LU502639B1 - A method and system for denoising valid real-time microseismic events by DAS Ko-Well monitoring - Google Patents

Abstract

The present invention relates to a method and a system for denoising valid real-time microseismic events by DAS co-well monitoring, which belongs to the field of microseismic monitoring technology. First the training data set is obtained, then the training data set is used to train the pre-built denoise model to obtain the trained denoise model, and finally the real-time events are obtained with microseismic by DAS Ko-well monitoring, the training denoise model is then used to remove the pipeline waves from the real-time microseismic events to obtain the denoised microseismic events, this enables the efficient removal of pipeline waves from the real-time microseismic events monitored by the DAS Ko-Well using a denoising model, to avoid the impact of pipeline waves on the subsequent processing of the microseismic events. In addition, the invention can be set to collect new pipeline wave data according to the cycle to update the trained denoise model and use it in the next cycle, so that the denoise model can always have good denoise performance as the hydraulic fracturing progresses.

Description

Description

A method and system for denoising valid real-time 02699 microseismic events by the DAS Ko-Well

monitoring

Technical part

The present invention relates to the technical field of microseismic monitoring and more particularly to a method and system for denoising valid real-time microseismic events by the

DAS Ko-Well monitoring based on deep learning.

technology in the background

Hydraulic fracturing uses high-pressure fluids in

Injected shale deposits to create complex man-made fractures that shape the

Increase reservoir connectivity and increase production per well. In hydraulic fracturing, a high-pressure pump is used to

Wellhead injected a fracturing fluid into the well being fractured, resulting in the underground reservoir as a result of stress changes

Microseismicity is generated. Microseismic monitoring technology monitors the fracturing process and evaluates the fracturing effect by monitoring the microseismic signals generated during the hydraulic fracturing

process are generated, which in turn controls and optimizes the technical parameters. Microseismic monitoring consists of the following components:

Effective event recording, effective event signal amplification (noise reduction), location of microseismic sources, microseismic

Source analysis and reservoir stress analysis.

DAS is an emerging data acquisition technology that has evolved in recent

years, enabling the measurement of strain signals along the axial direction of an optical fiber by detecting the phase change of the backscattered light signal generated by a laser pulse in the internal

Scatter of the fiber is generated. In DAS monitoring of hydraulic fracturing 1

Description, the optical fibers are placed along the horizontal borehole, which allows measurements 99 of the entire borehole section and a spatial scan of 1 m, therefore the DAS has the advantage of a wide orientation and a high

data density.

DAS monitoring is generally merged into monitoring of accounts

Wells and the monitoring of neighboring wells divided. At the Ko-Well

Monitoring is the monitoring well and the fracturing

Drilling around the same horizontal well, and the fracturing pump generates a strong pipeline wave during operation, so there is a strong pipeline wave of microseismic valid events originating from the ko-well monitoring. The pipeline wave has a high amplitude that exceeds the amplitude of the microseismic event; The pipeline wave propagates along the

Monitoring well and has a strong periodicity, so that the

Pipeline wave disturbs the microseismic event more and cannot be ignored. At the same time, the pipeline wave shows a high degree of similarity with the microseismic event, with the frequency band of the pipeline wave partially overlapping the frequency band of the microseismic event ım

Frequency range overlaps; In the time domain, the waveform resembles this

Pipeline wave of the valid microseismic event, so it is difficult to separate the pipeline wave from the valid microseismic event.

Also, in the DAS neighborhood survey, the survey well and the frac well are different horizontal wells, and there is no such pipeline wave propagating along the frac well in the data, i. H. only microseismically valid events are included. Since the same monitoring well and adjacent well monitoring, the same monitoring well and monitoring system for the

Using data acquisition, the valid microseismic events from this adjacent wellbore survey show a high degree of similarity to 2

Description valid microseismic events from the same borehole monitor 2 >.

Generally, when a valid microseismic event is detected, subsequent processing of that microseismic event is required, including source location, mechanism inversion, etc., but degrades the pipeline wave generated by DAS co-well monitoring however, as a uniform noise with high amplitude and

Periodicity significantly affects the subsequent analysis of the valid microseismic event, so it must be separated.

Therefore, there is an urgent need for a denoising method and system that extracts pipeline waves from real-time microseismic

can effectively remove events monitored by DAS Ko-Wells.

content of the invention

The purpose of the present invention is to provide a method and a system for

Denoising of valid real-time microseismic events by the DAS

To provide ko-well monitoring that pipeline waves from microseismic mixing

Effectively remove real-time events for DAS co-well monitoring and

Effects on the subsequent processing of microseismic events can be avoided.

To achieve this, the invention offers the following solutions:

A method for denoising valid real-time microseismic

Events through DAS Ko-Well monitoring, where the denoising process includes:

obtaining a training dataset; the training dataset contains a count of valid pipeline wave microseismic events and a valid non-

Pipeline wave microseismic event corresponding to any of the described valid

corresponds to pipeline wave microseismic events;

Training the ready-made denoising model using the 3

Description

training data set to obtain a trained denoising model; 1006688

Valid real-time microseismic clearances are provided by DAS Ko-Well

Obtained monitoring, and using the valid microseismic

Real-time exposures as input are the pipeline waves in the valid real-time microseismic events using the trained

Removed denoising model to get denoised valid microseismic events.

Before removing pipeline waves from the real-time microseismic

validity events using the trained denoising model with the real-time microseismic validity events as input, the

Denoising method further includes: determining whether the

Monitoring time for DAS co-well monitoring a preset

period has reached; if yes, the training dataset is created using the

Updated pipeline wave data obtained during DAS ko-well monitoring in the current cycle to obtain an updated dataset and the updated dataset is used to calculate the trained

Update the denoising model to get a trained denoising model for the next cycle.

A system for denoising valid real-time events by the

DAS Ko-Well monitoring, where the denoising system includes:

A data set acquisition module for acquiring a training data set; where the training data set contains a number of valid microseismic mixers

pipeline wave events and valid non-pipeline wave microseismic

events corresponding to each of the applicable pipeline wave microseismic

correspond to events;

A training module for training a ready-made denoising model using the training data set to create a trained

to obtain denoising model; 4

Description

A denoising module for capturing valid microseismic oo

Real-time events through DAS Ko-Well monitoring and to use the valid real-time microseismic events as input, to remove

Pipeline waves from the valid real-time microseismic events below

Using the trained denoising model to obtain denoised valid microseismic events.

Optionally, the denoising system further comprises an update module to determine if the monitoring time for the DAS co-well monitoring has reached a preset period of time; if so, the training data set is updated using the pipeline wave data collected during DAS co-

Well-monitoring received in the current cycle have been updated to an

Obtain data set, and the updated data set is used to update the trained denoising model to a trained

to obtain a denoising model for the next cycle.

According to the specific embodiments provided by the present

Invention are provided, the invention discloses the following technical

Effects:

The present invention is to provide a method and a

System for denoising valid real-time microseismic events by DAS Ko-Well monitoring, first a training data set is created, the training data set containing a number of valid microseismic

pipeline wave events and valid non-pipeline wave microseismic

events corresponding to each of the applicable pipeline wave microseismic

correspond to events. The training data set is then used to train the prefabricated = denoising model to the trained

to obtain a denoising model. Finally, the valid real-time microseismic events are obtained by DAS Ko-Well monitoring, and to use the valid real-time microseismic events as input,

Description for removing pipeline waves from valid real-time microseismic 902699

events using the trained denoising model to obtain denoised valid microseismic events such that the pipeline

Ripples in the valid real-time microseismic events can be effectively removed by building a denoising model to reduce the influence of the

Pipeline waves on the subsequent processing of valid microseismic

to avoid events. In addition, the invention can be set to collect new pipeline wave data in accordance with the cycle to update the training data set and the trained

Keep updating the noise reduction model and use it in the next cycle, so that the denoising model is always good

has denoising power as hydraulic fracturing progresses.

Description of the attached drawings

In order to make the technical solutions in the embodiments of the invention or in the prior art more clear, a brief

Description of the accompanying drawings given in the

Embodiments are to be used. It will be evident that the accompanying drawings in the following description are only a few

Illustrate embodiments of the invention and that others attendant

Drawings based on these drawings without any creative

Effort on the part of a person of ordinary skill in the field

technology can be obtained.

Figure 1 shows a process flow diagram of Example 1 of the present

Invention provided denoising method;

Figure 2 is a schematic block diagram of the denoising process

Example 1 of the present invention;

Figure 3 shows a schematic diagram of the lattice structure of the

denoising model of Example 1 of the present invention; 6

Description

Figure 4 shows a system block diagram of the denoising system of example 0 2 of the present invention.

Detailed description

The technical solutions in the embodiments of the present

Inventions are described clearly and fully in the following in connection with the attached drawings in the embodiments of the present invention, and it is clear that the described embodiments are only a part of the embodiments of the present invention and not all of them.

Based on the embodiments of the present invention, all other embodiments achieved by those skilled in the art without any creative work fall within the scope of the present invention.

The purpose of the present invention is to provide a method and a system for

Denoising of valid real-time microseismic events by the DAS

To provide ko-well monitoring that microseismicizes pipeline waves

Effectively remove real-time events for DAS co-well monitoring and

Effects on the subsequent processing of microseismic events can be avoided.

In order that the above objects, features and advantages of the invention may be more clearly and fully understood, the invention will be described in more detail below in connection with the accompanying drawings and specific embodiments.

Example 1:

When monitoring hydraulic fracturing with DAS, DAS has, in addition to the

The advantages of the broad orientation and the high data density also have the advantage that a very large amount of monitoring data can be obtained as a continuous, long-lasting process and also a large volume of data.

For example, the DAS sample is 1 millisecond, and at a single

Drilling of 2000 wells can be scanned at a spatial depth of 7

Description

Im yielding nearly 650GB of data per day, making it impossible to process by manual or traditional 902699 methods and requiring the use of artificial intelligence methods to improve processing efficiency.

The non-uniform temperature during the cooling process of the optical fiber during manufacture leads to non-uniform distribution of the

Material density, and the density of the doped to enhance the scattering effect

Particle is also distributed unevenly, resulting in a difference ım

Index of refraction in the fiber and thus leads to Rayleigh scattering throughout

space of the fiber is present. Because the density of the fibers at any point throughout

Fiber is random, the refractive index at each point is also the microscopic

fiber randomly distributed, and the Rayleigh scattered light does not drift due to

spread in frequency.

Before introducing the denoising method of this embodiment, the way the DAS is used to monitor the hydraulic

fracturing process is used, and the most important processes of hydraulic

fracturing described:

The DAS is structured as follows: The distributed fiber optic based microseismic monitoring system for horizontal wells includes a

Metal body, single-mode fiber, with an armored fiber optic cable attached to the

Attached to the outside of the metal case, a special single-mode fiber inside the armored fiber optic cable, a DAS modulation and

Demodulation instrument placed near the wellhead and the

Demodulator DAS signal port connected to the special single-mode fiber outside the case.

The key processes in hydraulic fracturing are: (1) The metal jacket and armored fiber optic cable are slowly and simultaneously lowered into the finished wellbore. 8th

Description (2) At the wellhead, attach the ring metal clamp to the joint between the two lengths of metal tubing to secure the armored fiber optic cable and protect it from movement and/or damage as the tubing is lowered. (3) With a high-pressure pump, the cement slurry from the

pumped from the well bottom to the well head so that the cement slurry along the

annulus between the outer wall of the metal casing and the wellbore. After the cement slurry sets, it permanently holds the metal casing, armored fiber optic cable, and formation rock together. (4) Connect the single-mode fiber outside the casing inside the armored fiber optic cable to the DAS signal input of the DAS modulation and demodulation equipment at the wellhead. (5) Collect 3D ground seismic data from the area around the horizontal borehole and preprocess it to generate a 3D seismic

longitudinal velocity data body, then the 3D seismic

Longitudinal velocity data body calibrated, adjusted and with the

Velocity data of the acoustic log updated to show the preliminary seismic longitudinal velocity field of the formation around the horizontal

to get a borehole. (6) The directional injection of the metal case is carried out sequentially at predetermined injection points in the borehole, and the injection signals generated during the directional injection are measured using optical

Single-mode fibers placed downhole outside of the casing, and the

DAS modulation and demodulation instrument near the wellhead and these signals were used to calibrate and update the initial seismic velocity field in step (5) to determine the

Velocity field for final microseismic analysis of

Receive hydraulic fracturing events. 9

Description (7) During hydraulic fracturing, this system provides 07699 microseismic monitoring using armored fiber optic cables permanently routed outside the metal housing. The data will be sent with

Single-mode fiber optic cables that are laid outside the borehole and to the

Demodulation to a DAS modulation and demodulation device near the

wellhead, a continuous record is established of valid microseismic events occurring during fracturing of the subsurface formation by hydraulic fracturing operations in adjacent or shared wells. (8) In combination with the longitudinal one obtained in step (6).

Velocity field of the subsurface layers, information about the microseismic effective event can be obtained, the information about the

Time of occurrence, the three-dimensional spatial location and the

Include energy magnitude of the microseismic effective event. (9) Upon completion of the hydraulic fracturing, an analysis of the seismic swelling mechanism and seismic magnitude will be performed on the

based on the recorded microseismic events to determine the fracture mechanism of the underground reservoir of the microseismic events and the distribution properties and laws of the fractures in the underground reservoir after transformation. The above

Information is integrated to provide a valid and reliable qualitative and quantitative assessment of the impact of hydraulic fracturing of the

reservoirs at this horizontal well.

In step (7), the valid microseismic events from the co-

Well monitoring strong pipeline wave interference that the subsequent

Analysis using valid microseismic events in steps (8) and (9), i.e. the microseismic valid ones

Events Obtained from Recording in DAS Ko-Well Surveillance

Description will contain pipeline wave interference, and the suppression of yon

Pipeline Waves is required to extract valid microseismic events that do not contain pipeline waves. However, there are currently none

Denoising techniques for real-time microseismic events generated by DAS

Ko-wells are monitored under similar procedures currently exist: 1)

Separation methods for conventional microseismic data, e.g. B. Non-ML

methods and ML-based methods; 2) separation methods for conventional

Geophones with waves from the well pipe, which are generally physical

Suppression methods and data separation algorithms include. physical

Suppression methods generally involve adding

Dampers to piping and data separation algorithms include im

Generally adapted phase subtraction methods, filtering in time-space

Area, vector median filtering, and SVD decomposition methods. However, none of these existing techniques can solve the problem of effectively eliminating

Resolve pipeline wave interference within the microseismic valid events obtained from DAS co-well monitoring.

Machine learning (ML) methods rely on probabilistic models for learning from data. Able to extract complex patterns and valid features from data streams, they have the potential to generalize and derive new features and mechanisms that can improve predictions about systems, and are a practical tool that complements existing data analysis methods. The

Complexity of the earth structure, the indirect character of geophysical

Observations and the sophistication of numerical tools make big

Challenges for the development of geophysical research. The

Combination of data-driven disciplines and model-driven techniques helps to study difficult problems in seismology in depth and the

to accelerate the acquisition of new knowledge and the application of the 11

Description of machine learning in seismology is increasing. In this context, machine learning methods are an excellent choice.

Based on this, this embodiment is used to

To provide a method for denoising valid real-time microseismic events through DAS co-well monitoring as illustrated in Figures 1 and 2, the denoising method comprising:

S1: Obtaining a training data set; where the training dataset is a

comprises a number of valid pipeline wave microseismic events and valid non-pipeline wave microseismic events corresponding to each of the valid pipeline wave microseismic events;

Specifically, S1 may include: (1) Data Generation: The pipeline wave-free valid microseismic

Events are obtained from the DAS neighborhood monitor performed in the

Ramp from this work area or other work areas, preferably from

Working areas with similar deposit formations. The

Pipeline wave data is then summed with the microseismic pipeline wave events from the neighboring wells to determine the microseismic

Obtain pipeline wave events corresponding to microseismic pipeline wave events.

It should be noted that the same parameters are used to collect data for adjacent and common well monitoring, but it is not necessary for the summed pipeline wave data and the non-pipeline wave microseismic events to be collected simultaneously. A single pipeline wave-free microseismic event contains a total of Nt (may be 1000) channels, each with a total of Ns (may be 1000) sample points.

The pipeline wave data also includes Nt (may be 1000) channels with a total of Ns (may be 1000) sample points.

Preferably, this embodiment can also store the data during the 12th

Description

Amend data generation process to improve generalizability of dataset 502639.

In particular, after obtaining the valid non-pipeline wave microseismic event, the denoising method of this embodiment further comprises: performing data augmentation on the valid microseismic event

Event that is not a pipeline wave for a new valid microseismic

Obtain event that is not a pipeline wave. Data scaling for pipe-wave free microseismic events may include: Scaling the pipe-wave free microseismic valid events to different ones

amplitude scales, where the amplitude scale is the data size scale of the

sampling points of the pipeline wave-free microseismically valid events, i.e. H.

scaling the data size of each sampling point of the pipeline wave-free microseismic valid events by a specified percentage; Shifting the position of each channel included in the microseismic valid event without the

pipeline wave is involved, e.g. B. by panning left and right or up and down to change the position of each channel, e.g. B. by swapping the positions of channel 1 and channel 3; replacing the normal channel in the non-pipelined wave microseismic event with an anomaly channel, d. H. Constructing an anomaly channel in non-pipeline microseismic

Wave event, and an anomaly channel can be a bad channel or a missing channel, a bad channel is a channel in which the data values are particularly large or small in the whole channel, and a missing channel is a

Channel missing data in the entire channel.

