BY 4.0 license Open Access Published by De Gruyter Open Access October 12, 2021

Lithology classification of volcanic rocks based on conventional logging data of machine learning: A case study of the eastern depression of Liaohe oil field

Wang Wenhua, Wang Zhuwen, Han Ruiyi, Xu Fanghui, Qi Xinghua and Cui Yitong
From the journal Open Geosciences

Abstract

The reservoirs in the eastern depression of Liaohe basin are formed by multistage igneous eruption. The lithofacies and lithology are complex, and the lithology is mainly intermediate and basic igneous rocks. Based on the integration of debris data of igneous rocks and logging data, this article selected 6,462 continuous logging data with complete cuttings data and five conventional logging curves (RLLD, AC, DEN, GR, and CNL) from four wells in the eastern depression of Liaohe basin as the training set. A variety of lithologic identification schemes based on support vector machine and random forest are established to classify the pure igneous strata and actual strata. By comparing the classification results with the identification data of core slice and debris slice, a practical lithologic classification scheme for igneous rocks in the eastern depression of Liaohe basin is obtained, and the classification accuracy reaches 97.46%.

1 Introduction

From Mesozoic to Cenozoic, the eastern depression has been regarded as the magmatic activity center of Liaohe Depression, where igneous rocks are widely developed [1]. The main lithologies in the eastern depression are trachyte, basalt, and diabase [2]. The igneous rock exploration with trachyte as the main reservoir in this area has realized the continuous oil and gas bearing situation in many areas. The practice of igneous oil and gas exploration shows that the diagenetic mode of igneous rocks has a direct impact on its reservoir properties [3,4]. The different chemical compositions of igneous rocks with similar diagenetic patterns lead to great differences in pore types and spatial distribution after diagenesis. Therefore, it is of practical significance to distinguish igneous rocks according to their chemical composition [5].

The types of igneous rocks are complex. Different eruption modes and periods and different regions and horizons have different lithologies. Not only are the mineral compositions of rocks in the same region or horizon different, but the mineral compositions of the same lithology are also very different [6]. Due to the limitation of quantity, the classification of igneous rocks cannot be evaluated systematically according to the actual core data. The conventional logging data are relatively easy to obtain, rich, and comprehensive, and can reflect the formation characteristics in the vertical direction. It is worth noting that different types of igneous rocks have great differences in conventional logging data, especially natural gamma, acoustic, compensated neutron, and resistivity curves logging data, which provide a basis for us to use conventional logging data for igneous rock lithology identification. There are many methods to identify igneous rocks by using logging data, such as cross-plot method, imaging logging method, and formation element logging method. Among them, the cross-plot method is the simplest. First, select the standard igneous rock data and cross-plot the conventional logging curves in pairs. Different types of igneous rocks will generally fall in different areas in the cross-plot, which can effectively distinguish the types of igneous rocks. However, the cross-plot method needs a large number of accurate standard lithologic data, and is limited by the geological region. At the same time, due to the complexity and particularity of the igneous rock, different types of igneous rock data may overlap in the cross-plot, so the cross-plot method cannot be used as a separate lithology identification method, but needs to be combined with other identification methods. Imaging logging method and formation element logging method can identify the lithology of volcanic rocks from the structure and chemical composition, respectively, but these two methods are expensive, and the cost of large-scale application is too high.

Machine learning is one of the most advanced research fields in artificial intelligence. As a way to realize artificial intelligence [7,8,9,10], machine learning has aroused wide interest in the field of artificial intelligence. Machine learning is not only applied in knowledge-based systems, but also widely used in natural language processing [11,12], machine vision, pattern recognition, and many other fields [13,14]. Support vector machine (SVM) is a widely used kernel machine learning algorithm [15,16], which has excellent network generalization performance in solving small sample learning and high-dimensional pattern recognition [17,18]. Random forest (RF) is a kind of statistical learning theory [19]. It uses bootstrap resampling method to extract multiple samples from the original samples, models the decision tree for each bootstrap sample and then combines the prediction of multiple decision trees to get the final prediction result by voting.

According to the conventional logging data of igneous rock area in Liaohe east depression, this article uses SVM and RF methods to classify the lithology under the condition of pure igneous rock and then designs different schemes to classify the lithology under the stratum condition, and divides the lithology of actual well section [20,21,22]. Finally, the correctness of identification results is verified by combining with actual core data and lithology section.

2 Method, classifiers, and data sets

2.1 Method

Methodology of this work is divided into three steps:

Step 1: Suppose a stratum has six kinds of igneous rock lithology under the condition of pure igneous rock, design two schemes for classification, and determine the better scheme.

Step 2: Sedimentary rock and coal are added to the formation to construct the formation under actual conditions. Based on the previous optimization scheme, four classification schemes are designed to determine an optimal scheme suitable for the actual formation conditions.

Step 3: The optimal scheme is applied to four wells for lithologic identification, and compared with lithologic section and thin section, so as to verify the accuracy of the scheme.

2.2 Classifiers

2.2.1 SVM

SVM is a common discriminant method. It is a supervised learning model in the field of machine learning, which is usually used for pattern recognition, classification, and regression analysis [23,24]. The SVM method maps the sample space to a high-dimensional or even infinite dimensional feature space (Hilbert space) through a nonlinear mapping P, so that the nonlinear separable problem in the original sample space can be transformed into a linear separable problem in the feature space. In the classification problem, for the sample set that may not be linearly processed in the low dimensional sample space, SVM first completes the calculation in the low dimensional space, then maps the input space to the high dimensional feature space through the kernel function, and finally constructs the optimal separation hyperplane in the high dimensional feature space, so as to separate the nonlinear data which is not easy to separate on the plane [25,26,27]. The process of finding hyperplane can be transformed into solving a quadratic programming problem [28,29,30].

SVM classifier is implemented by SVM module in scikit-learn library of Python programming language. SVM in scikit-learn library has four kernel functions: linear kernel function, polynomial kernel function, radial basis function (RBF), and sigmoid kernel function. In this article, the radial basis kernel function is used for the following reasons:

  1. (1)

    There are many parameters of polynomial kernel function, and the process of parameter optimization is complex.

  2. (2)

    Linear kernel function can only be used to deal with linear problems.

  3. (3)

    The sigmoid kernel function and the kernel function of RBF are similar under some conditions and can replace each other.

  4. (4)

    The five conventional logging data sets used in this article are not high-dimensional data sets, so it is more appropriate to use radial basis kernel function. There are two important parameters of RBF kernel function, which are penalty factor C and kernel parameter gamma.

In this article, 30% of the training set is taken as ten times cross-validation of the back-judgment test set, and the penalty factor C and the parameter gamma are optimized as the optimal values. The Z-score normalization of all the training set and test set data of SVM is carried out.

2.2.2 RF

RF is a machine learning algorithm based on decision tree, which combines the decision tree into a RF, that is, randomize the use of variables and data, generate many decision trees and then summarize the classification results. Like other classification methods, RF can explain the effect of several independent variables on the dependent variable y. If the dependent variable y has n observations and its k features are always related, the RF will put back n observations in the data set when constructing the decision tree, which is called bootstrap resampling method. At the same time, RF selects a part of k features randomly to construct the decision tree nodes, which can ensure that the decision tree generated each time is inconsistent. In general, a RF will generate several decision trees randomly for classification and then summarize the classification results of each tree and decide the classification results by voting.

RF classifier is implemented by RF classifier module in scikit-learn library of Python programming language. It needs to set three main parameters: the number of decision trees (n_estimators), the maximum depth of decision trees (max_depth), and the maximum number of features of decision trees (max_features). The optimization process of the three main parameters is divided into two parts:

  1. (1)

    Using the out-of-bag data generated by bootstrap resampling to estimate the internal error, and the out-of-bag error must be controlled to be low enough and stable.

  2. (2)

    Taking 30% of the training set as ten times of cross-validation to ensure that the cross-validation results are high enough.

RF classifier does not need to preprocess the data.

2.3 Selection of data sets

This article mainly studies the eastern depression of Liaohe basin, where the igneous rocks are mainly intermediate basic rocks [31,32]. According to the lithologic classification scheme of deep igneous rocks in Song Liao basin, combined with the feasibility of logging lithologic identification and the actual production demand, the igneous rocks in the study area are divided into six kinds of lithology: compact trachyte, non-compact trachyte, compact basalt, non-compact basalt, diabase, and gabbro. There are many kinds of mineral compositions in igneous reservoir rocks, and the pore and reservoir types are complex. The logging response characteristics are more complex. It is difficult to identify the lithology of igneous rocks effectively by using conventional logging interpretation methods.

Igneous rocks, coal, and sedimentary rocks coexist in the eastern depression of Liaohe basin. There are obvious differences among the three kinds of lithology in the characteristics of Genesis, structure, and mineral composition, which are difficult to be accurately distinguished by conventional logging interpretation. According to the actual needs of producers, igneous reservoirs in this area are mainly concentrated in non-compact trachyte and non-compact basalt, so the key to lithology classification is to accurately select non-compact trachyte and non-compact basalt. To identify the above two kinds of lithology as accurately as possible, this article divides the lithology classification of the area into two stages, first for the pure igneous strata and second for the actual strata.

In the conventional logging methods, RLLD, AC, CNL, DEN, and GR curves reflect the petrological characteristics of conductivity, sound velocity, porosity, density, and radioactivity, respectively, (Table 1). In this article, 6,462 continuous logging data of complete cuttings data from four wells in the eastern depression of Liaohe basin are selected as the training set, including 2,202 sedimentary rock data and 4,162 igneous rock data as well as continuous rock data of six lithology, compact trachyte, non-compact trachyte, compact basalt, non-compact basalt, diabase, and gabbro, and 98 coal seam data. It should be noted that because the actual selection of data is difficult, the data quantity of different lithologies in the training set varies greatly, that is, the training set used in this article is a typical unbalanced data set.

Table 1

Logging response of igneous rock in eastern depression, Liaohe oil field [33]

Lithology CNL (%) DEN (g cm−3) GR (API) AC (μs m−1) RLLD (Ω m)
Non-compact trachyte 16–24 2.1–2.3 152–160 223–263 42–80
Compact trachyte 4–8 2.3–2.6 140–155 171–197 44–2,000
Non-compact basalt 24–36 2.3–2.6 40–56 197–263 10–20
Compact basalt 24–29 2.6–2.8 28–40 184–236 8–12
Diabase 12–18 2.6–2.7 30–45 171–210 128–800
Gabbro 13–16 2.6–2.7 60–80 158–171 698–8,000

In addition, since the focus of this article is to classify the igneous rocks in the target area, all types of sedimentary rocks in this area are collected as sedimentary rock data set. It includes mudstone, carbonaceous mudstone, basaltic mudstone, silty mudstone, glutenite, argillaceous glutenite, argillaceous siltstone, gravelly sandstone, fine sandstone, and other common sedimentary rock data in the eastern depression area of Liaohe basin.

In this article, 5,070 continuous logging data from seven wells in the eastern depression of Liaohe basin, which are different from the training set, are selected as the test set, including 801 sedimentary rock data as well as common sedimentary rock types of the same type as the training set. There are 4,158 igneous rock data, including continuous strata data of six kinds of lithology, compact trachyte, non-compact trachyte, compact basalt, non-compact basalt, diabase, and gabbro, and 111 coal seam logging data.

3 Experimental verification

3.1 Lithologic classification of pure igneous strata

In this stage, it is assumed that there is an ideal pure igneous rock stratum, that is, it is assumed that sedimentary rock and coal in the stratum have been screened by other means, and only igneous rock exists. To get more accurate classification results, two sets of lithologic classification schemes are designed as follows (Figure 1):

Figure 1 
                  Flow chart of lithology classification of pure igneous strata.

Figure 1

Flow chart of lithology classification of pure igneous strata.

Scheme A: Direct use of SVM classifier and RF classifier to classify six lithologies of compact trachyte, non-compact trachyte, compact basalt, non-compact basalt, diabase, and gabbro.

Scheme B: Four lithologies of trachyte, basalt, diabase, and gabbro in igneous formation are classified by SVM classifier and RF classifier and then the compactness of basalt and trachyte is classified by the same classifier (Figure 2).

Figure 2 
                  Results of lithology classification of pure igneous strata.

Figure 2

Results of lithology classification of pure igneous strata.

Lithological classification of pure igneous formation is carried out according to the above two schemes (Table 2). It can be found that out of the four classification results of the two schemes mentioned above, the RF method used in scheme A, which directly classifies the six lithologies, has only 10.25% correct classification rate for non-compact trachyte and 25.6% correct comprehensive classification rate, which is obviously not competent for classification requirements. The classification accuracy of scheme A is 97.14%, but the classification accuracy of compact trachyte and non-compact basalt is 58.33 and 58.17%, respectively, which cannot meet the actual production needs.

Table 2

Statistics of lithological classification results of pure igneous formation

Lithology Data and results Scheme A SVM Scheme A RF Scheme B SVM Scheme B RF
Non-compact trachyte Return correct rate (%) 97.88 100 97.99 99.59
Number of test samples 3,179 3,179 3,179 3,179
Number of correct 3,179 326 3,179 3,166
Correctness (%) 100 10.25 100 99.59
Compact trachyte Return correct rate (%) 97.88 100 97.99 99.59
Number of test samples 48 48 48 48
Number of correct 28 48 48 48
Correctness (%) 58.33 100 100 100
Non-compact basalt Return correct rate (%) 97.88 100 97.99 99.59
Number of test samples 153 153 153 153
Number of correct 89 147 123 153
Correctness (%) 58.17 96.08 80.39 100
Compact basalt Return correct rate (%) 97.88 100 97.99 99.59
Number of test samples 178 178 178 178
Number of correct 178 177 170 178
Correctness (%) 100 99.44 95.51 100
Diabase Return correct rate (%) 97.88 100 99.64 99.59
Number of test samples 300 300 300 300
Number of correct 265 300 264 294
Correctness (%) 88.33 100 88 98
Gabbro Return correct rate (%) 97.88 100 99.64 99.59
Number of test samples 300 300 300 300
Number of correct 300 300 288 278
Correctness (%) 100 100 96 92.67
Total Return correct rate (%) 97.88 100
Number of test samples 4,158 4,158 4,158 4,158
Number of correct 4,039 1,298 4,072 4,117
Correctness (%) 97.14 25.60 97.93 99.01

Two classification methods are used in scheme B. The SVM method has a higher accuracy for non-compact trachyte and gabbro. For compact trachyte, non-compact basalt, compact basalt, and diabase, the classification accuracy is lower than that of RF method and the total classification accuracy is slightly lower than that of RF method. The RF classification method used in scheme B is more accurate for each lithology classification.

3.2 Lithological classification under actual stratigraphic conditions

This stage is lithological classification under actual formation conditions, i.e., formation is divided into sedimentary, coal, and igneous rocks according to the feasibility of logging lithological identification and actual production requirements. Because the conventional logging curves of some sedimentary rocks and igneous rocks are similar in characteristics, it is difficult to distinguish lithology effectively by one multi-classification, so four lithology classification schemes are designed in this stage, as follows (Figure 3):

Figure 3 
                  Flow chart of lithology classification under actual stratigraphic conditions.

Figure 3

Flow chart of lithology classification under actual stratigraphic conditions.

Scheme C: Sedimentary rocks, coal seams, dense trachyte, non-dense trachyte, dense basalt, non-dense basalt, diabase, gabbro, and other eight lithologies are directly classified by SVM and RF classifier.

Scheme D: By SVM and RF classifier, the strata are first divided into sedimentary rocks, igneous rocks, and coal and then the igneous rocks are classified by Scheme B using the same classifier.

Scheme E: Strata are divided into sedimentary rocks, igneous rocks, and coal by SVM classifier and then lithological classification is carried out by RF classifier of scheme B.

Scheme F: Strata are divided into sedimentary rocks, igneous rocks, and coal by RF classifier and then lithological classification is carried out by SVM classifier of scheme B (Figure 4).

Figure 4 
                  Results of lithology classification of actual strata.

Figure 4

Results of lithology classification of actual strata.

Lithological classification of actual strata is carried out according to the above four schemes (Table 3). It can be found that scheme C, which directly uses the classifier for one-off multi-classification, has poor classification results for both classifiers and cannot reach the classification accuracy required in actual production.

Table 3

Statistics of lithological classification results of actual strata

Lithology Data and results Scheme C SVM Scheme C RF Scheme D SVM Scheme D RF Scheme E Scheme F
Non-compact trachyte Return correct rate (%) 93.74 95.19 95.65 93.55 97.21 92.05
Number of test samples 3,179 3,179 3,179 3,179 3,179 3,179
Number of correct 1,906 287 3,083 1,665 3,083 1,665
Correctness (%) 59.96 9.03 96.98 52.37 96.98 52.37
Compact trachyte Return correct rate (%) 93.74 95.19 95.65 93.55 97.21 92.05
Number of test samples 48 48 48 48 48 48
Number of correct 0 47 48 48 48 48
Correctness (%) 0 97.92 100 100 100 100
Non-compact basalt Return correct rate (%) 93.74 95.19 95.65 93.55 97.21 92.05
Number of test samples 153 153 153 153 153 153
Number of correct 90 153 113 134 149 103
Correctness (%) 58.82 100 73.86 87.58 97.39 67.32
Compact basalt Return correct rate (%) 93.74 95.19 95.65 100 97.21 92.05
Number of test samples 178 178 178 178 178 178
Number of correct 176 178 178 175 178 175
Correctness (%) 98.88 100 100 98.31 100 98.31
Diabase Return correct rate (%) 93.74 95.19 95.65 100 97.21 93.60
Number of test samples 300 300 300 300 300 300
Number of correct 287 300 264 294 294 264
Correctness (%) 95.67 100 88.00 98.00 98.00 88.00
Gabbro Return correct rate (%) 93.74 95.19 95.65 93.55 97.21 93.60
Number of test samples 300 300 300 300 300 300
Number of correct 272 300 288 278 278 288
Correctness (%) 90.67 100 96.00 92.67 92.67 96.00
Sedimentary rocks Return correct rate (%) 93.74 95.19 97.61 93.94 97.61 93.94
Number of test samples 801 801 801 801 801 801
Number of correct 801 801 801 801 801 801
Correctness (%) 100 100 100 100 100 100
Coal Return correct rate (%) 93.74 95.19 97.61 93.94 97.61 93.94
Number of test samples 111 111 111 111 111 111
Number of correct 109 110 110 108 110 108
Correctness (%) 98.20 99.10 99.10 97.30 99.10 97.30
Total Return correct rate (%) 93.74 95.19
Number of test samples 5,070 5,070 5,070 5,070 5,070 5,070
Number of correct 3,641 2,176 4,885 3,503 4,941 3,552
Correctness (%) 71.81 42.92 96.35 69.09 97.46 70.06

Schemes D and F for classification of coal and igneous rocks using RFs have low accuracy in overall classification of igneous rocks, which indicates that RFs cannot effectively distinguish sedimentary rocks from igneous rocks. Compared with scheme E, which uses SVM to classify sedimentary rocks, coal, and igneous rocks, and scheme E, which uses RF to classify igneous rocks, the classification accuracy of scheme E for gabbro is slightly lower than that of scheme D, 92.67%, and the classification accuracy of other lithologies is not lower than that of scheme D, and the comprehensive classification accuracy reaches 97.46%.

4 Results and discussion

To validate the accuracy of the method described in scheme E, four wells in the east depression of Liaohe oil field are taken as examples to illustrate the igneous lithological characteristics and logging lithological identification results.

Figure 5 is the result map of lithological identification of Well X from 2,145 to 2,220 m. It can be seen from the figure that the classification method described in Scheme E can distinguish the formation boundary between sedimentary rock and trachyte in Well X according to the difference in logging responses. Based on the lithological profile, trachyte in target formation can be identified as non-compact trachyte and the thickness of the formation can be determined.

Figure 5 
               Lithology identification results of 2,145–2,220 m in well X.

Figure 5

Lithology identification results of 2,145–2,220 m in well X.

The depth of well section is 2,158–2,220 m, in which 2,200 m cuttings specimen is described as trachyte, of which 70% of the cuttings are trachyte (Figure 6a) with porphyry structure, block structure, and local polymorphic structure, mainly alkaline feldspar; and a few plagioclase phenocrysts are also observed. The matrix is of trachytic texture, alkaline feldspar microcrystal and glassy are found in the matrix. Feldspar clay mineralization is developing generally. 20% of them are trachyte igneous breccia (Figure 6c). Igneous breccia structure is a massive structure. 50% of the breccia content is trachyte and basalt clasts are also found. Igneous breccia is poorly sorted and rounded-subprismatic; another 10% is mudstone.

Figure 6 
               Well X, 2,200 m, cuttings slice. (a) Trachyte (monopolar), (b) trachyte (orthogonally polarized), (c) trachyte igneous breccia (single polarized light), and (d) trachyte igneous breccia (orthogonal polarized light).

Figure 6

Well X, 2,200 m, cuttings slice. (a) Trachyte (monopolar), (b) trachyte (orthogonally polarized), (c) trachyte igneous breccia (single polarized light), and (d) trachyte igneous breccia (orthogonal polarized light).

Core specimens at a depth of 2133.3 m are described: Lithology is trachyte, phenocryst is potassium feldspar, strongly dissolved, reaction edge is visible at the edge, matrix is directed by feldspar microcrystals of potassium feldspar, trachytic in texture, and a small amount of pyroxene is visible. The fracture is self-breaking seam and unfilled (Figure 7). The results show that the lithologic classification results of scheme E are consistent with the core cuttings data.

Figure 7 
               Well X, 2193.3 m, trachyte. (a) Core slice (single polarization) and (b) core slice (orthogonal polarization).

Figure 7

Well X, 2193.3 m, trachyte. (a) Core slice (single polarization) and (b) core slice (orthogonal polarization).

Figure 8 shows the lithologic identification results of well O 3,320–3,380 m. It can be seen from the figure that scheme E can distinguish the boundary between basalt and trachyte in Well O according to the difference in the logging response. Based on the lithologic profile, the trachyte in the target formation is identified as non-compact trachyte and basalt is non-compact basalt, and the thickness of the formation is determined.

Figure 8 
               Lithology identification results of Well O 3,320–3,380 m.

Figure 8

Lithology identification results of Well O 3,320–3,380 m.

The depth is 3,320–3,352 m, in which 3316.5 m cuttings are described as follows: The lithology is comprehensively named trachyte, in which 80% of the cuttings are trachyte (Figure 9), with porphyry structure, massive structure, and local porphyry structure. The phenocrysts are mainly alkaline feldspar (5%). A small amount of plagioclase phenocrysts can also be seen, and the matrix is of coarse texture. Alkali feldspar microcrystalline and glassy can be seen in the matrix, and feldspar clay mineralization is generally developed. Another 20% is mudstone.

Figure 9 
               Well O, 3316.5 m, trachyte. (a) Trachyte (monopolar) and (b) trachyte (orthogonally polarized).

Figure 9

Well O, 3316.5 m, trachyte. (a) Trachyte (monopolar) and (b) trachyte (orthogonally polarized).

Description of core specimen at depth of 3,360 m: The lithology is basalt (Figure 10), porphyritic texture, phenocryst is augite, and matrix is intergranular cryptic texture, and the plagioclase microcrystals are directionally distributed in the brown black igneous glass. The whole rock has developed dissolution pores, unfilled, and moderately altered. The results show that the lithologic classification results of scheme E are consistent with the core cuttings data.

Figure 10 
               Well O, 3,360 m, basalt. (a) Basalt (single polarization) and (b) basalt (orthogonal polarization).

Figure 10

Well O, 3,360 m, basalt. (a) Basalt (single polarization) and (b) basalt (orthogonal polarization).

Figure 11 shows the lithologic identification results of J Well from 2,020 to 2,080 m. It can be seen from the figure that scheme E can distinguish the boundary between the sedimentary rock and gabbro in Well J according to the difference in logging response and determine the thickness of the formation at the same time.

Figure 11 
               Lithological identification results of Well J from 2,020 to 2,080 m.

Figure 11

Lithological identification results of Well J from 2,020 to 2,080 m.

Depth of well section 2,041–2,080 m, including 2,050 m cuttings specimen description: the lithology is comprehensively named gabbro, 90% of which is gabbro structure (Figure 12). The main mineral compositions are basic plagioclase (50%), clinopyroxene (30%) and olivine (20%). Particle diameter is 1–2 mm with high degree of crystallization. Another 10% is mudstone. The result shows that the lithological classification result of scheme E is consistent with the data of core and cuttings.

Figure 12 
               Well J, 2,050 m gabbro. (a) gabbro (single polarized) and (b) gabbro (orthogonal polarized).

Figure 12

Well J, 2,050 m gabbro. (a) gabbro (single polarized) and (b) gabbro (orthogonal polarized).

Figure 13 shows the lithological identification results of Well L from 2,020 to 2,080 m. As can be seen from the figure, Scheme E can distinguish the boundary between the sedimentary rocks and diabase in Well L and determine the thickness of the formation according to the difference in logging responses.

Figure 13 
               Lithological identification results of Well L from 2,020 to 2,080 m.

Figure 13

Lithological identification results of Well L from 2,020 to 2,080 m.

The depth of the well section is 2,056–2,080 m, of which 2,058 m cuttings are described as follows: The lithology is comprehensively named amphibolite-diabase, 80% of which is amphibolite-diabase structure (Figure 14). The minerals are mainly pyroxene and plagioclase with a few biotite and amphibole and another 20% is mudstone. The result shows that the lithological classification result of scheme E is consistent with the data of core and cuttings.

Figure 14 
               L Well, 2,058 m, amphibolite-diabase. (a) diabase (single polarization) and (b) diabase (orthogonal polarization).

Figure 14

L Well, 2,058 m, amphibolite-diabase. (a) diabase (single polarization) and (b) diabase (orthogonal polarization).

5 Conclusion

In this article, the igneous rock types are summarized from the geological coring and mineral identification data, and the logging response combination characteristics of logging curves are analyzed from the logging data. Six igneous lithologies of basalt, non-compact basalt, trachyte, non-compact trachyte, diabase, and gabbro were effectively identified by SVM and RF methods. Finally, mutual proof between geological data and logging data is achieved, and the purpose of igneous lithology identification is achieved, and the following conclusions are drawn:

  1. (1)

    A classification scheme of igneous rocks is determined, which can be used in actual stratigraphic conditions, and the highest classification accuracy is 97.46%.

  2. (2)

    For lithological classification of pure igneous formation, the method of RF sectional classification used in scheme B has the highest classification accuracy, reaching 99.01%.

  3. (3)

    When the training set is unbalanced data, the RF classifier cannot effectively distinguish sedimentary rocks from igneous rocks, i.e., under unbalanced classification, the effect of RF classifier is significantly lower than that of SVM.

  1. Funding information: The work described in this article is supported by the National Natural Science Foundation of China (No. 41874135 and No. 41790453) and the National Key R&D Program of China (2019YFC0605402).

  2. Conflict of interest: Authors state no conflict of interest.

References

[1] Zongli L, Zhuwen W, Dapeng Z, Shuqin Z, Min X. Pore distribution characteristics of the igneous reservoirs in the eastern sag of the Liaohe depression. Open Geosci. 2017;9(1):161–73. 10.1515/geo-2017-0014. Search in Google Scholar

[2] Mou D, Wang ZW, Huang YL, Xu S, Zhou DP. Lithological identification of volcanic rocks from SVM well logging data: case study in the eastern depression of Liaohe Basin. Chin J Geophys (Acta Geophysica Sin). 2015;58(5):1785–93. 10.6038/cjg20150528. Search in Google Scholar

[3] Zou C, Zhao W, Jia C, Zhu R, Zhang G, Zhao X, et al. Formation and distribution of volcanic hydrocarbon reservoirs in sedimentary basins of China. Pet Expl Dev. 2008;35(3):257–71. 10.1016/S1876-3804(08)60071-3. Search in Google Scholar

[4] Jin C, Pan W, Qiao D. Volcanic facies and their reservoir characteristics in Eastern China Basins. J Earth Sci (Wuhan, China). 2013;24(6):935–46. 10.1007/s12583-013-0380-8. Search in Google Scholar

[5] Zhao W. Identification of the lithology of igneous rocks in central of the Junggar Basin. Nat Gas Ind. 2010;2(2):21–5. 10.1016/S1876-3804(11)60008-6. Search in Google Scholar

[6] Yujiao H, Chao Y, Yiren F, Xinmin G, Zhuoying F, Wenchao Y. Identification of igneous reservoir lithology based on empirical mode decomposition and energy entropy classification: a case study of Carboniferous igneous reservoir in Chunfeng oil field. Shíyóu Yŭ Tiānránqì Dìzhì. 2018;39(4):759–65. 10.11743/ogg20180413. Search in Google Scholar

[7] Saeed U, Jan SU, Lee YD, Koo I. Fault diagnosis based on extremely randomized trees in wireless sensor networks. Reliab Eng Syst Saf. 2021;205:107284. 10.1016/j.ress.2020.107284. Search in Google Scholar

[8] Onan A, KorukoGlu S. A feature selection model based on genetic rank aggregation for text sentiment classification. J Inf Sci. 2017;43(1):25–38. 10.1177/0165551515613226. Search in Google Scholar

[9] Yan D, Yasin Q, Cui M. A novel neural network for seismic anisotropy and fracture porosity measurements in carbonate reservoirs. Arab J Sci Eng. 2021;46:1–23. 10.1007/s13369-021-05970-4. Search in Google Scholar

[10] Yasin Q, Yan D, Ismail A, Du Q. Estimation of petrophysical parameters from seismic inversion by combining particle swarm optimization and multilayer linear calculator. Nat Resour Res. 2020;29:3291–317. 10.1007/s11053-020-09641-3. Search in Google Scholar

[11] Onan A. Classifier and feature set ensembles for web page classification. J Inf Sci. 2016;42(2):150–65. 10.1177/0165551515591724. Search in Google Scholar

[12] Onan A. Two-stage topic extraction model for bibliometric data analysis based on word embeddings and clustering. IEEE Access. 2019;7:145614–33. 10.1109/ACCESS.2019.2945911. Search in Google Scholar

[13] Onan A. An ensemble scheme based on language function analysis and feature engineering for text genre classification. J Inf Sci. 2018;44(1):28–47. 10.1177/0165551516677911. Search in Google Scholar

[14] Onan A, Tocoglu MA. A term weighted neural language model and stacked bidirectional LSTM based framework for sarcasm identification. IEEE Access. 2021;9:1–56. 10.1109/ACCESS.2021.3049734. Search in Google Scholar

[15] Tsoupos A, Khadkikar V. A novel SVM technique with enhanced output voltage quality for indirect matrix converters. IEEE Trans Ind Electr(1982). 2019;66(2):832–41. 10.1109/TIE.2018.2835404. Search in Google Scholar

[16] Al-Anazi A, Gates ID. On the capability of support vector machines to classify lithology from well logs. Nat Resour Res (N York, NY). 2010;19(2):125–39. 10.1007/s11053-010-9118-9. Search in Google Scholar

[17] Hsu C, Lin C. A comparison of methods for multiclass support vector machines. IEEE Trans Neural Netw. 2002;13(2):415–25. 10.1109/72.991427. Search in Google Scholar

[18] Opitz D, Maclin R. Popular ensemble methods: an empirical study. J Artif Intell Res. 1999;11:169–98. 10.1613/jair.614. Search in Google Scholar

[19] Breiman L. Bagging predictors. Mach Learn. 1996;24(2):123–40. 10.1007/BF00058655. Search in Google Scholar

[20] Suykens JAK, Vandewalle J. Least squares support vector machine classifiers. Neural Process Lett. 1999;9(3):293–300. 10.1023/A:1018628609742. Search in Google Scholar

[21] Nigam K, Mccallum AK, Thrun S, Mitchell T. Text classification from labeled and unlabeled documents using EM. Mach Learn. 2000;39(2):103–34. 10.1023/A:1007692713085. Search in Google Scholar

[22] Abedi M, Norouzi G, Bahroudi A. Support vector machine for multi-classification of mineral prospectivity areas. Comput Geosci. 2012;46:272–83. 10.1016/j.cageo.2011.12.014. Search in Google Scholar

[23] Nguyen-Sy T, To QD, Vu MN, Nguyen TD, Nguyen TT. Predicting the electrical conductivity of brine-saturated rocks using machine learning methods. J Appl Geophys. 2021;184:104238. 10.1016/j.jappgeo.2020.104238. Search in Google Scholar

[24] Vapnik VN. An overview of statistical learning theory. IEEE Trans Neural Netw. 1999;10(5):988–99. 10.1109/72.788640. Search in Google Scholar

[25] Li X, Wang L, Wang J, Zhang X. Multi-focus image fusion algorithm based on multilevel morphological component analysis and support vector machine. IET Image Process. 2017;11(10):919–26. 10.1049/iet-ipr.2016.0661. Search in Google Scholar

[26] Min X, Pengbo Q, Fengwei Z. Research and application of logging lithology identification for igneous reservoirs based on deep learning. J Appl Geophys. 2020;173:103929. 10.1016/j.jappgeo.2019.103929. Search in Google Scholar

[27] Chen Y, Wang G, Dong S. Learning with progressive transductive support vector machine. Pattern Recognit Lett. 2003;24(12):1845–55. 10.1016/S0167-8655(03)00008-4. Search in Google Scholar

[28] Yasin Q, Khalid P, Du Q. Application of machine learning tool to predict the porosity of clastic depositional system Indus Basin, Pakistan. J Pet Sci Eng. 2020;197:107975. 10.1016/j.petrol.2020.107975. Search in Google Scholar

[29] Qiang Z, Yasin Q, Du Q. Prediction of reservoir quality from log-core and seismic inversion analysis with an artificial neural network: a case study from the Sawan Gas Field, Pakistan. Energies. 2020;13(2):486. 10.3390/en13020486. Search in Google Scholar

[30] Du Q, Yasin Q. Combining classification and regression for improving shear wave velocity estimation in a highly heterogeneous reservoir from well logs data. J Pet Sci Eng. 2019;182:106260. 10.1016/j.petrol.2019.106260. Search in Google Scholar

[31] Feng Z. Volcanic rocks as prolific gas reservoir: a case study from the Qingshen gas field in the Songliao Basin, NE China. Mar Pet Geol. 2008;25(4):416–32. 10.1016/j.marpetgeo.2008.01.008. Search in Google Scholar

[32] Mao ZG, Zhu RK, Luo JL, Wang JH, Du ZH, Su L, et al. Reservoir characteristics, formation mechanisms and petroleum exploration potential of volcanic rocks in China. Pet Sci. 2015;12(1):54–66. 10.1007/s12182-014-0013-6. Search in Google Scholar

[33] Liu YL, Saraf A, Catanese B, Lee SM, Zhang Y, Connolly EP, et al. A comparison of binary and multiclass support vector machine models for volcanic lithology estimation using geophysical log data from Liaohe Basin, China. Explor Geophys (Melb). 2016;47(2):145–9. 10.1071/EG14114. Search in Google Scholar

Received: 2021-03-05
Revised: 2021-09-12
Accepted: 2021-09-14
Published Online: 2021-10-12

© 2021 Wang Wenhua et al., published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.