CN108952699B - Intelligent stratum lithology identification method in complex geological drilling process - Google Patents

Abstract

The invention discloses an intelligent identification method for lithology of strata in a complex geological drilling process, which comprises the steps of collecting logging data, analyzing the obtained logging data, selecting data closely related to the lithology, and obtaining a denoised sample data set by using a wavelet denoising technology; performing normalization processing on the basis of obtaining a sample data set, establishing an ECOC coding matrix, deleting the ECOC coding matrix by using a concurrency deletion algorithm, classifying lithology by using kernel Fisher linear discrimination as a base classifier, and optimizing kernel parameters and threshold parameters of the base classifier by using a bat optimization algorithm; and (5) finishing final lithology type output through Hamming distance decoding. The invention converts the multi-classification problem into a plurality of two-classification problems, can better distinguish lithology with similar logging characteristics, can overcome the problem that the class information with larger difference covers the class information with smaller difference in the kernel Fisher linear discrimination multi-classification method, and lays a good foundation for intelligent control of the complex drilling process.

Description

Intelligent stratum lithology identification method in complex geological drilling process

Technical Field

The invention relates to the field of intelligent control of geological drilling processes, in particular to an intelligent identification method for stratum lithology in a complex geological drilling process.

Background

At present, the demand of China on a large number of mineral products is still continuously increased, the contradiction between supply and demand of the existing resources is increasingly prominent, and the external dependence is continuously increased. In fact, however, the mineral resources in China are various and abundant in reserves. With the continuous exploration of new mineral resources, deep geological exploration and development become necessary.

Although deep resource energy development potential is large, due to the harsh environment in the deep stratum, the drilling process often encounters complex strata with different characteristics and different causes such as sensitive shale strata, rheological strata, various salts and the like. These formations are more susceptible to shrinkage, collapse or fracture than other formations due to the mechanism of formation of the formation itself. For example, well collapse occurs mostly in shale formations, accounting for over 90%. The diameter shrinkage mostly occurs in a salt-paste layer, a salt-paste-containing soft mud layer and sandstone, and the mud density is required to be specially adjusted to ensure that the well wall is not damaged. Therefore, stratum identification is carried out in the drilling process, the current stratum is detected in real time during real drilling, and the foundation for realizing safe and efficient drilling is achieved.

Lithology refers to some property that reflects a characteristic of rock, such as color, composition, structure, texture, etc. Lithology identification is the process of classifying reservoir rocks into different units based on the different characteristics and attributes of the rock presence. The formation lithology recognition is the first step of formation three-pressure modeling, and the accurate recognition of the distribution characteristics of the underground rock stratum is the key to realize the exploration and breakthrough of deep mineral resources.

The logging information contains rich stratum lithology information, which is the basic information of lithology analysis. The lithology identification modeling method using the logging data can be mainly divided into two types: firstly, an artificial drawing method; the second is an intelligent modeling method. There are many ways to identify lithology by artificial mapping, including cross mapping, spider web mapping, ladder mapping, etc. The intelligent modeling method mostly adopts a supervised machine learning method, and the intelligent modeling method aiming at the formation lithology recognition comprises methods such as a neural network, a support vector machine, ambiguity, kernel fisher linear discrimination and the like. The lithology identification by using an intelligent modeling method becomes an important direction for the development of well logging technology and drilling technology.

When the lithology identification is carried out by using an intelligent modeling method, most researches assume that the lithology distribution is uniform and the obtained logging data are balanced, and in fact, the characteristic of data imbalance widely exists between stratums. In this case, it is necessary to conduct an unbalanced data classification method study on the well log data. The intelligent identification method of the processing method of the fused unbalanced data can more effectively identify the lithology of the stratum.

Disclosure of Invention

In view of this, the embodiment of the invention provides an intelligent complex geological drilling process stratum lithology identification method which establishes a self-adaptive unbalanced multi-classification architecture of the stratum lithology in the complex geological drilling process, can effectively improve the accuracy of a lithology identification model, and lays a good foundation for intelligent control of the complex geological drilling process.

The embodiment of the invention provides an intelligent identification method for formation lithology in a complex geological drilling process, which comprises the following steps:

s1, collecting logging data, analyzing the obtained logging data, selecting data closely related to lithology, and obtaining a denoised sample data set by using a wavelet denoising technology;

s2, performing normalization processing on the basis of obtaining the sample data set, establishing an ECOC coding matrix, and deleting the ECOC coding matrix by utilizing a concurrency deletion algorithm;

s3, classifying the lithology by using kernel Fisher linear discrimination as a base classifier based on the ECOC coding matrix obtained in the step S2, and optimizing kernel parameters and threshold parameters of the base classifier by using a bat optimization algorithm;

and S4, finishing final lithology type output through Hamming distance decoding.

Further, in step S1, the data closely related to lithology includes density logging, acoustic logging, natural potential logging, deep and shallow three-lateral logging, micro-potential and micro-gradient.

Further, in step S2, the formula of the normalization process is:

in the formula: x is the number of_maxAnd x_minMaximum and minimum values of characteristic data, respectively, the characteristic data being data closely related to lithology, x_normIs the value after normalization, x_norm∈[0,1]。

Further, the step S3 specifically includes the following steps:

s3.1, selecting a Gaussian kernel by the kernel fisher-based classifier, and optimizing a kernel parameter K of each kernel fisher-based classifier according to the following formula by using a bat optimization algorithm and taking a Gmean value as a fitness function;

Gmean＝(TP/(TP+FN)×TN/(TN+FP))^1/2

K＝exp{-||x-x_c||∧2/(2*δ∧2)}

in the formula: TP, FP, FN and TN respectively represent the correct classification number of the positive classes, the positive classes are predicted but the true negative classes are predicted, the negative classes are predicted but the true positive classes are predicted, the correct classification number of the negative classes is determined, and x is nullAt a point in the middle, x_cIs the kernel function center, δ is the width parameter of the function;

s3.2, optimizing a threshold parameter lambda by using a bat optimization algorithm and taking a Gmean value as a fitness function according to the following formula;

μ＝λ×m₁+(1-λ)×m₂

in the formula: m is₁And m₂Is the average of the projections after ECOC-kernel fisher encoding and after two classes of projections, if the threshold parameter λ changes, the threshold μ will shift between mean1 and mean 2.

Further, the step S4 specifically includes the following steps:

s4.1, calculating the Hamming distance between the ECOC simplified code and a predicted value according to the following formula;

in the formula: x, y are the coding and prediction values of the ECOC matrix, i is the ith bit of the ECOC code,

represents either or;

and S4.2, searching the row number corresponding to the minimum value of the H, namely the predicted lithology label.

Compared with the prior art, the invention has the following beneficial effects:

(1) the ECOC algorithm and the kernel Fisher algorithm are combined, the multi-classification problem is converted into a plurality of two-classification problems, the advantages of kernel Fisher linear discrimination are brought into play, lithology with similar logging characteristics can be well distinguished, and the problem that class information with large difference covers class information with small difference in the kernel Fisher linear discrimination multi-classification method can be solved;

(2) the structure of ECOC codes is optimized, higher modeling precision can be obtained, and the model can obtain the effects of rapidness and accuracy;

(3) on the basis of establishing a base classifier by using ECOC codes, a bat optimization algorithm is used for optimizing kernel parameters and threshold parameters, the influence of unbalanced data can be considered, the threshold can be adaptively adjusted, the lithology identification method is verified by using actual logging data, and the application in actual production is facilitated.

Drawings

FIG. 1 is a well plot of collected well log data in an embodiment of the present invention.

Fig. 2 is a flow chart of a concurrent pruning algorithm in an embodiment of the present invention.

FIG. 3 is a diagram of an original ECOC encoding matrix according to an embodiment of the present invention.

FIG. 4 is a sequence diagram of a truncated ECOC encoding matrix according to an embodiment of the present invention.

FIG. 5 is a schematic diagram of an ECOC-BAKFDA experimental structure in an embodiment of the present invention.

FIG. 6 is a graph illustrating the relationship between the length of the ECOC encoding matrix and the classification accuracy according to an embodiment of the present invention.

FIG. 7 is a one-dimensional projection distribution diagram of the KFDA algorithm in an embodiment of the invention.

FIG. 8 is a flow chart of an intelligent method for identifying the lithology of the formation during the drilling process of complex geology.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be further described with reference to the accompanying drawings.

Referring to fig. 8, an embodiment of the present invention provides an intelligent identification method for formation lithology in a complex geological drilling process, including the following steps:

s1, collecting logging data as shown in figure 1, analyzing the obtained logging data, selecting data closely related to lithology, and obtaining a denoised sample data set by using a wavelet denoising technology; data closely related to lithology include density logs, sonic logs, natural potential logs, deep and shallow three lateral logs, micro-potentials and micro-gradients.

S2, performing normalization processing on the basis of obtaining the sample data set, establishing an ECOC coding matrix (as shown in figure 3), and deleting the ECOC coding matrix (as shown in figure 4) by utilizing a concurrency deletion algorithm;

the formula of the normalization process is:

in the formula: x is the number of_maxAnd x_minMaximum and minimum values of characteristic data, respectively, the characteristic data being data closely related to lithology, x_normIs the value after normalization, x_norm∈[0,1]。

As shown in fig. 2, the concurrent pruning algorithm:

listing all combinations of ECOC coding;

training a base classifier with training data;

for each base classifier Ci;

for each input sample;

if the combined classifier is misclassified and the base classifier is misclassified;

Metrici＝Metrici–2；

if the combined classifier is classified incorrectly and the base classifier is classified correctly;

Metrici＝Metrici+2；

if the combined classifier classification is correct and the base classifier classification is correct;

Metrici＝Metrici+1；

sorting according to the Metrici value, and selecting the base classifier with the highest Metrici value as the selected first base classifier;

for each of the remaining base classifiers;

selecting the top b-1/2 (adjustable) base classifiers with the highest Metrici value, and calculating the separation degree k between the top b-1/2 (adjustable) base classifiers and the selected base classifiers;

h represents different numbers of codes of each base classifier;

selecting the base classifier with the highest k value in the rest base classifiers;

circulating until the selected base classifiers reach the required number;

s3, classifying the lithology by using kernel Fisher linear discrimination as a base classifier based on the ECOC coding matrix obtained in the step S2, and optimizing kernel parameters and threshold parameters of the base classifier by using a bat optimization algorithm;

the method specifically comprises the following steps:

s3.1, selecting a Gaussian kernel by the kernel fisher-based classifier, and optimizing a kernel parameter K of each kernel fisher-based classifier according to the following formula by using a bat optimization algorithm and taking a Gmean value as a fitness function;

Gmean＝(TP/(TP+FN)×TN/(TN+FP))^1/2

K＝exp{-||x-x_c||∧2/(2*δ∧2)}

in the formula: TP, FP, FN and TN respectively represent the correct classification number of positive classes, are predicted to be positive classes but real negative classes, are predicted to be negative classes but real positive classes, and are correctly classified number of negative classes, x is a certain point in the space, x_cIs the kernel function center, δ is the width parameter of the function;

s3.2, optimizing a threshold parameter lambda by using a bat optimization algorithm and taking a Gmean value as a fitness function according to the following formula;

μ＝λ×m₁+(1-λ)×m₂

in the formula: m is₁And m₂Is the average of the projections after ECOC-kernel fisher encoding and after two classes of projections, if the threshold parameter λ changes, the threshold μ will shift between mean1 and mean 2.

And S4, finishing final lithology type output through Hamming distance decoding.

The method specifically comprises the following steps:

s4.1, calculating the Hamming distance between the ECOC simplified code and a predicted value according to the following formula;

in the formula: x, y are the coding and prediction values of the ECOC matrix, i is the ith bit of the ECOC code,

represents either or;

and S4.2, searching the row number corresponding to the minimum value of the H, namely the predicted lithology label.

According to the method, an ECOC algorithm and a kernel Fisher algorithm are combined, a multi-classification problem is converted into a plurality of two-classification problems, the advantages of kernel Fisher linear discrimination are brought into play, lithology with similar logging characteristics can be well distinguished, and the problem that class information with large difference covers class information with small difference in a kernel Fisher linear discrimination multi-classification method can be solved; the structure of ECOC codes is optimized, higher modeling precision can be obtained, and the model can obtain the effects of rapidness and accuracy; on the basis of establishing a base classifier by using ECOC codes, a bat optimization algorithm is used for optimizing kernel parameters and threshold parameters, the influence of unbalanced data can be considered, the threshold can be adaptively adjusted, the lithology identification method is verified by using actual logging data, and the application in actual production is facilitated.

Example 1

An intelligent stratum lithology identification method in a complex geological drilling process comprises the steps of firstly, analyzing logging data obtained through a graph 1, selecting data closely related to lithology, obtaining a denoised sample data set by using a wavelet denoising technology, using obtained 7 parameters as input of a stratum lithology identification model, establishing a coding matrix by using an improved ECOC algorithm, establishing a base classifier by using a kernel Fisher linear discrimination method, and optimizing kernel parameters and threshold parameters by using a bat algorithm. The method uses actual well logging data for simulation verification.

The method comprises the following specific steps:

(1) the original sample data consists of 153 groups of logging data including density logging, acoustic logging, natural potential logging, deep and shallow three-lateral logging and micro-potential and micro-gradient logging, 4 different lithologies are provided, and are respectively represented by labels 1,2,3 and 4, the proportion is 35:52:50:16, and the lithology data are unbalanced. Table 1 shows a part of original data, and a one-dimensional wavelet denoising technique is applied to obtain a denoised sample data set;

(2) based on the total 132 groups of sample data sets of the 7 parameters obtained in the step (1), respectively taking data of the number of the four lithology in any proportion of 15:25:25:5 as training sets, taking the remaining data in any proportion of 10:10:10:4 and the selected training set as verification sets to measure the accuracy of the classifier, and taking the remaining data as test sets. Fig. 2 shows a flowchart of an ECOC matrix pruning algorithm for pruning the ECOC matrix by using a concurrent pruning algorithm. The relationship between the length and accuracy of the ECOC matrix subtracted according to the Metric value is shown in fig. 6. Fig. 3 is an original ECOC coding matrix combination. Fig. 4 shows the truncated ECOC coding matrix. For each column of the coding matrix, a two-class transformation is performed. For the pruned ECOC codes, the class represented by code 1 can be regarded as a positive class, the class represented by code-1 can be regarded as a negative class, and the class samples represented by code 0 can be ignored.

TABLE 1 partial original logging parameter table

(3) Classifying the lithology of the stratum by using a kernel Fisher linear discrimination as a base classifier based on the ECOC coding matrix obtained in the step (2), and optimizing kernel parameters and threshold parameters of the base classifier by using a bat optimization algorithm by using a Gmean value as an objective function;

(4) and decoding by using the Hamming distance to obtain a final result. And ten-fold cross validation was used to calculate the average of their accuracy as shown in figure 5. The proposed IECOC-BAKFDA method is compared with a KFDA multi-classification method, an ECOC-BAKFDA algorithm without deletion algorithm improvement and an unoptimized ECOC-KFDA algorithm, and the results are respectively recorded in tables 2-4. The method is superior to other methods in classification precision, generalization ability and prediction speed;

TABLE 2 BA-KFDA confusion matrix without error correction coding

Final recognition rate: 76 percent.

TABLE 3 formation lithology recognition accuracy optimized by bat algorithm (ten-fold cross validation)

TABLE 4 results of stratigraphic lithology identification without bat algorithm optimization (ten-fold cross validation)

The features of the embodiments and embodiments described herein above may be combined with each other without conflict.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (4)

1. An intelligent identification method for stratum lithology in a complex geological drilling process is characterized by comprising the following steps:

s1, collecting logging data, analyzing the obtained logging data, selecting data closely related to lithology, and obtaining a denoised sample data set by using a wavelet denoising technology;

s2, performing normalization processing on the basis of obtaining the sample data set, establishing an ECOC coding matrix, and deleting the ECOC coding matrix by utilizing a concurrency deletion algorithm;

s3, classifying the lithology by using kernel Fisher linear discrimination as a base classifier based on the ECOC coding matrix obtained in the step S2, and optimizing kernel parameters and threshold parameters of the base classifier by using a bat optimization algorithm; the method specifically comprises the following steps:

s3.1, selecting a Gaussian kernel by the kernel fisher-based classifier, and optimizing a kernel parameter K of each kernel fisher-based classifier according to the following formula by using a bat optimization algorithm and taking a Gmean value as a fitness function;

Gmean＝(TP/(TP+FN)×TN/(TN+FP))^1/2

K＝exp{-||x-x_c||^∧2/(2*δ^∧2)}

in the formula: TP, FP, FN and TN respectively represent the correct classification number of positive classes, are predicted to be positive classes but real negative classes, are predicted to be negative classes but real positive classes, and are correctly classified number of negative classes, x is a certain point in the space, x_cIs the kernel function center, δ is the width parameter of the function;

s3.2, optimizing a threshold parameter lambda by using a bat optimization algorithm and taking a Gmean value as a fitness function according to the following formula;

μ＝λ×m₁+(1-λ)×m₂

in the formula: m is₁And m₂The average value is the projected average value after ECOC-kernel fisher coding and dividing into two types, if the threshold parameter lambda changes, the threshold mu can move between mean1 and mean 2;

and S4, finishing final lithology type output through Hamming distance decoding.

2. The method for intelligently identifying the lithology of the formation during the complex geological drilling process as claimed in claim 1, wherein the data closely related to the lithology in step S1 comprises density logging, acoustic logging, natural potential logging, deep and shallow three-lateral logging, micro-potential and micro-gradient.

3. The method for intelligently identifying the lithology of the stratum during the complex geological drilling process according to claim 1, wherein in the step S2, the formula of the normalization process is as follows:

in the formula: x is the number of_maxAnd x_minAre respectively speciallyMaximum and minimum values of feature data, which are data closely related to lithology, x_normIs the value after normalization, x_norm∈[0,1]。

4. The method for intelligently identifying the lithology of the stratum in the complex geological drilling process according to claim 1, wherein the step S4 specifically comprises the following steps:

s4.1, calculating the Hamming distance between the ECOC simplified code and a predicted value according to the following formula;

in the formula: x, y are the coding and prediction values of the ECOC matrix, i is the ith bit of the ECOC code,

represents either or;

and S4.2, searching the row number corresponding to the minimum value of the H, namely the predicted lithology label.

Images