CN114549899A

CN114549899A - Reservoir classification method and device, electronic equipment and computer-readable storage medium

Info

Publication number: CN114549899A
Application number: CN202210166156.1A
Authority: CN
Inventors: 任钰; 申瑞彩; 张兴聪; 方杰; 徐东兴
Original assignee: Beijing Yuexin Times Technology Co ltd
Current assignee: Beijing Yuexin Times Technology Co ltd
Priority date: 2022-02-23
Filing date: 2022-02-23
Publication date: 2022-05-27

Abstract

The application provides a reservoir classification method, a reservoir classification device, electronic equipment and a computer-readable storage medium, wherein the method comprises the following steps: acquiring a target imaging logging image and target conventional logging curve data corresponding to a target depth area in a target well; inputting the target imaging logging image into a first convolution neural network classification model, and outputting a first reservoir classification result of a target reservoir corresponding to a target depth region; inputting the target conventional well logging curve data into a second convolutional neural network classification model, and outputting a second reservoir classification result of the target reservoir; inputting a splicing result obtained by splicing the first reservoir classification result and the second reservoir classification result into a target full-connection layer to obtain a target reservoir classification result of a target reservoir; and determining the category of the target reservoir according to the classification result of the target reservoir. According to the scheme, the reservoir is automatically classified through the convolutional neural network classification model, so that the working efficiency of reservoir classification and the classification accuracy are improved.

Description

Reservoir classification method and device, electronic equipment and computer-readable storage medium

Technical Field

The present application relates to the field of computer technologies, and in particular, to a reservoir classification method, apparatus, electronic device, and computer-readable storage medium.

Background

A reservoir refers to a rock formation that stores hydrocarbons, containing interconnected porosity. The method has the advantages that the reservoir is classified, the oil and gas storage capacity of the reservoir can be objectively and generally expressed, the method has important significance for accurately and quantitatively evaluating the oil and gas reserves, and the method plays an important guiding role in oil and gas exploration and development.

In the prior art, when a reservoir is classified, the reservoir is generally classified by manually analyzing parameters of porosity, permeability, lithology and the like of the reservoir. The manual reservoir classification method mainly depends on expert experience, is influenced by subjective factors, has low classification accuracy, and has low working efficiency in the face of a large amount of logging data.

Disclosure of Invention

In view of the above, an object of the present application is to provide a reservoir classification method, apparatus, electronic device and computer-readable storage medium, so as to improve the accuracy of reservoir classification and improve the work efficiency.

In a first aspect, an embodiment of the present application provides a reservoir classification method, including:

acquiring a target imaging logging image and target conventional logging curve data corresponding to a target depth area in a target well; the target conventional logging curve data comprises target logging data corresponding to a plurality of target logging curves in the target depth region;

inputting the target imaging logging image into a pre-trained first convolution neural network classification model, and processing a first image feature of the target imaging logging image through the first convolution neural network classification model to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target well;

inputting the target conventional well logging curve data into a second convolutional neural network classification model trained in advance, and processing each target well logging data contained in the target conventional well logging curve data through the second convolutional neural network classification model to obtain a second reservoir classification result of the target reservoir corresponding to the target depth region in the target well drilling;

inputting a splicing result obtained by splicing the first reservoir classification result and the second reservoir classification result into a pre-trained target full-connection layer to obtain a target reservoir classification result of the target reservoir;

and determining the category of the target reservoir according to the target reservoir classification result.

With reference to the first aspect, an embodiment of the present application provides a first possible implementation manner of the first aspect, where before acquiring a target imaging log image corresponding to a target depth region in a target borehole, the method further includes:

acquiring an initial imaging log image corresponding to the target well drilling; the initial imaging logging image comprises imaging logging information corresponding to each depth in the target well;

segmenting the initial imaging log image according to a preset depth interval to obtain a plurality of imaging log images and the target depth area corresponding to each imaging log image;

and carrying out gray processing on the imaging log image aiming at each imaging log image to obtain the target imaging log image corresponding to the imaging log image.

With reference to the first aspect, the present application provides a second possible implementation manner of the first aspect, where before acquiring the target conventional log data corresponding to the target depth region position in the target wellbore, the method further includes:

acquiring initial conventional logging curve data corresponding to the target well drilling; the initial conventional logging curve data comprises initial logging curve data corresponding to a plurality of initial logging curves; the initial logging curve data comprises initial curve values corresponding to all depths in the target well;

for each initial logging curve data, performing smooth filtering processing on the initial logging curve data to obtain first logging curve data; the first logging curve data comprise first curve values corresponding to all depths in the target well drilling;

for each piece of first logging curve data, respectively carrying out normalization processing on each first curve value in the first logging curve data according to the maximum first curve value and the minimum first curve value in the first logging curve data to obtain a second curve value corresponding to each first curve value, and determining second logging curve data and the target logging curve corresponding to the second logging curve data according to the second curve value corresponding to each depth; the value range of the second curve value is 0-1;

and aiming at the second logging curve data corresponding to each target logging curve, selecting a second curve value corresponding to each target depth from the second logging curve data according to each target depth contained in the target depth area, carrying out mean processing on the selected second curve values to obtain a target curve value corresponding to the target depth area, and taking the target curve value corresponding to the target depth area as the target logging data corresponding to the target logging curve in the target depth area.

With reference to the first aspect, an embodiment of the present application provides a third possible implementation manner of the first aspect, where the processing, by the first convolutional neural network classification model, the first image feature of the target imaging log image to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target wellbore includes:

taking an initial image feature matrix corresponding to the target imaging logging image as an input image feature matrix, performing first convolution processing on the input image feature matrix for a first preset number of times, performing first nonlinear transformation after each first convolution processing, and performing first pooling processing on a result obtained by each first nonlinear transformation;

taking the image feature matrix obtained by the first pooling process each time as an input image feature matrix of the next first convolution process;

the first volume process includes: calculating the dot product of each first receptive field data in a first image characteristic matrix and a first convolution kernel respectively to obtain a first characteristic value corresponding to each first receptive field data, and constructing a second image characteristic matrix according to each first characteristic value; the matrix size corresponding to the first receptive field data is the same as the matrix size corresponding to the first convolution kernel;

the first non-linear transformation comprises: inputting each second characteristic value in the third image characteristic matrix into a preset activation function aiming at each third image characteristic matrix to obtain a third characteristic value corresponding to each second characteristic value in the third image characteristic matrix; in the preset activation function, for each second eigenvalue, when the second eigenvalue is greater than 0, determining the second eigenvalue as a third eigenvalue; when the second characteristic value is not more than 0, changing the second characteristic value to obtain a third characteristic value; constructing a fourth image feature matrix using the third feature values;

the first pooling process includes: extracting a maximum value from each second receptive field data in a fifth image characteristic matrix, and constructing a sixth image characteristic matrix according to the maximum value extracted from each second receptive field data;

and performing first full-connection processing on the output image feature matrix obtained by the last first pooling processing to obtain the first reservoir classification result corresponding to the target imaging logging image.

With reference to the first aspect, an embodiment of the present application provides a fourth possible implementation manner of the first aspect, where the processing, by the second convolutional neural network classification model, each target log data included in the target conventional log data to obtain a second reservoir classification result corresponding to the target depth region in the target wellbore includes:

taking each target logging data contained in the target conventional logging curve data as input logging data, performing second convolution processing on the input logging data for a second preset number of times, performing second nonlinear transformation after each second convolution processing, and performing second pooling processing on a result obtained by each second nonlinear transformation;

taking the characteristic data obtained by each second pooling as input logging data of the next second convolution processing;

the second convolution processing includes: dividing all the first logging data to obtain at least one third receptive field data; calculating the dot product of the first logging data and a second convolution kernel contained in the third receptive field data aiming at each third receptive field data to obtain a fourth characteristic numerical value corresponding to the third receptive field data, and constructing second logging data according to the fourth characteristic numerical value; the matrix size corresponding to the third receptive field data is the same as the matrix size corresponding to the second convolution kernel; the second non-linear transformation comprises: judging whether the third logging data is greater than 0 or not according to each third logging data; if the third logging data is larger than 0, determining the third logging data as fourth logging data; if the third logging data is not greater than 0, determining the product of the third logging data and a preset multiple as fourth logging data;

the second pooling process includes: dividing all the fifth logging data to obtain at least one fourth receptive field data; extracting a maximum value from the fourth receptive field data aiming at each fourth receptive field data, and constructing sixth logging data according to the maximum value extracted from each fourth receptive field data; and carrying out second full-connection processing on output logging data obtained by the last second pooling processing to obtain a second reservoir classification result corresponding to the target conventional logging curve data.

With reference to the first aspect, an embodiment of the present application provides a fifth possible implementation manner of the first aspect, where after determining respective corresponding categories of the target reservoirs corresponding to each target depth region in the target wellbore, the method further includes:

for each target reservoir, judging whether the category corresponding to the target reservoir is the same as the categories corresponding to other target reservoirs adjacent to the target reservoir;

and if the category corresponding to the target reservoir is the same as the categories corresponding to other target reservoirs adjacent to the target reservoir, merging the target reservoir and the other target reservoirs adjacent to the target reservoir to obtain a synthetic reservoir.

With reference to the first aspect, an embodiment of the present application provides a sixth possible implementation manner of the first aspect, where the first convolutional neural network classification model, the second convolutional neural network classification model, and the target fully-connected layer are obtained by training in the following manner:

constructing training samples for training the first convolutional neural network classification model, the second convolutional neural network classification model and the target fully-connected layer; the training samples include: the method comprises the steps that a sample imaging logging image and sample conventional logging curve data corresponding to a sample depth area in a sample drill well, and a class label of a sample reservoir layer corresponding to the sample depth area; the sample conventional logging curve data comprises sample logging data corresponding to a plurality of sample logging curves in the sample depth region;

inputting the sample imaging logging image into a first initial classification model, and processing a second image characteristic of the sample imaging logging image through the first initial classification model to obtain a first prediction result of the sample reservoir;

inputting the sample conventional well logging curve data into a second initial classification model, and processing each sample well logging data contained in the sample conventional well logging curve data through the second initial classification model to obtain a second prediction result of the sample reservoir;

inputting a third prediction result obtained by splicing the first prediction result and the second prediction result of the sample reservoir into an initial full-connection layer, and outputting a prediction classification result of the sample reservoir;

calculating a loss value according to the prediction classification result and the class label of the sample reservoir;

and updating parameters in the first initial classification model, the second initial classification model and the initial full-link layer by using the loss value, and training the updated first initial classification model, the second initial classification model and the initial full-link layer again until the training times reach a preset training time, so as to obtain the first convolutional neural network classification model, the second convolutional neural network classification model and the target full-link layer.

With reference to the first aspect, the present application provides a seventh possible implementation manner of the first aspect, where after determining the category of the target reservoir according to the target reservoir classification result, the method further includes:

evaluating the amount of storage of a target substance stored in the target reservoir according to the category of the target reservoir.

In a second aspect, an embodiment of the present application further provides a reservoir training device, including:

the first acquisition module is used for acquiring a target imaging logging image and target conventional logging curve data corresponding to a target depth area in a target well; the target conventional logging curve data comprises target logging data corresponding to a plurality of target logging curves in the target depth region;

the first input module is used for inputting the target imaging logging image into a first convolution neural network classification model trained in advance, and processing a first image feature of the target imaging logging image through the first convolution neural network classification model to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target well;

the second input module is used for inputting the target conventional logging curve data into a second convolutional neural network classification model which is trained in advance, and processing each target logging data contained in the target conventional logging curve data through the second convolutional neural network classification model to obtain a second reservoir classification result of the target reservoir corresponding to the target depth region in the target well;

the third input module is used for inputting a splicing result obtained after splicing the first reservoir classification result and the second reservoir classification result into a pre-trained target full-connection layer to obtain a target reservoir classification result of the target reservoir;

and the first determining module is used for determining the category of the target reservoir according to the target reservoir classification result.

With reference to the second aspect, the present application provides a first possible implementation manner of the second aspect, where the method further includes:

the second acquisition module is used for acquiring an initial imaging logging image corresponding to the target well before the first acquisition module acquires the target imaging logging image corresponding to the target depth region in the target well; the initial imaging logging image comprises imaging logging information corresponding to each depth in the target well;

the segmentation module is used for segmenting the initial imaging log image according to a preset depth interval to obtain a plurality of imaging log images and the target depth area corresponding to each imaging log image;

and the gray processing module is used for carrying out gray processing on the imaging logging image aiming at each imaging logging image to obtain the target imaging logging image corresponding to the imaging logging image.

With reference to the second aspect, embodiments of the present application provide a second possible implementation manner of the second aspect, where the method further includes:

the third acquisition module is used for acquiring initial conventional logging curve data corresponding to the target well before the first acquisition module acquires the target conventional logging curve data corresponding to the position of the target depth region in the target well; the initial conventional logging curve data comprises initial logging curve data corresponding to a plurality of initial logging curves; the initial logging curve data comprises initial curve values corresponding to all depths in the target well;

the filtering module is used for carrying out smooth filtering processing on the initial logging curve data aiming at each initial logging curve data to obtain first logging curve data; the first logging curve data comprise first curve values corresponding to all depths in the target well drilling;

a normalization module, configured to, for each first logging curve data, perform normalization processing on each first curve value in the first logging curve data according to a maximum first curve value and a minimum first curve value in the first logging curve data, to obtain a second curve value corresponding to each first curve value, and determine second logging curve data and the target logging curve corresponding to the second logging curve data according to the second curve value corresponding to each depth; the value range of the second curve value is 0-1;

and the mean value processing module is used for selecting the second curve value corresponding to each target depth from the second logging curve data according to each target depth contained in the target depth area aiming at the second logging curve data corresponding to each target logging curve, carrying out mean value processing on the selected second curve value to obtain the target curve value corresponding to the target depth area, and taking the target curve value corresponding to the target depth area as the target logging data corresponding to the target logging curve in the target depth area.

With reference to the second aspect, an embodiment of the present application provides a third possible implementation manner of the second aspect, wherein when the first input module is configured to calculate, through the first convolutional neural network classification model, a first image feature of the target imaging log image to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target wellbore, the first input module is specifically configured to:

With reference to the second aspect, the present application provides a fourth possible implementation manner of the second aspect, wherein the second input module, when being configured to process, through the second convolutional neural network classification model, each target log data included in the target conventional log data to obtain a second reservoir classification result of the target reservoir corresponding to the target depth region in the target wellbore, is specifically configured to:

the second convolution processing includes: dividing all the first logging data to obtain at least one third receptive field data; calculating the dot product of the first logging data and a second convolution kernel contained in the third receptive field data aiming at each third receptive field data to obtain a fourth characteristic numerical value corresponding to the third receptive field data, and constructing second logging data according to the fourth characteristic numerical value; the matrix size corresponding to the third receptive field data is the same as the matrix size corresponding to the second convolution kernel; the second non-linear transformation comprises: judging whether the third logging data is greater than 0 or not according to each third logging data; if the third logging data is larger than 0, determining the third logging data as fourth logging data; if the third logging data is not larger than 0, determining the product of the third logging data and the preset multiple as fourth logging data;

With reference to the second aspect, embodiments of the present application provide a fifth possible implementation manner of the second aspect, where the method further includes:

the judging module is used for judging whether the category corresponding to the target reservoir is the same as the categories corresponding to other target reservoirs adjacent to the target reservoir or not aiming at each target reservoir after determining the respective category corresponding to the target reservoir corresponding to each target depth region in the target drilling well;

and the merging module is used for merging the target reservoir stratum and other target reservoir stratums adjacent to the target reservoir stratum to obtain a synthetic reservoir stratum if the category corresponding to the target reservoir stratum is the same as the categories corresponding to other target reservoir stratums adjacent to the target reservoir stratum.

With reference to the second aspect, embodiments of the present application provide a sixth possible implementation manner of the second aspect, where the method further includes:

the building module is used for building training samples for training the first convolutional neural network classification model, the second convolutional neural network classification model and the target full-connection layer; the training samples include: the method comprises the steps that a sample imaging logging image and sample conventional logging curve data corresponding to a sample depth area in a sample drill well, and a class label of a sample reservoir layer corresponding to the sample depth area; the sample conventional logging curve data comprises sample logging data corresponding to a plurality of sample logging curves in the sample depth region;

the fourth input module is used for inputting the sample imaging logging image into the first initial classification model, and processing the second image characteristics of the sample imaging logging image through the first initial classification model to obtain a first prediction result of the sample reservoir;

the fifth input module is used for inputting the sample conventional well logging curve data into a second initial classification model, and processing each sample well logging data contained in the sample conventional well logging curve data through the second initial classification model to obtain a second prediction result of the sample reservoir;

the sixth input module is used for inputting a third prediction result obtained by splicing the first prediction result and the second prediction result of the sample reservoir into an initial full-connected layer and outputting a prediction classification result of the sample reservoir;

a calculation module for calculating a loss value according to the predicted classification result of the sample reservoir and the class label;

and the updating module is used for updating parameters in the first initial classification model, the second initial classification model and the initial full-link layer by using the loss value, and retraining the updated first initial classification model, the second initial classification model and the initial full-link layer until the training times reach a preset training time, so as to obtain the first convolutional neural network classification model, the second convolutional neural network classification model and the target full-link layer.

With reference to the second aspect, embodiments of the present application provide a seventh possible implementation manner of the second aspect, where the method further includes:

and the second determination module is used for evaluating the storage amount of the target substance stored in the target reservoir according to the category of the target reservoir after the first determination module determines the category of the target reservoir according to the classification result of the target reservoir.

In a third aspect, an embodiment of the present application further provides an electronic device, including: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating via the bus when the electronic device is running, the machine-readable instructions being executable by the processor to perform the steps of any one of the possible implementations of the first aspect.

In a fourth aspect, this application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to perform the steps in any one of the possible implementation manners of the first aspect.

According to the reservoir classification method, the device, the electronic equipment and the computer-readable storage medium, the first image characteristics of the target imaging log image are processed through the first convolutional neural network classification model, and a first reservoir classification result of a target reservoir corresponding to a target depth region in a target well is obtained; processing each target logging data contained in the target conventional logging curve data through a second convolutional neural network classification model to obtain a second reservoir classification result of the target reservoir; inputting a splicing result obtained by splicing the first reservoir classification result and the second reservoir classification result into a target full-connection layer to obtain a target reservoir classification result of a target reservoir; and determining the category of the target reservoir according to the classification result of the target reservoir. Compared with the method for reservoir classification manually in the prior art, the method has the advantages that the reservoir is automatically classified through the convolutional neural network model, the working efficiency of reservoir classification is improved, and the accuracy of reservoir classification is improved.

In order to make the aforementioned objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.

Fig. 1 illustrates a flow chart of a method of reservoir classification provided by an embodiment of the present application;

FIG. 2 is a schematic diagram of an initial image feature matrix and first and second convolution kernels provided by an embodiment of the present application;

FIG. 3 is a schematic diagram illustrating a fifth image feature matrix and a sixth image feature matrix provided in an embodiment of the present application;

fig. 4 is a schematic structural diagram illustrating a reservoir classification device provided in an embodiment of the present application;

fig. 5 shows a schematic structural diagram of an electronic device provided in an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.

In the prior art, methods for classifying reservoirs mainly include the following three categories: (1) the reservoirs are classified by manually analyzing parameters of porosity, permeability, lithology and the like of the reservoirs. (2) Conventional logging curve data are analyzed through machine learning methods such as multiple linear regression, support vector machines and fuzzy logic, and automatic classification of reservoir classes is achieved. (3) And the automatic classification of the reservoir classes is realized by analyzing imaging logging information.

Although the above methods achieve classification of reservoirs, there are still some problems: the manual reservoir classification method mainly depends on expert experience, is influenced by subjective factors, is low in classification accuracy, and is low in working efficiency in the face of a large amount of logging data. The method for analyzing the conventional logging curve data or the imaging logging data has certain singleness and low classification accuracy.

In view of the above problems, embodiments of the present application provide a reservoir classification method, device, electronic device, and computer-readable storage medium to improve the accuracy and work efficiency of reservoir classification, which are described below by way of example.

The first embodiment is as follows:

to facilitate understanding of the present embodiment, a reservoir classification method disclosed in the embodiments of the present application will be described in detail first. Fig. 1 is a flow chart illustrating a reservoir classification method provided in an embodiment of the present application, and as shown in fig. 1, the method includes the following steps:

s101: acquiring a target imaging logging image and target conventional logging curve data corresponding to a target depth area in a target well; the target conventional logging curve data comprises target logging data corresponding to a plurality of target logging curves in a target depth area.

The target well is stored with substances such as oil, natural gas and the like, wherein the oil or the natural gas is stored in the target well through a reservoir stratum. The reservoir has interconnected pores that allow oil or gas to be stored therein through the reservoir within a target wellbore.

The target well can be a newly drilled well, and for a newly drilled well (i.e., the newly drilled well), reservoir division results corresponding to different depths of the whole well need to be determined, i.e., the reservoirs in the target well are classified, and whether each reservoir in the target well is a good reservoir or a non-good reservoir is determined. The purpose of classifying the reservoirs is to analyze the potential of oil and gas development in the target well bore and determine the key to further increase the recoverable reserves in the target well bore.

The target depth refers to the distance between each location within the well and the surface, and the target depth zone refers to the distance between each zone within the well and the surface. Illustratively, the target depth zone is between 100 meters and 102 meters, between 102 meters and 104 meters, and the like.

The target imaging well logging image corresponding to the target depth area comprises an in-well logging image corresponding to the target depth area in the target well. The target conventional logging curve data corresponding to the target depth area comprises target logging data corresponding to a plurality of target logging curves corresponding to the target depth area in the target drilling well at the target depth area. The target well logging curves are multiple, the abscissa of each target well logging curve is the distance between the target well logging curve and the ground, namely the depth in the well, and the ordinate is the curve value corresponding to each depth. Wherein the number of data contained in the target conventional well log data is the same as the number of target well logs.

Illustratively, there are 7 target well logs, and each target well log corresponds to one curve value in the target depth region, so that the target conventional well log data corresponding to the target depth region includes 7 curve values. And according to the change trend of the target logging curve along with the depth, the reservoir properties of different depth areas can be reflected.

S102: inputting the target imaging logging image into a first convolutional neural network classification model trained in advance, and processing first image characteristics of the target imaging logging image through the first convolutional neural network classification model to obtain a first reservoir classification result of a target reservoir corresponding to a target depth region in a target well.

The first convolutional neural network classification model may be a two-dimensional convolutional neural network classification model. Each target depth zone corresponds to one target reservoir, for example, if the target depth zone is between 100 and 102 meters, the target reservoir corresponding to the target depth zone is a reservoir within the target borehole at a depth of 100 and 102 meters. The first reservoir classification result is the probability of the target reservoir on each category predicted by the first convolutional neural network classification model. Illustratively, when the reservoir categories are classified into a non-premium reservoir, a first-class premium treatment and a second-class premium reservoir, the first reservoir classification result is the probability that the target reservoir belongs to the non-premium reservoir, the first-class premium treatment and the second-class premium reservoir respectively.

S103: and inputting the target conventional well logging curve data into a second convolutional neural network classification model trained in advance, and processing each target well logging data contained in the target conventional well logging curve data through the second convolutional neural network classification model to obtain a second reservoir classification result of a target reservoir corresponding to a target depth region in the target well drilling.

The second convolutional neural network classification model may be a one-dimensional convolutional neural network classification model. The second reservoir classification result is the probability of the target reservoir predicted by the second convolutional neural network classification model on each category. Aiming at the same target reservoir, the first reservoir classification result and the second reservoir classification result corresponding to the target reservoir may be the same or different.

S104: and inputting a splicing result obtained by splicing the first reservoir classification result and the second reservoir classification result into a pre-trained target full-connection layer to obtain a target reservoir classification result of the target reservoir.

Illustratively, the first reservoir classification result is [0.2,0.5,0.3], the second reservoir classification result is [0.1,0.8,0.1], the splicing result obtained after splicing the two reservoir classification results is [0.2,0.5,0.3,0.1,0.8,0.1], the splicing result is input into a pre-trained target fully-connected stratum to obtain a target reservoir classification result, namely the probability that the target reservoir belongs to a non-premium reservoir, a first-class premium reservoir and a second-class premium reservoir.

S105: and determining the category of the target reservoir according to the classification result of the target reservoir.

The target reservoir classification result comprises the probability that the target reservoir belongs to a non-high-quality reservoir, a first-class high-quality reservoir and a second-class high-quality reservoir respectively, and the class with the highest probability is selected from the target reservoir classification result and is determined as the final class of the target reservoir.

Illustratively, the target reservoir classification result includes a probability 0.2 that the target reservoir belongs to a first-class high-quality reservoir, a probability 0.45 of a second-class high-quality reservoir, and a probability 0.1 of a non-high-quality reservoir, and the second-class high-quality reservoir corresponding to the probability 0.45 is determined as the class of the target reservoir, that is, the class of the target reservoir is the second-class high-quality reservoir.

Compared with the method for analyzing the conventional logging curve or the imaging logging data in the prior art, the reservoir classification method comprehensively considers two sets of data of the target imaging logging image and the target conventional logging curve data, and is favorable for improving the classification accuracy by comprehensively analyzing the imaging logging image and the conventional logging curve data.

In a possible embodiment, before the step S101 is executed to acquire the target imaging log image corresponding to the target depth region in the target borehole, the method further includes:

s1011: acquiring an initial imaging logging image corresponding to a target well drilling; the initial imaging logging image comprises imaging logging information corresponding to each depth in the target well.

The initial imaging logging image is a method for imaging physical parameters of a well wall and objects around the well according to the observation of a geophysical field in the drilled hole, and can visually and visibly reflect stratum characteristics corresponding to different depths in a target drilled hole.

The initial imaging log image comprises imaging log images corresponding to all depth areas in the whole target well, and the imaging log images are images shot by a logging instrument.

S1012: and segmenting the initial imaging log image according to a preset depth interval to obtain a plurality of imaging log images and a target depth area corresponding to each imaging log image.

In one embodiment, the initial imaging log image has annotated information (i.e., redundant information) on the head and tail, such as the number of the target well, the name of the target well, color system labels, north of guide labels, and the like. Therefore, when the initial imaging log image is segmented, firstly removing the labeling information on the head and the tail of the initial imaging log image (i.e. removing the redundant information on the initial imaging log image) to obtain the initial imaging log image with the labeling information removed. And then, segmenting the initial imaging log image without the labeling information again according to the preset depth interval to obtain a plurality of imaging log images.

The size of the imaging log image is smaller than that of the initial imaging log image, the preset depth interval can be 1 meter, 1.5 meters, 2 meters and the like, and empirical adjustment can be performed.

Illustratively, when the initial imaging log image includes a log image with a depth of 100 meters to 500 meters in the target borehole, if the preset depth interval is 2 meters, 200 segmented imaging log images are obtained, wherein the imaging log images include 1 imaging log image corresponding to a target depth region of 100 meters to 102 meters, 1 imaging log image corresponding to a target depth region of 102 meters to 104 meters, 1 imaging log image corresponding to a target depth region of 498 meters to 500 meters, and the like. Each of the imaged log images is the same size.

S1013: and performing gray processing on the imaging log image aiming at each imaging log image to obtain a target imaging log image corresponding to the imaging log image.

The initial imaging log image and the imaging log image are color images in RGB format (RGB stands for three color channels of red R, green G, and blue B). Each pixel point in the color image of the RGB color space is determined by the data of R, G, B channels, and since the data for processing three channels is complex, the influence of color on the determination of reservoir type in the imaging log image is too small, the imaging log image is grayed, that is, the three channels of RGB are unified into the data of one channel. Because human eyes have different sensitivities to RGB colors, the scheme adopts a weighting mode to obtain data after gray processing, and the specific formula is as follows:

gray(x)＝0.299×red(x)+0.587×green(x)+0.114×blue(x)

aiming at each pixel point x in the imaging log image, red (x) represents a pixel value corresponding to a red R channel of the pixel point x, green (x) represents a pixel value corresponding to a green G channel of the pixel point x, blue (x) represents a pixel value corresponding to a blue B channel of the pixel point x, and gray (x) represents a gray value of the pixel point x.

In one possible embodiment, before performing step S101 to obtain target conventional log data corresponding to a target depth region position in a target well, the method further includes:

s1014: acquiring initial conventional logging curve data corresponding to a target well; the initial conventional logging curve data comprises initial logging curve data corresponding to a plurality of initial logging curves; the initial logging curve data comprises initial curve values corresponding to all depths in the target drilling well.

The initial well logs may be 7 well logs, such as a natural gamma curve, a natural potential curve, a borehole diameter curve, a sound wave curve, a neutron curve, a density curve, a resistivity curve, and the like, wherein each initial well log includes an initial curve value corresponding to each depth in the target borehole, an abscissa of each initial well log is a depth in the target borehole, and an ordinate is a corresponding curve value. The initial logging curve data corresponding to each initial logging curve comprises curve values corresponding to all depths on the initial logging curve. The initial conventional well logging curve data comprises curve values corresponding to each initial well logging curve at each depth.

S1015: for each initial logging curve data, performing smooth filtering processing on the initial logging curve data to obtain first logging curve data; the first logging curve data comprises first curve values corresponding to all depths in the target drilling well.

In the logging operation, due to the random effects of nuclear decay, extra-nuclear electrons, gamma quanta and the like, or the random collision of a logging probe, multiple refraction and reflection of sound waves and the like, an initial logging curve contains a large amount of interference information. If we directly use these signals that are not related to geological features for well logging analysis, the analysis results may be greatly affected. Therefore, each initial logging curve is subjected to smooth filtering treatment in a curve smooth filtering mode, useful signals related to geological special effects are reserved to the maximum extent, and interference of short-period burr signals is effectively inhibited.

The smoothing filtering process adopts a Savitzky-Golay smoothing filtering (a filtering method based on least square fitting), and the filtering method eliminates noise under the condition of keeping the morphological characteristics of an original curve unchanged.

In a specific embodiment, when the Savitzky-Golay smoothing filtering method is used for smoothing filtering processing, specifically, a k-order polynomial fitting is performed on data points in a window with a certain length through a weighted average algorithm of a moving sliding window, so as to obtain an effect after fitting.

For example, for each initial logging curve, if there are curve values corresponding to 20 depths on the initial logging curve, the 20 depths are divided, and every 5 depths are a group to obtain 4 groups of data, that is, the width of the moving sliding window is 5 at this time. For each set of data (i.e. for the data in each moving sliding window), performing k-order polynomial fitting by using the set of data to obtain the parameter a in the following formula_i(i 1.. k), the curve values for each depth included in the set of data are then input into the following formula:

Y＝a₀+a₁x+a₂x²+…+a_kx^k

wherein, x represents a curve value on an initial logging curve corresponding to a certain depth, and Y represents a first curve value on first logging curve data corresponding to the depth after smoothing processing.

S1016: for each piece of first logging curve data, respectively carrying out normalization processing on each first curve value in the first logging curve data according to the maximum first curve value and the minimum first curve value in the first logging curve data to obtain a second curve value corresponding to each first curve value, and determining second logging curve data and a target logging curve corresponding to the second logging curve data according to the second curve value corresponding to each depth; the value range of the second curve value is 0-1. Because different first well logging curves have different fluctuation ranges, the first well logging curve with a smaller value is phagocytosed by the first well logging curve with a larger value due to different dimensional ranges (namely fluctuation ranges), normalization processing is carried out on data of each first well logging curve, the dimensional expression is changed into a dimensionless expression, and the convergence speed and the classification precision of the model are improved.

For example, the ordinate of some first well logs may fluctuate between 1 and 5, and the ordinate of another first well log may fluctuate between 40 and 50, in order to avoid the influence of different fluctuation ranges on the classification result, the present scheme normalizes each first curve value in the first well log data of each first well log, and specifically, for each first well log, inputs each first curve value in the first well log data of the first well log into the following formula:

wherein, P_maxRepresents the maximum value, P, of all the first curve values contained in the first log data_minRepresents the minimum value, P, of all the first curve values contained in the first log data_iEach first curve value Q included in the first log data_iAnd the normalized second curve values corresponding to the first curve values contained in the first logging curve data are represented. Since the second curve values included in the data of each second log are between 0 and 1, the fluctuation range of the ordinate of each second log is between 0 and 1.

S1017: and aiming at the second logging curve data corresponding to each target logging curve, selecting a second curve value corresponding to each target depth from the second logging curve data according to each target depth contained in the target depth area, carrying out mean value processing on the selected second curve values to obtain a target curve value corresponding to the target depth area, and taking the target curve value corresponding to the target depth area as the target logging data corresponding to the target logging curve in the target depth area.

For each target well log, the second well log data corresponding to the target well log includes second log values corresponding to each depth, for example, the second well log data includes: the second curve value for 100 meters is 40.22, the second curve value for 101 meters is 43.932, the second curve value for 102 meters is 56.33, the second curve value for 103 meters is 67.9, and so on. If the target depth region is 100-. By the method, the target curve value corresponding to each target depth area can be calculated.

In a possible embodiment, when the step S102 is executed to process the first image feature of the target imaging log image through the first convolutional neural network classification model to obtain the first reservoir classification result of the target reservoir corresponding to the target depth region in the target wellbore, the following steps may be specifically executed:

the method comprises the steps of taking an initial image feature matrix corresponding to a target imaging logging image as an input image feature matrix, carrying out first convolution processing on the input image feature matrix for a first preset number of times, carrying out first nonlinear transformation after each first convolution processing, and carrying out first pooling processing on a result obtained by each first nonlinear transformation.

And taking the image feature matrix obtained by the first pooling process each time as an input image feature matrix of the next first convolution process.

The first volume process includes: calculating the dot product of each first receptive field data in the first image characteristic matrix and the first convolution kernel respectively to obtain a first characteristic value corresponding to each first receptive field data, and constructing a second image characteristic matrix according to each first characteristic value; the matrix size corresponding to the first receptive field data is the same as the matrix size corresponding to the first convolution kernel.

The first non-linear transformation comprises: inputting each second characteristic value in the third image characteristic matrix into a preset activation function aiming at each third image characteristic matrix to obtain a third characteristic value corresponding to each second characteristic value in the third image characteristic matrix; in the preset activation function, for each second eigenvalue, when the second eigenvalue is greater than 0, determining the second eigenvalue as a third eigenvalue; when the second characteristic value is not more than 0, changing the second characteristic value to obtain a third characteristic value; constructing a fourth image feature matrix by using the third feature value;

the first pooling process includes: and extracting a maximum value from each second receptive field data in the fifth image characteristic matrix, and constructing a sixth image characteristic matrix according to the maximum value extracted from each second receptive field data.

And performing first full-connection processing on the output image characteristic matrix obtained by the last first pooling processing to obtain a first reservoir classification result corresponding to the target imaging logging image.

The first convolution neural network classification model takes the target imaging logging image as input, does not need to stretch and deform the target imaging logging image, and reserves the spatial structure of the target imaging logging image and information such as shape and texture in the image.

The first convolutional neural network classification model mainly comprises: the first rolling layer with the first preset times, the first pooling layer with the first preset times, the first nonlinear transformation with the first preset times and a first full-link layer. Performing a first convolution processing in the first convolution layer, specifically for performing feature extraction on the input characteristic features; the first pooling process is performed in the first pooling layer, and the first all-connected process is performed in the first all-connected layer.

In one embodiment, when there are two first winding layers, the first winding layer J is the first winding layer J₁And a first winding layer J₂The number of the first pooling layers is two, and the two first pooling layers are respectively a first pooling layer C₁And a first pooling layer C₂. Wherein the first coiled layer J₁As the first pooling layer C₁Input of (2), first pooling layer C₁As the first winding layer J₂Input of, the first winding layer J₂As the first pooling layer C₂Input of (2), first pooling layer C₂As an input to the first fully connected layer.

The convolutional layer is a core part of a convolutional neural network model and is used for extracting features of input features. The block simulates the working principle of human visual cortex, and the adjacent pixels of the image are considered to be closely related, and pixels with longer distances are not obviously related. The convolutional layer preserves the spatial structure of the image, using a locally perceptual convolution approach instead of the fully connected approach, this local region is also called the receptive field, where the size of the receptive field depends on the convolution kernel size.

Firstly, according to a target imaging logging image, an initial image characteristic matrix of the target imaging logging image is determined. Inputting the initial image feature matrix of the target imaging log image into the first convolution layer J₁Fig. 2 is a schematic diagram illustrating an initial image feature matrix, a first convolution kernel and a second image feature matrix provided in an embodiment of the present application, as shown in fig. 2, in a first convolution layer J₁In the above description, when the initial image feature matrix is a 5 × 5 matrix (i.e., a 5 × 5 matrix of input image features) and the first convolution kernel is a 3 × 3 matrix, a plurality of first field data are determined from the initial image feature matrix according to the size of the first convolution kernel. Here, since the first convolution kernel is a 3 × 3 matrix, the first field data is also a 3 × 3 matrix.

As shown in fig. 2, a 3 × 3 matrix of hatched portions (i.e., from the first row, the first column to the third row, the third column) serves as the first receptor field data Z₁And the second line, the second column, the third line and the fourth column from the first line to the third line as the second first field data Z₂And the third line from the third column on the first row to the fifth column on the third row as the third first field data Z₃And the first column from the second row to the third column from the fourth row as fourth first field data Z₄In this order, 9 pieces of first receptive field data can be obtained. I.e., there are duplicates between the first receptive field data.

Calculating first receptive field data Z₁Performing dot product on the first convolution kernel to obtain a first characteristic numerical value 9 corresponding to the first receptive field data Z1; calculating first receptive field data Z₂Dot product with the first convolution kernel to obtain the first receptive field data Z₂The corresponding first eigenvalue is 3; calculating first receptive field data Z₃Dot product with the first convolution kernel to obtain the first receptive field data Z₃The corresponding first eigenvalue is 9. In this order, 9 first eigenvalues may be obtained.

When calculating the dot product of the first receptive field data and the first convolution kernel, the dot product may be specifically obtained by dot-multiplying the numerical value at the corresponding position in the matrix corresponding to the first receptive field data and the matrix corresponding to the first convolution kernel, and calculating the sum of the dot-products. Using the first receptive field data Z₁Taking the dot product with the first convolution kernel as an example, the specific calculation process is as follows:

1×1+1×2+0×0+1×2+0×1+1×1+2×1+0×0+1×1＝9

and constructing a second image feature matrix according to the obtained 9 first feature values, namely determining the position of the first feature value corresponding to each first receptive field data on the second image feature matrix according to the position of each first receptive field data on the first feature matrix.

In this embodiment, the features of the imaging log are extracted by using the first convolution kernel with the size of 3 × 3, and it is considered that the superposition of a plurality of continuous small convolution kernels with the size of 3 × 3 has the same receptive field as that of the large convolution kernel, and introduces more first nonlinear transformations, so that the deep features of the input data can be better extracted.

Laminating the first winding layer J₁And taking the output second image feature matrix as a third image feature matrix, and performing first nonlinear transformation on the third image feature matrix, specifically, taking each numerical value in the matrix as a second characteristic numerical value in the third image feature matrix respectively for each third image feature matrix. Inputting each second feature value in the third image feature matrix into a preset activation function (i.e., an ELU activation function):

wherein y represents each second eigenvalue, ELU (y) represents a third eigenvalue obtained after the first nonlinear transformation,

is a fixed constant, and e is a fixed constant.

In the embodiment, the ELU activation function is adopted by the activation function of the network to replace the ReLU activation function, so that the problem of neuron necrosis is avoided, and the robustness of the model to noise is improved.

Inputting the fourth image feature matrix output by the first nonlinear transformation as a fifth image feature matrix to the first pooling layer C₁In (1). FIG. 3 is a schematic diagram of a fifth image feature matrix and a sixth image feature matrix provided in the embodiment of the present application, and the first pooling layer C is shown in FIG. 3₁The fifth image feature matrix is divided to obtain at least one second receptive field data, a maximum value is extracted from the second receptive field data aiming at each second receptive field data to obtain a maximum value corresponding to each second receptive field data, and a sixth image feature matrix is constructed by using the maximum value corresponding to each second receptive field data according to the position of each second receptive field data on the fifth image feature matrix. Wherein there is no overlapping portion between any two second receptive field data.

Illustratively, as shown in fig. 3, a 2 × 2 matrix of hatched portions (i.e., from a first row and a first column to a second row and a second column) is used as the first second receptive field data Y₁(ii) a From the third column on the first row to the fourth column on the second row as second field data Y₂(ii) a The third row, the first column, the fourth row, the second column as the third second field data Y₃(ii) a From the third row to the third column to the fourth row and the fourth column as fourth second field data Y₄. In this order, 4 second receptive field data can be obtained. Wherein the second receptive field data Y₁The maximum value in (1) is 8, and the second receptive field data Y₂The maximum value of 9, the second field data Y₃The maximum value in (1) is 7, and the second field data Y₄The maximum value of (3) is 5.

In a specific embodiment, if the second receptive field data is a 2 × 2 matrix and the fifth image feature matrix is a 3 × 3 matrix, it is necessary to complement 0 in the fifth image feature matrix, that is, complement 0 in the fourth column and the fourth row, and divide the fifth image feature matrix after complementing 0 to determine a plurality of second receptive field data.

The first pooling layer C₂As input to a first fully-connected layer, in which to the first pooling layer C₂And performing first full-connection processing on the output image characteristic matrix to obtain a first reservoir classification result corresponding to the target imaging logging image.

In a possible embodiment, when the step S103 is executed to process each target well logging data included in the target conventional well logging curve data through the second convolutional neural network classification model to obtain a second reservoir classification result corresponding to the target depth region in the target well, the following steps may be specifically executed:

the second convolution processing includes: dividing all the first logging data to obtain at least one third receptive field data; calculating the dot product of the first logging data and the second convolution kernel contained in the third receptive field data aiming at each third receptive field data to obtain a fourth characteristic numerical value corresponding to the first receptive field data, and constructing second logging data according to the fourth characteristic numerical value; the matrix size corresponding to the third receptive field data is the same as the matrix size corresponding to the second convolution kernel; the second nonlinear transformation includes: judging whether the third logging data are larger than 0 or not according to each third logging data; if the third logging data is larger than 0, determining the third logging data as fourth logging data; if the third logging data is not greater than 0, determining the product of the third logging data and a preset multiple as fourth logging data;

the second pooling process includes: dividing all the fifth logging data to obtain at least one fourth receptive field data; extracting a maximum value from the fourth receptive field data aiming at each fourth receptive field data, and constructing sixth logging data according to the maximum value extracted from each fourth receptive field data;

and performing second full-connection processing on the output logging data obtained by the last second pooling processing to obtain a second reservoir classification result corresponding to the target conventional logging curve data.

The second convolutional neural network classification model mainly comprises: the second convolutional layer for a second preset number of times, the second pooling layer for a second preset number of times, the leakage ReLU activation function, and the second fully-connected layer. Performing a second convolution process in the second convolution layer, performing a second pooling process in the second pooling layer, performing a second nonlinear transformation in the Leaky ReLU activation function, and performing a second fully-connected process in the second fully-connected layer.

In one embodiment, if there are three second convolution layers, there are respectively second convolution layers E₁Second convolution layer E₂Second convolution layer E₃(ii) a Three second pooling layers R₁Second pooling layer R₂Second pooling layer R₃. Wherein the second convolution layer E₁As a first input of the Leaky ReLU activation function, and as a second pooling layer R₁Of the second pooling layer R₁As a second convolution layer E₂Input of (2) a second convolution layer E₂As a second input to the Leaky ReLU activation function, and as a second output to the Leaky ReLU activation function as a second pooling layer R₂Of the second pooling layer R₂As a second convolution layer E₃Input of (2), the second convolution layer E₃As a third input of the Leaky ReLU activation function, and as a second pooling layer R₃Of the second pooling layer R₃As an input to the second fully connected layer.

Inputting each target logging data contained in the target conventional logging curve data corresponding to the target depth region into the second convolutional layer E₁In an exemplary case, when the target depth zone is 106-.

On the second convolution layer E₁In the method, target logging data is used as first logging data, all the target logging data are divided to obtain 3 third receptive field data which are respectively [0.144, 0.15 and 0.73 ]]、[0.66、0.137、0.475]、[0.48、0.118、0.504]。

And calculating the dot product of the first logging data and the second convolution kernel contained in the third receptive field data aiming at each third receptive field data to obtain a fourth characteristic numerical value corresponding to the third receptive field data. Specifically, the third receptive field data U is calculated₁[0.144、0.15、0.73]And the dot product of the first convolution kernel and the second convolution kernel obtains a fourth characteristic value. Calculating third receptive field data U₂[0.66、0.137、0.475]And the dot product of the first convolution kernel and the second convolution kernel to obtain a second fourth eigenvalue. Calculating third receptive field data U₃[0.48、0.118、0.504]And the dot product of the first convolution kernel and the second convolution kernel obtains a third fourth characteristic numerical value.

And constructing second logging data according to the fourth characteristic values, namely using the three fourth characteristic values as the second logging data. I.e. the second convolution layer E₁The output of (a) is second log data comprising three fourth eigenvalues.

A second convolution layer E₁The output includes three fourth characteristic numbersThe second log of values is input as third log data into a Leaky ReLU (recirculated linear unit, ReLU) activation function, specifically as follows:

wherein z is the third logging data input into the Leaky ReLU activation function, Leaky ReLU (z) is the fourth logging data output by the Leaky ReLU activation function, and β is the preset parameter, namely the preset multiple, in the Leaky ReLU activation function. The output of the Leaky ReLU activation function is fourth log data. The value range of the fourth logging data is between 0 and 1, and the decimal point is usually two digits after the decimal point.

Inputting the fourth logging data output by the Leaky ReLU activation function as fifth logging data into the second pooling layer R₁In another specific embodiment, if the input is to the second pooling layer R₁When the number of the fifth logging data is 6, the 6 fifth logging data are divided into 3 fourth receptive field data, which are (0.86, 0.44), (0.62, 0.46), (0.51, 0.65), respectively.

The maximum value of 0.86 was extracted from the fourth number of receptive fields (0.86, 0.44). The maximum value of 0.62 was extracted from the fourth receptive field data (0.62, 0.46). The maximum value of 0.65 was extracted from the fourth receptive field data (0.51, 0.65). The extracted 0.86, 0.62 and 0.65 are used to construct a sixth log. I.e. the second pooling layer R₁And outputting sixth logging data.

A second pooling layer R₃The output logging data is input to a second fully-connected layer, in which a second pooling layer R is formed₃And carrying out second full-connection processing on the output logging data, and outputting a second reservoir classification result corresponding to the target conventional logging curve data.

In a possible embodiment, after the step S105 is executed to determine the respective corresponding category of the target reservoir corresponding to each target depth region in the target wellbore, the following steps may be further specifically executed:

s1051: and judging whether the category corresponding to the target reservoir is the same as the categories corresponding to other target reservoirs adjacent to the target reservoir or not for each target reservoir.

And aiming at the target reservoir corresponding to each target depth region, 1 or 2 other target reservoirs adjacent to the target reservoir are provided. Illustratively, there are 1 other target reservoir adjacent to the target reservoir corresponding to the target depth zone 100-102, specifically the target reservoir corresponding to the target depth zone 102-104. The target reservoirs corresponding to the target depth zones 120-122 are 2 other adjacent target reservoirs, namely the target reservoir corresponding to the target depth zones 118-120 and the target reservoir corresponding to the target depth zones 122-124.

S1052: and if the category corresponding to the target reservoir is the same as the categories corresponding to other target reservoirs adjacent to the target reservoir, merging the target reservoir and the other target reservoirs adjacent to the target reservoir to obtain a synthetic reservoir.

In a specific embodiment, the category of the target reservoir corresponding to the target depth zone 120-.

Due to the limitation of the thickness of the reservoirs, the thickness of each type of high-quality reservoir or non-high-quality reservoir cannot be too thin, and in order to accurately determine the boundary of each high-quality reservoir and the non-high-quality reservoir, a region merging mode is adopted to further merge the target reservoir division results. Firstly, the depth difference corresponding to each divided reservoir is detected globally, for the fine reservoir division results, the reservoirs of the front part and the rear part of the layer are compared and merged with the surrounding reservoir categories, by using the technology, the fine division is merged, the abnormal classification results are eliminated, and the reservoir boundary is re-established.

In one possible embodiment, the first convolutional neural network classification model, the second convolutional neural network classification model and the target fully-connected layer are obtained by training in the following way:

s1001: constructing training samples for training a first convolutional neural network classification model, a second convolutional neural network classification model and a target full-connection layer; the training samples include: sample imaging logging images and sample conventional logging curve data corresponding to a sample depth area in a sample drill well, and a class label of a sample reservoir corresponding to the sample depth area; the sample conventional well logging curve data comprises sample well logging data corresponding to a plurality of sample well logging curves in a sample depth area.

When a sample imaging logging image is constructed, an initial sample imaging logging image corresponding to a sample well is obtained, wherein the initial sample imaging logging image comprises sample logging images corresponding to the depths of all samples in the sample well.

Removing the labeling information in the initial sample imaging log image, and segmenting the initial sample imaging log image after the labeling information is removed according to a preset depth interval to obtain a plurality of intermediate sample imaging log images and a sample depth area corresponding to each intermediate sample imaging log image.

And performing gray processing on the intermediate sample imaging log image aiming at each intermediate sample imaging log image to obtain a sample imaging log image corresponding to the intermediate sample imaging log image.

In this application, because the pixel of initial sample imaging log image is too big and contains the redundant information of mark nature, if directly use initial sample imaging log image to train first initial classification model, can lead to the model training difficulty, too much redundant information can bring too much noise, influences the final precision of first convolution neural network classification model. In order to improve the model training efficiency and precision, the image segmentation technology is adopted in the method, the initial sample imaging logging image is segmented, redundant information of the labeling property of the initial sample imaging logging image is removed, the initial sample imaging logging image is cut into a small-size sample imaging logging image beneficial to the training of the first convolution neural network classification model, and the training efficiency and precision of the first convolution neural network classification model are improved while the data volume is increased.

When the conventional well logging curve data of the sample is constructed, the method can be specifically executed according to the following steps:

acquiring conventional logging curve data of an initial sample corresponding to sample drilling; the conventional logging curve data of the initial sample comprises initial sample logging curve data corresponding to a plurality of initial sample logging curves; the initial sample logging curve data comprises initial sample curve values corresponding to all depths in the sample drilling;

carrying out smoothing filtering processing on the initial sample well logging curve aiming at the initial sample well logging curve corresponding to each initial sample well logging curve data to obtain a smooth sample well logging curve and first sample well logging curve data corresponding to the smooth sample well logging curve; the first sample logging curve data comprises first sample curve values corresponding to all depths in the sample drilling;

for each piece of first sample logging curve data, respectively carrying out normalization processing on each first sample curve value in the first sample logging curve data according to the maximum first sample curve value and the minimum first sample curve value in the first sample logging curve data to obtain a second sample curve value corresponding to each first sample curve value, and determining second sample logging curve data and a sample logging curve corresponding to the second sample curve data according to the second sample curve value corresponding to each depth; the value range of the second sample curve value is 0-1;

and aiming at the second sample logging curve data corresponding to each sample logging curve, selecting a second sample curve value corresponding to each sample depth from the second sample logging curve data according to each sample depth contained in the sample depth area, carrying out mean value processing on the selected second sample curve values to obtain a sample curve value corresponding to the sample depth area, and taking the sample curve value corresponding to the sample depth area as the sample logging data corresponding to the sample logging curve in the sample depth area.

And in the sample drilling, each sample depth area corresponds to a sample imaging logging image and a sample conventional logging curve data.

S1002: and inputting the sample imaging logging image into the first initial classification model, and processing the second image characteristics of the sample imaging logging image through the first initial classification model to obtain a first prediction result of the sample reservoir.

The first initial classification model includes: the method comprises the steps of performing first initial convolution layer with a first preset number of times, a first initial pooling layer with a first preset number of times, an ELU activation function and a first initial full-link layer, performing first sample convolution processing in the first initial convolution layer, performing first sample pooling processing in the first initial pooling layer, performing first initial nonlinear transformation in the ELU activation function, and performing first sample full-link processing in the first initial full-link layer.

Taking initial sample imaging logging image features corresponding to the sample imaging logging images as an input sample image feature matrix, performing first sample convolution processing on the input sample image feature matrix for a first preset number of times, performing first initial nonlinear transformation after the first sample convolution processing each time, and performing first sample pooling processing on a result obtained by the first initial nonlinear transformation each time;

taking a sample image feature matrix obtained by each time of first sample pooling as an input sample image feature matrix of next first sample convolution processing;

the first sample convolution process includes: in the first initial convolution layer, calculating the dot product of each fifth receptive field data in the first sample image characteristic matrix and the first sample convolution kernel respectively to obtain a fifth characteristic value corresponding to each fifth receptive field data, and constructing a second sample image characteristic matrix according to each fifth characteristic value; the matrix size corresponding to the fifth receptive field data is the same as the matrix size corresponding to the first sample convolution kernel;

the first initial nonlinear transformation comprises: taking a second sample image feature matrix output by the first initial convolutional layer as a third sample image feature matrix, and performing first initial nonlinear transformation on the third sample image feature matrix; specifically, for each third sample image feature matrix, inputting each sixth feature value in the third sample image feature matrix into the ELU activation function, and outputting a value corresponding to each sixth feature value; in the ELU activation function, for each sixth characteristic value, when the sixth characteristic value is greater than 0, determining the sixth characteristic value as a seventh characteristic value; when the sixth characteristic value is not greater than 0, transforming the sixth characteristic value to obtain a seventh characteristic value; constructing a fourth sample image feature matrix by using the seventh feature value;

the first sample pooling process comprises: in the first initial pooling layer, extracting a maximum value from each sixth receptive field data in the fifth sample image characteristic matrix, and constructing a sixth sample image characteristic matrix according to the maximum value extracted from each sixth receptive field data;

and in the first initial full-connection layer, performing first sample full-connection processing on an output sample image characteristic matrix obtained by the last first sample pooling processing to obtain a first prediction result corresponding to the sample imaging logging image. The first prediction result comprises the probability of the sample reservoir corresponding to the sample depth region predicted by the first initial classification model on each category.

S1003: and inputting the sample conventional well logging curve data into a second initial classification model, and processing each sample well logging data contained in the sample conventional well logging curve data through the second initial classification model to obtain a second prediction result of the sample reservoir.

The second initial classification model includes: a second initial convolutional layer for a second preset number of times, a second initial pooling layer for a second preset number of times, a Leaky ReLU activation function, and a second initial fully-connected layer. Performing a second sample convolution process in the second initial convolution layer, performing a second sample pooling process in the second initial pooling layer, performing a second initial nonlinear transformation in the Leaky ReLU activation function, and performing a second sample full-join process in the second initial full-join layer.

Taking each sample logging data contained in the sample conventional logging curve data as input sample logging data, performing second sample convolution processing on the input sample logging data for a second preset time, performing sample nonlinear transformation after each second sample convolution processing, and performing second sample pooling processing on a result obtained by each sample nonlinear transformation;

performing pooling processing on each second sample to obtain sample characteristic data serving as input sample logging data of the next second sample convolution processing;

the second sample convolution processing includes: in the second initial convolutional layer, dividing all the first sample logging data to obtain at least one seventh receptive field data; calculating the dot product of the first sample logging data and the second convolution kernel contained in the seventh receptive field data aiming at each seventh receptive field data to obtain an eighth characteristic numerical value corresponding to the seventh receptive field data, and constructing second sample logging data according to the eighth characteristic numerical value; the second sample nonlinear transformation includes: in the Leaky ReLU activation function, judging whether the third sample logging data is greater than 0 or not aiming at each third sample logging data; if the third sample logging data is larger than 0, determining the third sample logging data as fourth sample logging data; if the third sample logging data is not greater than 0, determining the product of the third sample logging data and a preset multiple as fourth sample logging data;

the second pooling process includes: dividing all the fifth sample well logging data to obtain at least one eighth receptive field data; extracting a maximum value from the eighth receptive field data aiming at each eighth receptive field data, and constructing sixth sample logging data according to the maximum value extracted from each eighth receptive field data;

and in the second initial full-connection layer, performing second sample full-connection processing on output sample logging data obtained by the last second pooling processing to obtain a second prediction result corresponding to the conventional logging curve data of the sample. The second prediction result comprises the probability of the sample reservoir corresponding to the sample depth region predicted by the second initial classification model on each category.

S1004: and inputting a third prediction result obtained by splicing the first prediction result and the second prediction result of the sample reservoir into the initial full-connection layer, and outputting a prediction classification result of the sample reservoir.

The prediction classification result comprises the probabilities of sample reservoirs corresponding to the sample depth region on each category, which are predicted after comprehensively considering two data, namely a sample imaging logging image and sample conventional logging curve data.

S1005: and calculating a loss value according to the prediction classification result and the class label of the sample reservoir.

Determining the prediction category of the sample reservoir according to the prediction classification result of the sample reservoir; and calculating a loss value through a cross entropy loss function according to the prediction category and the category label of the sample reservoir.

In one specific embodiment, in the case of binary classification, we assume the predicted class probabilities are p (e.g., probability of good reservoir) and 1-p (e.g., probability of non-good reservoir), then the loss function expression is:

wherein U is a loss value, M is the number of training samples, U_jFor the loss result corresponding to the jth training sample, t_jClass label for jth training sample, q_jAnd predicting the probability of being a good reservoir for the jth training sample.

In another specific embodiment, in the case of multi-classification, i.e. there are multiple reservoir categories, the loss function expression is:

wherein L is a loss value, N is the number of training samples, L_iFor the loss result corresponding to the ith training sample, G is the classified class quantity, y_icClass label for ith training sample，P_icThe probability of class C is predicted for the ith training sample.

S1006: and updating parameters in the first initial classification model, the second initial classification model and the initial full-connection layer by using the loss value, and training the updated first initial classification model, the second initial classification model and the initial full-connection layer again until the training times reach a preset training time, so as to obtain the first convolutional neural network classification model, the second convolutional neural network classification model and the target full-connection layer.

And after the training is finished, taking the first initial convolutional layer, the first initial pooling layer and the first initial fully-connected layer in the first initial classification model as the first convolutional layer, the first pooling layer and the first fully-connected layer in the first convolutional neural network classification model. And taking the second initial convolutional layer, the second initial pooling layer, the initial activation function and the second initial fully-connected layer in the second initial classification model as the second convolutional layer, the second pooling layer, the activation function and the second fully-connected layer in the second convolutional neural network classification model.

In a possible embodiment, after the step S105 is executed to determine the category of the target reservoir according to the target reservoir classification result, the following steps may be further executed:

and evaluating the storage amount of the target substance stored in the target reservoir according to the category of the target reservoir.

The target reservoirs differ in their categories, resulting in different amounts of the target substance stored in the target reservoirs. The target substance includes petroleum, natural gas, etc. In a specific embodiment, the categories of the target reservoirs are classified into a premium type reservoir, and a non-premium type reservoir. The storage amount of the target substances stored in the high-quality reservoirs is the largest, and the storage amount of the target substances stored in the non-high-quality reservoirs is the smallest.

In one possible embodiment, after the step S1052 is executed, if the category corresponding to the target reservoir is the same as the categories corresponding to other target reservoirs adjacent to the target reservoir, the target reservoir and the other target reservoirs adjacent to the target reservoir are merged to obtain a synthesized reservoir, the following steps may be further specifically executed:

and evaluating the storage amount of the target substance stored in the synthetic composition according to the number and the category of the target reservoirs contained in the synthetic reservoirs.

Example two:

based on the same technical concept, an embodiment of the present application further provides a reservoir classification device, and fig. 4 shows a schematic structural diagram of the reservoir classification device provided by the embodiment of the present application, and as shown in fig. 4, the device includes:

a first obtaining module 401, configured to obtain a target imaging logging image and target conventional logging curve data corresponding to a target depth region in a target borehole; the target conventional logging curve data comprises target logging data corresponding to a plurality of target logging curves in the target depth region;

a first input module 402, configured to input the target imaging log image into a first convolutional neural network classification model trained in advance, and process a first image feature of the target imaging log image through the first convolutional neural network classification model to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target wellbore;

a second input module 403, configured to input the target conventional well logging curve data into a second convolutional neural network classification model trained in advance, and process each target well logging data included in the target conventional well logging curve data through the second convolutional neural network classification model to obtain a second reservoir classification result of the target reservoir corresponding to the target depth region in the target well;

a third input module 404, configured to input a splicing result obtained by splicing the first reservoir classification result and the second reservoir classification result into a pre-trained target fully-connected stratum, so as to obtain a target reservoir classification result of the target reservoir;

a first determining module 405, configured to determine a category of the target reservoir according to the target reservoir classification result.

Optionally, the second obtaining module is configured to obtain an initial imaging log image corresponding to the target drilling well before the first obtaining module 401 obtains the target imaging log image corresponding to the target depth region in the target drilling well; the initial imaging logging image comprises imaging logging information corresponding to each depth in the target well;

the segmentation module is used for segmenting the initial imaging logging image according to a preset depth interval to obtain a plurality of imaging logging images and the target depth area corresponding to each imaging logging image;

Optionally, the third obtaining module is configured to obtain initial conventional well logging curve data corresponding to the target well before the first obtaining module 401 obtains target conventional well logging curve data corresponding to a target depth region in the target well; the initial conventional logging curve data comprises initial logging curve data corresponding to a plurality of initial logging curves; the initial logging curve data comprises initial curve values corresponding to all depths in the target well;

Optionally, the first input module 402, when configured to process the first image feature of the target imaging log image through the first convolutional neural network classification model to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target wellbore, is specifically configured to:

Optionally, when the second input module 403 is configured to process each target well logging data included in the target conventional well logging data through the second convolutional neural network classification model to obtain a second reservoir classification result of the target reservoir corresponding to the target depth region in the target well, specifically, the second input module is configured to:

taking each target logging data contained in the target conventional logging curve data as input logging data, performing second convolution processing on the input logging data for a second preset number of times, performing nonlinear transformation after each second convolution processing, and performing second pooling processing on a result obtained by each nonlinear transformation;

the second pooling process comprises: dividing all the fifth logging data to obtain at least one fourth receptive field data; extracting a maximum value from the fourth receptive field data aiming at each fourth receptive field data, and constructing sixth logging data according to the maximum value extracted from each fourth receptive field data; and carrying out second full-connection processing on output logging data obtained by the last second pooling processing to obtain a second reservoir classification result corresponding to the target conventional logging curve data.

Optionally, the method further includes:

the fourth input module is used for inputting the sample imaging logging image into the first initial classification model, and calculating the second image characteristics of the sample imaging logging image through the first initial classification model to obtain a first prediction result of the sample reservoir;

the fifth input module is used for inputting the sample conventional well logging curve data into a second initial classification model, and performing weighted calculation on each sample well logging data contained in the sample conventional well logging curve data through the second initial classification model to obtain a second prediction result of the sample reservoir;

the calculation module is used for calculating a loss value according to the prediction classification result of the sample reservoir and the class label;

Optionally, the method further includes:

For the specific implementation steps and principles, reference is made to the description of the first embodiment, which is not repeated herein.

Example three:

based on the same technical concept, an embodiment of the present application further provides an electronic device, and fig. 5 shows a schematic structural diagram of the electronic device provided in the embodiment of the present application, and as shown in fig. 5, the electronic device 500 includes: a processor 501, a memory 502 and a bus 503, wherein the memory stores machine-readable instructions executable by the processor, when the electronic device is operated, the processor 501 and the memory 502 communicate with each other through the bus 503, and the processor 501 executes the machine-readable instructions to execute the steps of the method described in the first embodiment. For the specific implementation steps and principles, reference is made to the description of the first embodiment, which is not repeated herein.

Example four:

based on the same technical concept, a computer-readable storage medium is further provided in a fourth embodiment of the present application, where a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to perform the method steps in the first embodiment. For the specific implementation steps and principles, reference is made to the description of the first embodiment, which is not repeated herein.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed coupling or direct coupling or communication connection between each other may be through some communication interfaces, indirect coupling or communication connection between devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present application and are intended to be covered by the appended claims. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method of reservoir classification, comprising:

inputting the target imaging logging image into a pre-trained first convolutional neural network classification model, and processing a first image feature of the target imaging logging image through the first convolutional neural network classification model to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target well;

2. The reservoir classification method according to claim 1, wherein before obtaining the target imaging log image corresponding to the target depth region in the target borehole, the method further comprises:

3. The reservoir classification method as claimed in claim 1, wherein before the obtaining of the target conventional well log data corresponding to the target depth zone position in the target wellbore, further comprising:

for each initial logging curve data, performing smooth filtering processing on the initial logging curve data to obtain first logging curve data; the first logging curve data comprise first curve values corresponding to all depths in the target well;

4. The reservoir classification method of claim 1, wherein the processing, by the first convolutional neural network classification model, the first image features of the target imaging log image to obtain a first reservoir classification result of a target reservoir corresponding to the target depth region in the target borehole comprises:

5. The reservoir classification method of claim 1, wherein the processing, by the second convolutional neural network classification model, each of the target well log data included in the target conventional well log data to obtain a second reservoir classification result corresponding to the target depth region in the target well comprises:

6. The reservoir classification method of claim 1, wherein after determining the respective category of the target reservoir corresponding to each of the target depth zones in the target wellbore, further comprising:

7. The reservoir classification method of claim 1, wherein the first convolutional neural network classification model, the second convolutional neural network classification model and the target fully-connected layer are trained by:

constructing training samples for training the first convolutional neural network classification model, the second convolutional neural network classification model and the target fully-connected layer; the training samples include: the method comprises the steps that sample imaging logging images and sample conventional logging curve data corresponding to a sample depth area in a sample drill well, and a class label of a sample reservoir layer corresponding to the sample depth area; the sample conventional logging curve data comprises sample logging data corresponding to a plurality of sample logging curves in the sample depth region;

8. The reservoir classification method as claimed in claim 1, wherein after determining the category of the target reservoir according to the target reservoir classification result, the method further comprises:

9. A reservoir sorting apparatus, comprising:

10. An electronic device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when the electronic device is operating, the machine-readable instructions when executed by the processor performing the steps of the method of any of claims 1 to 8.

11. A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, is adapted to carry out the steps of the method according to any one of claims 1 to 8.