Three-dimensional velocity field multivariate nonlinear regression method and device based on deep learning

Abstract

The invention relates to a three-dimensional velocity field multivariate nonlinear regression method and device based on deep learning. The method comprises the following steps: acquiring velocity field training set data formed by four characteristics of an X coordinate, a Y coordinate, time and velocity; preprocessing (i.e., standardizing) the training set data to enable the training set data to meet normal distribution; a DNN model is built and trained, and when the error between the prediction speed and the label speed is smaller than a certain threshold value, it can be considered that training of the DNN model is completed; and predicting each CMP point by using the trained DNN model to obtain a three-dimensional seismic exploration velocity field after interpolation and extrapolation of the whole region. The speed control point density and the manual picking time can be effectively reduced, and the picking precision comparable with that of manual picking is achieved, so that the data processing cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of seismic data processing in the field of earth science, and more specifically, to a three-dimensional velocity field multivariate nonlinear regression method and device based on deep learning. Background Art

[0002] Velocity analysis is a critical step in the seismic data processing workflow. A reasonable and accurate velocity model plays a vital role in various subsequent processing tasks.

[0003] There are two main purposes: first, to obtain accurate static correction (NMO) velocity for normal motion correction; second, to construct a starttime model based on velocity analysis results, laying the foundation for offset correction, time-depth conversion, and various wavefield inversions. If the velocity analysis is inaccurate, it will directly lead to errors in various subsequent processing results.

[0004] At present, speed analysis mainly uses the following two methods: manual speed analysis method and automatic speed analysis method, which are introduced below respectively.

[0005] 1. Manual speed analysis method

[0006] This is a traditional approach, which relies on seismic processing personnel to visually inspect the velocity spectrum, manually pick the best peak point, and calculate the corresponding velocity. The specific steps include:

[0007] (a) NMO stacking and preliminary velocity estimation

[0008] By changing the NMO speed, use multiple speeds to perform high-intensity similarity stacking, observe the stacking quality, and select the speed with the best effect

[0009] (b) On the selected optimal stacking velocity spectrum, the processor manually selects several peak points that can fully represent the formation information;

[0010] (c) Based on the manually selected peak points, the velocity value of the corresponding formation is calculated, and the relationship between the formation velocity and time is fitted to construct a velocity-time model.

[0011] This method requires the processing personnel to visually check a large number of velocity spectra, which is very time-consuming and highly dependent on the experience and subjectivity of the seismic processing personnel. It is inefficient and almost infeasible for processing large amounts of 3D seismic volumes. However, since it is directly based on the best velocity spectrum point, the accuracy is often more reliable.

[0012] 2. Automatic speed analysis method.

[0013] This type of method uses a computer to automatically scan a large amount of velocity spectrum data and select the best peak point according to certain preset rules. For example, the peak significance is judged by threshold, the peak connected area is judged, etc. This can achieve efficient large-scale data processing, but the accuracy of the automatic method cannot reach the level of manual analysis.

[0014] Traditional automated speed analysis methods mainly include the following categories:

[0015] (1) Based on statistical characteristics analysis method

[0016] This type of method analyzes the overall statistical characteristics of the velocity spectrum, such as calculating the reliability and relative strength of each sampling point, and uses this to determine the best peak. However, statistical characteristics cannot completely determine the best peak.

[0017] (2) Based on the method of tracking connected domains

[0018] After determining the initial peak seed point, the connected domain formed by the peak is tracked and the strongest point in the connected domain is selected. However, it is difficult to select the initial seed point.

[0019] (3) Based on layer tracking picker

[0020] Tracks the peak points of a single layer at different offsets, but is less effective for complex models.

[0021] (4) Tracking based on layer similarity

[0022] Calculate the similarity of adjacent layers and track the peaks of similar layers. This type of method is highly sensitive to noise.

[0023] In addition to the above-mentioned special automatic speed analysis algorithms, some methods attempt to combine manual methods with automatic methods to give full play to their respective advantages. The speed analysis method includes the following steps:

[0024] 1. Perform automatic speed picking first to obtain preliminary results

[0025] 2. The processing personnel then conduct verification and modification to obtain refined results.

[0026] 3. Combine the precision advantages of manual methods with the efficiency advantages of automatic methods.

[0027] Currently, most traditional automatic speed analysis technologies rely on preset parameters, are sensitive to input data, and cannot achieve the accuracy of manual analysis. This remains a major obstacle to the application of automated processes.

[0028] In summary, velocity analysis, as a key step in seismic data processing, directly affects the results of subsequent processing. Manual velocity analysis has high accuracy but low efficiency and is difficult to handle large-scale data. Traditional automatic methods have high efficiency but limited accuracy. How to automate velocity analysis while ensuring accuracy is still a technical problem that needs to be solved urgently.

[0029] These all limit the application of speed analysis. Summary of the invention

[0030] In view of this, the present invention discloses a solution for automatically picking speed-time pairs, which can match the accuracy of manual picking.

[0031] According to one aspect of the present invention, a three-dimensional velocity field multivariate nonlinear regression method based on deep learning is proposed, the method comprising:

[0032] Step 1, obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity;

[0033] Step 2, preprocessing the velocity field training set data to make it satisfy the normal distribution;

[0034] Step 3, building a deep neural network model, using the preprocessed velocity field training set data to train the deep neural network model, and using the manually picked speed as the label speed;

[0035] Step 4: Use the trained deep neural network model to predict each CMP point and obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

[0036] In some implementations, in step 2, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn library to satisfy a normal distribution.

[0037] In some embodiments, in step 3, when the error between the predicted speed and the labeled speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

[0038] In some embodiments, in step 3, a function in the Scikit-learn library is called to randomly divide the training set data into multiple subsets, and some subsets are used as test sets for cross-validation during training.

[0039] According to one aspect of the present invention, a three-dimensional velocity field multivariate nonlinear regression device based on deep learning is also proposed, and the method includes:

[0040] A training data acquisition unit is used to obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity;

[0041] A preprocessing unit, used for preprocessing the velocity field training set data to make it satisfy a normal distribution;

[0042] A DNN model training unit, used to build a deep neural network model, use the preprocessed velocity field training set data to train the deep neural network model, and use the artificial picking speed as the label speed;

[0043] The prediction unit is used to predict each CMP point using the trained deep neural network model to obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

[0044] In some implementations, in the preprocessing unit, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn library to satisfy a normal distribution.

[0045] In some embodiments, in the DNN model training unit, when the error between the predicted speed and the label speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

[0046] In some embodiments, in the DNN model training unit, a function in the Scikit-learn learning library is called to randomly divide the training set data into multiple subsets, and some subsets are used as test sets for cross-validation during training.

[0047] According to another aspect of the present invention, an electronic device is also provided, the electronic device comprising:

[0048] A memory storing executable instructions;

[0049] A processor runs the executable instructions in the memory to implement the three-dimensional velocity field multivariate nonlinear regression method based on deep learning as described above.

[0050] According to another aspect of the present invention, a computer-readable storage medium is also proposed, which stores a computer program. When the computer program is executed by a processor, the deep learning three-dimensional velocity field multivariate nonlinear regression method described above is implemented.

[0051] The three-dimensional velocity field multivariate nonlinear regression scheme based on deep learning disclosed in the present invention helps the processor to reduce the density of velocity control points and the time of manual picking, and obtain relatively accurate time-velocity pairs through multivariate nonlinear regression prediction. This technical solution has at least the following advantages:

[0052] 1. Realized the automation of speed picking

[0053] Traditional speed picking relies on manual operation by professionals, which is time-consuming and laborious. The present invention realizes the automation of speed picking by establishing a deep learning model, which greatly improves work efficiency.

[0054] 2. Pickup effect is comparable to artificial

[0055] The test results show that the automatic picking effect of the present invention can match the manual picking effect of professionals, ensuring the quality of the picking result;

[0056] 3. Reduce the density of control points

[0057] Relying on the prediction of the deep learning model, better interpolation and extrapolation effects can be achieved with fewer control points, reducing the demand for control point density;

[0058] 4. Accurate prediction speed

[0059] The present invention predicts the velocity through a deep learning model, which can obtain more accurate and reliable velocity field information compared with simple interpolation and extrapolation;

[0060] 5. Improve workflow efficiency

[0061] By automating the picking speed, time-consuming manual steps in the traditional workflow are skipped, greatly improving the automation and efficiency of the data processing process as a whole;

[0062] 6. Reduce data processing costs

[0063] Relying on automation technology to reduce manual operations can significantly reduce the labor cost and resource consumption of seismic data processing;

[0064] 7. Improve processing scale capabilities

[0065] The automated model can easily process massive 3D seismic data without being limited by the size of the data;

[0066] 8. Standardized processing flow

[0067] The present invention realizes the standardization and normalization of the key step of speed analysis, and reduces the subjective arbitrariness of the processing results;

[0068] In summary, the present invention has important progressive significance for seismic data processing in terms of improving velocity picking efficiency, ensuring picking quality, and reducing processing costs.

[0069] The methods and apparatus of the present invention have other features and advantages that will be apparent from or will be described in detail in the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0071] Figure 1 A flowchart of a deep learning three-dimensional velocity field multivariate nonlinear regression method according to an embodiment of the present invention is shown.

[0072] Figure 2 (a), (b) and (c) are schematic diagrams of the prediction effect of label control points in the entire study area obtained according to an exemplary embodiment of the present invention;

[0073] Figure 3 (a), (b) and (c) are prediction effect diagrams of a certain label control point obtained according to an exemplary embodiment of the present invention;

[0074] Figure 4 (a), (b) and (c) are comparison diagrams of the three-dimensional seismic exploration velocity field effects after interpolation and extrapolation in the entire area predicted according to the prior art and the embodiment of the present invention. DETAILED DESCRIPTION

[0075] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0076] Example 1

[0077] Figure 1 A flowchart of a deep learning three-dimensional velocity field multivariate nonlinear regression method according to an embodiment of the present invention is shown. As shown in the figure, the method includes steps 1 to 4.

[0078] Step 1: Obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity.

[0079] The goal of step 1 is to construct a training set of velocity field data.

[0080] The training set data according to the present invention is represented by four-dimensional features, namely X-coordinate, Y-coordinate, time and speed. Among them, the X-coordinate and the Y-coordinate represent the spatial position information of the sample. It is generally obtained by sampling the coordinates of the detection line or the common center point (CMP). Time represents the sampling time point at the position. As needed, the time can be discretized or sampled into multiple discrete values. Speed ​​is the label of the sample, which represents the actual speed value at the XY coordinate at a given time point. In the present invention, the label speed can be obtained by manually picking up the speed spectrum to improve the prediction accuracy of the deep neural network (DNN, Deep Neural Networks) model.

[0081] In step 1, a large number of position coordinate time combinations are used as input features, and the corresponding speeds are used as labels to form training data. The larger the training set, the richer the samples it contains, and the better the model training effect.

[0082] The sample positions can be designed to cover the entire area as much as possible, the sample time points can be discretized reasonably, and the sample label speed can be accurate and reliable. These can help improve the prediction accuracy and precision of the model.

[0083] Constructing high-quality training set data and providing sufficient information for the training of deep learning models is the basis for obtaining a good prediction model.

[0084] In step 1, by constructing a velocity field training set data with rich information, wide coverage, and accurate labels, it provides input for subsequent deep learning modeling and training.

[0085] Step 2: preprocess the velocity field training set data to make it satisfy the normal distribution.

[0086] Preprocessing the training set data to make it conform to the normal distribution is helpful to improve the performance of the model and the training effect. Normal distribution (also known as Gaussian distribution) is a common probability distribution model in statistics, which has the characteristics of a symmetrical bell-shaped curve.

[0087] Preprocessing the training set data to satisfy the normal distribution can be achieved by following the following steps:

[0088] Data standardization: The training set data is standardized to have zero mean and unit variance. The standardization formula is: (x-mean) / std, where x is the sample value of the training set data, mean is the mean of the training set data, and std is the standard deviation of the training set data. Standardization makes the distribution of data close to the standard normal distribution.

[0089] Data transformation: For skewed distribution data, mathematical transformation operations such as logarithmic transformation and square root transformation can be performed to make it closer to normal distribution. These transformations can be implemented through functions or algorithms.

[0090] Outlier processing: For possible outliers or outliers, appropriate processing methods can be adopted, such as deletion, replacement with mean or median, etc., to reduce the impact of data deviation.

[0091] According to some embodiments, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn library to satisfy a normal distribution.

[0092] Scikit-learn is a Python machine learning library that provides many commonly used machine learning algorithms and tools.

[0093] StandardScaler is a class in Scikit-learn for data standardization, which is also called normalization.

[0094] The main effect of normalization is to adjust the data to a standard normal distribution with a mean of 0 and a variance of 1, which can enhance the performance of many machine learning algorithms.

[0095] The main functions of the StandardScaler class include:

[0096] 1.fit(): Calculate the mean and standard deviation of the training data as reference parameters for subsequent standardization.

[0097] 2. transform(): Standardize the data using the parameters calculated by fit().

[0098] 3. fit_transform(): A combination of fit() and transform() to standardize the training data in one step.

[0099] 4.inverse_transform(): Convert the standardized data back to the original scale.

[0100] Standardization formula:

[0101] X_std=(X-X_mean) / X_std

[0102] Where X_mean is the mean and X_std is the standard deviation.

[0103] A StandardScaler object is only applicable to a single feature. In the present invention, the speed feature can be pre-processed by standardization to establish a StandardScaler of the speed feature.

[0104] In the present invention, specifically, preprocessing the velocity feature to satisfy the normal distribution includes:

[0105] Calculate the mean and standard deviation of the velocity feature;

[0106] For each speed data point, zero mean processing is performed, that is, the mean of all data is subtracted from the value of each data point;

[0107] The zero-mean data are converted to unit variance, that is, the zero-mean data are divided by the standard deviation of all the data.

[0108] Therefore, each preprocessed data point has zero mean and unit variance, making the velocity data satisfy the normal distribution.

[0109] Repeat the above preprocessing process to standardize all speed data in the training set to meet the needs of subsequent model training.

[0110] When using the model for prediction, the speed characteristics of the predicted data should also be subjected to the same preprocessing standardization using the same preprocessing factors to ensure that the predicted data is consistent with the model training data.

[0111] In implementation, as described above, the standardization process can be implemented using the StandardScaler normalization class in the Scikit-learn library. This class provides convenient fit() and transform() methods to first fit the data distribution parameters and then apply them to subsequent data.

[0112] In this embodiment, by preprocessing the training set data to satisfy the normal distribution, the training effect and convergence speed of the model can be improved. Because under the premise of normal distribution, the model can better learn the statistical characteristics and laws of the data, thereby improving the performance and generalization ability of the model.

[0113] Step 3: Build a deep neural network model, use the preprocessed velocity field training set data to train the deep neural network model, and use the manually picked speed as the label speed.

[0114] In some embodiments, when the error between the predicted speed and the labeled speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

[0115] A deep neural network model (DNN) is a type of artificial neural network structure containing multiple hidden layers. The network structure of DNN includes an input layer, a hidden layer, and an output layer, each layer containing multiple nodes. There are usually multiple hidden layers. The nodes are connected by weight coefficients, and the weights are adjusted and optimized through training. Commonly used deep network structures include fully connected networks, convolutional neural networks, etc. The activation function determines the output of the node, and commonly used ones include ReLU, Sigmoid, tanh, etc. During training, the input layer data is calculated to the output layer in sequence through forward propagation, and then each weight is updated through back propagation after evaluating the prediction result error. According to the present invention, the network is repeatedly trained until the error between the prediction speed and the label speed is less than a preset threshold, and it is judged that the deep neural network model has completed training to obtain a trained model.

[0116] Deep networks can perform function mapping approximation and learn complex patterns implicit in data sets. Compared with shallow networks, deep networks have stronger expressive power, but are also prone to overfitting. According to the present invention, in step 2, the training set data is preprocessed to achieve overfitting control.

[0117] Typically, a deep neural network model can be built through the following steps:

[0118] 1. Choose a deep neural network model framework suitable for handling the current problem, such as a multi-layer fully connected network, a convolutional neural network, etc.

[0119] 2. Design the input and output layers of the network according to the dimensions of the input and output features.

[0120] 3. Design the number of hidden layers of the network structure and the number of nodes in each hidden layer.

[0121] 4. Choose a suitable activation function, such as ReLU, Sigmoid, etc.

[0122] 5. Initialize network parameters such as weight matrix and bias vector.

[0123] 6. Use the deep learning framework to build the designed network model.

[0124] 7. Fine-tune the network structure based on specific problems to obtain the final deep neural network.

[0125] This will obtain the framework and parameters of a trainable deep neural network model, preparing for subsequent model training.

[0126] In some embodiments, a function in the Scikit-learn learning library is called to randomly divide the training set data into multiple subsets, and some subsets are used as test sets for cross-validation during training.

[0127] For example, the train_test_split function in the Scikit-learn library can be used to implement the function of randomly dividing the training data, thereby efficiently dividing the training data into a training set and a test set (also called a validation set).

[0128] The main parameters of the train_test_split function include:

[0129] -X: Feature data

[0130] -y: target label

[0131] -test_size: test set size, usually set to 0.2~0.3

[0132] -random_state: random seed to ensure consistency of each division

[0133] For example, calling method:

[0134] from sklearn.model_selection import train_test_split

[0135] X_train, X_test, y_train, y_test=train_test_split(X, y, test_size=0.3, random_state=42)

[0136] In this way, the training set and test set can be randomly divided for model evaluation.

[0137] The train_test_split function uses a random sampling method internally, which can effectively avoid the training set and test set containing only certain specific data distributions, which is more scientific and reasonable than simple sequential sampling.

[0138] In summary, the present invention utilizes deep neural networks and the powerful feature learning and mapping capabilities of multiple hidden layers to perform complex function approximation and pattern recognition, and can achieve high-precision prediction.

[0139] Step 4: Use the trained deep neural network model to predict each CMP point and obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

[0140] The control points are the limited sample points that we manually select to obtain speed information, and the trained model can automatically predict the speed information of any CMP point.

[0141] Predicting the velocities of all CMP points is equivalent to interpolating the three-dimensional velocity field of the entire area. For CMP points outside the control point range, the velocity information predicted by the DNN model is equivalent to extrapolation.

[0142] The model prediction can simultaneously realize the interpolation and extrapolation of CMP points in the whole area, and obtain a complete three-dimensional velocity field information. Compared with manual interpolation and extrapolation, the calculation efficiency is significantly improved.

[0143] In summary, the prediction of CMP points in the entire area with the help of the DNN model can effectively realize the automatic interpolation and extrapolation of the three-dimensional velocity field and obtain richer and more accurate velocity information.

[0144] This embodiment provides a three-dimensional velocity field multivariate nonlinear regression method based on deep learning. Compared with traditional methods, this technology not only realizes automated velocity picking, but also has a picking accuracy comparable to manual picking, significantly improving the accuracy of seismic velocity spectrum picking, thereby improving the quality of velocity analysis.

[0145] Example 2

[0146] According to an embodiment of the present invention, a three-dimensional velocity field multivariate nonlinear regression device based on deep learning is provided, and the method includes:

[0147] A training data acquisition unit is used to obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity;

[0148] A preprocessing unit, used for preprocessing the velocity field training set data to make it satisfy a normal distribution;

[0149] A DNN model training unit, used to build a deep neural network model, use the preprocessed velocity field training set data to train the deep neural network model, and use the artificial picking speed as the label speed;

[0150] The prediction unit is used to predict each CMP point using the trained deep neural network model to obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

[0151] In some implementations, in the preprocessing unit, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn library to satisfy a normal distribution.

[0152] In some embodiments, in the DNN model training unit, when the error between the predicted speed and the label speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

[0153] In some embodiments, in the DNN model training unit, a function in the Scikit-learn learning library is called to randomly divide the training set data into multiple subsets, and some subsets are used as test sets for cross-validation during training.

[0154] This embodiment provides a three-dimensional velocity field multivariate nonlinear regression device based on deep learning. Compared with traditional methods, this technology not only realizes automated velocity picking, but also has a picking accuracy comparable to manual picking, significantly improving the accuracy of seismic velocity spectrum picking, thereby improving the quality of velocity analysis.

[0155] For other detailed descriptions and advantages of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0156] Example 3

[0157] According to another aspect of the present invention, an electronic device is provided. The electronic device comprises:

[0158] Memory, which stores executable instructions:

[0159] A processor runs the executable instructions in the memory to implement the three-dimensional velocity field multivariate nonlinear regression method based on deep learning according to the present invention.

[0160] The method comprises the following steps:

[0161] Step 1, obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity;

[0162] Step 2, preprocessing the velocity field training set data to make it satisfy the normal distribution;

[0163] Step 3, building a deep neural network model, using the preprocessed velocity field training set data to train the deep neural network model, and using the manually picked speed as the label speed;

[0164] Step 4: Use the trained deep neural network model to predict each CMP point and obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

[0165] In some implementations, in step 2, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn library to satisfy a normal distribution.

[0166] In some embodiments, in step 3, when the error between the predicted speed and the labeled speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

[0167] In some embodiments, in step 3, a function in the Scikit-learn library is called to randomly divide the training set data into multiple subsets, and some subsets are used as test sets for cross-validation during training.

[0168] Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.

[0169] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present invention, the processor is used to run the computer-readable instructions stored in the memory.

[0170] The three-dimensional velocity field multivariate nonlinear regression solution based on deep learning provided in this embodiment not only realizes automated velocity picking compared to traditional methods, but also has a picking accuracy comparable to manual picking, significantly improving the accuracy of seismic velocity spectrum picking, thereby improving the quality of velocity analysis.

[0171] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0172] Example 4

[0173] According to another aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program. When the computer program is executed by a processor, the deep learning-based three-dimensional velocity field multivariate nonlinear regression method according to the present invention is implemented.

[0174] The method comprises the following steps:

[0175] Step 1, obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity;

[0176] Step 2, preprocessing the velocity field training set data to make it satisfy the normal distribution;

[0177] Step 3, building a deep neural network model, using the preprocessed velocity field training set data to train the deep neural network model, and using the manually picked speed as the label speed;

[0178] Step 4: Use the trained deep neural network model to predict each CMP point and obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

[0179] In some implementations, in step 2, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn library to satisfy a normal distribution.

[0180] In some embodiments, in step 3, when the error between the predicted speed and the labeled speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

[0181] In some embodiments, in step 3, a function in the Scikit-learn library is called to randomly divide the training set data into multiple subsets, and some subsets are used as test sets for cross-validation during training.

[0182] The computer-readable storage medium according to the embodiment of the present invention stores non-transitory computer-readable instructions, and when the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of the embodiments of the present invention are executed.

[0183] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0184] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present invention.

[0185] The embodiment provides a three-dimensional velocity field multivariate nonlinear regression solution based on deep learning. Compared with traditional methods, this technology not only realizes automated velocity picking, but also has a picking accuracy comparable to manual picking, significantly improving the accuracy of seismic velocity spectrum picking, thereby improving the quality of velocity analysis.

[0186] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0187] Example 5

[0188] The effects of the present invention are illustrated by taking a certain work area in Northwest China as an example.

[0189] Table 1 shows the statistical distribution of the training set data collected in this work area.

[0190] Table 1 Distribution statistics of training data set

[0191]

[0192] From the count row in Table 1, we can see that there are 83 time-speed pairs of control points, each with 60 time-speed pairs, and a total of 4980 sample feature vectors in the training set. The feature vector consists of the following variables: X, Y, Time is the input training data, and speed is the label.

[0193] According to the present invention, the training set data is subjected to standardization preprocessing so that it satisfies the normal distribution (i.e., a Gaussian distribution with zero mean and unit variance) to improve the training effect and convergence speed of the model. The factors used to standardize the training set must be applied to all subsequent data sets used for regression. The StandardScalar normalization class included in Scikit-learn can be used, which can be applied to the standardization of the training set, and can also be subsequently applied to standardize any input data. The normalization function provides a simple and fast method for the standardized processing of the data set.

[0194] When training a supervised learning algorithm, some data can be separated from the training set to evaluate the accuracy of the regression according to the present invention. We can perform cross-validation on the training set to adjust the parameters of the model. Scikit-learn contains functions that randomly divide the training data into many subsets. We use 5% of the data as the test set in cross-validation.

[0195] Table 2 shows the statistical distribution of label speed, prediction speed obtained according to the present invention and prediction error.

[0196] Table 2 Statistical distribution of labeling speed, prediction speed and error

[0197]

[0198] It can be seen from Table 2 that it is obvious that the error of the predicted speed obtained according to the present invention is small enough.

[0199] We use the DNN model to re-pick up the label speed points, such as Figure 2 (a), (b) and (c). Specifically, Figure 2 (a) shows the time-velocity pairs picked manually, Figure 2 (b) shows the predicted time-velocity pairs at the control points according to the present invention, Figure 2 (c) shows the prediction error. It can be seen that the time-speed pairs picked manually and the predicted time-speed pairs at the control points are not much different.

[0200] At the same time, for a certain label speed control point, such as Figure 3 As shown in (a), (b) and (c). Specifically, Figure 3 (a) shows the label time-velocity pair of the control point, Figure 3 (b) shows the time velocity pair of the control point predicted according to the present invention, Figure 3 (c) is the prediction error. It can be seen that the labeled time-speed pair and the predicted time-speed pair look exactly the same.

[0201] Figure 4 (a), (b) and (c) are comparison diagrams of the velocity field effects of three-dimensional seismic exploration after interpolation and extrapolation in the whole area predicted according to the prior art and the embodiment of the present invention. Specifically, Figure 4 (a) is the 3D seismic exploration velocity field after interpolation and extrapolation of the whole area obtained according to conventional existing technology, Figure 4 (b) is a three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area predicted according to an exemplary embodiment of the present invention, Figure 4 (c) is a cross-sectional and longitudinal section of the three-dimensional velocity field predicted according to an exemplary embodiment of the present invention. The entire three-dimensional seismic has 168,300 CMP points, and the prediction time according to the present invention does not exceed 2 minutes. That is, once the model is trained, the prediction will be more effective than conventional interpolation and extrapolation methods.

[0202] For other detailed descriptions of this exemplary embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0203] The three-dimensional velocity field multivariate nonlinear regression scheme based on deep learning disclosed in each embodiment of the present invention can help the processor reduce the density of velocity control points and the time of manual picking, and obtain relatively accurate time-velocity pairs through multivariate nonlinear regression prediction. This technical solution has at least the following advantages:

[0204] 1. Realized the automation of speed picking

[0205] Traditional speed picking relies on manual operation by professionals, which is time-consuming and laborious. The present invention realizes the automation of speed picking by establishing a deep learning model, which greatly improves work efficiency.

[0206] 2. Pickup effect is comparable to artificial

[0207] The test results show that the automatic picking effect of the present invention can match the manual picking effect of professionals, ensuring the quality of the picking result;

[0208] 3. Reduce the density of control points

[0209] Relying on the prediction of the deep learning model, better interpolation and extrapolation effects can be achieved with fewer control points, reducing the demand for control point density;

[0210] 4. Accurate prediction speed

[0211] The present invention predicts the velocity through a deep learning model, which can obtain more accurate and reliable velocity field information compared with simple interpolation and extrapolation;

[0212] 5. Improve workflow efficiency

[0213] By automating the picking speed, time-consuming manual steps in the traditional workflow are skipped, greatly improving the automation and efficiency of the data processing process as a whole;

[0214] 6. Reduce data processing costs

[0215] Relying on automation technology to reduce manual operations can significantly reduce the labor cost and resource consumption of seismic data processing;

[0216] 7. Improve processing scale capabilities

[0217] The automated model can easily process massive 3D seismic data without being limited by the size of the data;

[0218] 8. Standardized processing flow

[0219] The present invention realizes the standardization and normalization of the key step of speed analysis, and reduces the subjective arbitrariness of the processing results;

[0220] In summary, the present invention has important progressive significance for seismic data processing in terms of improving velocity picking efficiency, ensuring picking quality, and reducing processing costs.

[0221] The flow chart in the accompanying drawings shows the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square frame in the flow chart can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the logical function of the specification. In some alternative implementations, the function marked in the square frame can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square frames can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square frame in the flow chart, and the combination of the square frames in the flow chart, can be implemented with a special hardware-based system that performs the function or action of the specification, or can be implemented with a combination of special hardware and computer instructions.

[0222] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is embodied as a computer storage medium, and in another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (SDK) and the like.

[0223] It can be understood that the above embodiments mentioned in the present disclosure can be combined with each other to form a combined embodiment without violating the principle logic. Due to space limitations, the present disclosure will not repeat them. It can be understood by those skilled in the art that in the above methods of the specific implementation, the specific execution order of each step should be determined according to its function and possible internal logic.

[0224] Note that, unless otherwise directly stated, all features disclosed in this specification (including any attached claims, abstracts and drawings) may be replaced by alternative features for achieving the same, equivalent or similar purposes. Therefore, unless otherwise explicitly stated, each feature disclosed is only an example of a group of equivalent or similar features. Where used, further, preferably, further and more preferably are simple beginnings for elaborating another embodiment based on the aforementioned embodiment, and the content of the further, preferably, further or more preferably followed by the combination with the aforementioned embodiment constitutes a complete construction of another embodiment. Several further, preferably, further or more preferably settings following the same embodiment can be arbitrarily combined to form another embodiment.

[0225] It should be understood by those skilled in the art that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

[0226] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A three-dimensional velocity field multivariate nonlinear regression method based on deep learning, It is characterized in that The method comprises: Step 1, obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity; Step 2, preprocessing the velocity field training set data to make it satisfy the normal distribution; Step 3, building a deep neural network model, using the preprocessed velocity field training set data to train the deep neural network model, and using the manually picked speed as the label speed; Step 4: Use the trained deep neural network model to predict each CMP point and obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

2. The method according to claim 1, It is characterized in that In step 2, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn learning library to satisfy the normal distribution.

3. The method according to claim 1, It is characterized in that In step 3, when the error between the predicted speed and the labeled speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

4. The method according to claim 1, It is characterized in that In step 3, a function in the Scikit-learn library is called to randomly divide the velocity field training set data into a plurality of subsets, and some subsets are used as test sets for cross-validation in training.

5. A three-dimensional velocity field multivariate nonlinear regression device based on deep learning, It is characterized in that The method comprises: A training data acquisition unit is used to obtain velocity field training set data consisting of four features: X-coordinate, Y-coordinate, time and velocity; A preprocessing unit, used for preprocessing the velocity field training set data to make it satisfy a normal distribution; A DNN model training unit, used to build a deep neural network model, use the preprocessed velocity field training set data to train the deep neural network model, and use the artificial picking speed as the label speed; The prediction unit is used to predict each CMP point using the trained deep neural network model to obtain the three-dimensional seismic exploration velocity field after interpolation and extrapolation of the entire area.

6. The device according to claim 5, It is characterized in that In the preprocessing unit, the velocity field training set data is preprocessed using the StandardScalar normalization class in the Scikit-learn learning library to make it satisfy the normal distribution.

7. The device according to claim 5, It is characterized in that In the DNN model training unit, when the error between the predicted speed and the label speed is less than a preset threshold, it is determined that the deep neural network model has completed training.

8. The device according to claim 5, It is characterized in that In the DNN model training unit, a function in the Scikit-learn learning library is called to randomly divide the velocity field training set data into a plurality of subsets, and some subsets are used as test sets for cross-validation in training.

9. An electronic device, It is characterized in that The electronic device comprises: A memory storing executable instructions; A processor, wherein the processor runs the executable instructions in the memory to implement the method according to any one of claims 1 to 4.

10. A computer-readable storage medium storing a computer program, wherein the computer program implements the method according to any one of claims 1 to 4 when executed by a processor.

Images