Micro-seismic event intensity evaluation method and system based on deep learning

Abstract

The invention provides a micro-seismic event intensity evaluation method and system based on deep learning, and belongs to the field of hydraulic fracturing micro-seismic monitoring data interpretation. The method comprises the following steps: step 1, establishing a forward modeling model; 2, constructing a training data set; step 3, constructing a microseism event intensity evaluation network model; 4, taking the training data set as input, training the micro-seismic event intensity evaluation network model, and obtaining a trained micro-seismic event intensity evaluation network model; and 5, evaluating the intensity of the microseism event. According to the method, deep learning is utilized to automatically extract features of multiple microseismic events, and the features are utilized to classify microseismic event intensities; noise data is added into a built data set to simulate a false picking phenomenon in an event recognition process, and a false picking event can be distinguished.

Description

Technical Field

[0001] The present invention belongs to the field of hydraulic fracturing microseismic monitoring data interpretation, and specifically relates to a microseismic event intensity evaluation method and system based on deep learning. Background Art

[0002] In the field of oil extraction, microseismic monitoring technology has become a common technology in the field of unconventional oil and gas extraction. In the hydraulic fracturing of shale oil and gas, by real-time monitoring of underground rupture events, the crack expansion characteristics can be dynamically described to provide real-time guidance for engineering construction. At the same time, based on the microseismic monitoring effect, combined with geology and engineering, post-evaluation can be carried out to provide certain guidance for the optimization of subsequent engineering parameters of fracturing wells.

[0003] Most microseismic interpretation methods are based on the analysis of microseismic events, but they are limited by the surface reception conditions or shallow absorption attenuation. When the acquisition signal quality is poor and the signal-to-noise ratio is low, in order to take into account the recognition ability of weak signals, more microseismic "events" will be detected through multi-channel monitoring superposition, and these "events" need to be further classified and screened for intensity. Relying on manual judgment is highly subjective and inefficient in areas with developed fractures due to the large number of events. The existing quantitative calculation of the intensity of microseismic events is mainly based on natural earthquakes. The event characteristic parameters are extracted and the relevant empirical formulas are used to estimate the magnitude. However, the use of magnitude attributes to classify microseismic events has the problem of distinguishing false events. Summary of the invention

[0004] The purpose of the present invention is to solve the above-mentioned problems existing in the prior art and to provide a method and system for evaluating the intensity of microseismic events based on deep learning.

[0005] The present invention is achieved through the following technical solutions:

[0006] A first aspect of the present invention provides a method for evaluating the intensity of microseismic events based on deep learning, comprising the following steps:

[0007] Step 1, establish a forward model;

[0008] Step 2: construct a training data set;

[0009] Step 3, constructing a network model for evaluating the intensity of microseismic events;

[0010] Step 4, using the training data set as input, training the microseismic event intensity evaluation network model to obtain a trained microseismic event intensity evaluation network model;

[0011] Step 5: Evaluation of the intensity of microseismic events.

[0012] A further improvement of the present invention is:

[0013] In step 1, a forward model is established, and the specific operations include:

[0014] A reasonable layered velocity model is obtained based on known logging data or VSP data. The observation system simulates the surface microseismic observation system and the microseismic observation system in the well to construct the forward model. The source locations are randomly distributed near the target layer. It is assumed that S sources are randomly generated.

[0015] A further improvement of the present invention is:

[0016] In step 2, a training data set is constructed, and the specific operations include:

[0017] Step 21, combining the moment tensor, forward modeling the microseismic data of different focal mechanisms to obtain the microseismic data without noise and with event arrival time annotations;

[0018] Perform time difference correction and leveling based on arrival time information, and capture data within the event time window;

[0019] Assume that the sample data obtained by forward modeling is i is the sample number, and its value range is 1, 2, 3…S, j is the channel number, and k is the sampling point sequence;

[0020] Step 22, expanding the forward modeling data set;

[0021] The microseismic data obtained by forward simulation are flipped, scaled, and noise with different signal-to-noise ratios are added to enhance the data, and the noise is added to the data set as a sample;

[0022] The following formula is used to enhance the data:

[0023] TE=(-1) r *w*T+snr*noise

[0024] Where T is the simulated single-channel record, w is the weighting coefficient, when it is 0, it is a noise sample, and when it is not 0, it indicates the scaling factor of the signal; r is -1 or 1 to represent whether to flip; snr is the added signal-to-noise ratio; noise is the noise that satisfies the Gaussian distribution;

[0025] The enhanced sample data is TE (i) ,i=1,2,…N E ,N E is the number of samples after augmentation;

[0026] Step 23: Create TE (i) The corresponding label vector Label (i) ;

[0027] According to the added signal-to-noise ratio and whether it is noise, the classification result of the event intensity is judged, and the TE is established based on the classification result. (i) The corresponding label vector Label (i) ;

[0028] Step 24: Considering the potential effect of multi-channel superposition features, TE (i) Perform multi-channel superposition to obtain an average and copy the TS (i) , then [(TE (i) , TS (i) , Label (i) ] constitutes the training data set D,

[0029] in, N is the number of detectors.

[0030] A further improvement of the present invention is:

[0031] The input of the microseismic event intensity evaluation network model constructed in step 3 is a two-channel data, one of which is TE (i) , the other channel is TS (i) ;

[0032] The microseismic event intensity assessment network model is divided into a plurality of front convolution blocks for extracting data features of different scales and a rear fully connected layer. The convolution block includes a plurality of modules, which sequentially perform convolution, batch normalization, activation function, convolution, batch normalization, activation function, and pooling operations on the input; for the convolution block in front of the fully connected layer, pyramid pooling, cropping or stretching operations are used to solve the problem of inconsistent data size during actual prediction.

[0033] A further improvement of the present invention is:

[0034] The step 3 includes:

[0035] Cross entropy is used as the loss function of the model, and its formula is as follows:

[0036]

[0037] Among them, P is the probability vector of the network output, Loss is the loss value, which is the output probability of the network, and p i is the probability that the category is i, y is the label, if the category is i, y i =1, otherwise 0;

[0038] Adaptive moment estimation algorithm is used as the optimization algorithm.

[0039] The improvement of the brocade clothing of the present invention is:

[0040] The specific operations of step 4 include:

[0041] The training data set is divided into training set, validation set, and test set in a ratio of 7:1:2. The training set, validation set, and test set are used as input to train the microseismic event intensity evaluation network model to obtain a trained microseismic event intensity evaluation network model.

[0042] A further improvement of the present invention is:

[0043] The microseismic event intensity evaluation in step 5 specifically includes the following operations:

[0044] The preprocessed microseismic data containing rupture signals are input into the trained microseismic event intensity assessment network model to obtain the probability vector results, and the event category is discriminated according to the probability results of category prediction.

[0045] A second aspect of the present invention provides a microseismic event intensity assessment system based on deep learning, comprising:

[0046] A forward model building unit, used for building a forward model;

[0047] A training set construction unit, used to construct a training data set;

[0048] A model building unit, used to build a network model for evaluating the intensity of microseismic events;

[0049] A model training unit is used to train the microseismic event intensity evaluation network model using the training data set as input to obtain a trained microseismic event intensity evaluation network model;

[0050] Event evaluation unit, used for intensity evaluation of microseismic events.

[0051] The third object of the present invention is to provide a computer-readable storage medium, which stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the microseismic event intensity evaluation method based on deep learning as described above.

[0052] A fourth aspect of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the microseismic event intensity assessment method based on deep learning as described above.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention uses deep learning to automatically extract the features of multi-channel microseismic events, and uses these features to classify the intensity of microseismic events; noise data is added to the constructed data set to simulate the false-pick phenomenon in the event recognition process, and can identify false-pick events;

[0055] The present invention can perform classification prediction on data of indefinite length. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The present invention provides a flowchart of a method for evaluating the intensity of microseismic events based on deep learning;

[0057] Figure 2 It is an example of forward model building;

[0058] Figure 3 is a generated simulated microseismic data TE (i) ;

[0059] Figure 4 is the simulated microseismic data TE after eliminating the polarity effect (i) ;

[0060] Figure 5 It is a multi-channel superimposed and expanded data TS (i) ;

[0061] Figure 6 It is the structural diagram of the network model for evaluating the intensity of microseismic events;

[0062] Figure 7 It is the structure diagram of the convolution block. DETAILED DESCRIPTION

[0063] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0064] The present invention provides a method for evaluating the intensity of microseismic events based on deep learning. Figure 1 As shown, the method includes:

[0065] Step 1, establish a forward model;

[0066] Specific operations include:

[0067] A reasonable layered velocity model is obtained based on known logging data or VSP data. The observation system simulates the surface microseismic observation system and the microseismic observation system in the well to construct the forward model. The source locations are randomly distributed near the target layer. It is assumed that S sources are randomly generated.

[0068] The ground observation system adopts star ray or grid pattern, and the in-well observation system follows the well trajectory.

[0069] Step 2: construct a training data set;

[0070] Specific operations include:

[0071] (1) Combined with the moment tensor, forward modeling of microseismic data of different focal mechanisms is performed to obtain noise-free microseismic data with event arrival time annotations;

[0072] According to the arrival time information, the time difference correction and leveling is performed (in order to simulate the actual situation, a small random disturbance is added to the arrival time information). Specifically, the microseismic data after forward simulation is time-shifted, and the time shift amount is is the propagation time from source i to detector j;

[0073] Cut the data within the event time window and assume that the sample data obtained by forward modeling is i is the sample number, and its value range is 1, 2, 3…S, j represents the channel number, and k is the sampling point sequence.

[0074] (2) Expand the forward modeling data set;

[0075] The microseismic data obtained by forward simulation are flipped, scaled, and added with noise of different signal-to-noise ratios to enhance the data. By expanding the number of samples, the risk of overfitting of the network is reduced. In addition, noise is added as a sample to the data set to simulate the actual false event mispickup, so that the network model has the ability to judge false events.

[0076] The following formula is used to enhance the data:

[0077] TE=(-1) r *w*T+snr*noise

[0078] Where T is the simulated single-channel record, w is the weighting coefficient, when it is 0, it is a noise sample, and when it is not 0, it indicates the scaling factor of the signal; r is -1 or 1 to represent whether to flip; snr is the added signal-to-noise ratio; noise is the noise that satisfies the Gaussian distribution;

[0079] The enhanced sample data is TE (i) ,i=1,2,…N E ,N E is the number of samples after augmentation.

[0080] (3) Establish TE (i) The corresponding label vector Label (i) ;

[0081] Generate labels for corresponding event samples based on the added signal-to-noise ratio and whether it is noise.

[0082] According to the above data enhancement process, the added signal-to-noise ratio and whether it is noise are determined, for example, TE = (-1) rIn *w*T+snr*noise, if w is 0, it is a noise sample. The suspected event, weak event, medium intensity event and strong event are defined according to the size of the added snr.

[0083] The event intensity evaluation network classification results are divided into five categories, including false events, suspected events, weak events, medium intensity events, and strong events. A 1D vector is used to represent the classification results. For example, the sample label of a strong event can be represented as [0, 0, 0, 0, 1].

[0084] That is to say, the classification result of event intensity is judged according to the added signal-to-noise ratio and whether it is noise, and TE is established according to the classification result. (i) The corresponding label vector Label (i) For example, the sample label of a strong event can be expressed as [0,0,0,0,1], and the sample label of a suspected event can be expressed as [0,1,0,0,0].

[0085] (4) Considering the potential effect of multi-channel superposition characteristics, TE (i) Perform multi-channel superposition to obtain an average and copy the TS (i) , then [(TE (i) , TS (i) ), Label (i) ] constitutes the training data set D,

[0086] in, N is the number of detectors.

[0087] Step 3, constructing a network model for evaluating the intensity of microseismic events;

[0088] Specific operations include:

[0089] (1) Construct a network model for evaluating the intensity of microseismic events;

[0090] The input of the microseismic event intensity assessment network model is a two-channel data, one of which is TE (i) , the other channel is TS (i) The overall network model can be divided into multiple front convolution blocks to extract data features of different scales and a rear fully connected layer. The convolution block includes multiple modules, which sequentially perform convolution, batch normalization, activation function, convolution, batch normalization, activation function, and pooling operations on the input; for the convolution block in front of the fully connected layer, pyramid pooling, cropping or stretching operations are used to solve the problem of inconsistent data size during actual prediction.

[0091] Among them, the size of the convolution kernel of each convolution layer is set to (number of channels × 3 × 3), and the padding is set to 1; the pooling layer uses maximum pooling; and the activation function uses the ReLU function.

[0092] (2) Construction of loss function and selection of optimization algorithm;

[0093] Cross entropy is used as the loss function of the model, and its formula is as follows:

[0094]

[0095] Among them, P is the probability vector of the network output, Loss is the loss value, which is the output probability of the network, and p i is the probability that the category is i, y is the label, if the category is i, y i =1, otherwise 0.

[0096] The optimization algorithm uses the adaptive moment estimation algorithm, which has good stability and can converge quickly.

[0097] Step 4, using the training data set as input, training the microseismic event intensity evaluation network model to obtain a trained microseismic event intensity evaluation network model;

[0098] Specific operations include:

[0099] The attenuation of the learning rate is adjusted using cosine annealing, the initial learning rate can be set to 1e-4, the batch size can be set to 256, and the training rounds can be set to 200.

[0100] The training data set is divided into training set, validation set, and test set in a ratio of 7:1:2. The training set, validation set, and test set are used as input to train the microseismic event intensity evaluation network model to obtain a trained microseismic event intensity evaluation network model.

[0101] Step 5, microseismic event intensity assessment;

[0102] Specific operations include:

[0103] The long and short time window method can be used to automatically identify events and obtain microseismic data containing rupture signals. After a series of preprocessing, including but not limited to normalization, DC removal, bandpass filtering, etc., the preprocessed microseismic data containing rupture signals are input into the trained microseismic event intensity evaluation network model to obtain the probability vector result.

[0104] The event category is determined based on the probability results of category prediction.

[0105] The prediction result of the network is a discrete probability distribution of 5 categories, whose values ​​are between 0 and 1, and the category corresponding to the maximum value is taken as the prediction result.

[0106] Attached Figure 2 Example of forward modeling. Figure 2It is a horizontal layered model, and the velocity parameters are set based on the actual well data. The observation system is set to simulate the surface reception and well reception.

[0107] Construction of training sample data, including Figure 3 The simulated microseismic data TE generated for a (i) , Figure 4 To eliminate the polarity effect of simulated microseismic data TE (i) , attached Figure 5 TS for multi-channel superimposed and expanded data (i) The label classification result of this simulated event is a medium-intensity event, that is, its label vector Label (i) is [0,0,0,1,0]. [(TE (i) , TS (i) ), Label (i) ] constitutes a training sample.

[0108] The input of the network model is a two-channel data, in which the network as a whole includes 4 convolutional blocks and 2 fully connected layers, such as Figure 6 The convolution block sequentially performs convolution, batch normalization, activation function, convolution, batch normalization, activation function, and pooling operations on the input. Its structure is shown in Figure 7 For the pooling operation before the fully connected layer, pyramid pooling is used to fix the number of parameters to solve the problem of non-uniform input data size.

[0109] [Example 2]

[0110] The embodiment of the present invention provides a microseismic event intensity assessment system based on deep learning, comprising:

[0111] A forward model building unit, used for building a forward model;

[0112] A training set construction unit, used to construct a training data set;

[0113] A model building unit, used to build a network model for evaluating the intensity of microseismic events;

[0114] A model training unit is used to train the microseismic event intensity evaluation network model using the training data set as input to obtain a trained microseismic event intensity evaluation network model;

[0115] Event evaluation unit, used for intensity evaluation of microseismic events.

[0116] [Example 3]

[0117] An embodiment of the present invention provides a computer-readable storage medium, which stores at least one computer-executable program. When the at least one program is executed by the computer, the computer executes the steps in the microseismic event intensity assessment method based on deep learning as described in Example 1.

[0118] [Example 4]

[0119] An embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the microseismic event intensity assessment method based on deep learning as described in Example 1.

[0120] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0121] The above technical solution is only one implementation mode of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the principles disclosed in the present invention, and it is not limited to the technical solution described in the above specific embodiments of the present invention. Therefore, the above description is only preferred and does not have a restrictive meaning.

Claims

1. A method for evaluating the intensity of microseismic events based on deep learning, characterized in that: The following steps are involved: Step 1, establish a forward model; Step 2: construct a training data set; Step 3, constructing a network model for evaluating the intensity of microseismic events; Step 4, using the training data set as input, training the microseismic event intensity evaluation network model to obtain a trained microseismic event intensity evaluation network model; Step 5: Evaluation of the intensity of microseismic events.

2. The microseismic event intensity evaluation method based on deep learning according to claim 1 is characterized in that: In step 1, a forward model is established, and the specific operations include: A reasonable layered velocity model is obtained based on known logging data or VSP data. The observation system simulates the surface microseismic observation system and the microseismic observation system in the well to construct the forward model. The source locations are randomly distributed near the target layer, and it is assumed that s sources are randomly generated.

3. The method for evaluating the intensity of microseismic events based on deep learning according to claim 2, characterized in that: In step 2, a training data set is constructed, and the specific operations include: Step 21, combining the moment tensor, forward modeling the microseismic data of different focal mechanisms to obtain the microseismic data without noise and with event arrival time annotations; Perform time difference correction and leveling based on arrival time information, and capture data within the event time window; Assume that the sample data obtained by forward modeling is i is the sample number, and its value range is 1, 2, 3...S, j is the channel number, and k is the sampling point sequence; Step 22, expanding the forward modeling data set; The microseismic data obtained by forward simulation are flipped, scaled, and noise with different signal-to-noise ratios are added to enhance the data, and the noise is added to the data set as a sample; The following formula is used to enhance the data: TE=(-1)r* w *T+ snr * no in se Where T is the simulated single-channel record, w is the weighting coefficient. When it is 0, it is a noise sample. When it is not 0, it indicates the scaling coefficient of the signal. r takes -1 or 1 to indicate whether to flip. sn r is the added signal-to-noise ratio; no i se To satisfy the noise of Gaussian distribution; The enhanced sample data is TE( i ), i = 1, 2, ... NE, NE is the number of samples after augmentation; Step 23: Create TE ( i ) corresponding to the label vector La b e l(i); According to the added signal-to-noise ratio and whether it is noise, the classification result of the event intensity is judged, and the label vector L corresponding to TE(i0) is established according to the classification result. a b e l(i); Step 24: Considering the potential effect of multi-channel superposition features, TE (i) Perform multi-channel superposition, average and copy to obtain T S( i), then [(TE(i), TS(i)), Label(i)] constitutes the training data set D. in, N is the number of detectors.

4. The method for evaluating the intensity of microseismic events based on deep learning according to claim 3, characterized in that: The input of the microseismic event intensity evaluation network model constructed in step 3 is a two-channel data, one of which is TE ( i) , the other channel is TS( i ); The microseismic event intensity assessment network model is divided into a plurality of front convolution blocks for extracting data features of different scales and a rear fully connected layer. The convolution block includes a plurality of modules, which sequentially perform convolution, batch normalization, activation function, convolution, batch normalization, activation function, and pooling operations on the input; for the convolution block in front of the fully connected layer, pyramid pooling, cropping or stretching operations are used to solve the problem of inconsistent data size during actual prediction.

5. The method for evaluating the intensity of microseismic events based on deep learning according to claim 4, characterized in that: The step 3 includes: Cross entropy is used as the loss function of the model, and its formula is as follows: Among them, P is the probability vector of the network output, and Loss is the loss value, which is the output probability of the network. p i is the probability that category i is y is the label, if the category is i, y i=1, otherwise 0; Adaptive moment estimation algorithm is used as the optimization algorithm.

6. The method for evaluating the intensity of microseismic events based on deep learning according to claim 5, characterized in that: The specific operations of step 4 include: The training data set is divided into training set, validation set and test set in the ratio of 7:1:

2. The training set, validation set and test set are used as input to train the microseismic event intensity evaluation network model to obtain a trained microseismic event intensity evaluation network model.

7. The method for evaluating the intensity of microseismic events based on deep learning according to claim 6, characterized in that: The microseismic event intensity evaluation in step 5 specifically includes the following operations: The preprocessed microseismic data containing rupture signals are input into the trained microseismic event intensity assessment network model to obtain the probability vector results, and the event category is discriminated according to the probability results of category prediction.

8. A microseismic event intensity assessment system based on deep learning, characterized in that: include: A forward model building unit, used for building a forward model; A training set construction unit, used to construct a training data set; A model building unit, used to build a network model for evaluating the intensity of microseismic events; A model training unit is used to train the microseismic event intensity evaluation network model using the training data set as input to obtain a trained microseismic event intensity evaluation network model; Event evaluation unit, used for intensity evaluation of microseismic events.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the microseismic event intensity assessment method based on deep learning as described in any one of claims 1 to 7.

10. A computer device, characterized in that: It includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the microseismic event intensity evaluation method based on deep learning as described in any one of claims 1 to 7.

Images