After obtaining the pipeline ripple data, the denoising method of this embodiment further includes: performing data scaling on the pipeline ripple data to obtain the new pipeline ripple data. The

Data augmentation of the pipeline wave data may include: scaling the pipeline wave data to different amplitude scales, where the 13th

Description

amplitude scale is the data size scale of the sample points of the pipeline wave data => i.e. H. the data size of each sample point of the pipeline wave data is scaled by a specified percentage.

The method described above allows for data augmentation of the acquired pipeline wave data and valid non-pipeline wave microseismic events, and the resulting new pipeline wave data and new valid non-pipeline wave microseismic events are used to form the valid pipeline wave microseismic events . (2) Repeat the above steps until N valid microseismic

Non-pipeline wave events and valid microseismic wave events

pipeline waves corresponding to any valid non-pipeline-wave microseismic event are obtained to obtain the training data set.

The training dataset for this implementation is described as follows: includes 2D tube wave-containing microseismic valid events and 2D

Pipeline wave-free microseismic valid events that are both coincident and of the same magnitude. The training data set of this example may contain Nk (1000) valid microseismic events, so the data body (with and without valid pipeline wave microseismic events) is both Nk*Ns*Nt.

S2: Training the ready-made denoising model using the

training data set to obtain a trained denoising model;

Before S2, the denoising method of this embodiment can further

Include: pre-processing the training data set to obtain a pre-processed data set, and executing S2 with the pre-processed data set as a new training data set. If all actually monitored data within the training data set are pre-processed in the same steps, the pre-processing of the training data set can be 14

Description

Include: anomaly channeling, averaging, outlier removal, and normalization of the data for both valid microseismic

pipeline wave events as well as valid microseismic non-

pipeline wave events. The processing of abnormal channels refers to the

Identification of abnormal channels and subsequent interpolation and replacement with adjacent channels. The averaging refers to the

Averaging the current data Ns*Nt. When removing outliers

Outliers identified and then replaced by interpolation, where the

Outliers are values that fall outside the interval dividing 95% of the

maximum value and 95% of the minimum value of the current data Ns*Nt.

Data normalization can be done using any of the existing ones

Normalization methods are carried out, such as B. Maximum-minimum

normalization, mean normalization, etc. It should be noted that the

Training dataset with any combination of the above

Preprocessing methods can be preprocessed.

The denoising model used in this embodiment can be a deep learning model whose modular architecture can be as follows: The

Features of the training data are represented in spatial scales by a pair of

Downsampling and upsampling operations shrink and grow. The

Extraction of features at different spatial scales can be achieved by multiple pairs of such upsampling and downsampling operations, and to obtain the feature details at different scales, downsampling and upsampling are performed at the same spatial scales

scales connected by an attention-mechanism layer, and the

Attention Mechanism Layer contains a convolutional layer, a maximal

pooling layer, an average pooling layer, and a connection layer with an up-sampling function that takes the input and input dimensions of the

attention mechanism layer consistent. As shown in picture 3,

Description, the denoising model of this embodiment comprises a sequentially “9 connected input layer, a feature extraction layer, a fully connected layer and an output layer, where the

feature extraction layer a sequentially connected first convolution block, a maximum pooling layer, a second convolution block, a maximum

pooling layer, a third convolutional block, a maximum pooling layer, a fourth convolutional block, a fifth convolutional block, an upsampling layer, a sixth convolutional block, an upsampling layer, a seventh

convolution block, an upsampling layer and an eighth convolution block. The first

Convolution block is connected to the eighth convolution block by a

Attention mechanism layer connected and has the same spatial

Scale, the second convolution block is connected to the seventh convolution block by a

Attention mechanism layer connected and has the same spatial

Scale, the third convolution block is connected to the sixth convolution block by a

Attention mechanism layer connected and has the same spatial

scale, and the fourth convolution block has the same spatial scale as the fifth convolution block. The spatial scale of the second convolution block is half that of the first convolution block, the spatial scale of the third

The spatial scale of the fourth convolution block is half that of the third

convolution blocks.

More precisely, the first, the second and the third contain convolution blocks

Two first convolution layers each and the fourth convolution block a first

convolution layer. The first convolutional layer uses a 2D convolutional kernel with a convolution size of 3x3 and each first convolutional layer is activated with a corrected linear unit, i.e. with a ReLU activation function with padding and stride of 1 and 2, respectively. The parameters of the convolutional layers 1-7 are then : 32x3x3, 32x3x3, 64x3x3, 64x3x3, 128x3x3, 128x3x3, 256x3x3, where the 16th

Description first dimension the number of folds and the last two dimensions the 994699

folding dimensions are. The fifth convolution block contains a second

Convolutional layer, and the sixth, seventh and eighth convolutional block each contain two second convolutional layers with 2D convolution kernels and a convolution size of 2 x 2. Each second convolutional layer is activated with a corrected linear unit, i.e. H. a ReLU activation function with padding and stride of 1 and 2, respectively.

The input to the denoising model of this embodiment is a valid pipeline wave microseismic event, the output is the predicted denoised valid microseismic event, and the

Label is the valid unpipelined microseismic event, which corresponds to the valid microseismic event with a pipeline. When training the

In the denoising model, the error between the predicted effective microseismic events after denoising and the effective microseismic events without pipeline waves is used to calculate the

Update network parameters of the denoising model, in particular, the MSE (Mean Square Error) can be used as a loss function to calculate the error of the denoising model, and with the

Trained stochastic gradient descent optimization method. The input to the denoising model of this embodiment is a valid pipeline wave microseismic event, the output is the predicted denoised valid microseismic event, and the

Label is the valid unpipelined microseismic event, which corresponds to the valid microseismic event with a pipeline. during the

training, the dynamic learning rate is set to an initial value of 0.0001 and every 50 iterations, the dynamic learning rate is reduced by half

Stack size is set to 40 and number of iterations can be set to 200. 17

Description

In this embodiment, when training the denoising model, the 502639

The training data set can also be divided into a training set and a test set in an 8:2 ratio. The training set is used to train the denoise model and the test set is used to test the trained denoise model and the model with the highest accuracy is called the trained

Denoising model selected.

In this implementation, model training is performed entirely on the GPU

Image processing unit carried out.

S3: Detection of valid real-time microseismic events by DAS co-

Well monitoring and using valid real-time microseismic

Events as input, removing pipeline waves from the valid real-time microseismic events using the trained one

noise model to obtain denoised valid microseismic events.

When processing the actual monitoring data for analysis, the valid real-time microseismic events from the DAS Ko-Well

Monitoring recorded, using the valid real-time microseismic events

Pipeline ripple noise included. The valid real-time microseismic

Events are preprocessed in the same steps used for the

Training data set are carried out. The preprocessing can

Processing of damaged channels (interpolation and replacement with

adjacent channels), which include averaging, outlier removal (replacing outliers with interpolation), and normalizing the data.

The pre-processed real-time microseismic events are then fed into the trained denoising model to generate the denoised microseismic

get events.

Since hydraulic fracturing is a continuous process, the denoising method of this embodiment can be applied to the continuously acquired microseismic real-time valid events during the 18th

Description can be applied throughout the hydraulic fracturing process. To the trained 502639

Being able to adapt the denoising model to the whole process of hydraulic fracturing and always having the best denoising performance for the detected microseismic FE real-time valid events is demonstrated in this

Embodiment during the monitoring process a threshold for the

Model update set. if the monitoring time a certain

When the period (e.g. 1 day) is reached, a model update process is initiated to update the trained denoising model. In particular, this includes

Denoising method of this embodiment before removing

Pipeline waves from the valid real-time microseismic events below

Using the trained denoising model with the valid real-time microseismic events as input: (1) determining whether the monitoring time for the DAS co-well monitoring has reached a preset time period;

The default period can be set artificially, e.g. B. 1 day, 2

days. (2) If yes, the training data set is generated using the pipeline

Wave data updated during DAS Ko-Well monitoring in the current

cycle have been obtained to obtain an updated data set and the updated data set is used to update the trained denoise model to obtain a trained denoise model for the next cycle.

The training data set update process may include the following: Most of the data collected by DAS Ko-Well Monitoring is pipeline wave data, while the remaining part of the data is microseismic valid events containing pipeline waves. during one

Monitoring cycle, when a valid real-time microseismic event is monitored, the trained denoising model of the current cycle is used for 19

Description uses denoising. Pipeline wave data may also be collected periodically roo during the current cycle, i.e. as the fracturing process progresses, a certain length of pipeline wave data is collected according to a specified time to perform a pipeline wave data update, which is then used to update the training data set, and the updated pipeline wave data is used to update the pipeline wave containing valid microseismic events to obtain the updated data set.

The trained denoising model includes sequentially connected

Input layer, a feature extraction layer, a fully connected

layer and an output layer and updating the trained

denoising model using the updated data set

Include the following: Freezing the network parameters of the

Feature extraction layer, updating the mapping relationship between the

feature extraction layer and the fully connected layer below

Using the updated dataset to get new network parameters and updating the trained denoising model, which is next

cycle is used.

Specifically, during the update process, the network structure of the

Feature extraction layer frozen, i.e. the parameters of

feature extraction layer remain unchanged, and the mapping relationship between the feature extraction layer and the fully connected layer is trained using all the updated pipelines containing valid microseismic events and some of the old pipelines containing valid microseismic events in the training dataset (the number of data remains for both Nk to um to keep the number consistent), i.e. the features output from the last layer of the feature extraction layer are extracted and the model is fine-tuned using the new data to

Description to restore the mapping relationship from the feature extraction layer to the fully "200 connected layer. The training speed can be improved because only the mapping relationships between the

The feature output layer and the fully linked layer of the model can be trained by migration without training the entire model.

Example 2:

This embodiment is used to demonstrate a method for denoising valid real-time microseismic events by the DAS co-well

Provide monitoring as shown in Figure 4, where the denoising system includes:

A data set acquisition module M1 for acquiring a training data set; where the training data set contains a number of valid microseismic mixers

pipeline wave events and valid non-pipeline wave microseismic

events corresponding to each of the applicable pipeline wave microseismic

correspond to events;

A —training module M2 for training a ready-made

denoising model using the training data set to obtain a trained denoising model;

A denoising module M3 for acquiring valid microseismic

Real-time events through DAS Ko-Well monitoring and to use the valid real-time microseismic events as input, to remove

Pipeline waves from the valid real-time microseismic events below

Using the trained denoising model to obtain denoised valid microseismic events.

Preferably, the denoising system further comprises an update module to determine whether the monitoring time for the DAS co-well monitoring has reached a preset period of time; if so, the training data set is updated using the pipeline wave data collected during the DAS co-21st

Description

Well-monitoring received in the current cycle have been updated to an

Obtain data set, and the updated data set is used to update the trained denoising model to a trained

to obtain a denoising model for the next cycle.

Each of the embodiments in this description focuses on the

differences from the other embodiments, and it suffices to refer to the other for like and similar parts of each embodiment. For the system disclosed in the examples, the description is simpler since it is the one shown in FIGS

Examples corresponds to the method disclosed and the relevant parts are ım

Method section described.

The principles and the implementation of the invention are described in this paper by means of specific examples, the description of the above-mentioned embodiments only serving to illustrate the method of the invention and its

understand core ideas; At the same time, changes in the specific implementation and in the area of application based on the ideas of the invention may arise for a person skilled in the art. In summary, the content of this

Description should not be construed as limiting the present invention. 22

Claims (10)

Claim LU502639 Claims 1. A method for denoising valid real-time microseismic events through DAS Ko-well monitoring, characterized in that the denoising method comprises: obtaining a training data set; the training data set includes a number of valid pipeline wave microseismic events and one valid non-pipeline wave microseismic event corresponding to each of the described valid pipeline wave microseismic events; training the pre-built denoise model using the training data set to obtain a trained denoise model; Valid real-time microseismic events are obtained through DAS ko-well monitoring, and using the valid real-time microseismic events as input, the pipeline waves in the valid real-time microseismic events are removed using the trained denoising model to obtain denoised valid microseismic get events. 2. A method for denoising real-time microseismic valid events by DAS co-well monitoring according to claim 1, characterized in that obtaining a training data set comprises in particular: detecting a pipeline wave-free microseismic valid event by DAS neighboring well monitoring and pipeline wave data by DAS Ko-Well Monitoring; summing the pipeline wave data and the pipeline wave-free microseismic valid event to obtain a pipeline wave-containing microseismic valid event corresponding to the pipeline wave-free microseismic valid event; Repetition of step "Detection of a pipeline wave-free microseismic valid event by DAS neighboring well monitoring 1 claims LU502639 and pipeline wave data by DAS Ko-Well monitoring; Summing the pipeline wave data and the pipeline wave-free microseismic valid event to obtain a pipeline-wave-containing microseismic valid event corresponding to the pipeline-wave free microseismic valid event" through N non-containing pipeline microseismic waves events and the pipeline microseismic waves corresponding to each non-containing pipeline microseismic wave event -Events are received and the training data set is created. 3. A method for denoising real-time microseismic valid events by DAS Ko-Well monitoring according to claim 2, characterized in that after obtaining the pipeline wave-free microseismic valid events, the denoising method further comprises: data augmentation of the pipeline wave-free microseismic valid events by obtain new pipeline wave-free microseismic valid events; The data expansion of the excluded microseismic valid pipeline wave-free events includes in particular: scaling the excluded microseismic valid pipeline wave-free events to different amplitude scales; shifting the position of each channel included in the excluded pipeline wave-free microseismic events; and replacing the normal channels contained in the excluded pipeline wave-free microseismic events with anomalous channels. 4. A method for denoising real-time microseismic valid events through DAS co-well monitoring as claimed in claim 2, characterized in that after obtaining pipeline wave data, the denoising method further comprises: performing data augmentation on the pipeline wave data to get new pipeline wave data; 2 Claims LU502639 The data expansion of the pipeline wave data includes in particular: Scaling of the pipeline wave data to different amplitude scales. 5. A method for denoising real-time microseismic valid events by DAS Ko-Well monitoring according to claim 1, characterized in that prior to training a previously constructed denoising model using the training data set, the denoising method further comprises: pre-processing the training data set to obtain a pre-processed data set and use the pre-processed data set as a new training data set; The pre-processing of the training data set includes in particular: the processing of the anomaly channel, the averaging, the removal of outliers and the normalization of the data for the valid microseismic events of the pipeline waves and the valid microseismic events of the non-pipeline waves. 6. A method for denoising real-time microseismic valid events by DAS co-well monitoring according to claim 1, characterized in that the denoising model comprises a sequentially connected input layer, a feature extraction layer, a fully connected layer and an output layer; wherein the feature extraction layer comprises a first convolution block, a maximum pooling layer, a second convolution block, a maximum pooling layer, a third convolution block, a maximum pooling layer, a fourth convolution block, a fifth convolution block, an upsampling layer, a sixth convolution block, an upsampling layer, a seventh convolution block, an upsampling layer and an eighth convolution block connected in series; The first convolutional block is connected to the eighth convolutional block by an attention mechanism layer and on the same spatial scale; the second 3 Claims LU502639 convolution block is connected to the seventh convolution block by an attention mechanism layer and on the same spatial scale; the third convolutional block is connected to the sixth convolutional block by an attention mechanism layer and on the same spatial scale; and the fourth convolution block is on the same spatial scale as the fifth convolution block. 7. A method for denoising microseismic valid real-time events by DAS Ko-Well monitoring according to claim 1, characterized in that when training a previously created denoising model using the training data set, MSE is used as the loss function and a stochastic gradient descent optimization method used for training. 8. A method for denoising real-time microseismic valid events by DAS Ko-well monitoring according to claim 1, characterized in that prior to removing pipeline waves from the real-time microseismic valid events using the trained denoising model with the real-time microseismic valid events as input, the denoising method further comprises: determining whether the monitoring time for the DAS ko-well monitoring has reached a preset period of time; If yes, the training dataset is updated using the pipeline wave data obtained during the DAS ko-well monitor in the current cycle to obtain an updated dataset and the updated dataset is used to update the trained denoising model to get a trained denoising model for the next cycle. 9. A method for denoising valid real-time microseismic 4 Claims LU502639 events through the DAS Ko-Well monitoring according to claim 8, characterized in that the trained denoising model comprises a sequentially connected input layer, a feature extraction layer, a fully connected layer and an output layer, the updating of the trained denoising model using the updated In particular, the data set comprises: freezing the network parameters of the feature extraction layer and updating the mapping relationship between the feature extraction layer and the fully connected layer using the updated data set. 10. A system for denoising real-time valid events through DAS co-well monitoring, characterized in that the denoising system comprises: a data set acquisition module for acquiring a training data set; wherein the training data set comprises a number of valid pipeline-wave microseismic events and valid non-pipeline-wave microseismic events corresponding to each of the valid pipeline-wave microseismic events; A training module for training a pre-built denoise model using the training data set to obtain a trained denoise model; A denoising module for capturing valid real-time microseismic events through DAS ko-well monitoring and using the valid real-time microseismic events as input, removing pipeline waves from the valid real-time microseismic events using the trained denoising model to obtain denoised valid to receive microseismic events.