Well-seismic combined intelligent stratigraphic contrast method

Abstract

The invention provides a well-seismic combined intelligent stratigraphic contrast method. The well-seismic combined intelligent stratigraphic contrast method comprises the following steps: step 1, making a logging curve sample set; 2, encoding and decoding the original data, and constructing a deep learning framework; step 3, performing different proportion data set experiments and error analysis; and 4, optimizing and correcting an intelligent stratigraphic division result to obtain final horizon division, and realizing well-to-seismic joint intelligent stratigraphic comparison. According to the well-seismic combined intelligent stratigraphic comparison method, the result is analyzed and optimized by utilizing a well-seismic combined method, so that accurate division of geological stratification can be realized, the operation efficiency can be improved, and the research period can be shortened.

Description

Technical Field

[0001] The invention relates to the technical field of fine stratum research in oil reservoir engineering, and in particular to a method for combining well-seismic and intelligent stratum comparison. Background Art

[0002] The division of stratigraphic units is an indispensable and important step in oil and gas exploration and development. Well logging data has the characteristics of good continuity and high resolution, which is of great significance for identifying lithology, dividing strata, reservoirs and oil, gas and water logging phases. For a long time, in actual work, logging engineers have compared the responses of various curves at different depths in the logging data based on actual experience, and compared the correlation between the curves and the mutual dependence of the curves in different ranges at each position, thereby determining the formation attributes corresponding to different depth fragments. This method of manual stratification is relatively primitive, and the results of stratigraphic division will be inconsistent due to the differences in the logging curves used by different logging engineers to divide the strata or the differences in the standards for dividing the strata, thereby affecting the accuracy of the stratigraphic division. Therefore, it is of great significance to study a reasonable and effective stratigraphic unit automatic stratification method.

[0003] In the Chinese patent application with application number: CN201611242605.7, a method and device for stratigraphic correlation is involved, wherein the method includes: obtaining stratigraphic stratification of outcrops reflecting stratigraphic deposition in the study area, and determining the field standard section stratification of the study area according to the stratigraphic stratification of the outcrops; obtaining stratigraphic stratification of a designated well in the study area and stratigraphic correlation results between the stratigraphic stratification of the designated well and the field standard section stratification; obtaining logging data in the study area, and obtaining stratigraphic correlation results of the designated well and the remaining wells in the study area according to the logging data; obtaining stratigraphic correlation results in the study area according to the stratigraphic correlation results of the wells in the study area and the field standard section stratification. This application embodiment can conveniently and quickly obtain a fine correlation of the stratigraphic formations in the region.

[0004] In the Chinese patent application with application number: CN202011415038.7, a method for fine stratigraphic comparison of complex fault blocks in the later stage of development based on alternating well and seismic data is involved. It includes: establishing a preliminary structural model; conducting preliminary stratigraphic comparison with the help of regional and local markers; conducting fine stratigraphic calibration of well and seismic data, tracking stratigraphic positions in the entire area, using seismic attributes to guide the identification and description of low-order faults, and establishing a fine structural model; combining well and seismic data, conducting fine stratigraphic comparison of alternating well and seismic data, and resolving well and seismic contradictions; conducting fine closure interpretation of the entire area, and drawing structural maps; drawing small layer plans, and checking stratigraphic comparison data. If there are problems, the stratification or structure is adjusted to resolve the contradiction between oil and water development. This invention method solves the problems of difficult well comparison and serious well and seismic contradictions in complex fault blocks with high sand-to-land ratio and no marker layers, and divides the strata more finely and accurately, and clarifies the distribution law of remaining oil in complex fault blocks.

[0005] In the Chinese patent application with application number: CN202110525735.6, a stratigraphic division and comparison method based on regional structural analysis is involved, and the stratigraphic division and comparison method based on regional structural analysis includes: step 1, combining well and seismic data to determine the range of stratigraphic segment attribution; step 2, clarifying the local and regional structural evolution relationship horizontally; step 3, implementing the residual stratigraphic top and bottom contact relationship vertically; step 4, clarifying the attribution of residual stratigraphic sub-segments at different structural positions; step 5: performing fine stratigraphic division and comparison based on rock electrical signs. The stratigraphic division and comparison method based on regional structural analysis implements the stratigraphic overstripping relationship through regional structural analysis, and then performs fine comparative division of the strata. The stratigraphic division and comparison method based on regional structural analysis is scientific and reasonable, effectively reduces multi-solution, and provides a reliable working basis for oil and gas exploration and scientific research.

[0006] In the Chinese patent application with application number: CN201911043239.6, a method for modeling a stratigraphic sequence framework combined with well-seismic is involved. The method includes: step 1, using well logging data to divide the stratigraphic order according to the research purpose of the work area based on the key well sequence division; step 2, using different time-frequency analysis methods to perform time-frequency analysis on the seismic traces near the well; step 3, comparing the results obtained by different time-frequency analysis methods for the seismic traces near the well with the results of the well logging sequence analysis, and selecting the best results of the time-frequency analysis methods for sequences of different orders; step 4, using the selected best time-frequency analysis method to perform time-frequency analysis on the seismic data, and establishing a well-seismic combined sequence framework with the best well-seismic match from the well. The method for modeling a stratigraphic sequence framework combined with well-seismic improves the applicability of different time-frequency analysis methods, and establishes a more accurate well-seismic relationship and a three-dimensional spatial sequence stratigraphic framework.

[0007] The above existing technologies are all significantly different from the present invention and fail to solve the technical problem we want to solve. For this reason, we have invented a new method combining well-seismic and intelligent formation comparison. Summary of the invention

[0008] The purpose of the present invention is to provide a method for combining well-seismic and intelligent stratum comparison which can achieve accurate division of geological strata and improve computational efficiency.

[0009] The object of the present invention can be achieved by the following technical measures: a method for combining well-seismic and intelligent formation comparison, the method for combining well-seismic and intelligent formation comparison comprising:

[0010] Step 1, preparing a logging curve sample set;

[0011] Step 2: Encode and decode the original data and build a deep learning framework;

[0012] Step 3: Conduct experiments on data sets of different proportions and perform error analysis;

[0013] Step 4: Optimize and correct the intelligent stratigraphic division results to obtain the final stratigraphic division, and realize the intelligent stratigraphic comparison of well-seismic combined.

[0014] The purpose of the present invention can also be achieved by the following technical measures:

[0015] In step 1, a sample set for intelligent automatic division and comparison of formations is prepared. The original logging signal can highlight the local features of the curve at different depths after wavelet transformation.

[0016] In step 1, the natural potential curve SP and microelectrode curves ML1 and ML2, which are more sensitive to lithology, are selected as the reference curves for intelligent formation division and comparison, and then two types of high-frequency signals d2 and d4 processed by wavelet transform based on the ML2 logging curve are added to the sample set.

[0017] In step 2, the original data is encoded and decoded through the SegNet network to build a deep learning framework.

[0018] In step 2, the encoding network and decoding network functions in the SegNet network are applied to extract and process the feature information of the input data image, upsample the input data, and restore the original input data image.

[0019] In step 2, the SegNet network realizes the classification of the target body in an end-to-end manner. The SegNet network structure consists of two parts: the encoding network and the decoding network. The encoding network mainly extracts the feature information of the input data image and continuously reduces the scale of the input data; decoding is the reverse process of encoding, which upsamples the input data, restores the original input data image, and finally determines the accurate position of the logging layer line, realizing the intelligent model for high-precision intelligent formation comparison.

[0020] In step 3, experiments and error analysis are performed on data sets of different proportions. All samples are trained based on the SegNet model. The prediction results of the target layer of a single well are compared and visualized with the original stratified data, and the accuracy data is calculated to analyze and evaluate the results.

[0021] In step 4, the initial model of block wave impedance inversion is obtained by random optimization inversion method, and the intelligent stratigraphic division result is optimized and corrected to obtain the final stratigraphic division, thus realizing the intelligent stratigraphic comparison of well and seismic.

[0022] In step 4, when the well-seismic data are combined with the optimal layer division results, the initial model of block wave impedance inversion is obtained by the random optimization inversion method, and each sample point is used as a pseudo well.

[0023] In step 4, the sample set is selected for training, all data sets are trained and blind wells are predicted, the results of intelligent stratigraphic division are calculated, the error of the stratigraphic plane is calculated and interpolated according to the manual stratification of well logging in the work area, and finally the intelligent stratigraphic division results are optimized and corrected with the well stratigraphic position as the constraint condition to obtain the final stratigraphic division.

[0024] The method of combining well-seismic data with intelligent stratum comparison in the present invention uses the natural potential curve and the microelectrode curve as the reference curves for intelligent stratum division and comparison, and processes the curves at different scales and with multiple resolutions through the coif5 wavelet transform method, so as to more accurately identify a variety of continuously changing periodic information; the SegNet neural network learning method is used for intelligent comparison, and the well-seismic combination method is used to analyze and optimize the results, which can not only realize the accurate division of geological strata, but also improve the computing efficiency and shorten the research cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of a specific embodiment of the method of combining well-seismic and intelligent formation comparison of the present invention;

[0026] Figure 2 A decomposition diagram of a well logging curve in a specific embodiment of the present invention obtained by wavelet transformation;

[0027] Figure 3 A schematic diagram of the SegNet network structure used in a specific embodiment of the present invention;

[0028] Figure 4 A comparison diagram of formation division results of a single well in a specific embodiment of the present invention;

[0029] Figure 5 A comparison diagram of stratigraphic division results of a multi-well well-connected profile in a specific embodiment of the present invention;

[0030] Figure 6 It is a schematic diagram of an initial model of block wave impedance inversion in a specific embodiment of the present invention;

[0031] Figure 7 This is a comparison chart between intelligent prediction results and manual interpretation results in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0032] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0033] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0034] The method of combining well-seismic and intelligent formation comparison of the present invention comprises:

[0035] Step 1: The sample set for formation intelligent automatic division and comparison is prepared. The original logging signal can highlight the local features of the curve at different depths after wavelet transformation;

[0036] Step 2: Encode and decode the original data through the SegNet network to build a deep learning framework;

[0037] Step 3: Experiments and error analysis of different scale data sets. All samples are trained based on the SegNet model. The prediction results of the target layer of a single well are compared and visualized with the original stratified data, and the accuracy data is calculated to analyze and evaluate the results.

[0038] Step 4: The initial model of block wave impedance inversion is obtained through the random optimization inversion method, and the intelligent stratigraphic division results are optimized and corrected to obtain the final stratigraphic division, thus realizing the intelligent stratigraphic comparison of well-seismic joint.

[0039] In step 1, the natural potential curve (SP) and microelectrode curves (ML1, ML2) that are more sensitive to lithology are selected as the benchmark curves for intelligent formation division and comparison. Then, two types of high-frequency signals d2 and d4 based on the ML2 logging curve after wavelet transform processing are added to the sample set.

[0040] In step 2, the encoding network and decoding network functions in the SegNet network are applied to extract and process the feature information of the input data image, upsample the input data, and restore the original input data image.

[0041] In step 3, all samples are trained based on the SegNet model, the prediction results of the target layer of a single well are compared and visualized with the original stratification data, and the accuracy data is calculated to analyze and evaluate the results.

[0042] In step 4, the initial model of block wave impedance inversion is obtained by the random optimization inversion method, and finally the intelligent stratigraphic division result is optimized and corrected with the well position as the constraint condition to obtain the final stratigraphic division.

[0043] This well-seismic combined intelligent stratigraphic comparison method extracts the overall and local features of complex logging curves, builds a deep learning framework based on the features, collects and processes data to perform intelligent stratigraphic division of the wells in the entire block, evaluates the accuracy of the division results, and selects the final stratigraphic division results through the combination of well-seismic and well-logging methods.

[0044] The following are several specific embodiments of the present invention.

[0045] Example 1

[0046] In a specific embodiment 1 of the present invention, Figure 1 As shown, Figure 1 The flowchart of the method of combining well-seismic analysis with intelligent formation comparison of the present invention is as follows:

[0047] Step 101, preparation of logging curve sample set: selecting natural potential curve and microelectrode curve which are more sensitive to lithology as reference curves for formation division and comparison, adding two types of high-frequency signals processed by wavelet transform based on the curve, so that the neural network model can more efficiently extract and learn the overall and local characteristics of the logging curve, and thus obtain more accurate formation division results.

[0048] Step 102: Build a deep learning framework. As shown in 3, Figure 3This is a schematic diagram of the SegNet network structure used in a specific embodiment of the present invention. The network structure includes two parts: an encoding network and a decoding network. The encoding network is mainly composed of a series of layer groups based on nonlinear processing. Each layer group is composed of a convolution layer, a batch normalization layer, an excitation function layer, and a maximum pooling layer. It mainly extracts the feature information of the input data image through convolution, and continuously reduces the input data scale through pooling operations. Decoding is the reverse process of encoding, and is also composed of a series of layer groups based on nonlinear processing. Each layer group is composed of an upsampling layer, a convolution layer, a batch normalization layer, and an excitation function layer. The main purpose is to upsample the input data and restore the original input data image. From Figure 3 It can be seen that the SegNet network mainly realizes the classification of the target body in an end-to-end manner and is often used for pixel-level image segmentation tasks. The network structure consists of two parts: the encoding network and the decoding network. The encoding network mainly extracts the feature information of the input data image and continuously reduces the input data scale; decoding is the reverse process of encoding, which upsamples the input data, restores the original input data image, and finally determines the accurate position of the logging layer line, realizing the intelligent model for high-precision intelligent formation comparison.

[0049] Step 103, comparative evaluation of prediction results: all samples are trained based on the SegNet model, the prediction results of the target layer of a single well are compared and visualized with the manual stratification data, and the accuracy data is calculated to analyze and evaluate the results.

[0050] Step 104, combining well-seismic analysis with optimal stratum division results: the initial model of block wave impedance inversion is obtained by random optimization inversion method, and each sample point is used as a pseudo well. The sample set is selected for training, all data sets are trained and blind wells are predicted, the results of intelligent stratum division are calculated, the error of this stratum surface is calculated and interpolated according to the artificial stratification of well logging in the work area, and finally the intelligent stratum division results are optimized and corrected with the well stratum as the constraint condition to obtain the final stratum division.

[0051] Example 2

[0052] This embodiment is a unit in a block in Shengli. The block is a typical meandering river deposit of sand-packed mud type. The similarity of sand body development between wells is high. The error rate of manual comparison is large and the time consumption is long. Therefore, this well area is selected as the study area to carry out intelligent comparison technology research.

[0053] ① Production of logging curve sample set

[0054] The wavelet transform processing method selected is the coif5 wavelet method. Compared with other wavelet basis functions, the Coiflets wavelet has better orthogonality, tight time-frequency support, smoothness and is suitable for fast calculation of well logging data. Figure 2 As shown, Figure 2 This is a decomposition diagram of the logging curve in a specific embodiment of the present invention obtained through wavelet transformation; in (1), s represents the microelectrode curve, a represents the low-frequency signal, and d represents the high-frequency signal. The decomposition tree structure explains that after the first decomposition of the original signal, a low-frequency signal a1 and a high-frequency signal d1 are obtained. The second decomposition decomposes the low-frequency signal a1 into a low-frequency signal a2 and a high-frequency signal d2. The signal is decomposed 7 times in turn, and finally 7 low-frequency signals a1~a7 and 7 high-frequency signals d1~d7 are obtained. (2) is the complete decomposition mode of the logging signal, and the horizontal axis is the depth of the logging data, that is, the ordinal number of the sampling point. From Figure 2 It can be seen that the original logging signal can highlight the local characteristics of the curve at different depths after wavelet transformation, while reducing information redundancy, and ultimately provide more data features for intelligent stratigraphic division and comparison.

[0055] ②Build a deep learning framework

[0056] In combination with the data characteristics of the research area, the network is adjusted and a multi-task network structure is selected. That is, the types of loss functions are added to the network and optimized simultaneously during model training. Finally, the accurate position of the logging layer line is determined, and the intelligent model is used to perform high-precision intelligent stratigraphic comparison.

[0057] ③ Comparative evaluation of prediction results

[0058] Different proportions of data sets were prepared for the new sample sets. In the study, the study was divided into 7 groups according to the well data ratio of 90%, 80%, 70%, 40%, 20%, 10%, and 5% of the training samples, and the experimental wells were randomly selected as training and prediction samples. All samples were trained based on the SegNet model, and the prediction results of the single well target layer were compared with the original stratified data, and the accuracy data was calculated to analyze and evaluate the results (Table 1). Figure 5 This is a comparison chart between the intelligent division results and the manual division results on the well section along the provenance direction. The solid line is the manual interpretation result, and the dotted line is the artificial intelligence division result. The overall consistency rate is more than 86%. The differences in individual divisions are mainly due to differences in geological cognition.

[0059] Table 1. Evaluation table of different scale data sets and prediction results

[0060]

[0061]

[0062] ④Results of optimal layer division based on well-seismic combination

[0063] Based on existing research, the initial model of block impedance inversion is obtained by random optimization inversion method ( Figure 6), and each of its sample points is regarded as a pseudo well. The AC curve of the pseudo well and its high-frequency signal of the curve transformed by coif5 wavelet are selected to make a new sample set for training; all data sets are applied to the SegNet model for training. Based on the well logging stratigraphic division, the overall stratigraphic division results of the experimental block are obtained.

[0064] Example 3

[0065] In the specific embodiment 3 of the present invention, a Shahejie block oilfield in Shengli Oilfield is selected as the target block. The block is a dome anticline structure oil reservoir complicated by faults and is currently in a development stage with low oil production rate, medium production degree and medium to high water content.

[0066] ① Production of logging curve sample set

[0067] The wavelet transform processing method selected is the coif5 wavelet method. Compared with other wavelet basis functions, the Coiflets wavelet has better orthogonality, tight time-frequency support, smoothness, and is suitable for fast calculation of logging data. After the original logging signal is transformed by wavelet, it can highlight the local characteristics of the curve at different depths, providing more data features for intelligent formation division and comparison.

[0068] ②Build a deep learning framework

[0069] In combination with the data characteristics of the research area, the network is adjusted and a multi-task network structure is selected. That is, the types of loss functions are added to the network and optimized simultaneously during model training. Finally, the accurate position of the logging layer line is determined, and the intelligent model is used to perform high-precision intelligent stratigraphic comparison.

[0070] ③ Comparative evaluation of prediction results

[0071] Figure 4 The dashed line in the middle is the SegNet prediction result, the solid line is the manual interpretation result, and the dashed line is the model prediction result. The neural network made a large misjudgment in the division of the Ng74 sublayer. Further analysis revealed that there are two main factors: (1) The influence of data set production and its quality: Due to the differences in geological environment, the geological conditions in different regions are also reflected differently in the logging curves. However, the high similarity of the target layer sand and mud logging curve combination brings great challenges to the learning of multi-category data features; (2) The influence of differences in geological cognition: The river channel in the target work area changes frequently, and the vertical and lateral phases of the sand body change rapidly. Different division standards according to different geological conditions lead to misjudgment of the neural network model.

[0072] ④Results of optimal layer division based on well-seismic combination

[0073] The error of this layer is calculated according to the artificial stratification of well logging in the work area, and interpolation is performed based on the well point error calculation ( Figure 7), and finally the intelligent stratigraphic division results are optimized and corrected with the well positions as constraints to obtain the final stratigraphic division.

[0074] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0075] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. A method combining well-seismic analysis with intelligent formation comparison, characterized in that: The method of combining well-seismic and intelligent formation comparison includes: Step 1, preparing a logging curve sample set; Step 2: Encode and decode the original data and build a deep learning framework; Step 3: Conduct experiments on data sets of different proportions and perform error analysis; Step 4: Optimize and correct the intelligent stratigraphic division results to obtain the final stratigraphic division, and realize the intelligent stratigraphic comparison of well-seismic combined.

2. The method of combining well-seismic analysis with intelligent formation correlation according to claim 1, characterized in that: In step 1, a sample set for intelligent automatic division and comparison of formations is prepared. The original logging signal can highlight the local features of the curve at different depths after wavelet transformation.

3. The method of combining well-seismic analysis with intelligent formation correlation according to claim 2, characterized in that: In step 1, the natural potential curve SP and microelectrode curves ML1 and ML2, which are more sensitive to lithology, are selected as the reference curves for intelligent formation division and comparison, and then two types of high-frequency signals d2 and d4 processed by wavelet transform based on the ML2 logging curve are added to the sample set.

4. The method of combining well-seismic analysis with intelligent formation correlation according to claim 1, characterized in that: In step 2, the original data is encoded and decoded through the SegNet network to build a deep learning framework.

5. The method of combining well-seismic analysis with intelligent formation correlation according to claim 4 is characterized in that: In step 2, the encoding network and decoding network functions in the SegNet network are applied to extract and process the feature information of the input data image, upsample the input data, and restore the original input data image.

6. The method of combining well-seismic analysis with intelligent formation correlation according to claim 5, characterized in that: In step 2, the SegNet network realizes the classification of the target body in an end-to-end manner. The SegNet network structure consists of two parts: the encoding network and the decoding network. The encoding network mainly extracts the feature information of the input data image and continuously reduces the scale of the input data; decoding is the reverse process of encoding, which upsamples the input data, restores the original input data image, and finally determines the accurate position of the logging layer line, realizing the intelligent model for high-precision intelligent formation comparison.

7. The method of combining well-seismic analysis with intelligent formation correlation according to claim 1, characterized in that: In step 3, experiments and error analysis are performed on data sets of different proportions. All samples are trained based on the SegNet model. The prediction results of the target layer of a single well are compared and visualized with the original stratified data, and the accuracy data is calculated to analyze and evaluate the results.

8. The method of combining well-seismic analysis with intelligent formation correlation according to claim 1, characterized in that: In step 4, the initial model of block wave impedance inversion is obtained by random optimization inversion method, and the intelligent stratigraphic division result is optimized and corrected to obtain the final stratigraphic division, thus realizing the intelligent stratigraphic comparison of well and seismic.

9. The method of combining well-seismic analysis with intelligent formation correlation according to claim 8, characterized in that: In step 4, when the well-seismic data are combined with the optimal layer division results, the initial model of block wave impedance inversion is obtained by the random optimization inversion method, and each sample point is used as a pseudo well.

10. The method of combining well-seismic analysis with intelligent formation correlation according to claim 9, characterized in that: In step 4, the sample set is selected for training, all data sets are trained and blind wells are predicted, the results of intelligent stratigraphic division are calculated, the error of the stratigraphic plane is calculated and interpolated according to the manual stratification of well logging in the work area, and finally the intelligent stratigraphic division results are optimized and corrected with the well stratigraphic position as the constraint condition to obtain the final stratigraphic division.

Images