. my country's unconventional oil and gas reserves are about 107.5 billion tons, with huge development potential. The exploration and development of unconventional oil and gas is the only way to ensure my country's energy security. Microseismic monitoring technology is a key technology for evaluating the effect of hydraulic fracturing and obtaining parameters such as the orientation, length, and height of fractures. It is an important means to ensure the efficient development of unconventional oil and gas.

[0003] In engineering practice, ground microseismic geophones not only collect a large amount of random noise during data collection, but also contain strong interference caused by complex ground noise sources (such as 50Hz industrial AC interference, drilling interference, etc.), which makes the effective microseismic signals very easy to be submerged, seriously affecting the accuracy and reliability of the identification of effective microseismic events and the location of the source. Therefore, noise suppression is the core link of microseismic data preprocessing. However, when establishing the noise suppression model, we found that the traditional denoising method is usually based on subjective expert experience guidance when distinguishing effective signals from noise, that is, it is determined that the noise signal only exists in the high-frequency part of the overall signal, and the high-frequency noise is eliminated by low-pass filtering. However, the noise signal contained in the microseismic signal is coupled. Subjectively identifying the noise frequency and taking corresponding methods to eliminate it not only cannot eliminate all noise but also loses effective signals. In addition, the distribution of effective microseismic event characteristics is discrete and only exists in the local position of the time series. The microseismic signals collected by different geophones also show different characteristics due to geological changes, which leads to the loss of effective events in the process of noise suppression of microseismic data. Therefore, finding new methods to suppress the noise in microseismic data without losing valid events has become extremely important in unconventional oil and gas exploration and development. Summary of the invention

[0004] In order to solve the technical problems mentioned in the background technology: the traditional denoising method is usually based on subjective expert experience guidance when distinguishing effective signals from noise, that is, it is determined that the noise signal only exists in the high-frequency part of the overall signal, and the high-frequency noise is eliminated by low-pass filtering. However, the noise signal contained in the microseismic signal is coupled, and subjectively determining the noise frequency and taking corresponding methods to eliminate it not only fails to eliminate all noise but also loses effective signals. Therefore, a microseismic data noise suppression method and system based on a temporal attention mechanism mutual information generation adversarial network, a storage medium and a seismic information processing device are proposed.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generation adversarial network, characterized in that the method comprises the following steps:

[0006] Step 1: Use the geophone to collect microseismic data of the target block to obtain microseismic data X with noise and microseismic data without noise.

[0007] Step 2: The microseismic data collected in step 1 are divided into a training set, a validation set, and a test set in a ratio of 8:1:1; the training set is represented as where x k is the kth training set sample, with a total of K training samples; the validation set is represented by where x a is the a-th validation set sample, with a total of A validation samples; the test set is represented by where x b is the b-th test set sample, with a total of B test samples; the sampling frequency of the training set, validation set and test set is f;

[0008] Step 3: Extract the data collected in step 1 The structural semantic feature set Among them, c d ∈R H×1 is the dth microseismic structure semantic vector, with a length of H;

[0009] Step 4: Sample the random noise distribution (e.g., Gaussian distribution) with a sampling frequency of f and obtain a set of random noise vectors where z m ∈R N×1 is the mth random noise vector, consisting of N sampling points;

[0010] Step 5: Take the data obtained in step 3 and The result obtained in step 4 Merge dimensions to obtain feature vectors where vu ∈R (H+N)×1 is the u-th merged feature vector;

[0011] Step 6: Use the time-attention mechanism-based mutual information generative adversarial network to learn the training set data and output the trained microseismic noise suppression pre-training model. The specific path is as follows:

[0012] The microseismic noise suppression model used in this step includes a generation model, a discrimination model and a separation model;

[0013] The generation model consists of a temporal attention module and a generation module. The temporal attention module is specifically defined as:

[0014]

[0015] The input of the temporal attention module is v obtained in step 5. u,j ; The output of the temporal attention module is weighted α u ; σ t represents the tanh activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Z′×1; is the bias term;

[0016] The generation module consists of a Dense layer, a Batch Normalization layer, a ReLU layer, a Deconv layer, and a Tanh layer; the discriminant model consists of a temporal attention module and a discriminant module. The temporal attention module is specifically defined as:

[0017]

[0018] Among them, the input of the temporal attention module is obtained in step 2 The output of the temporal attention module is weighted β k ; σ s represents the softmax activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is F′×1; is the bias term;

[0019] The discriminant module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Softmax layer;

[0020] The separation model consists of two parts: the temporal attention module and the separation module. The temporal attention module is defined as:

[0021]

[0022] Among them, the input of the temporal attention module is G(z,c) obtained through the generative model h,j ; The output of the temporal attention module is the weight γ h ; σ sig Represents the sigmoid activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Y′×1; is the bias term;

[0023] The separation module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Sigmoid layer;

[0024] The obtained step 5 Input into the generation model to obtain the generated noise-free microseismic data set; the generated noise-free microseismic data set and the one obtained in step 2 are combined. The noise-free microseismic data are input into the separation model to extract the structural semantic features, the mutual information between the noise-free microseismic data and the extracted structural semantic features is calculated, the separation model is trained, and the gradient is passed back to the generation model; finally, when the maximum number of iterations is reached, the pre-trained model for microseismic noise suppression can be obtained;

[0025] Step 7: Use the validation set data obtained in step 2 Input into the noise suppression pre-training model obtained in step 6, use the signal-to-noise ratio to verify the performance of the trained noise suppression pre-training model, and obtain the optimal hyperparameter combination of the microseismic noise suppression model;

[0026] Step 8: Use the test set data obtained in step 2 The optimal hyperparameter combination of the microseismic noise suppression model obtained in step 7 is tested to obtain the final microseismic data noise suppression model;

[0027] Step 9: Input the field microseismic data into the final microseismic noise suppression model obtained in step 8 to obtain denoised microseismic data, thereby achieving microseismic data noise suppression.

[0028] Furthermore, the generation model training process in step 6 is performed according to the following path:

[0029] 2.1 The generation model consists of two parts: the temporal attention module and the generation module. The temporal attention module is defined as:

[0030]

[0031] The input of the temporal attention module is v obtained in step 5 u,j ; The output of the temporal attention module is weighted α u ; σ t represents the tanh activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Z′×1; is the bias term.

[0032] 2.2 The v obtained in step 5 u,j And α obtained by formula (1) u Input into formula (2) to obtain the enhanced feature map v att , which is defined as follows:

[0033]

[0034] Among them, v att is the enhanced feature map of the merged feature vector, v u is the u-th merged eigenvector, α u is the temporal attention weight of the u-th merged feature vector.

[0035] 2.3 The v obtained by formula (2) att Input into the generation module to obtain the generated noise-free microseismic data set Among them, G(z,c) h ∈R S× 1 is the hth generated noise-free microseismic data; the generation module consists of a Dense layer, a Batch Normalization layer, a ReLU layer, a Deconv layer, and a Tanh layer.

[0036] Furthermore, the discriminant model training process in step 6 is performed according to the following path:

[0037] 3.1 The discriminant model consists of two parts: the temporal attention module and the discriminant module. The temporal attention module is defined as:

[0038]

[0039] The input of the temporal attention module is obtained in step 2 The output of the temporal attention module is weighted β k ; σ s represents the softmax activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is F′×1; is the bias term.

[0040] 3.2 Obtained from step 2 And β obtained by formula (3) k Input into formula (4) to obtain the enhanced feature map x att , which is defined as follows:

[0041]

[0042] where x att for Enhanced feature map of is the kth training set sample, β k is the temporal attention weight of the kth training set sample;

[0043] 3.3 The x obtained by formula (4) att Input into the discrimination module to obtain The probability of belonging to a real sample. The discriminant module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Softmax layer.

[0044] Furthermore, the separation model training process in step 6 is performed according to the following path:

[0045] 4.1 The separation model consists of two parts: the temporal attention module and the separation module. The temporal attention module is defined as:

[0046]

[0047] The input of the temporal attention module is G(z,c) obtained through the generative model h,j ; The output of the temporal attention module is the weight γ h ; σ sig Represents the sigmoid activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Y′×1; is the bias term.

[0048] 4.2 Obtained through Feature 2.3 And γ obtained by formula (5) h Input into formula (6) to obtain the enhanced feature map q att , which is defined as follows:

[0049]

[0050] Among them, q att is the enhanced feature map of G(z,c); G(z,c) h is the noise-free microseismic data generated for the hth time, γ h is the temporal attention weight of the h-th generated noise-free microseismic data.

[0051] 4.3 The q obtained by formula (6) att Input into the separation module to obtain The structural semantic vector in is the wth extracted structural semantic feature vector, and calculates G(z,c) and c q The mutual information of the two layers is obtained by averaging 100% and 100% respectively; the separation module is composed of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Sigmoid layer.

[0052] Furthermore, the model verification process in step 7 is performed according to the following path:

[0053] The validation set data obtained in step 2 The noise suppression pre-training model obtained in step 6 is input, and the performance of the trained noise suppression pre-training model is verified using the signal-to-noise ratio, and the optimal hyperparameter combination of the microseismic noise suppression model is obtained.

[0054] Furthermore, the model testing process in step 8 is performed according to the following path:

[0055] The test set data obtained in step 2 The optimal hyperparameter combination of the microseismic noise suppression model obtained in step 7 is tested to obtain the final microseismic data noise suppression model.

[0056] Furthermore, the noise suppression process in step 9 is performed according to the following path: the field microseismic data is input into the final microseismic noise suppression model obtained in step 8 to obtain denoised microseismic data, thereby achieving noise suppression of the microseismic data.

[0057] A microseismic data noise suppression system based on a temporal attention mechanism mutual information generation adversarial network, the system has a program module corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned microseismic data noise suppression method based on a temporal attention mechanism mutual information generation adversarial network during operation.

[0058] A computer-readable storage medium stores a computer program, wherein the computer program is configured to implement the steps of the above-mentioned method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generation adversarial network when called by a processor.

[0059] A seismic information processing device, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned microseismic data noise suppression method based on a temporal attention mechanism mutual information generation adversarial network.

[0060] The present invention has the following beneficial technical effects:

[0061] The microseismic noise suppression method described in the present invention overcomes the shortcomings of traditional filtering methods that cannot completely remove coupled noise and are prone to cause the loss of effective microseismic events. Traditional denoising methods are usually based on subjective expert experience guidance when distinguishing effective signals from noise, that is, it is determined that the noise signal only exists in the high-frequency part of the overall signal, and then the high-frequency noise is eliminated by low-pass filtering. However, the noise signal contained in the microseismic signal is coupled. Subjectively determining the noise frequency and taking corresponding methods to eliminate it not only fails to eliminate all noise but also loses effective signals. In addition, the distribution of effective event characteristics of microseisms is discrete and only exists in local positions of the time series. The microseismic signals collected by different detectors also show different characteristics due to geological changes, which leads to the loss of effective events in the process of noise suppression of microseismic data.

[0062] The method of the present invention comprises the following steps: using a detector to collect microseismic data of a target block; dividing the obtained microseismic data into a training set, a validation set and a test set according to a certain ratio, and inputting the training set data into a temporal attention mechanism mutual information generation adversarial network model to learn and output a microseismic noise suppression pre-training model; inputting the validation set data into the trained model, using the signal-to-noise ratio to verify the model parameters, and obtaining the optimal microseismic noise suppression model hyperparameter combination; then using the test set data to test the optimal hyperparameter combination and obtain the final microseismic noise suppression model; inputting the field microseismic data into the final microseismic noise suppression model to obtain denoised microseismic data, thereby realizing microseismic data noise suppression.

[0063] The present invention takes into account that the noise contained in the microseismic data has the characteristics of "strong coupling and large energy", that is, the noise characteristics and the microseismic data characteristics are encoded in a complex and disordered state in the data space, which makes it difficult for the current method to achieve the ideal denoising effect. Therefore, the present invention introduces a potential noise vector and maximizes the mutual information between the potential noise vector distribution and the generated microseismic data to learn the noise distribution in the microseismic data to achieve the purpose of removing noise. Finally, considering that the distribution of effective event features in the microseismic monitoring data is discretized and only exists in the local position of the time series. Therefore, the present invention adopts a time attention mechanism to amplify the effective event features to avoid the loss of effective events during the denoising process, thereby obtaining a better denoising effect. The present invention proposes a generative adversarial network based on mutual information to solve the problem that strong coupling noise is difficult to remove in microseismic noise suppression, and introduces a time attention mechanism to amplify the effective events in the microseismic data to avoid the loss of effective events during the denoising process. A set of systematic and scientific microseismic data noise suppression methods are proposed, which are operable and practical.

[0064] Therefore, the method of the present invention can effectively solve the above-mentioned problems and amplify effective microseismic events while achieving microseismic noise suppression.

[0065] The effectiveness of the method of the present invention is embodied in the content of the invention:

[0066] 1. This method is built using a deep convolutional neural network, which has excellent nonlinear characterization capabilities and improves the accuracy of microseismic noise suppression at the network structure level.

[0067] 2. This method uses the temporal attention mechanism to amplify the local characteristics of effective events in the microseismic data time series, which can avoid the loss of effective signals in the microseismic data in the process of noise suppression and improve the accuracy of microseismic noise suppression at the level of data characteristics themselves.

[0068] 3. This method uses the calculation of mutual information to strengthen the correlation between the effective characteristics of microseismic data and the microseismic signals, which can effectively suppress the coupling noise contained in the microseismic data and improve the accuracy of microseismic noise suppression at the algorithm level.

[0069] In addition, this method is a data-driven unsupervised learning method, which gets rid of the reliance on manual annotation and reduces more manpower and material resources. The microseismic data on site can be directly noise-reduced through the trained model, providing on-site construction personnel with a more user-friendly microseismic data processing technology, improving on-site work efficiency and achieving the best results. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1is a flowchart of a microseismic noise suppression model (method and system) according to an embodiment of the present invention;

[0071] Figure 2 is a core graph of a mutual information generation adversarial network using a temporal attention mechanism according to an embodiment of the present invention;

[0072] Figure 3 and Figure 4 is a diagram showing an example of visualization of noise suppression of microseismic data according to an embodiment of the present invention. Figure 3 This is a visualization example of noise suppression of microseismic data (microseismic noise data containing noise). Figure 4 This is a visualization example of noise suppression of microseismic data (microseismic noise data after denoising); Figure 5 and Figure 6 This is another set of visualization examples of noise suppression of microseismic data (noisy microseismic noise data);

[0073] exist Figure 4 and Figure 6 In the figure, the four lines after denoising represent the event axes (the event axes are the lines connecting the extreme values ​​of the same vibration phase of each channel on the seismic record), representing the stratification of the strata; the two groups of event axes in the two figures have different directions and different potentials, and the geological structures they represent are different. It can be seen from the provided drawings that the present invention can be used to suppress multiple different geological structure data, which can better reflect the practicality of the algorithm of the present invention. The present invention is not only easy to use on one set of data, but also has a good effect on suppressing multiple different geological structure data. DETAILED DESCRIPTION

[0074] The present invention will be further described below in conjunction with the accompanying drawings:

[0075] In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application.

[0076] Figure 1 This is the microseismic noise suppression flow chart in the technical solution of the present invention. The entire flow chart is completely implemented by Python language programming (of course, other programming languages ​​can also be used to implement it). Figure 1It can be seen that first, the microseismic noise-free data and noisy data at the scene must be collected, and then divided into training set, validation set and test set according to a specific ratio, and the training set data is input into the temporal attention mechanism mutual information generative adversarial network model to learn and output the microseismic noise suppression pre-training model; the validation set data is input into the trained model, the model parameters are verified using the signal-to-noise ratio, and the optimal microseismic noise suppression model hyperparameter combination is obtained; the test set data is then used to test the optimal hyperparameter combination and obtain the final microseismic noise suppression model; the field microseismic data is input into the final microseismic noise suppression model to obtain the denoised microseismic data and realize the noise suppression of the microseismic data.

[0077] Embodiment: Taking the field microseismic data as an example, the effectiveness of the method of the present invention is verified.

[0078] Step 1: Use the geophone to collect microseismic data of the target block to obtain microseismic data signal A with noise and microseismic data signal B without noise.

[0079] Step 2: Divide the microseismic data collected in step 1 into a training set, a validation set, and a test set in a ratio of 8:1:1. The training set is represented as Among them, x k is the kth training set sample; the validation set is represented by Among them, x a is the ath validation set sample; the test set is represented by Among them, x b is the bth test set sample. The sampling point lengths of the training set, validation set and test set are all S.

[0080] Step 3: Extract the structural semantic feature set of signal B collected in step 1 Among them, c d ∈R H×1 is the dth microseismic structure semantic vector with a length of H.

[0081] Step 4: Sample a random noise distribution (e.g., Gaussian distribution) and obtain a set of random noise vectors Among them, z m ∈R N×1 is the mth random noise vector, consisting of N sampling points.

[0082] Step 5: Take the data obtained in step 3 and The result obtained in step 4 Merge dimensions to obtain merged feature vectors Among them, v u ∈R (H+N)×1 is the u-th merged feature vector.

[0083] Step 6: Reference Figure 2 , the training set data is learned using the mutual information generation adversarial network based on the temporal attention mechanism and the trained microseismic noise suppression pre-training model is output. The specific path is as follows:

[0084] The microseismic noise suppression model used in this step includes a generation model, a discrimination model and a separation model.

[0085] The generation model consists of a temporal attention module and a generation module. The temporal attention module is specifically defined as:

[0086]

[0087] The input of the temporal attention module is v obtained in step 5. u,j ; The output of the temporal attention module is weighted α u ; σ t represents the tanh activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Z′×1; is the bias term.

[0088] The generation module consists of a Dense layer, a Batch Normalization layer, a ReLU layer, a Deconv layer, and a Tanh layer, which belongs to the scope of the existing technology.

[0089] The discriminant model consists of two parts: the temporal attention module and the discriminant module. The temporal attention module is specifically defined as:

[0090]

[0091] Among them, the input of the temporal attention module is obtained in step 2 The output of the temporal attention module is weighted β k ; σ s represents the softmax activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is F′×1; is the bias term.

[0092] The discriminant module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Softmax layer.

[0093] The separation model consists of two parts: the temporal attention module and the separation module. The temporal attention module is defined as:

[0094]

[0095] Among them, the input of the temporal attention module is G(z,c) obtained through the generative model h,j ; The output of the temporal attention module is the weight γ h ; σ sig Represents the sigmoid activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Y′×1; is the bias term.

[0096] The separation module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Sigmoid layer.

[0097] Figure 2 The microseismic noise suppression model is explained: the Input into the generation model to obtain the generated noise-free microseismic data set; the generated noise-free microseismic data set and the one obtained in step 2 are combined. The microseismic data are input into the discriminant model respectively, the discriminant model is trained and the gradient is fed back to the generative model; the generated noise-free microseismic data set is input into the separation model to extract the structural semantic features, the mutual information between the generated noise-free microseismic data and the extracted structural semantic features is calculated and the separation model is trained, and the gradient is fed back to the generative model, which can enhance the microseismic data. This method can enhance the correlation between the effective features of the microseismic data and the microseismic signal, and effectively suppress the coupling noise in the microseismic data.

[0098] Step 7: Use the validation set data obtained in step 2 The noise suppression pre-training model obtained in step 6 is input, and the performance of the trained noise suppression pre-training model is verified using the signal-to-noise ratio, and the optimal hyperparameter combination of the microseismic noise suppression model is obtained.

[0099] Step 8: Use the test set data obtained in step 2 The optimal hyperparameter combination of the microseismic noise suppression model obtained in step 7 is tested to obtain the final microseismic data noise suppression model.

[0100] Table 1 Comparison of different denoising models

[0101]

[0102] Step 9: Input the field microseismic data into the final microseismic noise suppression model obtained in step 8 to obtain the denoised microseismic data and achieve microseismic data noise suppression. The experimental results of microseismic data noise suppression are shown in Figure 9. Figure 3 and Figure 4 As shown Figure 5 and Figure 6 shown.

[0103] As can be seen from the provided figures, the present invention can be used to suppress multiple different geological structure data, which can better reflect the practicality of the algorithm of the present invention. The present invention is not only useful on one set of data, but also has a good effect on suppressing multiple different geological structure data. The present invention uses a temporal attention mechanism to amplify the effective event features to avoid the loss of effective events during the denoising process, thereby obtaining a better denoising effect and having good universality.

[0104] The method of the present invention has been verified through simulation experiments and practical applications, and the technical effects claimed by the present invention have been verified.

[0105] The algorithm (method) proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0106] Based on the algorithm (method) proposed in the present invention, a microseismic data noise suppression system based on a temporal attention mechanism mutual information generation adversarial network is developed using a programming language. The system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned microseismic data noise suppression method based on a temporal attention mechanism mutual information generation adversarial network during operation.

[0107] The computer program of the developed system (software) is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-mentioned microseismic data noise suppression method based on the temporal attention mechanism mutual information generation adversarial network when called by the processor. That is, the present invention is materialized on a carrier to become a computer program product.

[0108] A seismic information processing device, comprising at least one processor and a memory connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned microseismic data noise suppression method based on the temporal attention mechanism mutual information generation adversarial network. The seismic information processing device is used as a terminal product of the present invention.

[0109] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0110] The computer programs (also referred to as programs, software, software applications, or codes) of the present invention include machine instructions for programmable processors, and these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or device (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0111] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, they are all within the scope of protection of the present invention.

Claims

1. A method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generative adversarial network, characterized in that The method comprises the following steps: Step 1: Use the geophone to collect microseismic data of the target block to obtain microseismic data X with noise and microseismic data without noise. Step 2: The microseismic data collected in step 1 are divided into a training set, a validation set, and a test set in a ratio of 8:1:1; the training set is represented as where x k is the kth training set sample, with a total of K training samples; the validation set is represented by where x a is the a-th validation set sample, with a total of A validation samples; the test set is represented by where x b is the b-th test track set sample, with a total of B test samples; the sampling frequency of the training set, validation set and test set is f; Step 3: Extract the data collected in step 1 The structural semantic feature set Among them, c d ∈R H×1 is the dth microseismic structure semantic vector, with a length of H; Step 4: Sample the random noise distribution with a sampling frequency of f and obtain a set of random noise vectors where z m ∈R N×1 is the mth random noise vector, consisting of N sampling points; Step 5: Take the data obtained in step 3 and The result obtained in step 4 Merge dimensions to obtain feature vectors where v u ∈R (H+N)×1 is the u-th merged feature vector; Step 6: Use the time-attention mechanism-based mutual information generative adversarial network to learn the training set data and output the trained microseismic noise suppression pre-training model. The specific path is as follows: The microseismic noise suppression model used in this step includes a generation model, a discrimination model and a separation model; The generation model consists of a temporal attention module and a generation module. The temporal attention module is specifically defined as: The input of the temporal attention module is v obtained in step 5. u,j ; The output of the temporal attention module is weighted α u ; σ t represents the tanh activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Z′×1; is the bias term; The generation module consists of a Dense layer, a Batch Normalization layer, a ReLU layer, a Deconv layer, and a Tanh layer; The discriminant model consists of two parts: the temporal attention module and the discriminant module. The temporal attention module is specifically defined as: Among them, the input of the temporal attention module is obtained in step 2 The output of the temporal attention module is weighted β k ; σ s Represents the softmax activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is F′×1; is the bias term; The discriminant module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Softmax layer; The separation model consists of two parts: the temporal attention module and the separation module. The temporal attention module is defined as: Among them, the input of the temporal attention module is G(z,c) obtained through the generative model h,j ; The output of the temporal attention module is the weight γ h ; σ sig Represents the sigmoid activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Y′×1; is the bias term; The separation module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Sigmoid layer; The obtained step 5 Input into the generation model to obtain the generated noise-free microseismic data set; the generated noise-free microseismic data set and the one obtained in step 2 are combined. The noise-free microseismic data are input into the separation model to extract the structural semantic features, the mutual information between the noise-free microseismic data and the extracted structural semantic features is calculated, the separation model is trained, and the gradient is passed back to the generation model; finally, when the maximum number of iterations is reached, the pre-trained model for microseismic noise suppression can be obtained; Step 7: Use the validation set data obtained in step 2 Input into the noise suppression pre-training model obtained in step 6, use the signal-to-noise ratio to verify the performance of the trained noise suppression pre-training model, and obtain the optimal hyperparameter combination of the microseismic noise suppression model; Step 8: Use the test set data obtained in step 2 The optimal hyperparameter combination of the microseismic noise suppression model obtained in step 7 is tested to obtain the final microseismic data noise suppression model; Step 9: Input the field microseismic data into the final microseismic noise suppression model obtained in step 8 to obtain denoised microseismic data, thereby achieving microseismic data noise suppression.

2. According to claim 1, a method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generative adversarial network is characterized by: The generation model training process in step 6 is performed according to the following path: 2.1 The generation model consists of two parts: the temporal attention module and the generation module. The temporal attention module is defined as: The input of the temporal attention module is v obtained in step 5 u,j ; The output of the temporal attention module is weighted α u ; σ t represents the tanh activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Z′×1; is the bias term; 2.2 The v obtained in step 5 u,j And α obtained by formula (1) u Input into formula (2) to obtain the enhanced feature map v att , specifically defined as follows: Among them, v att is the enhanced feature map of the merged feature vector, v u is the u-th merged eigenvector, α u is the temporal attention weight of the u-th merged feature vector; 2.3 The v obtained by formula (2) att Input into the generation module to obtain the generated noise-free microseismic data set Among them, G(z,c) h ∈R S×1 is the hth generated noise-free microseismic data; the generation module consists of a Dense layer, a Batch Normalization layer, a ReLU layer, a Deconv layer, and a Tanh layer.

3. A method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generative adversarial network according to claim 1 or 2, characterized in that: The discriminant model training process in step 6 is performed according to the following path: 3.1 The discriminant model consists of two parts: the temporal attention module and the discriminant module. The temporal attention module is defined as: The input of the temporal attention module is obtained in step 2 The output of the temporal attention module is weighted β k ; σ s Represents the softmax activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is F′×1; is the bias term; 3.2 Obtained from step 2 And β obtained by formula (3) k Input into formula (4) to obtain the enhanced feature map x att , specifically defined as follows: where x att for Enhanced feature map of is the kth training set sample, β k is the temporal attention weight of the kth training set sample; 3.3 The x obtained by formula (4) att Input into the discrimination module to obtain The probability of belonging to a real sample, where the discriminant module consists of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Softmax layer.

4. According to claim 3, a method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generation adversarial network is characterized by: The separation model training process in step 6 is performed according to the following path: 4.1 The separation model consists of two parts: the temporal attention module and the separation module. The temporal attention module is defined as: The input of the temporal attention module is G(z,c) obtained through the generative model h,j ; The output of the temporal attention module is the weight γ h ; σ sig Represents the sigmoid activation function; Represents the jth channel of the convolution kernel, and the convolution kernel size is Y′×1; is the bias term; 4.2 Obtained through Feature 2.3 And γ obtained by formula (5) h Input into formula (6) to obtain the enhanced feature map q att , specifically defined as follows: Among them, q att is the enhanced feature map of G(z,c); G(z,c) h is the noise-free microseismic data generated for the hth time, γ h is the time attention weight of the hth generated noise-free microseismic data; 4.3 The q obtained by formula (6) att Input into the separation module to obtain The structural semantic vector in is the wth extracted structural semantic feature vector, and calculates G(z,c) and c q The mutual information of the two layers is obtained by averaging 100% and 100% respectively; the separation module is composed of a Dense layer, a Batch Normalization layer, a LeakyReLU layer, a Conv layer, and a Sigmoid layer.

5. According to claim 4, a method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generative adversarial network is characterized by: The model verification process in step 7 is performed as follows: The validation set data obtained in step 2 The noise suppression pre-training model obtained in step 6 is input, and the performance of the trained noise suppression pre-training model is verified using the signal-to-noise ratio, and the optimal hyperparameter combination of the microseismic noise suppression model is obtained.

6. According to claim 5, a method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generative adversarial network is characterized by: The model testing process in step 8 is performed according to the following path: The test set data obtained in step 2 The optimal hyperparameter combination of the microseismic noise suppression model obtained in step 7 is tested to obtain the final microseismic data noise suppression model.

7. According to claim 6, a method for suppressing noise in microseismic data based on a temporal attention mechanism mutual information generation adversarial network is characterized by: The noise suppression process in step 9 is performed according to the following path: The field microseismic data are input into the final microseismic noise suppression model obtained in step 8 to obtain denoised microseismic data, thereby achieving microseismic data noise suppression.

8. A microseismic data noise suppression system based on a temporal attention mechanism mutual information generation adversarial network, characterized by: The system has a program module corresponding to the steps of any one of claims 1 to 7 above, and executes the steps in the above-mentioned microseismic data noise suppression method based on the temporal attention mechanism mutual information generation adversarial network when running.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is configured to implement the steps of the microseismic data noise suppression method based on the temporal attention mechanism mutual information generation adversarial network described in any one of claims 1 to 7 when called by a processor.

10. A seismic information processing device, characterized in that: The seismic information processing device includes at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a microseismic data noise suppression method based on a temporal attention mechanism mutual information generation adversarial network as described in any one of claims 1-7.

Images