CN114817356A - An oilfield core data and logging data fusion method - Google Patents

Abstract

The invention relates to a method for processing oilfield exploration data, in particular to a method for fusing oilfield core data and logging data. The method of the invention uses the density-based DBSCAN algorithm to perform clustering analysis on the core data and the logging data respectively, obtains the clustering results, finds a common well set, and then takes each coring well in the common well set as a label class, and adopts the The K-nearest neighbor algorithm of supervised learning subdivides the corresponding logging categories until all logging data is classified, ensuring that each logging data can be assigned to the closest coring well, realizing both logging data and core data. Fusion applications of scale and different density data. The method of the invention can be widely applied to the research of reservoir engineering and geological modeling, and can effectively improve the accuracy and efficiency of geological research.

Description

An oilfield core data and logging data fusion method

technical field

The invention relates to a method for processing oilfield exploration data, in particular to a method for fusing oilfield core data and logging data.

Background technique

Oilfield research is an interdisciplinary and multi-field research involving geophysical exploration, geological research, drilling, logging, logging, and development testing. There are significant differences in data acquisition methods and technologies, resulting in very different data representations. At the same time, different levels of reservoirs, such as the division of different levels of geological units, form data of different scales in geological research. Oilfield core data and logging data are two types of data that differ in form, scale, space and density. The core data is the performance of various attribute values of the geological object core, and the vertical data is mostly expressed in the centimeter-level range; the logging data is the performance of various attributes of the geological object well layer, and the vertical data is expressed within the kilometer range; The distribution of core and logging data on the oilfield plane is basically in a density ratio of 1:15 according to the number of wells. However, the two types of data are the most direct reflections of the verification of reservoir conditions, and the two types of data are the most basic requirements for integrated applications in oilfields. The core work of fusion is core homing.

Core homing refers to uniformly assigning the drilling depth of cores taken from drilling to a certain standard depth. The standard depth used in scientific research work generally refers to the logging depth.

Chinese patent application CN108227036A discloses a method for homing fine-grained sedimentary rock cores, which includes the following steps: 1) data collection and data preparation; 2) use of sensitive logging curves to establish lithology logging charts; 3) use of logging lithology Carry out the selection and calibration of typical lithological sections and lithological combination sections, and compare and analyze the core centimeter-level fine description or core scanning image, and carry out the overall preliminary positioning of the core; 4) Use the analytical laboratory data and the centimeter-level fine description of the core to analyze the preliminary The homing core is subjected to local fine homing. The purpose of the invention is to provide a method for locating fine-grained sedimentary rock cores based on the multi-parameter combination of "assay data-core description-core image-lithology combination-logging curve", in order to reduce unconventional oil and gas in fine-grained sedimentary rock formations It provides a practical technical system to improve the exploration risk of high-yield and high-efficiency well locations.

Chinese patent application CN110134918A discloses an automatic core homing method based on the sliding window method; the method sequentially includes a preliminary depth matching step, a sliding window construction step, a correlation coefficient calculation step and a screening step. The sliding window method is used in combination with logging data. Preliminary depth matching with the experimental data measured by the core and calculation of the correlation coefficient between the two when the porosity parameter slides along the logging curve. Intelligent calculation processing of automatic core positioning is performed to minimize the difference between the depth of the cable and the depth of the drill pipe. systematic error to improve the calculation efficiency and accuracy of core homing.

At present, the corresponding relationship between lithology and electrical properties is mainly used to uniformly classify the cores on the logging curve. This method is cumbersome, and is greatly affected by human factors, and the accuracy is relatively poor. Later, the core ground gamma test technology was developed, which is a test technology developed for the reuse and in-depth research of core data. The ground gamma ray homing of the core is to perform the ground gamma ray test on the core, and compare the measured curve with the logging gamma ray curve for homing. Convenient. But there is a big disadvantage that the number of wells for surface core gamma scanning is very small, and the ratio of wells to coring wells is 1:80. With such a large ratio, it is of little significance to complete the core homing work immediately and realize the fusion of the two kinds of data, because the ratio of the fusion of core gamma data and logging data is 1:1200. Geological research is a global work, and it is necessary to obtain sufficient reservoir physical property data reflected by the core and conduct research in combination with logging data.

Therefore, it is necessary to find a way to generalize the core data to all logging data, so as to realize the comprehensive integration of the reservoir physical properties reflected by the core and the reservoir state reflected by the logging data.

SUMMARY OF THE INVENTION

The main purpose of the present invention is to provide a new method for fusion of oilfield core data and logging data. Make full use of the existing oilfield full logging data and coring data, optimize the application of the density-based method DBSCAN algorithm and the supervised learning K-nearest neighbor algorithm, and establish logging data and The coring data fusion application classification model forms a complete set of coring data obtained from all logged single wells in the oilfield, making it an essential path for obtaining reservoir physical property data for geological research. The method of the invention has comprehensive research data and can accurately reflect the current situation of oil reservoirs and reservoirs.

To achieve the above object, the present invention adopts the following technical solutions:

The invention provides an oilfield core data and logging data fusion method, which comprises the following steps:

1) Establish a machine learning data training set, collect full logging data and core data, and establish a logging data training set for cluster analysis and a core data training set for cluster analysis respectively;

2) Based on the data training set, use the DBSCAN algorithm to perform cluster analysis, and obtain the cluster analysis result set of logging data and the cluster analysis result set of core data respectively;

3) Determine a common clustering well set;

4) Label each well in a common cluster well set;

5) Use supervised K-nearest neighbor algorithm for further classification;

6) Complete the supervised learning classification of all logging data;

7) Obtain the core data with the closest reservoir physical properties.

Further, in step 1), collect all logging data, analyze common logging curve types, unify data units, complete data cleaning, standardize and normalize data, and establish logging data for cluster analysis. Training set.

Further, in step 1), the core data includes pore, infiltration, saturation, carbon conventional experimental data, oil-water phase infiltration data, wettability data, sensitivity analysis data, mercury intrusion data; clean and normalize the data. First, establish a core data training set for cluster analysis.

Further, in step 2), based on the logging data training set, further feature engineering analysis is performed, and the DBSCAN algorithm is used for cluster analysis to obtain a log data cluster analysis result set A={a ₁ , a ₂ ,  … a _m }; Based on the core data training set, further carry out feature engineering analysis, use the DBSCAN algorithm to perform cluster analysis, and obtain the core data cluster analysis result set B={b ₁ ,b ₂ ,......b _n }.

Further, in step 3), the logging data class a _i is taken, the coring data class b _j with a common hash number is automatically found, and a common well set C _i ={c ₁ ,c ₂ ,...c _k is established }, k<n, k<m.

Further, in step 4), each well in the well set C _i ={c ₁ ,c ₂ ,...c _k } is taken as a class, and used as a label for the logging data to be further subdivided according to the core data in the next step, Create a label set.

Further, using the coordinate data of each well as a constraint, the wells in logging class a _i are further classified according to the C _i label set using the supervised K-nearest neighbor algorithm to complete the classification of all wells in logging class a _i . , continue to iteratively optimize to reach the target value.

Further, repeat steps 3) to 5) until each log finds the closest coring well.

Compared with the prior art, the present invention has the following advantages:

(1) The method of the present invention can realize the real-time fusion of logging data and core data, two types of data of different scales and different distribution densities. The method is not limited by experimental conditions, oil development stages, etc., and is not limited by traditional empirical formulas. It realizes the fusion of two types of data in real time, and realizes the on-demand integration of basic data for oilfield geological research business.

(2) The method of the present invention can form a complete set of physical property data of all wells in the oilfield. The method undergoes two classification learning, the first unsupervised learning obtains a preliminary classification, the second supervised learning obtains a more accurate classification, and professional constraints are added, so that all wells in the oil field can obtain the closest physical parameter values, making the oil field The physical property data is extended from point to surface, which can be used as a new data resource of oil field. Solve the problem that the most important data is missing in the process of intelligent oilfield and intelligent reservoir research.

(3) The method of the present invention can be applied to the research of reservoir engineering and geological modeling, and effectively improves the accuracy and efficiency of geological research.

Description of drawings

The accompanying drawings forming a part of the present invention are used to provide further understanding of the present invention, and the exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

FIG. 1 is a flow chart of a method for fusing oilfield core data and logging data according to a specific embodiment of the present invention.

Detailed ways

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, and/or combinations thereof.

In order to enable those skilled in the art to understand the technical solutions of the present invention more clearly, the technical solutions of the present invention will be described in detail below with reference to specific embodiments.

Example 1

As shown in Figure 1, the oilfield core data and logging data fusion method includes the following steps:

Step 110. Create a machine learning data training set. Collect full logging data, analyze common logging curve types, including SP, R4, RIID/RIIS, CAL, AC, GR, POR, etc., unify data units, complete data cleaning, standardize and normalize data, and establish Logging data training set used for cluster analysis; at the same time, the core data is analyzed, mainly various physical property data of the core, including pore, permeability, saturation, carbon conventional experimental data, oil-water phase permeability data, wettability data, sensitivity data Analyze data, mercury intrusion data, etc., clean and normalize the data, and establish a core data training set for cluster analysis.

Step 120. Use DBSCAN algorithm to perform cluster analysis. Based on the logging data training set, further carry out feature engineering analysis, use the DBSCAN algorithm to do cluster analysis, and obtain the logging data cluster analysis result set A={a ₁ , a ₂ , ... a _m }; based on the core data training set , further carry out feature engineering analysis, use DBSCAN algorithm to do cluster analysis, and obtain a set of core data cluster analysis results B = {b ₁ , b ₂ , ...... b _n }.

Step 130. Determine a common cluster well set. It is necessary to write an automatic corresponding program, take the logging data class a _i , automatically find the coring data class b _j with a common hash number, and establish a common well set C _i ={c ₁ ,c ₂ ,...c _k },k <n, k<m.

Step 140. Label each well in the common cluster well set. Take each well in the well set C _i ={c ₁ ,c ₂ ,...c _k } as a class, and use it as a label for the logging data to be further subdivided according to the core data in the next step to establish a label set.

Step 150. Use a supervised K-nearest neighbor algorithm for further classification. Using the coordinate data of each well as a constraint, the wells in logging class a _i are further classified according to the C _i label set using the supervised K-nearest neighbor algorithm to complete the classification of all wells in logging class a _i , that is, Corresponding specific core wells. The algorithm is evaluated according to the algorithm evaluation indicators such as EVS, MAE, MSE, and R2, and the iterative optimization is continuously carried out to achieve the target value.

Step 160. All logging data are classified by supervised learning. Repeat steps 3 to 5 until all logging data are classified into N-T categories (where: 0<=T<N), that is, to find the closest coring well for each logging well.

Step 170. Obtain core data with the closest reservoir physical properties for each well with logging data. Logging uses the core data of coring wells with the same classification to realize the fusion of two types of data of different scales and different distribution densities.

In the method described in this embodiment, the two types of data fusion processes are automated, while the traditional method is basically in a manual state. The traditional method is generally implemented in the following steps: borrowing the quality data of the core logging drawings; Text record, if there is no text record, it is also necessary to manually compare the difference between the logging depth and the coring depth on the core logging map to obtain the homing value of the core at different depths; then manually homing the depth of each sample in the core; finally Corresponding sample homing depth and logging depth to realize the fusion of the two. Comparing the analysis result of the method of this embodiment with the analysis result of the traditional method: the model prediction accuracy rate of the method of this embodiment reaches 92% of the traditional method, but the data fusion application efficiency is improved by 8-9 times.

The above-mentioned embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above-mentioned embodiments, and any other changes, modifications, substitutions, combinations, The simplification should be equivalent replacement manners, which are all included in the protection scope of the present invention.

Claims (8)

1. A method for fusing core data and logging data of an oil field is characterized by comprising the following steps:

1) establishing a machine learning data training set, collecting full logging data and core data, and respectively establishing a logging data training set for cluster analysis and a core data training set for cluster analysis;

2) based on the data training set, performing cluster analysis by using a DBSCAN algorithm to respectively obtain a logging data cluster analysis result set and a rock core data cluster analysis result set;

3) determining a common set of clustered wells;

4) labeling each well in a common cluster well set;

5) adopting a supervised K-nearest neighbor algorithm for further classification;

6) completing supervised learning classification of all logging data;

7) and obtaining the core data with the closest reservoir physical property.

2. The method of claim 1, wherein in step 1), the full-scale logging data is collected, common logging curve types are analyzed, data units are unified, data cleaning is completed, and the data is standardized and normalized to establish a training set of logging data for cluster analysis.

3. The method according to claim 1, wherein in step 1), the core data comprises pore, permeability, saturated carbon routine experiment data, oil-water phase permeability data, wettability data, sensitivity analysis data and mercury intrusion data; and cleaning and normalizing the data, and establishing a core data training set for cluster analysis.

4. The method as claimed in claim 1, wherein in step 2), based on the training set of logging data, further performing feature engineering analysis, and performing cluster analysis using DBSCAN algorithm to obtain a set of logging data cluster analysis results a ═ a ₁ ,a ₂ ,……a _m }; based on the core data training set, further performing characteristic engineering analysis, performing cluster analysis by using a DBSCAN algorithm, and obtaining a core data cluster analysis result set B ═ B ₁ ,b ₂ ,……b _n }。

5. The method of claim 4, wherein in step 3), the logging data a is taken _i Class, automatic finding coring data b with common well number _j Class, establishing a common well set C _i ＝{c ₁ ,c ₂ ,……c _k }，k<n，k<m。

6. The method of claim 5, wherein in step 4), a well set C is taken _i ＝{c ₁ ,c ₂ ,……c _k And taking each well as a class, and establishing a label set as next-step logging data according to labels subdivided by the core data.

7. The method of claim 6, wherein a is logged using coordinate data for each well as a constraint _i Well in the class C _i The label set is further classified by adopting a supervised K-nearest neighbor algorithm to finish logging a _i And (4) continuously performing iterative optimization on the classification of all wells in the class to reach a target value.

8. The method of any one of claims 1-7, wherein steps 3) through 5) are repeated until each well log finds the closest coring well.

Images