CN117388933A

CN117388933A - Neural network nuclear magnetic logging curve prediction method and device based on feature enhancement

Info

Publication number: CN117388933A
Application number: CN202210766852.6A
Authority: CN
Inventors: 王兵; 王贵重; 郭翔; 阴学彬; 赵亮; 易贝贝
Original assignee: China National Petroleum Corp; BGP Inc
Current assignee: China National Petroleum Corp; BGP Inc
Priority date: 2022-06-30
Filing date: 2022-06-30
Publication date: 2024-01-12

Abstract

The application discloses a neural network nuclear magnetic logging curve prediction method and device based on feature enhancement, and relates to the technical field of seismic exploration. The method comprises the following steps: acquiring target wells Duan Quxian corresponding to the conventional logging curves respectively; performing curve intersection analysis on target wells Duan Quxian corresponding to the conventional logging curves respectively, and removing logging data of the expanded section in the target well section curve to obtain a stable well Duan Quxian; performing standard deviation normalization processing and combined analysis on the stable well section curve to obtain a normalized well Duan Quxian; and carrying out porosity prediction analysis on the normalized well Duan Quxian through a target long-short term memory network to obtain a nuclear magnetic porosity prediction curve of the target well section. The change trend of conventional logging data along with depth and the front-back association of the data can be analyzed through a target long-short-term memory network, and the characteristic dimension is expanded through combined analysis, so that the accuracy of the nuclear magnetic porosity curve obtained through prediction is improved.

Description

Neural network nuclear magnetic logging curve prediction method and device based on feature enhancement

Technical Field

The embodiment of the application relates to the technical field of seismic exploration, in particular to a neural network nuclear magnetic logging curve prediction method and device based on characteristic enhancement.

Background

Compared with the conventional logging curve, the nuclear magnetic logging curve more accurately and intuitively reflects the characteristics of porosity, permeability, saturation and the like of the reservoir, has a good corresponding relation with an oil layer, and has good positive correlation with the porosity of the core. The nuclear magnetic logging curve can be used as an important logging curve for classifying and evaluating tight oil and shale oil dessert reservoirs in oil fields. However, due to the complex underground condition, various unexpected and unavoidable problems such as well diameter expansion, instrument faults and the like exist in the measuring process, and partial well section data distortion or missing often occurs in practical application. These missing logging data can present significant challenges for subsequent reservoir evaluation and prediction efforts, and re-logging presents significant limitations in actual production. Reconstruction of distorted or missing nuclear magnetic logs is therefore particularly important without adding additional measurement costs.

In the related art, based on information such as geological parameters and mechanical parameters, a missing nuclear magnetic logging curve can be directly generated through inversion of a physical model.

However, the physical model is often based on strong assumptions, which are a great simplification of the true formation information. And different physical models are required to be selected according to different situations, and the selection process has strong subjectivity and depends on expert experience and field knowledge. Thus, the accuracy of the nuclear magnetic log generated based on the physical model is low.

Disclosure of Invention

The embodiment of the application provides a neural network nuclear magnetic logging curve prediction method and device based on characteristic enhancement, which improves the accuracy of predicting the nuclear magnetic logging curve, and the technical scheme is as follows:

in one aspect, a neural network nuclear magnetic logging curve prediction method based on feature enhancement is provided, and the method comprises the following steps:

acquiring a plurality of conventional well logging curves, wherein the conventional well logging curves are all-well section well logging data measured on a target well aiming at different measurement parameters;

performing well-seismic joint interpretation on a plurality of conventional well logging curves to obtain target wells Duan Quxian corresponding to the conventional well logging curves respectively, wherein the target well section curves are well logging data corresponding to target well sections in the target wells;

performing curve intersection analysis on target wells Duan Quxian corresponding to a plurality of conventional logging curves respectively, and removing the logging data of the expanded diameter section in the target well section curve to obtain a stable well section curve, wherein the logging data of the expanded diameter section is abnormal data of well diameter expansion in the target well section curve;

performing standard deviation normalization processing and combined analysis on the stable well section curve to obtain a normalized well Duan Quxian;

and carrying out porosity prediction analysis on the normalized well Duan Quxian through a target long-short term memory network to obtain a nuclear magnetic porosity prediction curve of the target well section.

In another aspect, a feature enhancement-based nuclear magnetic log prediction apparatus is provided, the apparatus comprising:

the data acquisition module is used for acquiring a plurality of conventional logging curves, wherein the conventional logging curves are all-well section logging data measured on a target well aiming at different measurement parameters;

the joint interpretation module is used for performing well-seismic joint interpretation on a plurality of conventional well logging curves to obtain target wells Duan Quxian corresponding to the conventional well logging curves respectively, and the target well section curves are well logging data corresponding to target well sections in the target wells;

the intersection analysis module is used for carrying out curve intersection analysis on target wells Duan Quxian corresponding to a plurality of conventional logging curves respectively, removing the expanded diameter section logging data in the target well section curves to obtain stable well section curves, wherein the expanded diameter section logging data are abnormal data of well diameter expansion in the target well section curves;

the normalization processing module is used for carrying out standard deviation normalization processing and combination analysis on the stable well section curve to obtain a normalization well Duan Quxian;

and the prediction analysis module is used for carrying out porosity prediction analysis on the normalized well Duan Quxian through a target long-short-term memory network to obtain a nuclear magnetic porosity prediction curve of the target well section.

In another aspect, a computer device is provided, the computer device including a processor and a memory having at least one instruction, at least one program, a set of codes, or a set of instructions stored therein, the at least one instruction, the at least one program, the set of codes, or the set of instructions loaded and executed by the processor to implement a feature-enhanced neural network nuclear magnetic log prediction method according to any of the embodiments of the present application.

In another aspect, a computer readable storage medium having at least one program code stored therein is provided, the at least one program code loaded and executed by a processor to implement a feature-enhanced based neural network nuclear magnetic log curve prediction method according to any of the embodiments of the present application.

In another aspect, a computer program product or computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the neural network nuclear magnetic log prediction method based on feature enhancement according to any one of the embodiments of the present application.

The beneficial effects that technical scheme that this application embodiment provided include at least:

by carrying out well-seismic joint interpretation on a plurality of conventional logging curves, a target well Duan Quxian corresponding to a target well section in a target well is obtained, and by analyzing a part of well sections in the target well instead of the whole well section, the value range difference of data needing to be analyzed at a time is reduced, and the accuracy of a nuclear magnetic porosity prediction curve is improved; the method comprises the steps of performing curve intersection analysis on a target well Duan Quxian, removing abnormal data in the target well Duan Quxian, and performing standard deviation normalization on the curve from which the abnormal data are removed to obtain a normalized well Duan Quxian; finally, the curve is predicted and analyzed through the target long-short-term memory network to obtain a nuclear magnetic porosity prediction curve of the target well section, and the short-term memory and the long-term memory are combined through a gate mechanism by the target long-short-term memory network, so that the change trend of conventional logging data along with depth and the front-back association of the data can be analyzed through the target long-short-term memory network, and the accuracy of the predicted nuclear magnetic porosity curve is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic illustration of an implementation environment provided by an exemplary embodiment of the present application;

FIG. 2 is a flow chart of a feature-enhanced based neural network nuclear magnetic log prediction method provided in one exemplary embodiment of the present application;

FIG. 3 is a flow chart of a feature-enhanced based neural network nuclear magnetic log prediction method provided in another exemplary embodiment of the present application;

FIG. 4 is a block diagram of elements of an LSTM network provided in accordance with one exemplary embodiment of the present application;

FIG. 5 is a flow chart of a feature-enhanced based neural network nuclear magnetic log prediction method provided in another exemplary embodiment of the present application;

FIG. 6 is a schematic illustration of a feature-enhanced based nuclear magnetic log prediction interface provided in one exemplary embodiment of the present application;

FIG. 7 is a comparison of a true nuclear magnetic log and a predicted nuclear magnetic log provided in one exemplary embodiment of the present application;

FIG. 8 is a block diagram of a feature-enhanced based nuclear magnetic log prediction device provided in one exemplary embodiment of the present application;

FIG. 9 is a block diagram of a feature-enhanced based nuclear magnetic log prediction device provided in accordance with another exemplary embodiment of the present application;

fig. 10 is a block diagram of a server according to an exemplary embodiment of the present application.

Detailed Description

For the purposes of making the objects, technical solutions and advantages of the present application more apparent, the embodiments of the present application will be described in further detail below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.

The terms "first," "second," and the like in this application are used for distinguishing between similar elements or items having substantially the same function and function, and it should be understood that there is no logical or chronological dependency between the terms "first," "second," and no limitation on the amount or order of execution.

The recurrent neural network (Recurrent Neural Network, RNN) is a type of recurrent neural network which takes sequence data as input, performs recursion in the evolution direction of the sequence, and all nodes (circulation units) are connected in a chained manner. The long short term memory (Long Short Term Memory, LSTM) network provided in the embodiments of the present application is a special RNN that adds element states and gate structures to the hidden layer of the original RNN. Wherein the cell state is used to hold information and carry information along a time series, even if information of an earlier time step can be transferred to a later moment; the gate structure may add or delete information to the cell state. The LSTM network combines short-term memory and long-term memory through a gate mechanism, solves the problem that only short-term history input can not be effectively solved, and solves the problems of gradient disappearance and gradient explosion to a certain extent.

In the related art, other missing curves are predicted through the existing partial complete well logging curves, and the missing well logging curves are generated through direct inversion through physical model trial mainly according to information such as geological parameters and mechanical parameters. However, these physical models are often based on strong assumptions, which are a great simplification of the true formation information. And different physical models are required to be selected according to different situations, and the selection process has strong subjectivity and depends on expert experience and field knowledge. Therefore, the quality of the log generated based on the physical model is difficult to be effectively ensured. The embodiment of the application provides a characteristic-enhanced neural network nuclear magnetic logging curve prediction method, which is mainly used for predicting and generating a usable nuclear magnetic logging curve through a long-term and short-term memory network aiming at wells with unmeasured or partially missing nuclear magnetic logging data. Firstly, acquiring a plurality of conventional well logging curves of a target well, performing well-shock joint interpretation on the conventional well logging curves, and selecting a well logging curve corresponding to a target well section of the target well as a target well Duan Quxian; secondly, eliminating abnormal data of well diameter expansion in a target well section curve to obtain a stable well Duan Quxian; then, standard deviation normalization processing is carried out on the stable well section curve, so that the data distribution range is consistent; and finally, inputting the standard deviation normalized stable logging curve into an LSTM network, and predicting to obtain a nuclear magnetic porosity prediction curve of the target well section.

Fig. 1 is a schematic diagram of an implementation environment provided in an exemplary embodiment of the present application, as shown in fig. 1, where the implementation environment includes a terminal 110, a server 120, and a communication network 130, where the terminal 110 and the server 120 are connected through the communication network 130, and in some alternative embodiments, the communication network 130 may be a wired network or a wireless network, and this embodiment is not limited to this.

In some alternative embodiments, terminal 110 is a smart phone, tablet, notebook, desktop computer, smart home appliance, smart car terminal, smart speaker, digital camera, etc., but is not limited thereto. Optionally, the terminal 110 is provided with a target application, which may be a conventional application, a cloud application, an applet or an application module in a host application, or a web platform, which is not limited in this embodiment. Optionally, the target application program is provided with a nuclear magnetic porosity curve prediction function, which is schematically shown in fig. 1, when the nuclear magnetic porosity curve of the target well needs to be predicted, the terminal 110 uploads a plurality of conventional logging curves of the target well to the server 120, the server 120 analyzes the plurality of conventional logging curves, predicts the nuclear magnetic porosity curve of the target well, and feeds back the nuclear magnetic porosity curve to the terminal 110.

In some alternative embodiments, server 120 is configured to provide background services for a target application installed in terminal 110, and optionally, a target long-short-term memory network is provided in server 120. Illustratively, after receiving the plurality of conventional log curves of the target well, the server 120 performs a well-to-seismic joint interpretation on the plurality of conventional log curves, and selects a log curve corresponding to a target well section of the target well as a target well Duan Quxian; secondly, eliminating abnormal data of well diameter expansion in a target well section curve to obtain a stable well Duan Quxian; then, standard deviation normalization processing is carried out on the stable well section curve to obtain a normalized well Duan Quxian; and finally, inputting the normalized well section curve into a target long-short-term memory network, outputting to obtain a nuclear magnetic porosity prediction curve of the target well section, and transmitting the nuclear magnetic porosity prediction curve to the terminal 110, wherein the terminal 110 can display the nuclear magnetic porosity prediction curve.

In some alternative embodiments, the target long-short term memory network may also be deployed on the terminal 110 side, where the nuclear magnetic porosity curve prediction function is implemented locally by the terminal 110, without the aid of the server 120, which is not limited in the embodiments of the present application.

It should be noted that the server 120 can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and can also be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs (Content Delivery Network, content delivery networks), and basic cloud computing services such as big data and artificial intelligence platforms.

It should be noted that, information (including but not limited to user equipment information, user personal information, etc.), data (including but not limited to data for analysis, stored data, presented data, etc.), and signals referred to in this application are all authorized by the user or are fully authorized by the parties, and the collection, use, and processing of relevant data is required to comply with relevant laws and regulations and standards of relevant countries and regions. For example, conventional logs referred to in this application are acquired with sufficient authorization.

With reference to the above description and the implementation environment, the method for predicting a nuclear magnetic logging curve of a neural network based on feature enhancement provided in the embodiments of the present application is described, and fig. 2 is a flowchart of the method for predicting a nuclear magnetic logging curve of a neural network based on feature enhancement provided in the embodiments of the present application, where the method is applied to a terminal or a server shown in fig. 1, and as shown in fig. 2, the method includes:

In step 201, a plurality of conventional log curves are acquired.

The plurality of conventional log curves are full interval log data measured for the target well for different measured parameters.

Full interval log data refers to log data for all depth intervals of the target well. Logging refers to a method for measuring physical parameters such as conductive properties, acoustic properties, radioactivity, electrochemical properties and the like of underground rock formations along a drilling section by using underground special instruments, and different logging methods and corresponding logging curves are adopted for different physical parameters.

Conventional logs include, among others, resistivity-like logs and non-resistivity-like logs. The resistivity logging curves include micro resistivity logging curves, lateral logging curves, induction logging curves and the like; the non-resistivity logging curves include natural gamma logging curves, natural potential logging curves, borehole logging curves, acoustic time difference logging curves, compensation density logging curves, compensation neutron logging curves and the like. Alternatively, the conventional log is plotted on the abscissa as each physical parameter (e.g., borehole diameter size, natural gamma value, resistivity, etc.), and on the ordinate as the measured depth of the target well.

Illustratively, nine conventional log curves for the target well (i.e., the nine conventional log curves mentioned in the above description) are acquired.

Step 202, performing well-seismic joint interpretation on the plurality of conventional well-logging curves to obtain target wells Duan Quxian corresponding to the plurality of conventional well-logging curves respectively.

The target interval curve is log data corresponding to a target interval in the target well.

Optionally, the target well is subjected to geological stratification by well-to-seismic joint interpretation, and well-logging data of the well section in the plurality of conventional well-logging curves is selected as the target well Duan Quxian according to the uniform geological stratification well section.

Optionally, the method for acquiring the target well section curves corresponding to the plurality of conventional well logging curves respectively further includes the following steps:

1. seismic data of the target well is acquired.

The seismic data is seismic data reflecting the geological features of the target well.

Illustratively, the seismic data is a seismic wave acquired by manually exciting a tiny harmless earthquake in a target well, and the ordinate of the seismic data is the reflection time of the seismic wave.

2. And carrying out well-seismic joint interpretation on the seismic data and a plurality of conventional logging curves to obtain joint interpretation results.

The joint interpretation results are used to indicate geologic stratification conditions in the target well.

Illustratively, the stratum interpretation is performed on the target well based on the seismic data and the conventional logging curve of the target well in combination with geological knowledge, geophysical knowledge and the like, and the obtained stratum interpretation result is the joint interpretation result. Optionally, the joint interpretation result includes a geological stratification identification and a depth interval corresponding to the geological stratification identification.

3. Based on the joint interpretation results, a target depth interval is selected, the target depth interval being used to indicate a depth position on a section of the target well Duan Zaiquan.

Logging data of the plurality of conventional logging curves in the target depth interval is selected, and target wells Duan Quxian corresponding to the plurality of conventional logging curves are obtained.

Illustratively, by well-to-seismic joint interpretation, the target well is divided into three sections A, B, C, and if the section A is selected, logging data of a depth interval corresponding to the section A (i.e., corresponding top depth and bottom depth) is obtained as the target well Duan Quxian.

And 203, performing curve intersection analysis on the target wells Duan Quxian corresponding to the conventional logging curves respectively, and removing the logging data of the expanded section in the target section curve to obtain a stable section curve.

The well logging data of the expanded diameter section are abnormal data of well diameter expansion in a target well section curve.

In some alternative embodiments, the target interval curves include a compensated neutron curve and a compensated density curve, and then curve intersection analysis is illustrated with the compensated neutron curve and the compensated density curve:

constructing a density neutron intersection plan based on the compensation neutron curve and the compensation density curve; removing abnormal data points distributed in a designated area of a quartz model line in a density neutron intersection plan view; and taking the compensation neutron curve with the abnormal data points removed and the compensation density curve with the abnormal data points removed as stable well section curves.

Schematically, constructing a density neutron intersection plan by taking the abscissa as compensation density data of a target well section and the ordinate as compensation neutron data of the target well section, and marking a quartz model line on the intersection plan; the compensated neutron data is used for indicating the hydrogen index of the stratum of the target well section, the compensated density data is used for indicating the density of the stratum of the target well section, and the quartz model line is data obtained by measuring the compensated density and the compensated neutrons of single quartz ore and is generally a straight line with an approximately 45-degree included angle with the abscissa of a density neutron intersection plan view. The data points distributed at the upper left corner of the quartz model line represent expanded diameter section data, namely, abnormal data of well diameter expansion, the data points are removed, and the rest data points form a compensation neutron curve and a compensation density curve, namely, a stable well section curve.

And 204, performing standard deviation normalization processing and combined analysis on the stable well section curve to obtain a normalized well Duan Quxian.

In some alternative embodiments, standard deviation normalization is performed on the stable well section curves to obtain candidate normalized wells Duan Quxian; performing combined analysis on the candidate normalized wells Duan Quxian to construct a combined log; performing standard deviation normalization processing on the combined log curve to obtain a combined normalized well Duan Quxian; candidate normalized well Duan Quxian and combined normalized interval curve are determined to be normalized well Duan Quxian.

Schematically, after standard deviation normalization processing is performed on a single stable well section curve, the curves can be subjected to combined analysis to construct a new curve, namely a combined well logging curve, then standard deviation normalization processing is performed on the combined well logging curve, and finally, the normalization result of the single stable well section curve and the normalization result of the combined well logging curve are input into a target long-short-term memory network together for analysis.

Alternatively, after combining two or more of the stability wells Duan Quxian to obtain a plurality of combined log curves, geological awareness may be combined to select a curve that reflects the reservoir characteristics of the target well as much as possible from the plurality of combined log curves as the combined log curve that is ultimately needed for analysis.

In some alternative embodiments, the stable wellbore section curves include resistivity-like logs and non-resistivity-like logs; the resistivity logging curve refers to a micro resistivity logging curve, a lateral logging curve and an induction logging curve, the non-resistivity logging curve refers to a natural gamma logging curve, a natural potential logging curve, a borehole diameter logging curve, a sonic time difference logging curve, a compensation density logging curve and a compensation neutron logging curve, and standard deviation normalization processing is performed on the stable well section curve to obtain a candidate normalized well section curve, which comprises the following steps:

Carrying out logarithmic transformation on the resistivity logging curves to obtain transformed resistivity logging curves; and carrying out standard deviation normalization processing on the resistivity logging curves and the non-resistivity logging curves after transformation to obtain candidate normalized wells Duan Quxian corresponding to the resistivity logging curves and the non-resistivity logging curves after transformation respectively.

Illustratively, first, since the abscissa in the resistivity log is resistivity, and is exponential, the log transformation of the resistivity log is required, and the transformation formula is as follows:

equation one: x' =log ₁₀ x

Where x is the resistivity log to be converted and x' is the normalized well Duan Quxian after the resistivity log is converted.

Secondly, standard deviation normalization processing is carried out on the resistivity logging curves and the non-resistivity logging curves after transformation, the mean value of the data of each logging curve after processing is changed to 0, the variance is changed to 1, normal distribution is obeyed, and the standard deviation normalization formula is as follows:

formula II:

wherein y is a resistivity logging curve and a non-resistivity logging curve which need to be subjected to standard deviation normalization, y' is logging data subjected to standard deviation normalization, u is the average value of all logging data, and sigma is the standard deviation of all logging data.

In some alternative embodiments, the performing a combinatorial analysis on candidate normalized wells Duan Quxian to construct a combinatorial well log further comprises:

performing four arithmetic operations on candidate normalized wells Duan Quxian to construct a combined log; alternatively, the candidate normalized wells Duan Quxian are linearly combined to construct a combined log; alternatively, the candidate normalized wells Duan Quxian are polynomial combined to construct a combined log.

Illustratively, at least one of an addition, a subtraction, a multiplication, a division is performed on two or more candidate normalization wells Duan Quxian, e.g.: and calculating a+b, a-b, a multiplied by b or a multiplied by b when the candidate normalized interval curves are a and b.

Alternatively, two or more candidate normalized interval curves are weighted summed (i.e., linearly combined), for example: and calculating ka+gb if the candidate normalized well section curves are a and b.

Alternatively, two or more candidate normalization wells Duan Quxian are polynomial combined, for example: candidate normalized interval curves are a, b and c, and ka is calculated ² +gb+c。

And 205, carrying out porosity prediction analysis on the normalized well Duan Quxian through a target long-short term memory network to obtain a nuclear magnetic porosity prediction curve of a target well section.

Illustratively, after standard deviation normalization treatment is carried out on the single stable well Duan Quxian and the combined well logging curves, the standard deviation normalization treatment is input into a target long-short-term memory network, and porosity prediction analysis is carried out on each stable well section curve and each combined well logging curve, so as to obtain a nuclear magnetic porosity prediction curve of the target well section.

In summary, according to the neural network nuclear magnetic logging curve prediction method based on feature enhancement provided by the embodiment of the application, through performing well-seismic joint interpretation on a plurality of conventional logging curves, a target well Duan Quxian corresponding to a target well section in a target well is obtained, and through analyzing a part of well sections in the target well instead of a whole well section, the value range difference of data required to be analyzed at a time is reduced, and the accuracy of the nuclear magnetic porosity prediction curve is improved; the method comprises the steps of performing curve intersection analysis on a target well Duan Quxian, removing abnormal data in the target well Duan Quxian, and performing standard deviation normalization on the curve from which the abnormal data are removed to obtain a normalized well Duan Quxian; finally, the curve is predicted and analyzed through the target long-short-term memory network to obtain a nuclear magnetic porosity prediction curve of the target well section, and the short-term memory and the long-term memory are combined through a gate mechanism by the target long-short-term memory network, so that the change trend of conventional logging data along with depth and the front-back association of the data can be analyzed through the target long-short-term memory network, and the accuracy of the predicted nuclear magnetic porosity curve is improved.

According to the method provided by the embodiment of the application, the standard deviation normalization processing is carried out on the curve, so that the consistency of the characteristics is improved; and a combined well logging curve is constructed by a combined analysis method, and a single stabilizing well Duan Quxian and the combined well logging curve are synthesized to conduct predictive analysis, so that the characteristic dimension is expanded, and the accuracy of the nuclear magnetic porosity curve obtained by prediction is further improved.

In some optional embodiments, the target long-short term memory network is a network obtained by training a sample data set, and the method for predicting a nuclear magnetic logging curve of a neural network based on feature enhancement further includes a training process of the target long-short term memory network, and fig. 3 is a flowchart of a method for predicting a nuclear magnetic logging curve of a neural network based on feature enhancement according to an embodiment of the present application, and the method is described by using the method in a terminal or a server shown in fig. 1 as an example, and as shown in fig. 3, the method includes:

step 301, a first reference magnetic porosity curve of a sample well in a sample data set and a plurality of sample conventional log curves corresponding to the sample well are obtained.

Illustratively, a well in a sample dataset that simultaneously measures a plurality of conventional log curves and a nuclear magnetic porosity curve is selected as a sample well. The measured plurality of conventional well logging curves are a plurality of sample conventional well logging curves corresponding to the sample well, and the measured nuclear magnetic porosity curve is a first reference nuclear magnetic porosity curve of the sample well.

Step 302, initialize a sample long and short term memory network.

The sample long-term and short-term memory network comprises model parameters to be trained.

First, the cell structure of the sample long-short-term memory network will be described:

referring to fig. 4, schematically, a cell 400 of a sample long and short term memory network at time t is shown, the gate structure inside the cell comprising: a forget gate 401, an input gate 402, and an output gate 403; the forget gate 401 is used for discarding or retaining information in the unit state, the input gate 402 is used for updating information of the unit state, and the output gate 403 is used for calculating hidden layer state information to be output at the current moment.

As shown in fig. 4, the unit of the sample long-short-term memory network at time t receives two pieces of information from the unit of the sample long-short-term memory network at time t-1, namely: h is a _t-1 (hidden layer state information output by the network is memorized for the sample long-short period at the time of t-1), C _t-1 (the cell state information output at time t-1); the unit of the sample long short-term memory network at the time t also receives the characteristic vector x _t . The gate structure inside the unit of the sample long-short-term memory network passing through the time t is calculated to obtain the time h of the time t _t And C _t The calculation formula of the forgetting gate 401 is as follows:

And (3) a formula III: f (f) _t ＝σ(W _f [h _t-1 ，x _t ]+b _f )

Wherein σ is a sigmoid activation function; w (W) _f For forgetting the gate weight matrix, W _f The dimension of the hidden layer state and the dimension of the unit state output at the time t-1 and the dimension of the feature vector input at the time t-1 are determined; b _f Is biased; [ h ] _t-1 ，x _t ]H is output at time t-1 _t-1 X input with time t _t Splicing the feature vectors; ft is the output of the forgetting gate, and is a profile value with a value range of 0-1, and the unit state multiplied by ft can determine how much of the input hidden layer state information needs to be reserved.

Input gate 402 is a calculation of the cell state for updating the information of the cell stateThe method comprises two steps, wherein the first step is to calculate h _t-1 And x _t Which information needs to be emphasized or decremented, the activation function uses the tanh function, and the first step of the input gate 402 is calculated as follows:

equation four:

wherein W is _c A first weight matrix is used as an input gate; b _c Is biased;the state information inputted at time t.

Second, using sigmoid activation function to output probability value for decidingHow much information is to be updated into the cell state, the second step of the input gate 402 is calculated as follows:

formula five: i.e _t ＝σ(W _i [h _t-1 ，x _t ]+b _i )

Wherein σ is a sigmoid activation function; w (W) _i A second weight matrix is input to the gate; b _i Is biased; i.e _t Is the output of the input gate.

At this time, the long-term state information and the current state information can be combined by the operations of the forgetting gate 401 and the input gate 402 to calculate the updated cell state C at the time t _t The calculation formula is as follows:

formula six:

the output gate 403 determines which information the elements of the sample long and short-term memory network at time t have to output; output value O _t The unit state is related to the information to be output can be determined by the tanh activation function, and then the sigmoid function is used for determining the quantity of the output information, wherein the calculation formula is as follows:

formula seven: o (O) _t ＝σ(W _o [h _t-1 ，x _t ]+b _o )

Formula eight: h is a _t ＝O _t tanh(C _t )

Wherein h is _t And the hidden layer state information output by the network is memorized in a long-term and short-term mode and represents a sample at the time t.

Secondly, before training a sample long-short-term memory network, a developer needs to set super parameters in the long-short-term memory network, namely initial model parameters; the super parameters comprise the number of loop iterations of the sample long-short term memory network, the number of layers of the hidden layer and the like.

Optionally, if the sample long-short-term memory network is a trained network, the parameters in the sample long-short-term memory network need to be initialized when the sample long-short-term memory network needs to be retrained.

And step 303, analyzing the conventional log curves of a plurality of samples through a sample long-term and short-term memory network to obtain a predicted nuclear magnetic porosity curve.

Illustratively, the analyzing the plurality of sample conventional log curves through the sample long-short term memory network further comprises a pretreatment process of the plurality of sample conventional log curves:

(1) The sample well and well section are preselected.

The deposition environment of the sample well and the color, composition, structure and construction of the rock stratum in different depth intervals are different, so that the whole well section of the sample well often comprises different deposition phase bands and different lithology combination characteristics (such as clastic rock combination and volcanic rock combination), the range of values of conventional well logging curves of the different deposition phase bands and the different lithology combination characteristics is large, and the range of values of the nuclear magnetic porosity curves is not large. The mapping between the features of the sample data (sample conventional log) and the labels (first reference magnetic porosity curve) containing different depositional bands and different lithology combinations is therefore blurred, resulting in low accuracy of network model predictions.

Optionally, analyzing the deposition environment of the sample well and the well bore (such as the track data of the well bore) through well-seismic joint interpretation, and selecting the sample well according to the requirement of sample long-short term memory network training; and according to the geological stratification result of the sample well and the uniform geological stratification well section, selecting a sample conventional well logging curve corresponding to the proper well section to participate in training, namely, a plurality of sample target wells Duan Quxian respectively corresponding to the sample conventional well logging curves.

In the process of drilling a sample well, if the drilling encounters a mudstone stratum or a loose stratum, the well wall tends to collapse easily, and the well diameter is enlarged, so that the measured conventional well logging data (such as acoustic time difference, compensation density, compensation neutrons and the like) of the sample are abnormal, and the abnormal data with the enlarged well diameter is required to be removed from the sample target well Duan Quxian.

Optionally, abnormal data in the sample target well section curve is removed through curve intersection analysis. Illustratively, on the intersection plan of the compensated neutron curve and the compensated density curve, data points distributed at the upper left corner of the quartz model line represent abnormal data of the expanded diameter section, and after the data points are removed, a preliminary data sample for this training is formed, namely a sample stable well section curve corresponding to each of a plurality of sample target wells Duan Quxian.

(2) Data preprocessing and feature enhancement.

First, preprocessing data in a sample stabilization interval curve:

illustratively, lateral, induction, and other resistivity logs in the sample stabilization log are exponentially characterized, while other non-resistivity logs are linearly characterized. Alternatively, the log of the resistivity log is converted to linear data and the logarithmic transformation formula can refer to formula one.

It is noted that there may be a significant difference in magnitude between the multiple sample stability log curves, which may lead to a dominant magnitude attribute, and a slow rate of iterative convergence of the sample long and short term memory network. Therefore, by normalizing the plurality of sample stabilizing wells Duan Quxian, the optimization range of the long-short-term memory network of the sample can be reduced, the optimization process becomes gentle, and the optimal solution can be more easily and correctly converged.

Optionally, a standard deviation normalization formula is adopted to perform standard deviation normalization processing on the data in each sample stable well section curve, the mean value of each sample stable well section curve after processing becomes 0, the variance becomes 1, normal distribution is obeyed, and a calculation formula can refer to a formula II.

Second, the characteristics of the input network can be enhanced by combining analysis methods:

the feature quantity and quality of the training samples determine the learning effect of the long-term and short-term memory network of the samples.

Optionally, a combination analysis is performed on the sample stable well section curves, a new curve is constructed through four arithmetic operations, linear combination or polynomial combination, and the characteristic dimension of the sample is preferably expanded by combining geology knowledge and preferably reflecting the curve of the reservoir characteristics as far as possible, so that the consistency of the sample characteristics is improved, and the important characteristics of reservoir information indicating the sample well are learned by a long-short-term memory network of the sample.

Illustratively, a new curve formed by the sample stabilizing well Duan Quxian and the sample stabilizing well section curve subjected to standard deviation normalization is input into a sample long-term and short-term memory network, optionally, feature vectors corresponding to the curves are extracted, and feature analysis is performed on the feature vectors to obtain a predicted nuclear magnetic porosity curve. For example: if the number of the stable well section curves of the sample is 9 and the number of the new curves is 3, the number of the feature vectors obtained by initial extraction is 13.

Step 304, calculating a contrast loss value based on the predicted nuclear magnetic porosity curve and the first reference magnetic porosity curve.

The contrast loss value is used to indicate a differential outcome between the predicted nuclear magnetic porosity curve and the first reference magnetic porosity curve.

Optionally, the comparison loss value is calculated by a loss function in the sample long-short-term memory network, and illustratively, a mean square error is adopted as the loss function, and the loss function has the following formula:

formula nine:

wherein a is predicted nuclear magnetic porosity curve data, y is first reference magnetic porosity curve data, L _mse For comparison loss values. And calculating an error between the predicted value and the true value by using the loss function, so as to evaluate the learning degree of the sample long-term and short-term memory network on the characteristics.

And step 305, updating model parameters in the sample long-short-period memory network based on the comparison loss value to obtain the target long-short-period memory network.

Illustratively, updating a weight matrix and a bias in the sample long-short-term memory network based on the comparison loss value to obtain the target long-short-term memory network.

In some alternative embodiments, the sample dataset includes data corresponding to n sample wells; and iteratively updating model parameters in the sample long-short-term memory network based on the comparison loss values respectively corresponding to the n sample wells to obtain the target long-short-term memory network.

Illustratively, taking a sample curve corresponding to each of the n sample wells as a sequence, inputting the n sequences corresponding to the n sample wells into a sample long-short-term memory network for training, and iterating four processes of forward propagation, calculation loss, reverse propagation and parameter updating in the internal loop of the sample long-short-term memory network. Optionally, the number of loop iterations may be set when initializing the network, for example: setting the circulation times as 200 times; then carrying out internal loop iteration for 200 times in the sample long-short-period memory network, observing the change trend of the comparison loss value, taking the loop times corresponding to the minimum value of the comparison loss value as the optimal training times, and taking the parameters corresponding to the sample long-short-period memory network corresponding to the optimal loop times as target parameters; and obtaining the target long-short-term memory network based on the target parameter.

Optionally, in the loop iteration process, determining the model parameter when the contrast loss value is minimum based on a random gradient descent algorithm. Illustratively, a random gradient descent optimizer is disposed in the sample long-short-term memory network, and the parameter θ when the loss function L is minimum can be found by using the random gradient descent optimizer, where the formula of the objective function is as follows:

formula ten:

wherein θ ^* Is the value of the parameter θ under optimal conditions.

Searching for the value of the parameter under the optimal condition, namely finding the minimum point of the loss function L under the condition of meeting the objective function, schematically, randomly setting a parameter when initializing the sample long-short-term memory network Then calculate the gradient vector at this point +.>Setting a learning rate parameter eta; based on learning rate parameter eta and gradient vector->Updating the parameter theta to obtain a new theta ¹ The calculation formula is as follows:

formula eleven:

based on the number of cycles, the process is repeated, and the parameter θ at which the loss function L is minimum is found.

In summary, according to the neural network nuclear magnetic logging curve prediction method based on feature enhancement provided by the embodiment of the application, a sample long-short-term memory network is trained through a first reference magnetic porosity curve and a plurality of sample conventional logging curves corresponding to sample wells, so that a target long-short-term memory network is obtained; before training, performing well-shock joint interpretation, abnormal data rejection, logarithmic conversion, standard deviation normalization and other preprocessing operations on conventional well-logging curves of a plurality of samples, so that the quality of sample data input into a long-term and short-term memory network of the samples is improved; and the feature dimension of the sample data is expanded through combined analysis, and the consistency of the sample features is improved, so that the training efficiency of the long-short-period memory network of the sample is improved, and the prediction accuracy of the target long-short-period memory network obtained through training is higher.

In some optional embodiments, the method for predicting a nuclear magnetic logging curve of a neural network based on feature enhancement further includes a test process of a target long-short term memory network to check whether the target long-short term memory network meets an application standard, and fig. 5 is a flowchart of a method for predicting a nuclear magnetic logging curve of a neural network based on feature enhancement according to an embodiment of the present application, and the method is applied to a terminal or a server shown in fig. 1 for illustration, as shown in fig. 5, and includes:

and step 3051, updating model parameters in the sample long-period memory network based on the comparison loss value to obtain a candidate long-period memory network.

The model parameters obtained at this time are model parameters in the sample long-short-term memory network corresponding to the obtained minimum contrast loss value based on the number of cycles.

Step 3052, obtaining a second reference magnetic porosity curve and a plurality of candidate conventional log curves corresponding to candidate wells in the test dataset.

Illustratively, a well in the test dataset that simultaneously measures a plurality of conventional log curves and a nuclear magnetic porosity curve is selected as a candidate well. The measured plurality of conventional logging curves are a plurality of candidate conventional logging curves corresponding to the candidate well, and the measured nuclear magnetic porosity curve is a first reference nuclear magnetic porosity curve of the candidate well.

And step 3053, analyzing the conventional log curves of the plurality of samples through the candidate long-short-term memory network to obtain candidate nuclear magnetic porosity curves.

Alternatively, the process of obtaining the candidate nuclear magnetic porosity curve may refer to steps 202 to 205, which are not described herein.

Schematically, if the whole well section of the candidate well is divided into a first well section and a second well section according to geological stratification, respectively obtaining a candidate nuclear magnetic porosity curve of the first well section and a candidate nuclear magnetic porosity curve of the second well section, and finally combining the candidate nuclear magnetic porosity curve of the first well section and the candidate nuclear magnetic porosity curve of the second well section to obtain the candidate nuclear magnetic porosity curve of the whole well section of the candidate well.

Step 3054, calculating a correlation coefficient between the candidate nuclear magnetic porosity curve and the second reference magnetic porosity curve.

The correlation coefficient is used to indicate the similarity between the candidate nuclear magnetic porosity curve and the second reference magnetic porosity curve.

Illustratively, the candidate nuclear magnetic porosity curve is a full-interval nuclear magnetic porosity curve of the candidate well; and the second reference magnetic porosity curve is a full-well section nuclear magnetic porosity curve of the candidate well, whether data of the two curves under the same depth are in an error range is compared, and if the data are in the error range, the data points representing the candidate nuclear magnetic porosity curve under the depth meet the condition. And calculating the ratio of the data points meeting the conditions in the data points of the candidate nuclear magnetic porosity curve to all the data points of the candidate nuclear magnetic porosity curve, namely, the correlation coefficient.

In some alternative embodiments, the correlation coefficient is an average of the predictions for the plurality of candidate wells. That is, the correlation coefficients corresponding to the candidate wells are calculated, the correlation coefficients corresponding to the candidate wells are averaged, and the obtained average correlation coefficient is the coefficient which is finally compared with the expected value.

In step 3055, training the sample long-term and short-term memory network is continued in response to the correlation coefficient being less than the expected value.

Illustratively, the expected value is set to 80%, and if the ratio of the data point satisfying the condition among the data points of the candidate nuclear magnetic porosity curve to all the data points of the candidate nuclear magnetic porosity curve is less than 80%, which indicates that the prediction rate of the candidate long-short term memory network is not as expected, the sample long-short term memory network needs to be trained.

Optionally, model parameters in the candidate long-short-term memory network are suitable, the candidate long-short-term memory network with the parameters adjusted is used as an initialized sample long-short-term memory network to be retrained until the contrast loss value of the trained network is reduced to a proper range as much as possible, and the correlation coefficient reaches an expected value.

And step 3056, in response to the correlation coefficient reaching the expected value, taking the candidate long-term memory network as a target long-term memory network.

Illustratively, the expected value is set to 80%, and the model parameter at that time is saved in response to the ratio of the data point meeting the condition in the data points of the candidate nuclear magnetic porosity curve to all the data points of the candidate nuclear magnetic porosity curve being greater than or equal to 80%, and the candidate long-short-period memory network corresponding to the model parameter is used as the target long-short-period memory network.

In summary, according to the neural network nuclear magnetic logging curve prediction method based on feature enhancement provided by the embodiment of the application, before a target long-short-term memory network is obtained through training, the candidate long-short-term memory network obtained through training is tested based on the logging curve of the candidate well in the test data set, and only when the correlation coefficient between the candidate nuclear magnetic porosity curve obtained through prediction of the candidate long-short-term memory network and the second reference magnetic porosity curve reaches a desired value, training is stopped, so that the accuracy of the target long-short-term memory network obtained through final training is further improved.

Based on the above flow, the embodiment of the application provides a visual software interface diagram of man-machine interaction, which integrates key steps of data input, data preprocessing, model training, parameter optimization, data output and the like related to the embodiment of the application into a software device, and can efficiently and intelligently generate a nuclear magnetic logging curve. Referring to FIG. 6, a schematic diagram of a characteristic-enhanced nuclear magnetic log prediction interface is shown, and the following description describes the functions in the interface:

(1) A training file is selected.

Clicking on the select training file button 601 may select the training file stored in the terminal. Illustratively, log data for sample wells in a sample dataset is selected as a training file.

(2) Data preparation.

Clicking on the import and display data button 602, the data in the selected training file may be selected and displayed in the data display box 603, where each column represents a log.

Illustratively, a log transformation button 604 and a normalization button 605 are displayed below the data display frame 603, the log transformation button 604 is selected, and a column 3 is input in the following text input frame, so that log transformation is required for logging curves of the 3 rd column of the data display frame 603; selecting the normalization button 605 and inputting the labels 1, 2, and 3 in the text input boxes, it represents that standard deviation normalization operation is required to be performed on the log curves of the 1 st, 2 nd, and 3 rd columns of the data display box 603.

Illustratively, input column 3 in text box 606, which represents the feature of the log that needs to be extracted for column 3 of data display box 603; inputting a list 5 in a text box 607, wherein the logging curve representing the 3 rd column of the data display box 603 is a reference logging curve; inputting a number 1 in text box 608, which indicates that the number of parameters that are transmitted to the network for training in a single time is 1; in text box 609, the number 50 is entered, indicating that the length of the sequence that is transferred to the network for training (one sequence for each log for a sample well) is 50.

Illustratively, clicking the application button 610 indicates that all operations set up at the data preparation interface are performed on the log in the data display box 603; clicking the initialization button 611 indicates that the log in the data display box 603 is restored to the original state.

(3) And (5) setting a neural network.

Illustratively, page 612 may make settings for the neural network, the loss function, and the optimizer that need to be applied.

Illustratively, at the neural network setup interface, a type of neural network may be selected in selection block 613, for example: RNN, LSTM, etc.; inputting the number 3 in the text box 614, representing the number of features of the input neural network as 3; inputting the number 1 in the text box 615, wherein the number of features representing the output neural network is 1; inputting the number 5 in the text box 616, which represents the number of hidden layers being 5; a number 2 is entered in text box 617, representing a number of network layer layers of 2.

Illustratively, the type of the loss function may be set at the setting interface of the loss function, for example: setting a loss function of the neural network as a mean square error function; the objective function may be set at a setting interface of the optimizer.

(4) And (5) training a network.

Illustratively, the number 200 is entered in text box 618, representing a number of cycles of 200 for this training.

Illustratively, clicking the training and display loss function button 619 trains the network according to the parameters and number of cycles set on page 612; and a change curve of the loss function is displayed in a function display box 620.

(5) And (5) network testing.

Illustratively, clicking on the select test file button 621 may select a test file stored in the terminal. Illustratively, well log data for candidate wells in the test dataset is selected as the test file.

Illustratively, clicking on import data and application button 622 indicates that candidate nuclear magnetic porosity curves are to be predicted from the well log data of the candidate well over the trained network; clicking on the run and generate contrast curve button 623 generates a comparison map and analysis data of the candidate nuclear magnetic porosity curve and a reference nuclear magnetic porosity curve corresponding to the candidate nuclear magnetic porosity curve.

(6) Network applications.

Illustratively, clicking on the select predicted file button 624 may select a predicted file stored in the terminal. Illustratively, log data of a target well in the prediction dataset is selected as the prediction file.

Illustratively, clicking the import data and the application button 625 indicates that the log data of the target well will be predicted via the trained network to obtain a nuclear magnetic porosity prediction curve; clicking on the run button 626 generates a nuclear magnetic porosity prediction graph.

Illustratively, the software is applied to the logging data of 10 wells in the 12 wells with conventional logging curves and nuclear magnetic porosity curves as training samples, the logging data of 2 wells are left as test samples, as shown in fig. 7, the real nuclear magnetic porosity logging curves of the well 701 and the well 702 and the nuclear magnetic porosity logging prediction curves obtained by the method provided by the embodiment of the application are compared, and as shown in fig. 7, the real nuclear magnetic porosity logging curves and the nuclear magnetic porosity logging prediction curves of the well 701 and the well 702 are well matched, and the correlation is above 88%.

In summary, the characteristic-enhancement-based neural network nuclear magnetic logging curve prediction method provided by the embodiment of the application can predict and obtain a nuclear magnetic logging curve with higher precision, so that reliable data is provided for the fine evaluation and prediction of a subsequent reservoir. Illustratively, the distribution range of the dessert reservoir can be accurately predicted by carrying out seismic inversion based on the curve, and horizontal well drilling and storage increasing production are effectively promoted.

Referring to fig. 8, a block diagram of a characteristic-enhanced nuclear magnetic log prediction apparatus according to an exemplary embodiment of the present application is shown, where the apparatus includes the following modules:

A data acquisition module 810 for acquiring a plurality of conventional log curves, the plurality of conventional log curves being full interval log data measured for a target well for different measurement parameters;

the joint interpretation module 820 is configured to perform a well-seismic joint interpretation on a plurality of conventional well-logging curves, and obtain target wells Duan Quxian corresponding to the conventional well-logging curves, where the target well section curves are well-logging data corresponding to target well sections in the target wells;

the intersection analysis module 830 is configured to perform curve intersection analysis on target wells Duan Quxian corresponding to multiple conventional logging curves, and remove expanded-diameter section logging data in the target well section curve to obtain a stable well section curve, where the expanded-diameter section logging data is abnormal data of well diameter expansion in the target well section curve;

the normalization processing module 840 is configured to perform standard deviation normalization processing and combination analysis on the stable well section curve to obtain a normalized well Duan Quxian;

the prediction analysis module 850 is configured to perform a porosity prediction analysis on the normalized well Duan Quxian through a target long-short term memory network, so as to obtain a nuclear magnetic porosity prediction curve of the target well section.

As shown in fig. 9, in some alternative embodiments, the data acquisition module 810 is further configured to acquire a first reference magnetic porosity curve of a sample well in a sample data set and a plurality of sample conventional log curves corresponding to the sample well; the device further comprises:

The initialization module 860 is configured to initialize a sample long-short-term memory network, where the sample long-term memory network includes model parameters to be trained;

the prediction analysis module 850 is further configured to analyze a plurality of conventional log curves of samples through the long-short-term memory network of samples to obtain a predicted nuclear magnetic porosity curve;

a calculation module 870 for calculating a contrast loss value based on the predicted nuclear magnetic porosity curve and the first reference nuclear magnetic porosity curve, the contrast loss value being indicative of a differentiation result between the predicted nuclear magnetic porosity curve and the reference nuclear magnetic porosity curve;

and an updating module 880, configured to update the model parameters in the sample long-short-term memory network based on the comparison loss value, so as to obtain the target long-short-term memory network.

In some alternative embodiments, the sample dataset includes data corresponding to n sample wells; the updating module 880 is further configured to iteratively update the model parameters in the sample long-short-term memory network based on the comparison loss values corresponding to the n sample wells, so as to obtain the target long-short-term memory network.

In some optional embodiments, the updating module 880 is configured to update the model parameters in the sample long-term memory network based on the comparison loss value to obtain a candidate long-term memory network; the update module 880 further includes:

An obtaining unit 881, configured to obtain a second reference magnetic porosity curve and a plurality of candidate conventional log curves corresponding to candidate wells in the test dataset;

an analysis unit 882, configured to analyze the conventional log curves of the plurality of samples through the candidate long-short term memory network to obtain candidate nuclear magnetic porosity curves;

a calculating unit 883 for calculating a correlation coefficient between the candidate nuclear magnetic porosity curve and the second reference magnetic porosity curve, the correlation coefficient being used to indicate a degree of similarity between the candidate nuclear magnetic porosity curve and the second reference magnetic porosity curve;

a training unit 884, configured to continue training the sample long-short-term memory network in response to the correlation coefficient being less than an expected value;

and a determining unit 885, configured to use the candidate long-short term memory network as the target long-short term memory network in response to the correlation coefficient reaching the expected value.

In some alternative embodiments, the data acquisition module 810 is configured to acquire seismic data of the target well, where the seismic data is seismic data reflecting geological features of the target well; the joint interpretation module 820 is configured to perform a well-seismic joint interpretation on the seismic data and the plurality of conventional log curves to obtain a joint interpretation result, where the joint interpretation result is used to indicate a geological stratification condition in the target well; the joint interpretation module 820 further includes:

A selecting unit 821, configured to select a target depth interval based on the joint interpretation result, where the target depth interval is used to indicate a depth position of the target well segment on the whole well segment;

the selecting unit 821 is further configured to select logging data of a plurality of conventional logging curves in the target depth interval, so as to obtain target wells Duan Quxian corresponding to the conventional logging curves respectively.

In some alternative embodiments, the target interval curves include a compensated neutron curve and a compensated density curve; the intersection analysis module 830 is configured to construct a density neutron intersection plan based on the compensated neutron curve and the compensated density curve; the intersection analysis module 830 is configured to reject abnormal data points distributed in a designated area of the quartz model line in the density neutron intersection plan; the intersection analysis module 830 is configured to use the compensated neutron curve from which the abnormal data points are removed and the compensated density curve from which the abnormal data points are removed as the stable log curve.

In some optional embodiments, the normalization processing module 840 is configured to perform standard deviation normalization processing on the stable well section curve to obtain candidate normalized wells Duan Quxian, where the normalization processing module 840 includes:

A combination unit 841, configured to perform a combination analysis on the candidate normalized wells Duan Quxian, and construct a combined log;

the normalization processing module 840 is further configured to perform standard deviation normalization processing on the combined log curve to obtain a combined normalized well Duan Quxian;

the normalization processing module 840 is further configured to determine the candidate normalized interval curves and the combined normalized interval curve as the normalized well Duan Quxian.

In some alternative embodiments, the stable wellbore interval curves include resistivity-like well logs and non-resistivity-like well logs; the normalization processing module 840 includes:

a transformation unit 842, configured to perform logarithmic transformation on the resistivity log to obtain a transformed resistivity log;

the normalization processing module 840 is further configured to perform standard deviation normalization processing on the transformed resistivity log and the non-resistivity log, to obtain normalized wells Duan Quxian corresponding to the transformed resistivity log and the non-resistivity log, respectively.

In some alternative embodiments, a combination unit 841 is configured to perform four operations on the stability well Duan Quxian to construct the combined log; or, for linearly combining the stability wells Duan Quxian to construct the combined log; alternatively, a polynomial combination is used to construct the combined log for the stability well Duan Quxian.

In summary, according to the characteristic enhancement-based nuclear magnetic logging curve prediction device provided by the embodiment of the application, through performing well-shock joint interpretation on a plurality of conventional logging curves, a target well Duan Quxian corresponding to a target well section in a target well is obtained, and through analyzing a part of well sections in the target well instead of a whole well section, the value range difference of data needing to be analyzed at a time is reduced, and the accuracy of a nuclear magnetic porosity prediction curve is improved; the method comprises the steps of performing curve intersection analysis on a target well Duan Quxian, removing abnormal data in the target well Duan Quxian, and performing standard deviation normalization on the curve from which the abnormal data are removed to obtain a normalized well Duan Quxian; finally, the curve is predicted and analyzed through the target long-short-term memory network to obtain a nuclear magnetic porosity prediction curve of the target well section, and the short-term memory and the long-term memory are combined through a gate mechanism by the target long-short-term memory network, so that the change trend of conventional logging data along with depth and the front-back association of the data can be analyzed through the target long-short-term memory network, and the accuracy of the predicted nuclear magnetic porosity curve is improved.

It should be noted that: the characteristic enhancement-based nuclear magnetic log prediction device provided in the above embodiment is only exemplified by the division of the above functional modules, and in practical application, the above functional allocation may be performed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to perform all or part of the functions described above. In addition, the embodiment of the characteristic-enhanced nuclear magnetic logging curve prediction device and the characteristic-enhanced neural network nuclear magnetic logging curve prediction method provided in the foregoing embodiments belong to the same concept, and detailed implementation processes of the device and the method are detailed in the embodiments, and are not described herein.

Fig. 10 shows a schematic structural diagram of a server according to an exemplary embodiment of the present application. The server may be a server as shown in fig. 1. Specifically, the structure comprises the following structures:

the server 1000 includes a central processing unit (Central Processing Unit, CPU) 1001, a system Memory 1004 including a random access Memory (Random Access Memory, RAM) 1002 and a Read Only Memory (ROM) 1003, and a system bus 1005 connecting the system Memory 1004 and the central processing unit 1001. The server 1000 also includes a mass storage device 1006 for storing an operating system 1013, application programs 1014, and other program modules 1015.

The mass storage device 1006 is connected to the central processing unit 1001 through a mass storage controller (not shown) connected to the system bus 1005. The mass storage device 1006 and its associated computer-readable media provide non-volatile storage for the server 1000. That is, the mass storage device 1006 may include a computer readable medium (not shown) such as a hard disk or compact disc read only memory (Compact Disc Read Only Memory, CD-ROM) drive.

Computer readable media may include computer storage media and communication media without loss of generality. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, erasable programmable read-only memory (Erasable Programmable Read Only Memory, EPROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read Only Memory, EEPROM), flash memory or other solid state memory technology, CD-ROM, digital versatile disks (Digital Versatile Disc, DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Of course, those skilled in the art will recognize that computer storage media are not limited to the ones described above. The system memory 1004 and mass storage device 1006 described above may be referred to collectively as memory.

According to various embodiments of the present application, the server 1000 may also operate by a remote computer connected to the network through a network, such as the Internet. I.e., the server 1000 may be connected to the network 1012 through a network interface unit 1011 connected to the system bus 1005, or may be connected to other types of networks or remote computer systems (not shown) using the network interface unit 1011.

The memory also includes one or more programs, one or more programs stored in the memory and configured to be executed by the CPU.

Embodiments of the present application also provide a computer device that may be implemented as a terminal or server as shown in fig. 3. The computer device comprises a processor and a memory, wherein at least one instruction, at least one section of program, code set or instruction set is stored in the memory, and the at least one instruction, the at least one section of program, the code set or the instruction set is loaded and executed by the processor to realize the neural network nuclear magnetic logging curve prediction method based on feature enhancement provided by the method embodiments.

Embodiments of the present application further provide a computer readable storage medium having at least one instruction, at least one program, a code set, or an instruction set stored thereon, where the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the neural network nuclear magnetic log prediction method based on feature enhancement provided by the foregoing method embodiments.

Embodiments of the present application also provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the neural network nuclear magnetic logging curve prediction method based on feature enhancement provided by the above method embodiments.

Alternatively, the computer-readable storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), solid state disk (SSD, solid State Drives), or optical disk, etc. The random access memory may include resistive random access memory (ReRAM, resistance Random Access Memory) and dynamic random access memory (DRAM, dynamic Random Access Memory), among others. The foregoing embodiment numbers of the present application are merely for describing, and do not represent advantages or disadvantages of the embodiments.

It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

The foregoing description of the preferred embodiments is merely exemplary in nature and is in no way intended to limit the invention, since it is intended that all modifications, equivalents, improvements, etc. that fall within the spirit and scope of the invention.

Claims

1. A neural network nuclear magnetic log prediction method based on feature enhancement, the method comprising:

2. The method of claim 1, wherein the training process of the target long-short term memory network comprises:

acquiring a first reference magnetic porosity curve of a sample well in a sample data set and a plurality of sample conventional logging curves corresponding to the sample well;

initializing a sample long-term and short-term memory network, wherein the sample long-term and short-term memory network comprises model parameters to be trained;

analyzing a plurality of conventional log curves of the samples through the long-period memory network of the samples to obtain a predicted nuclear magnetic porosity curve;

calculating a contrast loss value based on the predicted nuclear magnetic porosity curve and the first reference nuclear magnetic porosity curve, the contrast loss value being used to indicate a differentiation result between the predicted nuclear magnetic porosity curve and the reference nuclear magnetic porosity curve;

and updating the model parameters in the sample long-short-term memory network based on the comparison loss value to obtain the target long-short-term memory network.

3. The method of claim 2, wherein the sample dataset comprises data corresponding to n sample wells;

The updating the model parameters in the sample long-short term memory network based on the contrast loss value comprises:

and circularly and iteratively updating the model parameters in the sample long-short-period memory network based on the comparison loss values respectively corresponding to the n sample wells to obtain the target long-short-period memory network.

4. The method of claim 2, wherein updating the model parameters in the sample long-short-term memory network based on the contrast loss value results in the target long-short-term memory network, comprising:

updating the model parameters in the sample long-term and short-term memory network based on the comparison loss value to obtain a candidate long-term and short-term memory network;

acquiring a second reference magnetic porosity curve and a plurality of candidate conventional logging curves corresponding to candidate wells in the test data set;

analyzing the conventional log curves of a plurality of samples through the candidate long-term and short-term memory network to obtain candidate nuclear magnetic porosity curves;

calculating a correlation coefficient between the candidate nuclear magnetic porosity curve and the second reference magnetic porosity curve, wherein the correlation coefficient is used for indicating the similarity between the candidate nuclear magnetic porosity curve and the second reference magnetic porosity curve;

Continuing training the sample long-term and short-term memory network in response to the correlation coefficient being less than an expected value;

and responding to the correlation coefficient reaching the expected value, and taking the candidate long-short-term memory network as the target long-short-term memory network.

5. The method of claim 1, wherein performing a well-log joint interpretation on the plurality of conventional well-log curves to obtain target interval curves respectively corresponding to the plurality of conventional well-log curves comprises:

acquiring seismic data of the target well, wherein the seismic data are seismic data reflecting geological features of the target well;

performing well-seismic joint interpretation on the seismic data and a plurality of conventional logging curves to obtain joint interpretation results, wherein the joint interpretation results are used for indicating geological stratification in the target well;

selecting a target depth interval based on the joint interpretation result, wherein the target depth interval is used for indicating the depth position of the target well section on the whole well section;

and selecting logging data of a plurality of conventional logging curves in the target depth interval to obtain target wells Duan Quxian corresponding to the conventional logging curves respectively.

6. The method of claim 1, wherein the target interval curve comprises a compensated neutron curve and a compensated density curve;

Performing curve intersection analysis on target wells Duan Quxian corresponding to the conventional logging curves respectively, removing the logging data of the expanded section in the target well section curve to obtain a stable well section curve, including:

constructing a density neutron intersection plan based on the compensation neutron curve and the compensation density curve;

removing abnormal data points distributed in a designated area of a quartz model line in the density neutron intersection plan view;

and taking the compensation neutron curve from which the abnormal data points are eliminated and the compensation density curve from which the abnormal data points are eliminated as the stable well section curves.

7. The method of claim 1, wherein the performing standard deviation normalization and combination analysis on the stable wellsite curves to obtain normalized wellsite curves comprises:

performing standard deviation normalization processing on the stable well section curve to obtain candidate normalized wells Duan Quxian;

performing combined analysis on the candidate normalized wells Duan Quxian to construct a combined log;

performing standard deviation normalization processing on the combined logging curve to obtain a combined normalized well Duan Quxian;

the candidate normalized interval curve and the combined normalized interval curve are determined as the normalized well Duan Quxian.

8. The method of claim 7, wherein the stable wellsite curves comprise resistivity-class well logs and non-resistivity-class well logs;

and performing standard deviation normalization processing on the stable well section curve to obtain a candidate normalized well section curve, wherein the method comprises the following steps:

performing logarithmic transformation on the resistivity logging-like curve to obtain a transformed resistivity logging-like curve;

and carrying out standard deviation normalization processing on the resistivity logging curves and the non-resistivity logging curves after transformation to obtain candidate normalized wells Duan Quxian corresponding to the resistivity logging curves and the non-resistivity logging curves after transformation respectively.

9. The method of claim 7, wherein the performing a combinatorial analysis of the candidate normalized well Duan Quxian to construct a combinatorial well log comprises:

performing four arithmetic operations on the candidate normalized wells Duan Quxian to construct the combined log; or,

linearly combining the candidate normalized wells Duan Quxian to construct the combined log; or,

polynomial combinations are performed on the candidate normalized wells Duan Quxian to construct the combined log.

10. A nuclear magnetic log prediction apparatus based on feature enhancement, the apparatus comprising: