CN114444393A

CN114444393A - Logging curve construction method and device based on time convolution neural network

Info

Publication number: CN114444393A
Application number: CN202210095698.4A
Authority: CN
Inventors: 岳明; 王鸣川; 宋田茹; 朱维耀; 宋洪庆; 范琳
Original assignee: University of Science and Technology Beijing USTB
Current assignee: University of Science and Technology Beijing USTB
Priority date: 2022-01-26
Filing date: 2022-01-26
Publication date: 2022-05-06

Abstract

The disclosure provides a model building method and a logging curve building method. The model establishing method comprises the following steps: acquiring label data and characteristic data; randomly mixing the label data and the characteristic data, and dividing the mixed data into a training set and a verification set according to a proportion; performing sequence modeling based on a time convolution neural network according to the tag data and the characteristic data; training the established model based on a training set, and verifying the effectiveness of the model through a verification set; and when the value of the loss function of the model meets the preset condition and is not reduced any more, saving the model. The well logging curve construction method comprises the following steps: acquiring data of a known logging curve of a target well; and inputting the data of the known logging curve of the target well into the model, and constructing the unknown logging curve of the target well. The method and the device can realize prediction of the unknown logging curve of the shale gas reservoir with strong geological heterogeneity.

Description

Logging curve construction method and device based on time convolution neural network

Technical Field

The disclosure relates to the technical field of oilfield development and logging, in particular to a logging curve construction method and device based on a time convolution neural network.

Background

The well logging curve contains a large amount of lithology information, provides an important data source for solving the problem of lithology identification, and is of great importance for exploration and development research of oil and gas resources.

In practical application, due to the influence of factors such as hole diameter expansion, well wall collapse and instrument failure, the problem of partial logging data distortion or loss is often caused, and certain difficulty is brought to subsequent interpretation work. Re-logging is not only expensive, but may even be impossible for a section of the wellbore that has already been completed. Therefore, it is of great significance to explore and develop a logging curve prediction method and predict logging data of distorted or missing well sections so as to increase the accuracy of logging interpretation.

The well logging curve prediction is a method for predicting an unknown curve by using the relationship between the well logging curve and the unknown curve existing in the data. Because the underground condition is complex and the heterogeneity is strong, the logging data of the same well often presents extremely strong nonlinear mapping relation, the traditional interpolation method can not accurately predict the logging curve, and the method based on the physical model has very strong hypothesis conditions, thereby greatly simplifying the real stratum information.

With the development of big data and artificial intelligence, deep learning is widely applied to various industries, wherein a neural network has strong nonlinear data processing capacity, so that the neural network can be used for effectively constructing unknown logging curves.

Disclosure of Invention

The invention provides a time convolution neural network-based unknown logging curve construction method and device, aiming at the problem of unknown logging curve prediction, in particular to the unknown logging curve prediction of a shale gas reservoir with strong geological heterogeneity.

In one aspect, some embodiments of the present disclosure provide a model building method, which includes S1 to S5.

And S1, acquiring the label data and the characteristic data. Wherein the label data is data of one logging curve, and the characteristic data is data of at least one other logging curve; and the data of any one logging curve is a data set of the data of the same logging curve of a plurality of sample wells in the same block.

And S2, randomly mixing the label data and the feature data, and dividing the mixed data into a training set and a verification set according to a proportion.

And S3, performing sequence modeling based on the time convolution neural network according to the label data and the characteristic data.

And S4, training the established model based on the training set, and verifying the validity of the model through the verification set.

S5, when the loss function value of the model meets the preset condition and does not decrease any more, saving the model.

In another aspect, some embodiments of the present disclosure further provide a modeling apparatus, which includes a first processor and a first memory, where the first memory stores computer program instructions adapted to be executed by the first processor, and the computer program instructions, when executed by the first processor, perform the steps of the modeling method according to any one of the above embodiments.

In yet another aspect, some embodiments of the present disclosure also provide a method for constructing a well log. The method utilizes the model established by the model establishing method in any embodiment; the method comprises S10-S20.

S10, acquiring the data of the known logging curve of the target well;

and S20, inputting the data of the known logging curve of the target well into the model, and constructing the unknown logging curve of the target well.

And the type of the known well logging curve of the target well is the same as that of the well logging curve in the characteristic data, and the type of the unknown well logging curve of the target well is the same as that of the well logging curve in the label data.

In yet another aspect, some embodiments of the present disclosure further provide a well log construction apparatus, which includes a second processor and a second memory, where the second memory stores computer program instructions adapted to be executed by the second processor, and the computer program instructions, when executed by the second processor, perform the steps of the well log construction method according to any of the above embodiments.

In yet another aspect, some embodiments of the present disclosure also provide a computer-readable storage medium having stored therein computer program instructions, which, when executed by a processor of a user equipment, cause the user equipment to perform a method of well log construction as described in any of the above embodiments.

Drawings

The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the disclosure and together with the description serve to explain the principles of the disclosure.

FIG. 1 is a flow diagram of a model building method according to some embodiments;

FIG. 2 is a flow diagram of a method of log construction according to some embodiments;

FIG. 3 is a schematic diagram of the construction of a DEN curve of a well log construction method according to some embodiments.

Detailed Description

The present disclosure will be described in further detail with reference to the drawings and embodiments. It is to be understood that the specific embodiments described herein are for purposes of illustration only and are not to be construed as limitations of the present disclosure. It should be further noted that, for the convenience of description, only the portions relevant to the present disclosure are shown in the drawings.

It should be noted that the embodiments and features of the embodiments in the present disclosure may be combined with each other without conflict. The present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

It should be noted that, the step numbers in the text are only for convenience of explanation of the specific embodiments, and do not serve to limit the execution sequence of the steps.

The methods provided by some embodiments of the present disclosure may be executed by a relevant processor, and are all described below by taking the processor as an example of an execution subject. The execution subject can be adjusted according to the specific case, such as a server, an electronic device, a computer, and the like.

The geological heterogeneity of the shale gas reservoir is strong, the nonlinear mapping relation among the logging data of the same well is complex, and the conventional logging method has large errors in the aspect of predicting unknown logging curves of the same well of the shale gas reservoir. The invention provides a model establishing method and device, a logging curve constructing method and device and a computer readable storage medium based on a time convolution neural network, aiming at the problem of unknown logging curve prediction, in particular to the unknown logging curve prediction of a shale gas reservoir with strong geological heterogeneity.

As shown in FIG. 1, some embodiments of the present disclosure provide a model building method, which includes S1-S5.

And S1, acquiring the label data and the characteristic data. The label data is data of one logging curve, and the characteristic data is data of at least one other logging curve; the data of any kind of well logging curve is a data set of data of the same well logging curve of a plurality of sample wells in the same block.

Exemplary, the types of logs may include depth curves, caliper curves, density curves, natural potential curves, natural gamma curves, compensated neutron curves, litho-density curves, neutron porosity curves, sonic moveout curves, deep lateral resistivity curves, shallow lateral resistivity curves, and the like. The data of any kind of well logging curve can be used as label data, and the data corresponding to other kinds of well logging curves can be used as characteristic data. The data for any of the logs may be a consolidated data set of log data for a plurality of wells. For example, A, B, C three wells are located in the same block, and the data of the depth curve can be a data set formed by combining the depth curve data of the A well, the depth curve data of the B well and the depth curve data of the C well.

Alternatively, the test ratio of the training set to the validation set may be 8: 2.

And S5, when the value of the loss function of the model meets the preset condition, saving the model.

According to the model establishing method provided by some embodiments of the disclosure, the known well logging curves of one or more wells in the same block are used as data sources for modeling training, modeling is performed based on a time convolution neural network, and a model capable of effectively predicting unknown well logging curves of target wells in the same block is established. For well logging data, the data varies with the depth of the well bore, and the well bore usually has a length of 2000m to 4000m, and the data is measured every 0.125m or 0.25m to generate a group of data, so that the data volume of each well logging curve is larger. The geological information reflected by the logging parameters of the same reservoir at different depths is basically consistent, and the data correlation is strong, so that the historical information is considered to be crucial to the construction of an unknown logging curve, and the time convolution neural network model can retain the historical information to a greater extent, and is suitable for solving the problem of the construction of the unmeasured logging curve. The complex nonlinear relation between different logging curves can be extracted and learned through machine learning, and compared with the traditional interpolation method and other methods, the model established by the method is higher in accuracy.

The model established by the model establishing method disclosed by the invention can be applied to the establishment of complex logging curves of shale gas fields and the like, and has the advantages of high establishing accuracy, simple method and high calculating speed. The logging curve constructed by the model established by the method can provide important technical support for oil field development and has a guiding effect on field operation.

In some embodiments, the model building method further includes S6-S7.

And S6, acquiring feature source data. The characteristic source data includes data of at least one log.

And S7, for the characteristic source data, respectively analyzing the correlation between the data of each logging curve and the label data, and taking the data of which the correlation analysis result meets the preset condition as the characteristic data.

Here, in step S7, the correlation analysis may employ at least one of Pearson correlation coefficient, Kendall rank correlation coefficient, and Spearman rank correlation coefficient.

The three methods of the correlation analysis and the predetermined condition for determining the high correlation between the data of a certain log and the tag data are described in detail below.

The linear correlation between the data of a certain logging curve and the tag data is quantitatively analyzed by using a Pearson correlation coefficient, which mainly comprises the following steps:

two variables are provided, x and y, for example, x is data of a first log and y is data of a second log, wherein the data of the second log may be tag data. The Pearson correlation coefficient between x and y is defined as the ratio of the covariance between the two variables x, y and the product of the two standard deviations. The Pearson correlation coefficient may be calculated by the following formula:

wherein cov (x, y) is the covariance of x and y, σ_xIs the standard deviation of x, σ_yIs the standard deviation of y, and E is the mathematical expectation.

ρ_xyRepresenting Pearson correlation coefficient with the value range of [ -1,1](ii) a The positive and negative represent the linear direction of correlation between the variables, p_xy> 0 denotes a linear positive correlation, p_xy< 0 indicates a linear negative correlation. For well log data, consider 0.5 < | ρ_xyI.ltoreq.1 indicates high linear correlation.

The Kendall rank correlation coefficient τ is:

wherein i, j is 1,2, … …, n;

(x_i，y_i) Is observed data, i.e. actual measurements, x, of a log of some kind_iCorresponding to the i-th measured data, y, on a certain log_iCorresponding to the ith measured data on the other logging curve;

is a sign function.

And the absolute value of the Kendall rank correlation coefficient tau is equal to 1, and the variables are considered to have high correlation.

Spearman rank correlation coefficient δ is:

in the formula, R_iIs the order of variable x, S_iRank of variable y;

is R_iThe average value of (a) of (b),

is S_i1,2, … …, n.

The absolute value of the Spearman rank correlation coefficient δ is greater than or equal to 0.5, the correlation between variables is considered high, otherwise the correlation is considered low.

In some embodiments of the present disclosure, one or more correlation analysis methods may be adopted, and the correlation analysis result of the feature data may satisfy any one of the preset conditions.

In some embodiments, before step S3, the model building method includes: and S8, performing multi-scale feature noise reduction processing on the label data and the feature data.

Here, the multi-scale processing means that data is divided into two types, i.e., a high-frequency signal and a low-frequency signal, and processed separately when data is denoised.

The data on the logging curve has a plurality of statistical fluctuations and burr interference which are irrelevant to the stratum property, the multi-scale characteristic noise reduction processing is carried out on the logging curve data, only the information reflecting the stratum characteristic is reserved, the interference of the interference data on the logging curve prediction result can be eliminated to the maximum extent, the precision of the trained model is improved, and therefore the prediction accuracy of the unknown logging curve of the target well is improved.

For example, local features of the log may be multiscale analyzed by continuous wavelet transform to achieve de-noising of the signature data and feature data. The wavelet transform adopts a method of changing the shape of a time-frequency window, has self-adaptive characteristics, can analyze the characteristics of a logging curve in a multi-scale mode, and after wavelet decomposition, most of wavelet coefficients with larger amplitude values are useful signals, and the coefficients with smaller amplitude values are noise generally.

For any one

Satisfy the requirements of

For basic wavelets (wavelet mother functions), scale factors (scale factors) a and translation factors b are introduced, and a and b satisfy the following conditions: a, b ∈ R, a ≠ 0, and the basic wavelet is subjected to shrinkage and translation to obtain the following function family:

to analyze the wavelet, the coefficient | a Y^-1/2Is a normalization constant.

If it is not

Satisfies a tolerable condition

Wherein R represents the whole of non-zero real numbers,

is a wavelet mother function. For an arbitrary real pair (a, b), where the parameter a must be a non-zero real number, the wavelet basis function is:

for any energy limited signal f (t), the wavelet transform is defined as:

if the length of the noise-contaminated signal is n, the expression of the noise-contaminated signal is as follows:

e(t)＝x(t)+z(t)

where x (t) represents the desired signal and z (t) represents the noise signal. The wavelet de-noising process is to remove the noise signal from the noise-contaminated signal as much as possible without damaging the useful signal.

As a possible implementation, a wavelet threshold denoising method may be adopted to achieve denoising of data in the original data set.

The wavelet threshold denoising method mainly comprises the following steps:

(1) and (3) decomposition: determining the size of wavelet decomposition, selecting wavelet signal with N decomposition layers for wavelet decomposition to obtain a group of wavelet coefficients omega_j,kWavelet coefficients are mainly divided into high-frequency coefficients and low-frequency coefficients. For the high-frequency coefficient, the following steps (2) and (3) are carried out step by step, and for the low-frequency coefficient, the step (3) is directly carried out, namely reconstruction is carried out through wavelet inverse transformation.

(2) Threshold processing: setting threshold value for decomposition value under high resolution, i.e. high frequency coefficient, quantizing wavelet coefficient to obtain estimation value of wavelet coefficient

Wherein, the threshold function adopts a Garrote function:

selecting a general threshold (VisuShrink threshold), and aiming at the multidimensional independent normal variable joint distribution, obtaining an optimal threshold under the limitation of maximum-minimum estimation, wherein the optimal threshold meets the following conditions:

where λ represents a threshold, σ represents a noise standard deviation, and N is a length of the signal.

The noise standard deviation is estimated by:

wherein, mean represents the median of the absolute values of the coefficients obtained by performing the wavelet decomposition of the first layer on the logging data, and d (k) represents the wavelet coefficient values of the k layers of the wavelet decomposition.

(3) Wavelet reconstruction: and performing wavelet inverse transformation on the estimated value of the wavelet coefficient to obtain a denoised signal.

And after denoising is finished, evaluating the denoising effect through the signal-to-noise ratio.

The signal-to-noise ratio calculation mode is as follows:

wherein f is_iFor denoised logging signals (data), g_iIs the original log signal (data). When the signal-to-noise ratio is larger than 15, the good denoising effect can be considered to be achieved.

The data processing process before modeling will be described below by taking A, B, C, D wells in the same block as an example.

A. B, C, D the depth curve, the well diameter curve, the density curve and the natural potential curve of A, B, C three wells are all known logging curves. And D, the well depth curve, the well diameter curve and the density curve are known logging curves, and the natural potential curve is an unknown logging curve. After the model is established by the model establishing method disclosed by the invention, the natural potential curve of the D well can be established or predicted by utilizing the model. The type of the logging curve corresponding to the label data is the same as that of the logging curve of the D well to be predicted, namely a natural potential curve.

Therefore, the label data is data corresponding to the natural potential curves of A, B, C wells, that is, a data set obtained by combining the natural potential curve data of the well a, the natural potential curve data of the well B, and the natural potential curve data of the well C.

Merging A, B, C the data of the depth curves of the three wells as the data of the depth curves; combining A, B, C well diameter curve data of three wells to be used as well diameter curve data; the density curve data of the A, B, C three wells were merged as density curve data. And taking the data of the depth curve, the data of the hole diameter curve and the data of the density curve after combination as characteristic source data.

And analyzing the correlation between the data of the depth curve and the tag data, the correlation between the data of the well diameter curve and the tag data and the correlation between the data of the density curve and the tag data by adopting at least one method of a Pearson correlation coefficient, a Kendall rank correlation coefficient and a Spearman rank correlation coefficient, and selecting the data of the depth curve and the data of the well diameter curve which meet the preset condition of high correlation as characteristic data.

The data of the depth curve, the data of the well diameter curve and the label data can be used as a data set X; wherein, X is an M multiplied by N matrix, M is the number of the collected data sample points, N is the number of the types of the logging curves of which the correlation degree with the label data meets the preset condition, and N is a positive integer greater than or equal to 1. Here, N is 2.

Carrying out multi-scale characteristic noise reduction processing on the data of the depth curve, the data of the well diameter curve and the label data to obtain a denoised data set X₁，X₁Is a matrix of order mxn.

Data set X₁The data in (1) are randomly mixed and divided into a training set X _ train and a verification set X _ validation in a ratio of 8: 2.

A logging curve prediction model based on a time convolution neural network is constructed based on the denoised data, the unknown logging curve can be well predicted and constructed, and the curve construction accuracy is improved.

A TCN (Temporal convolutional network) model can receive an input sequence of arbitrary length and map it to an output sequence of equal length, with each time instant being computed simultaneously, rather than serially in time. For well log data, the data varies with the depth of the wellbore, and the wellbore usually has a length of 2000m to 4000m, and the data is measured every 0.125m or 0.25m to generate a set of data, so that the data volume of each well log is relatively large. The geological information reflected by the logging parameters of the same reservoir at different depths is basically consistent, and the data correlation is strong, so that the historical information is considered to be important for constructing an unknown logging curve, and the TCN model can retain the historical information to a greater extent, thereby being beneficial to solving the problem of constructing the unmeasured logging curve.

The TCN model mainly comprises a causal convolutional layer, an expansion convolutional layer and a residual connection module.

The causal convolution is a one-way structure and is a strict time constraint model, and the value at the time t of the upper layer depends only on the value at the time t of the lower layer and the value before the time t of the lower layer. Filter F ═ F₁,f₂,L,f_K) For the sequence problem X ═ X₁,x₂,L,x_t) At x_iThe causal convolution of (a) is:

where K denotes the kth filter and K denotes the total of K filters.

The expansion convolution is to inject holes on the basis of standard convolution so as to increase the receptive field and capture longer dependency relationship, and the related hyper-parameter is the expansion rate d, namely the interval number of convolution kernels. Note that the image size is n × n, the convolution kernel size is f × f, the step size is s, and the padding is pa, then the size of the normal convolution output is:

on the basis, the expansion convolution is to fill 0 into a common convolution kernel, the size of the convolution kernel is changed by the expansion rate d, and other methods are the same as the common convolution, and the calculation formula of the convolution kernel is as follows:

kernal_size＝d×(kernal_size-1)+1

filter F ═ F₁,f₂,L,f_k) For the sequence problem X ═ X₁,x₂,L,x_t) At x_iThe swell convolution at swell ratio d is:

when the expansion ratio d is 1, the convolution is normal.

Aiming at the problem of gradient disappearance or explosion easily appearing in the neural network, the residual error network is added into the identity mapping of cross-layer connection, and the learning transformation function H (x) is changed into the learning residual error function F (x) H (x) -x.

Based on this, in some embodiments, the model established in step S3 may include a one-dimensional full convolution network and a causal convolution, and meanwhile, a residual connection and an inflation convolution are used to obtain longer historical sequence information, where the specific network structure is: the device comprises an input layer, a plurality of hidden layers, a full connection layer and an output layer which are connected in sequence.

The number N of the hidden layers can be set according to the complexity of the practical problem handling; each hidden layer is formed by stacking a plurality of same residual modules, and each residual module comprises a convolutional layer, a nonlinear mapping unit, a Dropout layer and a residual connecting unit which are sequentially connected. The residual error connection unit enables the network to transfer information in a cross-layer mode by learning an identity mapping function.

In addition, the convolution kernel size of the convolution layer can be further set, and if the expansion convolution is adopted, the expansion rate and the step length are also required to be set; selecting a certain nonlinear mapping unit, such as a ReLU function or a Softmax function; the ratio of Dropout is set and includes values of 0.2, 0.3, 0.4, 0.5, etc.

Let X _ temp be the input value of the residual error module, and the identity mapping function adopted during network cross-layer is F (·), the Output value Output of the residual error module can be expressed as:

Output＝Activation(X_temp+F(X_temp))

wherein Activation represents a non-linear mapping function.

The built model may then be trained based on a training set, and the validity of the model may be verified by a validation set.

Illustratively, a training set X _ train and a verification set X _ validation are respectively converted into a three-dimensional input format B × T × N set by the model, where B is a sample size and T is a time step; the transformed training set is named as X _ train _3D, and the verification set is named as X _ evaluation _ 3D. And inputting the training set X _ train _3D into the established model, training model parameters, and verifying the validity of the model through the verification set X _ validation _ 3D. Alternatively, the loss function of the model may adopt a Mean Absolute Error (MAE), and the loss function may intuitively reflect the magnitude of the Absolute Error between the predicted value and the true value.

The MAE calculation formula is:

wherein, y_iThe actual value is represented by the value of,

representing the predicted value, and n is the number of samples.

The loss function is

Here, the preset condition of the loss function is that the MAE value is less than 1 and does not decrease any more. That is, when the MAE value is smaller than 1, the training of the model is continued, and the MAE value remains unchanged or slightly increases, so that the trained model satisfies the preset condition of the loss function.

Illustratively, if the Loss value (MAE value) is less than 1 and does not decrease any more during the training process, the model is successfully trained, and the model is saved. The model may then be used to construct an unknown log of the target well in the same block.

As shown in fig. 2, an embodiment of the present disclosure further provides a well log construction method, which constructs an unknown well log of a target well by using a model established by any one of the above-mentioned model establishment methods. The method comprises S10-S20.

And S10, acquiring the data of the known logging curve of the target well.

The type of the known logging curve of the target well is the same as that of the logging curve in the characteristic data, and the type of the unknown logging curve of the target well is the same as that of the logging curve in the label data.

Again taking A, B, C, D wells above as an example, the natural potential profile of the target well D-well is the unknown log. As can be seen from the above, the data of the depth curve and the data of the hole diameter curve are characteristic data having a high correlation with the tag data.

And taking the data of the known well logging curves of the target well D well, namely the data of the depth curve and the data of the well diameter curve of the D well as a test set. Before the test set is substituted into the built model, the test set can be preprocessed to obtain a denoised test set X _ test.

And inputting the denoised test set X _ test into the established model, so as to construct an unknown logging curve of the target well, namely a natural potential curve of the D well.

The well logging curve construction method based on the time convolution neural network can be applied to construction of complex well logging curves such as shale gas fields, an unknown well logging curve of a target well is constructed based on a known well logging curve of a block where the target well is located, and the method is high in construction accuracy, simple and fast in calculation speed. The logging curve constructed by the method can provide important technical support for oil field development and has a guiding effect on field operation.

The method of the present disclosure will be described in detail below with a horizontal well W of the Fuling shale gas field as a target well.

The logging data of each horizontal well in the well block mainly comprises six curves of DEPTH (DEPTH), well diameter (CAL), Density (DEN), natural Gamma (GR), Compensation Neutron (CNL) and sound wave time difference (AC). The depth measurement is 2400 m-4000m, the measurement interval is 0.125m, so that 12800 groups of data are totally collected, and the size of an original data set is 12800 multiplied by 6.

And taking the DEN log as an unknown log of the W well, and taking the other logs as known logs.

Therefore, the data set of the DEN logs of each horizontal well other than the target well is used as the label data, and the log data of the DEN, CAL, GR, CNL, and AC logs are used as the feature source data.

The method for selecting the Spearson rank correlation coefficient performs correlation analysis on the data of the DEPTH, CAL, GR, CNL and AC well logs and the data of the DEN well log.

The results of the correlation analysis are as follows:

spearson rank correlation coefficient delta	DEPTH	CAL	CNL	GR	DEN	AC
							DEN	0.83	0.22	0.63	0.2	1.0	0.82

As can be seen from the table, the logging curves having high correlation with the DEN logging curve are the dept, CNL and AC logging curves, and therefore, data corresponding to the dept, CNL and AC logging curves of the horizontal wells other than the target well are used as characteristic data of the model, data corresponding to the DEN logging curve are used as tag data, and the size of the data set X is 12800 × 4.

And denoising the data of the four well logging curves of the data set X and the data corresponding to the DEPTH, CNL and AC well logging curves of the W well. And carrying out multi-scale characteristic noise reduction on the logging curve data by adopting a continuous wavelet method, and calculating the signal-to-noise ratio of the logging curve data before and after noise reduction. Aiming at the DEN logging curve, the calculated signal-to-noise ratio is larger than 20, the good noise reduction effect is considered to be achieved, and the method can be used for establishing a subsequent model and predicting an unknown logging curve.

The data in the noise-reduced data set X are randomly mixed, then the first 80% of the data is used as a training set X _ train, and the next 20% is used as a verification set X _ evaluation.

And taking data corresponding to DEPTH, CNL and AC well logging curves of the W well after noise reduction as a test set X _ test.

The two-dimensional matrices X _ train, X _ evaluation, and X _ test are converted into three-dimensional matrices X _ train _3D, X _ evaluation _3D and X _ test _ 3D. Specifically, the time step is set to 1, and thus X _ train _3D is 256 × 1 × 4 in size, X _ evaluation _3D is 12800 × 1 × 4, and X _ test _3D is 12800 × 1 × 4.

And further, constructing a time convolution neural network model. And setting hidden layers of the model as 8 completely identical residual modules, and adopting expansion convolution according to the convolution kernel size of the convolution layer of each residual module. The convolution kernel size of the first convolution layer and the second convolution layer is set to be 7 x 1, the expansion rate d is 1, and the step length is 2; the convolution kernel size of the third convolution layer and the fourth convolution layer is 5 × 1, the expansion rate d is 2, and the step length is 4; the convolution kernels of the fifth convolution layer, the sixth convolution layer and the seventh convolution layer are 3 x 1, the expansion rate d is 4, and the step length is 8; the convolution kernel size of the eighth convolution layer is 1 × 1, the expansion d is 8, and the step size is 16. Further, the nonlinear mapping unit of each residual block uses the ReLU function, setting the ratio of Dropout to 0.2.

Inputting the training set X _ train _3D into the constructed model based on the time convolution neural network, training model parameters, and verifying the effectiveness of the model through the verification set X _ validation _ 3D. And (3) adopting the average absolute error MAE as a loss function of the model, and when the MAE value is less than 1 and is in a stable state (not reduced any more), considering that the training of the model is finished, and storing the model.

And predicting the test set X _ test _3D by using the stored model, namely inputting the test set X _ test _3D into the model, namely outputting the DEN logging curve of the W well, thereby realizing the construction of the unknown logging curve of the target well. The results of the DEN logs predicted by the model are shown in figure 3. The unknown logging curve of the target well is predicted by the model established by the method, and the construction accuracy is very high.

Some embodiments of the present disclosure also provide a modeling apparatus comprising a first processor and a first memory, the first memory having stored therein computer program instructions adapted to be executed by the first processor, the computer program instructions, when executed by the first processor, performing the steps of the modeling method according to any of the above embodiments.

Some embodiments of the present disclosure further provide a well log construction device, which includes a second processor and a second memory, where the second memory stores computer program instructions adapted to be executed by the second processor, and the computer program instructions, when executed by the second processor, perform the steps of the well log construction method according to any of the above embodiments.

The first/second processor may be a Central Processing Unit (CPU), or other general-purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or the like. Wherein a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The first/second Memory may be a Read-Only Memory (ROM) or other types of static storage devices that can store static information and instructions, a Random Access Memory (RAM) or other types of dynamic storage devices that can store information and instructions, an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM) or other optical Disc storage, optical Disc storage (including Compact Disc, laser Disc, optical Disc, digital versatile Disc, blu-ray Disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory may be self-contained and coupled to the processor via a communication bus. The memory may also be integral to the processor.

Some embodiments of the present disclosure also provide a computer-readable storage medium having computer program instructions stored therein, which, when executed by a processor of a user equipment, cause the user equipment to perform the log construction method of any of the above embodiments.

Computer-readable storage media provided by any embodiment of the present disclosure include permanent and non-permanent, removable and non-removable media, and information storage may be implemented by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

From the above description of the embodiments, it is clear to those skilled in the art that the embodiments of the present disclosure can be implemented by software plus necessary general hardware platform. Based on such understanding, the technical solutions of the embodiments of the present specification may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments of the present specification.

In the description of the present specification, reference to the description of "one embodiment/mode", "some embodiments/modes", "example", "specific example", or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment/mode or example is included in at least one embodiment/mode or example of the present application. In this specification, the schematic representations of the terms used above are not necessarily intended to be the same embodiment/mode or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments/modes or examples. Furthermore, the various embodiments/aspects or examples and features of the various embodiments/aspects or examples described in this specification can be combined and combined by one skilled in the art without conflicting therewith.

In addition, in the description of the present disclosure, "a plurality" means at least two, e.g., two, three, etc., unless explicitly specified otherwise. Meanwhile, in the description of the present disclosure, unless otherwise explicitly specified or limited, the terms "connected" and "connected" should be interpreted broadly, e.g., as being fixedly connected, detachably connected, or integrally connected; the connection can be mechanical connection or electrical connection; may be directly connected or indirectly connected through an intermediate. The specific meaning of the above terms in the present disclosure can be understood by those of ordinary skill in the art as appropriate.

It will be understood by those skilled in the art that the foregoing embodiments are merely for clarity of illustration of the disclosure and are not intended to limit the scope of the disclosure. Other variations or modifications may be made to those skilled in the art, based on the above disclosure, and still be within the scope of the present disclosure.

Claims

1. A model building method, characterized in that the model building method comprises:

acquiring label data and characteristic data; wherein the label data is data of one logging curve, and the characteristic data is data of at least one other logging curve; the data of any one logging curve is a data set of data of the same logging curve of a plurality of sample wells in the same block;

randomly mixing the label data and the feature data, and dividing the mixed data into a training set and a verification set according to a proportion;

performing sequence modeling based on a time convolution neural network according to the tag data and the characteristic data;

training the established model based on the training set, and verifying the validity of the model through the verification set;

and when the value of the loss function of the model meets a preset condition, saving the model.

2. The model building method according to claim 1, wherein before the feature data is acquired, the model building method further comprises:

acquiring characteristic source data, wherein the characteristic source data comprises data of at least one logging curve;

and respectively analyzing the correlation between the data of each logging curve included in the characteristic source data and the label data, and taking the data of which the correlation analysis result meets the preset condition as the characteristic data.

3. The model building method of claim 2, wherein the correlation analysis uses at least one of Pearson correlation coefficient, Kendall rank correlation coefficient, and Spearman rank correlation coefficient.

4. The model building method according to claim 3,

when the correlation analysis adopts Pearson correlation coefficients, the correlation analysis result of the feature data satisfies the following conditions: 0.5 < | ρ_xy| is less than or equal to 1, wherein rho_xyRepresenting Pearson correlation coefficients, wherein x is label data, and y is data of any one logging curve in the characteristic source data;

when the correlation analysis adopts Kendall rank correlation coefficients, the correlation analysis result of the feature data meets the following conditions: the absolute value of the Kendall rank correlation coefficient tau is equal to 1;

when the correlation analysis adopts a Spearman rank correlation coefficient, the correlation analysis result of the characteristic data meets the following conditions: the absolute value of the Spearman rank correlation coefficient δ is greater than or equal to 0.5.

5. The model building method according to any one of claims 1 to 4, wherein before performing the time-convolutional-neural-network-based sequence modeling based on the tag data and the feature data, the model building method further comprises: and carrying out multi-scale feature noise reduction processing on the tag data and the feature data.

6. The model building method according to claim 5, wherein the built model comprises an input layer, a plurality of hidden layers, a fully connected layer and an output layer which are connected in sequence; each hidden layer is formed by stacking a plurality of same residual modules, and each residual module comprises a convolution layer, a nonlinear mapping unit, a Dropout layer and a residual connecting unit which are sequentially connected.

7. The method according to claim 5, wherein the loss function is an average absolute error MAE, and the predetermined condition of the loss function is that the MAE value is less than 1 and does not decrease.

8. A model building apparatus, comprising a first processor and a first memory having stored therein computer program instructions adapted to be executed by the first processor, the computer program instructions when executed by the first processor performing the steps of the model building method according to any one of claims 1 to 7.

9. A well log construction method, characterized in that the method utilizes a model established by the model establishment method of any one of claims 1 to 7; the method comprises the following steps:

acquiring data of a known logging curve of a target well;

inputting the data of the known logging curve of the target well into the model, and constructing the unknown logging curve of the target well;

10. A well log construction apparatus comprising a second processor and a second memory, the second memory having stored therein computer program instructions adapted to be executed by the second processor, the computer program instructions when executed by the second processor performing the steps of the well log construction method of claim 9.