CN116413783A - Seismic data interpretation method and device - Google Patents

Abstract

The invention discloses a seismic data interpretation method and a seismic data interpretation device, wherein the method comprises the following steps: establishing a meshed seismic data body, selecting seismic data on the meshed seismic data body as sample data, and calibrating geological layers of the sample data according to logging data and a side-well seismic wave group; establishing a multi-level label according to geological horizon calibration of the sample data, and converting the amplitude characteristics of the sample data into characteristic values of the multi-level label; training an image semantic segmentation network U-Net according to the sample data, the multi-level labels and the characteristic values of the multi-level labels to obtain a seismic data interpretation model; inputting uncalibrated seismic data into a seismic data interpretation model, and predicting the probability of the uncalibrated seismic data corresponding to each level label; according to the probability that the geological horizon of the uncalibrated seismic data corresponds to each level label, the geological horizon and the layer sequence interface of the uncalibrated seismic data are determined, and the efficiency and the accuracy of seismic interpretation can be improved.

Description

Seismic data interpretation method and device

technical field

The invention relates to the technical field of seismic interpretation, in particular to a method and device for establishing an intelligent seismic sequence stratigraphic framework.

Background technique

This section is intended to provide a background or context to embodiments of the invention that are recited in the claims. The descriptions herein are not admitted to be prior art by inclusion in this section.

At present, artificial seismic interpretation technology is very mature, such as horizon calibration, horizon interpretation and fault identification, which have the advantages of accurate interpretation and little influence of interference factors. However, with the deepening of oil and gas exploration, high-precision 3D seismic data are increasing year by year , the traditional seismic structure interpretation method has low work efficiency, huge workload and strong multi-solution. Most of the current processing and interpretation systems are based on mathematical statistics and filtering methods, and the accuracy of automatic horizon tracking cannot be compared with manual interpretation. On the other hand, with the repeated acquisition, processing and interpretation of seismic data, the current situation of seismic data has caused the problems of high cost and low efficiency of upstream oil and gas development.

Seismic interpretation has entered the era of big data, and the massive data generated by workstations has exceeded the effective analysis capabilities of people, and the field of seismic interpretation is facing the problem of being unable to effectively interpret and efficiently extract massive data information. In this context, the traditional manual processing and interpretation methods of stratigraphic correlation, seismic horizon interpretation, and fault identification have exposed the disadvantages of low work efficiency and being easily affected by human factors of different analysts. The current seismic interpretation software system has been unable to meet the requirements. growing production needs.

For the above problems, no effective solution has been proposed yet.

Contents of the invention

An embodiment of the present invention provides a seismic data interpretation method for performing intelligent seismic interpretation, improving the efficiency and accuracy of seismic interpretation, and reducing the workload of seismic interpretation personnel. The method includes:

Establish a gridded seismic data volume, select seismic data as sample data on the gridded seismic data volume, and calibrate the geological horizon of the sample data according to well logging data and side-hole seismic wave groups, the gridded seismic data volume Seismic data volume is composed of a plurality of spaced main survey lines and network lines;

Establishing a multi-level label according to the geological horizon calibration of the sample data, converting the amplitude feature of the sample data into a feature value of the multi-level label, and the feature value of the multi-level label reflects the horizon information and sequence information of the sample data;

According to the sample data, multi-level labels and eigenvalues of multi-level labels, the image semantic segmentation network U-Net is trained to obtain the seismic data interpretation model;

Obtain the uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability of the uncalibrated seismic data corresponding to the labels of each level;

According to the probability that the geological horizon of the uncalibrated seismic data corresponds to each layer label, the geological horizon and the sequence boundary of the uncalibrated seismic data are determined.

An embodiment of the present invention also provides a seismic data interpretation device, which is used for intelligent seismic interpretation, improves the efficiency and accuracy of seismic interpretation, and reduces the workload of seismic interpretation personnel. The device includes:

The selection and calibration module is used to establish a gridded seismic data volume, select seismic data as sample data on the gridded seismic data volume, and calibrate the geological horizon of the sample data according to well logging data and side-hole seismic wave groups , the gridded seismic data volume is composed of a plurality of spaced main survey lines and network lines;

The establishment and conversion module is used to establish a multi-level label according to the geological horizon calibration of the sample data, and convert the amplitude feature of the sample data into a feature value of the multi-level label, and the feature value of the multi-level label reflects the horizon information of the sample data and sequence information;

The training module is used to train the image semantic segmentation network U-Net according to the sample data, the multi-level labels and the feature values of the multi-level labels to obtain the seismic data interpretation model;

The acquisition and prediction module is used to acquire uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability that the uncalibrated seismic data corresponds to labels of each level;

The determining module is configured to determine the geological horizon and sequence interface of the uncalibrated seismic data according to the probability that the geological horizon of the uncalibrated seismic data corresponds to each layer label.

An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and operable on the processor, and the processor implements the above seismic data interpretation method when executing the computer program.

An embodiment of the present invention also provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the above seismic data interpretation method is implemented.

An embodiment of the present invention also provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the above seismic data interpretation method is realized.

In the embodiment of the present invention, a gridded seismic data volume is established, seismic data is selected as sample data on the gridded seismic data volume, and the geological horizon of the sample data is calibrated according to well logging data and side-hole seismic wave groups, The gridded seismic data volume is composed of a plurality of spaced main survey lines and network lines; multi-level labels are established according to the geological horizon calibration of the sample data, and the amplitude characteristics of the sample data are converted into eigenvalues of the multi-level labels , the eigenvalues of the multi-level labels reflect the layer information and sequence information of the sample data; according to the sample data, the multi-level labels and the eigenvalues of the multi-level labels, the image semantic segmentation network U-Net is trained to obtain the seismic data interpretation model ; Obtain the uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability that the uncalibrated seismic data corresponds to each level label; according to the geological layer of the uncalibrated seismic data Bits correspond to the probability of each level label, determine the geological horizon and sequence interface of uncalibrated seismic data, and use it for intelligent seismic interpretation, improve the efficiency and accuracy of seismic interpretation, and greatly reduce the work of seismic interpretation personnel quantity.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work. In the attached picture:

Fig. 1 is the processing flowchart of seismic data interpretation method in the embodiment of the present invention;

Fig. 2 is a schematic diagram of a specific example of calibrating sample data geological horizons in an embodiment of the present invention;

FIG. 3 is a schematic diagram of a specific example of establishing a multi-level label in an embodiment of the present invention;

Fig. 4 is a schematic diagram of the construction process of the seismic data interpretation model in the embodiment of the present invention;

Fig. 5a, Fig. 5b, Fig. 5c are schematic diagrams of specific examples of closed inspection of sequence interface in the embodiment of the present invention;

Fig. 6 is a schematic diagram of a sequence stratigraphic framework manually interpreted in an embodiment of the present invention;

Fig. 7A is a schematic diagram of a sequence stratigraphic framework for intelligent interpretation in an embodiment of the present invention;

Fig. 7B is a schematic diagram of a sequence stratigraphic framework for intelligent interpretation in an embodiment of the present invention;

Fig. 8 is a schematic structural diagram of a seismic data interpretation device in an embodiment of the present invention;

Fig. 9 is a structural schematic diagram of a specific example of the seismic data interpretation device in the embodiment of the present invention;

FIG. 10 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments and descriptions of the present invention are used to explain the present invention, but not to limit the present invention.

First, the technical terms in the embodiments of the present invention are explained:

Well-seismic calibration: Using logging information, through the comparison of forward modeling records and actual seismic data, the reflection event of the measured seismic wave is given geological significance. During the calibration, it is generally realized by comparing the logging geological horizon and the reflection event of the seismic wave, so the well-seismic calibration is also called the horizon calibration.

Seismic trace profile: It is composed of a series of seismic waveform traces, each seismic trace is a one-dimensional signal trace, the analysis and processing of seismic traces is of great theoretical significance for studying stratigraphic geological structures and judging the location of reservoirs and application value.

Sedimentary sequence: A sequence is a set of relatively complete, genetically related stratigraphic units whose top and bottom are bounded by unconformities or their corresponding integration surfaces. Sedimentary sequences can contain several different types of sedimentary systems domain and parasequence group and parasequence. Due to the high vertical resolution of well logging curves, multi-level stratigraphic units can be divided, such as sequences, parasequences, parasequence groups, layer groups, etc. According to logging stratification and well-seismic calibration, the stratigraphic units divided by logging curves can be used for classification and establishment of labels with different degrees of fineness on the seismic section.

Sequence stratigraphic framework: According to the characteristics of sedimentary sequence cycle, sequence stratigraphic units can be divided into grades and different degrees of fineness, so as to establish an isochronic stratigraphic framework.

Fig. 1 is a processing flowchart of the seismic data interpretation method in the embodiment of the present invention. As shown in Figure 1, the seismic data interpretation method in the embodiment of the present invention may include:

Step 101. Establish a gridded seismic data volume, select seismic data as sample data on the gridded seismic data volume, and calibrate the geological horizon of the sample data according to well logging data and side-hole seismic wave groups. The gridded seismic data volume is composed of multiple spaced main survey lines and network lines;

Step 102: Establish a multi-level label according to the geological horizon calibration of the sample data, convert the amplitude characteristics of the sample data into the feature values of the multi-level label, and the feature values of the multi-level label reflect the horizon information and sequence information of the sample data ;

Step 103, according to the sample data, the multi-level labels and the feature values of the multi-level labels, train the image semantic segmentation network U-Net to obtain the seismic data interpretation model;

Step 104, obtaining uncalibrated seismic data in the gridded seismic data volume, inputting the uncalibrated seismic data into the seismic data interpretation model, and predicting the probability that the uncalibrated seismic data corresponds to labels of each level;

Step 105: Determine the geological horizon and sequence boundary of the unmarked seismic data according to the probability that the geological horizon of the unmarked seismic data corresponds to each layer label.

Since the seismic horizon interpretation is based on well-seismic calibration, the process of seismic horizon interpretation can be regarded as dividing the seismic trace into different parts according to the main measuring line and the tie line, and each trace of seismic data is separated at the interlayer and horizon interface. The positions of each can reflect the depositional process, and their characteristics can be distinguished one by one according to the principle of sequence stratigraphy interpretation. Therefore, single-trace seismic horizon division can be regarded as a target detection problem, and the U-shaped convolutional neural network is used to detect the "seismic facies" contained in the seismic trace, and the position of the horizon interface is regarded as the boundary to distinguish the seismic facies.

In order to generate convolutional neural network training data and training labels, the seismic horizon can be manually calibrated and interpreted on a gridded seismic data volume. Taking the logging layered data as the standard, only 1% of the three-dimensional data volume needs to be interpreted manually , for example, the main survey line and tie line direction of the seismic data volume can be separated by 100 traces as seed points.

In specific implementation, firstly, a gridded seismic data volume can be established, and seismic data can be selected as sample data on the gridded seismic data volume, and the geological horizon of the sample data can be calibrated according to well logging data and side-hole seismic wave groups , where the gridded seismic data volume consists of multiple spaced main survey lines and network lines.

In one embodiment, calibrating the geological horizon of the sample data according to the well logging data and the side-hole seismic wave group may include: synthesizing seismic records according to the well-logging data, and comparing the seismic records with the side-hole seismic wave groups; According to the results, the corresponding relationship between the seismic record and the side-hole seismic wave group is determined; according to the corresponding relationship, the geological horizon of the sample data is calibrated.

Fig. 2 is a schematic diagram of a specific example of calibrating geological horizons of sample data in an embodiment of the present invention. As shown in Figure 2, the seismic records of a certain well can be artificially synthesized according to the logging data of a certain well, and the stratum distribution of a certain well in the seismic records can be obtained. The stratum, conglomerate stratum and silt conglomerate stratum, and the corresponding relationship between layer 1, layer 2, layer 3 and layer 4 of the seismic profile of a well, the corresponding geological horizons are calibrated on the seismic profile of a well, according to the measured Well calibration, geological laws, and wave group characteristics of seismic sections are used to interpret horizons for sample data in seismic data volumes. Then, a multi-level label is established based on the geological horizon calibration of the sample data, that is, a multi-level sub-level label can be established based on the logging layered interpretation data of the sample data, and the fineness of the layered interpretation of the label data can be adjusted according to the actual production demand.

Fig. 3 is a schematic diagram of a specific example of establishing multi-level labels in the embodiment of the present invention. As shown in Figure 3, the logging interpretation data of the sample data is divided into levels 1, 2, 3, and 4. In the example, the level 2 interpretation framework is selected to establish label data, and 4 levels are established according to actual production needs. , where each level corresponds to multiple level labels, for example: S under level 1, S1, S2, S3, S4, S5 under level 2, S11, S12, S21, S22 and so on.

After the multi-level labels are established according to the geological horizon calibration of the sample data, the amplitude characteristics of the sample data can be converted into the eigenvalues of the multi-level labels, where the eigenvalues of the multi-level labels reflect the horizon information and sequence information of the sample data.

In one embodiment, the amplitude feature of the sample data can be converted into the feature value of the multi-level label according to the following formula:

Among them, s(i) is the amplitude of the sample data point i, i is the data point of the sample data, Z(i) is the feature value of the multi-level label of the sample data point i, mean_s is the average amplitude value of the sample data, and max_s is The maximum value of the sample data amplitude value, min_s is the minimum value of the sample data amplitude value.

In the specific implementation, the label eigenvalues are divided according to the above formula, and each layer inside the sequence framework is given an eigenvalue, and the training label data is produced according to the sedimentary sequence calibrated by the logging data, which represents the sedimentary sequence and sequence stratum grid. Different eigenvalue numbers are then assigned to adjacent seismic layered data to generate training labels. The size of the training label is the same as that of the training data, and the label data is divided based on the logging layered data and the layer interface of manual interpretation. For example, the training label is divided into five different parts according to the layer interface. From the above The numbers are marked 0, 1, 2, 3 and 4 in turn.

After establishing multi-level labels based on the geological horizon calibration of sample data, and converting the amplitude characteristics of sample data into feature values of multi-level labels, image semantics can be trained based on sample data, multi-level labels, and feature values of multi-level labels Split the network U-Net to get the seismic data interpretation model.

In one embodiment, the hidden layers of the encoder and decoder in the seismic data interpretation model include a convolutional layer, a batch normalization layer, and a maximum pooling layer, and the ReLU function is used as an activation function to activate the features of multi-level labels The features contained in the value.

In one embodiment, the ReLU function is used as the activation function to activate the features contained in the feature values of the multi-level labels according to the following formula:

为单道地震数据最大值。Among them, _xi is the single-channel seismic data,

is the maximum value of single-trace seismic data.

Fig. 4 is a schematic diagram of the construction process of the seismic data interpretation model in the embodiment of the present invention, as shown in Fig. 4, wherein the U-net network structure is a symmetrical network framework, including two main parts of the encoder network and the decoder network. The decoder can handle pixel-level image or data segmentation tasks. The encoder network consists of a series of hidden layers, and each layer includes operations such as convolutional layers, batch normalization, ReLU activation functions, max pooling layers, and upsampling.

The main purpose of the convolution layer is to extract the features of the input seismic data. Researchers in the field of image processing usually select a constant size convolution filter in each layer, but according to the loss function, in the case of fewer eigenvalues, Repeated use of convolution and pooling to extract features and feature optimization can better optimize the horizon features in seismic data. The convolution kernel dimensionality reduction methods can be selected in the encoder, respectively 256×16, 128×32 , 64×64, 32×128 and 16×256. Downsampling uses maximum pooling, and the size of the convolution kernel is 2×1.

The role of the decoder network is to reconstruct the image according to the feature map, including the number of hidden layers corresponding to the encoder network, each layer consists of an upsampling operator, a convolution filter, a batch normalization regularization function and a ReLU activation operator, where Upsampling restores the label data volume to the same size as the original seismic data volume. The size of a sample data and corresponding labels in the sample is 3×256×1, the total feature map of the first layer is 16, the size of each feature map is 3×256, and the total feature map of the last layer is 256, The size of each feature map is 3×8. The goal of ReLU is to activate the main features contained in the eigenvalues using the above formula.

The upsampling operator is used to reconstruct the feature map, and the size of all convolution kernels is 2×1. In the decoder, the uptrend sizes used for the convolution kernels are chosen to be 16×256, 32×128, 64×64, 128×32 and 256×16. The final step of the network is an element-wise softmax classification layer of size 1×1, whose role is to define the weight of each pixel using the feature values of the multi-level labels of the sample data points.

In one embodiment, the image semantic segmentation network U-Net is trained according to the sample data, the multi-level labels and the feature values of the multi-level labels to obtain the seismic data interpretation model, which may include: according to the sample data, the multi-level labels and the multi-level labels eigenvalues, construct training set and test set; use the training set to train the image semantic segmentation network U-Net to obtain the seismic data interpretation model; use the test set to test the seismic data interpretation model.

In one embodiment, using the training set to train the image semantic segmentation network U-Net to obtain the seismic data interpretation model may include: using the cross-entropy loss function to calculate the eigenvalues of the multi-level labels corresponding to the training set, and earthquake The data interpretation model outputs the loss value between the prediction values corresponding to the training set; when the loss value is not less than the preset threshold, the parameters of the seismic data interpretation model are updated according to the loss value, and the updated seismic data interpretation model is used to continue outputting the prediction corresponding to the training set value until the calculated loss value is less than the preset threshold, and the optimized seismic data interpretation model is obtained; the optimized seismic data interpretation model is tested using the test set.

In one embodiment, the cross-entropy loss function can be used to calculate the loss value between the eigenvalues of the multi-level labels corresponding to the training set and the predicted values corresponding to the seismic data interpretation model output training set according to the following formula:

Among them, i is the data point of the sample data, x _i is the seismic data of data point i, θ is the network parameter, T is the transpose symbol, h _θ ( _xi ) is the predicted probability output by the seismic data interpretation model, J(θ ) is the cross-entropy loss value, Z(i) is the feature value of the multi-level label of the sample data point i, and m is a natural number.

In the specific implementation, in order to avoid the overfitting problem in the training process, the sample data for training and testing can be randomly selected in each update optimization period. A certain sample data may belong to the training set of the current update optimization period, or it may belong to the following The test set in the optimization period is updated for the second time, wherein the percentage of the training set in the sample data may be 70%, and the percentage of the test set in the sample data may be 30%.

After obtaining the seismic data interpretation model that has passed the test, the seismic data interpretation model can be applied to the uncalibrated seismic data to predict the probability that the uncalibrated seismic data corresponds to each layer label, and according to the geological horizon of the uncalibrated seismic data Corresponding to the probability of each layer label, the geological horizon and sequence boundary of the uncalibrated seismic data are determined.

In one embodiment, determining the geological horizon and sequence interface of the unmarked seismic data according to the probability that the geological horizon of the unmarked seismic data corresponds to each layer label may include: according to the geological layer of the unmarked seismic data According to the probability of each level label, the geological horizon of the uncalibrated seismic data is determined; according to the geological horizon of the uncalibrated seismic data, the characteristic boundary of the uncalibrated seismic data is extracted; according to the characteristic boundary of the uncalibrated seismic data, Identify sequence boundaries for unscaled seismic data.

In an embodiment, it may further include: performing a closure check on the sequence boundary of the uncalibrated seismic data.

In one embodiment, the closure inspection of the sequence boundary of the uncalibrated seismic data may include: determining two sequence boundaries of the uncalibrated seismic data in the direction of the main survey line and the direction of the network line; interpreting the model according to the seismic data According to the predicted results of the prediction results, it is judged whether the probabilities of each layer label corresponding to the seismic data intersecting the two sequence interfaces are consistent; if they are consistent, the closure check is passed, and the sequence boundary of the uncalibrated seismic data is determined to be correct; if they are not consistent, the closure inspection Failed, the sequence boundary of the uncalibrated seismic data was incorrectly determined.

Fig. 5a, Fig. 5b, and Fig. 5c are schematic diagrams of specific examples of the closure inspection of the sequence interface in the embodiment of the present invention. Assuming that the seismic data intersecting two sequence boundaries includes seismic trace 5 and seismic trace 7, as shown in Fig. 5a, Fig. 5b, and Fig. 5c, the specific steps for checking the closure of these two sequence boundaries may include:

As shown in Fig. 5a, based on the target seismic section data and sequence stratigraphic interpretation framework, the sequence interface is traced along the label eigenvalue interfaces inline A'-A and crossline BB', and the algorithm considers that seismic trace 2 -1, the intelligent interpretation sequence interface between 1-3 is closed.

As shown in Figure 5b, based on the inline A-A', crossline BB' label eigenvalue interface, the sequence interface on the crossline D-D' is traced, and the algorithm considers the intelligent sequence interface between seismic traces 2-4 Interpret results closed.

As shown on the left side of Fig. 5c, check the consistency of the sequence boundary results of crossline B-B' and inline C-C' on seismic trace 5. If there is a difference in the interpretation results of the two sequence boundaries in the seismic trace 5, it is considered that the horizon interpretation results on the four sections are under-closed interpretations, and the seismic data range is reduced according to the given grid density, and the closure of the interpretations is further checked sex.

As shown on the right side of Figure 5c, narrow down the range of seismic data, after the intelligent sequence interface interpretation between 2-4 in the previous step of the seismic trace, check the sequence of crossline BB', inline EE' on seismic trace 7 The interface interpretation results are consistent. If there is no difference in the interpretation results of the two sections in trace 7, the program considers the horizon interpretation results on the four sections as closed interpretations.

The above steps are repeated cyclically until the four directions of inline and crossline are completely closed, and it can be confirmed that the prediction result of the seismic data interpretation model is correct. During specific implementation, after the sequence interface of the uncalibrated seismic data passes the closed inspection, the seismic data interpretation model can be used to perform intelligent interpretation of the seismic data, which can greatly improve the efficiency of seismic interpretation.

Fig. 6 is a schematic diagram of the sequence stratigraphic framework manually interpreted in the embodiment of the present invention. It is assumed that a sequence framework in a certain work area is manually calibrated with 100 seismic lines.

Fig. 7A and Fig. 7B are schematic diagrams of the sequence stratigraphic framework for intelligent interpretation in the embodiment of the present invention, and the intelligent interpretation results of the sequence stratigraphic framework output by using the target seismic data and the convolutional neural network structure model.

As shown in Figures 6 and 7, the interpretation accuracy of the intelligent seismic sequence stratigraphic framework has reached the level of manual interpretation, but the interpretation efficiency has increased several times.

An embodiment of the present invention also provides a seismic data interpretation device, as described in the following embodiments. Since the problem-solving principle of the device is similar to that of the seismic data interpretation method, the implementation of the device can refer to the implementation of the seismic data interpretation method, and the repetition will not be repeated.

Fig. 8 is a schematic structural diagram of a seismic data interpretation device in an embodiment of the present invention. As shown in Figure 8, the seismic data interpretation device in the embodiment of the present invention may specifically include:

The selecting and calibrating module 801 is used to establish a gridded seismic data volume, select seismic data as sample data on the gridded seismic data volume, and carry out geological stratification of the sample data according to well logging data and side-hole seismic wave groups Calibration, the gridded seismic data volume is composed of a plurality of spaced main survey lines and network lines;

The establishment and conversion module 802 is used to establish a multi-level label according to the geological horizon calibration of the sample data, and convert the amplitude feature of the sample data into the feature value of the multi-level label, and the feature value of the multi-level label reflects the horizon of the sample data information and sequence information;

The training module 803 is used to train the image semantic segmentation network U-Net according to the sample data, the multi-level labels and the feature values of the multi-level labels to obtain the seismic data interpretation model;

The acquisition and prediction module 804 is used to acquire uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability that the uncalibrated seismic data corresponds to labels of each level;

The determining module 805 is configured to determine the geological horizon and sequence boundary of the unmarked seismic data according to the probability that the geological horizon of the unmarked seismic data corresponds to each layer label.

In one embodiment, the selection and calibration module 801 is specifically used for:

Synthesize seismic records based on well logging data, and compare the seismic records with the seismic wave groups beside the well;

According to the comparison results, determine the corresponding relationship between the seismic record and the seismic wave group beside the well;

According to the corresponding relationship, the geological horizon of the sample data is calibrated.

In one embodiment, the establishment and conversion module 802 is specifically configured to convert the amplitude feature of the sample data into the feature value of the multi-level label according to the following formula:

Among them, s(i) is the amplitude of the sample data point i, i is the data point of the sample data, Z(i) is the feature value of the multi-level label of the sample data point i, mean_s is the average amplitude value of the sample data, and max_s is The maximum value of the sample data amplitude value, min_s is the minimum value of the sample data amplitude value.

In one embodiment, the hidden layers of the encoder and decoder in the seismic data interpretation model include a convolutional layer, a batch normalization layer, and a maximum pooling layer, and the ReLU function is used as an activation function to activate the features of multi-level labels The features contained in the value.

In one embodiment, the ReLU function is used as the activation function to activate the features contained in the feature values of the multi-level labels according to the following formula:

是单道地震数据最大值。Among them, _xi is the single-channel seismic data;

is the maximum value of single-trace seismic data.

In one embodiment, the training module 803 is specifically used for:

Construct a training set and a test set based on sample data, multi-level labels, and feature values of multi-level labels;

Using the training set to train the image semantic segmentation network U-Net to obtain the seismic data interpretation model;

The seismic data interpretation model is tested using a test set.

In one embodiment, the training module 803 is specifically used for:

The cross-entropy loss function is used to calculate the loss value between the eigenvalue of the multi-level label corresponding to the training set and the predicted value corresponding to the output training set of the seismic data interpretation model;

When the loss value is not less than the preset threshold, the parameters of the seismic data interpretation model are updated according to the loss value, and the updated seismic data interpretation model is used to continue outputting the prediction value corresponding to the training set until the calculated loss value is less than the preset threshold value, and the optimized earthquake is obtained data interpretation model;

The optimized seismic data interpretation model is tested on the test set.

In one embodiment, the training module 803 is specifically used to: use the cross-entropy loss function to calculate the eigenvalue of the multi-level label corresponding to the training set according to the following formula, and the loss value between the predicted value corresponding to the seismic data interpretation model output training set :

Among them, i is the data point of the sample data, x _i is the seismic data of data point i, θ is the network parameter, T is the transpose symbol, h _θ ( _xi ) is the predicted probability output by the seismic data interpretation model, J(θ ) is the cross-entropy loss value, z _i is the feature value of the multi-level label of the sample data point i, and m is a natural number.

In one embodiment, the determining module 805 is specifically configured to:

According to the probability that the geological horizon of the uncalibrated seismic data corresponds to each level label, the geological horizon of the uncalibrated seismic data is determined;

According to the geological horizon of the uncalibrated seismic data, the characteristic boundary of the uncalibrated seismic data is extracted;

The sequence boundary of the uncalibrated seismic data is determined according to the characteristic boundaries of the uncalibrated seismic data.

Fig. 9 is a structural schematic diagram of a specific example of the seismic data interpretation device in the embodiment of the present invention. As shown in Figure 9, in an embodiment, the seismic data interpretation device shown in Figure 8 may also include:

Closure checking module 901, used for:

Closure checks are performed on sequence boundaries of uncalibrated seismic data.

In one embodiment, the closure checking module 901 is specifically used to:

Determine the two sequence boundaries of the uncalibrated seismic data in the direction of the main survey line and the direction of the network line;

According to the prediction results of the seismic data interpretation model, it is judged whether the probabilities of each level label corresponding to the seismic data intersecting the two sequence interfaces are consistent;

If they are consistent, the closure check is passed, and the sequence boundary of the uncalibrated seismic data is determined correctly; if not, the closure check is not passed, and the sequence boundary of the uncalibrated seismic data is determined incorrectly.

Based on the foregoing inventive concepts, as shown in FIG. 10 , the present invention also proposes a computer device 1000, including a memory 1010, a processor 1020, and a computer program 1030 stored in the memory 1010 and operable on the processor 1020. When the processor 1020 executes the computer program 1030, the aforementioned seismic data interpretation method is implemented.

An embodiment of the present invention also provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the above seismic data interpretation method is realized.

An embodiment of the present invention also provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the above seismic data interpretation method is realized.

In summary, in the embodiment of the present invention, a gridded seismic data volume is established, and seismic data is selected as sample data on the gridded seismic data volume, and the geological data of the sample data are analyzed according to well logging data and side-hole seismic wave groups. Horizontal calibration, the gridded seismic data volume is composed of a plurality of spaced main survey lines and network lines; multi-level labels are established according to the geological horizon calibration of the sample data, and the amplitude characteristics of the sample data are converted into multiple The feature value of the hierarchical label, the feature value of the multi-level label reflects the layer information and the sequence information of the sample data; according to the sample data, the multi-level label and the feature value of the multi-level label, the image semantic segmentation network U-Net is trained, Obtain the seismic data interpretation model; obtain the uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability of the uncalibrated seismic data corresponding to each level label; according to the uncalibrated seismic data The geological horizon of the seismic data corresponds to the probability of each layer label, and the geological horizon and sequence interface of the uncalibrated seismic data can be determined, which can intelligently perform seismic interpretation, improve the efficiency and accuracy of seismic interpretation, and greatly reduce the number of seismic interpretation personnel workload.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. Protection scope, within the spirit and principles of the present invention, any modification, equivalent replacement, improvement, etc., shall be included in the protection scope of the present invention.

Claims (25)

1. A seismic data interpretation method, characterized in that, comprising: Establish a gridded seismic data volume, select seismic data as sample data on the gridded seismic data volume, and calibrate the geological horizon of the sample data according to well logging data and side-hole seismic wave groups, the gridded seismic data volume Seismic data volume is composed of a plurality of spaced main survey lines and network lines; Establishing a multi-level label according to the geological horizon calibration of the sample data, converting the amplitude feature of the sample data into a feature value of the multi-level label, and the feature value of the multi-level label reflects the horizon information and sequence information of the sample data; According to the sample data, multi-level labels and eigenvalues of multi-level labels, the image semantic segmentation network U-Net is trained to obtain the seismic data interpretation model; Obtain the uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability of the uncalibrated seismic data corresponding to the labels of each level; According to the probability that the geological horizon of the uncalibrated seismic data corresponds to each layer label, the geological horizon and the sequence boundary of the uncalibrated seismic data are determined. 2. The method according to claim 1, characterized in that, according to the logging data and the side-hole seismic wave group, the geological horizon of the sample data is demarcated, including: Synthesize seismic records based on well logging data, and compare the seismic records with the seismic wave groups beside the well; According to the comparison results, determine the corresponding relationship between the seismic record and the seismic wave group beside the well; According to the corresponding relationship, the geological horizon of the sample data is calibrated. 3. The method according to claim 1, wherein the amplitude feature of the sample data is converted into the feature value of the multi-level label according to the following formula:

Among them, s(i) is the amplitude of the sample data point i, i is the data point of the sample data, Z(i) is the feature value of the multi-level label of the sample data point i, mean_s is the average amplitude value of the sample data, and max_s is The maximum value of the sample data amplitude value, min_s is the minimum value of the sample data amplitude value. 4. The method according to claim 1, wherein the hidden layers of the encoder and the decoder in the seismic data interpretation model include a convolutional layer, a batch normalization layer and a maximum pooling layer, using the ReLU function Activate the features contained in the feature values of the multi-level labels as activation functions. 5. The method according to claim 4, wherein the ReLU function is used as the activation function to activate the features contained in the feature values of the multi-level labels according to the following formula:

Among them, _xi is the single-channel seismic data,

is the maximum value of single-trace seismic data. 6. The method according to claim 1, wherein, according to sample data, multi-level labels and eigenvalues of multi-level labels, training image semantic segmentation network U-Net obtains seismic data interpretation model, comprising: Construct a training set and a test set based on sample data, multi-level labels, and feature values of multi-level labels; Using the training set to train the image semantic segmentation network U-Net to obtain the seismic data interpretation model; The seismic data interpretation model is tested using a test set. 7. method as claimed in claim 6, is characterized in that, utilizes training set to carry out training to image semantic segmentation network U-Net, obtains described seismic data interpretation model, comprising: The cross-entropy loss function is used to calculate the loss value between the eigenvalue of the multi-level label corresponding to the training set and the predicted value corresponding to the output training set of the seismic data interpretation model; When the loss value is not less than the preset threshold, the parameters of the seismic data interpretation model are updated according to the loss value, and the updated seismic data interpretation model is used to continue outputting the prediction value corresponding to the training set until the calculated loss value is less than the preset threshold value, and the optimized earthquake is obtained Data Interpretation Model. 8. The method according to claim 7, characterized in that, the eigenvalue of the multi-level label corresponding to the training set is calculated using the cross-entropy loss function according to the following formula, and the difference between the predicted value corresponding to the seismic data interpretation model output training set Loss value:

Among them, i is the data point of the sample data, x _i is the seismic data of data point i, θ is the network parameter, T is the transpose symbol, h _θ ( _xi ) is the predicted probability output by the seismic data interpretation model, J(θ ) is the cross-entropy loss value, z _i is the feature value of the multi-level label of the sample data point i, and m is a natural number. 9. The method according to claim 1, wherein, according to the probability that the geological horizon of the unmarked seismic data corresponds to each level label, determining the geological horizon and the sequence interface of the unmarked seismic data includes: According to the probability that the geological horizon of the uncalibrated seismic data corresponds to each level label, the geological horizon of the uncalibrated seismic data is determined; According to the geological horizon of the uncalibrated seismic data, the characteristic boundary of the uncalibrated seismic data is extracted; The sequence boundary of the uncalibrated seismic data is determined according to the characteristic boundaries of the uncalibrated seismic data. 10. The method of claim 9, further comprising: Closure checks are performed on sequence boundaries of uncalibrated seismic data. 11. The method as claimed in claim 10, characterized in that, the sequence boundary of uncalibrated seismic data is checked for closure, comprising: Determine the two sequence boundaries of the uncalibrated seismic data in the direction of the main survey line and the direction of the network line; According to the prediction results of the seismic data interpretation model, it is judged whether the probabilities of each level label corresponding to the seismic data intersecting the two sequence interfaces are consistent; If they are consistent, the closure check is passed, and the sequence boundary of the uncalibrated seismic data is determined correctly; if not, the closure check is not passed, and the sequence boundary of the uncalibrated seismic data is determined incorrectly. 12. A seismic data interpretation device, characterized in that it comprises: The selection and calibration module is used to establish a gridded seismic data volume, select seismic data as sample data on the gridded seismic data volume, and calibrate the geological horizon of the sample data according to well logging data and side-hole seismic wave groups , the gridded seismic data volume is composed of a plurality of spaced main survey lines and network lines; The establishment and conversion module is used to establish a multi-level label according to the geological horizon calibration of the sample data, and convert the amplitude feature of the sample data into a feature value of the multi-level label, and the feature value of the multi-level label reflects the horizon information of the sample data and sequence information; The training module is used to train the image semantic segmentation network U-Net according to the sample data, the multi-level labels and the feature values of the multi-level labels to obtain the seismic data interpretation model; The acquisition and prediction module is used to acquire uncalibrated seismic data in the gridded seismic data volume, input the uncalibrated seismic data into the seismic data interpretation model, and predict the probability that the uncalibrated seismic data corresponds to labels of each level; The determining module is configured to determine the geological horizon and sequence interface of the uncalibrated seismic data according to the probability that the geological horizon of the uncalibrated seismic data corresponds to each layer label. 13. The device according to claim 12, wherein the selection and calibration module is specifically used for: Synthesize seismic records based on well logging data, and compare the seismic records with the seismic wave groups beside the well; According to the comparison results, determine the corresponding relationship between the seismic record and the seismic wave group beside the well; According to the corresponding relationship, the geological horizon of the sample data is calibrated. 14. The device according to claim 12, wherein the establishment and conversion module is specifically used to convert the amplitude feature of the sample data into the feature value of the multi-level label according to the following formula:

Among them, s(i) is the amplitude of the sample data point i, i is the data point of the sample data, Z(i) is the feature value of the multi-level label of the sample data point i, mean_s is the average amplitude value of the sample data, and max_s is The maximum value of the sample data amplitude value, min_s is the minimum value of the sample data amplitude value. 15. The device according to claim 12, wherein the hidden layers of the encoder and the decoder in the seismic data interpretation model include a convolutional layer, a batch normalization layer and a maximum pooling layer, using the ReLU function Activate the features contained in the feature values of the multi-level labels as activation functions. 16. The device according to claim 15, wherein the ReLU function is used as the activation function to activate the features contained in the feature values of the multi-level labels according to the following formula:

Among them, _xi is the single-channel seismic data;

is the maximum value of single-trace seismic data. 17. The device according to claim 12, wherein the training module is specifically used for: Construct a training set and a test set based on sample data, multi-level labels, and feature values of multi-level labels; Using the training set to train the image semantic segmentation network U-Net to obtain the seismic data interpretation model; The seismic data interpretation model is tested using a test set. 18. The device according to claim 17, wherein the training module is specifically used for: The cross-entropy loss function is used to calculate the loss value between the eigenvalue of the multi-level label corresponding to the training set and the predicted value corresponding to the output training set of the seismic data interpretation model; When the loss value is not less than the preset threshold, the parameters of the seismic data interpretation model are updated according to the loss value, and the updated seismic data interpretation model is used to continue outputting the prediction value corresponding to the training set until the calculated loss value is less than the preset threshold value, and the optimized earthquake is obtained data interpretation model; The optimized seismic data interpretation model is tested on the test set. 19. The device according to claim 18, wherein the training module is specifically used to: use the cross-entropy loss function to calculate the eigenvalues of the multi-level labels corresponding to the training set according to the following formula, corresponding to the seismic data interpretation model output training set The loss value between the predicted values of :

Among them, i is the data point of the sample data, x _i is the seismic data of data point i, θ is the network parameter, T is the transpose symbol, h _θ ( _xi ) is the predicted probability output by the seismic data interpretation model, J(θ ) is the cross-entropy loss value, z _i is the feature value of the multi-level label of the sample data point i, and m is a natural number. 20. The device according to claim 12, wherein the determination module is specifically used for: According to the probability that the geological horizon of the uncalibrated seismic data corresponds to each level label, the geological horizon of the uncalibrated seismic data is determined; According to the geological horizon of the uncalibrated seismic data, the characteristic boundary of the uncalibrated seismic data is extracted; The sequence boundary of the uncalibrated seismic data is determined according to the characteristic boundaries of the uncalibrated seismic data. 21. The apparatus of claim 20, further comprising a closure check module for: Closure checks are performed on sequence boundaries of uncalibrated seismic data. 22. The device according to claim 21, wherein the closure checking module is specifically used for: Determine the two sequence boundaries of the uncalibrated seismic data in the direction of the main survey line and the direction of the network line; According to the prediction results of the seismic data interpretation model, it is judged whether the probabilities of each level label corresponding to the seismic data intersecting the two sequence interfaces are consistent; If they are consistent, the closure check is passed, and the sequence boundary of the uncalibrated seismic data is determined correctly; if not, the closure check is not passed, and the sequence boundary of the uncalibrated seismic data is determined incorrectly. 23. A computer device, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, characterized in that, when the processor executes the computer program, any one of claims 1 to 11 is realized. the method. 24. A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented. 25. A computer program product, characterized in that the computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images