Intelligent seismic high-frequency sequence interpretation method and apparatus using logging information

Abstract

The present invention belongs to the technical field of petroleum and natural gas exploration. Provided are an intelligent seismic high-frequency sequence interpretation method and apparatus using logging information. The method comprises: inputting, into a sequence stratigraphy prediction model in a current state, seismic attribute data of determined point locations, which are determined on the basis of a sequential path, along a well-tie seismic trace of a known conditioning well in a target block, so as to obtain high-frequency sequence stratigraphy data of the determined point locations; and determining whether any high-frequency sequence stratigraphy data of a point location in the target block has not been obtained, and if any high-frequency sequence stratigraphy data of a point location in the target block has not been obtained, adding the high-frequency sequence stratigraphy data of the determined point locations to sequential simulation conditioning data, using updated sequential simulation conditioning data to construct a learning sample to train the sequence stratigraphy prediction model in the current state, and then jumping to the previous step to perform high-frequency sequence stratigraphy prediction on the next determined point location determined on the basis of the sequential path. The method can be applied to onshore early-exploration blocks and offshore sparse-well blocks, etc., so as to realize high-frequency sequence interpretation of these blocks with high efficiency, low costs and high accuracy.

Description

Seismic high-frequency sequence intelligent interpretation method and device using logging information

[0001] Cross-reference to Related Applications

[0002] This application claims the benefit of Chinese Patent Application No. 202411138990.5, filed August 19, 2024, the contents of which are incorporated by reference herein. TECHNICAL FIELD

[0003] The present application belongs to the technical field of oil and gas exploration, and specifically relates to a seismic high-frequency sequence intelligent interpretation method using logging information, a seismic high-frequency sequence intelligent interpretation device using logging information, a computer device, and a machine-readable storage medium. BACKGROUND

[0004] In the field of oil and gas exploration, seismic data structural interpretation is a fundamental and important work, and using seismic profiles to trace stratum interfaces is one of the key steps for identifying and evaluating oil and gas reservoirs. With the continuous increase in domestic oil and gas exploration intensity, the exploration targets have gradually shifted from onshore mature blocks to offshore areas, and passive continental margin basins in deep water / ultra-deep water have become the main battlefield for the discovery of global conventional large and medium-sized oil and gas fields. With the continuous deepening of deepwater oil and gas exploration, evaluation, and development, seismic structural interpretation for offshore few-well / sparse-well blocks has greatly increased in terms of interpretation difficulty, operation workload, accuracy, precision, and timeliness requirements, especially for seismic high-frequency sequence interpretation in offshore few-well blocks. Due to the thin thickness of high-frequency cyclic strata, how to quickly and accurately carry out high-frequency cyclic strata interpretation using limited drilling, logging, and seismic information is also a difficult problem in seismic structural interpretation. Fine structural delineation and geological modeling based on seismic high-frequency sequence interpretation have become an important challenge for oil companies to efficiently explore, evaluate, and develop conventional large and medium-sized oil and gas fields.

[0005] For mature exploration and development blocks, drilling data is very rich, and the block is well controlled by wells. High-frequency sequence stratigraphic data can be interpreted using abundant drilling and logging data, and the lateral trend information of seismic data in the exploration stage can be used for auxiliary constraint, so that high-frequency sequence interpretation can be completed. For early-stage onshore exploration blocks and offshore few-well blocks, drilling and logging data are very limited, and well control is very low. Therefore, on the basis of limited high-frequency cyclic stratigraphic information of drilling and logging, high-frequency cyclic stratigraphic information in seismic data needs to be fully tapped to realize seismic high-frequency sequence prediction using drilling and logging information.

[0006] The existing technology research for identifying high-frequency sequences by earthquakes shows that the existing technical solutions for identifying high-frequency sequences by earthquakes mainly include two types. The first type of solution mainly relies on making synthetic records, establishing the correlation between seismic horizons and logging horizons, and identifying and determining the stratigraphic boundaries through manual interpretation. The second type of solution is a seismic fine sequence prediction method based on neural networks. The first type of solution mainly has the following defects: 1) due to the limitation of seismic data resolution, it is impossible to calibrate high-frequency stratification information on the logging scale to seismic data, and thus it is impossible to perform high-precision division of the geological stratification of the target block. Even under the guidance of the theory and technical system of seismic sedimentology, after frequency raising processing and 90° phase transformation of the seismic data, the target block is interpreted for high-frequency sequences, there are still defects such as long time-consuming, low efficiency and insufficient accuracy of manual interpretation; 2) for early exploration offshore few-well blocks, usually the seismic data area is large, and multiple geophysical interpreters are required to cooperatively perform horizon tracking interpretation. However, due to the differences in the understanding of the regional geology and sedimentary sequence of the research area by different interpreters, the seismic interpretation results may be significantly affected by human factors, which makes the accuracy and reliability of the horizon interpretation results uncertain. Therefore, the interpreters need to have a deep understanding of the seismic response characteristics of the regional geology and sedimentary sequence of the research area. It can be seen that high requirements are placed on the professional level of the interpreters. The second type of solution mainly has the following defects: for high-frequency sequence interpretation of strata under the condition of less drilling and logging data in early exploration onshore blocks and offshore few-well blocks, the existing artificial intelligence model has limited data learning and modeling capacity for large-area seismic data, multi-attribute seismic information is not fully utilized, and it is difficult to fully exploit high-frequency sequence information in seismic data. SUMMARY

[0007] The purpose of the embodiments of the present application is to provide a seismic high-frequency sequence intelligent interpretation method using logging information, a seismic high-frequency sequence intelligent interpretation device using logging information, a computer device and a machine-readable storage medium, to overcome one or more defects in the use of manual interpretation means and deep learning technology in the seismic high-frequency sequence interpretation of early exploration blocks and offshore few-well blocks in the prior art.

[0008] In order to achieve the above object, the first aspect of the embodiment of the present application provides a seismic high-frequency sequence intelligent interpretation method using logging information, which comprises: inputting seismic attribute data associated with depth information of a determined seismic trace point into a sequence layering prediction model in a current state to obtain high-frequency sequence layering data of the determined seismic trace point, the determined seismic trace point being a well-side seismic trace point of a known condition well in a target block determined according to a sequential path; judging whether there is still high-frequency sequence layering data of a seismic trace point not obtained in the target block, if yes, adding the high-frequency sequence layering data of the determined seismic trace point into sequential simulation condition data to update the sequential simulation condition data, and using the updated sequential simulation condition data to construct a learning sample to train the sequence layering prediction model in the current state to obtain a trained sequence layering prediction model, then jumping to the previous step to perform high-frequency sequence layering prediction on a next determined seismic trace point determined according to the sequential path, otherwise, the high-frequency sequence layering prediction for the target block is completed. The sequence layering prediction model in the initial state is obtained by training an initial sequence prediction neural network using an initial state sequential simulation condition data, the initial state sequential simulation condition data comprising logging high-frequency sequence layering data and seismic attribute data of a well point seismic trace of a known condition well in the target block having established a corresponding relationship, and seismic attribute data of a well point seismic trace of a non-known condition well in the target block; the seismic attribute data being seismic attribute data sensitive to high-frequency sequence.

[0009] Optionally, the inputting of the seismic attribute data associated with depth information of the determined seismic trace point into the sequence layering prediction model in the current state to obtain the high-frequency sequence layering data of the determined seismic trace point comprises: determining a determined seismic trace point of a well-side seismic trace of a known condition well in a target block to be subjected to high-frequency sequence layering prediction according to a sequential path; determining layering information of the determined seismic trace point according to stratum top and bottom surface information obtained from seismic data of the target block; composing a seismic attribute sequence from seismic attribute data corresponding to each layering in the determined seismic trace point; inputting the seismic attribute sequence associated with depth information of each layering in the determined seismic trace point into the sequence layering prediction model in the current state to obtain the high-frequency sequence layering data of the determined seismic trace point.

[0010] Optionally, the high-frequency sequence layering data of the determined seismic trace point is added to the sequential simulation condition data to update the sequential simulation condition data, comprising: establishing a corresponding relationship between the seismic attribute data and the high-frequency sequence layering data of the determined seismic trace point; and adding the seismic attribute data and the high-frequency sequence layering data of the determined seismic trace point with the established corresponding relationship to the sequential simulation condition data to update the sequential simulation condition data, wherein the updated sequential simulation condition data comprises the seismic attribute data and the high-frequency sequence layering data with the established corresponding relationship, and the seismic attribute data in the target block without corresponding high-frequency sequence layering data.

[0011] Optionally, the sequence layering prediction model in the initial state is obtained by training the initial sequence prediction neural network using the learning sample constructed by the initial state of the sequential simulation condition data, comprising: encoding each stratum in the well point seismic trace logging high-frequency sequence layering data of the known condition well in the target block to obtain the stratum encoding sequence of each point of the well point seismic trace, and taking the stratum encoding sequence of each point as the sample label of the seismic attribute data of each point in a one-to-one manner; and training the initial sequence prediction neural network using the seismic attribute data of each point of the well point seismic trace of the known condition well and the sample label corresponding to each seismic attribute data, and obtaining the sequence layering prediction model in the initial state after the training is completed.

[0012] Optionally, when the seismic attribute sequence and the depth information of each stratum in the determined seismic trace point are associated and input into the sequence layering prediction model in the current state, the seismic attribute sequence of the determined seismic trace point further embeds the planar position encoding of the determined seismic trace point.

[0013] Optionally, the method further comprises: obtaining the high-frequency sequence division result of the target block by performing sequence classification on the high-frequency sequence layering data using the constructed classification model.

[0014] Optionally, the sequence prediction neural network is a sequence prediction neural network with a CRF layer added.

[0015] Optionally, the sequence prediction neural network comprises a BiLSTM network and a CRF layer connected to the output end of the BiLSTM network.

[0016] Optionally, the method further comprises: after the high-frequency sequence layering data of the determined seismic trace point is predicted, performing geological rationality judgment on the high-frequency sequence layering data of the determined seismic trace point using the high-frequency sequence layering data of the well point of the known condition well, and if it is not reasonable, modifying the high-frequency sequence layering data of the determined seismic trace point.

[0017] Optionally, the correcting the high-frequency sequence layering data of the determined seismic trace point comprises: generating a correction instruction for manually correcting an unreasonable sequence of the determined seismic trace point according to the unreasonable sequence of the determined seismic trace point reflected by the high-frequency sequence layering data of the determined seismic trace point and the high-frequency sequence layering data of the well point of the known condition well; or generating a correction instruction for adjusting the constraint condition of the CRF layer, updating the sequence layering prediction model based on the adjusted constraint condition of the CRF layer, inputting the seismic attribute data associated with the depth information of the determined seismic trace point into the updated sequence layering prediction model, and obtaining the corrected high-frequency sequence layering data of the determined seismic trace point.

[0018] Optionally, the seismic trace points predicted by the sequence layering prediction model of each state are limited in a circular region with the point on the sequential path as the center, and if there is an overlapping region between the circular regions, the seismic trace points in the circular region predicted in the later are not included in the seismic trace points in the overlapping region.

[0019] Optionally, the sequence layering prediction model of each state predicts the high-frequency sequence layering data of the point where the initial center on the sequential path is located.

[0020] The second aspect of the embodiment of the present application provides a device for intelligent interpretation of seismic high-frequency sequence by using logging information, which comprises: a first module for inputting the seismic attribute data associated with the depth information of the determined seismic trace point of the well seismic trace of the known condition well of the target block determined according to the sequential path into the sequence layering prediction model of the current state to obtain the high-frequency sequence layering data of the determined seismic trace point; a second module for judging whether there is still high-frequency sequence layering data of the seismic trace point not obtained in the target block, if there is still high-frequency sequence layering data of the seismic trace point not obtained in the target block, adding the high-frequency sequence layering data of the determined seismic trace point into the sequential simulation condition data to update the sequential simulation condition data, and using the updated sequential simulation condition data to construct a learning sample to train the sequence layering prediction model of the current state to obtain the trained sequence layering prediction model, so that the first module performs high-frequency sequence layering prediction on the next determined seismic trace point determined according to the sequential path, otherwise, the high-frequency sequence layering prediction for the target block is completed. The sequence layering prediction model of the initial state is obtained by training the initial sequence prediction neural network using the learning sample constructed by the sequential simulation condition data of the initial state, the sequential simulation condition data of the initial state comprises the logging high-frequency sequence layering data and the seismic attribute data of the well point seismic trace of the known condition well of the target block having established a corresponding relationship, and the seismic attribute data of the seismic trace point in the target block which is not a known condition well; the seismic attribute data is the seismic attribute data sensitive to high-frequency sequence.

[0021] A third aspect of the embodiments of the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for intelligent interpretation of high-frequency sequence of seismic data using well logging information according to the first aspect of the embodiments of the present application.

[0022] A fourth aspect of the embodiments of the present application provides a machine readable storage medium, having a computer program stored thereon, wherein the computer program is executable on a processor to implement the method for intelligent interpretation of high-frequency sequence of seismic data using well logging information according to the first aspect of the embodiments of the present application.

[0023] In the above technical solution, the high-frequency sequence layering data obtained by well logging interpretation at the well point of the known condition well in the target block and the seismic attribute sensitive to the high-frequency sequence extracted from the seismic data of the well point seismic trace are used as initial learning samples to learn and train the sequence prediction neural network, so as to construct the initial state sequence layer prediction model. Based on the sequential simulation technology, the high-frequency sequence layering data of the point of the seismic trace beside the known condition well is predicted by using the sequence layer prediction model, and the prediction result of the seismic trace point is expanded to the learning sample. Then, according to the steps of "learning-prediction-updating-learning", the learning sample is sequentially expanded, the high-frequency sequence layering data of the known condition well point seismic trace is gradually expanded to the adjacent seismic trace of the known condition well, and finally the high-frequency sequence layering data of the entire target block is predicted.

[0024] Therefore, by combining the sequence prediction neural network and the sequential simulation, the high-frequency sequence space prediction of the entire target block is realized by using a small amount of well logging information and a large amount of seismic data in the target block, the limitation of the traditional artificial intelligence model on the data learning and modeling ability of the large area seismic data is overcome, the seismic attribute data sensitive to the high-frequency sequence is fully utilized, and compared with the artificial interpretation and the deep learning scheme based on the traditional artificial intelligence model, the high-frequency sequence space prediction has high efficiency, high accuracy and low cost, and has important application value for the risk exploration, fine structure modeling, oil and gas reservoir evaluation and development of the oil and gas field in the few well / sparse well block.

[0025] Other features and advantages of the embodiments of the present application will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0026] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used together with the following specific embodiments to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the drawings:

[0027] Fig. 1 schematically shows a flow chart of a seismic high-frequency sequence intelligent interpretation method using well logging information according to an embodiment of the present application;

[0028] Fig. 2 schematically shows a technical roadmap of the seismic high-frequency sequence intelligent interpretation method using well logging information in a specific application example;

[0029] Fig. 3 schematically shows a schematic diagram of the shape of a target prediction block and the location of a known condition well in a specific application example;

[0030] Fig. 4 schematically shows a first seismic attribute profile extracted from seismic data in a specific application example;

[0031] Fig. 5 schematically shows a second seismic attribute profile extracted from seismic data in a specific application example;

[0032] Fig. 6 schematically shows a third seismic attribute profile extracted from seismic data in a specific application example;

[0033] Fig. 7 schematically shows a fourth seismic attribute profile extracted from seismic data in a specific application example;

[0034] Fig. 8 schematically shows a horizon label setting diagram in a specific application example;

[0035] Fig. 9 schematically shows a prediction example of a BiLSTM+CRF model in a specific application example;

[0036] Fig. 10 schematically shows a comparison diagram of prediction effects of a sequence horizon prediction model without adding a CRF layer and a sequence horizon prediction model adding a CRF layer in a specific application example, wherein part (a) is a prediction effect diagram of the sequence horizon prediction model without adding the CRF layer, and part (b) is a prediction effect diagram of the sequence horizon prediction model adding the CRF layer;

[0037] Fig. 11 schematically shows a sequential simulation process diagram according to a sequential path in a specific application example, wherein parts (a)-(d) are schematic diagrams of four stages of the sequential simulation process;

[0038] Fig. 12 schematically shows a working principle diagram of a Softmax classifier in a specific application example;

[0039] Fig. 13 schematically shows a seismic profile display example diagram after frequency extraction of prediction results of six layer interfaces in a specific application example;

[0040] Fig. 14 schematically shows a three-dimensional structure layer model example diagram of six layer interfaces in a specific application example;

[0041] Fig. 15 schematically shows a component block diagram of a seismic high-frequency sequence intelligent interpretation device using logging information according to an embodiment of the present application;

[0042] Fig. 16 schematically shows a structural block diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] The specific embodiments of the embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely intended to illustrate and explain the embodiments of the present application, and are not intended to limit the embodiments of the present application.

[0044] Embodiment one

[0045] The seismic high-frequency sequence intelligent interpretation method using logging information provided by the embodiments of the present application can be applied to seismic high-frequency sequence interpretation of onshore exploration early blocks and offshore few-well blocks, which have large-area seismic data and less logging data. It should be noted that the above description does not constitute a limitation on the application scenarios of the embodiments of the present application. It can be known that the seismic high-frequency sequence interpretation of other blocks with large-area seismic data and less logging data is also applicable, so the application scenarios of the embodiments of the present application are relatively wide.

[0046] Referring to Fig. 1, the embodiments of the present application provide a seismic high-frequency sequence intelligent interpretation method using logging information, which includes the following implementation steps S100 to S200.

[0047] Step S100, inputting seismic attribute data associated with depth information of a determined seismic trace point into a constructed sequence layering prediction model in a current state to obtain high-frequency sequence layering data of the determined seismic trace point, the determined seismic trace point being a well-side seismic trace point of a known condition well in a target block determined according to a sequential path. Wherein, the sequence layering prediction model in an initial state is obtained by training and parameter optimization of an initial sequence prediction neural network using learning samples constructed by using sequence simulation condition data in the initial state; the sequence simulation condition data in the initial state includes logging high-frequency sequence layering data and seismic attribute data of a well point seismic trace of a known condition well in the target block having established a corresponding relationship, and seismic attribute data of a well point seismic trace of a non-known condition well in the target block; the seismic attribute data is seismic attribute data sensitive to high-frequency sequence.

[0048] Based on the above description, it should be understood that:

[0049] 1) The known condition well is a drilled well with known conditions.

[0050] 2) The seismic attribute data is generally divided into multiple types according to motion dynamics, according to picking methods, etc., that is, the types of seismic attributes that can be extracted from seismic data are very rich. Here, the seismic attribute data as the input data of the sequence layering prediction model is the seismic attribute data sensitive to high-frequency sequences. Among them, sensitive to high-frequency sequences means that the seismic attribute type has a strong correlation with the high-frequency sequence horizon change. Generally, the seismic attribute data sensitive to high-frequency sequences is multiple, and the multiple seismic attribute data as the input data of the sequence layering prediction model can be determined based on expert experience from multiple seismic attribute data sensitive to high-frequency sequences, or the correlation analysis method such as PCA principal component analysis method can be used to analyze multiple seismic attribute data, and multiple seismic attribute data with strong correlation with high-frequency sequence layering can be obtained as the input data of the sequence layering prediction model.

[0051] 3) Sequential simulation (SS, English full name: Sequential Simulation) is a conditional simulation method, and the simulation idea is to sequentially calculate the simulation value of each grid node along the pre-prepared path.

[0052] 4) Establishing a corresponding relationship between the well point seismic trace of the known condition well of the target block and the logging high-frequency sequence layering data and the seismic attribute data means that in order to constitute the initial state learning sample, the corresponding relationship between the sample and the sample label needs to be established in advance, that is, the seismic attribute data extracted from the seismic data belonging to a certain seismic trace point is associated with the logging high-frequency sequence layering data of the seismic trace point, for example, the association can be established through depth information, the seismic attribute data of a certain seismic trace point is taken as input, and the logging high-frequency sequence layering data of the point is taken as the sample label of training. It is known that for a certain seismic trace point, in order to take the seismic attribute sequence as the input data of the sequence prediction neural network, the seismic attribute data of each layer determined by the top and bottom surfaces of the seismic data of the seismic trace point needs to be collectively constituted as a seismic attribute sequence, and then the seismic attribute sequence is taken as the input sequence of the sequence prediction neural network. In order to take the logging high-frequency sequence layering data of the seismic trace point as the sample label of training, the sequence data needs to be constituted by each layering information in the logging high-frequency sequence layering data of the seismic trace point, for example, each layer in the logging high-frequency sequence layering data can be given a coding value, and the sequence composed of each coding value is taken as the sample label. Exemplarily, in a specific embodiment, the layer numbers of the strata of the target block can be coded as 1, 2, 3, 4, and 5, a total of five layers.

[0053] Accordingly, for example, in one or more specific embodiments, the seismic attribute data of the determined seismic trace point position associated with depth information is input into the current state of the sequence layering prediction model to obtain high-frequency sequence layering data of the determined seismic trace point position, specifically including the following implementation steps: determining the determined seismic trace point position of the well seismic trace of the known condition well of the target block to be subjected to high-frequency sequence layering prediction according to the sequential path; determining the layering information of the determined seismic trace point position according to the stratum top and bottom surface information obtained from the seismic data of the target block; forming a seismic attribute sequence from the seismic attribute data corresponding to each layering in the determined seismic trace point position; and inputting the seismic attribute sequence associated with the depth information of each layering in the determined seismic trace point position into the current state of the sequence layering prediction model to obtain high-frequency sequence layering data of the determined seismic trace point position.

[0054] Illustratively, in one or more specific embodiments, the initial sequence prediction neural network is trained using the initial state of the sequential simulation condition data to obtain the initial state of the sequence layering prediction model, specifically including the following implementation steps: encoding each stratum in the logging high-frequency sequence layering data of the well seismic trace of the known condition well of the target block to obtain the stratum encoding sequence of each point position of the well seismic trace of the known condition well, and taking the stratum encoding sequence of each point position as the sample label of the seismic attribute data of each point position in a one-to-one correspondence; training the initial sequence prediction neural network using the seismic attribute data of each point position of the well seismic trace of the known condition well and the sample label corresponding to each seismic attribute data, and obtaining the initial state of the sequence layering prediction model after training.

[0055] Illustratively, in one or more specific embodiments, when the seismic attribute sequence and the depth information of each corresponding layering in the determined seismic trace point position are input into the current state of the sequence layering prediction model, the seismic attribute sequence of the determined seismic trace point position also embeds the planar position encoding of the seismic trace point position.

[0056] It needs to be understood that, for a certain seismic trace point position, the seismic attribute data input into the initial sequence prediction neural network is a seismic attribute sequence composed of the seismic attribute of each layering of the seismic trace point position, and by embedding the planar position encoding of the seismic trace point position in the seismic attribute sequence, the sensitivity of the sequence prediction neural network to the position of each element in the seismic attribute sequence is enhanced, thereby improving the robustness of the network. For example, the planar position encoding can be the (XLINE, INLINE) coordinates corresponding to the seismic trace point position on the plane, such as (633, 784), or a new coordinate system can be established according to the requirements of the target block, such as when the current coordinate system cannot meet the high accuracy requirement of the target block on high-frequency sequence.

[0057] In step S200, it is determined whether there is high-frequency sequence layering data of seismic trace points in the target block that has not been obtained. If yes, the high-frequency sequence layering data of the determined seismic trace points obtained in step S100 is added to the sequential simulation condition data to update the sequential simulation condition data, and the updated sequential simulation condition data is used to train the sequence layering prediction model of the current state using the learning sample to obtain the trained sequence layering prediction model, and then jump to step S100 to perform high-frequency sequence layering prediction on the next determined seismic trace point according to the sequential path. Otherwise, the high-frequency sequence layering prediction for the target block is completed.

[0058] For example, in one or more specific embodiments, the high-frequency sequence layering data of the determined seismic trace points 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: establishing a corresponding relationship between the high-frequency sequence layering data of the determined seismic trace points obtained in step S100 and the seismic attribute data; adding the seismic attribute data and the high-frequency sequence layering data of the determined node with the established corresponding relationship to the sequential simulation condition data to update the sequential simulation condition data, and the updated sequential simulation condition data includes the seismic attribute data and the high-frequency sequence layering data with the established corresponding relationship, and the seismic attribute data in the target block for which the corresponding high-frequency sequence layering data has not been obtained.

[0059] Exemplarily, in one or more specific embodiments, the sequence prediction neural network is a sequence prediction neural network added with a CRF (Conditional Random Field) layer. It is known that the sequence prediction neural network added with the CRF layer means that after feature extraction by the sequence prediction neural network, the output is passed to the CRF layer for global optimization and labeling, thereby establishing an end-to-end layer sequence horizon prediction model. The CRF layer is a kind of undirected graph network, which can realize labeling or classification of sequence data. The CRF layer can consider the relationship between adjacent sample points as a constraint condition for layer sequence horizon prediction, thereby improving the accuracy of sequence prediction, avoiding unreasonable prediction, preventing layer sequence prediction disorder, and making the prediction result of layer sequence horizon more consistent with geological laws. It is known that the constraint condition of the CRF layer is generally realized by defining a feature function, which limits the output sequence obtained after feature extraction of the sequence prediction neural network by the feature function, and each limit corresponds to a feature function. For example, if there are 5 sub-layers in the stratigraphic sequence, a feature function is defined to ensure that the output sequence predicted by the sequence prediction neural network contains all 5 layers. This feature function can check whether the predicted output sequence contains the labels 1, 2, 3, 4, and 5. If not all are included, the model score is penalized. For another example, the stratigraphic sequence cannot have a cross-layer or jump-layer phenomenon. Another feature function is defined to ensure that the predicted output sequence conforms to the rule of gradual increase, such as checking whether the layer number coding of adjacent two layers conforms to the above rule, for example, the unreasonable phenomenon of layer 1 to layer 3 cannot occur, the unreasonable phenomenon of layer 3 to layer 2 cannot occur, and only the increasing situation of layer 1 to layer 2 and layer 2 to layer 3 can occur. When training the CRF layer, the above defined feature functions are added to the layer sequence horizon prediction model, and the weight adjustment of these feature functions is completed through training. It should be known that the performance of the CRF layer is affected by feature selection, so appropriate feature functions need to be designed in combination with the specific target block prediction requirements and the characteristics of the sequential simulation condition data. The specific feature functions involved in this embodiment are not described in detail.

[0060] In one specific application, for example, after adding the feature function to the layer sequence horizon prediction model and after training, the probability of transferring from layer 2 to layer 1 is set to 0, the probability of transferring from layer 1 to layer 3 is set to 0, and so on. Each horizon has a transfer score for indicating the size of the possibility of transferring from the current layer to another layer. The transfer score can help the prediction model based on the sequence prediction neural network to better understand the relationship between the horizons, thereby improving the performance of the prediction model based on the sequence prediction neural network in tasks such as seismic sequence labeling.

[0061] In one comparative example, the sequence prediction neural network is a sequence prediction neural network without a CRF layer, as shown in FIG. 10, the prediction result of the layer sequence horizon prediction model without the CRF layer shows a layer skipping phenomenon, while the prediction result of the layer sequence horizon prediction model with the CRF layer does not show a layer skipping and layer skipping phenomenon.

[0062] As can be seen, in the above examples, by introducing a CRF layer in the sequence prediction neural network, the prediction accuracy of the high-frequency layer sequence horizon is improved, thereby improving the accuracy of the seismic high-frequency layer sequence interpretation using a small amount of well logging information.

[0063] Exemplarily, in one or more specific embodiments, the sequence prediction neural network comprises 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 a forward direction, and the backward LSTM processes the seismic attribute sequence in a reverse direction. After processing, the outputs of the two LSTMs are spliced together, thereby fully utilizing the context information of the seismic attribute sequence, effectively extracting the seismic data features, and better finding and utilizing the relationship between the stratigraphic coding sequence and the changes before and after the seismic attribute sequence.

[0064] As can be known, although the above examples mention the BiLSTM network as the preferred core network of the layer sequence horizon prediction model, it is not a limitation on the layer sequence horizon prediction model. In application, a suitable sequence prediction neural network should be selected according to the specific scene conditions, for example, other sequence prediction neural networks with lower requirements for device computing power can be selected, or the known sequence prediction neural network can be modified to achieve the same or better prediction accuracy as the BiLSTM network. The present application embodiments cannot exhaustively describe various sequence prediction neural networks applicable to the present application, but at least for ordinary skilled persons in the art, the applicability of commonly used sequence prediction neural networks in the present application cannot be ruled out, such as RNN, LSTM, etc.

[0065] As in the above examples, the BiLSTM network and the CRF layer are combined to form a sequence prediction neural network, and the high-frequency layer sequence data prediction using this network has high accuracy, meeting the demand for seismic high-frequency layer sequence interpretation of onshore exploration early blocks and offshore few-well blocks.

[0066] Exemplarily, in one or more specific embodiments, after predicting the high-frequency layer sequence data of the determined seismic trace point, the high-frequency layer sequence data of the known condition well point is used to make a geological rationality judgment on the high-frequency layer sequence data of the determined seismic trace point. If it is not reasonable, the high-frequency layer sequence data of the determined seismic trace point is modified.

[0067] As in the above embodiment, it is determined whether the predicted high-frequency sequence layering data is reasonable by comparing with the high-frequency sequence layering data of the known condition well points, for example, the geological reasonability determination includes the stratum sequence reasonability determination and the stratum thickness reasonability determination, by comparing the high-frequency sequence layering data of the predicted seismic trace points of the adjacent well position of the condition well with the high-frequency sequence layering data of the seismic trace of the condition well point, if the layer position ratio difference is too large or the sequence prediction is unreasonable, at this time, it is determined that the predicted high-frequency sequence layering data is unreasonable, and the high-frequency sequence layering data of the determined seismic trace points needs to be corrected.

[0068] Exemplarily, in one or more specific embodiments, the correction manner can be: correction by manual assistance; correction by adjusting the constraint conditions of the CRF layer. Specifically, the specific process of correction by manual assistance includes: generating a correction instruction for manually correcting the unreasonable sequence of the determined seismic trace points according to the unreasonable sequence of the determined seismic trace points reflected by the high-frequency sequence layering data of the determined seismic trace points and the high-frequency sequence layering data of the known condition well points. The specific process of correction by adjusting the constraint conditions of the CRF layer includes: generating a correction instruction for adjusting the constraint conditions of the CRF layer according to the unreasonable sequence of the determined seismic trace points reflected by the high-frequency sequence layering data of the determined seismic trace points and the high-frequency sequence layering data of the known condition well points, updating the sequence layering prediction model based on the adjusted constraint conditions of the CRF layer, inputting the seismic attribute data associated with the depth information of the determined seismic trace points into the updated sequence layering prediction model, and obtaining the corrected high-frequency sequence layering data of the determined seismic trace points.

[0069] For example, when determining the stratum sequence reasonability, it is determined whether the layer penetration or layer skipping phenomenon occurs, whether it starts from a specified 1-sublayer and ends at a 5-sublayer, etc., if the above situations occur, the unreasonable high-frequency sequence layering prediction result needs to be manually corrected, for example, the sequence code corresponding to the layer penetration or layer skipping of the seismic trace points is corrected, thereby providing reliable learning samples for the next round of model learning. For another example, when determining the stratum thickness reasonability, if the thickness difference of the same sublayer of the adjacent seismic traces is too large, at this time, the geological features of the target block at this position need to be combined, for example, whether there is a fault line, if there is no fault line, it needs to be manually corrected to a reasonable range.

[0070] As in the above embodiment, the prediction accuracy of the sequence layering prediction model on the seismic high-frequency sequence layering data is improved by the geological reasonability determination and correction on the predicted high-frequency sequence layering data, thereby the accuracy of the seismic high-frequency sequence interpretation based on a small amount of well logging information is improved.

[0071] Exemplarily, in one or more specific embodiments, the seismic trace points predicted by the layer sequence hierarchical prediction model of each state are limited in a circular region with the point on the sequential path as the center, and if there is an overlapping region between the circular regions, the seismic trace points in the overlapping region are not included in the seismic trace points in the circular region predicted later. Preferably, each circular region has the same radius.

[0072] As an improvement of the above embodiment, in the following specific embodiments, an information base is introduced, i.e., the high-frequency layer sequence hierarchical data and the seismic attribute data of the well point seismic trace points of the known condition well are used as the information base, regardless of the change of the center of the prediction range (circular region) of the layer sequence hierarchical prediction model, the information base is always used to construct the learning sample, i.e., the information base is always used as the sample point in the learning sample. As in the above embodiment, because the information base represents the most reliable high-frequency layer sequence hierarchical information, by always using the most reliable high-frequency layer sequence hierarchical information as the sample point of the learning sample, the prediction accuracy of the layer sequence horizon prediction model is improved.

[0073] Embodiment two

[0074] The embodiment of the present application provides a seismic high-frequency layer sequence intelligent interpretation method using logging information, which is different from the embodiment one in that after the step S200, the following steps are further included:

[0075] In step S300, the layer sequence classification is performed on the high-frequency layer sequence hierarchical data by using the constructed classification model, and the high-frequency layer sequence division result of the target block is obtained.

[0076] For example, if the high-frequency layer sequence division rule has been defined for the known condition well position, and the high-frequency layer sequence division has been performed on the seismic trace of the well point by using the rule, the high-frequency layer sequence division can be performed on other high-frequency layer sequence hierarchical data of the target block by using the high-frequency layer sequence division rule, and at this time, the layer sequence type defined by the high-frequency layer sequence division rule can be used as the layer sequence classification type of the classification model.

[0077] In a comparative example, after obtaining the high-frequency layer sequence hierarchical data of each point of the target block, the high-frequency layer sequence hierarchical data can be manually interpreted to obtain the high-frequency layer sequence division result of the target block. However, the efficiency and accuracy of the manual high-frequency layer sequence division are obviously lower than those of the classification model-based high-frequency layer sequence division. Therefore, the layer sequence classification is performed on the high-frequency layer sequence hierarchical data by using the constructed classification model, which takes into account the high efficiency and high accuracy.

[0078] Exemplarily, in one or more specific embodiments, the classification model is a Softmax classifier. The Softmax classifier is a typical classifier that can complete a multi-classification task. It needs to be understood that, although the above description mentions the Softmax classifier as the classification model, it is not a limitation on the classification model. In application, a suitable classification model needs to be selected in combination with specific scene conditions. The embodiments of the present application cannot exhaustively describe various classification models that can be applied to the present application, but at least for ordinary skilled persons in the art, the applicability of commonly used classification models in the present application cannot be ruled out, such as SVM, Bayesian classifier, etc.

[0079] Embodiment three

[0080] The embodiment of the present application is an application example of the above-mentioned embodiment two. In combination with the technical roadmap shown in FIG. 2, the embodiment of the present application provides a seismic high-frequency sequence intelligent interpretation method using well logging information, which includes the following implementation steps:

[0081] Step SS1, using the high-frequency sequence layering data of the well point seismic trace of the known condition well in the target block obtained through well logging data, the target block plane position information and the plane position information of the known condition well to establish a condition information library for storing sequential simulation condition data, and encoding the plane position of the known condition well and storing it in the condition information library, and reading the top and bottom surface information of the target block obtained from the seismic data into the condition information library.

[0082] Step SS2, extracting seismic attribute data in the SEGY format seismic data volume, optimizing depth domain seismic attribute data according to the difference in sensitivity to high-frequency layering, then extracting the seismic attribute data of the seismic trace beside the known condition well, and storing these seismic attribute data in the condition information library.

[0083] Step SS3, classifying and extracting the data in the condition information library, and transferring the well logging high-frequency sequence layering data of the known condition well point seismic trace to the seismic attribute data at the same depth, completing the correspondence and coding between the seismic attribute data and the high-frequency sequence layering data of the known condition well point seismic trace, setting sample labels according to the correspondence relationship between the high-frequency sequence layering data and the seismic attribute data of the known condition well, and performing normalization mathematical operation after sample data screening. Among them, because the seismic attribute data is discrete when it is used as sample data, in order to improve the training efficiency of the subsequent sequence prediction neural network, the seismic attribute data is normalized.

[0084] Step SS4, learn the characteristics of the sample data around the high-frequency sequence stratification data and the planar position coding information of the known condition well, establish a BiLSTM+CRF neural network model (referring to the sequence prediction neural network composed of the BiLSTM network and the CRF layer connected with the output end of the BiLSTM network in the above embodiment), and carry out learning and training on the BiLSTM+CRF neural network model for the high-frequency sequence stratification problem, so as to ensure a high self-test accuracy rate for the learning sample, obtain an initial state sequence stratigraphic prediction model, and then save the sequence stratigraphic prediction model containing the weight value information between each network node after learning and training.

[0085] Step SS5, obtain the seismic attribute data of the point (referred to as: sample point to be predicted) of the condition well to be predicted for high-frequency sequence stratification from the seismic trace beside the known condition well in the condition information library, use the sequence stratigraphic prediction model after learning and training to predict the high-frequency sequence stratification of the sample point, and combine the high-frequency sequence stratification information of the known condition well well point to evaluate and correct the prediction result, so as to meet the rationality in the geological sense, that is, to meet the rationality standard of the high-frequency sequence stratification of the seismic trace beside the known condition well in the geological sense, thereby completing the high-frequency sequence stratification prediction of the seismic attribute of the adjacent position of the limited condition well.

[0086] Step SS6, record the predicted high-frequency sequence stratification data and position information of the adjacent position of the known condition well, implant them into the condition information library, and update the condition content and structure in the condition information library, use the extended condition information library to update the learning sample, achieve the purpose of synchronous updating and expansion of the learning sample, and use the sequence stratigraphic prediction model after updating the learning sample to predict other sample points of the seismic trace adjacent to the known condition well, and correct the prediction result. Through the operation of "learning-prediction-correction-updating-learning", the cyclic updating of the condition information library and the sequential expansion of the learning sample are realized.

[0087] Among them, the sample point range predicted by the sequence stratigraphic prediction model of each state is a circular area with the point on the sequential path as the center, and the sequential prediction principle expressed in step SS6 is as follows:

[0088] Suppose the coordinate of the initial sample point is (x0, y0), and there is corresponding seismic trace stratification information S and depth information D, then according to the pre-defined circular area radius R, the sample point range predicted by the initial state sequence stratigraphic prediction model can be obtained, which is called the initial circular area.

[0089] Suppose the function expression of the sequence stratigraphic prediction model is f0(K, S, D), where θ represents the parameters of the neural network, and K represents the seismic attribute data of the initial sample point, then the prediction result of the initial circular area can be expressed as: S pred(x, y) = f0(K, S, D) · I(x, y), I(x, y) is an indicator function and is used to represent whether the point (x, y) is within the initial circular region, and

[0090] In order to predict the new circular region within the target block, the center of the circle (x0, y0) is translated along the sequential path, assuming that the translation vector is represented as (Δx, Δy), then the coordinates of the center of the new circular region are represented as (x0+Δx, y0+Δy);

[0091] The sample points within the new circular region which have not been predicted are predicted by high-frequency layer sequence stratification, and the prediction result of the new circular region can be represented as: S pred (x', y') = f θ (K, S, D) · I(x', y'), I(x', y') represents whether the point (x', y') is within the new circular region, and

[0092] In addition, it needs to be understood that in actual application, the input of the layer sequence stratigraphic prediction model can also include other more characteristic parameters, for example, when the influence of lithofacies on high-frequency layer sequence is considered, the lithofacies is taken as the input of the layer sequence stratigraphic prediction model. Moreover, in actual application, the boundary condition and the updating strategy of the prediction result can be considered, the boundary condition refers to the boundary and the overall shape of the target block, and the boundary condition can affect the rules of sequential prediction, for example, if the boundary is extremely irregular, the operation of reducing the circular region where the learning sample and the prediction sample point are located can be taken, in addition, in order to obtain the next round of learning sample, the strategy for updating the current learning sample after updating the conditional information library based on the current prediction result is adjusted according to the morphology of the target block. At the same time, the selection of the learning sample capacity, the prediction step and the radius of the circular region needs to be determined according to the size of the target block and the computer computing power, so as to reduce the learning sample capacity and improve the prediction step under the premise of ensuring the prediction accuracy as much as possible, thereby reducing the occupation of computing resources.

[0093] In step SS7, the high-frequency layer sequence stratification rule is determined by analyzing the high-frequency layer sequence stratification data of the known conditional well of the target block, so as to obtain the category type of the high-frequency layer sequence, and then a classification model for classifying the high-frequency layer sequence stratification information is designed according to the category type of the high-frequency layer sequence, and then the high-frequency layer sequence stratification information of each point in the target block is classified by using the classification model, so as to obtain the high-frequency layer sequence stratification result of the target block.

[0094] The classification model adopts a Softmax classifier. In addition, the construction principle of the Softmax classifier can be combined with the process in the ordinary embodiment, which will not be described here.

[0095] Embodiment Four

[0096] The embodiment of the present application is the specific application of the above-mentioned embodiment three in a few-well block (hereinafter referred to as the target prediction block) in a certain sea area abroad. In combination with the method shown in Figs. 3 to 14, in the application, the seismic high-frequency sequence layering interpretation method using the logging information comprises the following specific steps:

[0097] Step A1, by analyzing the data of the target prediction block, obtaining the plane position data of each seismic trace of the target prediction block, establishing a virtual coordinate axis in the computer, reproducing the shape of the target prediction block, and at the same time, calibrating and coding 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 position of the condition wells are shown in Fig. 3. Save the high-frequency sequence layering data of the two condition wells. Establish a condition information library and store the above-mentioned data in it. At the same time, store the top and bottom surface data in the vertical direction of the target prediction block in the condition information library for subsequent use. The reading of information in different seismic attribute data bodies (SEGY format files) includes: extraction of seismic trace data, reading of its (X, Y, Z) coordinates; reading of the required seismic attribute data body; coordinate conversion of seismic attribute coordinates according to the coordinates read by the seismic trace, converting the (X, Y) series of seismic attributes into (XLINE, INLINE) series in the seismic body.

[0098] Step A2, more than ten seismic attributes are selected by correlation analysis method, and four seismic attributes are selected as high-frequency sequence layering prediction basis by combining with the well seismic profile, the four selected seismic attribute maps are shown in FIGS. 4 to 7, and the four seismic attributes are: quadrature amplitude (Quar), trace AGC (TAGC), relative acoustic impedance (RAI) and RMS amplitude (RMS). At the same time, the relationship between the ground elevation in the seismic attribute data and the logging depth of the compensation core is balanced, and the processed seismic attribute data is stored in the condition information library to provide data support for the learning samples of the neural network model. At this time, the condition information library contains the planar position data, seismic attribute data and target prediction block of each seismic trace sample. The top and bottom surface depth and position data of the seismic trace of the condition well point are extracted, and the corresponding relationship between the high-frequency sequence layering data and the extracted seismic attribute data is established, different layers are numbered, the layer coding corresponding to each sample point is obtained according to the existing high-frequency sequence layering information and the depth information of the sample point, and the layer coding sequence is used as the sample label of the corresponding sample point. The coding of the five layers of the target prediction block is 1, 2, 3, 4 and 5, and the layer label setting is shown in FIG. 8. In FIG. 8, the codes of the respective sublayers of the layers of the target prediction block are 1, 2, 3, 4 and 5, and attributes A to D represent four seismic attributes. Because the extracted seismic attribute values are very discrete in the horizontal coordinate, in order to make the sequence layer prediction model better extract the internal relationship between the seismic attribute and the high-frequency sequence layering data, the seismic attribute values are normalized.

[0099] Step A3, based on the characteristics of the high-frequency sequence layering data and the seismic attribute data of the known condition well, an initial sequence prediction neural network based on BiLSTM+CRF is established, the initial sequence prediction neural network based on BiLSTM+CRF is trained through the learning samples, and an initial state sequence layer prediction model is obtained. The operation example of the model is shown in FIG. 9. The sequence layer prediction model is saved in the root directory of the program file in h5 format, and the model can be directly called when the model is used to predict new data next time.

[0100] Step A4, according to the top and bottom surface data in the condition information base, search for the top and bottom surface data of the seismic trace in the vicinity of the known condition well, extract the seismic attribute data between the top and bottom surfaces of the target seismic trace in different seismic attribute data bodies according to the top and bottom surface information and the seismic trace position data in the top and bottom surface data. Take the seismic attribute data of each point of the seismic trace and the depth information of each point as the input data of the sequence stratigraphic position prediction model, and use the above-mentioned saved sequence stratigraphic position prediction model to predict the high-frequency sequence stratigraphic layer of the point of the seismic trace in the vicinity of the condition well. Compare and analyze the prediction results with the high-frequency sequence stratigraphic layer data of the seismic trace of the known condition well, evaluate whether the prediction results are ideal according to the high-frequency sequence stratigraphic layer data of the seismic trace of the known condition well, through comparison, if it is found that the prediction results of the seismic trace in the vicinity of the known condition well and the high-frequency sequence stratigraphic layer data of the seismic trace of the condition well appear to have a large difference in high-frequency sequence stratigraphic layer proportion or unreasonable sequence prediction phenomenon, then the constraint conditions of CRF layer need to be manually corrected or adjusted to reach the standard of basically consistent with the high-frequency sequence stratigraphic layer information of the seismic trace beside the known condition well. The high-frequency sequence stratigraphic layer prediction results of the seismic attribute in the vicinity of the condition well and the corresponding position information are implanted into the condition information base according to the format required by the condition information base. Update the information of the newly implanted condition information base to the learning samples of the sequence stratigraphic position prediction model, to establish more corresponding relationship between the seismic attribute and the high-frequency sequence stratigraphic layer information, to better predict the high-frequency sequence stratigraphic layer condition of the next adjacent seismic trace, until the high-frequency sequence stratigraphic layer prediction results and the corresponding position information of all selected adjacent seismic traces in the preliminary delineated prediction range are implanted into the condition information base. It needs to be emphasized that the selection of the capacity of the condition information base depends on the size of the delineated prediction range radius (the radius of the circular area), the size of the target prediction block and the computer computing power. The selection of the prediction sample point depends on the size of the target prediction block, the required sample seismic trace density to reach the structural level and the computer computing power. The subsequent gradually increased conditions are executed according to the above flow, the range of the prediction sample point is delineated in a certain area or direction, the range of the prediction sample point is moved on the target prediction block according to the established rules, the high-frequency sequence stratigraphic layer prediction operation of the seismic attribute in the vicinity of the known condition well is repeated, until the prediction operation traverses the entire target prediction block, at this time, all selected seismic traces in the entire target prediction block have experienced the prediction process, finally realize the conditional cyclic update and learning sample sequential expansion of the seismic high-frequency sequence stratigraphic layer information. And in the sequential simulation, define the information base, the information base as a resident fixed sample point in the learning sample, no matter where the center of the prediction range radius is located, the information base is always as a sample point in the learning sample, the prediction process is shown in FIG. 11.In Figure 11, 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.

[0101] Step A5: Based on the high-frequency sequence partitioning structure of the known conditions wells of the target prediction block, the Softmax classification network is selected to construct the classification model, and a Softmax classifier suitable for this target prediction block is obtained. The working principle of the Softmax classifier is shown in Figure 12.

[0102] Through the above steps, intelligent interpretation of high-frequency seismic sequences for a target prediction block with few wells was achieved using a limited amount of high-frequency sequence logging information. Figure 13 shows the seismic profiles identified by tracking five short-term cycle sequences within the target prediction block, and Figure 14 shows a plan view of six stratigraphic interfaces from top to bottom. Furthermore, the intelligent interpretation of the high-frequency sequence for the target prediction block achieved through the above steps took only tens of hours, while traditional manual seismic horizon tracking methods required nearly 500 km of three-dimensional seismic stratigraphy. 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.

[0103] Example 5

[0104] Referring to Figure 15, based on the above-described method embodiments, this embodiment of the invention also provides a seismic high-frequency sequence layering intelligent interpretation device 400 utilizing well logging information, including a first module 410 and a second module 420.

[0105] 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 determined by the sequential path into the current state sequence stratigraphic prediction model to obtain high-frequency sequence stratigraphic data for the determined seismic trace points. The second module 420 is used to determine whether there are still high-frequency sequence stratigraphic data for the seismic trace points in the target block that have not been obtained. If there are still high-frequency sequence stratigraphic data for the 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 to obtain the trained sequence stratigraphic prediction model. This allows the first module to perform high-frequency sequence stratigraphic prediction for the next determined seismic trace point determined by the sequential path. Otherwise, the high-frequency sequence stratigraphic prediction for the target block is completed.

[0106] The layer sequence hierarchical prediction model of the initial state is obtained by training an initial sequence prediction neural network by using learning samples constructed by the sequential simulation condition data of the initial state, the sequential simulation condition data of the initial state comprising well point seismic trace well logging high-frequency sequence hierarchical data and seismic attribute data of a known condition well in the target block, seismic attribute data of a non-known condition well in the target block, and seismic attribute data of a well point seismic trace of the known condition well in the target block having established a corresponding relationship; and the seismic attribute data is high-frequency sequence sensitive seismic attribute data.

[0107] In one specific embodiment, the seismic attribute data of the determined seismic trace point position associated with depth information is input into the layer sequence hierarchical prediction model of the current state to obtain high-frequency sequence hierarchical data of the determined seismic trace point position, comprising: determining a determined seismic trace point position of a well seismic trace of a known condition well in a target block to be subjected to high-frequency sequence hierarchical prediction according to a sequential path; determining hierarchical information of the determined seismic trace point position according to stratum top and bottom surface information obtained from seismic data of the target block; forming a seismic attribute sequence from seismic attribute data corresponding to each hierarchical layer in the determined seismic trace point position; and inputting the seismic attribute sequence associated with depth information of each hierarchical layer in the determined seismic trace point position into the layer sequence hierarchical prediction model of the current state to obtain high-frequency sequence hierarchical data of the determined seismic trace point position.

[0108] In one specific embodiment, the high-frequency sequence hierarchical data of the determined seismic trace point position is added to the sequential simulation condition data to update the sequential simulation condition data, comprising: establishing a corresponding relationship between seismic attribute data and high-frequency sequence hierarchical data of the determined seismic trace point position; and adding the seismic attribute data and high-frequency sequence hierarchical data of the determined seismic trace point position having established a corresponding relationship to the sequential simulation condition data to update the sequential simulation condition data, the updated sequential simulation condition data comprising seismic attribute data and high-frequency sequence hierarchical data having established a corresponding relationship, and seismic attribute data of a seismic trace in the target block for which corresponding high-frequency sequence hierarchical data has not been obtained.

[0109] In one specific embodiment, the layer sequence hierarchical prediction model of the initial state is obtained by training an initial sequence prediction neural network by using learning samples constructed by the sequential simulation condition data of the initial state, comprising: encoding each stratum in well point seismic trace well logging high-frequency sequence hierarchical data of a known condition well in a target block to obtain stratum encoding sequences of each point position of the well point seismic trace, and using the stratum encoding sequences of each point position as sample labels of seismic attribute data of each point position in a one-to-one correspondence; and training the initial sequence prediction neural network by using seismic attribute data of each point position of the well point seismic trace of the known condition well and the sample labels corresponding to each seismic attribute data to obtain the layer sequence hierarchical prediction model of the initial state after the training is completed.

[0110] In one specific embodiment, when training the initial sequence prediction neural network by using the seismic attribute data of each point of the well point seismic trace under the known condition and the sample label corresponding to each seismic attribute data, the seismic attribute data of each point is embedded with the planar position coding of the point.

[0111] In one specific embodiment, the seismic high-frequency sequence intelligent interpretation device using logging information further comprises a third module configured to obtain the high-frequency sequence division result of the target block by performing sequence classification on the high-frequency sequence layered data by using the constructed classification model.

[0112] In one specific embodiment, the sequence prediction neural network is a sequence prediction neural network added with a CRF layer.

[0113] In one specific embodiment, the sequence prediction neural network comprises a BiLSTM network and a CRF layer connected to the output end of the BiLSTM network.

[0114] In one specific embodiment, the seismic high-frequency sequence intelligent interpretation device using logging information further comprises a fourth module configured to perform geological rationality judgment on the high-frequency sequence layered data of the determined seismic trace point by using the high-frequency sequence layered data of the well point under the known condition after predicting the high-frequency sequence layered data of the determined seismic trace point, and if the high-frequency sequence layered data of the determined seismic trace point is not reasonable, the high-frequency sequence layered data of the determined seismic trace point is modified.

[0115] In one specific embodiment, the modification of the high-frequency sequence layered data of the determined seismic trace point comprises manual modification and adjustment of the constraint condition of the CRF layer.

[0116] In one specific embodiment, the point position predicted by the sequence layered prediction model of each state is limited in a circular region with the point on the sequential path as the center, and if there is an overlapping region between the circular regions, the point position in the overlapping region is not included in the point position in the circular region predicted later.

[0117] In one specific embodiment, the sequence layered prediction model of each state predicts the high-frequency sequence layered data of the point position where the initial center of the sequential path is located.

[0118] In another aspect, the embodiments of the present application also provide a machine readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned seismic high-frequency sequence intelligent interpretation method using logging information.

[0119] In another aspect, the embodiment of the present application further provides a computer device, which can be a terminal, and an internal structure diagram of the computer device can be shown in FIG. 16. The computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05 and a memory (not shown in the figure) connected through a system bus. The processor A01 of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes an 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 operating system B01 and the computer program B02 in the non-volatile storage medium A06. The network interface A02 of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor A01 to implement the above-mentioned method for intelligent interpretation of high-frequency sequence of seismic data using logging information. The display screen A04 of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device A05 of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0120] In one embodiment, the device for intelligent interpretation of high-frequency sequence of seismic data using logging information 400 provided by the present application can be implemented in the form of a computer program, which can run on a computer device as shown in FIG. 15. The memory of the computer device can store various program modules constituting the device for intelligent interpretation of high-frequency sequence of seismic data using logging information 400. The computer program constituted by various program modules enables the processor to execute the steps of the method for intelligent interpretation of high-frequency sequence of seismic data using logging information described in the specification.

[0121] The device embodiments described above are only schematic, and the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. they can be located in one place, or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement it without creative labor.

[0122] It should also be noted that the terms "comprising", "comprises" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent interpretation of high-frequency sequences in seismic data using well log information, characterized in that, The method comprises: inputting the seismic attribute data associated with depth information of the determined seismic trace point into the current state's sequence layering prediction model to obtain high-frequency sequence layering data of the determined seismic trace point, the determined seismic trace point being a well trace point of a known condition well in a target block determined according to a sequential path; determining whether there is still high-frequency sequence layering data of a seismic trace point not obtained in the target block, if yes, adding the high-frequency sequence layering data of the determined seismic trace point into the sequential simulation condition data to update the sequential simulation condition data, and using the updated sequential simulation condition data to construct a learning sample to train the current state's sequence layering prediction model to obtain a trained sequence layering prediction model, and then jumping to the previous step to perform high-frequency sequence layering prediction on a next determined seismic trace point determined according to the sequential path, otherwise, the high-frequency sequence layering prediction for the target block is completed; wherein the sequence layering prediction model in the initial state is obtained by training an initial sequence prediction neural network using a learning sample constructed from initial state's sequential simulation condition data, the initial state's sequential simulation condition data comprising logging high-frequency sequence layering data and seismic attribute data of a well trace of a known condition well in a target block having established a corresponding relationship, seismic attribute data of a well trace in the target block which is not a known condition well; the seismic attribute data being high-frequency sequence sensitive seismic attribute data. 2.The method of claim 1, wherein, The inputting the seismic attribute data associated with depth information of the determined seismic trace point into the current state's sequence layering prediction model to obtain high-frequency sequence layering data of the determined seismic trace point comprises: determining a determined seismic trace point of a well trace of a known condition well in a target block to be subjected to high-frequency sequence layering prediction according to a sequential path; determining layering information of the determined seismic trace point according to formation top and bottom surface information obtained from seismic data of the target block; composing a seismic attribute sequence from seismic attribute data corresponding to each layer in the determined seismic trace point; inputting the seismic attribute sequence associated with depth information of each layer in the determined seismic trace point into the current state's sequence layering prediction model to obtain high-frequency sequence layering data of the determined seismic trace point. 3.The method of claim 1, wherein, The adding the high-frequency sequence layering data of the determined seismic trace point into the sequential simulation condition data to update the sequential simulation condition data comprises: establishing a corresponding relationship between seismic attribute data and high-frequency sequence layering data of the determined seismic trace point; adding the seismic attribute data and high-frequency sequence layering data of the determined seismic trace point having established a corresponding relationship into the sequential simulation condition data to update the sequential simulation condition data, the updated sequential simulation condition data comprising the seismic attribute data and high-frequency sequence layering data having established a corresponding relationship, and seismic attribute data in the target block for which corresponding high-frequency sequence layering data has not been obtained. 4.The method of claim 1, wherein, The sequence layering prediction model in the initial state is obtained by training an initial sequence prediction neural network using a learning sample constructed from initial state's sequential simulation condition data, the initial state's sequential simulation condition data comprising logging high-frequency sequence layering data and seismic attribute data of a well trace of a known condition well in a target block having established a corresponding relationship, seismic attribute data of a well trace in the target block which is not a known condition well; the seismic attribute data being high-frequency sequence sensitive seismic attribute data. The strata in the well point seismic trace of the known condition well in the target block are encoded in the high-frequency sequence stratification data of the well point, to obtain the strata encoding sequence of each point of the known condition well well point seismic trace, and the strata encoding sequence of each point is correspondingly used as the sample label of the seismic attribute data of each point. The initial sequence prediction neural network is trained by using the seismic attribute data of each point of the known condition well well point seismic trace and the sample label corresponding to each seismic attribute data, and the initial state sequence stratification prediction model is obtained after the training is completed.

5. The method for intelligent interpretation of seismic high-frequency sequence using well logging information according to claim 2, characterized in that, When the seismic attribute sequence and the depth information of each stratification in the determined seismic trace point are associated and input into the current state sequence stratification prediction model, the seismic attribute sequence of the determined seismic trace point also embeds the planar position encoding of the determined seismic trace point.

6. The method for intelligent interpretation of seismic high-frequency sequence using well logging information according to claim 1, characterized in that, The method further comprises: After the sequence classification of the high-frequency sequence stratification data is performed by using the constructed classification model, the high-frequency sequence division result of the target block is obtained.

7. The method for intelligent interpretation of seismic high-frequency sequences using well logging information according to claim 1, characterized in that, The sequence prediction neural network is a sequence prediction neural network with a CRF layer. 8.The method of claim 7, wherein, The sequence prediction neural network comprises a BiLSTM network and a CRF layer connected to the output end of the BiLSTM network. 9.The method of claim 6, wherein, The classification model is a Softmax classifier. 10.The method of claim 7, wherein, The method further comprises: After the high-frequency sequence stratification data of the determined seismic trace point is predicted, the high-frequency sequence stratification data of the determined seismic trace point is judged for geological rationality by using the high-frequency sequence stratification data of the known condition well well point, and if it is not reasonable, the high-frequency sequence stratification data of the determined seismic trace point is modified. 11.The method of claim 10, wherein, The modification of the high-frequency sequence stratification data of the determined seismic trace point comprises: According to the unreasonable sequence of the determined seismic trace point reflected by the high-frequency sequence stratification data of the determined seismic trace point and the high-frequency sequence stratification data of the known condition well well point, a modification instruction is generated for artificially modifying the unreasonable sequence; Or; According to the unreasonable sequence of the determined seismic trace point reflected by the high-frequency sequence stratification data of the determined seismic trace point and the high-frequency sequence stratification data of the known condition well well point, a modification instruction is generated for adjusting the constraint condition of the CRF layer, updating the sequence stratification prediction model based on the adjusted constraint condition of the CRF layer, inputting the seismic attribute data associated with the depth information of the determined seismic trace point into the updated sequence stratification prediction model, and obtaining the modified high-frequency sequence stratification data of the determined seismic trace point.

12. The method for intelligent interpretation of seismic high-frequency sequences using well logging information according to claim 1, characterized in that, The seismic trace points predicted by the sequence stratification prediction model of each state are limited in a circular region with the point on the sequential path as the center, and if there is an overlapping region between the circular regions, the seismic trace points in the circular region predicted later do not include the seismic trace points in the overlapping region.

13. An apparatus for intelligent interpretation of seismic high-frequency sequences using well log information, characterized by, The device comprises: A first module is configured to input the seismic attribute data associated with the depth information of the determined seismic trace point of the well point seismic trace of the known condition well in the target block determined according to the sequential path into the current state sequence stratification prediction model, to obtain the high-frequency sequence stratification data of the determined seismic trace point. A second module is configured to input the high-frequency sequence stratification data of the determined seismic trace point into the sequence stratification prediction model of the next state, to obtain the high-frequency sequence stratification data of the determined seismic trace point of the next state. The second module is configured to determine whether high-frequency sequence layering data of seismic trace points in the target block is obtained, and if the high-frequency sequence layering data of the seismic trace points in the target block is not obtained, the high-frequency sequence layering data of the determined seismic trace points is added to the sequential simulation condition data to update the sequential simulation condition data, and the updated sequential simulation condition data is used to train the sequence layering prediction model of the current state by using a learning sample to obtain a trained sequence layering prediction model, so that the first module performs high-frequency sequence layering prediction on the next determined seismic trace point determined according to the sequential path, and otherwise, the high-frequency sequence layering prediction for the target block is completed. The sequence layering prediction model of the initial state is obtained by training the initial sequence prediction neural network by using the learning sample constructed by the sequential simulation condition data of the initial state, and the sequential simulation condition data of the initial state includes the logging high-frequency sequence layering data and the seismic attribute data of the well point seismic trace of the known condition well of the target block, and the seismic attribute data of the well point seismic trace of the non-known condition well in the target block; and the seismic attribute data is high-frequency sequence sensitive seismic attribute data.

14. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the method for intelligent interpretation of seismic high-frequency sequences by using logging information according to any one of claims 1 to 12.

15. A machine-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method for intelligent interpretation of seismic high-frequency sequences by using logging information according to any one of claims 1 to 12.

Images