Methods and apparatus for seismic fault and horizon interpretation based on multi-task deep learning

Abstract

This specification provides a method and apparatus for interpreting seismic faults and horizons based on multi-task deep learning. The method includes: synthesizing a velocity model with spatial relationships of faults and horizons based on three-dimensional seismic numerical simulation, and performing forward modeling on the velocity model to obtain synthetic seismic data; automatically labeling the synthetic seismic data with fault and horizon labels as a training dataset; using the training dataset to perform inference learning on a multi-task deep learning network model to obtain an initial network model; acquiring a small amount of fault and horizon label data from the seismic data to be interpreted; using the small amount of fault and horizon label data from the seismic data to be interpreted to perform transfer learning on the initial network model to obtain a target interpretation model; and interpreting the seismic data to be interpreted based on the target interpretation model to obtain fault and horizon interpretation results. The embodiments in this specification can improve the accuracy of fault and horizon interpretation of seismic data.

Description

Technical Field

[0001] This specification relates to the field of seismic data interpretation technology, and in particular to a method and apparatus for interpreting seismic faults and horizons based on multi-task deep learning. Background Technology

[0002] Seismic interpretation is a crucial part of oil and gas exploration, with fault and stratigraphic interpretation being two key tasks. Traditionally, fault and stratigraphic interpretation is performed manually by interpreters through a human-computer interaction process, which is extremely labor-intensive, inefficient, and prone to human error.

[0003] To address this, researchers have proposed several methods for automatic interpretation of seismic data. Regarding seismic fault interpretation, Marfurt et al. (1999) proposed a characteristic structure coherence algorithm to construct a covariance matrix and eigenvalue matrix with eigenvalue decomposition matrices; Pedersen et al. (2002) proposed an ant-tracking method to extract small faults and then merge them to form larger fault planes; Wu et al. (2016) proposed a 3D image processing method to automatically calculate the fault plane and tilt slip vector of image samples adjacent to the fault. Regarding seismic horizon interpretation, Zeng et al. (1998) proposed an interpolation method that first manually selects multiple reference horizons and then constructs the entire seismic horizon by interpolating the interpreted horizon plane; Borgos et al. (2003) first uses the troughs of seismic amplitudes to generate horizon blocks, and then merges the similarity of each horizon block to form a single horizon. Lomask et al. (2006) first used reflection dip to flatten the seismic reflection phase axis, and then generated a relative geological time (RGT) volume based on the flattened seismic data volume for horizon picking. Wu et al. (2012) proposed a 3D graphical phase unfolding method with horizontal and unconformity constraints to generate the relative geological time volume, which improves the horizon interpretation effect in tectonically complex areas. Wu et al. (2018) proposed a new automatic seismic horizon interpretation method with local slope and multi-grid correlation.

[0004] However, in the process of realizing this application, the inventors of this application discovered that when interpreting faults and strata respectively, the above methods did not consider the spatial structural combination relationship of faults and strata, which resulted in the final obtained strata and faults being non-closed in spatial structure. Summary of the Invention

[0005] The purpose of the embodiments in this specification is to provide a method and apparatus for interpreting seismic faults and horizons based on multi-task deep learning, so as to improve the accuracy of seismic data fault and horizon interpretation.

[0006] To achieve the above objectives, on the one hand, embodiments of this specification provide a method for interpreting seismic faults and horizons based on multi-task deep learning, including:

[0007] A velocity model with spatial combination relationships of faults and horizons is synthesized based on three-dimensional seismic numerical simulation, and forward modeling is performed on the velocity model to obtain synthetic seismic data.

[0008] The synthetic seismic data is automatically labeled with fault labels and horizon labels to serve as a training dataset;

[0009] The training dataset is used to perform inference learning on a multi-task deep learning network model to obtain an initial network model.

[0010] Obtain fault and horizon label data from a small amount of uninterpreted seismic data;

[0011] The initial network model is transferred to the fault and horizon label data of the small amount of earthquake data to be interpreted in order to obtain the target interpretation model.

[0012] Based on the target interpretation model, fault and horizon interpretation is performed on the seismic data to be interpreted, and the fault and horizon interpretation results are obtained.

[0013] In the embodiments of this specification, the step of using the training dataset to perform inference learning on a multi-task deep learning network model includes:

[0014] The training dataset is input into a multi-task deep learning network model for inference learning to obtain feature maps of faults and layers.

[0015] The fault and horizon feature maps are classified to obtain the fault and horizon category probabilities for each earthquake data in the training dataset.

[0016] The loss values ​​between the fault and layer category probabilities and their corresponding labels are calculated using a preset loss function.

[0017] When the loss value is greater than a preset threshold, the weight parameters of each layer in the multi-task deep learning network model are updated using the loss value, and training continues until the loss value is no greater than the preset threshold.

[0018] In the embodiments of this specification, the multi-task deep learning network model includes: a backbone network for learning the relationship between faults and layers, a fault detection branch network for learning fault features, and a layer interpretation branch network for learning layer features; wherein, the shared input features between layers in the backbone network, the fault detection branch network, and the layer interpretation branch network are formed by tensor concatenation.

[0019] In the embodiments of this specification, the tensor concatenation of shared input features between layers in the backbone network is obtained by the following formula:

[0020]

[0021] Among them, X n_input X represents the input data of the nth layer of the backbone network. Concatenate represents tensor concatenation, N represents the number of layers in the backbone network, and X represents the input data of the nth layer of the backbone network. n-1_output This is the output data of the (n-1)th layer network within the backbone network.

[0022] In the embodiments of this specification, the tensor concatenation of shared input features among the layers in the tomographic detection branch network is obtained by the following formula:

[0023]

[0024] Among them, X F,m_intput This represents the input data for the m-th layer of the tomography detection branch network. Concatenate indicates tensor concatenation, N represents the number of layers in the backbone network, M represents the number of layers in the tomography detection branch network, K represents the number of layers in the layer interpretation branch network, and X... n_output The output data of the nth layer network within the backbone network, X F,m-1_output X represents the output data of the (m-1)th layer network within the tomographic detection branch network. L,k-1_output This explains the output data of the (k-1)th layer of the branch network.

[0025] In the embodiments of this specification, the tensor concatenation of shared input features between layers in the layered interpretation branch network is obtained by the following formula:

[0026]

[0027] Among them, X L,k_intput For the input data of the k-th layer of the layer interpretation branch network, Concatenate represents tensor concatenation, N represents the number of layers in the backbone network, M represents the number of layers in the tomography detection branch network, K represents the number of layers in the layer interpretation branch network, and X... n_output The output data of the nth layer network within the backbone network, X L,k-1_output To interpret the output data of the (k-1)th layer network within the branch network, X F,m-1_output This is the output data of the (m-1)th layer network within the fault detection branch network.

[0028] In the embodiments of this specification, the loss function includes:

[0029]

[0030] Where loss is the loss function, loss F Let loss be the loss function of the tomography detection branch network. L Let β be the loss function of the layer interpretation branch network, and λ be the weights of the loss functions of the fault detection branch network and the layer interpretation branch network, respectively, used to adjust the training and optimization effect of the network model. P is the number of input seismic data samples. Y is the predicted tomographic data for the i-th sample data. i For the labeled tomographic data of the i-th sample, Z represents the predicted stratigraphic data for the i-th sample data. i Let represent the label layer data of the i-th sample data, α be the positive and negative sample importance balance factor, and γ be the simple and complex sample modulation factor.

[0031] On the other hand, embodiments of this specification also provide a seismic fault and horizon interpretation device based on multi-task deep learning, including:

[0032] The data synthesis module is used to synthesize a velocity model with spatial combination relationships of faults and horizons based on three-dimensional seismic numerical simulation, and to perform forward modeling on the velocity model to obtain synthetic seismic data.

[0033] The first labeling module is used to automatically label the synthetic seismic data with fault labels and horizon labels as a training dataset;

[0034] The first training module is used to perform inference learning on a multi-task deep learning network model using the training dataset to obtain an initial network model.

[0035] The second label module is used to obtain fault and horizon label data for a small amount of earthquake data to be interpreted.

[0036] The second training module is used to perform transfer learning on the initial network model using the fault and horizon label data of the small amount of earthquake data to be interpreted, so as to obtain the target interpretation model.

[0037] The data interpretation module is used to interpret faults and horizons in the seismic data to be interpreted according to the target interpretation model, and to obtain the interpretation results of faults and horizons.

[0038] On the other hand, embodiments of this specification also provide a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the computer program, when run by the processor, executes instructions for the above-described method.

[0039] On the other hand, embodiments of this specification also provide a computer storage medium storing a computer program thereon, which, when run by the processor of a computer device, executes instructions for the above-described method.

[0040] As can be seen from the technical solutions provided in the embodiments of this specification above, in these embodiments, the spatial combination relationship of faults and strata is used in the three-dimensional numerical simulation synthetic velocity model (the spatial combination relationship of faults and strata refers to the spatial structural combination relationship that is inherently and closely related to geological strata, and this relationship determines the accuracy of seismic tectonic interpretation); thus, the synthetic seismic data obtained by forward modeling the velocity model will also contain the spatial combination relationship of faults and strata; in this way, when the synthetic seismic data is automatically labeled with fault labels and strata labels and used as a training dataset, and then used for training a multi-task deep learning network model, the multi-task deep learning network model can learn the accurate combination relationship of faults and strata, thereby making the predicted spatial structural combination relationship of strata and faults more accurate. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0042] Figure 1 The flowcharts of seismic fault and horizon interpretation methods based on multi-task deep learning in some embodiments of this specification are shown.

[0043] Figure 2 The flowcharts shown here illustrate inference learning of multi-task deep learning network models using training datasets in some embodiments of this specification.

[0044] Figure 3 This specification shows schematic diagrams illustrating the structure of multi-task deep learning network models in some embodiments.

[0045] Figure 4a A schematic diagram of preprocessed seismic data is shown in one embodiment of this specification;

[0046] Figure 4b A schematic diagram of layer label data in one embodiment of this specification is shown;

[0047] Figure 4c A schematic diagram of tomographic label data in one embodiment of this specification is shown;

[0048] Figure 5a This specification shows a schematic diagram of preprocessed seismic data to be interpreted in one embodiment.

[0049] Figure 5b This specification shows a schematic diagram of predicted stratigraphic label data in one embodiment;

[0050] Figure 5c A schematic diagram of predicted fault label data is shown in one embodiment of this specification;

[0051] Figure 6 The present specification shows a structural block diagram of a seismic fault and horizon interpretation device based on multi-task deep learning in some embodiments;

[0052] Figure 7 A structural block diagram of a computer device in some embodiments of this specification is shown.

[0053] [Explanation of Labels in the Attached Image]

[0054] 61. Data synthesis module;

[0055] 62. First tag module;

[0056] 63. First Training Module;

[0057] 64. Second tag module;

[0058] 65. Second training module;

[0059] 66. Data Interpretation Module;

[0060] 702. Computer equipment;

[0061] 704, Processor;

[0062] 706. Memory;

[0063] 708. Drive mechanism;

[0064] 710. Input / output interfaces;

[0065] 712. Input devices;

[0066] 714. Output devices;

[0067] 716. Presentation equipment;

[0068] 718. Graphical User Interface;

[0069] 720. Network interface;

[0070] 722. Communication link;

[0071] 724. Communication bus. Detailed Implementation

[0072] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0073] Given the low accuracy of existing automatic fault and horizon interpretation techniques for seismic data, this manual provides an improved automatic interpretation scheme for seismic data faults and horizons.

[0074] This specification provides an embodiment of a seismic fault and horizon interpretation method based on multi-task deep learning, which can be applied to any suitable computer device. References Figure 1 As shown, in some embodiments, the seismic fault and horizon interpretation method based on multi-task deep learning may include the following steps:

[0075] S101. Based on the three-dimensional seismic numerical simulation, a velocity model with spatial combination relationships of faults and horizons is synthesized, and the velocity model is subjected to forward modeling to obtain synthetic seismic data.

[0076] S102. Automatically label the synthetic seismic data with fault labels and horizon labels to use as a training dataset.

[0077] S103. Use the training dataset to perform inference learning on the multi-task deep learning network model to obtain an initial network model.

[0078] S104. Obtain fault and horizon label data for a small amount of earthquake data to be interpreted.

[0079] S105. Using the fault and horizon label data of the small amount of earthquake data to be interpreted, the initial network model is transferred to obtain the target interpretation model.

[0080] S106. Based on the target interpretation model, perform fault and horizon interpretation on the seismic data to be interpreted to obtain the fault and horizon interpretation results.

[0081] In the embodiments of this specification, the spatial combination relationship of faults and strata is used in the synthetic velocity model obtained by three-dimensional numerical simulation (the spatial combination relationship of faults and strata refers to the spatial structural combination relationship that is inherently and closely related to geological strata, and this relationship determines the accuracy of seismic tectonic interpretation); thus, the synthetic seismic data obtained by forward modeling the velocity model will also contain the spatial combination relationship of faults and strata; in this way, when the synthetic seismic data is automatically labeled with fault labels and strata labels and used as a training dataset, and then used to train a multi-task deep learning network model, the multi-task deep learning network model can learn the accurate combination relationship of faults and strata, thereby making the predicted spatial structural combination relationship of strata and faults more accurate.

[0082] Furthermore, based on the initial network model obtained by training a multi-task deep learning network model on the training dataset, when it is necessary to interpret faults and horizons in the seismic data to be interpreted, the initial network model is first transferred to a small amount of fault and horizon label data of the seismic data to be interpreted to obtain a target interpretation model that is more suitable for the seismic data scenario. Therefore, interpreting faults and horizons in the seismic data to be interpreted based on the target interpretation model can further improve the accuracy of seismic data layer and horizon interpretation, making the interpretation results more likely to be closer to the real faults and horizons.

[0083] Seismic numerical simulation is a seismic simulation method that, assuming the subsurface medium structure model and corresponding physical parameters are known, uses computer numerical calculations to study the propagation laws of seismic waves in various subsurface media and calculates the numerical seismic records that should be observed at various observation points on the surface or underground. Seismic numerical simulation can include one-dimensional simulation, two-dimensional simulation, and three-dimensional simulation. Three-dimensional simulation is used in the embodiments of this specification. A velocity model, also known as a seismic velocity model (e.g., a surface velocity model), is a mathematical model used to characterize geological structures. In exploration seismic data processing, the accuracy of the velocity model directly affects the imaging accuracy of depth migration imaging.

[0084] In some embodiments, the step of synthesizing a velocity model with spatial combinations of faults and horizons based on three-dimensional seismic numerical simulation includes the following steps:

[0085] First, a horizontally layered three-dimensional velocity model is formed using a synthesis algorithm. Then, a Gaussian perturbation is added to the horizontally layered three-dimensional velocity model (i.e., a construction perturbation is applied to the horizontally layered three-dimensional velocity model) to form a non-horizontally layered three-dimensional velocity model. Based on this, a fault is added to the non-horizontally layered three-dimensional velocity model using a position displacement method to form a faulted non-horizontally layered three-dimensional velocity model. In this way, an accurate spatial combination relationship between faults and layers can be formed.

[0086] Since synthetic seismic data contains spatial relationships of faults and horizons, any suitable automatic fault identification and horizon picking method can be used to automatically label faults and horizons in the synthetic seismic data. In this way, the automatically labeled synthetic seismic data can be used as a training dataset.

[0087] For example, in an exemplary embodiment, the synthetic seismic data obtained based on the above method can be as follows: Figure 4a As shown (the process of obtaining synthetic seismic data is the preprocessing process). And by processing data such as... Figure 4a The synthetic seismic data shown can be labeled with faults and horizons to obtain the corresponding results as follows: Figure 4b The tomographic label data shown and such Figure 4c The tomographic label data shown.

[0088] It should be noted that, in the embodiments of this specification, multi-task learning refers to a network model being able to perform multiple tasks. Transfer learning, on the other hand, refers to applying knowledge learned from the source domain (source domin) to the target domain (target domin) to improve the learning performance in the target domain.

[0089] refer to Figure 2 As shown, in some embodiments, the inference learning of the multi-task deep learning network model using the training dataset may include the following steps:

[0090] S201. Input the training dataset into the multi-task deep learning network model for inference learning to obtain feature maps of faults and layers.

[0091] like Figure 3 As shown, in some embodiments, the multi-task deep learning network model may include a backbone network for learning the relationship between faults and layers, a fault detection branch network for learning fault features, and a layer interpretation branch network for learning layer features. The backbone network, fault detection branch network, and layer interpretation branch network are all composed of multiple layers of deep learning networks. For example, the backbone network may consist of N layers of deep learning networks, the fault detection branch network may consist of M layers of deep learning networks, and the layer interpretation branch network may consist of K layers of deep learning networks, where N, M, and K are all positive integers greater than 1.

[0092] In some embodiments, the shared input features among the layers in the backbone network, fault detection branch network, and layer interpretation branch network are formed by tensor concatenation. The multi-task deep learning network model formed by tensor concatenation and the interconnected parts can be beneficial for effectively learning the spatial structural combination relationship of faults and layers in seismic data.

[0093] In some embodiments, the tensor concatenation of shared input features among layers in the backbone network can be obtained using the following formula:

[0094]

[0095] Among them, X n_input X represents the input data of the nth layer of the backbone network. Concatenate represents tensor concatenation, N represents the number of layers in the backbone network, and X represents the input data of the nth layer of the backbone network. n-1_output This is the output data of the (n-1)th layer network within the backbone network.

[0096] In some embodiments, the tensor concatenation of shared input features among layers within the tomography detection branch network can be obtained using the following formula:

[0097]

[0098] Among them, X F,m_intput This represents the input data for the m-th layer of the tomography detection branch network. Concatenate indicates tensor concatenation, N represents the number of layers in the backbone network, M represents the number of layers in the tomography detection branch network, K represents the number of layers in the layer interpretation branch network, and X... n_output The output data of the nth layer network within the backbone network, X F,m-1_output X represents the output data of the (m-1)th layer network within the tomographic detection branch network. L,k-1_output This explains the output data of the (k-1)th layer of the branch network.

[0099] In some embodiments, the tensor concatenation of shared input features among layers within the layered interpretation branch network can be obtained using the following formula:

[0100]

[0101] Among them, X L,k_intput For the input data of the k-th layer of the layer interpretation branch network, Concatenate represents tensor concatenation, N represents the number of layers in the backbone network, M represents the number of layers in the tomography detection branch network, K represents the number of layers in the layer interpretation branch network, and X... n_output The output data of the nth layer network within the backbone network, X L,k-1_output To interpret the output data of the (k-1)th layer network within the branch network, X F,m-1_output This is the output data of the (m-1)th layer network within the fault detection branch network.

[0102] S202. Classify the fault and layer feature maps to obtain the fault and layer category probabilities for each earthquake data in the training dataset.

[0103] Among them, the fault and horizon category probabilities of each earthquake data are the fault probability and horizon category probability of each earthquake data; the fault probability of each earthquake data is used to characterize the probability that the corresponding location of the earthquake data is a fault structure; the horizon category probability of each earthquake data is used to characterize the probability that the corresponding location of the earthquake data is a certain horizon.

[0104] S203. Calculate the loss value between the fault and layer category probabilities and the corresponding labels using a preset loss function.

[0105] The loss function is used to evaluate the degree to which the model's predictions differ from the actual values. A smaller loss value indicates a smaller discrepancy between the two. Any suitable loss function can be used to calculate the loss value as needed. For example, in some embodiments, the following loss function can be used to calculate the loss between the fault and slice class probabilities and their corresponding labels:

[0106]

[0107] Where loss is the loss function, loss F Let loss be the loss function of the tomography detection branch network. L Let β be the loss function of the layer interpretation branch network, and λ be the weights of the loss functions of the fault detection branch network and the layer interpretation branch network, respectively, used to adjust the training and optimization effect of the network model. P is the number of input seismic data samples. Y is the predicted tomographic data for the i-th sample data. i For the labeled tomographic data of the i-th sample, Z represents the predicted stratigraphic data for the i-th sample data. i Let represent the label layer data of the i-th sample data, α be the positive and negative sample importance balance factor, and γ be the simple and complex sample modulation factor.

[0108] S204. Determine whether the loss value is greater than the preset threshold. If the loss value is not greater than the preset threshold, proceed to step S205; otherwise, proceed to step S206.

[0109] S205. When the loss value is not greater than a preset threshold, the current multi-task deep learning model is used as the initial network model.

[0110] S206. When the loss value is greater than a preset threshold, the weight parameters of each layer in the multi-task deep learning network model are updated using the loss value, and the process jumps to step S201 to continue training.

[0111] When the loss value is greater than the preset threshold, it indicates that the current multi-task deep learning network model does not meet the requirements. At this time, the weight parameters of each layer in the multi-task deep learning network model can be updated using the loss value (i.e., the multi-task deep learning network model is updated) to continue training.

[0112] When the loss value is not greater than the preset threshold, it indicates that the current multi-task deep learning network model meets the requirements. At this time, the current multi-task deep learning model can be used as the initial network model.

[0113] The method of obtaining fault and horizon label data of a small amount of earthquake data to be interpreted in step S105 is similar to the process of obtaining synthetic earthquake data and automatically labeling it (i.e., fault labels and horizon labels) in steps S101 and S102 above. The difference is that the step of obtaining fault and horizon label data of a small amount of earthquake data to be interpreted only involves a small part of the earthquake data to be interpreted.

[0114] The process of using the fault and stratigraphic label data of the small amount of earthquake data to be interpreted to perform transfer learning on the initial network model in step S105 is somewhat similar to the process of using the training dataset to perform inference learning on the multi-task deep learning network model to obtain the initial network model in step S103 above. It also requires the loss function to determine under what conditions the training will terminate, which will not be elaborated here.

[0115] In step S105, interpreting faults and horizons in the seismic data to be interpreted according to the target interpretation model means inputting the seismic data to be interpreted into the target interpretation model, and the corresponding output results are the fault and horizon interpretation results. For example, in an exemplary embodiment, using... Figure 5a Taking the seismic data to be interpreted shown as an example (the process of synthesizing the seismic data to be interpreted is the preprocessing process), based on the seismic fault and horizon interpretation method based on multi-task deep learning in the embodiments of this specification, it is possible to predict the following: Figure 5b The stratigraphic interpretation results shown and as follows Figure 5c The fault interpretation results are shown.

[0116] Although the process described above includes multiple operations that occur in a specific order, it should be clearly understood that these processes may include more or fewer operations, which may be executed sequentially or in parallel (e.g., using parallel processors or a multithreaded environment).

[0117] Corresponding to the aforementioned method for interpreting seismic faults and horizons based on multi-task deep learning, this specification also provides a device for interpreting seismic faults and horizons based on multi-task deep learning. (Reference) Figure 6As shown, in some embodiments, the seismic data tomography and stratigraphic interpretation device may include: a data synthesis module 61, a first label module 62, a first training module 63, a second label module 64, a second training module 65, and a data interpretation module 66. Wherein:

[0118] The data synthesis module 61 can be used to synthesize a velocity model with spatial combination relationships of faults and horizons based on three-dimensional seismic numerical simulation, and to perform forward modeling on the velocity model to obtain synthesized seismic data.

[0119] The first labeling module 62 can be used to automatically label the synthetic seismic data with fault labels and horizon labels as a training dataset;

[0120] The first training module 63 can be used to perform inference learning on a multi-task deep learning network model using the training dataset to obtain an initial network model.

[0121] The second label module 64 can be used to obtain fault and horizon label data for a small amount of seismic data to be interpreted.

[0122] The second training module 65 can be used to perform transfer learning on the initial network model using the fault and horizon label data of the small amount of earthquake data to be interpreted, so as to obtain the target interpretation model.

[0123] The data interpretation module 66 can be used to interpret faults and horizons in the seismic data to be interpreted according to the target interpretation model, and obtain the interpretation results of faults and horizons.

[0124] In the device embodiments described in this specification, the first training module may include:

[0125] The feature map acquisition submodule is used to input the training dataset into a multi-task deep learning network model for inference learning, and obtain feature maps of faults and layers.

[0126] The classification processing submodule is used to classify the fault and layer feature maps and obtain the fault and layer category probabilities of each earthquake data in the training dataset.

[0127] The loss value calculation submodule is used to calculate the loss value between the fault and layer category probabilities and the corresponding labels using a preset loss function.

[0128] The training control submodule is used to update the weight parameters of each layer in the multi-task deep learning network model using the loss value when the loss value is greater than a preset threshold, and continue training until the loss value is no greater than the preset threshold.

[0129] In the device embodiments described in this specification, the multi-task deep learning network model may include: a backbone network for learning the relationship between faults and layers, a fault detection branch network for learning fault features, and a layer interpretation branch network for learning layer features; wherein, the shared input features between layers in the backbone network, the fault detection branch network, and the layer interpretation branch network are formed by tensor concatenation.

[0130] In the device embodiments described in this specification, the tensor concatenation of shared input features between layers in the backbone network can be obtained using the following formula:

[0131]

[0132] Among them, X n_input X represents the input data of the nth layer of the backbone network. Concatenate represents tensor concatenation, N represents the number of layers in the backbone network, and X represents the input data of the nth layer of the backbone network. n-1_output This is the output data of the (n-1)th layer network within the backbone network.

[0133] In the device embodiments described in this specification, the tensor concatenation of shared input features among the layers in the tomographic detection branch network can be obtained using the following formula:

[0134]

[0135] Among them, X F,m_intput This represents the input data for the m-th layer of the tomography detection branch network. Concatenate indicates tensor concatenation, N represents the number of layers in the backbone network, M represents the number of layers in the tomography detection branch network, K represents the number of layers in the layer interpretation branch network, and X... n_output The output data of the nth layer network within the backbone network, X F,m-1_output X represents the output data of the (m-1)th layer network within the tomographic detection branch network. L,k-1_output This explains the output data of the (k-1)th layer of the branch network.

[0136] In the device embodiments described in this specification, the tensor concatenation of shared input features between layers within the layered interpretation branch network can be obtained using the following formula:

[0137]

[0138] Among them, X L,k_intput For the input data of the k-th layer of the layer interpretation branch network, Concatenate represents tensor concatenation, N represents the number of layers in the backbone network, M represents the number of layers in the tomography detection branch network, K represents the number of layers in the layer interpretation branch network, and X... n_output The output data of the nth layer network within the backbone network, XL,k-1_output To interpret the output data of the (k-1)th layer network within the branch network, X F,m-1_output This is the output data of the (m-1)th layer network within the fault detection branch network.

[0139] In the device embodiments described in this specification, the loss function may include:

[0140]

[0141] Where loss is the loss function, loss F Let loss be the loss function of the tomography detection branch network. L Let β be the loss function of the layer interpretation branch network, and λ be the weights of the loss functions of the fault detection branch network and the layer interpretation branch network, respectively, used to adjust the training and optimization effect of the network model. P is the number of input seismic data samples. Y is the predicted tomographic data for the i-th sample data. i For the labeled tomographic data of the i-th sample, Z represents the predicted stratigraphic data for the i-th sample data. i Let represent the label layer data of the i-th sample data, α be the positive and negative sample importance balance factor, and γ be the simple and complex sample modulation factor.

[0142] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0143] Embodiments of this specification also provide a computer device. For example... Figure 7As shown, in some embodiments of this specification, the computer device 702 may include one or more processors 704, such as one or more central processing units (CPUs) or graphics processing units (GPUs), each of which may implement one or more hardware threads. The computer device 702 may also include any memory 706 for storing information of any kind, such as code, settings, data, etc. In one specific embodiment, a computer program is stored on the memory 706 and can run on the processor 704. When the computer program is run by the processor 704, it can execute instructions of the seismic fault and horizon interpretation method based on multi-task deep learning described in any of the above embodiments. Non-limitingly, for example, the memory 706 may include any type of RAM, any type of ROM, flash memory, hard disk, optical disk, etc. More generally, any memory can use any technology to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 702. In one scenario, when processor 704 executes associated instructions stored in any memory or combination of memories, computer device 702 can perform any operation of the associated instructions. Computer device 702 also includes one or more drive mechanisms 708 for interacting with any memory, such as hard disk drive mechanisms, optical disk drive mechanisms, etc.

[0144] Computer device 702 may also include an input / output interface 710 (I / O) for receiving various inputs (via input device 712) and providing various outputs (via output device 714). A specific output mechanism may include a presentation device 716 and an associated graphical user interface 718 (GUI). In other embodiments, the input / output interface 710 (I / O), input device 712, and output device 714 may be omitted, and the device may function solely as a computer device within a network. Computer device 702 may also include one or more network interfaces 720 for exchanging data with other devices via one or more communication links 722. One or more communication buses 724 couple the components described above together.

[0145] Communication link 722 can be implemented in any way, such as via a local area network, a wide area network (e.g., the Internet), a point-to-point connection, or any combination thereof. Communication link 722 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc., governed by any protocol or combination of protocols.

[0146] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to some embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processor to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processor, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0147] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processor to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions may also be loaded onto a computer or other programmable data processor, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0149] In a typical configuration, a computer device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0150] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0151] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by computer equipment. As defined in this specification, computer-readable media does not include transient media, such as modulated data signals and carrier waves.

[0152] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of computer program products implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0153] The embodiments described in this specification can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. The embodiments of this specification can also be practiced in distributed computing environments where tasks are performed by remote processors connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0154] It should also be understood that, in the embodiments of this specification, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0155] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, system embodiments are basically similar to method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. In the description of this specification, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments in this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0156] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for seismic fault and horizon interpretation based on multi-task deep learning, characterized in that, The method comprises the following steps: synthesizing a velocity model with spatial combination relations of faults and horizons according to three-dimensional seismic numerical simulation, and performing forward modeling on the velocity model to obtain synthetic seismic data; automatically labeling fault labels and horizon labels of the synthetic seismic data as a training data set; performing inference learning on a multi-task deep learning network model by using the training data set to obtain an initial network model; obtaining fault and horizon label data of a small amount of seismic data to be interpreted; performing transfer learning on the initial network model by using the fault and horizon label data of the small amount of seismic data to be interpreted to obtain a target interpretation model; performing fault and horizon interpretation on seismic data to be interpreted according to the target interpretation model to obtain fault and horizon interpretation results; wherein the multi-task deep learning network model comprises a backbone network for learning the mutual relations of faults and horizons, a fault detection branch network for learning fault features, and a horizon interpretation branch network for learning horizon features; wherein the shared input features between layers in the backbone network, the fault detection branch network and the horizon interpretation branch network are obtained by tensor concatenation; the tensor concatenation of the shared input features between layers in the backbone network is obtained by the following formula: wherein, is input data of the i-th layer network within the backbone network, is output data of the i-th layer network within the backbone network, denotes a tensor concatenation, denotes the number of network layers of the backbone network, is input data of the i-th layer network within the backbone network, is output data of the i-th layer network within the backbone network. the tensor concatenation of the shared input features between layers in the fault detection branch network is obtained by the following formula: wherein, is the input data to the layer of the fault detection branch network, is the output data of the layer of the backbone network, represents the number of network layers of the fault detection branch network, represents the number of network layers of the layer interpretation branch network, is the input data to the layer of the fault detection branch network, is the output data of the layer of the backbone network, is the output data of the layer of the fault detection branch network, is the output data of the layer of the fault detection branch network, is the output data of the layer of the layer interpretation branch network, is the output data of the layer of the layer interpretation branch network. the tensor concatenation of the shared input features between layers in the horizon interpretation branch network is obtained by the following formula: wherein, is the input data for the layer network to interpret the branch within the layer network.

2. The multi-task deep learning based seismic fault and horizon interpretation method of claim 1, wherein, the inference learning on the multi-task deep learning network model by using the training data set comprises: inputting the training data set into the multi-task deep learning network model for inference learning to obtain feature maps of faults and horizons; performing classification processing on the fault and horizon feature maps to obtain fault and horizon category probabilities of each seismic data in the training data set; calculating loss values between the fault and horizon category probabilities and corresponding labels by using a preset loss function; when the loss value is greater than a preset threshold, updating weight parameters of each layer in the multi-task deep learning network model by using the loss value and continuing training until the loss value is not greater than the preset threshold.

3. The multi-task deep learning based seismic fault and horizon interpretation method of claim 2, wherein, the loss function comprises: in For loss function, The loss function of the fault detection branch network is... To explain the loss function of the branch network at each level, and These are the weights of the loss functions for the tomography detection branch network and the layer interpretation branch network, respectively, used to adjust the training and optimization performance of the network model. To input the number of earthquake data samples, For the first Predicted tomographic data for each sample data point For the first Labeled tomographic data of each sample data point, For the first Predicted stratigraphic data for each sample data point For the first Label layer data for each sample data point As a balancing factor for the importance of positive and negative samples, Modulation factors for simple and complex samples.

4. A multi-task deep learning based seismic fault and horizon interpretation apparatus, characterized by, The method comprises the following steps: a data synthesis module is configured to synthesize a velocity model with spatial combination relations of faults and horizons according to three-dimensional seismic numerical simulation, and perform forward modeling on the velocity model to obtain synthetic seismic data; a first label module is configured to automatically label fault labels and horizon labels of the synthetic seismic data as a training data set; a first training module is configured to perform inference learning on a multi-task deep learning network model by using the training data set to obtain an initial network model; a second label module is configured to obtain fault and horizon label data of a small amount of seismic data to be interpreted; a second training module is configured to perform transfer learning on the initial network model by using the fault and horizon label data of the small amount of seismic data to be interpreted to obtain a target interpretation model; and a data interpretation module, configured to interpret the target seismic data according to the target interpretation model to obtain fault and horizon interpretation results; The multi-task deep learning network model comprises: a backbone network configured to learn the mutual relationship between faults and horizons, a fault detection branch network configured to learn fault features, and a horizon interpretation branch network configured to learn horizon features; shared input features between layers in the backbone network, the fault detection branch network, and the horizon interpretation branch network are obtained through tensor concatenation. Tensor concatenation of the shared input features between layers in the backbone network is obtained through the following formula: wherein, is input data of the i-th network layer within the backbone network, is output data of the i-th network layer within the backbone network, denotes a tensor concatenation, denotes the number of network layers of the backbone network, is input data of the i-th network layer within the backbone network, is output data of the i-th network layer within the backbone network. Tensor concatenation of the shared input features between layers in the fault detection branch network is obtained through the following formula: wherein, is the input data to the layer of the fault detection branch network, is the output data of the layer of the fault detection branch network, represents the number of network layers of the fault detection branch network, represents the number of network layers of the layer position interpretation branch network, is the input data to the layer of the backbone network, is the output data of the layer of the backbone network, is the input data to the layer of the fault detection branch network, is the output data of the layer of the fault detection branch network, is the input data to the layer of the layer position interpretation branch network, is the output data of the layer of the layer position interpretation branch network. Tensor concatenation of the shared input features between layers in the horizon interpretation branch network is obtained through the following formula: wherein, is the input data for the layer network to interpret the branch network within the layer network.

5. A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein, The computer program, when executed by the processor, executes instructions of the method according to any one of claims 1-3.

6. A computer storage medium having stored thereon a computer program, characterized in that The computer program, when executed by the processor of the computer device, executes instructions of the method according to any one of claims 1-3.

Images