An oil and gas data distortion repair method and system based on data enhancement

Abstract

The present application relates to the technical field of geological data processing, and particularly relates to an oil and gas data distortion repair method and system based on data enhancement, comprising: performing distortion processing on high-quality oil and gas data to form distorted oil and gas data, and performing enhancement processing on the distorted oil and gas data to form diversified distorted data; performing distortion repair on the diversified distorted data according to physical constraints by using a Kalman filtering algorithm to obtain a distorted coarse repair result conforming to the physical constraints; performing distortion repair supplement on the distorted coarse repair result by using a convolutional neural network to obtain a distorted fine repair result; and performing interpolation repair on the distorted oil and gas data based on the distorted fine repair result to restore the high-quality oil and gas data. The present application adopts a two-stage distortion repair mode, which is a coarse repair stage using a Kalman filtering algorithm, distortion repair is performed according to the physical constraint characteristics of the oil and gas data itself, the distortion repair result conforms to the geological signal law, and the credibility of the distortion repair is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological data processing, and particularly relates to an oil and gas data distortion repair method and system based on data enhancement. BACKGROUND

[0002] In the field of oil and gas exploration and development, high-precision geological modeling, reserve estimation and development plan optimization are highly dependent on the integrity and accuracy of key data such as seismic data, well logging curves and production dynamics. However, in actual operation, data may be locally distorted or missing due to factors such as sensor failure, environmental noise, acquisition blind area or transmission interference (such as abnormal jump of well logging curve, blank band of seismic profile, interruption of production pressure signal). If such distorted data is directly used for subsequent analysis, it will significantly reduce the reliability of reservoir characterization, and even cause development decision errors.

[0003] Existing data repair techniques can use machine learning algorithms for distortion repair, which can compensate for data loss and improve efficiency, but still have significant limitations in handling complex distortion scenarios. The main limitation is that the physical laws of oil and gas reservoirs (such as stratigraphic continuity, fluid percolation equation, rock physical property correlation) are ignored, that is, the physical constraints are missing in the distortion repair process, which ultimately affects the reliability of distortion repair. SUMMARY

[0004] The present application aims to provide an oil and gas data distortion repair method and system based on data enhancement, to solve the technical problem that the existing technology lacks physical constraints in the distortion repair process using machine learning algorithms, affecting the reliability of distortion repair.

[0005] To solve the above technical problems, the present application specifically provides the following technical solutions:

[0006] An oil and gas data distortion repair method based on data enhancement, comprising the following steps:

[0007] Distorting high-quality oil and gas data to form distorted oil and gas data, and enhancing the distorted oil and gas data to form diversified distortion data;

[0008] Using Kalman filter algorithm to repair distortion based on physical constraints on diversified distortion data to obtain a distorted rough repair result that meets the physical constraints;

[0009] Using convolutional neural network to repair distortion in the distorted rough repair result to obtain a distorted fine repair result;

[0010] Based on the distorted fine repair result, the distorted oil and gas data is interpolated and repaired to restore high-quality oil and gas data.

[0011] As a preferred embodiment of the present application, the construction method of the diversified distortion data comprises:

[0012] high-quality oil and gas data distortion processing is performed to obtain distorted oil and gas data , wherein is a distortion operator, and the distortion operator includes random missing and strip missing;

[0013] a perturbation operator is added to the distorted oil and gas data to obtain diversified distorted data , wherein the perturbation operator includes local occlusion, noise injection, geometric deformation and Gaussian blur.

[0014] As a preferred scheme of the present application, the method for distortion repair based on physical constraints by using Kalman filtering algorithm includes:

[0015] system modeling is performed by using Kalman filtering algorithm to obtain a state equation for distortion repair and an observation equation for distortion observation;

[0016] the state equation is: , A is a state transition matrix, , is the kth and (k-1)th data in X, is a process excitation noise, ;

[0017] the observation equation is: , H is an observation matrix, is the distorted data , is an observation noise;

[0018] a prediction process is formed according to the state equation and the observation equation:

[0019] ;

[0020] ;

[0021] in the formula, , are prediction values of the kth and (k-1)th data in X, , are error covariance matrices, and Q is a process noise covariance matrix;

[0022] an updating process is formed according to the state equation and the observation equation:

[0023] ;

[0024] ; ​

[0025] ;

[0026] In the formula, is the optimal estimation value of the kth data in X, is the Kalman gain, R is the measurement noise covariance matrix, is the updated error covariance matrix, and I is the unit matrix.

[0027] The is taken as the rough repair result of distortion .

[0028] As a preferred scheme of the present application, the method for supplementing the rough repair result of distortion by using a convolutional neural network comprises:

[0029] The rough repair result of distortion and the diversified distortion data and the high-quality oil and gas data are used to form a data set, and the data set is divided into a training set and a test set;

[0030] The training set is used to train the convolutional neural network, and a distortion repair supplement model is obtained, wherein the distortion repair supplement model is: , is the fine repair result of distortion output by the distortion repair supplement model, is the convolutional neural network;

[0031] The loss function of the distortion repair supplement model is: , wherein is an L1 norm formula, , is a hyperparameter, is a structural similarity index between and .

[0032] The distortion repair supplement model is tested for performance and optimized for parameters by using the test set.

[0033] As a preferred scheme of the present application, the fine repair result of distortion output by the distortion repair supplement model after training is taken as the repair and restoration result of the high-quality oil and gas data.

[0034] As a preferred scheme of the present application, the high-quality oil and gas data X in has been normalized.

[0035] As a preferred scheme of the present application, the present application provides a data enhancement-based data distortion interpolation repair system, which is applied to a data enhancement-based oil and gas data distortion repair method, and the system comprises:

[0036] a data preprocessing unit, configured to perform distortion processing on high-quality oil and gas data to form distorted oil and gas data, and perform enhancement processing on the distorted oil and gas data to form diversified distortion data;

[0037] a coarse repair unit, configured to perform distortion repair on the diversified distortion data according to physical constraints by using a Kalman filtering algorithm to obtain a distorted coarse repair result conforming to the physical constraints;

[0038] a fine repair unit, configured to perform distortion repair supplement on the distorted coarse repair result by using a distortion repair supplement model pre-established by a convolutional neural network to obtain a distorted fine repair result;

[0039] a repair output unit, configured to output high-quality oil and gas data obtained by performing interpolation repair and restoration on the distorted oil and gas data.

[0040] As a preferred scheme of the present application, the method for constructing the diversified distortion data by the data preprocessing unit comprises:

[0041] performing distortion processing on the high-quality oil and gas data to obtain distorted oil and gas data , wherein is a distortion operator, and the distortion operator comprises random missing and strip missing;

[0042] adding a disturbance operator to the distorted oil and gas data to obtain the diversified distortion data , wherein the disturbance operator comprises local occlusion, noise injection, geometric deformation and Gaussian blur.

[0043] As a preferred scheme of the present application, the distortion repair supplement model constructed by the convolutional neural network comprises: , is the distorted coarse repair result, is the diversified distortion data, is a distorted fine repair result output by the distortion repair supplement model, is the convolutional neural network.

[0044] As a preferred scheme of the present application, the high-quality oil and gas data output by the repair output unit is the distorted fine repair result predicted by the distortion repair supplement model. .

[0045] Compared with the prior art, the present application has the following beneficial effects:

[0046] This invention employs a two-stage distortion repair method. The first stage utilizes the Kalman filter algorithm for coarse repair, while the second stage performs distortion repair based on the physical constraints of the oil and gas data itself. This results in distortion repair results that conform to geological signal patterns, thereby improving the reliability of the distortion repair.

[0047] This invention, based on coarse restoration, utilizes a convolutional neural network to supplement the distortion restoration results, thereby repairing complex distortions that the Kalman filter algorithm cannot handle and restoring microscopic geological features, achieving high-quality restoration of distortion restoration. Attached Figure Description

[0048] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0049] Figure 1 A flowchart of a data-enhanced oil and gas data distortion repair method provided in an embodiment of the present invention;

[0050] Figure 2 A block diagram of an oil and gas data distortion repair system based on data augmentation provided in an embodiment of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, this invention provides a method for repairing oil and gas data distortion based on data augmentation, comprising the following steps:

[0053] High-quality oil and gas data is distorted to form distorted oil and gas data, and then the distorted oil and gas data is enhanced to form diversified distorted data.

[0054] The Kalman filter algorithm is used to repair distortion based on physical constraints on diverse distorted data, resulting in a coarse distortion repair result that conforms to the physical constraints; and a coarse distortion repair result that conforms to physical constraints (such as the spatiotemporal continuity of seismic wave propagation, etc.).

[0055] The distortion fine repair result is obtained by using a convolutional neural network to supplement the distortion repair in the distortion coarse repair result.

[0056] In the distortion coarse repair, a neural network algorithm is used to obtain a distortion fine repair result (microscopic geological feature recovery, processing complex distortion that cannot be processed by KF (such as non-Gaussian noise, irregular missing)).

[0057] Based on the distortion fine repair result, the distortion oil and gas data is interpolated and repaired to restore high-quality oil and gas data.

[0058] The present application first uses a physically constrained method for distortion repair, which meets the signal evolution rule (such as the spatiotemporal continuity of seismic wave propagation) or the physical law of oil and gas reservoir (such as stratigraphic continuity, fluid percolation equation, and rock physical property correlation), and uses a Kalman filter algorithm for physically constrained distortion repair.

[0059] The present application completes the physically constrained distortion repair by Kalman filtering to obtain a first-stage repair result. Although the repair result meets the physical constraint, it is difficult to handle complex distortion such as non-Gaussian noise and irregular missing, and it is also difficult to restore geological micro features (such as thin sand layers, small faults, and pores) in distortion repair. Therefore, the present application further uses a machine learning model to supplement the repair based on the physically constrained distortion repair, makes up for the shortcomings of the physically constrained distortion repair, and obtains a more accurate repair result.

[0060] In order to improve the generalization of the distortion repair model based on machine learning / neural network, the diversity of the distortion repair model training data needs to be ensured. Therefore, in the data sample collection stage, a data augmentation technique is combined in the distortion processing to obtain diversified distortion data, thereby forming a data set for training the model to ensure the generalization of the model, as follows:

[0061] The construction method of the diversified distortion data includes:

[0062] high-quality oil and gas data is subjected to distortion processing to obtain distortion oil and gas data , wherein is a distortion operator, and the distortion operator includes random missing and strip missing.

[0063] A perturbation operator is added to the distortion oil and gas data to obtain diversified distortion data , wherein the perturbation operator includes local occlusion, noise injection, geometric deformation, and Gaussian blur.

[0064] ​The whole process firstly simulates the distortion condition that may be generated in the real scene data collection by using the basic distortion method, and then on this basis, the distortion condition is diversified by using some disturbance process, and finally a large number of sample data with sufficient quantity and diverse distortion conditions are obtained, so that a large number of enhanced samples can be generated by using relatively less high-quality data, and the generalization ability and repair robustness of the model to unknown / complex distortion modes are significantly improved.

[0065] The application firstly adopts a distortion repair method with physical constraints to meet the signal evolution rules (such as the space-time continuity of seismic wave propagation) or the physical rules of oil and gas reservoirs (such as stratum continuity, fluid percolation equation, and rock physical property correlation), and adopts a Kalman filter algorithm to perform distortion repair according to physical constraints, as follows:

[0066] The method for performing distortion repair according to physical constraints by using the Kalman filter algorithm comprises:

[0067] A state equation and an observation equation for distortion repair are obtained by using the Kalman filter algorithm to perform prediction system modeling;

[0068] The state equation is: A is a state transition matrix, 、 X is a data sequence, is a process excitation noise, ;

[0069] The observation equation is: H is an observation matrix, is the data sequence after distortion processing, , is an observation noise;

[0070] A prediction process is formed according to the state equation and the observation equation:

[0071] ;

[0072] ;

[0073] In the formula, 、 are prediction values of the kth and (k-1)th data in X, 、 are error covariance matrices, and Q is a process noise covariance matrix;

[0074] An update process is formed according to the state equation and the observation equation:

[0075] ;

[0076] ;

[0077] ;

[0078] wherein, is the optimal estimation value of the kth data in X, is the Kalman gain, and R is the measurement noise covariance matrix, is the updated error covariance matrix, and I is the unit matrix;

[0079] the rough distortion repair result is taken as the rough distortion repair result .

[0080] The present application completes the distortion repair conforming to the physical constraint through the Kalman filter to obtain the repair result of the first stage, and the repair result conforms to the physical constraint, but it is difficult to deal with the complexity distortion such as the non-Gaussian noise, irregular missing, and it is also difficult to restore the geological microscopic features (such as thin sand layer, micro fault, pore and crack, etc.) in the distortion repair. Therefore, the present application further utilizes the machine learning model for repair and supplement on the basis of the distortion repair conforming to the physical constraint, makes up for the deficiency of the distortion repair of the physical constraint, and obtains a repair result with higher accuracy, as follows:

[0081] The method for repairing and supplementing the distortion in the rough distortion repair result by utilizing the convolutional neural network comprises:

[0082] the rough distortion repair result and the diversified distortion data and the high-quality oil and gas data are formed into a data set, and the data set is divided into a training set and a test set;

[0083] the training set is utilized to train the convolutional neural network to obtain a distortion repair and supplement model, and the distortion repair and supplement model is: , is the fine distortion repair result output by the distortion repair and supplement model, is the convolutional neural network;

[0084] the loss function of the distortion repair and supplement model is: , wherein is an L1 norm formula, , is a hyperparameter, is a structural similarity index between and ;

[0085] the performance test and parameter optimization of the distortion repair and supplement model are performed through the test set.

[0086] The present application constructs a distortion repair supplementary model (machine learning model) on the basis of physical constraint distortion repair, can repair the distortion which is difficult for Kalman filtering algorithm to do, and thus obtains a more accurate repair result on the basis of physical constraint repair.

[0087] The loss term for training the distortion repair supplementary model in the present application includes three parts, the first part is the precision constraint term, that is, The distortion repair result output by the distortion repair supplementary model on the basis of the distortion repair result of physical constraint With high-quality data before distortion The difference is minimal, thereby ensuring that the two-stage distortion repair result of Kalman filtering algorithm and distortion repair supplementary model realizes optimal distortion repair accuracy, the second part is the structure constraint term, that is, the structural similarity index , which constrains the two-stage distortion repair process to maintain the spatial structure characteristics of oil and gas data when repairing distortion, so that the restoration degree of visual structure in the two-stage distortion repair process is optimized, and unreasonable repair results in geology are avoided, and the third part is the physical constraint term, that is, which ensures that the repair result of the distortion repair supplementary model still satisfies the physical constraint in the distortion repair result of Kalman filtering algorithm, maintains the physical consistency in the repair process, avoids the disappearance of the physical constraint, and produces random constraint.

[0088] The present application trains the distortion repair supplementary model by using a loss function including three parts, and finally can obtain a fine repair result which conforms to the physical constraint and has higher repair accuracy.

[0089] The present application adopts a two-stage repair process, which is complementary to each other, Kalman filtering algorithm supplements the defects of machine learning algorithm in lacking physical constraint, machine learning algorithm makes up for the defects of Kalman filtering algorithm in being unable to undertake complex repair, and the combination of the two finally realizes a high-quality distortion repair process.

[0090] The distortion fine repair result output by the distortion repair supplementary model after training As a repair and restoration result of high-quality oil and gas data.

[0091] The high-quality oil and gas data X has undergone normalization processing.

[0092] As Figure 2 shown, the present application provides a data enhancement-based especially data distortion interpolation repair system, which is applied to a data enhancement-based oil and gas data distortion repair method, and the system comprises:

[0093] A data preprocessing unit is configured to perform distortion processing on high-quality oil and gas data to form distorted oil and gas data, and to perform enhancement processing on the distorted oil and gas data to form diversified distorted data.

[0094] A coarse repair unit is configured to perform distortion repair on the diversified distorted data according to physical constraints by using a Kalman filter algorithm to obtain a distorted coarse repair result that conforms to the physical constraints.

[0095] A fine repair unit is configured to perform distortion repair supplement on the distorted coarse repair result by using a distortion repair supplement model pre-established by a convolutional neural network to obtain a distorted fine repair result.

[0096] A repair output unit is configured to output high-quality oil and gas data obtained by performing interpolation repair and restoration on the distorted oil and gas data.

[0097] The method for constructing diversified distorted data by the data preprocessing unit comprises the following steps:

[0098] Performing distortion processing on the high-quality oil and gas data to obtain distorted oil and gas data . Among them, is a distortion operator, and the distortion operator includes random missing and strip missing.

[0099] Adding a disturbance operator to the distorted oil and gas data to obtain diversified distorted data .Among them, the disturbance operator includes local occlusion, noise injection, geometric deformation and Gaussian blur.

[0100] The distortion repair supplement model constructed by the convolutional neural network is as follows: , is the distorted coarse repair result, is the diversified distorted data, is the distorted fine repair result output by the distortion repair supplement model, and the convolutional neural network is

[0101] The high-quality oil and gas data output by the repair output unit is the distorted fine repair result predicted by the distortion repair supplement model .

[0102] The present application adopts a two-stage distortion repair method, which is a coarse repair stage using a Kalman filter algorithm, and performs distortion repair according to the physical constraint characteristics of the oil and gas data itself, so as to obtain a distortion repair result that conforms to the geological signal rule and improves the credibility of the distortion repair.

[0103] The application uses a convolutional neural network to supplement the distortion repair result of the rough repair, repair the complexity distortion that the Kalman filter algorithm cannot process, and restore the micro geological features, so that the high-quality restoration of the distortion repair is realized.

[0104] The above examples are only exemplary embodiments of the application and are not intended to limit the application, and the protection scope of the application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the application within the spirit and protection scope of the application, and such modifications or equivalent replacements are also considered to fall within the protection scope of the application.

Claims

1. A method for oil and gas data distortion repair based on data augmentation, characterized in that, The method comprises the following steps: The high-quality oil and gas data is distorted to form distorted oil and gas data, and the distorted oil and gas data is enhanced to form diversified distorted data; The distorted data is repaired according to physical constraints by using a Kalman filtering algorithm to obtain a distorted rough repair result conforming to the physical constraints; The distorted rough repair result is supplemented by using a convolutional neural network to obtain a distorted fine repair result; The distorted oil and gas data is repaired and restored to the high-quality oil and gas data by interpolation based on the distorted fine repair result; The method for constructing the diversified distorted data comprises: high-quality oil and gas data perform distortion processing to obtain distorted oil and gas data wherein is a distortion operator, the distortion operator including random missing, strip missing; In distorted oil and gas data Adding perturbation operators on top Obtaining diversified distorted data Wherein the perturbation operators include local occlusion, noise injection, geometric deformation, Gaussian blur; The method for supplementing the distorted rough repair result by using the convolutional neural network comprises: distortion rough repair results and diversified distortion data with high-quality oil and gas data form a dataset and divide the dataset into a training set and a test set; The convolutional neural network is trained by using the training set to obtain a distortion repair supplement model, and the distortion repair supplement model is: , is a distortion repair result output by the distortion repair supplement model, is the convolutional neural network; Loss function of the distortion repair complementary model is: , where is an L1 norm formula, , is a hyperparameter, is and is a structural similarity index between The performance of the distorted repair supplement model is tested and the parameters are optimized by using a test set.

2. The method of claim 1, wherein: The method for repairing the distorted data according to the physical constraints by using the Kalman filtering algorithm comprises: A state equation and an observation equation for the distorted repair are obtained by using the Kalman filtering algorithm to predict a system model; The state equation is: , A is a state transition matrix, , is the k, k-1 data in X, is the process excitation noise, ; The observation equation is: H is an observation matrix, is a distortion-processed , is an observation noise; A prediction process is formed according to the state equation and the observation equation: ； ； wherein, , are the prediction of the k, k-1 data in X respectively, , are the error covariance matrix respectively, and Q is the process noise covariance matrix. An updating process is formed according to the state equation and the observation equation: ； ； ； wherein is the optimal estimate of the kth data in X, is the Kalman gain, and R is the measurement noise covariance matrix, is the updated error covariance matrix, and I is the identity matrix. will be described below. as a distorted coarse repair result .

3. The method of claim 1, wherein: The distortion fine repair result output by the distortion repair supplementary model after training is completed As a repair and restoration result of high-quality oil and gas data.

4. The method of claim 1, wherein: High quality oil and gas data X Normalized.

5. A data augmentation based, in particular data distortion interpolation, inpainting system, characterized in that The system is applied to the oil and gas data distortion repair method based on data enhancement according to any one of claims 1-4, and the system comprises: A data preprocessing unit is configured to distort the high-quality oil and gas data to form distorted oil and gas data, and to enhance the distorted oil and gas data to form diversified distorted data; A rough repair unit is configured to repair the distorted data according to physical constraints by using a Kalman filtering algorithm on the diversified distorted data to obtain a distorted rough repair result conforming to the physical constraints; A fine repair unit is configured to supplement the distorted rough repair result by using a distorted repair supplement model pre-established by a convolutional neural network to obtain a distorted fine repair result; A repair output unit is configured to output high-quality oil and gas data obtained by repairing and restoring the distorted oil and gas data by interpolation.

6. The data augmentation based oil and gas data distortion repair system of claim 5, wherein: The method for constructing the diversified distorted data by the data preprocessing unit comprises: high-quality oil and gas data perform distortion processing to obtain distorted oil and gas data wherein is a distortion operator, the distortion operator including random missing, strip missing; In distorted oil and gas data Adding perturbation operators on top , resulting in diversified distorted data Wherein the perturbation operators include local occlusion, noise injection, geometric deformation, Gaussian blur.

7. The data augmentation based oil and gas data distortion repair system of claim 6, wherein: The distortion repair supplement model constructed by the convolutional neural network is as follows: , is a distortion rough repair result, is diversified distortion data, is a distortion fine repair result output by the distortion repair supplement model, is a convolutional neural network.

8. The data augmentation based oil and gas data distortion repair system of claim 7, wherein: The high-quality oil and gas data output by the repair output unit is used as a distortion repair supplement model prediction .

Images