CN113128477A

CN113128477A - Clastic rock lithology identification method and system based on deep learning method

Info

Publication number: CN113128477A
Application number: CN202110537681.5A
Authority: CN
Inventors: 丁熊; 朱家伟; 杨曦冉; 何鑫洋; 张洋洋
Original assignee: Southwest Petroleum University
Current assignee: Southwest Petroleum University
Priority date: 2021-05-18
Filing date: 2021-05-18
Publication date: 2021-07-16

Abstract

The invention discloses a clastic rock lithology identification method and system based on a deep learning method, wherein the clastic rock lithology identification method comprises the following steps: acquiring slice image data of a clastic rock core; inputting the preprocessed clastic rock core slice image data into a pre-created LeNet-5 model to obtain the probability of each lithology of the current clastic rock; determining clastic lithology from the maximum probability. A system, comprising: the data set establishing module is used for acquiring clastic rock core image data; the data preprocessing module is mainly used for dividing a data set into lines, enhancing the data and identifying the data; the data identification module is used for inputting the data of the preprocessing module into a pre-established LeNet-5 model and outputting the identification probability of the lithology of each clastic rock; and the lithology obtaining module is used for determining the lithology of the clastic rock corresponding to the maximum probability. The invention has the advantages that: the lithology recognition efficiency and accuracy of the clastic rock are improved, the investment cost is reduced, and beneficial geological basis is provided for oil-gas geological exploration.

Description

Clastic rock lithology identification method and system based on deep learning method

Technical Field

The invention relates to the technical field of geophysical exploration of petroleum, in particular to a method and a system for identifying lithology of clastic rock based on a deep learning method.

Background

With the expansion of the breadth and depth of oil exploration and development, the oil exploration technology is increasingly mature in the basin of continental facies in China, and meanwhile, the general trend of oil and gas exploration is shifting from the oil and gas reservoir construction to the lithologic oil and gas reservoir. Therefore, it is a hot spot of research today to identify lithology of different strata. The main identification methods at present are: gravity exploration, well logging data, seismic data, remote sensing images, geochemistry, core samples, slice analysis and the like. The following Chinese invention patents:

a method for identifying lithology of whole well section, application No. CN 201410037682.3;

a lithology recognition method and system, application No. CN 201711319523.2;

a lithology recognition method for well logging based on a deep belief network, with the application number of CN 201811626064.7;

a lithology identification method and device based on a logging curve, application number CN 201810633814.7;

a lithology recognition method based on support vector regression and nuclear FISER analysis, application No. CN 201510679235.2;

a well logging lithology identification method and system based on a convolutional neural network, with the application number of CN 202010410798.2;

the above-mentioned patents have focused on qualitative identification of well logs, with less identification of core images.

In the exploration and development of the continental basin in China, clastic rock research is still the focus of research. Therefore, the geological data of the clastic rock is continuously accumulated, and the mass clastic rock geological data is used as a carrier, so that additional benefits can be brought to oil exploration and development. On one hand, the traditional mathematical theory, statistics, rock physics and geophysical exploration theory are not good for developing and utilizing a large amount of clastic rock geological data; on the other hand, for complex lithology data, the problems of lithology recognition and the like cannot be effectively solved by conventional methods such as intersection graphs, linear regression, multivariate discriminant analysis and the like. Therefore, deep learning is mainly used in oil exploration and development, and the deep learning automatically extracts features from big data and then solves the problems of complex recognition classification or prediction and the like through feature change.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a clastic rock lithology identification method and system based on a deep learning method, and solves the defects in the prior art.

In order to realize the purpose, the technical scheme adopted by the invention is as follows:

a clastic rock lithology identification method based on a deep learning method comprises the following steps:

and step A, acquiring slice image data of the clastic rock core.

And carrying out image denoising and normalization pretreatment on the obtained clastic rock core image, and dividing lithology according to the particle size of clastic rock particles into conglomerate, sandstone, siltstone and mudstone. The sample data set was partitioned by 20% test set and 80% training set.

And B, inputting the preprocessed clastic rock core slice image data into a pre-created LeNet-5 model, and outputting the probability of each lithology of the clastic rock.

And B, on the basis of preprocessing the sample data and dividing the test set and the training set in the step A, inputting the preprocessed sample data into a pre-created LeNet-5 model, training the data and outputting the probability of each lithology of the clastic rock.

And step C, determining the lithology of the clastic rock according to the maximum probability.

And on the basis of obtaining the probability of each lithology in the step B, determining the lithology according to the maximum probability and naming.

Further, the LeNet-5 model framework includes:

and inputting preprocessed image data into the input layer, wherein the size of the image is normalized to 32 x N, and N represents the type of lithology.

The C1 layer convolution layer performs the first convolution operation with 6 convolution kernel input images of 5 × 5 steps, the step size is 1, 6C 1 feature maps are obtained, the feature map size is 28 × 28, and (32-5+2 × 0)/1+1 ═ 28 can be obtained according to the output matrix size calculation formula r ═ n-f +2p)/s + 1. Since the convolution kernel size is 5 × 5, there is an offset per kernel size, so there are 6 × 5+1 — 156 parameters, 156 × 28 — 122304 connections.

And S2, performing pooling operation after the first convolution, and reducing overfitting degree of the network training parameters and the model by using 2x2 regional downsampling. Since pooling was performed using 2 × 2 nuclei, step size was 2, and thus the resulting S2 was 6 signatures of 14 × 14 ((28-2+2 × 0)/2+1 ═ 14). The S2 layer sums the pixels in the 2x2 region of the C1 layer and multiplies a weight factor plus an offset, and finally maps the result again. This layer had a total of 6 × 2+1 × 14 ═ 5880 connections.

C3 layer — convolutional layer, which is the second convolutional layer, the layer is obtained by calculating 16 10 × 10 feature maps by special combination of feature maps of S2, and is specifically calculated as follows: the first 6 signatures of C3 are connected to the 3 signatures connected to the S2 level, the last 6 signatures are connected to the 4 signatures connected to the S2 level, the last 3 signatures are connected to the 4 signatures not connected to the S2 level, and the last signature is connected to all the signatures at the S2 level. The convolution kernel size is still 5x5, so the number of parameters is: 6 (3 × 5+1) +6 (4 × 5+1) +3 (4 × 5+1) +1 (6 × 5+1) ═ 1516, and 1516 × 10 × 10 is 151600 in total.

S4 layers-pooling layers, which were pooled using 2 × 2 nuclei as the second pooling layer, to obtain 16 characteristic maps, and 16 characteristic maps of 5 × 5 were obtained by pooling 16 characteristic maps of 10 × 10 in C3 layers with 2 × 2 as a unit. There are a total of 16x (2x2+1) x5x5 ═ 2000 linkages.

C5 layers-convolutional layers, the last convolutional layer, the convolutional kernel size is 5x 5. Since the size of the 16 signatures of the S4 layer is also 5 × 5, which is the same as the size of the convolution kernel, the signature size formed after convolution is 1 × 1. This layer will produce 120 convolution results and each is connected to 16 profiles of the S4 layer, thus sharing (5x5x16+1) x 120-48120 parameters and 48120 connections.

Layer F6-fully connected layer, which is responsible for classifying the 120 features of C5. The F6 level has a total of 84 nodes, corresponding to a 7 x12 bitmap. Where-1 represents white and 1 represents black, such that the black and white of each conforming bitmap corresponds to one code. This layer has a total of (120+1) × 84 ═ 10164 training parameters and 10164 links.

And the output layer belongs to a full connection layer, and has 10 nodes which correspond to 0-9 respectively.

The invention also discloses a clastic rock lithology identification system based on deep learning, which comprises the following steps: the device comprises a data set establishing module, a data preprocessing module, a data identification module and a lithology obtaining module.

And the data set establishing module is used for acquiring image data of the core of the clastic rock, dividing the lithology of the same type into a folder and marking the folder.

The data preprocessing module is used for carrying out image denoising, normalization, dimension and enhancement on the data in the data set establishing module, and the enhancement comprises the following steps: the method comprises the steps of picture rotation, picture translation, picture cutting, picture zooming and picture turning; according to the test set: training set 2: 8, dividing the test set and the training set, for example, 1155 sample images, 231 test sets and 924 training sets.

And the data identification module is used for taking the divided test set and training set in the data preprocessing module as input, guiding the input into a pre-established LeNet-5 model, and outputting the probability of each lithology.

And the lithology obtaining module is used for determining the lithology corresponding to the maximum probability and naming the lithology on the basis of obtaining the lithology probability in the data identification module.

Compared with the prior art, the invention has the advantages that:

the lithology recognition efficiency and accuracy of the clastic rock are improved, the investment cost is reduced, and beneficial geological basis is provided for oil-gas geological exploration.

Drawings

FIG. 1 is a flow chart of a clastic rock lithology identification method based on deep learning according to an embodiment of the invention;

FIG. 2 is a schematic diagram of a LeNet-5 model framework in an embodiment of the present invention;

FIG. 3 is a block diagram of a clastic rock lithology identification system based on deep learning according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a clastic rock core sample training in accordance with an embodiment of the present disclosure;

FIG. 5 is an illustration of an example of the use of a clastic core in northwest, Chuan, in accordance with an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings by way of examples.

As shown in fig. 1, a clastic rock lithology identification method based on a deep learning method includes the following steps:

and step A, acquiring slice image data of the clastic rock core.

And B, on the basis of preprocessing the sample data and dividing the test set and the training set in the step A, inputting the preprocessed sample data into a pre-created LeNet-5 model, training the data, maximizing the recognition probability by adjusting parameters, iteration times and an activation function, and outputting the lithology probability of the test data.

As shown in fig. 2, the LeNet-5 model framework in the embodiment of the present invention includes:

The C1 layer convolution layer performs a first convolution operation with 6 convolution kernel input images of 5 × 5 in size, with a step size of 1, to obtain 6C 1 feature maps (6 feature maps of 28 × 28 in size), and calculates a formula of r ═ n-f +2p)/s +1 according to the output matrix size, to obtain (32-5+2 × 0)/1+1 ═ 28. Since the convolution kernel size is 5 × 5, there is an offset per kernel size, so there are 6 × 5+1 — 156 parameters, 156 × 28 — 122304 connections.

And S2, performing pooling operation after the first convolution, and reducing overfitting degree of the network training parameters and the model by using 2x2 regional downsampling. Since pooling was performed using 2 × 2 nuclei, step size was 2, and thus the resulting S2 was 6 signatures of 14 × 14 ((28-2+2 × 0)/2+1 ═ 14). The pooling of S2 layers is obtained by summing the pixels in the 2x2 region of C1 layer, multiplying by a weight factor plus an offset, and mapping the result again. This layer had a total of 6 × 2+1 × 14 ═ 5880 connections.

S4 layers-pooling layers, as the second pooling layer, which was still pooled using 2 × 2 nuclei, for a total of 16 signatures, 16 signatures were obtained by pooling 2 × 2 of 16 10 signatures of C3 layers, respectively, to obtain 16 signatures of 5 × 5. The connection is similar to the S2 layer, and there are 2000 connections of 16x (2x2+1) x5x 5.

C5 layers-convolutional layers, the last convolutional layer, with the convolutional kernel size still 5x 5. Since the size of the 16 signatures of the S4 layer is also 5 × 5, which is the same as the size of the convolution kernel, the signature size formed after convolution is 1 × 1. This layer will produce 120 convolution results and each is connected to 16 profiles of the S4 layer, thus sharing (5x5x16+1) x 120-48120 parameters and 48120 connections.

As shown in fig. 3, the clastic rock lithology identification system based on deep learning in the embodiment of the present invention mainly includes a data set establishing module, a data preprocessing module, a data identification module, and a lithology obtaining module.

And the data set establishing module is mainly used for acquiring image data of the core of the clastic rock, dividing the lithology of the same type into a folder and marking the folder.

The data preprocessing module is mainly used for carrying out image denoising, normalization, dimension and enhancement (including picture rotation, picture translation, picture shearing, picture scaling and picture turning) on the data in the data set establishing module; according to the test set: training set 2: 8, dividing the test set and the training set, for example, 1155 sample images, 231 test sets and 924 training sets.

And the data identification module is used for inputting the divided test set and training set in the data preprocessing module, guiding the test set and the training set into a pre-established LeNet-5 model, and outputting the probability of each lithology.

And the lithology acquisition module is mainly used for determining the lithology corresponding to the maximum probability and naming the lithology on the basis of obtaining the lithology probability in the data identification module.

As shown in fig. 4, the schematic diagram of training a clastic rock core sample in the embodiment of the present invention includes the following points;

taking a core sample of the Bessen family river group in northwest of Sichuan as an example, preprocessing a slice on the basis of obtaining a core slice sample;

inputting the preprocessed sample data into a pre-created LeNet-5 model, setting an activation function as ReLU, a learning rate as 0.01, an optimizer as cross entropy and iteration times as 100.

As shown in fig. 5, in the application of the clastic rock core of the beard family river group in northwest china, the following results are obtained by identifying the core slice in the embodiment of the present invention: the conglomerate recognition probability reaches 95.47%, the mudstone recognition probability reaches 95%, the sandstone recognition probability reaches 94%, and the siltstone recognition probability reaches 95%.

It will be appreciated by those of ordinary skill in the art that the examples described herein are intended to assist the reader in understanding the manner in which the invention is practiced, and it is to be understood that the scope of the invention is not limited to such specifically recited statements and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims

1. A clastic rock lithology identification method based on a deep learning method is characterized by comprising the following steps:

step A, acquiring slice image data of a clastic rock core;

carrying out image denoising and normalization pretreatment on the obtained clastic rock core image, and dividing lithology according to the particle size of clastic rock particles into conglomerate, sandstone, siltstone and mudstone; dividing the sample data set according to a 20% test set and an 80% training set;

b, inputting the preprocessed clastic rock core slice image data into a pre-established LeNet-5 model, and outputting the probability of each lithology of the clastic rock;

on the basis of preprocessing sample data and dividing a test set and a training set in the step A, inputting the preprocessed sample data into a pre-created LeNet-5 model, training the data and outputting the probability of each lithology of clastic rock;

step C, determining the lithology of the clastic rock according to the maximum probability;

2. The clastic lithology identification method of claim 1, wherein the LeNet-5 model framework comprises:

an input layer inputting the preprocessed image data, the size of the image being normalized to 32 × N, N representing the type of lithology;

a C1 layer-convolution layer, which performs a first convolution operation on 6 convolution kernel input images with a size of 5 × 5, wherein the step size is 1, so as to obtain 6C 1 feature maps with a size of 28 × 28, and (32-5+2 × 0)/1+1 × 28 can be obtained according to an output matrix size calculation formula r ═ n-f +2p)/s + 1; since the convolution kernel size is 5 × 5, there is an offset per kernel size, so there are 6 × 5+1 — 156 parameters, 156 × 28 — 122304 connections;

s2, performing pooling operation after the first convolution, and reducing overfitting degree of the network training parameters and the model by using 2x2 regional downsampling; since pooling was performed using 2 × 2 nuclei, step size was 2, the resulting S2 was 6 signatures of 14 × 14 ((28-2+2 × 0)/2+1 ═ 14); the S2 layer is to sum up the pixels in the 2x2 area in the C1 layer and multiply a weight coefficient and add an offset, and finally, the result is mapped again; this layer had a total of 6 × 2+1 × 14 ═ 5880 connections;

c3 layer — convolutional layer, which is the second convolutional layer, the layer is obtained by calculating 16 10 × 10 feature maps by special combination of feature maps of S2, and is specifically calculated as follows: the first 6 feature maps of C3 are connected to the 3 feature maps connected to the S2 level, the last 6 feature maps are connected to the 4 feature maps connected to the S2 level, the last 3 feature maps are connected to the 4 feature maps not connected to the S2 level, and the last feature map is connected to all the feature maps of the S2 level; the convolution kernel size is still 5x5, so the number of parameters is: 6 (3 × 5+1) +6 (4 × 5+1) +3 (4 × 5+1) +1 (6 × 5+1) ═ 1516, 1516 × 10 × 10 ═ 151600 connections;

s4 layers-pooling layers, as second pooling layers, the layers were pooled using 2 × 2 nuclei for 16 feature maps in total, and 16 feature maps of 5 × 5 were obtained by performing pooling operations in units of 2 × 2 on 16 feature maps of 10 × 10 of C3 layers, respectively; there are a total of 16x (2x2+1) x5x5 ═ 2000 linkages;

c5 layers-convolutional layers, the last convolutional layer, with a convolutional kernel size of 5x 5; since the size of the 16 characteristic maps of the S4 layer is also 5 × 5, which is the same as the size of the convolution kernel, the size of the characteristic map formed after convolution is 1 × 1; this layer will produce 120 convolution results and each is connected to 16 signatures at layer S4, thus sharing (5x5x16+1) x 120-48120 parameters and 48120 connections;

layer F6-full connectivity layer, which is responsible for classifying the 120 features of C5; the F6 level has 84 nodes in total, corresponding to a 7 x12 bitmap; wherein, -1 represents white and 1 represents black, so that the black and white of each coincident bit map corresponds to one code; this layer had a total of (120+1) × 84 ═ 10164 training parameters and 10164 links;

3. The clastic rock lithology identification method of claim 2, wherein:

the clastic rock lithology identification method is realized based on a clastic rock lithology identification system, and the clastic rock lithology identification system comprises the following steps: the device comprises a data set establishing module, a data preprocessing module, a data identification module and a lithology obtaining module;

the data set establishing module is used for acquiring image data of the core of the clastic rock, dividing lithology of the same type into a folder and marking the folder;

the data preprocessing module is used for carrying out image denoising, normalization, dimension and enhancement on the data in the data set establishing module, and the enhancement comprises the following steps: the method comprises the steps of picture rotation, picture translation, picture cutting, picture zooming and picture turning; according to the test set: training set 2: dividing a test set and a training set;

the data identification module is used for taking the divided test set and training set in the data preprocessing module as input, guiding the input into a pre-established LeNet-5 model and outputting the probability of each lithology;