Complex river shallow water delta reservoir configuration prediction method based on CNN (Convolutional Neural Network) algorithm

Abstract

The invention discloses a complex interlaced channel type shallow water delta reservoir configuration prediction method based on a CNN convolutional neural network algorithm. The method comprises the following steps: establishing a configuration unit sample data set; preferably selecting effective sample points; seismic attribute and spectrum feature extraction; preferably selecting attributes which are highly related to the configuration units; preferably selecting a machine learning algorithm; constructing a convolutional neural network prediction model; inputting seismic attributes into the trained network model, outputting a configuration unit prediction model, and predicting the configuration unit distribution of the well control-free area; and identifying and dividing a single split-flow river channel and the like. According to the method, the plane contact relation and the vertical combination style of the composite sand body are finely dissected, so that the fine description of the single river channel in the multi-stage stacked complex interlaced composite river channel is realized, the reservoir recognition precision is improved, the fine implementation of an offshore oilfield horizontal well and the efficient development of an oilfield are guided, and the development of the offshore oilfield horizontal well is facilitated. The result of the prediction method is not greatly different due to subjectivity of researchers, and the prediction result is more reasonable and reliable.

Description

Technical Field

[0001] The invention belongs to the technical field of oil and gas field exploration and development, and specifically relates to a method for predicting the complex interwoven river channel type shallow water delta reservoir configuration based on a CNN convolutional neural network algorithm. Background Art

[0002] The shallow water delta reservoir type of complex interwoven river channel in the lower section of the Ming Dynasty is an important part of the current domestic oil and gas reserves, and is the backbone of the stable oil production of offshore oil fields. This type of reservoir type oil fields are mostly developed by horizontal wells, but the well network of offshore oil fields is sparse, and the reservoir prediction in the area without well control is difficult. To achieve the efficient development of this type of oil field, the spatial distribution law of the reservoir configuration unit must be analyzed first. At present, for this type of reservoir prediction, the amplitude attribute analysis method based on seismic data can only reflect the changes in sand and mudstone properties and the morphology of the composite river channel, but the distribution of different types of configuration units with sedimentary genesis and the boundary of the river channel are not clearly described. Therefore, it is urgent to establish a technical method that can effectively describe the reservoir boundary and realize the prediction of reservoir configuration units in the area without well control to guide the fine implementation of horizontal wells and the efficient development of oil fields. Summary of the invention

[0003] The present invention is proposed to solve the problem in the prior art that it is difficult to identify the composite channel superposition pattern and plane single channel boundary of multi-stage complex interwoven channel type shallow water delta reservoir under the conditions of sparse well network in the early stage of offshore oil field development. Its purpose is to provide a method for predicting the configuration of complex interwoven channel type shallow water delta reservoir based on CNN convolutional neural network algorithm.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for predicting reservoir configuration of shallow delta in complex river channel based on CNN algorithm includes the following steps:

[0006] S1. Establishing a sample data set of configuration units;

[0007] S2, select effective sample points;

[0008] S3, extracting seismic attributes and spectral features;

[0009] S4, preferably attributes with strong correlation with configuration units;

[0010] S5. Select the convolutional neural network algorithm as the machine learning algorithm;

[0011] S6. Construct a convolutional neural network prediction model;

[0012] S7, inputting seismic attributes into the trained convolutional neural network prediction model, outputting a configuration unit prediction model, and predicting the distribution of configuration units in the area without well control;

[0013] S8. Identify and demarcate a single distributary channel.

[0014] In the above technical scheme, the specific method of establishing the configuration unit sample data set in step S1 is: collecting basic information of the research object, identifying and dividing the configuration units of all exploration wells in the work area according to the identification characteristics of the configuration units, and establishing sample points of the complex interwoven channel type shallow water delta reservoir configuration unit; the identification characteristics refer to the logging curve characteristics of the complex interwoven channel type shallow water delta reservoir configuration unit; the complex interwoven channel type shallow water delta reservoir configuration unit includes underwater distributary channels and sheet sands, the underwater distributary channel sand body has a large thickness, generally 6 meters to 12 meters, and its logging curve is box-shaped or bell-shaped, with typical homogeneous rhythm or positive rhythm characteristics; the sheet sand is thinner, with a single layer thickness of about 2 meters, and mostly shows thin-layer sand and mudstone alternating deposition characteristics, and the sheet sand logging curve is finger-shaped or serrated, and the rhythmic characteristics are not obvious.

[0015] In the above technical solution, the basic data of the research object includes high-precision three-dimensional seismic data, drilling data, coring data and logging data.

[0016] In the above technical solution, the specific method of selecting effective sample points in step S2 is:

[0017] S21, improve the signal-to-noise ratio of the three-dimensional seismic data volume in the work area through random and linear noise suppression technology;

[0018] S22. Broaden the frequency band of the 3D seismic data volume in the work area by using time-varying wavelet resolution enhancement technology;

[0019] S23, make synthetic seismic records for all exploration wells in the work area for well-seismic calibration, and select the points with matching well-seismic correspondence from all configuration unit sample points established in step S1 as valid sample points. The target layer in the target area is shallowly buried, and generally different configuration units correspond to different lithology and thickness, and have a certain correlation with different forms of logging curves and different characteristics of seismic attributes (for example, sandstone generally corresponds to the low value of GR curve and the bright spot area of ​​seismic attributes), so this property can be used to select sample points with matching well-seismic relationship.

[0020] In the above technical solution, the method for extracting seismic attributes and spectral features in step S3 is specifically as follows: extracting 10Hz to 70Hz seismic attributes of the target sand body, and then performing time-frequency analysis on the seismic data to extract its spectral features. The method for extracting spectral features includes generalized S transform, wavelet transform, etc., among which S transform is a classic time-frequency analysis method widely used in the industry, and its accuracy can meet the feature extraction work of the time-frequency spectrum required by the subsequent application.

[0021] In the above technical solution, the preferred method for the highly correlated attributes of the configuration unit in step S4 is specifically: extracting the seismic attributes and spectral characteristics of the well bypass, combining the selected effective sample points in step S2 with their corresponding various seismic attributes for correlation analysis, using the seismic attributes and spectral characteristics as input for machine learning, and using the corresponding configuration units as output to form a training sample set.

[0022] In the above technical solution, the method for constructing the convolutional neural network prediction model in step S6 is specifically: using the training sample set constructed in step S4 to train the convolutional neural network algorithm selected in step S5 to obtain a trained convolutional neural network model, and using the trained convolutional neural network model to characterize the mapping relationship between seismic attributes and configuration units.

[0023] In the above technical solution, the specific method of measuring the distribution of configuration units in the area without well control in step S7 is: inputting the seismic attributes into the convolutional neural network model trained in step S6, outputting the reservoir configuration unit prediction model, and using the output configuration unit prediction model to predict the distribution of configuration units in the area without well control.

[0024] In the above technical solution, the specific method of step S8 for identifying and dividing a single distributary channel is as follows: in the reservoir configuration unit prediction model established in step S7, the reservoir configuration unit prediction data body within the time window of the target sand body is extracted, and different colors represent different configuration unit types, and the contact interface of the single distributary channel sand body of different periods is identified, and the distribution range and contact relationship of the single distributary channel sand body are divided on the plane based on this.

[0025] The beneficial effects of the present invention are:

[0026] The present invention provides a method for predicting the reservoir configuration of a shallow-water delta of a complex interwoven channel type based on a convolutional neural network (CNN) algorithm. The method innovatively proposes to characterize the deterministic mapping relationship between seismic attributes and configuration units by constructing a convolutional neural network model. After neural network training and prediction, a configuration unit prediction model is output, and conventional seismic attributes are converted into a reservoir configuration unit prediction model. Geophysical information is endowed with geological information with a sedimentary genesis concept, and the prediction of the distribution law of reservoir configuration units in areas without well control is realized. The planar contact relationship and vertical combination style of the composite sand body are finely dissected, and the fine characterization of a single channel inside a composite river channel with multiple periods of superposition and complex interweaving is realized, which significantly improves the accuracy of reservoir recognition.

[0027] The prediction method of the present invention has a high degree of consistency in the prediction of configuration unit types and configuration unit boundaries, and the results are reliable. It can be used to guide pre-drilling well site optimization and improve the drilling rate of high-quality reservoirs. It can significantly improve the accuracy of pre-drilling reservoir prediction, better guide pre-drilling well site optimization and development well implementation, especially for horizontal well development of oil fields to improve the drilling rate of high-quality reservoirs has a good application effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of the method of the present invention;

[0029] Figure 2 It is a schematic cross-sectional diagram of a vertically stacked pattern of reservoir configuration units in Example 1 of the present invention;

[0030] Figure 3 It is a schematic diagram of the plane layout of the reservoir configuration unit in Example 1 of the present invention;

[0031] Figure 4 It is a schematic diagram of the development well verification of the reservoir configuration prediction results in Example 1 of the present invention (A is the minimum amplitude attribute diagram, A' is the 90° phase section diagram, B is the plane slice diagram of the reservoir configuration prediction model, and B' is the reservoir configuration prediction section diagram);

[0032] Figure 5 This is a layout diagram of a single diversion channel in Example 1 of the present invention.

[0033] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0035] Example 1

[0036] The specific implementation process of the complex interwoven river channel type shallow water delta reservoir configuration prediction method based on the CNN convolutional neural network algorithm of the present invention is as follows: Figure 1 As shown:

[0037] S1. Establishing sample data set of configuration units

[0038] The basic data of the research object are collected, multiple information is combined, different configuration units are divided according to the identification characteristics of the configuration units, and a sample data set of complex interwoven channel-type shallow-water delta reservoir configuration units is established.

[0039] Basic data include high-precision three-dimensional seismic data, drilling data, coring data, logging data, etc. These data lay a solid foundation for subsequent reservoir configuration prediction and analysis.

[0040] The research object selected in this embodiment is Bohai B Oilfield, which is located in the southeastern waters of Bohai Sea. The three-dimensional seismic data has good quality, clear attribute characteristic response, good continuity of seismic phase axis, and high reservoir resolution, which can meet the needs of geophysical and geological research. Bohai B Oilfield has a total of 13 drilling wells, including 1 core well, rich logging data, and complete conventional logging projects. Through core observation and rock electrical calibration, the type of configuration unit and the electrical characteristics of different configuration units are determined. The complex interwoven channel type shallow water delta reservoir configuration unit mainly includes underwater distributary channels and sheet sands. The underwater distributary channel sand body has a large thickness, generally 6 meters to 12 meters, and the logging curve is box-shaped or bell-shaped, with typical homogeneous rhythm or positive rhythm characteristics. The sheet sand is thin, with a single layer thickness of about 2 meters, mostly showing thin-layer sand and mudstone alternating deposition characteristics, and the logging curve is finger-shaped or serrated, and the rhythm characteristics are not obvious.

[0041] According to the characteristics of the well logging curves, the configuration units of the drilled sand bodies were identified and divided, and the configuration unit sample data set of each well was established. According to the electrical characteristics of different configuration units, the configuration units of 13 exploration wells in the main area were identified and divided, and the configuration unit sample data set was established. The configuration unit sample data set includes a total of 125 configuration unit sample points.

[0042] S2. Optimize effective sample points

[0043] The three-dimensional seismic data volume in the work area is subjected to denoising and frequency enhancement processing to improve the vertical resolution of the reservoir, and then the synthetic seismic records of the exploration well are calibrated, and then all the configuration unit sample points in step S1 are screened, and the configuration unit sample points with better matching of the well-seismic correspondence are selected as valid sample points.

[0044] In order to achieve a better match between seismic data and the configuration units divided on the well, it is necessary to improve the resolution of seismic data for thin layers. Therefore, it is necessary to first perform denoising and frequency-boosting operations on the data to improve the quality of seismic data in the study area. The signal-to-noise ratio of seismic data is improved through random and linear noise suppression technology. Secondly, the frequency band of seismic data of Bohai B Oilfield is broadened through time-varying wavelet resolution enhancement technology, the main frequency is increased from 30HZ to 40HZ, and the vertical resolution of the reservoir is increased from 10m to 5m. On this basis, artificial synthetic seismic records are made for the 13 exploration wells in step S1 for well-seismic calibration, and points with matching well-seismic correspondence are selected from 125 configuration unit sample points as valid sample points. After screening, there are 80 valid sample points of configuration units in the study area.

[0045] S3, seismic attributes and spectrum feature extraction

[0046] It is known that the frequency division attributes can better characterize reservoirs of different thicknesses. High-frequency information has a better response to thin reservoirs, and low-frequency information has a better response to thick reservoirs. Seismic attributes such as frequency division instantaneous frequency, frequency division instantaneous amplitude and frequency division instantaneous phase of 10Hz-70Hz are extracted for the target sand body. Secondly, time-frequency analysis is performed on the seismic data to extract its spectral characteristics.

[0047] For the processed seismic data of Bohai B Oilfield, 48 post-stack seismic attributes such as frequency division instantaneous frequency, frequency division instantaneous amplitude and frequency division instantaneous phase attributes, minimum amplitude attributes, and sweet spot attributes are extracted for the target layer segment. The two-dimensional time-frequency spectrum is extracted using S transform for the seismic data of the target layer segment.

[0048] S4. Prefer attributes with strong correlation with configuration units

[0049] The seismic attributes and spectral characteristics of the well bypass are extracted, and the correlation analysis is performed between the attributes of each layer and the configuration unit of the valid sample point selected in step S2.

[0050] Since attributes such as frequency and amplitude are more sensitive to reservoirs of different thicknesses, and different configuration units are highly correlated with reservoir thickness and its rock physical properties, frequency and amplitude attributes are highly sensitive to reservoir configuration units.

[0051] The 80 valid sample points selected in step S2 are combined with their corresponding seismic attributes for intersection sensitivity analysis.

[0052] For the Bohai B oilfield, the frequency-divided instantaneous amplitude, frequency-divided instantaneous frequency and spectrum feature map are selected as the input of machine learning, and the corresponding configuration units (distributary channels, sheet sands, etc.) are used as outputs to constitute the training sample set.

[0053] S5. Optimal machine learning algorithm

[0054] Different machine learning methods have certain effects on the prediction results.

[0055] The method of the present invention is result-oriented, uses geological thinking, and combines the complex interwoven channel-type shallow-water delta sedimentary pattern and reservoir development characteristics to guide the optimization of machine learning algorithms.

[0056] Random forest, genetic algorithm, DNN deep neural network (Deep Neural Network) and CNN convolutional neural network (Convolutional Neural Network) algorithms are selected to train and predict the training sample set in step S4 respectively. By comparing the prediction results of the training set by different algorithms, it is found that the output of the CNN convolutional neural network algorithm is more in line with the strip-like distribution characteristics of the shallow delta reservoir of the complex interwoven river type. Therefore, the CNN convolutional neural network (Convolutional Neural Network) algorithm is preferred as the machine learning algorithm to be used.

[0057] S6. Convolutional Neural Network (CNN) prediction model construction

[0058] The CNN convolutional neural network selected in step S5 is trained using the training sample set in step S4, and the trained CNN convolutional neural network model is used to characterize the mapping relationship between seismic attributes and configuration units.

[0059] The training sample set consists of 80 sample points. Each sample point takes three types of seismic attributes as input, and its corresponding configuration unit code is output. The trained network contains two hidden layers. The first layer contains 10 convolution kernels, and the second layer contains 3 convolution kernels. The number of iterations is 50.

[0060] S7. Input seismic attributes into the trained network model, output the reservoir configuration unit prediction model, and predict the distribution of configuration units in the area without well control.

[0061] The seismic attributes are input into the trained CNN convolutional neural network model, and the configuration unit prediction model is output to predict the distribution of configuration units in the area without well control.

[0062] Different colors of the prediction results represent different configuration units, and the depth of the color represents the differences in reservoir quality such as thickness and scale of different configuration units.

[0063] The reservoir configuration prediction technology of the present invention transforms the conventional 3D seismic data of Bohai B Oilfield into a reasonable and reliable 3D prediction data volume of reservoir configuration units, which can clearly depict the superposition pattern of the composite sand body in the target layer on the section (such as Figure 2 The distribution range and change interface of different configuration units can be predicted on the plane (as shown in Figure 3 As shown in the figure, within the same configuration unit, the variation trend of good, medium and poor reservoir quality can be identified according to the change of color depth. This method can well characterize the spatial heterogeneity of the reservoir.

[0064] This embodiment is described in detail by taking the main target sand body B-1383 of Bohai B Oilfield as an example. There are 6 horizontal wells and 4 directional wells in the sand body, and the reservoir configuration prediction model is verified by actual drilling. The reservoir thickness encountered by 2 directional wells is 8m-10m, and the logging interpretation permeability is 2000mD-2500mD. The corresponding logging curve shape is a homogeneous box type, indicating that the 2 directional wells are drilled with typical distributary channel deposits, which corresponds to the main position of the red distributary channel in the reservoir configuration prediction model. The other 2 directional wells are drilled with a reservoir thickness of 2m-4m, and the logging interpretation permeability is 600mD-800mD. The corresponding logging curve shape is finger-shaped, indicating that the 2 directional wells are drilled with thin sheet sand deposits, which corresponds to the yellow sheet sand position in the reservoir configuration prediction model. The type and thickness of the reservoir configuration unit and physical properties of the 4 directional wells are 100% consistent with the prediction results in the reservoir configuration prediction model.

[0065] The horizontal section length of the six horizontal wells in the main target sand body of Bohai B Oilfield is 350m-450m. According to the actual drilling situation of the horizontal wells, the boundaries of different configuration units in the reservoir configuration prediction model can be verified (such as Figure 4 As shown in Figure 2). Taking Well A6H as an example, the horizontal section of Well A6H encountered 15m oil layer, 4m mudstone and 371.5m oil layer in sequence from the landing point to the end of the horizontal section. On the conventional seismic attribute map, A6H is located within the scope of a single river channel, and the seismic axis reflection on the seismic section is relatively homogeneous. However, according to the reservoir configuration prediction model, the cross-well trajectory section and plane map both show the splicing and superposition relationship of two river channels. The horizontal section of A6H landed in a slightly smaller secondary distributary river channel, and after passing through the contact interface of the two river channels, it was mainly distributed in the main distributary river channel. Combined with the horizontal section logging curve analysis, it is believed that the river channel contact interface in the reservoir configuration prediction model corresponds to the 4m mudstone encountered after the horizontal section of A6H landed, proving that the response of the reservoir configuration prediction model to the reservoir boundary is reliable and can be used as a geological basis for pre-drilling well position optimization. The prediction coincidence rate of the configuration unit boundary of the horizontal section position of the six horizontal wells in the reservoir configuration prediction model reached 95%.

[0066] S8. Identify and divide a single diversion channel

[0067] In the reservoir configuration unit prediction model established in step S7, the reservoir configuration unit prediction data body within the time window of the target sand body is extracted, and the contact interface of the single distributary channel sand body of different periods is identified according to different colors representing different configuration unit types. Based on this, the distribution range and contact relationship of the single distributary channel sand body are divided on the plane, such as Figure 5 shown.

[0068] This example takes the main target sand body B-1383 of Bohai Oilfield B as an example for detailed description. The plane slice diagram of the predicted reservoir architecture of the B-1383 sand body is extracted, and combined with the sedimentary pattern of the shallow water delta reservoir of the complex interweaving channel type, a single channel is divided by identifying the contact interface of different distributary channels. The single channel has continuity on the plane. According to this method, 14 distributary channels are identified on the plane of the B-1383 sand body.

[0069] When predicting the reservoir configuration of a shallow water delta with complex interwoven river channels under sparse well network conditions in the early stage of offshore oilfield development, the present invention innovatively proposes to construct a convolutional neural network model to characterize the deterministic mapping relationship between seismic attributes and configuration units in combination with the analysis of the reservoir configuration units that have been drilled. After neural network training and prediction, the configuration unit prediction model is output, and the conventional seismic attributes are converted into a reservoir configuration unit prediction model, realizing the reservoir configuration prediction under the phase control constraint of the non-well controlled area, and finely dissecting the planar contact relationship and vertical combination style of the composite sand body, realizing the fine characterization of a single channel inside a composite river channel with multiple superimposed and complex interweaving periods, and significantly improving the accuracy of reservoir recognition. The prediction method has a high degree of consistency in the prediction of the configuration unit type and the configuration unit boundary, and the results are reliable, which can be used to guide the optimization of well positions before drilling and improve the drilling rate of high-quality reservoirs. Most of the existing methods reflect the changes in sandstone and mudstone properties and the composite river channel morphology. It is rare to establish a deterministic relationship between seismic attributes and configuration units through a convolutional neural network training model, and then clearly identify and characterize the distribution of different types of configuration units with sedimentary genesis and the boundaries of a single river channel. The evaluation results of the present invention do not vary greatly due to the subjectivity of researchers, and the prediction results are more reasonable and reliable.

[0070] The applicant declares that the above is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention are within the protection scope and disclosure scope of the present invention.

Claims

1. A method for predicting reservoir configuration of shallow delta in complex river channel based on CNN algorithm, characterized by: The following steps are involved: S1. Establishing a sample data set of configuration units; S2, select effective sample points; S3, extracting seismic attributes and spectral features; S4, preferably attributes with strong correlation with configuration units; S5. Select the convolutional neural network algorithm as the machine learning algorithm; S6. Construct a convolutional neural network prediction model; S7, inputting seismic attributes into the trained convolutional neural network prediction model, outputting a configuration unit prediction model, and predicting the distribution of configuration units in the area without well control; S8. Identify and demarcate a single distributary channel.

2. The method for predicting reservoir configuration of a complex river shallow delta based on CNN algorithm according to claim 1 is characterized by: The specific method of establishing the configuration unit sample data set in step S1 is: collecting basic information of the research object, identifying and dividing the configuration units of all exploration wells in the work area according to the identification characteristics of the configuration units, and establishing a complex interwoven river channel type shallow water delta reservoir configuration unit sample set.

3. The method for predicting reservoir configuration of a complex river shallow delta based on CNN algorithm according to claim 2 is characterized by: The basic data of the research object include high-precision three-dimensional seismic data, drilling data, coring data and logging data.

4. The method for predicting reservoir configuration of a complex river shallow delta based on CNN algorithm according to claim 1, characterized in that: The specific method of selecting effective sample points in step S2 is: S21, improving the signal-to-noise ratio of the high-precision three-dimensional seismic data in the basic data of the research object collected in step S1 by using random and linear noise suppression technology; S22, broadening the frequency band of the high-precision three-dimensional seismic data processed in step S21 by using a time-varying wavelet resolution enhancement technique; S23, making artificial synthetic seismic records for all exploration wells in the work area to perform well-seismic calibration, and selecting points with matching well-seismic correspondence from the reservoir configuration unit sample set established in step S1 as valid sample points.

5. The method for predicting reservoir configuration of a shallow delta in a complex river channel based on a CNN algorithm according to claim 1, characterized in that: The method for extracting seismic attributes and spectral features in step S3 is specifically as follows: extracting 10Hz to 70Hz seismic attributes of the target sand body respectively, and then performing time-frequency analysis on the high-precision three-dimensional seismic data in the basic data of the research object collected in step S1 to extract its spectral features.

6. The method for predicting reservoir configuration of a complex river shallow delta based on CNN algorithm according to claim 1, characterized in that: The preferred method for selecting attributes with strong correlation of the configuration units in step S4 is specifically as follows: extracting the seismic attributes and spectral characteristics of the well bypass, combining the configuration units of the effective sample points selected in step S2 with their corresponding various seismic attributes for correlation analysis, using the strongly correlated frequency division instantaneous amplitude, frequency division instantaneous frequency and spectral feature diagram as input for machine learning, and using the corresponding configuration units as output to form a training sample set.

7. The method for predicting reservoir configuration of a complex river shallow delta based on CNN algorithm according to claim 1, characterized in that: The method for constructing the convolutional neural network prediction model in step S6 is specifically as follows: using the training sample set constructed in step S4 to train the convolutional neural network algorithm selected in step S5 to obtain a trained convolutional neural network model, and using the trained convolutional neural network model to characterize the mapping relationship between seismic attributes and configuration units.

8. The method for predicting reservoir configuration of a complex river shallow delta based on CNN algorithm according to claim 1, characterized in that: The specific method of measuring the distribution of configuration units in the area without well control in step S7 is: inputting seismic attributes into the convolutional neural network model trained in step S6, outputting a reservoir configuration unit prediction model, and using the output configuration unit prediction model to predict the distribution of configuration units in the area without well control.

9. The method for predicting reservoir configuration of a shallow delta in a complex river channel based on a CNN algorithm according to claim 1, characterized in that: The specific method for identifying and dividing a single distributary channel in step S8 is as follows: in the reservoir configuration unit prediction model established in step S7, the reservoir configuration unit prediction data body within the time window of the target sand body is extracted, and the contact interface of the single distributary channel sand body of different periods is identified according to different colors representing different configuration unit types, and the distribution range and contact relationship of the single distributary channel sand body are divided on the plane based on this.

Images