Method and device for intelligently explaining earthquake high-frequency sequence by using logging information

Abstract

The invention provides an intelligent interpretation method and device for a seismic high-frequency sequence by using logging information, and belongs to the technical field of oil and gas exploration, and the method comprises the steps: inputting seismic attribute data of a determined point location of a seismic channel beside a well with a known condition in a target block determined according to a sequential path into a sequence layering prediction model in a current state, obtaining high-frequency sequence hierarchical data of the determined point location; judging whether the high-frequency sequence hierarchical data of the point location exists in the target block or not, if yes, adding the determined high-frequency sequence hierarchical data of the point location into sequential simulation condition data, constructing a learning sample by using the updated sequential simulation condition data to train a current state sequence hierarchical prediction model, and then skipping to the previous step, and performing high-frequency sequence hierarchical prediction on the next determined point location determined according to the sequential path. The method can be applied to land exploration early-stage blocks, offshore blocks with few wells and the like, and high-frequency sequence interpretation of the blocks with high efficiency, low cost and high accuracy is achieved.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas exploration technology, specifically relating to a seismic high-frequency sequence intelligent interpretation method using well logging information, a seismic high-frequency sequence intelligent interpretation device using well logging information, a computer device, and a machine-readable storage medium. Background Technology

[0002] In the field of oil and gas exploration, seismic structural interpretation is a fundamental and crucial task. Tracing stratigraphic interfaces using seismic profiles is a key step in identifying and assessing oil and gas reservoirs. With the increasing intensity of oil and gas exploration in China, exploration targets have gradually shifted from mature onshore blocks to offshore areas. Deepwater / ultra-deepwater passive continental margin basins have become the main battleground for the discovery of large and medium-sized conventional oil and gas fields globally. As deepwater oil and gas exploration, evaluation, and development deepen, seismic structural interpretation of sparsely populated offshore blocks has significantly increased in terms of interpretation difficulty, workload, accuracy, precision, and timeliness requirements. This is particularly true for high-frequency sequence seismic interpretation in sparsely populated offshore blocks. Due to the thin thickness of high-frequency cyclic strata, how to quickly and accurately interpret the stratigraphic positions of high-frequency cyclic strata using limited drilling, logging, and seismic information is a challenging problem in the field of seismic structural interpretation. Detailed structural characterization and geological modeling based on high-frequency sequence seismic interpretation have become a significant challenge restricting the efficient exploration, evaluation, and development of conventional large and medium-sized oil and gas fields for oil companies.

[0003] For mature exploration and development blocks, drilling data is abundant, and well control is high. High-frequency sequence stratigraphy data interpreted from this rich drilling and logging data can be fully utilized, along with lateral trend information from seismic data from the exploration phase for auxiliary constraints, thus completing the interpretation of high-frequency sequences. However, for early-stage onshore exploration blocks and offshore blocks with few wells, drilling and logging data are very limited, and well control is very low. Therefore, it is necessary to fully extract high-frequency cyclic stratigraphy information from seismic data based on the limited high-frequency cyclic stratigraphy information from drilling and logging data to achieve high-frequency seismic sequence stratigraphy prediction using drilling and logging information.

[0004] A survey of existing technologies for identifying high-frequency seismic sequences reveals that traditional approaches mainly fall into two categories. The first category relies on creating synthetic records, establishing correlations between seismic and well-logging horizons, and then manually interpreting these records to identify and determine stratigraphic boundaries. The second category comprises neural network-based methods for predicting detailed seismic sequences. The first type of scheme has the following main drawbacks: 1) Due to the limitation of seismic data resolution, it is impossible to calibrate high-frequency stratigraphic information at the well logging scale into the seismic data, thus making it impossible to perform high-precision geological stratigraphic division of the target block. Even under the guidance of seismic sedimentology theory and technology, after frequency upscaling and 90° phase transformation of seismic data, high-frequency seismic sequence interpretation of the target block still suffers from drawbacks such as long manual interpretation time, low efficiency, and insufficient accuracy; 2) For offshore blocks with few wells in the early stages of exploration, the seismic data area is usually large, requiring multiple geophysical interpreters to work together to conduct stratigraphic tracing interpretation. However, due to differences in the understanding of regional geology and sedimentary sequence among different interpreters, the seismic interpretation results may be significantly affected by human factors, leading to uncertainty in the accuracy and reliability of the stratigraphic interpretation results. Therefore, interpreters need to have a deep understanding of the seismic response characteristics of regional geology and sedimentary sequence in the study area, which places high demands on the professional level of interpreters. The second type of scheme has the following main defects: When interpreting high-frequency sequence stratigraphy in areas with limited drilling and logging data, such as early-stage onshore exploration blocks and offshore blocks with few wells, existing artificial intelligence models have limited data learning and modeling capabilities for large-area seismic data, fail to fully utilize multi-attribute seismic information, and are unable to fully extract high-frequency sequence information from seismic data. Summary of the Invention

[0005] The purpose of this invention is to provide a seismic high-frequency sequence intelligent interpretation method using well logging information, a seismic high-frequency sequence intelligent interpretation device using well logging information, a computer device, and a machine-readable storage medium, in order to overcome one or more defects in the existing technology for seismic high-frequency sequence interpretation of early exploration blocks and offshore blocks with few wells, which rely on manual interpretation methods and deep learning technology.

[0006] To achieve the above objectives, a first aspect of the present invention provides a method for intelligent interpretation of high-frequency seismic sequence data using well logging information, the method comprising:

[0007] The seismic attribute data with depth information associated with the determined seismic trace locations are input into the current state sequence stratification prediction model to obtain high-frequency sequence stratification data with determined seismic trace locations. The determined seismic trace locations are the well-side seismic trace locations of the wells with known conditions in the target block determined according to the sequential path.

[0008] If it is determined whether there is still high-frequency sequence stratigraphic data for seismic trace points in the target block that has not been obtained, the high-frequency sequence stratigraphic data for the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data is then used to construct learning samples to train the current state sequence stratigraphic prediction model, and the trained sequence stratigraphic prediction model is obtained. Then, the process jumps to the previous step to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace point determined according to the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed.

[0009] Specifically, the initial sequence stratification prediction model is obtained by constructing learning samples using the sequential simulation condition data of the initial state and training the initial sequence prediction neural network. The sequential simulation condition data of the initial state includes well logging high-frequency sequence stratification data and seismic attribute data of well point seismic traces of wells with known conditions in the target block with established correspondence, and seismic attribute data of well point seismic traces of wells with unknown conditions in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence stratification.

[0010] Optionally, the step of inputting the seismic attribute data with associated depth information that determines the seismic trace location into the current state of the sequence stratigraphic prediction model to obtain high-frequency sequence stratigraphic data that determines the seismic trace location includes:

[0011] Based on the sequential path, the location of the seismic traces near the wells of the target block with known conditions is determined for high-frequency sequence layering prediction.

[0012] The stratigraphic information of the determined seismic trace points is determined based on the information on the top and bottom surfaces of the strata obtained from the seismic data of the target block.

[0013] The seismic attribute sequence is composed of the seismic attribute data corresponding to each layer within the determined seismic trace location;

[0014] The seismic attribute sequence is associated with the depth information of each layer within the determined seismic trace location and then input into the current state sequence stratification prediction model to obtain the high-frequency sequence stratification data of the determined seismic trace location.

[0015] Optionally, the high-frequency sequence stratigraphic data of the determined seismic trace locations are added to the sequential simulation condition data to update the sequential simulation condition data, including:

[0016] Establish the correspondence between the seismic attribute data and high-frequency sequence layering data of the determined seismic trace locations;

[0017] The seismic attribute data and high-frequency sequence layering data of the determined seismic trace locations with established correspondences are added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data includes the seismic attribute data and high-frequency sequence layering data with established correspondences, as well as the seismic attribute data of the target block for which no corresponding high-frequency sequence layering data has been obtained.

[0018] Optionally, the hierarchical prediction model of the initial state is obtained by training the initial sequence prediction neural network with learning samples constructed using the sequential simulation conditional data of the initial state, including:

[0019] Encode each stratum in the high-frequency sequence layering data of the well point seismic traces of the well with known conditions in the target block to obtain the stratum coding sequence of each point in the well point seismic trace. Use the stratum coding sequence of each point as the sample label of the seismic attribute data of each point.

[0020] Using the seismic attribute data of each point in the wellpoint seismic trace under known conditions and the sample labels corresponding to each seismic attribute data, an initial sequence prediction neural network is trained, and after training, an initial state sequence hierarchical prediction model is obtained.

[0021] Optionally, when the seismic attribute sequence is associated with the depth information of each layer within the determined seismic trace location and input into the current state of the sequence stratification prediction model, the seismic attribute sequence of the determined seismic trace location is also embedded with the planar location code of the determined seismic trace location.

[0022] Optionally, the method further includes:

[0023] The high-frequency sequence classification results of the target block are obtained by performing sequence classification on the high-frequency sequence data using the established classification model.

[0024] Optionally, the sequence prediction neural network is a sequence prediction neural network with a CRF layer added.

[0025] Optionally, the sequence prediction neural network includes a BiLSTM network and a CRF layer connected to the output of the BiLSTM network.

[0026] Optionally, the classification model is a Softmax classifier.

[0027] Optionally, the method further includes:

[0028] After obtaining the high-frequency sequence stratigraphic data for the determined seismic trace locations, the geological rationality of the high-frequency sequence stratigraphic data for the determined seismic trace locations is judged using the high-frequency sequence stratigraphic data of the known well points. If it is not reasonable, the high-frequency sequence stratigraphic data for the determined seismic trace locations is corrected.

[0029] Optionally, the high-frequency sequence layering data for determining the seismic trace locations can be corrected through manual correction and adjustment of the constraints of the CRF layers.

[0030] Optionally, the seismic trace locations predicted by the sequence stratification prediction model for each state are limited to a circular region centered on a point on the sequential path. If there are overlapping regions between the circular regions, the seismic trace locations in the later predicted circular regions will not include the seismic trace locations in the overlapping regions.

[0031] Optionally, the sequence stratification prediction model for various states can predict high-frequency sequence stratification data for the initial center point on the sequential path.

[0032] A second aspect of this invention provides a seismic high-frequency sequence intelligent interpretation device utilizing well logging information, the device comprising:

[0033] The first module is used to input the seismic attribute data with depth information of the seismic trace points determined by the well-side seismic traces of the target block with known conditions determined according to the sequential path into the current state of the sequence stratification prediction model to obtain high-frequency sequence stratification data that determines the seismic trace points.

[0034] The second module is used to determine whether there are still high-frequency sequence stratigraphic data for seismic trace points in the target block that have not been obtained. If there are still high-frequency sequence stratigraphic data for seismic trace points in the target block that have not been obtained, the high-frequency sequence stratigraphic data for the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data is then used to construct learning samples to train the current state sequence stratigraphic prediction model, so as to obtain the trained sequence stratigraphic prediction model. This enables the first module to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace point determined according to the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed.

[0035] Specifically, the initial sequence stratification prediction model is obtained by constructing learning samples using the sequential simulation condition data of the initial state and training the initial sequence prediction neural network. The sequential simulation condition data of the initial state includes well logging high-frequency sequence stratification data and seismic attribute data of well point seismic traces of wells with known conditions in the target block with established correspondence, and seismic attribute data of well point seismic traces of wells with unknown conditions in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence stratification.

[0036] A third aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the intelligent interpretation method for high-frequency seismic sequence using well logging information as described in the first aspect of the present invention.

[0037] A fourth aspect of the present invention provides a machine-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intelligent interpretation method for high-frequency seismic sequence using well logging information as described in the first aspect of the present invention.

[0038] In the above technical solution, high-frequency sequence stratigraphy data obtained from well logging interpretation at wellpoints of known condition wells within the target block and seismic attributes sensitive to high-frequency sequence stratigraphy extracted from the seismic data of the wellpoint seismic traces are used as initial learning samples to train the sequence prediction neural network, thereby constructing an initial state sequence stratigraphy prediction model. Based on sequential simulation technology, the sequence stratigraphy prediction model is used to predict the high-frequency sequence stratigraphy data of the points near the wellpoints of the known condition wells, and the prediction results of the points of the seismic traces are expanded into the learning samples. Then, following the steps of "learning-prediction-update-learning", the learning samples are sequentially expanded to achieve a gradual expansion from the wellpoint seismic traces of the known condition wells to the seismic traces adjacent to the condition wells, and finally to a traversal prediction of the high-frequency sequence stratigraphy data of all seismic traces in the entire target block.

[0039] Based on this, a technical concept combining sequence prediction neural networks and sequential simulation was adopted to achieve high-frequency sequence spatial prediction of the entire target block using a small amount of well logging information and a large amount of seismic data within the target block. This overcomes the limitations of traditional artificial intelligence models in data learning and modeling capabilities for large-area seismic data, and makes full use of seismic attribute data sensitive to high-frequency sequences. Compared with manual interpretation and deep learning schemes based on traditional artificial intelligence models, it has high efficiency, high accuracy and low cost, and has important application value for risk exploration, fine structural modeling, oil and gas reservoir evaluation and development in blocks with few or sparse wells.

[0040] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0041] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0042] Figure 1A flowchart illustrating a high-frequency seismic sequence intelligent interpretation method utilizing well logging information according to an embodiment of the present invention is shown.

[0043] Figure 2 This schematically illustrates the technical roadmap for intelligent interpretation of seismic high-frequency sequence stratigraphy using well logging information in a specific application example.

[0044] Figure 3 This illustration shows a schematic diagram of the target predicted block shape and the location of wells with known conditions in a specific application example;

[0045] Figure 4 This illustration schematically shows the first type of seismic attribute profile extracted from seismic data in a specific application example;

[0046] Figure 5 This illustration schematically shows a second seismic attribute profile extracted from seismic data in a specific application example.

[0047] Figure 6 This illustration schematically shows a third seismic attribute profile extracted from seismic data in a specific application example.

[0048] Figure 7 This illustration schematically shows a fourth seismic attribute profile extracted from seismic data in a specific application example.

[0049] Figure 8 This diagram illustrates the layer label settings in a specific application example.

[0050] Figure 9 The illustration shows a prediction example of the BiLSTM+CRF model in a specific application instance;

[0051] Figure 10 The illustration shows a comparison of the prediction performance of the sequence land level prediction model without CRF layer and with CRF layer in a specific application example. Part (a) shows the prediction performance of the sequence land level prediction model without CRF layer, and part (b) shows the prediction performance of the sequence land level prediction model with CRF layer.

[0052] Figure 11 The diagram illustrates the sequential simulation process based on the sequential path in a specific application example, where parts (a) to (d) are schematic diagrams of the four stages of the sequential simulation process, respectively.

[0053] Figure 12 This diagram illustrates the working principle of the Softmax classifier in a specific application example.

[0054] Figure 13This illustration shows an example of a seismic profile after frequency upscaling the prediction results of six layered interfaces in a specific application instance.

[0055] Figure 14 The diagram illustrates a three-dimensional structural layer model example of six layered interfaces in a specific application instance.

[0056] Figure 15 This schematic diagram illustrates the components of a seismic high-frequency sequence intelligent interpretation device utilizing well logging information according to an embodiment of the present invention.

[0057] Figure 16 A schematic block diagram of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0058] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0059] Example 1

[0060] The intelligent interpretation method for high-frequency seismic sequence logging using well logging information provided in this invention can be applied to the interpretation of high-frequency seismic sequence logging in early-stage onshore exploration blocks and offshore blocks with few wells. These blocks have large areas of seismic data but limited well logging data. It should be noted that the above description does not constitute a limitation on the application scenarios of this invention. It is understood that the method is also applicable to the interpretation of high-frequency seismic sequence logging in other blocks with large areas of seismic data and limited well logging data. Therefore, the application scenarios of this application are quite broad.

[0061] See Figure 1 This invention provides a method for intelligent interpretation of high-frequency seismic sequence data using well logging information, comprising the following implementation steps:

[0062] Step S100 involves inputting the seismic attribute data with depth information associated with the determined seismic trace locations into the constructed current-state sequence stratigraphy prediction model to obtain high-frequency sequence stratigraphy data for the determined seismic trace locations. These determined seismic trace locations are the well-side seismic trace locations of wells with known conditions in the target block, determined based on the sequential path. Specifically, the initial-state sequence stratigraphy prediction model is obtained by training and optimizing the initial sequence prediction neural network using learning samples constructed from the initial-state sequential simulation condition data. The initial-state sequential simulation condition data includes well logging high-frequency sequence stratigraphy data and seismic attribute data of wellpoint seismic traces from wells with known conditions in the target block with established correspondences, as well as seismic attribute data of wellpoint seismic traces from wells with unknown conditions within the target block. The seismic attribute data is sensitive to high-frequency sequence stratigraphy.

[0063] Based on the above statement, it should be understood that:

[0064] 1) A well with known conditions is a drilled well whose conditions are known.

[0065] 2) Seismic attribute data are typically categorized into various types based on dynamics, picking methods, etc., meaning that the types of seismic attributes that can be extracted from seismic data are very rich. Here, the seismic attribute data used as input data for the sequence stratigraphy prediction model are seismic attribute data sensitive to high-frequency sequence stratigraphy. Sensitive to high-frequency sequence stratigraphy, as the name suggests, refers to seismic attribute types that have a strong correlation with high-frequency sequence stratigraphic changes. Generally, there are multiple types of seismic attribute data sensitive to high-frequency sequence stratigraphy. Multiple seismic attribute data can be identified from these sensitive data based on expert experience, or correlation analysis methods such as PCA (principal component analysis) can be used to analyze these multiple seismic attribute data to derive multiple seismic attribute data with strong correlation to high-frequency sequence stratigraphy, which can then be used as input data for the sequence stratigraphy prediction model.

[0066] 3) Sequential Simulation (SS) is a conditional simulation method. Its simulation idea is to sequentially obtain the simulation values ​​of each grid node along a pre-defined path.

[0067] 4) Establishing a correspondence between the high-frequency sequence stratigraphic data of well logging of well points with known conditions in the target block and the seismic attribute data means that, in order to form the initial learning samples, it is necessary to establish a correspondence between the samples and the sample labels in advance. That is, to associate the seismic attribute data extracted from the seismic data belonging to a certain seismic trace point with the high-frequency sequence stratigraphic data of well logging of that seismic trace point. For example, the association can be established through depth information, and the seismic attribute data of a certain seismic trace point can be used as input, and the high-frequency sequence stratigraphic data of well logging of that point can be used as the training sample label. As is known, for a specific seismic trace location, in order to serve as input data for a sequence prediction neural network, the seismic attribute data of each layer determined by the top and bottom surfaces of the seismic data for that seismic trace location needs to be collectively constructed into a seismic attribute sequence. This seismic attribute sequence is then used as the input sequence for the sequence prediction neural network. To use the high-frequency sequence logging data of that seismic trace location as training sample labels, the sequence data needs to be composed of the information from each layer in the high-frequency sequence logging data of that seismic trace location. For example, each layer in the high-frequency sequence logging data can be assigned an encoding value, and the sequence composed of these encoding values ​​serves as the sample label. For instance, in one specific embodiment, the stratigraphic layer numbers of the target block can be encoded as 1, 2, 3, 4, and 5, for a total of five layers.

[0068] Accordingly, for example, in one or more specific embodiments, seismic attribute data with associated depth information that determines the location of seismic traces is input into the current state of the sequence stratigraphic prediction model to obtain high-frequency sequence stratigraphic data that determines the location of seismic traces. This specifically includes the following implementation steps:

[0069] Based on the sequential path, the location of the seismic traces near the wells of the target block with known conditions is determined for high-frequency sequence layering prediction.

[0070] The stratigraphic information of the seismic trace point is determined based on the information on the top and bottom surfaces of the strata obtained from the seismic data of the target block.

[0071] The seismic attribute sequence is composed of the seismic attribute data corresponding to each layer within the determined seismic trace location.

[0072] The seismic attribute sequence is associated with the depth information of each layer within the determined seismic trace location and then input into the current state sequence stratification prediction model to obtain the high-frequency sequence stratification data of the determined seismic trace location.

[0073] For example, in one or more specific embodiments, a hierarchical prediction model of the initial state is obtained by training the initial sequence prediction neural network with learning samples constructed using the sequential simulation conditional data of the initial state, specifically including the following implementation steps:

[0074] Encode each stratum in the high-frequency sequence layering data of the well point seismic traces of the well point of the known condition well in the target block, and obtain the stratum coding sequence of each point. Use the stratum coding sequence of each point as the sample label of the seismic attribute data of each point.

[0075] Using the seismic attribute data of each point in the wellpoint seismic trace under known conditions and the sample labels corresponding to each seismic attribute data, an initial sequence prediction neural network is trained, and after training, an initial state sequence hierarchical prediction model is obtained.

[0076] For example, in one or more specific embodiments, when the seismic attribute sequence is associated with the depth information of each layer corresponding to the determined seismic trace point and input into the current state of the sequence stratification prediction model, the seismic attribute sequence of the determined seismic trace point is also embedded with the planar position code of the seismic trace point.

[0077] It is important to understand that, for a given seismic trace location, the seismic attribute data used as input to the initial sequence prediction neural network is a sequence of seismic attribute data composed of the seismic attributes of each layer of that trace location. By embedding the planar location code of the seismic trace location into this sequence of seismic attribute data, the sensitivity of the sequence prediction neural network to the position of each element within the sequence is enhanced, thereby improving the network's robustness. For example, the planar location code can be the (XLINE, INLINE) coordinates of the seismic trace location on the plane, such as (633, 784). Alternatively, a new coordinate system can be established based on the requirements of the target block, for example, when the target block has high requirements for high-frequency sequence accuracy, and the current coordinate system cannot meet these requirements.

[0078] Step S200: Determine whether there are still high-frequency sequence stratigraphic data for seismic trace locations within the target block that have not been obtained. If so, add the high-frequency sequence stratigraphic data for the determined seismic trace locations obtained in step S100 to the sequential simulation condition data to update the sequential simulation condition data. Then, use the updated sequential simulation condition data to construct learning samples to train the current state sequence stratigraphic prediction model, obtain the trained sequence stratigraphic prediction model, and then jump to step S100 to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace location determined according to the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed.

[0079] For example, in one or more specific embodiments, the high-frequency sequence stratigraphic data of the seismic trace locations obtained in step S100 is added to the sequential simulation condition data to update the sequential simulation condition data, specifically including the following implementation steps:

[0080] Establish the correspondence between the high-frequency sequence layering data and seismic attribute data of the seismic trace points obtained in step S100;

[0081] The seismic attribute data and high-frequency sequence layering data of the identified node with established correspondence are added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data includes the seismic attribute data and high-frequency sequence layering data with established correspondence, as well as the seismic attribute data of the target block for which no corresponding high-frequency sequence layering data has been obtained.

[0082] For example, in one or more specific embodiments, the sequence prediction neural network is a sequence prediction neural network with a CRF (Conditional Random Field) layer added. It is understood that a sequence prediction neural network with a CRF layer means that after feature extraction through the sequence prediction neural network, the output is passed to the CRF layer for global optimization and annotation, thereby establishing an end-to-end sequence stratigraphic prediction model. The CRF layer is an undirected graphical network that can label or classify sequence data. The CRF layer can consider the relationships between adjacent sample points, using these relationships as constraints for sequence stratigraphic prediction, thereby improving the accuracy of sequence prediction, avoiding unreasonable predictions, preventing sequence prediction disorder, and making the predicted sequence stratigraphic levels more consistent with geological laws. It is understood that the constraints of the CRF layer are generally implemented by defining feature functions. These feature functions impose restrictions on the output sequence obtained after feature extraction by the sequence prediction neural network; each restriction corresponds to a feature function. For example, if a stratigraphic sequence has five sub-layers, a feature function is defined to ensure that the output sequence predicted by the sequence prediction neural network contains all five layers. This feature function checks whether the predicted output sequence contains the five labels 1, 2, 3, 4, and 5. If not, the model score is penalized. Another example is that stratigraphic sequences cannot exhibit layer crossing or skipping. Another feature function is defined to ensure that the predicted output sequence follows a sequentially increasing pattern. For instance, a feature function is defined to check whether the layer number encoding of adjacent layers conforms to this pattern. For example, there cannot be unreasonable transitions from layer 1 to layer 3, or from layer 3 to layer 2; only sequential transitions from layer 1 to layer 2, and from layer 2 to layer 3 are allowed. When training the CRF layer, the aforementioned feature functions are added to the stratigraphic sequence prediction model, and the weights of these feature functions are adjusted during training. It is important to know that the performance of the CRF layer is affected by feature selection. Therefore, it is necessary to design appropriate feature functions based on the specific target block prediction requirements and the characteristics of the sequential simulation condition data. This embodiment does not provide a detailed description of the specific feature functions involved.

[0083] In a specific application, for example, after adding feature functions to the sequence prediction model, the probability of transferring from layer 2 to layer 1 is set to 0 after training, and the probability of transferring from layer 1 to layer 3 is set to 0, etc. Each layer has a transfer score, which is used to represent the probability of transferring from the current layer to another layer. The transfer score can help the prediction model based on sequence prediction neural network to better understand the relationship between layers, thereby improving the performance of the prediction model based on sequence prediction neural network in tasks such as earthquake sequence annotation.

[0084] In one comparative example, the sequence prediction neural network is a sequence prediction neural network without added CRF layers, such as... Figure 10As shown, the sequence strata prediction model without the CRF layer exhibited layer skipping in its prediction results, while the sequence strata prediction model with the CRF layer did not exhibit layer skipping or layer skipping in its prediction results.

[0085] Therefore, in the above embodiments, by introducing a CRF layer into the sequence prediction neural network, the prediction accuracy of high-frequency sequence horizons is improved, thereby improving the accuracy of interpreting high-frequency seismic sequences using a small amount of well logging information.

[0086] For example, in one or more specific embodiments, the sequence prediction neural network includes a BiLSTM network (Bidirectional Long Short-Term Memory network) and a CRF layer connected in sequence. The BiLSTM network is composed of a forward LSTM and a backward LSTM. The forward LSTM processes the seismic attribute sequence in the forward direction, and the backward LSTM processes the seismic attribute sequence in the reverse direction. After processing, the outputs of the two LSTMs are concatenated, thereby making full use of the contextual information of the seismic attribute sequence, effectively extracting seismic data features, and better finding and utilizing the relationship between the stratigraphic coding sequence and the changes in the seismic attribute sequence.

[0087] As can be seen, although the above embodiments mention using a BiLSTM network as the preferred core network for the sequence prediction model, this is not intended to limit the sequence prediction model. In application, a suitable sequence prediction neural network should be selected based on specific scenario conditions. For example, considering the computing power of the device, other sequence prediction neural networks with lower requirements for device computing power can be selected, or known and publicly available sequence prediction neural networks can be modified to achieve the same or better prediction accuracy as the BiLSTM network. The embodiments of this invention cannot exhaustively describe all the sequence prediction neural networks applicable to this invention, but at least those skilled in the art will understand that the applicability of commonly used sequence prediction neural networks, such as RNNs and LSTMs, cannot be excluded.

[0088] As in the above embodiment, a sequence prediction neural network is formed by combining a BiLSTM network and a CRF layer. The high-frequency sequence data prediction using this network has high accuracy and meets the needs of seismic high-frequency sequence interpretation for early exploration blocks on land and low-well blocks at sea.

[0089] For example, in one or more specific embodiments, after predicting and obtaining high-frequency sequence stratigraphic data for determining seismic trace locations, the high-frequency sequence stratigraphic data for determining seismic trace locations is geologically reasonable using high-frequency sequence stratigraphic data for well points with known conditions. If it is unreasonable, the high-frequency sequence stratigraphic data for determining seismic trace locations is corrected.

[0090] As in the above embodiment, by comparing with the high-frequency sequence stratigraphy data of known well points, it is determined whether the predicted high-frequency sequence stratigraphy data is reasonable. For example, the geological rationality judgment includes the rationality judgment of the stratigraphic sequence and the rationality judgment of the stratigraphic thickness. By comparing the high-frequency sequence stratigraphy data of the seismic trace points near the predicted well with the high-frequency sequence stratigraphy data of the seismic trace at the well point, if there is a large difference in the stratigraphic ratio or the sequence prediction is unreasonable, it can be determined that the predicted high-frequency sequence stratigraphy data is unreasonable and the high-frequency sequence stratigraphy data of the determined seismic trace point needs to be corrected.

[0091] For example, in one or more specific embodiments, the correction may be performed by: manual assistance; or by adjusting the constraints of the CRF layer.

[0092] For example, when judging the rationality of stratigraphic sequence, it is necessary to determine whether there are instances of crossing or skipping layers, and whether it starts from the specified first sub-layer and ends at the specified fifth sub-layer. If any of these situations exist, the unreasonable high-frequency sequence stratification prediction results need to be manually corrected. For example, the sequence codes corresponding to the cross-layer or skipped-layer seismic trace points may be corrected to provide reliable learning samples for the next round of model learning. As another example, when judging the rationality of stratigraphic thickness, if the thickness difference of the same sub-layer between adjacent seismic traces is too large, it is necessary to consider the geological characteristics of the target block at that location, such as whether there are fault lines. If there are no fault lines, manual correction is required to bring it within a reasonable range.

[0093] As in the above embodiment, by performing geological rationality judgment and correction on the predicted high-frequency sequence stratigraphic data, the prediction accuracy of the sequence stratigraphic prediction model for seismic high-frequency sequence stratigraphic data is improved, thereby improving the accuracy of seismic high-frequency sequence interpretation based on a small amount of well logging information.

[0094] For example, in one or more specific embodiments, the seismic trace locations predicted by the sequence stratification prediction model for each state are limited to a circular region centered on a point on the sequential path. If there are overlapping areas between the circular regions, the seismic trace locations in the subsequently predicted circular regions will not include those in the overlapping areas. Preferably, the circular regions have the same radius.

[0095] As an improvement to the above embodiment, the following specific embodiment introduces an information source, which is the high-frequency sequence stratigraphic data and seismic attribute data of the wellpoint seismic traces of known wells. Regardless of how the center of the prediction range (circular area) of the sequence stratigraphic prediction model changes, this information source is always used to construct the learning samples; that is, the information source always serves as the sample point in the learning samples. As in the above embodiment, because the information source represents the most reliable high-frequency sequence stratigraphic information, by always using the most reliable high-frequency sequence stratigraphic information as the sample point in the learning samples, the prediction accuracy of the sequence stratigraphic prediction model is improved.

[0096] Example 2

[0097] This invention provides a method for intelligent interpretation of high-frequency seismic sequence data using well logging information. The difference between this method and Embodiment 1 is that, after step S200, the method further includes the following steps:

[0098] Step S300: Use the constructed classification model to perform stratified classification on the high-frequency stratified data to obtain the high-frequency stratified partitioning result of the target block.

[0099] For example, if a high-frequency sequence partitioning rule has been defined for the known well location and the high-frequency sequence partitioning of the seismic trace at that well point has been performed using this rule, then the high-frequency sequence partitioning rule can be used to perform high-frequency sequence partitioning on other high-frequency sequence stratified data of the target block. In this case, the sequence type defined by the high-frequency sequence partitioning rule can be used as the sequence classification type of the classification model.

[0100] In a comparative example, after obtaining the high-frequency stratified hierarchical data of each point in the target block, this high-frequency stratified hierarchical data can be manually interpreted to obtain the high-frequency stratified partitioning result of the target block. However, the efficiency and accuracy of manual high-frequency stratified partitioning are significantly lower than those based on classification models. Therefore, using the established classification model to perform stratified classification on high-frequency stratified hierarchical data achieves both high efficiency and high accuracy.

[0101] For example, in one or more specific embodiments, the classification model is a Softmax classifier. The Softmax classifier is a typical classifier capable of performing multi-class classification tasks. It should be understood that while the above description mentions using the Softmax classifier as a classification model, this is not a limitation. In application, a suitable classification model should be selected based on the specific scenario conditions. This invention cannot exhaustively describe all classification models applicable to this invention, but at least those skilled in the art will recognize that commonly used classification models, such as SVM and Bayesian classifiers, are applicable in this invention.

[0102] Example 3

[0103] This embodiment of the invention is an application example of the above-described embodiment two. Combined with... Figure 2 The technical roadmap shown in this invention illustrates a seismic high-frequency sequence intelligent interpretation method utilizing well logging information, which includes the following implementation steps:

[0104] Step SS1 involves establishing a condition information database for storing sequential simulation condition data using high-frequency sequence layering data of wellpoint seismic traces of known condition wells within the target block obtained through well logging data, the planar location information of the target block, and the planar location information of known condition wells. The planar locations of the known condition wells are then encoded and stored in the condition information database. Additionally, the top and bottom surface information of the target block obtained from seismic data is read into the condition information database.

[0105] Step SS2 involves extracting seismic attribute data from the SEGY format seismic data volume, optimizing the depth domain seismic attribute data based on the differences in sensitivity to high-frequency layers, then extracting the seismic attribute data of the well-side seismic traces with known conditions, and storing these seismic attribute data in the condition information database.

[0106] Step SS3 involves classifying and extracting data from the condition information database, and transferring the high-frequency sequence stratigraphic data from the wellpoint seismic traces of known condition wells to the seismic attribute data at the same depth. This completes the correspondence and encoding between the seismic attribute data and the high-frequency sequence stratigraphic data of the known condition wells. Sample labels are then set based on the correspondence between the high-frequency sequence stratigraphic data and the seismic attribute data, and the sample data is filtered and normalized mathematically. Specifically, because the specific values ​​of the seismic attribute data are relatively discrete when used as sample data, normalization is performed on the seismic attribute data to improve the training efficiency of the subsequent sequence prediction neural network.

[0107] Step SS4 involves learning the features of the high-frequency sequence hierarchical data and planar position encoding information of the wells under known conditions, and establishing a BiLSTM+CRF neural network model (referring to the sequence prediction neural network composed of a BiLSTM network and a CRF layer connected to the output of the BiLSTM network in the above embodiment). For the high-frequency sequence hierarchical problem, the BiLSTM+CRF neural network model is trained to ensure a high self-test accuracy on the learning samples, thus obtaining the initial state of the sequence hierarchical prediction model. Then, the sequence hierarchical prediction model containing the weight value information between each network node after training is saved.

[0108] Step SS5: Obtain the seismic attribute data of the points (referred to as the sample points to be predicted) near the well based on the data in the condition information database. Use the sequence stratigraphy prediction model after learning and training to predict the high-frequency sequence stratigraphy of the sample points. Combine the high-frequency sequence stratigraphy information of the known well points to evaluate and correct the prediction results to meet the geological rationality, that is, to meet the geological rationality standard of the high-frequency sequence stratigraphy of the well points near the known wells. Thus, the high-frequency sequence stratigraphy prediction of the seismic attributes of the location near the well with limited conditions is completed.

[0109] Step SS6 records the high-frequency sequence stratigraphic data and location information of the wells near the known conditions obtained from the prediction, inserts them into the condition information database, and updates the condition content and structure in the condition information database. The expanded condition information database is used to update the learning samples, so as to achieve the purpose of synchronous updating and expansion of the learning samples. After updating the learning samples, the sequence stratigraphic prediction model predicts other sample points of the seismic traces near the wells with known conditions according to the sequential path, and corrects the prediction results. Through the operation of "learning-prediction-correction-update-learning", the condition information database is updated cyclically and the learning samples are sequentially expanded.

[0110] The sample point range predicted by the sequence floor prediction model for each state is a circular region centered on points along the sequential path. The sequential prediction principle described in step SS6 above is as follows:

[0111] Assuming the initial sample point coordinates are represented as (x0, y0), and there is corresponding seismic trace stratigraphic information S and depth information D, then according to the predefined circular region radius R, the range of sample points predicted by the initial state sequence stratigraphic prediction model can be obtained, which is called the initial circular region.

[0112] Assume the functional expression of the sequence stratigraphic prediction model is f θ (K,S,D), where θ represents the parameters of the neural network and K represents the seismic attribute data of the initial sample points, then the prediction result of the initial circular region can be expressed as: S pred (x,y)=f θ (k,S,D)·I(x,y), where I(x,y) is an indicator function used to indicate whether the point (x,y) is within the initial circular region, and

[0113]

[0114] In order to predict the new circular region within the target block, the center (x0, y0) is translated along the sequential path. Assuming the translation vector is represented as (Δx, Δy), the coordinates of the center of the new circular region are represented as (x0+Δx, y0+Δy).

[0115] For sample points within the new circular region that have not been predicted before, high-frequency sequence stratification prediction is performed. The prediction result for the new circular region can be expressed as: S pred (x′,y′)=f θ (K,S,D)·I(x′,y′), where I(x′,y′) represents whether the point (x′,y′) is within the new circular region, and

[0116]

[0117] Furthermore, it's important to understand that in practical applications, the input to sequence stratigraphy prediction models may include additional feature parameters. For example, when considering the influence of lithofacies on high-frequency sequences, lithofacies can be used as input to the sequence stratigraphy prediction model. Moreover, practical applications may require consideration of boundary conditions and prediction update strategies. Boundary conditions refer to the boundaries and overall shape of the target block, which can affect the rules of sequential prediction. For instance, if the boundaries are highly irregular, operations such as reducing the circular area containing the learning and prediction sample points may be necessary. Additionally, to obtain the next round of learning samples, the strategy for updating the current learning samples after updating the condition information base using the current prediction results should be adjusted as the shape of the target block changes. Simultaneously, the selection of learning sample size, prediction step size, and circular area radius needs to be determined based on the size of the target block and computing power, aiming to reduce the learning sample size and increase the prediction step size while maximizing prediction accuracy, thereby reducing computational resource consumption.

[0118] Step SS7 involves analyzing the known conditions and high-frequency sequence data of the target block to determine the high-frequency sequence division rules, thereby obtaining the category type of the high-frequency sequence. Based on the category type of the high-frequency sequence, a classification model for classifying the high-frequency sequence hierarchical information is designed. Then, the classification model is used to perform hierarchical classification on the high-frequency sequence hierarchical information of each point in the target block to obtain the high-frequency sequence division result of the target block.

[0119] The classification model uses the Softmax classifier. Furthermore, the construction principle of the Softmax classifier can be explained using the process described in the standard implementation, and will not be elaborated further here.

[0120] Example 4

[0121] This invention embodiment is a specific application of Embodiment 3 above in a low-well block in a certain overseas sea area (hereinafter referred to as the target prediction block). Combined with... Figures 3 to 14 As shown, the seismic high-frequency sequence stratigraphic interpretation method using well logging information in this application includes the following specific steps:

[0122] Step A1: By analyzing the target prediction block data, obtain the planar location data of each seismic trace in the target prediction block, establish a virtual coordinate axis in the computer, reproduce the shape of the target prediction block, and simultaneously mark and encode the coordinates of two known condition wells (condition well A and condition well B) in the target prediction block. The shape of the target prediction block and the location of the condition wells are as follows: Figure 3 As shown. High-frequency sequence stratigraphic data from the two condition wells are saved. A condition information database is established and the above data is stored therein. Simultaneously, the top and bottom surface data of the target prediction block in the vertical direction are stored in the condition information database for later use. Reading information from different seismic attribute data volumes (SEGY format files) includes: extracting seismic trace data and reading its (X,Y,Z) coordinates; reading the required seismic attribute data volumes; and converting the seismic attribute coordinates based on the coordinates read from the seismic traces, transforming the (X,Y) series seismic attributes into the (XLINE, INLINE) series in the seismic volume.

[0123] Step A2 involves using correlation analysis to optimize the selection of over ten candidate seismic attributes, and combining this with in-ground seismic profile maps to select four seismic attributes as the basis for high-frequency sequence stratigraphy prediction. The selected four seismic attribute maps are shown below. Figures 4 to 7 As shown, the four seismic attributes are: Quadrature Amplitude (Quadra), Trace AGC (Trace AGC), Relative Acoustic Impedance (RAI), and Root Mean Square Amplitude (RMS). Simultaneously, the relationship between ground elevation and well logging core depth in the seismic attribute data is balanced. The processed seismic attribute data is stored in a conditional information database to provide data support for the learning samples of the neural network model. At this point, the conditional information database contains the planar location data, seismic attribute data, and the top and bottom surface depth and location data of all seismic traces in the target prediction block for each seismic trace sample. A correspondence is established between the high-frequency sequence stratigraphic data of the conditional well point seismic traces and the extracted seismic attribute data. Different strata are numbered, and based on the existing high-frequency sequence stratigraphic information and the depth information of the sample points, the stratigraphic level corresponding to each sample point is encoded to obtain a stratigraphic encoding sequence. This stratigraphic encoding sequence serves as the sample label for the corresponding sample point. The five strata of the target prediction block are coded as 1, 2, 3, 4, and 5, and the stratum labels are set as follows: Figure 8 As shown. Figure 8In the model, the sub-layers of the target prediction block are coded as 1, 2, 3, 4, and 5, respectively, and attributes A to D represent four seismic attributes. Because the extracted seismic attribute values ​​are highly discrete on the horizontal axis, the seismic attribute values ​​are normalized to enable the sequence stratigraphy prediction model to better extract the intrinsic relationship between seismic attributes and high-frequency sequence stratigraphy data.

[0124] Step A3: Based on the characteristics of high-frequency sequence stratigraphic data and seismic attribute data from wells with known conditions, an initial sequence prediction neural network based on BiLSTM+CRF is established. The initial sequence stratigraphic prediction model is trained using training samples to obtain the initial state sequence stratigraphic prediction model. An example of this model's operation can be found in [link to example]. Figure 9 As shown. Save the stratigraphic prediction model to the root directory of the program file in h5 format. It can be directly called when the model is used to predict new data next time.

[0125] Step A4: Based on the top and bottom surface data in the condition information database, search for and obtain the top and bottom surface data of seismic traces near known condition wells. Extract seismic attribute data between the top and bottom surfaces of the target seismic trace in different seismic attribute data volumes based on the top and bottom surface information and location data of the seismic trace in the top and bottom surface data. Use the seismic attribute data and depth information of each point of the seismic trace as input data for the sequence stratigraphy prediction model. Utilize the previously saved sequence stratigraphy prediction model to perform high-frequency sequence stratigraphy prediction for the points of the seismic traces near the condition well. Compare and analyze the prediction results with the high-frequency sequence stratigraphy data of the wellpoint seismic traces of known condition wells. Evaluate the ideality of the prediction results based on the high-frequency sequence stratigraphy data of the wellpoint seismic traces of known condition wells. If, through comparison, a significant difference in the high-frequency sequence stratigraphy ratio or unreasonable sequence prediction is found between the predicted results of the seismic traces near the known condition wells and the high-frequency sequence stratigraphy data at the wellpoint, manual correction or adjustment of the CRF layer constraints is required to achieve a standard that is basically consistent with the high-frequency sequence stratigraphy information of the seismic traces near the known condition wells. The high-frequency sequence stratigraphy prediction results of seismic attributes near the corrected well, along with their corresponding location information, are then incorporated into the conditional information database according to the required format. The newly incorporated information is then used to update the learning samples of the sequence stratigraphy prediction model to establish more correspondences between seismic attributes and high-frequency sequence stratigraphy information. This aims to better predict the high-frequency sequence stratigraphy status of the next nearby seismic trace, until all high-frequency sequence stratigraphy prediction results and their corresponding location information for all selected nearby seismic traces within the initially defined prediction range are incorporated into the conditional information database. It is important to emphasize that the capacity of the conditional information database depends on the radius of the defined prediction range (circular area radius), the size of the target prediction block, and the computing power. The selection of the sample points to be predicted depends on the size of the target prediction block, the density of sample seismic traces required to reach the structural plane, and the computing power. Subsequent conditions are gradually added, executed sequentially according to the above procedure. A range of sample points to be predicted is defined in a certain area or direction. Following predetermined rules, the range of sample points to be predicted is shifted across the target prediction block. This process of repeatedly performing high-frequency sequence stratigraphic prediction of seismic attributes at the location of wells near known conditions is repeated until the prediction operation traverses the entire target prediction block. At this point, all selected seismic traces within the entire target prediction block have undergone the prediction process, ultimately achieving conditional cyclical updating of high-frequency seismic sequence stratigraphic information and progressive sequential expansion of the learning samples. Furthermore, during sequential simulation, an information source is defined. The information source serves as a permanently fixed sample point in the learning samples, regardless of the location of the center of the prediction range radius. The prediction process is shown in the attached figure. Figure 11 As shown. Figure 11In the above, the learning sample queue container contains the sample points of the current learning sample, the elimination sample queue container contains the eliminated sample points, and the sample points in the overlapping area of ​​the learning sample queue container and the elimination sample queue container are included in the sample points of the current learning sample.

[0126] Step A5: Based on the high-frequency sequence partitioning structure of the wells with known conditions of the target prediction block, a Softmax classification network is selected to construct the classification model, resulting in a Softmax classifier suitable for this target prediction block. The working principle of the Softmax classifier is as follows: Figure 12 As shown.

[0127] Through the above steps, intelligent interpretation of seismic high-frequency sequence data for a target prediction block with few wells was achieved using a limited amount of high-frequency sequence data from well logging. For example... Figure 13 Seismic profiles were identified by tracking five short-term sequence stratigraphic cycles within the target prediction block. Figure 14 This is a plan view of six stratigraphic interfaces from top to bottom. Furthermore, the high-frequency sequence intelligent interpretation of the target prediction block achieved through the above steps only took tens of hours, while traditional manual seismic horizon tracing methods for nearly 500 km of three-dimensional stratigraphy took significantly less. 2 The interpretation of high-frequency sequence data for the target prediction block of the six vertical sub-layer interfaces takes at least several months. Therefore, the intelligent interpretation method for seismic high-frequency sequence data using well logging information implemented in this embodiment of the invention greatly improves interpretation efficiency, shortens the project research cycle, reduces interpretation costs, and effectively saves human resources.

[0128] Example 5

[0129] See Figure 15 This invention provides a seismic high-frequency sequence stratigraphy intelligent interpretation device 400 utilizing well logging information, comprising a first module 410 and a second module 420, wherein:

[0130] The first module 410 is used to input the seismic attribute data with depth information of the seismic trace points determined by the well-side seismic traces of the target block known conditions wells determined according to the sequential path into the current state of the sequence stratification prediction model to obtain high-frequency sequence stratification data that determines the seismic trace points.

[0131] The second module 420 is used to determine whether there are still high-frequency sequence stratification data of seismic trace points in the target block that have not been obtained. If there are still high-frequency sequence stratification data of seismic trace points in the target block that have not been obtained, the high-frequency sequence stratification data of the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data is then used to construct learning samples to train the current state sequence stratification prediction model, so as to obtain the trained sequence stratification prediction model. This enables the first module to perform high-frequency sequence stratification prediction for the next determined seismic trace point determined according to the sequential path. Otherwise, the high-frequency sequence stratification prediction for the target block is completed.

[0132] Specifically, the initial sequence stratification prediction model is obtained by constructing learning samples using the sequential simulation condition data of the initial state and training the initial sequence prediction neural network. The sequential simulation condition data of the initial state includes well logging high-frequency sequence stratification data and seismic attribute data of well point seismic traces of wells with known conditions in the target block with established correspondence, and seismic attribute data of well point seismic traces of wells with unknown conditions in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence stratification.

[0133] In one specific embodiment, seismic attribute data with associated depth information, which determines the location of seismic traces, is input into the current state of the sequence stratigraphic prediction model to obtain high-frequency sequence stratigraphic data that determines the location of seismic traces, including:

[0134] Based on the sequential path, the location of the seismic traces near the wells of the target block with known conditions is determined for high-frequency sequence layering prediction.

[0135] The stratigraphic information of the determined seismic trace points is determined based on the information on the top and bottom surfaces of the strata obtained from the seismic data of the target block.

[0136] The seismic attribute sequence is composed of the seismic attribute data corresponding to each layer within the determined seismic trace location;

[0137] The seismic attribute sequence is associated with the depth information of each layer within the determined seismic trace location and then input into the current state sequence stratification prediction model to obtain the high-frequency sequence stratification data of the determined seismic trace location.

[0138] In one specific embodiment, high-frequency sequence stratigraphic data for determining seismic trace locations are added to the sequential simulation condition data to update the sequential simulation condition data, including:

[0139] Establish the correspondence between the seismic attribute data and high-frequency sequence layering data of the determined seismic trace locations;

[0140] The seismic attribute data and high-frequency sequence layering data of the determined seismic trace locations with established correspondences are added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data includes the seismic attribute data and high-frequency sequence layering data with established correspondences, as well as the seismic attribute data of the target block for which no corresponding high-frequency sequence layering data has been obtained.

[0141] In one specific embodiment, the hierarchical prediction model of the initial state is obtained by training the initial sequence prediction neural network with learning samples constructed using the sequential simulation conditional data of the initial state, including:

[0142] Encode each stratum in the high-frequency sequence layering data of the well point seismic traces of the well with known conditions in the target block to obtain the stratum coding sequence of each point in the well point seismic trace. Use the stratum coding sequence of each point as the sample label of the seismic attribute data of each point.

[0143] Using the seismic attribute data of each point in the wellpoint seismic trace under known conditions and the sample labels corresponding to each seismic attribute data, an initial sequence prediction neural network is trained, and after training, an initial state sequence hierarchical prediction model is obtained.

[0144] In one specific embodiment, when training an initial sequence prediction neural network using the seismic attribute data of each point in a known wellpoint seismic trace and the sample labels corresponding to each seismic attribute data, the seismic attribute data of each point is embedded with the planar position code of that point.

[0145] In one specific embodiment, the seismic high-frequency sequence intelligent interpretation device utilizing well logging information further includes a third module, which is used to perform sequence classification on the high-frequency sequence layered data using the constructed classification model to obtain the high-frequency sequence division result of the target block.

[0146] In one specific embodiment, the sequence prediction neural network is a sequence prediction neural network with a CRF layer added.

[0147] In one specific embodiment, the sequence prediction neural network includes a BiLSTM network and a CRF layer connected to the output of the BiLSTM network.

[0148] In one specific embodiment, the classification model is the Softmax classifier.

[0149] In one specific embodiment, the seismic high-frequency sequence intelligent interpretation device utilizing well logging information further includes a fourth module. The fourth module is used to perform a geological rationality judgment on the high-frequency sequence stratification data of the determined seismic trace location after predicting and obtaining the high-frequency sequence stratification data of the determined seismic trace location using the high-frequency sequence stratification data of the known well points. If it is not reasonable, the high-frequency sequence stratification data of the determined seismic trace location is corrected.

[0150] In one specific embodiment, the high-frequency sequence layering data for determining seismic trace locations is corrected through manual correction and adjustment of the constraints of the CRF layer.

[0151] In one specific embodiment, the points predicted by the hierarchical prediction model for each state are limited to a circular region centered on the points on the sequential path. If there is an overlapping region between the circular regions, the points in the later predicted circular region will not include the points in the overlapping region.

[0152] In one specific embodiment, the sequence stratification prediction model for various states performs high-frequency sequence stratification data prediction for the initial center position on the sequential path.

[0153] On the other hand, embodiments of the present invention also provide a machine-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described intelligent interpretation method for high-frequency seismic sequence using well logging information.

[0154] In another aspect, embodiments of the present invention also provide a computer device, which may be a terminal, and its internal structure diagram may be as follows: Figure 16 As shown in the figure, the computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05, and a memory (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A06. The network interface A02 is used for communication with external terminals via a network connection. When the computer program is executed by the processor A01, it implements the aforementioned intelligent interpretation method of seismic high-frequency sequence using well logging information. The display screen A04 can be a liquid crystal display or an e-ink display. The input device A05 can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0155] In one embodiment, the intelligent seismic high-frequency sequence interpretation device 400 utilizing well logging information provided by the present invention can be implemented as a computer program, which can be implemented in the form of, for example... Figure 15 The computer device shown operates on this device. The computer device's memory can store various program modules that make up the intelligent interpretation device 400 based on well logging information. The computer program, composed of these program modules, causes the processor to execute the steps of the intelligent interpretation method for seismic high-frequency sequence using well logging information described in this specification.

[0156] This invention also provides a computer program product that, when executed on a data processing device, is suitable for executing a program that initializes the following method steps:

[0157] The seismic attribute data with depth information associated with the determined seismic trace locations are input into the current state sequence stratification prediction model to obtain high-frequency sequence stratification data with determined seismic trace locations. The determined seismic trace locations are the well-side seismic trace locations of the wells with known conditions in the target block determined according to the sequential path.

[0158] If it is determined whether there is still high-frequency sequence stratigraphic data for seismic trace points in the target block that has not been obtained, the high-frequency sequence stratigraphic data for the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data is then used to construct learning samples to train the current state sequence stratigraphic prediction model, and the trained sequence stratigraphic prediction model is obtained. Then, the process jumps to the previous step to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace point determined according to the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed.

[0159] Specifically, the initial sequence stratification prediction model is obtained by constructing learning samples using the sequential simulation condition data of the initial state and training the initial sequence prediction neural network. The sequential simulation condition data of the initial state includes well logging high-frequency sequence stratification data and seismic attribute data of well point seismic traces of wells with known conditions in the target block with established correspondence, and seismic attribute data of well point seismic traces of wells with unknown conditions in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence stratification.

[0160] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0161] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0162] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent interpretation of high-frequency seismic sequence data using well logging information, characterized in that, The method includes: The seismic attribute data with depth information associated with the determined seismic trace locations are input into the current state sequence stratification prediction model to obtain high-frequency sequence stratification data with determined seismic trace locations. The determined seismic trace locations are the well-side seismic trace locations of the wells with known conditions in the target block determined according to the sequential path. If it is determined whether there is still high-frequency sequence stratigraphic data for seismic trace points in the target block that has not been obtained, the high-frequency sequence stratigraphic data for the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data is then used to construct learning samples to train the current state sequence stratigraphic prediction model, and the trained sequence stratigraphic prediction model is obtained. Then, the process jumps to the previous step to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace point determined according to the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed. Specifically, the initial sequence stratification prediction model is obtained by constructing learning samples using the sequential simulation condition data of the initial state and training the initial sequence prediction neural network. The sequential simulation condition data of the initial state includes well logging high-frequency sequence stratification data and seismic attribute data of well point seismic traces of wells with known conditions in the target block with established correspondence, and seismic attribute data of well point seismic traces of wells with unknown conditions in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence stratification.

2. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1, characterized in that, The process involves inputting seismic attribute data with associated depth information, which determines the location of seismic traces, into the current state of the sequence stratigraphic prediction model to obtain high-frequency sequence stratigraphic data that determines the location of seismic traces, including: Based on the sequential path, the location of the seismic traces near the wells of the target block with known conditions is determined for high-frequency sequence layering prediction. The stratigraphic information of the determined seismic trace points is determined based on the information on the top and bottom surfaces of the strata obtained from the seismic data of the target block. The seismic attribute sequence is composed of the seismic attribute data corresponding to each layer within the determined seismic trace location; The seismic attribute sequence is associated with the depth information of each layer within the determined seismic trace location and then input into the current state sequence stratification prediction model to obtain the high-frequency sequence stratification data of the determined seismic trace location.

3. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1, characterized in that, Adding the high-frequency sequence stratigraphic data of the determined seismic trace locations to the sequential simulation condition data to update the sequential simulation condition data includes: Establish the correspondence between the seismic attribute data and high-frequency sequence layering data of the determined seismic trace locations; The seismic attribute data and high-frequency sequence layering data of the determined seismic trace locations with established correspondences are added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data includes the seismic attribute data and high-frequency sequence layering data with established correspondences, as well as the seismic attribute data of the target block for which no corresponding high-frequency sequence layering data has been obtained.

4. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1, characterized in that, The hierarchical prediction model for the initial state is obtained by constructing learning samples using sequential simulation conditional data of the initial state and training the initial sequence prediction neural network, including: Encode each stratum in the high-frequency sequence layering data of the well point seismic traces of the well with known conditions in the target block to obtain the stratum coding sequence of each point in the well point seismic trace. Use the stratum coding sequence of each point as the sample label of the seismic attribute data of each point. Using the seismic attribute data of each point in the wellpoint seismic trace under known conditions and the sample labels corresponding to each seismic attribute data, an initial sequence prediction neural network is trained, and after training, an initial state sequence hierarchical prediction model is obtained.

5. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 2, characterized in that, When the seismic attribute sequence is associated with the depth information of each layer within the determined seismic trace location and input into the current state of the sequence stratification prediction model, the seismic attribute sequence of the determined seismic trace location is also embedded with the planar location code of the determined seismic trace location.

6. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1, characterized in that, The method further includes: The high-frequency sequence classification results of the target block are obtained by performing sequence classification on the high-frequency sequence data using the established classification model.

7. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1, characterized in that, The sequence prediction neural network is a sequence prediction neural network with a CRF layer added.

8. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 7, characterized in that, The sequence prediction neural network includes a BiLSTM network and a CRF layer connected to the output of the BiLSTM network.

9. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 6, characterized in that, The classification model is the Softmax classifier.

10. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1 or 7, characterized in that, The method further includes: After obtaining the high-frequency sequence stratigraphic data for the determined seismic trace locations, the geological rationality of the high-frequency sequence stratigraphic data for the determined seismic trace locations is judged using the high-frequency sequence stratigraphic data of the known well points. If it is not reasonable, the high-frequency sequence stratigraphic data for the determined seismic trace locations is corrected.

11. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 10, characterized in that, The methods for correcting the high-frequency sequence layering data for determining seismic trace locations include manual correction and adjustment of the constraints of the CRF layer.

12. The intelligent interpretation method for high-frequency seismic sequence logging information according to claim 1, characterized in that, The seismic trace locations predicted by the sequence stratification prediction model for each state are limited to a circular region centered on a point on the sequential path. If there are overlapping areas between the circular regions, the seismic trace locations in the later predicted circular regions will not include those in the overlapping areas.

13. A seismic high-frequency sequence intelligent interpretation device utilizing well logging information, characterized in that, The device includes: The first module is used to input the seismic attribute data with depth information of the seismic trace points determined by the well-side seismic traces of the target block with known conditions determined according to the sequential path into the current state of the sequence stratification prediction model to obtain high-frequency sequence stratification data that determines the seismic trace points. The second module is used to determine whether there are still high-frequency sequence stratigraphic data for seismic trace points in the target block that have not been obtained. If there are still high-frequency sequence stratigraphic data for seismic trace points in the target block that have not been obtained, the high-frequency sequence stratigraphic data for the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data. The updated sequential simulation condition data is then used to construct learning samples to train the current state sequence stratigraphic prediction model, so as to obtain the trained sequence stratigraphic prediction model. This enables the first module to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace point determined according to the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed. Specifically, the initial sequence stratification prediction model is obtained by constructing learning samples using the sequential simulation condition data of the initial state and training the initial sequence prediction neural network. The sequential simulation condition data of the initial state includes well logging high-frequency sequence stratification data and seismic attribute data of well point seismic traces of wells with known conditions in the target block with established correspondence, and seismic attribute data of well point seismic traces of wells with unknown conditions in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence stratification.

14. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent interpretation method for high-frequency seismic sequence using well logging information as described in any one of claims 1 to 12.

15. A machine-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent interpretation method for high-frequency seismic sequence using well logging information as described in any one of claims 1 to 12.

Images