CN115932947A - Microseism P wave S wave separation method based on CNN-LSTM and multi-head attention mechanism - Google Patents
Microseism P wave S wave separation method based on CNN-LSTM and multi-head attention mechanism Download PDFInfo
- Publication number
- CN115932947A CN115932947A CN202211578756.5A CN202211578756A CN115932947A CN 115932947 A CN115932947 A CN 115932947A CN 202211578756 A CN202211578756 A CN 202211578756A CN 115932947 A CN115932947 A CN 115932947A
- Authority
- CN
- China
- Prior art keywords
- layer
- wave
- data
- attention mechanism
- head attention
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention belongs to the field of microseism monitoring, and relates to a microseism P-wave S-wave separation method based on a CNN-LSTM and multi-head attention mechanism. The technology comprises the following steps: constructing a data set, adopting micro-seismic data acquired on a dry hot rock hydraulic fracturing field, and manually marking the data, wherein the data set comprises micro-seismic data of different time periods and different background noise intensities; inputting the data set into a neural network consisting of a convolutional layer, an excitation layer, a pooling layer, a random hidden neuron (Dropout) layer, a long-short term memory (LSTM) layer, a multi-head attention mechanism layer and a full-connection layer for training, and storing the model after the training is finished; and inputting the micro seismic data to be separated into the trained model for separation, and finally obtaining the separation result of the P wave S wave.
Description
Technical Field
The invention belongs to the field of microseism monitoring, and particularly relates to a microseism P wave S wave separation method based on a CNN-LSTM and multi-head attention mechanism.
Background
At present, the key technology for developing the dry and hot rock resources is a hydraulic fracturing technology, high-pressure fluid is injected into a reservoir stratum to generate complex artificial fractures, and then the internal heat of the dry and hot rock is extracted. In the hydraulic fracturing construction process, the problems of difficult positioning, difficult observation of the size of an event, unclear fracture pressing space trend and the like are often encountered. In order to solve the problems, the monitoring and analysis of the whole hydraulic fracturing process by using a large number of observable micro-seismic events generated in the fracturing process become important, and the method is the premise of realizing the efficient development of the hot dry rock and realizing the safe production.
The microseism monitoring technology mainly comprises effective event picking, effective event signal strengthening, microseism seismic source positioning and the like. During actual monitoring, the situation of low signal-to-noise ratio is often encountered, and due to weak energy of the micro-seismic signal, the micro-seismic signal is easily submerged by background noise, so that the micro-seismic signal cannot be accurately identified, and further subsequent analysis work is influenced. For example, when a microseismic source is positioned, because the microseismic source has strong sensitivity to the arrival time of P waves and S waves, an error of 0.01 second can be converted into a large position error, so that guidance and optimization of engineering parameters cannot be provided for hydraulic fracturing.
In general, microseismic monitoring is a long and continuous process, and data acquired by a detector is mass data with TB as a unit. At this time, the requirement of real-time monitoring of the micro-earthquake cannot be guaranteed by manually picking up signals or using a conventional method.
In summary, it is important to provide a method capable of overcoming noise interference and accurately and efficiently separating P-waves and S-waves under site construction conditions.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention provides a microseism P wave S wave separation method based on a CNN-LSTM and a multi-head attention mechanism.
The present invention has been accomplished in such a manner that,
a microseism P wave S wave separation method based on a CNN-LSTM and multi-head attention mechanism comprises the following steps:
s1: adopting effective microseism data actually collected on a dry hot rock hydraulic fracturing site to obtain 10000 channels of microseism data, wherein each channel of microseism data comprises P wave and S wave signals;
s2: preprocessing each trace of micro seismic data, cutting the length of the data into 5000 sampling points, and manually labeling the data to finally form a training data set;
s3: building a neural network model consisting of a convolutional layer, an excitation layer, a pooling layer, a random hidden neuron layer, a long-short term memory layer, a multi-head attention mechanism layer and a full connection layer, and starting training by adopting the training data set in the step S2;
s4: after the training is finished, the neural network model is stored, the data to be separated is input into the model for separation, the probability value of whether each sampling point is a P wave or an S wave is output, and finally the separation result of the P wave and the S wave is obtained.
Further, step S2 specifically includes:
and intercepting the acquired data into a uniform length, marking the position of P wave or S wave in the data as 1, and marking the rest positions as 0.
Further, S3 specifically includes:
the neural network model comprises a convolution layer, an excitation layer, a pooling layer, a Dropout layer, an LSTM layer, a multi-head attention mechanism layer and a full-connection layer in sequence from input to output, wherein the convolution kernel parameter settings in the convolution layer are respectively 128 multiplied by 16 multiplied by 1, 64 multiplied by 32 multiplied by 1, 32 multiplied by 1 and 16 multiplied by 1; the excitation layer adopts an activation function which is a Sigmoid function, a Tanh function and a ReLU function respectively; the pooling layer adopts maximum pooling; the Dropout probability in the Dropout layer is selected to be 0.3; the input size in the LSTM layer is set to 16 and the hidden size is set to 16; the total size of the multi-head attention mechanism layer is set to be 32, and the number of heads of the multi-head attention mechanism layer is set to be 4; the input characteristic of the full connection layer is set to be 32, and the output characteristic is set to be 1; the loss function is selected from a mean square loss function:
loss(x i ,y i )＝(x i -y i ) 2
in the formula x i Representing the predicted value, y, output by the model i Representing the actual tag of the data.
Compared with the prior art, the invention has the beneficial effects that:
the method utilizes the Convolutional Neural Network (CNN) to extract and learn the high-latitude characteristics of the microseism data, and can better overcome the influence of background noise on the separation accuracy of the P wave S wave. The combination of the long-short term memory model (LSTM) on the basis of the CNN enables the model to consider the relevance before and after the data, and eliminates the adverse effect of irrelevant data on the separation result. A multi-head attention mechanism is introduced on the basis of CNN-LSTM, micro seismic data acquired in real time on a dry hot rock hydraulic fracturing site usually contain a large amount of background noise, and the separation accuracy of P wave S wave is greatly influenced, so that the requirement of a model on the signal-to-noise ratio of original data is reduced by using the multi-head attention mechanism, and the processing process of data in an early stage is reduced. The multi-head attention mechanism can increase the weight of important information in the micro-seismic data and reduce the weight of unimportant information, so that the model can learn which part of the micro-seismic data contains the important information for P-wave S-wave separation, and the model can improve the accuracy of P-wave S-wave separation under the condition of background noise interference.
Compared with the traditional method, the method does not need to manually change parameters or set thresholds after the network training is finished, and is beneficial to later-stage deployment and field real-time use. The Qinghai hot dry rock hydraulic fracturing monitoring data is used for verification, and the higher separation accuracy can be guaranteed while P waves and S waves are rapidly separated.
Additional aspects and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 shows a flow chart of the method of the present invention;
FIG. 2 illustrates a data set label style in an embodiment of the invention;
FIG. 3 is a diagram of a CNN-LSTM and multi-head attention mechanism model architecture in accordance with the present invention;
FIG. 4 shows the actual separation of the P-wave and S-wave by the method of the present invention;
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
FIG. 1 shows a flow chart of the method of the present invention. The method comprises the steps of constructing a data set, building a CNN-LSTM model, training, storing the model and separating the micro-seismic data through the model. The method specifically comprises the following steps:
a microseism P wave S wave separation method based on CNN-LSTM and a multi-head attention mechanism comprises the following steps:
s1: adopting effective microseism data actually collected on a dry hot rock hydraulic fracturing site to obtain 10000 channels of microseism data, wherein each channel of microseism data comprises P wave and S wave signals;
s2: preprocessing each trace of micro-seismic data, intercepting the data length into 5000 sampling points, and manually marking the data to finally form a training data set;
s3: building a neural network model consisting of a convolutional layer, an excitation layer, a pooling layer, a random hidden neuron layer, a long-short term memory layer, a multi-head attention mechanism layer and a full connection layer, and starting training by adopting the training data set in the step S2;
s4: after the training is finished, the neural network model is stored, the data to be separated is input into the model for separation, the probability value of whether each sampling point is a P wave or an S wave is output, and finally the separation result of the P wave and the S wave is obtained.
The step S2 specifically includes: and intercepting the acquired data into a uniform length, marking the position of P wave or S wave in the data as 1, and marking the rest positions as 0.
S3 specifically comprises the following steps: the neural network model comprises a convolution layer, an excitation layer, a pooling layer, a Dropout layer, an LSTM layer, a multi-head attention mechanism layer and a full-connection layer in sequence from input to output, wherein the convolution kernel parameter settings in the convolution layer are respectively 128 multiplied by 16 multiplied by 1, 64 multiplied by 32 multiplied by 1, 32 multiplied by 1 and 16 multiplied by 1; the excitation layer adopts an activation function which is a Sigmoid function, a Tanh function and a ReLU function respectively; the pooling layer adopts maximum pooling; the Dropout probability in the Dropout layer is selected to be 0.3; the input size in the LSTM layer is set to 16 and the hidden size is set to 16; the total size of the multi-head attention mechanism layer is set to be 32, and the number of heads of the multi-head attention mechanism layer is set to be 4; the input characteristic of the full connection layer is set to be 32, and the output characteristic is set to be 1; the loss function is selected from a mean square loss function:
loss(x i ,y i )＝(x i -y i ) 2
in the formula x i Representing the predicted value, y, output by the model i Representing the actual label of the data.
FIG. 2 illustrates a data set label style of the present invention. The P-wave and S-wave positions are labeled as 1, and the remaining positions are labeled as 0.
FIG. 3 shows the detailed structure of the CNN-LSTM and multi-head attention mechanism model of the present invention. Wherein the dimension of the input data is 5000 × 1, and the dimension of the input label is 5000 × 1. The 1 st, 2 nd, 3 rd and 4 th layers of the model are convolution layers, and the sizes of convolution kernels are 128 multiplied by 16 multiplied by 1, 64 multiplied by 32 multiplied by 1, 32 multiplied by 1 and 16 multiplied by 1 respectively; each layer of convolution is followed by an activation function, the 1 st layer is a Sigmoid function and is used for placing input in an interval of 0-1 so as to prevent the ReLU function from being inactivated, and the 2 nd, 3 rd and 4 th layers are ReLU functions and are used for solving the problem of gradient disappearance; the 5 th layer is a pooling layer, and the largest pooling is selected; the 6 th layer is a Dropout layer, and the Dropout probability is set to be 0.3; the 7 th layer is an LSTM layer, the input size is set to be 16, and the hidden size is set to be 16; the 8 th layer is a multi-head attention mechanism layer, the total size is set to be 32, and the number of heads of the multi-head attention mechanism is set to be 4; the 8 th layer is followed by a Tanh function as an activation function for changing the output into an output with 0 as the center so as to facilitate the final probability output of the model; the 9 th layer is a full connection layer, the input characteristic is set to be 32, and the output characteristic is set to be 1; layer 9 is followed by the Tanh function as the activation function for the same reason as layer 8.
Fig. 4 shows the actual separation result of the P-wave and S-wave according to the present invention. The method is characterized in that data actually acquired by hydraulic fracturing of Qinghai dry hot rock are used for demonstration, when the probability value of P wave or S wave is continuously greater than 0.5, P wave or S wave is successfully separated, the probability value continuously greater than 0.5 is set as 1, and the other probability values are set as 0, so that observation is facilitated. The experimental result shows that the model has good separation effect on the P wave and the S wave and has higher pickup precision on the first arrival position.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (3)
1. A microseism P wave S wave separation method based on CNN-LSTM and a multi-head attention mechanism is characterized by comprising the following steps:
s1: adopting effective microseism data actually collected on a dry hot rock hydraulic fracturing site to obtain 10000 channels of microseism data, wherein each channel of microseism data comprises P wave and S wave signals;
s2: preprocessing each trace of micro seismic data, cutting the length of the data into 5000 sampling points, and manually labeling the data to finally form a training data set;
s3: building a neural network model consisting of a convolutional layer, an excitation layer, a pooling layer, a random hidden neuron layer, a long-short term memory layer, a multi-head attention mechanism layer and a full connection layer, and starting training by adopting the training data set in the step S2;
s4: after the training is finished, the neural network model is stored, the data to be separated are input into the neural network model for separation, the probability value of whether each sampling point is a P wave or an S wave is output, and finally the separation result of the P wave and the S wave is obtained.
2. The technique according to claim 1, wherein step S2 specifically comprises:
the acquired data is intercepted to be uniform in length, the position of the P wave or the S wave in the data is marked as 1, and the positions of the other positions are marked as 0.
3. The technique of claim 1, wherein S3 specifically comprises:
the neural network model comprises a convolution layer, an excitation layer, a pooling layer, a Dropout layer, an LSTM layer, a multi-head attention mechanism layer and a full-connection layer in sequence from input to output, wherein the convolution kernel parameter settings in the convolution layer are respectively 128 multiplied by 16 multiplied by 1, 64 multiplied by 32 multiplied by 1, 32 multiplied by 1 and 16 multiplied by 1; the excitation layer adopts an activation function which is a Sigmoid function, a Tanh function and a ReLU function respectively; the pooling layer adopts maximum pooling; the Dropout probability in the Dropout layer is selected to be 0.3; the input size in the LSTM layer is set to 16 and the hidden size is set to 16; the total size of the multi-head attention mechanism layer is set to be 32, and the number of heads of the multi-head attention mechanism layer is set to be 4;
the input characteristic of the full connection layer is set to be 32, and the output characteristic is set to be 1; the loss function is a mean square loss function:
loss(x i ,y i )＝(x i -y i ) 2
in the formula x i Representing the predicted value, y, output by the model i Representing the actual label of the data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211578756.5A CN115932947A (en) | 2022-12-05 | 2022-12-05 | Microseism P wave S wave separation method based on CNN-LSTM and multi-head attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211578756.5A CN115932947A (en) | 2022-12-05 | 2022-12-05 | Microseism P wave S wave separation method based on CNN-LSTM and multi-head attention mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115932947A true CN115932947A (en) | 2023-04-07 |
Family
ID=86648664
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211578756.5A Pending CN115932947A (en) | 2022-12-05 | 2022-12-05 | Microseism P wave S wave separation method based on CNN-LSTM and multi-head attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115932947A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117390948A (en) * | 2023-08-10 | 2024-01-12 | 苏州黑盾环境股份有限公司 | Multi-head attention long-short-term memory neural network based water chilling unit monitoring method |
-
2022
- 2022-12-05 CN CN202211578756.5A patent/CN115932947A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117390948A (en) * | 2023-08-10 | 2024-01-12 | 苏州黑盾环境股份有限公司 | Multi-head attention long-short-term memory neural network based water chilling unit monitoring method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805269B (en) | Method for picking seismic facies arrival time based on LSTM recurrent neural network | |
| US11789173B1 (en) | Real-time microseismic magnitude calculation method and device based on deep learning | |
| CN112884134B (en) | Time domain-based convolutional neural network model for seismic facies identification and application | |
| CN111580151B (en) | SSNet model-based earthquake event time-of-arrival identification method | |
| CN104777418B (en) | A kind of analog-circuit fault diagnosis method based on depth Boltzmann machine | |
| CN111562612B (en) | Deep learning microseismic event identification method and system based on attention mechanism | |
| CN109670533A (en) | A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks | |
| CN111458746B (en) | Tunnel microseismic waveform arrival time picking method based on U-Net neural network | |
| CN115758212A (en) | Mechanical equipment fault diagnosis method based on parallel network and transfer learning | |
| Yanwei et al. | Deep learning for P-wave arrival picking in earthquake early warning | |
| CN110632662A (en) | Algorithm for automatically identifying microseism signals by using DCNN-inclusion network | |
| He et al. | PickCapsNet: Capsule network for automatic P-wave arrival picking | |
| CN115932947A (en) | Microseism P wave S wave separation method based on CNN-LSTM and multi-head attention mechanism | |
| Liao et al. | First arrival time identification using transfer learning with continuous wavelet transform feature images | |
| CN113568043A (en) | Three-stage seismic phase picking method based on deep convolutional neural network | |
| Zhang et al. | Automatic microseismic detection and location via the deep-convolutional neural network | |
| Liu et al. | Microseismic first-arrival picking using fine-tuning feature pyramid networks | |
| CN113792685A (en) | Microseism event detection method based on multi-scale convolution neural network | |
| Zhao et al. | A hybrid recognition model of microseismic signals for underground mining based on CNN and LSTM networks | |
| Mustika et al. | Comparison of keras optimizers for earthquake signal classification based on deep neural networks | |
| CN114357372A (en) | Aircraft fault diagnosis model generation method based on multi-sensor data driving | |
| Xu et al. | An automatic P-wave onset time picking method for mining-induced microseismic data based on long short-term memory deep neural network | |
| He et al. | Enhancing seismic p-wave arrival picking by target-oriented detection of the local windows using faster-rcnn | |
| CN113671564B (en) | NARX dynamic neural network-based microseism effective event automatic pickup method | |
| CN112801073B (en) | Acoustic emission first-break signal classification and identification method based on deep neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |