CN114167497B - SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label - Google Patents

SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label Download PDF

Info

Publication number
CN114167497B
CN114167497B CN202111369646.3A CN202111369646A CN114167497B CN 114167497 B CN114167497 B CN 114167497B CN 202111369646 A CN202111369646 A CN 202111369646A CN 114167497 B CN114167497 B CN 114167497B
Authority
CN
China
Prior art keywords
seismic
data
samples
attribute data
marked
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202111369646.3A
Other languages
Chinese (zh)
Other versions
CN114167497A (en
Inventor
张冬梅
殷鹏
李江
李洋
曹弘
汪校锋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Geosciences
Original Assignee
China University of Geosciences
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China University of Geosciences filed Critical China University of Geosciences
Priority to CN202111369646.3A priority Critical patent/CN114167497B/en
Publication of CN114167497A publication Critical patent/CN114167497A/en
Application granted granted Critical
Publication of CN114167497B publication Critical patent/CN114167497B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
    • G01V1/30—Analysis
    • G01V1/301—Analysis for determining seismic cross-sections or geostructures
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
    • G01V1/30—Analysis
    • G01V1/306—Analysis for determining physical properties of the subsurface, e.g. impedance, porosity or attenuation profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V2210/00—Details of seismic processing or analysis
    • G01V2210/60—Analysis
    • G01V2210/62—Physical property of subsurface
    • G01V2210/624—Reservoir parameters
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V2210/00—Details of seismic processing or analysis
    • G01V2210/60—Analysis
    • G01V2210/64—Geostructures, e.g. in 3D data cubes

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Remote Sensing (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Geophysics (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geology (AREA)
  • Environmental & Geological Engineering (AREA)
  • Acoustics & Sound (AREA)
  • Geophysics And Detection Of Objects (AREA)

Abstract

The invention relates to an SSL-CNN reservoir oil gas detection method based on a similarity measurement geological structure label, wherein the traditional reservoir oil gas detection method only considers the characteristics of a single seismic channel, and ignores the spatial structure information of a seismic body. According to the method, main control factors of reservoir geological structures and production dynamic data are combined, the main components are adopted to analyze and mine low-dimensional structural features of seismic attribute data, aiming at the problem of insufficient labeling samples, according to the spatial correlation of geological structures between adjacent seismic channels, a neighborhood similarity measurement algorithm is used for judging the homogeneity of the seismic channels adjacent to the samples for the first time, so that the training samples are fused with the geological spatial structural features of the data. In order to fully utilize a large number of unlabeled samples to improve algorithm detection performance, a fracture-cavity type oil reservoir oil gas detection model is further constructed based on pseudo labels fused with geological space structural features and combined with a semi-supervised convolutional neural network technology, and reservoir oil gas detection is achieved.

Description

SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label
Technical Field
The invention relates to the field of oil reservoir engineering, in particular to an SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure labels.
Background
In hydrocarbon reservoir evaluation, how to develop oil-bearing property predictions has become one of the most challenging hot topics in hydrocarbon exploration. The pore structure was explained by using log data in Yang Jinlin, 1998. Reservoir oil gas detection evaluation based on geophysical information interpretation is carried out in 2004 Li Zongjie and the like, and the oil gas detection of most reservoirs can be reasonably interpreted by the geophysical information-based interpretation method, but the oil gas detection recognition effect on low-permeability reservoirs, low-resistance reservoirs, thin layers and special lithology such as limestone, dolomite, mudstone cracks and the like is poor. In 2017 Wei Qianqian, a fuzzy clustering fusion analysis technology is utilized to establish nonlinear connection between fusion attributes and reservoir characteristics, and finally, prediction of a target layer oil gas enrichment region is realized. In 2019 Shi Wei and other methods for predicting a multi-attribute fusion reservoir by utilizing the multi-attribute fusion reservoir, firstly, a proper time window (P1 zone) is determined according to the oil reservoir characteristics of a carbolic industrial zone, then, correlation and attribute characteristic intensity comprehensive analysis are carried out on different types of attributes, 4 attributes which are more sensitive are selected and multi-element linear regression fitting is carried out, finally, a multi-attribute fusion attribute prediction plan is obtained, and the positions of favorable zones can be well matched.
In recent years, the application of machine learning technology in various aspects has attracted close attention, and many domestic students develop oil gas detection through a machine learning method, so that certain effects are achieved. However, the seismic attribute data has extremely strong nonlinear characteristics, has huge data volume and contains redundant information and noise, and the traditional method has weaker capability of expressing nonlinear structures and weaker processing capability on high-dimensional data.
In summary, although there is a certain knowledge in the development of fracture-cavity carbonate reservoirs in the current country, a relatively mature theory and a matching technology corresponding to the theory are not formed yet.
Disclosure of Invention
Aiming at the defects, the invention provides an SSL-CNN reservoir oil gas detection method based on a similarity measurement geological structure label.
In order to solve the technical problems, the invention adopts the following technical scheme:
SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label comprises the following steps:
Step 1, acquiring seismic attribute data, and selecting the seismic attribute data matched with the bottom hole coordinates of a production well to perform artificial sample marking to obtain the seismic attribute data containing artificial marked samples;
Step 2, performing expansion marking on the artificial marking sample in the seismic attribute data obtained in the previous step by adopting a gray correlation analysis technology to obtain the seismic attribute data containing the artificial marking sample and the expansion marking sample; taking the artificial marked sample and the expanded marked sample as marked samples to obtain seismic attribute data containing the marked samples;
Step 3, performing principal component analysis dimension reduction on the seismic attribute data of the extended marked sample containing the artificial marked sample to obtain the seismic attribute data containing the marked sample after dimension reduction;
Step 4, randomly selecting seismic attribute data containing marked samples after dimension reduction in a certain proportion as training samples, wherein unmarked samples in the training samples form an unmarked sample set S u, and marked samples form a marked sample set S l;
Step 5, training a classifier by using a marked sample set S l to obtain a classifier h i, wherein i is the iteration number, and the initial value of i is 1;
Step 6, selecting m training samples from the unlabeled sample set S u by adopting a random sampling strategy, sequentially sending the m training samples into a classifier h i, calculating the probability selection maximum value of each class of the training samples by the classifier h i, if the probability selection maximum value corresponding to a certain class is greater than a threshold value, giving the training samples to class labels of the corresponding class, and putting the obtained class labels into a pseudo-label sample set U p;
Step 7, adding 1 to the iteration number, training a classifier h j by using a mark sample set S l and a pseudo-tag sample set U p, randomly selecting n training samples from the pseudo-tag sample set U p, and classifying the n training samples by using a classifier h i and a classifier h j; the training samples with the same classification result of the two classifiers are removed from the pseudo tag sample set U p and added into the marked sample set S l, and the classifier h i is updated to h j;
Step 8, repeating the steps 6-7 until the iteration number reaches the highest iteration number or the number of samples in the marked sample set S l reaches the maximum threshold value, training the classifier h j in the last iteration by using the latest marked sample set S l to obtain a final classifier, and turning to the next step;
and 9, carrying out oil gas detection on the researched work area by utilizing a final classifier, and outputting an oil gas detection matrix of the researched work area.
Further, the step 1 specifically includes the following steps:
step 1.1, original seismic attribute data are read, various types of seismic attribute data related to oil gas are extracted, including energy half-decay time, root mean square amplitude and instantaneous frequency, normalization processing is carried out, and geological and fluid characteristics represented by various seismic attribute data are analyzed;
Step 1.2, reading production dynamic data, calculating accumulated oil production of a production well, marking the accumulated oil production of the production well as a high-yield well, wherein the accumulated oil production of the production well is between a low-yield threshold and the high-yield threshold and is lower than the low-yield threshold;
and 1.3, selecting seismic attribute data matched with the bottom hole coordinates of the production well, and marking samples.
Further, in the method for randomly selecting a certain proportion of seismic attribute data in the step 4, the seismic data are sequentially read according to the sequence from left to right and from top to bottom based on the d×d sliding window, and 20% of data blocks in the seismic attribute data are randomly selected as training samples.
Further, the method for performing expansion marking on the manually marked samples in the seismic attribute data by adopting the gray correlation analysis technology in the step2 comprises the following steps:
Step 2.1, reading seismic attribute data, setting a similarity coefficient threshold value and distinguishing an initial value of a coefficient;
Step 2.2, determining a parent sequence and a child sequence of analysis seismic attribute data, wherein the parent sequence is energy half-decay time data X 0＝[x0(1),x0(2),x0(3),...,x0(n)]T corresponding to a bottom hole coordinate of a marked sample, n represents a depth domain millisecond number, and the child sequence is seismic trace data (excluding a coordinate of the marked sample) of 8 neighbors of the bottom hole coordinate:
Wherein m is the number of subsequences of the seismic trace, X i is the ith subsequence (0 is less than i and less than m), and X i (k) is the k (1 is less than k and less than n) millisecond energy half-decay time value of the ith subsequence;
Step 2.3, calculating gray correlation coefficients of corresponding elements of the marked sample bottom coordinate 8 neighborhood subsequence and the parent sequence seismic attribute data, wherein the calculating method comprises the following steps:
Wherein x 0 (k) is the corresponding value of k milliseconds of the energy half-decay data corresponding to the parent sequence, x i (k) is the corresponding value of k milliseconds of the energy half-decay data of the ith subsequence, ρ is a resolution coefficient, 0 < ρ <1, and the default ρ value is 0.5;
Step 2.4, calculating the association degree between the bottom hole coordinates of the marked sample and the seismic channel data of the 8 neighborhood, and respectively calculating the average value of the association coefficient of each index of the seismic channel data (comparison sequence) in the 8 neighborhood and the corresponding element of the bottom hole coordinates so as to reflect the association relation between the adjacent seismic channel data and the seismic channel data of the bottom hole coordinates, wherein the method comprises the following steps:
wherein r (0,i) is the association degree between the bottom hole coordinates and the data of the ith neighborhood seismic trace, n is the millisecond number of the depth domain of the seismic trace, k is the kt millisecond (1 < k < n) of the seismic trace, and the geological space structures are more dissimilar when the difference between the association degrees is larger;
and 2.5, outputting the seismic trace data which are larger than the association threshold in the neighborhood, and marking the seismic trace data as an expansion marking sample.
The beneficial effects are that: the invention adopts the principal component analysis dimension reduction technology, the similarity measurement technology and the SSL-CNN fused with the geological space structural features, the fracture-cavity type oil reservoir has strong heterogeneity, the supervised machine learning algorithm marks the production well sample labels with high cost, the number of marked samples is seriously lacking, and the accuracy of the model is difficult to ensure. The traditional reservoir oil gas detection method only considers the characteristics of a single seismic channel, and ignores the space structure information of a seismic body. According to the method, main control factors and production dynamic data of reservoir geological structures are combined, low-dimensional structural features of seismic attribute data are mined by adopting principal component analysis, aiming at the problem of insufficient labeling samples, according to the spatial correlation of geological structures between adjacent seismic channels, a neighborhood similarity measurement algorithm is used for carrying out homogeneity judgment on the seismic channels adjacent to the samples for the first time, and a similarity measurement algorithm based on gray correlation is used for expanding training samples, so that the training samples are fused with the geological spatial structural features of the data; aiming at the problems that prior information is not utilized and noise is easy to interfere in the traditional unsupervised machine learning method, redundant information and noise of original data are removed through a dimension reduction method, classification accuracy is improved, and further a model is trained to achieve high-efficiency oil gas detection.
The invention will now be described in detail with reference to the drawings and examples.
Drawings
FIG. 1 is a flow chart of an SSL-CNN reservoir oil gas detection method based on similarity measure geologic structure labels of the invention;
FIG. 2 is a schematic representation of seismic attribute data markers;
FIG. 3 is a SSL-CNN network structure diagram fusing geologic spatial structural features;
FIG. 4 is a flow chart of a method for SSL-CNN reservoir hydrocarbon detection based on similarity metric geologic structure labels;
FIG. 5 is a TK744 well daily yield curve;
FIG. 6 is a TK744 well daily water cut curve;
FIG. 7 shows the results of TK744 well oil and gas detection;
FIG. 8 is a TK613 well daily yield curve;
FIG. 9 is a TK613 well daily moisture content curve;
FIG. 10 shows the results of TK613 well oil and gas detection.
Detailed Description
The principles and features of the present invention are described below with reference to the drawings, the examples are illustrated for the purpose of illustrating the invention and are not to be construed as limiting the scope of the invention.
The invention mainly comprises the following parts:
(1) Data preprocessing
1) Reading original seismic data, extracting various seismic attribute data related to oil gas based on the original seismic data, such as energy half-decay time, root mean square amplitude, instantaneous frequency and the like, carrying out normalization processing, and analyzing characteristics of geology, fluid and the like described by various seismic attribute data;
2) Reading production dynamic data, and calculating accumulated oil production of the production well, wherein the accumulated oil production is marked as a high-yield well, the accumulated oil production is marked as a medium-yield well, and the accumulated oil production is marked as a low-yield well, wherein the accumulated oil production is 3-10 ten thousand tons;
3) And selecting seismic attribute data matched with the bottom hole coordinates, and marking the samples.
(2) Training sample expansion module based on gray correlation analysis
The geologic structure of adjacent areas of seismic data has spatial structural similarity, and oil-water distribution characteristics near the marked samples tend to have strong correlation. And researching 8 neighborhood samples of 3*3 areas around the selection mark sample to perform similarity calculation, quantifying the similarity between the seismic trace where the mark sample is positioned and surrounding seismic trace waveform data by adopting a gray correlation analysis technology, and if the similarity is high, considering the seismic trace to belong to the same type of geological structure and reservoir body, and obtaining the spatial structure characteristics of the seismic data by calculation. The basic steps are as follows:
1) Reading seismic attribute data, setting a similarity threshold value and a resolution coefficient initial value;
2) Determining a parent sequence and a child sequence of the analysis seismic attribute data, wherein the parent sequence is energy half-decay time data X 0＝[x0(1),x0(2),x0(3),...,x0(n)]T corresponding to the bottom coordinates of a marked sample, n represents the millisecond number of a depth domain, and the child sequence is seismic trace data (excluding the coordinates of the marked sample) of 8 neighbors of the bottom coordinates:
Where m is the number of subsequences of the seismic trace, X i is the ith subsequence (0 < i < m), and X i (k) is the k (1 < k < n) millisecond energy half-decay time value of the ith subsequence.
3) The gray correlation coefficient of the corresponding element of the seismic attribute data of the marked sample bottom hole coordinate 8 neighborhood subsequence and the parent sequence is calculated by the following steps:
Wherein x 0 (k) is the corresponding value of k milliseconds of the energy half-decay data corresponding to the parent sequence, x i (k) is the corresponding value of k milliseconds of the energy half-decay data of the ith subsequence, ρ is a resolution coefficient, 0 < ρ <1, and the default ρ value is 0.5.
3) Calculating the association degree between the bottom hole coordinates of the marked sample and the seismic channel data of the 8 neighborhood, and respectively calculating the average value of the association coefficients of the corresponding elements of the bottom hole coordinates and the indexes of the seismic channel data (comparison sequences) of the 8 neighborhood so as to reflect the association relation between the adjacent seismic channel data and the seismic channel data of the bottom hole coordinates, wherein the method comprises the following steps:
wherein r (0,i) is the association degree between the bottom hole coordinates and the data of the ith neighborhood seismic trace, n is the millisecond number of the depth domain of the seismic trace, k is the kt millisecond (1 < k < n) of the seismic trace, and the geological space structures are more dissimilar when the difference between the association degrees is larger;
4) Outputting the seismic channel data in the neighborhood, which is greater than the association threshold, giving the seismic channel data the same label value as the corresponding bottom hole coordinates, and completing the fusion of the geological space structural features of the sample, as shown in fig. 2, so as to obtain the marked seismic attribute data.
(3) Seismic data dimension reduction module based on principal component analysis
The seismic data has higher data dimension, the problem of information redundancy exists, and the main components in the seismic data are extracted through dimension reduction processing, so that redundant information is removed. The original energy half-decay time data is seismic data with the size of x multiplied by y multiplied by n, wherein x is x/ines, y is in/ines, and n is the number of time domain sampling points. And obtaining data with the size of x multiplied by y multiplied by k after the main component analysis is used for reducing the dimension, wherein k is the number of features after the dimension reduction. The basic steps are as follows:
1) The marked seismic attribute data is input, and each trace of seismic data is described by a vector X i＝[xi(1),xi(2),xi(3),…,xi(n)]T.
Wherein X i is the ith seismic trace attribute data (0 < i < N), X i (k) is the ith seismic trace k (1 < k < N) millisecond seismic attribute value, and N is the total number of seismic traces N=x×y.
2) Calculating a mean vector u of the seismic trace sample data set N:
3) Centering all seismic trace samples:
4) Constructing a covariance matrix of the seismic channel sample data set after the centralization treatment:
5) Performing feature decomposition on the matrix V, and obtaining a feature value lambda i and a corresponding feature vector w i, and arranging the feature values lambda i in a descending order;
6) Taking the first d eigenvalues and corresponding eigenvectors as the basis of subspaces as d main components of the seismic trace data to be extracted according to the contribution rate;
7) And outputting the low-dimensional representation space characteristics of the seismic attribute data obtained by the principal component analysis algorithm.
(4) SSL-CNN oil gas detection semi-supervised classification algorithm module integrating geologic space structural features
And reading the low-dimensional space characteristics of the seismic attribute data, and further classifying the reduced-dimension data by utilizing an SSL-CNN algorithm fused with the geological space structural characteristics. The method comprises the following specific steps:
1) Initializing an SSL-CNN algorithm network structure fused with geological space structural features, and setting the maximum iteration number of network training, a well type label, a learning rate alpha and a marked sample threshold value, wherein a pseudo mark sample U p is initially set to be empty as shown in figure 3;
2) Inputting the seismic attribute data processed by principal component analysis and adopting a sample expanded by gray correlation analysis as a marked sample of semi-supervised learning;
3) Sequentially reading seismic data according to the sequence from left to right and from top to bottom based on a d multiplied by d sliding window, randomly selecting 20% of data blocks as convolutional neural network CNN training samples, wherein an unlabeled sample set is set as S u, and a labeled sample set is set as S l; as shown in fig. 2.
4) Training an initial classifier h i based on the labeled sample S l and the pseudo-labeled sample U p;
5) Selecting m samples from unlabeled samples by a random sampling strategy, sending the m samples into an initial classifier h i, calculating probability selection maximum values belonging to each category, and if the probability selection maximum values are larger than a threshold value, giving corresponding class labels to the samples, so that samples of the class labels form a sample set S w;
6) Updating the pseudo-labeled sample set and the unlabeled sample set in the seismic attribute data according to the formula:
Up＝Up∪Sw,Su＝Su-Sw；
7) Retraining the classifier h j, j=i+1 using the updated seismic attribute data marker samples;
8) Labeling the n samples selected in the step 6) by using a classifier h i,hj to obtain label 1,label2;
9) For the sample with the same classification result of the classifier h i,hj, adding the sample into the label sample, removing the sample from the pseudo label sample, and updating the classifier h i to h j;
10 Repeating steps 4-9 until the number of iterations is met or the number of marked samples reaches a threshold;
12 Acquiring the last classifier after training;
13 Using a classifier to detect oil gas in the research work area;
14 Outputting the oil gas detection matrix of the research work area.
(5) SSL-CNN reservoir oil gas detection algorithm flow based on similarity measurement geological structure feature tag
As shown in fig. 4, the specific steps of the algorithm are designed as follows:
1. Examples
The experimental study object is to perform oil and gas detection on the 67 th area of the Tahe oil field, and the total 222 production wells in the study work area are selected as labeling samples of the high-water-content oil well, the oil-water mixing well and the high-water-content oil well according to the production dynamic and structural characteristics. Seismic attribute data closely related to oil gas such as energy half-decay time, root mean square amplitude, instantaneous frequency and the like are extracted from the original seismic data, and data 70ms below a T74 horizon are calculated.
(1) Software and hardware environment
Experimental software and hardware configuration: the operating system is Windows10 (64 bit); the processor is Intel Core i5-6300HQ,2.30GHz; the memory is 12G, and the hard disk is 480G; the deep learning platform is an open source framework Tensorflow1.8 based on Google; the programming language is python3.6.
(2) Parameter setting
The model design comprises 4 convolution layers, 3*3 convolution kernels, 4 maximum pooling layers and 2 full connection layers, classification is carried out through a Softmax classifier, adam functions are selected as optimizers of an algorithm, cross entropy functions are selected as loss functions, iteration times are selected for 20000 times, the maximum expansion threshold of marked samples is 20% of the total number of samples, the types are 3 types in total, and the similarity threshold is set to be 0.9.
(3) Example analysis
1) TK744 well production curve analysis
TK744 well starts to produce in 4 th year 2000, the oil yield is higher from 11 th year 2007, and the water content is lower; the yield of the plant is lower from 11 months in 2007 to 10 months in 2008, and the water content is gradually increased; after this, the oil production is maintained at a low level and the water content is high and continuously fluctuates. The total oil production of the well is 109424.8 tons, and the daily average yield is 28.0 tons.
As shown in fig. 5-7, the TK744 is in the black region (predicted reservoir favorable zone) analyzed from the oil and gas detection results, and its cumulative oil production is 109424.8 tons, and is classified as high-oil well according to the production, so that the predicted result is consistent with the actual production. And analyzing a TK744 production curve, wherein the daily water content curve is a period of no aquatic oil in the early period, the yield reaches 150 tons per day, the water content is increased soon, and the water content is increased after the water content is increased by the oil-water interface after the water is extracted for a period of time in accordance with the situation that the TK744 well in the predicted result is positioned near a water area (white area).
2) TK613 well production well analysis
TK613 well started production in 10 months 2001, oil production was higher in the early short period, then water was high all the time, and the well was shut in after one month of production. The well produced 314.3 tons of oil altogether, and the daily average yield was 1.49 tons.
As shown in fig. 8-10, the TK613 was in a white zone (predicted reservoir unfavorable zone) with a cumulative oil production of 314.3 tons, consistent with the predicted results, as analyzed from the oil and gas detection results. And analyzing a TK613 production curve, wherein the daily water content curve is that the oil production is higher in a shorter time in the first 2 days, the water content is increased in the next 2 days, and the oil production is drastically reduced, so that the water content is consistent with the area of the TK613 well in a large water area in a predicted result.
The foregoing is illustrative of the best mode of carrying out the invention, and is not presented in any detail as is known to those of ordinary skill in the art. The protection scope of the invention is defined by the claims, and any equivalent transformation based on the technical teaching of the invention is also within the protection scope of the invention.

Claims (4)

1. The SSL-CNN reservoir oil gas detection method based on the similarity measurement geological structure label is characterized by comprising the following steps of:
Step 1, acquiring seismic attribute data, and selecting the seismic attribute data matched with the bottom hole coordinates of a production well to perform artificial sample marking to obtain the seismic attribute data containing artificial marked samples;
Step 2, performing expansion marking on the artificial marking sample in the seismic attribute data obtained in the previous step by adopting a gray correlation analysis technology to obtain the seismic attribute data containing the artificial marking sample and the expansion marking sample; taking the artificial marked sample and the expanded marked sample as marked samples to obtain seismic attribute data containing the marked samples;
Step 3, performing principal component analysis dimension reduction on the seismic attribute data of the extended marked sample containing the artificial marked sample to obtain the seismic attribute data containing the marked sample after dimension reduction;
Step 4, randomly selecting seismic attribute data containing marked samples after dimension reduction in a certain proportion as training samples, wherein unmarked samples in the training samples form an unmarked sample set S u, and marked samples form a marked sample set S l;
Step 5, training a classifier by using a marked sample set S l to obtain a classifier h i, wherein i is the iteration number, and the initial value of i is 1;
Step 6, selecting m training samples from the unlabeled sample set S u by adopting a random sampling strategy, sequentially sending the m training samples into a classifier h i, calculating the probability selection maximum value of each class of the training samples by the classifier h i, if the probability selection maximum value corresponding to a certain class is greater than a threshold value, giving the training samples to class labels of the corresponding class, and putting the obtained class labels into a pseudo-label sample set U p;
Step 7, adding 1 to the iteration number, training a classifier h j by using a mark sample set S l and a pseudo-tag sample set U p, randomly selecting n training samples from the pseudo-tag sample set U p, and classifying the n training samples by using a classifier h i and a classifier h j; the training samples with the same classification result of the two classifiers are removed from the pseudo tag sample set U p and added into the marked sample set S l, and the classifier h i is updated to h j;
Step 8, repeating the steps 6-7 until the iteration number reaches the highest iteration number or the number of samples in the marked sample set S l reaches the maximum threshold value, training the classifier h j in the last iteration by using the latest marked sample set S l to obtain a final classifier, and turning to the next step;
and 9, carrying out oil gas detection on the researched work area by utilizing a final classifier, and outputting an oil gas detection matrix of the researched work area.
2. The SSL-CNN reservoir hydrocarbon detection method based on similarity metric geologic structure labels according to claim 1, wherein step 1 specifically comprises the steps of:
step 1.1, original seismic attribute data are read, various types of seismic attribute data related to oil gas are extracted, including energy half-decay time, root mean square amplitude and instantaneous frequency, normalization processing is carried out, and geological and fluid characteristics represented by various seismic attribute data are analyzed;
Step 1.2, reading production dynamic data, calculating accumulated oil production of a production well, marking the accumulated oil production of the production well as a high-yield well, wherein the accumulated oil production of the production well is between a low-yield threshold and the high-yield threshold and is lower than the low-yield threshold;
and 1.3, selecting seismic attribute data matched with the bottom hole coordinates of the production well, and marking samples.
3. The method for detecting oil and gas in an SSL-CNN reservoir based on a similarity metric geologic structure label according to claim 1, wherein the method for randomly selecting a certain proportion of the seismic attribute data in step 4 is to sequentially read the seismic data in order from left to right and from top to bottom based on a d×d sliding window, and randomly selecting 20% of the data blocks in the seismic attribute data as training samples.
4. The method for detecting oil and gas in SSL-CNN reservoirs based on similarity metric geologic structure labels according to claim 3, wherein the method for performing expansion marking on the artificially marked samples in the seismic attribute data by using gray correlation analysis technique in step 2 comprises the following steps:
Step 2.1, reading seismic attribute data, setting a similarity coefficient threshold value and distinguishing an initial value of a coefficient;
Step 2.2, determining a parent sequence and a child sequence of analysis seismic attribute data, wherein the parent sequence is energy half-decay time data X 0＝[x0(1),x0(2),x0(3),...,x0(n)]T corresponding to a bottom hole coordinate of a marked sample, n represents a depth domain millisecond number, and the child sequence is seismic trace data of 8 neighborhoods of the bottom hole coordinate of the marked sample itself:
Wherein m is the number of the subsequences of the seismic trace, X i is the ith subsequence (0 < i < m), and X i (k) is the value of the k (1 < k < n) millisecond energy half-decay time of the ith subsequence;
Step 2.3, calculating gray correlation coefficients of corresponding elements of the marked sample bottom coordinate 8 neighborhood subsequence and the parent sequence seismic attribute data, wherein the calculating method comprises the following steps:
Wherein x 0 (k) is the corresponding value of k milliseconds of the energy half-decay data corresponding to the parent sequence, x i (k) is the corresponding value of k milliseconds of the energy half-decay data of the ith subsequence, ρ is a resolution coefficient, 0< ρ <1, and the default ρ value is 0.5;
Step 2.4, calculating the association degree between the bottom hole coordinates of the marked sample and the seismic channel data of the 8 neighborhood, and respectively calculating the average value of the association coefficients of the corresponding elements of the bottom hole coordinates and the indexes of the seismic channel data of the 8 neighborhood so as to reflect the association relation between the adjacent seismic channel data and the seismic channel data of the bottom hole coordinates, wherein the method comprises the following steps:
Wherein r (0,i) is the association degree between the bottom hole coordinates and the data of the ith neighborhood seismic trace, n is the millisecond number of the depth domain of the seismic trace, k is the kth millisecond (1 < k < n) of the seismic trace, and the geological space structures are more dissimilar when the difference between the association degrees is larger;
and 2.5, outputting the seismic trace data which are larger than the association threshold in the neighborhood, and marking the seismic trace data as an expansion marking sample.
CN202111369646.3A 2021-11-16 2021-11-16 SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label Active CN114167497B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111369646.3A CN114167497B (en) 2021-11-16 2021-11-16 SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111369646.3A CN114167497B (en) 2021-11-16 2021-11-16 SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label

Publications (2)

Publication Number Publication Date
CN114167497A CN114167497A (en) 2022-03-11
CN114167497B true CN114167497B (en) 2024-05-07

Family

ID=80479631

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111369646.3A Active CN114167497B (en) 2021-11-16 2021-11-16 SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label

Country Status (1)

Country Link
CN (1) CN114167497B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2259575C1 (en) * 2004-10-28 2005-08-27 Общество с ограниченной ответственностью "Инжиниринговый центр" Method for determining oil bed productiveness in three-dimensional inter-well space
CN109597129A (en) * 2019-01-07 2019-04-09 中国地质大学（武汉） Fracture-pore reservoir beading reflectance signature recognition methods based on target detection
CN112083498A (en) * 2020-10-16 2020-12-15 山东科技大学 Multi-wave earthquake oil and gas reservoir prediction method based on deep neural network
CN113095352A (en) * 2021-03-01 2021-07-09 中国地质大学（武汉） Oil-gas-containing property evaluation method for fracture-cavity type oil reservoir semi-supervised learning reservoir fused with static and dynamic characteristics
CN113534261A (en) * 2021-08-20 2021-10-22 成都理工大学 Reservoir gas content detection method and device based on intelligent optimization integrated network
CN113608265A (en) * 2021-08-13 2021-11-05 成都理工大学 Reservoir prediction method based on deep hybrid neural network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2259575C1 (en) * 2004-10-28 2005-08-27 Общество с ограниченной ответственностью "Инжиниринговый центр" Method for determining oil bed productiveness in three-dimensional inter-well space
CN109597129A (en) * 2019-01-07 2019-04-09 中国地质大学（武汉） Fracture-pore reservoir beading reflectance signature recognition methods based on target detection
CN112083498A (en) * 2020-10-16 2020-12-15 山东科技大学 Multi-wave earthquake oil and gas reservoir prediction method based on deep neural network
CN113095352A (en) * 2021-03-01 2021-07-09 中国地质大学（武汉） Oil-gas-containing property evaluation method for fracture-cavity type oil reservoir semi-supervised learning reservoir fused with static and dynamic characteristics
CN113608265A (en) * 2021-08-13 2021-11-05 成都理工大学 Reservoir prediction method based on deep hybrid neural network
CN113534261A (en) * 2021-08-20 2021-10-22 成都理工大学 Reservoir gas content detection method and device based on intelligent optimization integrated network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于EEMD高斯过程自回归模型的缝洞型油藏开发动态指标预测;张冬梅;林子航;康志江;王吉祥;邢路通;;地质科技情报;20190515(第03期);全文 *
基于双向循环神经网络的河流相储层预测方法及应用;朱剑兵;王兴谋;冯德永;隋志强;;石油物探;20200325(第02期);全文 *

Also Published As

Publication number Publication date
CN114167497A (en) 2022-03-11

Similar Documents

Publication Publication Date Title
CN108952699B (en) Intelligent stratum lithology identification method in complex geological drilling process
Ren et al. Lithology identification using principal component analysis and particle swarm optimization fuzzy decision tree
CN102819688A (en) Two-dimensional seismic data full-layer tracking method based on semi-supervised classification
CN113902861A (en) Three-dimensional geological modeling method based on machine learning
Eskandari et al. Reservoir Modelling of Complex Geological Systems--A Multiple-Point Perspective
Kemajou et al. Wellbore schematics to structured data using artificial intelligence tools
CN113095352B (en) Fracture-cavity type oil reservoir semi-supervised learning reservoir stratum oil-gas-bearing evaluation method integrating static and dynamic characteristics
Chang et al. SegLog: Geophysical logging segmentation network for lithofacies identification
CN115308799A (en) Seismic imaging free gas structure identification method and system
Mukhamediev et al. Estimation of Filtration Properties of Host Rocks in Sandstone-Type Uranium Deposits Using Machine Learning Methods
Ye et al. Drilling formation perception by supervised learning: Model evaluation and parameter analysis
Roustazadeh et al. Estimating hydrocarbon recovery factor at reservoir scale via machine learning: Database-dependent accuracy and reliability
CN114167497B (en) SSL-CNN reservoir oil gas detection method based on similarity measurement geological structure label
Morozov et al. Machine Learning on Field Data for Hydraulic Fracturing Design Optimization: Digital Database and Production Forecast Model
CN115879647B (en) Stratum lithology prediction method based on combination of domain knowledge and deep learning
CN113534283B (en) Quantitative evaluation method for ore-forming element characteristics of sandstone-type uranium ores
CN113761914B (en) Internet text weather disaster event identification method based on SVM model
Tong et al. Learning to generate rock descriptions from multivariate well logs with hierarchical attention
Sun et al. Swin Transformer based fluid classification using Gram angle field-converted well logging data: A novel approach
Katterbauer et al. A Deep Learning Wag Injection Method for Co2 Recovery Optimization
Roustazadeh et al. Estimating oil and gas recovery factors via machine learning: Database-dependent accuracy and reliability
CN118673368B (en) Shale gas horizontal well segmentation clustering method based on graph convolution neural network
CN110659685B (en) Well position optimization method based on statistical error active learning
Dwihusna Seismic and well log based machine learning facies classification in the PANOMA-hugoton field, Kansas and Raudhatain Field, North Kuwait
CN117633658B (en) Rock reservoir lithology identification method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant