Neural network based on feature enhancement for predicting nuclear magnetic logging curve and device

Abstract

The application discloses a feature enhancement-based neural network nuclear magnetic logging curve prediction method and device, and relates to the technical field of seismic exploration. The method comprises the following steps: obtaining target well section curves corresponding to a plurality of conventional logging curves respectively; performing curve cross analysis on the target well section curves corresponding to the plurality of conventional logging curves respectively, removing logging data of a diameter expansion section in the target well section curves, and obtaining stable well section curves; performing standard deviation normalization processing and combination analysis on the stable well section curves, and obtaining normalized well section curves; and performing porosity prediction analysis on the normalized well section curves through a target long short-term memory network, and obtaining a nuclear magnetic porosity prediction curve of a target well section. The target long short-term memory network can analyze the change trend of conventional logging data with depth and the correlation of data before and after the data, and the feature dimension is expanded through combination analysis, so that the accuracy of the predicted nuclear magnetic porosity curve is improved.

Description

Technical Field

[0001] This application relates to the field of seismic exploration technology, and in particular to a method and apparatus for predicting nuclear magnetic resonance logging curves based on feature-enhanced neural networks. Background Technology

[0002] Compared to conventional logging curves, nuclear magnetic resonance (NMR) logging curves more accurately and intuitively reflect reservoir characteristics such as porosity, permeability, and saturation, showing a good correlation with oil-bearing layers and a strong positive correlation with core porosity. NMR logging curves can serve as important logging curves for classifying and evaluating tight oil and shale oil sweet spots in oilfields. However, due to complex subsurface conditions and unpredictable and unavoidable problems during measurement, such as wellbore enlargement and instrument malfunctions, data distortion or missing data often occurs in certain well sections in practical applications. These missing logging data pose significant challenges to subsequent reservoir evaluation and prediction work, and re-logging is severely limited in actual production. Therefore, reconstructing distorted or missing NMR logging curves without increasing additional measurement costs is particularly important.

[0003] In related technologies, missing nuclear magnetic resonance logging curves can be directly generated through physical modeling based on geological and mechanical parameters.

[0004] However, physical models are often based on strong assumptions and represent a significant simplification of real formation information. Furthermore, different physical models need to be selected for different situations, a selection process that is inherently subjective and relies on expert experience and domain knowledge. Therefore, the accuracy of nuclear magnetic resonance logging curves generated based on physical models is relatively low. Summary of the Invention

[0005] This application provides a method and apparatus for predicting nuclear magnetic resonance logging curves based on feature-enhanced neural networks, which improves the accuracy of predicting nuclear magnetic resonance logging curves. The technical solution is as follows:

[0006] On the one hand, a neural network-based NMR logging curve prediction method based on feature enhancement is provided, the method comprising:

[0007] Acquire multiple conventional logging curves, which are full-section logging data of the target well measured for different measurement parameters;

[0008] Multiple conventional logging curves are interpreted in a combined well-seismic analysis to obtain target well section curves corresponding to the multiple conventional logging curves. The target well section curves are logging data corresponding to the target well section in the target well.

[0009] A curve intersection analysis is performed on the target well section curves corresponding to multiple conventional logging curves. The logging data of the enlarged diameter section in the target well section curve is removed to obtain the stable well section curve. The logging data of the enlarged diameter section is the abnormal data of well diameter enlargement in the target well section curve.

[0010] The standard deviation of the stable well section curve is normalized and combined analysis is performed to obtain the normalized well section curve;

[0011] The porosity prediction curve of the target well section is obtained by performing porosity prediction analysis on the normalized well section curve using a target long short-term memory network.

[0012] On the other hand, a feature-enhanced nuclear magnetic resonance logging curve prediction device is provided, the device comprising:

[0013] The data acquisition module is used to acquire multiple conventional logging curves, which are full-section logging data of the target well measured for different measurement parameters;

[0014] The joint interpretation module is used to perform joint well-seismic interpretation of multiple conventional well logging curves and obtain the target well section curves corresponding to the multiple conventional well logging curves respectively. The target well section curves are the well logging data corresponding to the target well section in the target well.

[0015] The cross-cutting analysis module is used to perform cross-cutting analysis on target well section curves corresponding to multiple conventional logging curves, remove the logging data of the enlarged diameter section in the target well section curve, and obtain the stable well section curve. The logging data of the enlarged diameter section is the abnormal data of well diameter enlargement in the target well section curve.

[0016] The normalization module is used to perform standard deviation normalization and combined analysis on the stable well section curve to obtain the normalized well section curve;

[0017] The predictive analysis module is used to perform porosity prediction analysis on the normalized well section curve through the target long short-term memory network to obtain the nuclear magnetic resonance porosity prediction curve of the target well section.

[0018] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to implement the feature-enhanced neural network nuclear magnetic logging curve prediction method described in any of the embodiments of this application.

[0019] On the other hand, a computer-readable storage medium is provided, wherein at least one piece of program code is stored in the computer-readable storage medium, the at least one piece of program code being loaded and executed by a processor to implement the feature-enhanced neural network nuclear magnetic logging curve prediction method according to any embodiment of this application.

[0020] On the other hand, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the feature-enhanced neural network-based NMR logging curve prediction methods described in the embodiments of this application.

[0021] The beneficial effects of the technical solutions provided in this application include at least the following:

[0022] By performing joint well-seismic interpretation on multiple conventional logging curves, the target well section curve corresponding to the target well section in the target well was obtained. By analyzing a portion of the target well section instead of the entire well section, the value range difference of the data to be analyzed in a single instance was reduced, improving the accuracy of the NMR porosity prediction curve. Through curve intersection analysis of the target well section curves, outliers were removed, and the curves with outliers were normalized by standard deviation to obtain normalized well section curves. Finally, the curves were predicted and analyzed using the target long short-term memory network to obtain the NMR porosity prediction curve for the target well section. Since the target long short-term memory network combines short-term and long-term memory through a gating mechanism, it can analyze the variation trend of conventional logging data with depth and the correlation between data, thereby improving the accuracy of the predicted NMR porosity curve. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of an implementation environment provided by an exemplary embodiment of this application;

[0025] Figure 2 This is a flowchart of a feature-enhanced neural network-based NMR logging curve prediction method provided in an exemplary embodiment of this application;

[0026] Figure 3This is a flowchart of a feature-enhanced neural network-based NMR logging curve prediction method provided in another exemplary embodiment of this application;

[0027] Figure 4 This is a cell structure diagram of an LSTM network provided in an exemplary embodiment of this application;

[0028] Figure 5 This is a flowchart of a feature-enhanced neural network-based NMR logging curve prediction method provided in another exemplary embodiment of this application;

[0029] Figure 6 This is a schematic diagram of a feature-enhanced nuclear magnetic resonance logging curve prediction interface provided in an exemplary embodiment of this application;

[0030] Figure 7 This is a comparison chart of real and predicted NMR logging curves provided in an exemplary embodiment of this application;

[0031] Figure 8 This is a structural block diagram of a feature-enhanced nuclear magnetic resonance logging curve prediction device provided in an exemplary embodiment of this application;

[0032] Figure 9 This is a structural block diagram of a feature-enhanced nuclear magnetic resonance logging curve prediction device provided in another exemplary embodiment of this application;

[0033] Figure 10 This is a structural block diagram of a server provided in an exemplary embodiment of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] In this application, the terms "first" and "second" are used to distinguish between identical or similar items that have essentially the same function. It should be understood that there is no logical or temporal dependency between "first" and "second", nor is there any limitation on the quantity or execution order.

[0036] Recurrent Neural Networks (RNNs) are a type of recurrent neural network that takes sequential data as input, recursively processes data in the direction of sequence evolution, and connects all nodes (recurrent units) in a chain-like manner. The Long Short-Term Memory (LSTM) network provided in this embodiment is a special type of RNN that adds unit states and gate structures to the hidden layers of the original RNN. The unit states are used to store and carry information along the time sequence; even information from earlier time steps can be passed to later time steps. The gate structures can add or delete information from the unit states. The LSTM network combines short-term and long-term memory through the gate mechanism, solving the problem that only short-term historical inputs can be effectively stored and thus not effectively addressing long-term memory issues. It also mitigates the vanishing and exploding gradient problems to some extent.

[0037] In related technologies, predicting missing logging curves from existing partially complete logging curves mainly relies on geological and mechanical parameters, attempting to directly invert and generate missing logging curves using physical models. However, these physical models are often based on strong assumptions, greatly simplifying the actual formation information. Furthermore, different physical models need to be selected for different situations, a selection process inherently subjective and reliant on expert experience and domain knowledge. Therefore, the quality of logging curves generated based on physical models is difficult to guarantee effectively. This application provides a feature-enhanced neural network-based NMR logging curve prediction method, primarily targeting wells with untested or partially missing NMR logging data, using a long short-term memory network to predict and generate usable NMR logging curves. First, multiple conventional logging curves of the target well are acquired, and a combined well-seismic interpretation is performed on the conventional logging curves. The logging curve corresponding to the target well section of the target well is selected as the target well section curve. Second, abnormal data with enlarged well diameter in the target well section curve are removed to obtain the stable well section curve. Then, the standard deviation of the stable well section curve is normalized to make the data distribution range consistent. Finally, the standard deviation normalized stable logging curve is input into the LSTM network to predict the nuclear magnetic porosity prediction curve of the target well section.

[0038] Figure 1 This is a schematic diagram of an implementation environment provided by an exemplary embodiment of this application, such as... Figure 1 As shown, the implementation environment includes a terminal 110, a server 120, and a communication network 130. The terminal 110 and the server 120 are connected through the communication network 130. In some optional embodiments, the communication network 130 can be a wired network or a wireless network. This embodiment does not limit this.

[0039] In some optional embodiments, terminal 110 may be a smartphone, tablet, laptop, desktop computer, smart home appliance, smart vehicle terminal, smart speaker, digital camera, etc., but is not limited thereto. Optionally, terminal 110 may have a target application installed. Schematic, the target application may be a traditional application, a cloud application, a mini-program or application module within a host application, or a web platform; this embodiment does not limit this. Optionally, the target application may provide a nuclear magnetic resonance porosity curve prediction function. Schematic, when it is necessary to predict the nuclear magnetic resonance porosity curve of a target well, such as... Figure 1 As shown, terminal 110 uploads multiple conventional logging curves of the target well to server 120. Server 120 analyzes the multiple conventional logging curves, predicts the nuclear magnetic porosity prediction curve of the target well, and feeds back the nuclear magnetic porosity prediction curve to terminal 110.

[0040] In some optional embodiments, server 120 is used to provide background services for the target application installed in terminal 110. Optionally, server 120 is equipped with a target long short-term memory network. Illustratively, after receiving multiple conventional logging curves from the target well, server 120 first performs a joint well-seismic interpretation of the multiple conventional logging curves, selecting the logging curve corresponding to the target well section as the target well section curve; secondly, it removes abnormal data indicating well diameter enlargement from the target well section curve to obtain a stable well section curve; then, it performs standard deviation normalization on the stable well section curve to obtain a normalized well section curve; finally, it inputs the normalized well section curve into the target long short-term memory network, outputs the nuclear magnetic resonance porosity prediction curve of the target well section, and sends the nuclear magnetic resonance porosity prediction curve to terminal 110, which can then display the nuclear magnetic resonance porosity prediction curve.

[0041] In some alternative embodiments, the target long short-term memory network can also be deployed on the terminal 110 side, and the terminal 110 can implement the nuclear magnetic porosity curve prediction function locally without the need for the server 120. This application embodiment does not limit this.

[0042] It is worth noting that server 120 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0043] It should be noted that all information (including but not limited to user equipment information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the conventional well logging curves involved in this application were obtained with full authorization.

[0044] Based on the above introduction and implementation environment, the neural network-based NMR logging curve prediction method provided in this application embodiment will be described. Figure 2 This is a flowchart illustrating a feature-enhanced neural network-based NMR logging curve prediction method provided in this application embodiment, demonstrating its application in situations such as... Figure 1 The following explanation uses the terminal or server shown as an example. Figure 2 As shown, the method includes:

[0045] Step 201: Obtain multiple conventional logging curves.

[0046] Multiple conventional logging curves are logging data for the entire well section measured with different measurement parameters for the target well.

[0047] Full-section logging data refers to logging data across the entire depth range of the target well. Well logging is a method of measuring physical parameters of underground rock formations, such as electrical conductivity, acoustic properties, radioactivity, and electrochemical properties, along the drilling profile using specialized downhole instruments. Different logging methods and corresponding logging curves exist for different physical parameters.

[0048] Conventional logging curves include resistivity logging curves and non-resistivity logging curves. Resistivity logging curves include micro-resistivity logging curves, lateral logging curves, and induction logging curves; non-resistivity logging curves include natural gamma logging curves, spontaneous potential logging curves, caliper logging curves, sonic transit time logging curves, compensated density logging curves, and compensated neutron logging curves. Optionally, the horizontal axis of a conventional logging curve represents various physical parameters (e.g., caliper size, natural gamma value, resistivity, etc.), and the vertical axis represents the measurement depth of the target well.

[0049] To illustrate, nine conventional logging curves of the target well (i.e., the nine conventional logging curves mentioned in the above description) are obtained.

[0050] Step 202: Perform a combined well-seismic interpretation on multiple conventional logging curves to obtain the target well section curves corresponding to each conventional logging curve.

[0051] The target well section curve is the logging data corresponding to the target well section in the target well.

[0052] Optionally, through joint well-seismic interpretation, the target well is geologically stratified, and according to a unified geological stratification well section, the logging data located in that well section from multiple conventional logging curves are selected as the target well section curve.

[0053] Optionally, the above method for obtaining target well section curves corresponding to multiple conventional logging curves further includes the following steps:

[0054] 1. Obtain seismic data for the target well.

[0055] Seismic data refers to seismic data that reflects the geological characteristics of the target well.

[0056] Indicatively, the seismic data consists of seismic waves acquired through a small, harmless earthquake artificially induced in the target well. The vertical axis of the seismic data represents the reflection time of the seismic waves.

[0057] 2. A combined well-seismic interpretation was performed on the seismic data and multiple conventional well logging curves to obtain the combined interpretation results.

[0058] The joint interpretation results are used to indicate the geological stratification in the target well.

[0059] Indicatively, based on seismic data and conventional logging curves from the target well, combined with geological and geophysical knowledge, stratigraphic interpretation is performed on the target well. The resulting stratigraphic interpretation is the joint interpretation result. Optionally, the joint interpretation result includes geological stratification markers and the corresponding depth intervals.

[0060] 3. Based on the joint interpretation results, select the target depth interval, which is used to indicate the depth position of the target well section in the whole well section.

[0061] By selecting logging data from multiple conventional logging curves within the target depth range, the target well section curves corresponding to the multiple conventional logging curves are obtained.

[0062] Indicatively, through joint well-seismic interpretation, the target well is divided into three segments: A, B, and C. If segment A is selected, the logging data of the corresponding depth range (i.e., the corresponding top and bottom depths) of segment A is obtained as the target well segment curve.

[0063] Step 203: Perform curve intersection analysis on the target well section curves corresponding to multiple conventional logging curves, remove the logging data of the enlarged diameter section in the target well section curve, and obtain the stable well section curve.

[0064] Among them, the logging data of the enlarged section is the abnormal data of the enlarged well diameter in the target well section curve.

[0065] In some optional embodiments, the target well section curve includes a compensated neutron curve and a compensated density curve, and the analysis is illustrated by curve intersection analysis of the compensated neutron curve and the compensated density curve:

[0066] Based on the compensated neutron curve and the compensated density curve, a density neutron intersection planar diagram is constructed; abnormal data points distributed in a specified area of ​​the quartz model line in the density neutron intersection planar diagram are removed; the compensated neutron curve and the compensated density curve after removing abnormal data points are used as the stable well section curves.

[0067] Schematic diagram: A density-neutron cross-plot is constructed with the compensated density data of the target well section on the horizontal axis and the compensated neutron data of the target well section on the vertical axis. A quartz model line is marked on this plot. The compensated neutron data indicates the hydrogen content of the formation in the target well section, and the compensated density data indicates the density of the formation. The quartz model line is obtained by measuring the compensated density and compensated neutrons of a single quartz ore sample; it is typically a straight line forming an angle of approximately 45 degrees with the horizontal axis of the density-neutron cross-plot. The data points located in the upper left corner of the quartz model line represent the enlarged diameter section data, i.e., the anomalous data indicating well diameter enlargement. After removing these data points, the remaining data points form the compensated neutron curve and compensated density curve, which represent the stable well section curves.

[0068] Step 204: Perform standard deviation normalization and combination analysis on the stable well section curve to obtain the normalized well section curve.

[0069] In some optional embodiments, the standard deviation of the stable well section curve is normalized to obtain candidate normalized well section curves; the candidate normalized well section curves are combined and analyzed to construct a combined logging curve; the standard deviation of the combined logging curves is normalized to obtain a combined normalized well section curve; the candidate normalized well section curves and the combined normalized well section curves are determined as the normalized well section curves.

[0070] Indicatively, after normalizing the standard deviation of a single stable well section curve, these curves can be combined for analysis to construct a new curve, namely a combined logging curve. Then, the combined logging curve is normalized again. Finally, the normalized results of the single stable well section curve and the combined logging curve are input into the target long short-term memory network for analysis.

[0071] Optionally, after combining two or more stable well section curves to obtain multiple combined logging curves, the curve that best reflects the reservoir characteristics of the target well can be selected from the multiple combined logging curves based on geological understanding as the final combined logging curve to be analyzed.

[0072] In some optional embodiments, the aforementioned stable well section curves include resistivity logging curves and non-resistivity logging curves; wherein, resistivity logging curves refer to micro-resistivity logging curves, lateral logging curves, and induction logging curves, and non-resistivity logging curves refer to natural gamma logging curves, spontaneous potential logging curves, caliper logging curves, sonic transit time logging curves, compensated density logging curves, and compensated neutron logging curves. Then, the above-mentioned standard deviation normalization processing of the stable well section curves yields candidate normalized well section curves, including:

[0073] Logarithmic transformation is performed on the resistivity logging curves to obtain the transformed resistivity logging curves. Standard deviation normalization is then performed on the transformed resistivity logging curves and the non-resistivity logging curves to obtain the candidate normalized well section curves corresponding to the transformed resistivity logging curves and the non-resistivity logging curves, respectively.

[0074] To illustrate, firstly, since the horizontal axis of resistivity logging curves is all resistivity, which is an exponential representation, a logarithmic transformation of the resistivity logging curves is required. The transformation formula is as follows:

[0075] Formula 1:

[0076] in, For resistivity-based logging curves that need to be converted, This is the normalized well section curve after conversion of resistivity-based logging curves.

[0077] Secondly, the standard deviation of the transformed resistivity logging curves and non-resistivity logging curves is normalized. After processing, the mean of each logging curve becomes 0, the variance becomes 1, and they follow a normal distribution. The standard deviation normalization formula is as follows:

[0078] Formula 2:

[0079] in, For resistivity logging curves and non-resistivity logging curves that require standard deviation normalization, To perform standard deviation normalization on the logging data, The mean of all well logging data. The standard deviation of all well logging data.

[0080] In some optional embodiments, the above-described combined analysis of candidate normalized well interval curves to construct combined logging curves further includes:

[0081] Perform arithmetic operations on the candidate normalized well section curves to construct a combined logging curve; or perform a linear combination of the candidate normalized well section curves to construct a combined logging curve; or perform a polynomial combination of the candidate normalized well section curves to construct a combined logging curve.

[0082] Indicatively, perform at least one of the following operations on two or more candidate normalized well section curves: addition, subtraction, multiplication, or division. For example, if the candidate normalized well section curves are a and b, then calculate a+b, ab, a×b, or a÷b.

[0083] Alternatively, a weighted summation (i.e., linear combination) can be performed on two or more candidate normalized well segment curves. For example, if the candidate normalized well segment curves are a and b, then ka+gb can be calculated.

[0084] Alternatively, a polynomial combination can be performed on two or more candidate normalized well section curves. For example, if the candidate normalized well section curves are a, b, and c, calculate ka. 2 +gb+c.

[0085] Step 205: Perform porosity prediction analysis on the normalized well section curve using the target long short-term memory network to obtain the nuclear magnetic porosity prediction curve of the target well section.

[0086] In a schematic manner, the standard deviation of the above-mentioned single stable well section curve and combined logging curve is normalized and then input into the target long short-term memory network to perform porosity prediction analysis on each stable well section curve and each combined logging curve, thereby obtaining the nuclear magnetic porosity prediction curve of the target well section.

[0087] In summary, the feature-enhanced neural network-based NMR logging curve prediction method provided in this application obtains the target well section curve corresponding to the target well section in the target well by performing joint well-seismic interpretation of multiple conventional logging curves. By analyzing a portion of the target well section instead of the entire well section, the value range difference of the data to be analyzed in a single instance is reduced, improving the accuracy of the NMR porosity prediction curve. By performing curve intersection analysis on the target well section curve, outlier data is removed, and the curves with outlier data are normalized according to their standard deviation to obtain normalized well section curves. Finally, the curves are predicted and analyzed using a target long short-term memory network to obtain the predicted NMR porosity curve of the target well section. Since the target long short-term memory network combines short-term memory and long-term memory through a gating mechanism, it can analyze the changing trend of conventional logging data with depth and the correlation between data, thereby improving the accuracy of the predicted NMR porosity curve.

[0088] The method provided in this application improves the consistency of features by normalizing the standard deviation of the curves; and by constructing a combined logging curve through a combined analysis method, it expands the feature dimensions by combining the single stable well section curve and the combined logging curve for predictive analysis, thereby further improving the accuracy of the predicted nuclear magnetic porosity curve.

[0089] In some optional embodiments, the target long short-term memory network is a network trained using a sample dataset. Therefore, the feature-enhanced neural network NMR logging curve prediction method further includes a training process for the target long short-term memory network. Figure 3 This is a flowchart illustrating a feature-enhanced neural network-based NMR logging curve prediction method provided in this application embodiment, demonstrating its application in situations such as... Figure 1 The following explanation uses the terminal or server shown as an example. Figure 3 As shown, the method includes:

[0090] Step 301: Obtain the first reference NMR porosity curve of the sample well in the sample dataset and the corresponding conventional logging curves of multiple samples.

[0091] To illustrate, wells in the sample dataset that have simultaneously measured multiple conventional logging curves and NMR porosity curves are selected as sample wells. The measured conventional logging curves are the sample conventional logging curves corresponding to the sample wells, and the measured NMR porosity curve is the first reference NMR porosity curve for the sample wells.

[0092] Step 302: Initialize the sample long short-term memory network.

[0093] The sample long short-term memory network includes the model parameters to be trained.

[0094] First, the unit structure of the sample long short-term memory network will be explained:

[0095] This is illustrative; please refer to it. Figure 4 It shows a unit 400 of a sample Long Short-Term Memory network at time t. The gate structure inside the unit includes a forget gate 401, an input gate 402, and an output gate 403. The forget gate 401 is used to discard or retain information in the unit state, the input gate 402 is used to update the information in the unit state, and the output gate 403 is used to calculate the hidden layer state information to be output at the current time.

[0096] like Figure 4 As shown, 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: (The hidden layer state information output by the Long Short-Term Memory network for the sample at time t-1) (This refers to the unit state information output at time t-1); the units of the Long Short-Term Memory network at time t also receive feature vectors. The gate structure within the cells of the Long Short-Term Memory network at time t is calculated to obtain the value at time t. and The calculation formula for Forget Gate 401 is as follows:

[0097] Formula 3: ( )

[0098] in, It is the sigmoid activation function; Here is the forget gate weight matrix. The dimension depends on the dimension of the hidden layer state output at time t-1, the dimension of the unit state, and the dimension of the input feature vector at time t-1; For bias; It is the output at time t-1. Input at time t Concatenation of eigenvectors; The output of the forget gate is a general value ranging from 0 to 1, multiplied by the cell state. This allows you to decide how much of the hidden layer state information needs to be retained.

[0099] Input gate 402 is used to update the cell state information. Its calculation is divided into two steps. The first step is to calculate... and To determine which information needs to be strengthened or weakened, the hyperbolic tangent function tanh is used as the activation function. The first step calculation formula for the input gate 402 is as follows:

[0100] Formula 4: tanh( )

[0101] in, This is the first weight matrix of the input gate; For bias; The input status information at time t.

[0102] The second step is to use the sigmoid activation function to output probability values, which are then used to determine... The second step of the calculation formula for input gate 402 is as follows: How much information needs to be updated in the cell state?

[0103] Formula 5:

[0104] in, It is the sigmoid activation function; This is the second weight matrix of the input gate; For bias; The output of the input gate.

[0105] At this point, the long-term state information and the current state information can be combined through the operations of forget gate 401 and input gate 402 to calculate the updated cell state at time t. The calculation formula is as follows:

[0106] Formula Six:

[0107] Output gate 403 determines what information a unit in the Long Short-Term Memory network needs to output at time t; the output value... The information to be output is related to the cell state. The tanh activation function determines the information to be output, and the sigmoid function determines the amount of information to be output. The calculation formula is as follows:

[0108] Formula 7:

[0109] Formula 8:

[0110] in, The hidden layer state information represents the output of the Long Short-Term Memory network at time t.

[0111] Secondly, before training the sample long short-term memory network, developers need to set the hyperparameters in this long short-term memory network, that is, the initial model parameters; among them, the hyperparameters include the number of iterations of the sample long short-term memory network, the number of hidden layers, etc.

[0112] 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 it needs to be retrained.

[0113] Step 303: Analyze the conventional logging curves of multiple samples using a sample long short-term memory network to obtain the predicted nuclear magnetic porosity curve.

[0114] This illustration demonstrates the analysis of multiple sample conventional logging curves using a sample long short-term memory network, and also includes the preprocessing of the multiple sample conventional logging curves:

[0115] (1) Pre-selected sample wells and well sections.

[0116] The sedimentary environment, as well as the color, composition, structure, and texture of the rock strata, vary across different depth ranges in the sample well. Consequently, the entire well section often contains different sedimentary facies zones and lithological assemblages (such as clastic rock assemblages and volcanic rock assemblages). The value ranges of conventional logging curves for different sedimentary facies zones and lithological assemblages differ significantly, while the value ranges of nuclear magnetic resonance (NMR) porosity curves do not differ much. Therefore, the mapping relationship between the characteristics (conventional logging curves of the sample) and the labels (first reference NMR porosity curve) of sample data containing different sedimentary facies zones and lithological assemblages becomes blurred, leading to low prediction accuracy of the network model.

[0117] Optionally, through well-seismic joint interpretation, the sedimentary environment and wellbore (e.g., wellbore trajectory data) of the sample wells are analyzed. Sample wells are selected according to the training requirements of the sample long short-term memory network. Based on the geological stratification results of the sample wells, the conventional logging curves corresponding to the appropriate well sections are selected for training according to the unified geological stratification well sections. That is, the sample target well section curves are respectively corresponding to the conventional logging curves of multiple samples.

[0118] During the drilling process of a sample well, if mudstone or loose formations are encountered, wellbore collapse and well diameter enlargement are likely to occur. This results in abnormalities in the measured conventional logging data (e.g., sonic transit time, compensated density, compensated neutrons, etc.), and the abnormal data with enlarged well diameter need to be removed from the target well section curve of the sample.

[0119] Optionally, abnormal data in the target well section curves can be removed through curve intersection analysis. Illustratively, on the intersection plane of the compensated neutron curve and the compensated density curve, the data points located in the upper left corner of the quartz model line represent abnormal data in the diameter expansion section. Removing these data points constitutes the initial data sample for this training, which consists of the sample stable well section curves corresponding to the multiple target well section curves.

[0120] (2) Data preprocessing and feature enhancement.

[0121] First, the data in the stable well section curves of the sample are preprocessed:

[0122] Indicatively, in the stable well section of the sample, lateral logging, induction logging, and other resistivity logging curves are characterized exponentially, while other non-resistivity logging curves are characterized linearly. Optionally, the resistivity logging curves can be logarithmically transformed into linear data; the formula for the logarithmic transformation can be found in Formula 1.

[0123] It is worth noting that there can be significant differences in magnitude between the stable well section curves of multiple samples. This difference in magnitude can lead to the dominance of attributes with larger magnitudes, slowing down the iterative convergence speed of the sample long short-term memory network. Therefore, normalizing the stable well section curves of multiple samples can reduce the optimization range of the sample long short-term memory network, making the optimization process smoother and more likely to converge to the optimal solution correctly.

[0124] Optionally, the standard deviation normalization formula is used to normalize the standard deviation of the data in the stable well section curves of each sample. After processing, the mean of the stable well section curves of each sample becomes 0 and the variance becomes 1, which follows a normal distribution. The calculation formula can be found in Formula 2.

[0125] Secondly, the features of the input network can be enhanced through combinatorial analysis:

[0126] The quantity and quality of features in the training samples determine the learning performance of the long short-term memory network.

[0127] Optionally, a combination analysis can be conducted on the curves of the stable well sections of the sample. New curves can be constructed through arithmetic operations, linear combinations, or polynomial combinations. Combined with geological understanding, curves that best reflect reservoir characteristics can be selected to expand the feature dimensions of the sample, thereby improving the consistency of sample features. This is beneficial for the sample's long short-term memory network to learn important features indicating reservoir information of the sample wells.

[0128] Indicatively, the sample stable well section curves, after standard deviation normalization, and the new curve formed by the sample stable well section curves are input into the sample long short-term memory network. Optionally, the feature vectors corresponding to each curve are extracted, and feature analysis is performed on multiple feature vectors to obtain the predicted NMR porosity curve. For example, if there are 9 sample stable well section curves and 3 new curves are formed, then the initially extracted feature vectors are 13.

[0129] Step 304: Calculate the contrast loss value based on the predicted NMR porosity curve and the first reference NMR porosity curve.

[0130] The contrast loss value is used to indicate the difference between the predicted NMR porosity curve and the first reference NMR porosity curve.

[0131] Optionally, the above contrastive loss value is calculated using the loss function in the sample's long short-term memory network. Illustratively, the mean squared error is used as the loss function, and the formula for the loss function is as follows:

[0132] Formula Nine:

[0133] Where 'a' represents the predicted NMR porosity curve data, and 'y' represents the first reference NMR porosity curve data. To compare the loss values, the error between the predicted and actual values ​​is calculated using this loss function, thereby evaluating the extent to which the sample's Long Short-Term Memory network has learned the features.

[0134] Step 305: Update the model parameters in the sample long short-term memory network based on the contrastive loss value to obtain the target long short-term memory network.

[0135] Indicatively, the target long short-term memory network is obtained by updating the weight matrix and bias in the sample long short-term memory network based on the contrastive loss value.

[0136] In some optional embodiments, the sample dataset includes data corresponding to n sample wells; then, based on the contrastive loss values ​​corresponding to the n sample wells, the model parameters in the sample long short-term memory network are iteratively updated to obtain the target long short-term memory network.

[0137] Indicatively, the sample curves corresponding to each of the aforementioned n sample wells are treated as a sequence. These n sequences are then input into a sample long short-term memory network for training. The training process within the sample long short-term memory network involves four iterative steps: forward propagation, loss calculation, backpropagation, and parameter update. Optionally, the number of iterations can be set during network initialization. For example, setting the number of iterations to 200 would result in 200 iterations within the sample long short-term memory network. The trend of the comparative loss value is observed, and the iteration count corresponding to the minimum comparative loss value is taken as the optimal training count. The parameters of the sample long short-term memory network corresponding to the optimal number of iterations are then taken as the target parameters. Based on these target parameters, the target long short-term memory network is obtained.

[0138] Optionally, during the above iterative process, the model parameters that minimize the comparative loss value are determined based on the stochastic gradient descent algorithm. Illustratively, if a stochastic gradient descent optimizer is set in the sample long short-term memory network, the parameters θ that minimize the loss function L can be found using the stochastic gradient descent optimizer, where the formula for the objective function is as follows:

[0139] Formula 10:

[0140] in, The parameters under optimal conditions The value of .

[0141] Finding the optimal parameter values, i.e., finding the point where the loss function L is minimized while satisfying the objective function, is illustrated by randomly setting a parameter during the initialization of the sample Long Short-Term Memory network. Then calculate the gradient vector at that point. And set a learning rate parameter η; based on the learning rate parameter η and the gradient vector Update parameter θ to obtain new The calculation formula is as follows:

[0142] Formula 11:

[0143] Based on the above number of iterations, repeat the process to find the parameter θ that minimizes the loss function L.

[0144] In summary, the feature-enhanced neural network NMR logging curve prediction method provided in this application trains the sample long short-term memory network (LSTM) using a first reference NMR porosity curve and multiple sample conventional logging curves corresponding to the sample wells to obtain the target LTM network. Before training, multiple preprocessing operations are performed on the multiple sample conventional logging curves, including joint well-seismic interpretation, outlier removal, logarithmic transformation, and standard deviation normalization, improving the quality of the sample data input to the sample LTM network. Furthermore, through combinatorial analysis, the feature dimensions of the sample data are expanded, improving the consistency of sample features, thereby increasing the training efficiency of the sample LTM network and resulting in higher prediction accuracy for the trained target LTM network.

[0145] In some optional embodiments, the above-described feature-enhanced neural network-based NMR logging curve prediction method further includes a testing process for the target long short-term memory network to verify whether the target long short-term memory network meets the application standards. Figure 5 This is a flowchart illustrating a feature-enhanced neural network-based NMR logging curve prediction method provided in this application embodiment, demonstrating its application in situations such as... Figure 1 The following explanation uses the terminal or server shown as an example. Figure 5 As shown, the method includes:

[0146] Step 3051: Update the model parameters in the sample long short-term memory network based on the contrastive loss value to obtain the candidate long short-term memory network.

[0147] Indicatively, the target long short-term memory network is obtained by updating the weight matrix and bias in the sample long short-term memory network based on the contrastive loss value.

[0148] It should be noted that the model parameters obtained at this time are the model parameters in the sample Long Short-Term Memory network corresponding to the minimum contrastive loss value obtained based on the number of iterations.

[0149] Step 3052: Obtain the second reference NMR porosity curve and multiple candidate conventional logging curves corresponding to the candidate wells in the test dataset.

[0150] To illustrate, wells in the test dataset that have simultaneously measured multiple conventional logging curves and nuclear magnetic resonance (NMR) porosity curves are selected as candidate wells. The measured conventional logging curves are the candidate conventional logging curves corresponding to the candidate wells, and the measured NMR porosity curve is the first reference NMR porosity curve for the candidate wells.

[0151] Step 3053: Analyze the conventional logging curves of multiple samples using a candidate long short-term memory network to obtain candidate nuclear magnetic porosity curves.

[0152] Optionally, the process of obtaining the candidate NMR porosity curve can be referred to steps 202 to 205, which will not be repeated here.

[0153] To illustrate, if the entire well section of a candidate well is divided into a first well section and a second well section according to geological stratification, then the candidate NMR porosity curves for the first well section and the second well section are obtained respectively. Finally, the candidate NMR porosity curves for the first well section and the second well section are combined to obtain the candidate NMR porosity curve for the entire well section of the candidate well.

[0154] Step 3054: Calculate the correlation coefficient between the candidate NMR porosity curve and the second reference NMR porosity curve.

[0155] The correlation coefficient is used to indicate the similarity between the candidate NMR porosity curve and the second reference NMR porosity curve.

[0156] Indicatively, the candidate NMR porosity curve represents the NMR porosity curve of the entire well section of the candidate well; the second reference NMR porosity curve also represents the NMR porosity curve of the entire well section of the candidate well. The data from the two curves at the same depth are compared to see if they are within the error range. If they are within the error range, it means that the data points of the candidate NMR porosity curve at that depth meet the condition. The correlation coefficient is calculated as the ratio of the data points that meet the condition in the candidate NMR porosity curve to all data points in the candidate NMR porosity curve.

[0157] In some optional embodiments, the correlation coefficient is the average of the prediction results of multiple candidate wells. That is, the correlation coefficients corresponding to multiple candidate wells are calculated, and the average correlation coefficient is the final coefficient compared with the expected value.

[0158] Step 3055: In response to the correlation coefficient being less than the expected value, continue training the sample long short-term memory network.

[0159] Indicatively, with an expected value of 80%, if the ratio of the number of data points that meet the condition to the total number of data points in the candidate NMR porosity curve is less than 80%, it indicates that the prediction rate of the candidate long short-term memory network has not met expectations, and the sample long short-term memory network needs to be trained.

[0160] Optionally, by adjusting the model parameters in the candidate long short-term memory network, the candidate long short-term memory network with adjusted parameters is used as the initial sample long short-term memory network for retraining until the contrastive loss value of the trained network is reduced to a suitable range and the correlation coefficient reaches the expected value.

[0161] Step 3056: In response to the correlation coefficient reaching the expected value, the candidate long short-term memory network is selected as the target long short-term memory network.

[0162] Indicatively, the expected value is set to 80%. This means that the ratio of the data points that meet the condition among the data points of the candidate NMR porosity curve to all data points of the candidate NMR porosity curve is greater than or equal to 80%. The model parameters at this time are saved, and the candidate long short-term memory network corresponding to the model parameters is used as the target long short-term memory network.

[0163] In summary, the feature-enhanced neural network NMR logging curve prediction method provided in this application tests the trained candidate long short-term memory network based on the logging curves of candidate wells in the test dataset before training the target long short-term memory network. Training stops only when the correlation coefficient between the candidate NMR porosity curve predicted by the candidate long short-term memory network and the second reference NMR porosity curve reaches the expected value, thereby further improving the accuracy of the finally trained target long short-term memory network.

[0164] Based on the above process, this application provides a human-computer interaction visualization software interface diagram, which integrates the key steps involved in this application embodiment, such as data input, data preprocessing, model training, parameter optimization, and data output, into a single software device, enabling efficient and intelligent generation of nuclear magnetic resonance logging curves. Please refer to... Figure 6 The diagram illustrates a feature-enhanced nuclear magnetic resonance logging curve prediction interface. The functions within this interface are explained below:

[0165] (1) Select training file.

[0166] Clicking the "Select Training File" button (601) allows you to choose a training file stored in the terminal. For example, well logging data from sample wells in the sample dataset is selected as the training file.

[0167] (2) Data preparation.

[0168] Clicking the Import and Display Data button 602 allows you to select data from the training file mentioned above and display it in the data display box 603. In the data display box 603, each column represents a logging curve.

[0169] As an illustration, a logarithmic transformation button 604 and a normalization button 605 are displayed below the data display box 603. Selecting the logarithmic transformation button 604 and entering column number 3 in the subsequent text input box indicates that the logging curve in the third column of the data display box 603 needs to be logarithmically transformed. Selecting the normalization button 605 and entering column numbers 1, 2, and 3 in the subsequent text input box indicates that the logging curves in the first, second, and third columns of the data display box 603 need to be normalized according to their standard deviation.

[0170] For illustration, entering column label 3 in text box 606 indicates that features of the logging curve in the third column of data display box 603 need to be extracted; entering list 5 in text box 607 indicates that the logging curve in the third column of data display box 603 is a reference logging curve; entering the number 1 in text box 608 indicates that the number of parameters passed to the network for training in a single pass is 1; entering the number 50 in text box 609 indicates that the length of the sequence (logging curve corresponding to one sample well is one sequence) passed to the network for training in a single pass is 50.

[0171] For illustration purposes, clicking the Apply button 610 indicates that all operations set in the data preparation interface are performed on the logging curve in the data display box 603; clicking the Initialize button 611 indicates that the logging curve in the data display box 603 is restored to its initial state.

[0172] (3) Neural network settings.

[0173] As an illustration, page 612 allows you to set the neural network, loss function, and optimizer that you wish to apply.

[0174] Indicatively, in the neural network settings interface, the type of neural network can be selected in selection box 613, such as RNN, LSTM, etc.; entering the number 3 in text box 614 represents that the number of features input to the neural network is 3; entering the number 1 in text box 615 represents that the number of features output to the neural network is 1; entering the number 5 in text box 616 represents that the number of hidden layers is 5; and entering the number 2 in text box 617 represents that the number of network layers is 2.

[0175] As an illustration, the loss function settings interface allows you to configure the type of loss function, for example, setting the loss function of the neural network to the mean squared error function; the optimizer settings interface allows you to configure the objective function.

[0176] (4) Online training.

[0177] For illustration, entering the number 200 in text box 618 represents 200 iterations for this training session.

[0178] As an illustration, clicking the "Train and Display Loss Function" button 619 allows the network to be trained according to the parameters and number of iterations set on page 612; and the loss function's curve is displayed in the function display box 620.

[0179] (5) Network testing.

[0180] As an illustration, clicking the "Select Test File" button 621 allows you to choose a test file stored in the terminal. As an illustration, you can select logging data from candidate wells in the test dataset as the test file.

[0181] For illustration purposes, clicking the "Import Data and Apply" button 622 indicates that the candidate NMR porosity curve will be obtained by predicting the logging data of the candidate well through the trained network; clicking the "Run and Generate Comparison Curve" button 623 will generate a comparison chart and analysis data of the candidate NMR porosity curve and the corresponding reference NMR porosity curve.

[0182] (6) Network applications.

[0183] As an illustration, clicking the "Select Prediction File" button 624 allows you to choose a prediction file stored in the terminal. As an illustration, you can select the logging data of the target well in the prediction dataset as the prediction file.

[0184] For illustration purposes, clicking the "Import Data and Apply" button 625 indicates that the well logging data of the target well will be used by the trained network to predict the nuclear magnetic porosity prediction curve; clicking the "Run" button 626 will generate a schematic diagram of the nuclear magnetic porosity prediction curve.

[0185] To illustrate, logging data from 10 out of 12 wells with measured conventional logging curves and nuclear magnetic resonance porosity curves were used as training samples for the above software application. Logging data from the remaining 2 wells were used as test samples. Figure 7 As shown, comparing the actual NMR porosity logging curves of wells 701 and 702 with the predicted NMR porosity logging curves obtained by the method provided in this application embodiment, from... Figure 7 As can be seen from the data, the actual nuclear magnetic porosity logging curves and the predicted nuclear magnetic porosity logging curves of wells 701 and 702 match well, with a correlation of over 88%.

[0186] In summary, the feature-enhanced neural network-based NMR logging curve prediction method provided in this application can predict NMR logging curves with high accuracy, thus providing reliable data for subsequent fine reservoir evaluation and prediction. Illustratively, seismic inversion based on this curve can accurately predict the distribution range of sweet spot reservoirs, effectively promoting horizontal well drilling and reservoir enhancement.

[0187] Please refer to Figure 8 The diagram illustrates a structural block diagram of a feature-enhanced nuclear magnetic resonance logging curve prediction device provided in an exemplary embodiment of this application. The device includes the following modules:

[0188] The data acquisition module 810 is used to acquire multiple conventional logging curves, which are full-section logging data of the target well measured for different measurement parameters;

[0189] The joint interpretation module 820 is used to perform joint well-seismic interpretation of multiple conventional well logging curves and obtain target well section curves corresponding to the multiple conventional well logging curves respectively. The target well section curve is the well logging data corresponding to the target well section in the target well.

[0190] The cross-sectional analysis module 830 is used to perform cross-sectional analysis on the target well section curves corresponding to multiple conventional logging curves, remove the logging data of the enlarged diameter section in the target well section curve, and obtain the stable well section curve. The logging data of the enlarged diameter section is the abnormal data of the enlarged well diameter in the target well section curve.

[0191] The normalization processing module 840 is used to perform standard deviation normalization processing and combined analysis on the stable well section curve to obtain the normalized well section curve;

[0192] The predictive analysis module 850 is used to perform porosity prediction analysis on the normalized well section curve through the target long short-term memory network to obtain the nuclear magnetic porosity prediction curve of the target well section.

[0193] like Figure 9 As shown, in some optional embodiments, the data acquisition module 810 is further configured to acquire the first reference NMR porosity curve of the sample well in the sample data set and multiple sample conventional logging curves corresponding to the sample well; the device further includes:

[0194] Initialization module 860 is used to initialize the sample long short-term memory network, which includes the model parameters to be trained;

[0195] The predictive analysis module 850 is also used to analyze the conventional logging curves of multiple samples through the sample long short-term memory network to obtain the predicted nuclear magnetic porosity curve.

[0196] The calculation module 870 is used to calculate a contrast loss value based on the predicted NMR porosity curve and the first reference NMR porosity curve, wherein the contrast loss value is used to indicate the difference between the predicted NMR porosity curve and the reference NMR porosity curve;

[0197] The update module 880 is used to update the model parameters in the sample long short-term memory network based on the contrastive loss value, so as to obtain the target long short-term memory network.

[0198] In some optional embodiments, the sample dataset includes data corresponding to n sample wells; the update module 880 is further configured to iteratively update the model parameters in the sample long short-term memory network based on the contrast loss values ​​corresponding to the n sample wells respectively, to obtain the target long short-term memory network.

[0199] In some optional embodiments, the update module 880 is used to update the model parameters in the sample long short-term memory network based on the contrastive loss value to obtain a candidate long short-term memory network; the update module 880 further includes:

[0200] The acquisition unit 881 is used to acquire the second reference nuclear magnetic porosity curve and multiple candidate conventional logging curves corresponding to the candidate wells in the test dataset.

[0201] Analysis unit 882 is used to analyze the conventional logging curves of multiple samples through the candidate long short-term memory network to obtain candidate nuclear magnetic porosity curves;

[0202] The calculation unit 883 is used to calculate the correlation coefficient between the candidate NMR porosity curve and the second reference NMR porosity curve, the correlation coefficient being used to indicate the similarity between the candidate NMR porosity curve and the second reference NMR porosity curve;

[0203] Training unit 884 is used to continue training the sample long short-term memory network in response to the correlation coefficient being less than the expected value;

[0204] The determination unit 885 is used to determine 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.

[0205] In some optional embodiments, the data acquisition module 810 is used to acquire seismic data of the target well, wherein the seismic data is seismic data reflecting the geological characteristics of the target well; the joint interpretation module 820 is used to perform well-seismic joint interpretation of the seismic data and multiple conventional logging curves to obtain joint interpretation results, wherein the joint interpretation results are used to indicate the geological stratification in the target well; the joint interpretation module 820 further includes:

[0206] Selection unit 821 is used to select a target depth interval based on the joint interpretation results, wherein the target depth interval is used to indicate the depth position of the target well section in the whole well section;

[0207] The selection unit 821 is also used to select logging data of multiple conventional logging curves in the target depth range to obtain the target well section curves corresponding to the multiple conventional logging curves respectively.

[0208] In some optional embodiments, the target well section curve includes a compensated neutron curve and a compensated density curve; the cross-plot analysis module 830 is used to construct a density neutron cross-plot based on the compensated neutron curve and the compensated density curve; the cross-plot analysis module 830 is used to remove abnormal data points distributed in a specified area of ​​the quartz model line in the density neutron cross-plot; the cross-plot analysis module 830 is used to use the compensated neutron curve and the compensated density curve (after removing the abnormal data points) as the stable logging curve.

[0209] In some optional embodiments, the normalization processing module 840 is used to perform standard deviation normalization processing on the stable well section curve to obtain candidate normalized well section curves. The normalization processing module 840 includes:

[0210] Combination unit 841 is used to perform combination analysis on the candidate normalized well section curves to construct combined logging curves;

[0211] The normalization processing module 840 is also used to perform standard deviation normalization processing on the combined logging curves to obtain combined normalized well section curves.

[0212] The normalization processing module 840 is further configured to determine the candidate normalized well section curve and the combined normalized well section curve as the normalized well section curve.

[0213] In some optional embodiments, the stable well section curve includes resistivity logging curves and non-resistivity logging curves; the normalization processing module 840 includes:

[0214] Transformation unit 842 is used to perform a logarithmic transformation on the resistivity logging curve to obtain the transformed resistivity logging curve;

[0215] The normalization processing module 840 is further used to perform standard deviation normalization processing on the transformed resistivity logging curve and the non-resistivity logging curve to obtain the normalized well section curves corresponding to the transformed resistivity logging curve and the non-resistivity logging curve, respectively.

[0216] In some optional embodiments, the combination unit 841 is used to perform arithmetic operations on the stable well section curves to construct the combined logging curve; or, it is used to perform a linear combination of the stable well section curves to construct the combined logging curve; or, it is used to perform a polynomial combination of the stable well section curves to construct the combined logging curve.

[0217] In summary, the feature-enhanced NMR logging curve prediction device provided in this application obtains the target well section curve corresponding to the target well section in the target well by performing joint well-seismic interpretation of multiple conventional logging curves. By analyzing a portion of the target well section instead of the entire well section, the value range difference of the data to be analyzed in a single instance is reduced, improving the accuracy of the NMR porosity prediction curve. By performing curve intersection analysis on the target well section curve, abnormal data is removed, and the curves with removed abnormal data are normalized according to their standard deviation to obtain normalized well section curves. Finally, the curves are predicted and analyzed using a target long short-term memory network to obtain the NMR porosity prediction curve of the target well section. Since the target long short-term memory network combines short-term memory and long-term memory through a gating mechanism, it can analyze the changing trend of conventional logging data with depth and the correlation between data, thereby improving the accuracy of the predicted NMR porosity curve.

[0218] It should be noted that the feature-enhanced NMR logging curve prediction device provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the feature-enhanced NMR logging curve prediction device and the feature-enhanced neural network NMR logging curve prediction method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0219] Figure 10 This illustration shows a schematic diagram of a server provided in an exemplary embodiment of this application. The server may be as follows: Figure 1 The server shown here. Specifically, it includes the following structure:

[0220] Server 1000 includes a Central Processing Unit (CPU) 1001, a system memory 1004 including Random Access Memory (RAM) 1002 and Read Only Memory (ROM) 1003, and a system bus 1005 connecting the system memory 1004 and the CPU 1001. Server 1000 also includes a mass storage device 1006 for storing an operating system 1013, application programs 1014, and other program modules 1015.

[0221] Mass storage device 1006 is connected to central processing unit 1001 via a mass storage controller (not shown) connected to system bus 1005. Mass storage device 1006 and its associated computer-readable media provide non-volatile storage for server 1000. That is, mass storage device 1006 may include computer-readable media (not shown) such as hard disk or compact disc read-only memory (CD-ROM) drives.

[0222] Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state storage technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, 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 above-mentioned types. The system memory 1004 and mass storage device 1006 described above can be collectively referred to as memory.

[0223] According to various embodiments of this application, server 1000 can also be connected to a remote computer on a network, such as the Internet. That is, server 1000 can be connected to network 1012 via network interface unit 1011 connected to system bus 1005, or it can also use network interface unit 1011 to connect to other types of networks or remote computer systems (not shown).

[0224] The aforementioned memory also includes one or more programs, which are stored in the memory and configured to be executed by the CPU.

[0225] Embodiments of this application also provide a computer device that can be implemented as follows: Figure 3 The terminal or server shown. The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, code set, or instruction set, the at least one instruction, at least one program, code set, or instruction set being loaded and executed by the processor to implement the feature-enhanced neural network nuclear magnetic logging curve prediction method provided in the above-described method embodiments.

[0226] Embodiments of this application also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the feature-enhanced neural network nuclear magnetic logging curve prediction method provided in the above-described method embodiments.

[0227] Embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the feature-enhanced neural network-based NMR logging curve prediction method provided in the above-described method embodiments.

[0228] Optionally, the computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), solid-state drives (SSDs), or optical discs, etc. The random access memory may include resistive random access memory (ReRAM) and dynamic random access memory (DRAM). The sequence numbers of the embodiments in this application are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0229] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0230] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A neural network-based NMR logging curve prediction method based on feature enhancement, characterized in that, The method includes: Acquire multiple conventional logging curves, which are full-section logging data of the target well measured for different measurement parameters; Multiple conventional logging curves are interpreted in a combined well-seismic analysis to obtain target well section curves corresponding to the multiple conventional logging curves. The target well section curves are logging data corresponding to the target well section in the target well. A curve intersection analysis is performed on the target well section curves corresponding to multiple conventional logging curves. The logging data of the enlarged diameter section in the target well section curve is removed to obtain the stable well section curve. The logging data of the enlarged diameter section is the abnormal data of well diameter enlargement in the target well section curve. The standard deviation of the stable well section curve is normalized and combined analysis is performed to obtain the normalized well section curve; The porosity prediction analysis of the normalized well section curve is performed by the target long short-term memory network to obtain the nuclear magnetic resonance porosity prediction curve of the target well section. The target well section curve includes a compensated neutron curve and a compensated density curve; The process involves performing curve cross-plot analysis on the target well section curves corresponding to multiple conventional logging curves, removing the logging data for the enlarged diameter section of the target well section curves, and obtaining the stable well section curves, including: Based on the compensated neutron curve and the compensated density curve, a density neutron intersection planar diagram is constructed. Remove outlier data points distributed in a specified region of the quartz model line in the density neutron intersection plane diagram; The compensated neutron curve and the compensated density curve after removing the abnormal data points are used as the stable well section curve.

2. The method according to claim 1, characterized in that, The training process of the target long short-term memory network includes: Obtain the first reference nuclear magnetic porosity curve of the sample well in the sample dataset and multiple sample conventional logging curves corresponding to the sample well; Initialize the sample long short-term memory network, which includes the model parameters to be trained; By analyzing the conventional logging curves of multiple samples using the sample long short-term memory network, a predicted nuclear magnetic porosity curve is obtained. Based on the predicted NMR porosity curve and the first reference NMR porosity curve, a contrast loss value is calculated, which is used to indicate the difference between the predicted NMR porosity curve and the reference NMR porosity curve; The model parameters in the sample long short-term memory network are updated based on the contrastive loss value to obtain the target long short-term memory network.

3. The method according to claim 2, characterized in that, The sample dataset includes data corresponding to n sample wells; Updating the model parameters in the sample long short-term memory network based on the contrastive loss value includes: The model parameters in the sample long short-term memory network are updated iteratively based on the contrastive loss values ​​corresponding to n sample wells to obtain the target long short-term memory network.

4. The method according to claim 2, characterized in that, The step of updating the model parameters in the sample long short-term memory network based on the contrastive loss value to obtain the target long short-term memory network includes: Based on the contrastive loss value, the model parameters in the sample long short-term memory network are updated to obtain the candidate long short-term memory network; Obtain the second reference NMR porosity curve and multiple candidate conventional logging curves corresponding to candidate wells in the test dataset; Candidate NMR porosity curves were obtained by analyzing the conventional logging curves of multiple samples using the candidate long short-term memory network. Calculate the correlation coefficient between the candidate NMR porosity curve and the second reference NMR porosity curve, the correlation coefficient being used to indicate the similarity between the candidate NMR porosity curve and the second reference NMR porosity curve; In response to the correlation coefficient being less than the expected value, the long short-term memory network of the sample continues to be trained; In response to the correlation coefficient reaching the expected value, the candidate long short-term memory network is selected as the target long short-term memory network.

5. The method according to claim 1, characterized in that, The method of performing joint well-seismic interpretation of multiple conventional well logging curves to obtain the target well section curves corresponding to each of the multiple conventional well logging curves includes: Obtain seismic data of the target well, wherein the seismic data is seismic data reflecting the geological characteristics of the target well; The seismic data and multiple conventional well logging curves are subjected to joint well-seismic interpretation to obtain joint interpretation results, which are used to indicate the geological stratification in the target well. Based on the joint interpretation results, a target depth range is selected, which is used to indicate the depth position of the target well section in the entire well section; By selecting logging data from multiple conventional logging curves within the target depth range, the target well section curves corresponding to the multiple conventional logging curves are obtained.

6. The method according to claim 1, characterized in that, The standard deviation normalization and combined analysis of the stable well section curves are performed to obtain normalized well section curves, including: The standard deviation of the stable well section curve is normalized to obtain the candidate normalized well section curve; The candidate normalized well section curves are combined and analyzed to construct combined logging curves; The combined logging curves are normalized by standard deviation to obtain the combined normalized well section curves; The candidate normalized well section curve and the combined normalized well section curve are determined as the normalized well section curve.

7. The method according to claim 6, characterized in that, The stable well section curves include resistivity logging curves and non-resistivity logging curves; The step of normalizing the standard deviation of the stable well section curve to obtain candidate normalized well section curves includes: Perform a logarithmic transformation on the resistivity logging curves to obtain the transformed resistivity logging curves; The standard deviation of the transformed resistivity logging curve and the non-resistivity logging curve is normalized to obtain the candidate normalized well section curves corresponding to the transformed resistivity logging curve and the non-resistivity logging curve, respectively.

8. The method according to claim 6, characterized in that, The step of performing combined analysis on the candidate normalized well interval curves to construct combined logging curves includes: Perform arithmetic operations on the candidate normalized well section curves to construct the combined logging curves; or, The candidate normalized well interval curves are linearly combined to construct the combined logging curve; or, The candidate normalized well section curves are combined using a polynomial to construct the combined logging curve.

9. A feature-enhanced nuclear magnetic resonance logging curve prediction device, characterized in that, The device includes: The data acquisition module is used to acquire multiple conventional logging curves, which are full-section logging data of the target well measured for different measurement parameters; The joint interpretation module is used to perform joint well-seismic interpretation of multiple conventional well logging curves and obtain the target well section curves corresponding to the multiple conventional well logging curves respectively. The target well section curves are the well logging data corresponding to the target well section in the target well. The cross-cutting analysis module is used to perform cross-cutting analysis on target well section curves corresponding to multiple conventional logging curves, remove the logging data of the enlarged diameter section in the target well section curve, and obtain the stable well section curve. The logging data of the enlarged diameter section is the abnormal data of well diameter enlargement in the target well section curve. The normalization module is used to perform standard deviation normalization and combined analysis on the stable well section curve to obtain the normalized well section curve; The predictive analysis module is used to perform porosity prediction analysis on the normalized well section curve through the target long short-term memory network to obtain the nuclear magnetic porosity prediction curve of the target well section. The target well section curve includes a compensated neutron curve and a compensated density curve; The cross-intersection analysis module is specifically used for: constructing a density neutron cross-intersection planar diagram based on the compensated neutron curve and the compensated density curve; removing abnormal data points distributed in a specified area of ​​the quartz model line in the density neutron cross-intersection planar diagram; and using the compensated neutron curve and the compensated density curve after removing the abnormal data points as the stable well section curve.

Images