An intelligent construction method and device for unmeasured logging curves based on small layer information

Abstract

The present disclosure provides a training method and device for an unmeasured well log curve construction model based on small layer information, and an intelligent construction method and device for an unmeasured well log curve based on small layer information. Among them, the training method includes: obtaining training data, test data, and small layer division information; coupling the small layer division information as prior information with the training data and test data respectively to form a first training set and a first test set correspondingly; saving the training data and test data corresponding to each small layer into the data body corresponding to the small layer to form a second training set and a second test set correspondingly; training an LSTM model using the first training set and the second training set respectively to obtain a soft constraint LSTM model and a hard constraint LSTM model; inputting the first test set into the first sub-model, inputting the second test set into the second sub-model, comparing the calculation results of the target well log curves obtained by the first and second sub-models, and selecting the sub-model with higher construction accuracy as the well log curve construction model corresponding to the target well log curve.

Description

Technical Field

[0001] The present disclosure relates to the technical fields of oilfield development and well logging, and particularly relates to an intelligent construction method and device for unmeasured well logging curves based on sub-layer information. Background Art

[0002] Geophysical well logging technology is a very important exploration technology in geophysical exploration methods. There are many types of well logging methods, and each well logging curve contains rich information. By comprehensively using single or multiple well logging curves, rich geological data can be obtained, and then a series of geological problems can be solved. With the gradual deepening and increasing difficulty of oil and gas exploration, various well logging problems have been caused, including the problem of quickly and efficiently constructing unmeasured curves. Summary of the Invention

[0003] Embodiments of the present disclosure provide a training method and device for a construction model of unmeasured well logging curves based on sub-layer information, an intelligent construction method and device for unmeasured well logging curves based on sub-layer information, and a computer-readable storage medium, so as to realize the construction of unmeasured or unknown well logging curves in complex shale gas fields, with high prediction accuracy, simple method, and fast calculation speed.

[0004] On the one hand, a training method for a construction model of unmeasured well logging curves based on sub-layer information is provided. The target block has multiple sample wells, and each sample well corresponds to data of multiple well logging curves; the sample wells include multiple training wells and at least one test well. The training method for the construction model of unmeasured well logging curves based on sub-layer information includes:

[0005] Obtaining the well logging curve data corresponding to multiple training wells as training data, and obtaining the well logging curve data corresponding to at least one test well as test data;

[0006] Obtaining the sub-layer division information of multiple sub-layers in the reservoir depth direction of the target block;

[0007] Performing soft constraint processing on the training data and the test data respectively to form a first training set and a first test set correspondingly; the soft constraint processing is to couple the sub-layer division information as prior information with the data to be processed in a column expansion manner;

[0008] Performing hard constraint processing on the training data and the test data respectively to form a second training set and a second test set correspondingly; the hard constraint processing is to save the data to be processed corresponding to each sub-layer into the data body corresponding to the sub-layer;

[0009] Training an LSTM model with the first training set to obtain a soft constraint LSTM model; the soft constraint LSTM model includes at least a first sub-model corresponding to the type of the target well logging curve;

[0010] Train the LSTM model using the second training set to obtain a hard-constrained LSTM model; the hard-constrained LSTM model includes at least a second sub-model corresponding to the type of the target logging curve.

[0011] Input the first test set into the first sub-model, input the second test set into the second sub-model, compare the calculation results of the target logging curves obtained by the first sub-model and the second sub-model, and select the sub-model with higher construction accuracy as the logging curve construction model corresponding to the target logging curve.

[0012] In some embodiments, training the LSTM model using the first training set to obtain a soft-constrained LSTM model includes:

[0013] Use the logging curve data corresponding to the type of the target logging curve in the first training set as label data, use the other data in the first training set as feature data, input them into the LSTM model for training, and when the LSTM model meets the preset termination condition, obtain a first sub-model corresponding to the type of the target logging curve.

[0014] In some embodiments, the data of multiple logging curves corresponding to the sampling points at the same depth are a set of sampling point group data, and each small layer corresponds to multiple sampling point groups; saving the data to be processed corresponding to each small layer into the data body corresponding to the small layer includes:

[0015] Starting from the first sampling point group, traverse all sampling point groups, determine the small layer corresponding to the current sampling point group, and save the data of the current sampling point group into the data body corresponding to the small layer.

[0016] In some embodiments, training the LSTM model using the second training set to obtain a hard-constrained LSTM model includes:

[0017] Use the logging curve data corresponding to the type of the target logging curve in the second training set as label data, use the other data in the second training set as feature data, input them into the LSTM model for training, and when the LSTM model meets the preset termination condition, obtain a second sub-model corresponding to the type of the target logging curve.

[0018] In some embodiments, the preset termination condition for model training is that the change of the mean absolute percentage error (MAPE) value with the increase of the training iteration times is less than 10%.

[0019] In some embodiments, comparing the calculation results of the target logging curves obtained by the first sub-model and the second sub-model, and selecting the sub-model with higher construction accuracy as the logging curve construction model corresponding to the target logging curve includes:

[0020] Calculate the mean absolute percentage error (MAPE) values corresponding to the first sub-model and the second sub-model, and select the model with the lower MAPE value as the logging curve construction model corresponding to the target logging curve.

[0021] On the other hand, a logging curve construction model training device is provided. The device includes a processor and a memory. The memory stores computer program instructions suitable for execution by the processor. When the computer program instructions are run by the processor, the steps in any of the above-mentioned logging curve construction model training methods based on sub-layer information are executed.

[0022] On another aspect, an intelligent construction method for unmeasured logging curves based on sub-layer information is provided. The method includes:

[0023] Obtain the logging curve data of the target well in the target block;

[0024] According to the type of the logging curve construction model, perform soft constraint processing or hard constraint processing on the logging curve data of the target well, and input the processed data into the logging curve construction model to obtain the target logging curve corresponding to the target well;

[0025] Wherein, the logging curve construction model is the logging curve construction model in any of the above embodiments.

[0026] On another aspect, a logging curve construction device is provided. The device includes a processor and a memory. The memory stores computer program instructions suitable for execution by the processor. When the computer program instructions are run by the processor, the steps in any of the above-mentioned intelligent construction methods for unmeasured logging curves based on sub-layer information are executed.

[0027] On another aspect, a computer-readable storage medium is provided. The storage medium stores computer program instructions. When the computer program instructions are executed by the processor of the user device, the user device is caused to execute any of the above-mentioned logging curve construction model training methods based on sub-layer information, and / or any of the above-mentioned intelligent construction methods for unmeasured logging curves based on sub-layer information.

[0028] The method for training a model for constructing an unmeasured logging curve based on small layer information provided by some embodiments of the present disclosure is as follows. Based on small layer information, a first training set is formed by performing soft constraint processing on training data, and an LSTM model is trained using the first training set to obtain a soft constraint LSTM model. A second training set is formed by performing hard constraint processing on training data, and an LSTM model is trained using the second training set to obtain a hard constraint LSTM model. Among them, the characteristic of soft constraint processing is that the small layer information only has an impact by being added to the logging curve dataset. At this time, the nature of the small layer information is the same as that of other logging curves, and the machine learning model can only find the mapping relationship between the small layer information and other logging curves through training. Hard constraint processing is to put the data of the same small layer together and find the mapping relationship between the data through machine learning. And the formation properties reflected by the data of the same small layer are the same or highly similar. Therefore, the data mapping relationship of hard constraint is more obvious, and the mapping relationship fully conforms to this small layer. The difference between soft constraint and hard constraint lies in whether the small layer information is added to the dataset or the data is classified according to the small layer information. The data dimension of the soft constraint dataset is M×(N + 1), and the data dimension of the hard constraint dataset is M×N. The method for training a model for constructing an unmeasured logging curve based on small layer information provided by some embodiments of the present disclosure takes into account the advantages of both soft constraint and hard constraint. By comparing the construction results of the soft constraint LSTM model and the hard constraint LSTM model for a target logging curve, the model with higher result accuracy is selected as the logging curve construction model corresponding to the final target logging curve. When the logging curve construction model is applied to predict the target logging curve of a target well in the same block, the target logging curve can be effectively constructed, and its prediction accuracy is high, the method is simple, and the calculation speed is fast. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, are used to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are included in this specification and form a part of this specification.

[0030] Figure 1 It is a flowchart of a method for training a model for constructing an unmeasured logging curve based on small layer information according to some embodiments;

[0031] Figure 2 It is a flowchart of a method for intelligently constructing an unmeasured logging curve based on small layer information according to some embodiments;

[0032] Figure 3 It is a comparison chart of the DEN curve predicted by the first sub-model of a method for intelligently constructing an unmeasured logging curve based on small layer information according to some embodiments and the true value;

[0033] Figure 4Comparison chart of the DEN curve predicted by the second sub-model of an intelligent construction method for unmeasured logging curves based on small layer information and the true value according to some embodiments. Detailed implementation manners

[0034] The present disclosure will be further described in detail below with reference to the accompanying drawings and implementation manners. It can be understood that the specific implementation manners described herein are only used to explain the relevant content and do not limit the present disclosure. Additionally, it should be noted that for the convenience of description, only parts related to the present disclosure are shown in the drawings.

[0035] It should be noted that, without conflict, the implementation manners and features in the implementation manners in the present disclosure can be combined with each other. The present disclosure will be described in detail below with reference to the drawings and in combination with the implementation manners.

[0036] It should be noted that the step numbers in the text are only for the convenience of explaining specific embodiments and do not serve to limit the execution order of the steps.

[0037] The method provided in this embodiment can be executed by a related server, and hereinafter, the server is taken as an example of the execution entity for description. Among them, the execution entity can be adjusted according to specific cases, such as electronic devices, computers, etc.

[0038] As described in the background art, with the gradual deepening of oil and gas exploration, the limitations of traditional methods for constructing unmeasured curves are becoming prominent, and they are unable to handle the complex non-linear mapping relationship of logging data.

[0039] In recent years, artificial intelligence has developed rapidly. Neural networks have the characteristics of self-organization and self-learning in processing information and are suitable for processing such complex mapping relationships as logging data. In particular, the long short-term memory neural network model can "remember" historical information and is more suitable for the construction of logging sequence data. However, traditional neural network methods do not consider reservoir information and lack the support of real geological data, and ultimately the prediction accuracy cannot fully meet the requirements.

[0040] At present, there are many methods for constructing unmeasured curves, mainly including the formula method and the artificial intelligence method. However, with the increasing complexity of geological problems, the limitations of traditional empirical formula methods are becoming prominent; in addition, most artificial intelligence methods focus on exploring the relationships between different logging curves and do not consider the small layer division information. And the small layer division is crucial for reservoir description. It can further provide geological conditions, making it easier to analyze the characteristics of the reservoir and guide oilfield development, which is of great significance for the exploitation of oil and gas fields. Therefore, it is very necessary to apply the small layer division information to the intelligent construction of unmeasured logging curves.

[0041] Based on this, embodiments of the present disclosure provide a method for constructing an unmeasured logging curve based on small layer information, which is applied to the construction of unmeasured or unknown logging curves in complex shale gas fields. The unmeasured or unknown logging curve can be completely missing or partially missing. It has high prediction accuracy, simple method, and fast calculation speed.

[0042] The application scenario of the present disclosure is introduced below. Since the geological conditions of different blocks are usually quite different, the method in the present disclosure is applicable to predicting and constructing unmeasured or unknown logging curves (target logging curves) of target wells in the same block. That is, the LSTM model is trained with the known logging curve data corresponding to multiple sample wells in the same block to obtain a logging curve construction model for predicting the target logging curve, and this logging curve construction model can be used to construct the target logging curve of the target well in the same block.

[0043] Here, each target well or sample well corresponds to data of multiple logging curves, where the target logging curve of the target well is an unknown logging curve, and the target logging curve of the sample well is a known logging curve. Other types of logging curves corresponding to the target well and the sample well are all known logging curves. The data of each type of logging curve is data collected at multiple sampling points distributed in the reservoir depth direction of the target block, and arranging the data of multiple sampling points according to depth can form a logging curve.

[0044] Before model training, multiple sample wells can be divided into training wells and test wells. That is, the sample wells include multiple training wells and at least one test well.

[0045] The target block is divided into multiple small layers in the reservoir depth direction. Here, each small layer can correspond to multiple sampling points.

[0046] As Figure 1 shown, some embodiments of the present disclosure provide a method for training an unmeasured logging curve construction model based on small layer information. The model training method includes:

[0047] S100, obtaining the logging curve data corresponding to multiple training wells as training data, and obtaining the logging curve data corresponding to at least one test well as test data.

[0048] S200, obtaining the small layer division information of multiple small layers in the reservoir depth direction of the target block.

[0049] Exemplarily, for S100 and S200, well logging curve data Data_log of multiple training wells and formation division information Data_zone corresponding to different training wells can be obtained. Among them, the Data_log data includes multiple well logging curves, and the multiple well logging curves are, for example, CAL curve, DEN curve, AC curve, GR curve, etc. Data_log can be expressed as Data_log = {log_CAL, log_DEN, log_AC, log_GR, …}, with a dimension of M×N, where M represents the sum of the sampling points of all wells, and N represents the number of well logging curve types. Thus, training data can be obtained. The process of obtaining test data is similar and will not be elaborated here.

[0050] The dimension of the Data_zone data is M×1, and M is also the sum of the sampling points of all wells. Here, the i-th formation can be marked as zone_i.

[0051] S300, perform soft constraint processing on the training data and test data respectively, and correspondingly form a first training set and a first test set.

[0052] The soft constraint processing is to use the formation division information as prior information and couple it with the data to be processed in a column expansion manner. Here, the data to be processed is training data or test data, which can be determined according to the actual data to be processed with soft constraints.

[0053] Taking the formation of the first training set as an example, the formation division information Data_zone is used as prior information and coupled with the well logging curve data Data_log into the input data in a column expansion manner. At this time, the first training set is Data_log_zone = {Data_log, Data_zone}, and the first training set is used as the input data for subsequent model training, with a dimension size of M×(N + 1).

[0054] Here, since the formation division data acts on the model in the form of input data and the constraint effect is not strong, the present disclosure calls it soft constraint.

[0055] S400, perform hard constraint processing on the training data and test data respectively, and correspondingly form a second training set and a second test set.

[0056] The hard constraint processing is to save the data to be processed corresponding to each formation into the data body corresponding to the formation. Here, the data to be processed is training data or test data, which can be determined according to the actual data to be processed with soft constraints.

[0057] The data of multiple well logging curves corresponding to the sampling points at the same depth is a set of sampling point group data, and each formation corresponds to multiple sampling point groups. Optionally, in the hard constraint processing, saving the data to be processed corresponding to each formation into the data body corresponding to the formation includes:

[0058] Starting from the first sampling point group, traverse all sampling point groups, determine the corresponding sub-layer of the current sampling point group, and save the data of the current sampling point group into the data body corresponding to the sub-layer.

[0059] Taking the formation of the first test set as an example, filter the corresponding logging curve data according to the sub-layer division information Data_zone and save it into the data body corresponding to the sub-layer. The number of data bodies corresponds to the total number of sub-layers, with a total of n_zone, and the i-th sub-layer is marked as zone_i.

[0060] The dimensionality sizes of each data body are different, which are M i ×N, M i indicating the number of sampling point groups of each data body, satisfying

[0061] A sampling point group refers to a combination of different logging curves at the same depth. For example, at a depth of 2000m, the data of N curves such as CAL, CNL, DEN, GR, and AC at the sampling point of the 2000m depth jointly form a sampling point group. Therefore, the dimensionality size of each sampling point group is 1×N. Each sub-layer zone_i corresponds to M i sampling point groups. For example, the second sub-layer zone_2 has a total of M i = 3000 sampling point groups, and the data dimensionality of each sampling point group is 1×5, where 5 indicates that there are 5 different logging curves. The M corresponding to different zone_i may be the same or different, and N is the same.

[0062] Starting from the first sampling point group, traverse all sampling point groups of the logging curves, determine the corresponding sub-layer of the current sampling point group, and save the data of the current sampling point group into the data body corresponding to the sub-layer. The dimensionality size of each sampling point group is 1×N. If the sub-layer corresponding to the sampling point group is the zone_i layer, then save the data of this group into the Data_zone_i data body.

[0063] Since each sampling point group is divided into different data bodies according to the actual corresponding sub-layer information, the logging curve data is strictly classified according to the sub-layer information, and the constraint effect is strong. Therefore, this disclosure calls it a hard constraint.

[0064] S500, train the LSTM model using the first training set to obtain a soft-constrained LSTM model. The soft-constrained LSTM model includes at least a first sub-model corresponding to the type of the target logging curve.

[0065] The S500 specifically includes: taking the well logging curve data corresponding to the type of the target well logging curve in the first training set as label data, taking the other data in the first training set as feature data, inputting them into the LSTM model for training, and when the LSTM model meets the preset termination condition, obtaining a first sub-model corresponding to the type of the target well logging curve.

[0066] Exemplarily, the dataset Data_log_zone is divided into two parts: the feature data and the label data required by the neural network model framework. The label data is the data of the target well logging curve, and the feature data is the data of other well logging curves and the sub-layer information of the corresponding well. Assuming that the target prediction curve is the neutron curve (CNL), the feature data is Feature = {Data_log after removing CNL, Data_zone}, and the label data is Label = {CNL}. The LSTM model adopts a supervised learning method, inputs the Feature and Label data into the LSTM model for training, and saves it as the Soft_LSTM_model when the model meets the termination condition.

[0067] Here, in the case where it is known which well logging curve the trained soft constraint model will be used to predict and construct, the type of the target well logging curve can be the same as the type of the well logging curve to be constructed; in the case where the type of the well logging curve to be constructed is unknown, any well logging curve can be used as the target well logging curve. For example, the well logging curves corresponding to the sample well are CAL, CNL, DEN, GR, and AC curves. CAL can be used as the target well logging curve, that is, the label data; the well logging curve data of CNL, DEN, GR, and AC and the sub-layer division information in the first training set are used as feature data, input into the LSTM model for training, and when the LSTM model meets the preset termination condition, the obtained first sub-model is the soft constraint CAL sub-model. Another example is to use CNL as the target well logging curve, that is, the label data; the well logging curve data of CAL, DEN, GR, and AC and the sub-layer division information in the first training set are used as feature data, input into the LSTM model for training, and when the LSTM model meets the preset termination condition, the obtained first sub-model is the soft constraint CNL sub-model.

[0068] As a possible implementation, during the process of training to obtain the soft constraint LSTM model, the model can be trained separately for multiple well logging curves corresponding to the sample well to obtain multiple corresponding first sub-models, so as to make full use of the well logging curve data of the sample well and obtain as many soft constraint sub-models of different types of well logging curves as possible when it is not determined which well logging curve will be predicted, thus facilitating the subsequent prediction and construction of different types of well logging curves.

[0069] Specifically, the types of target prediction curves can be used as the first sub-model discrimination criteria, and corresponding first sub-models are constructed for different target curves respectively. That is, parameter settings and model training are performed on different sub-models respectively. When the sub-model meets the end condition, it is saved as Soft_LSTM_model. Therefore, Soft_LSTM_model = {Soft_LSTM_model_CNL, Soft_LSTM_model_DEN, …}.

[0070] S600, train the LSTM model using the second training set to obtain a hard-constrained LSTM model.

[0071] The hard-constrained LSTM model includes at least a second sub-model corresponding to the type of target logging curve.

[0072] S600 specifically includes:

[0073] Use the logging curve data corresponding to the type of target logging curve in the second training set as label data, and use the other data in the second training set as feature data, and input them into the LSTM model for training. When the LSTM model meets the preset termination condition, a second sub-model corresponding to the type of target logging curve is obtained.

[0074] Exemplarily, the dataset Data_zone_i is divided into two parts: feature data and label data required by the neural network model framework. The label data is the target logging curve data, and the feature data is the other data. The LSTM model is also used for learning, and when the model meets the termination condition, it is saved as Hard_LSTM_model.

[0075] Here, when it is known which logging curve the trained hard-constrained model will be used to predict and construct, the type of target logging curve can be the same as the type of logging curve to be constructed; when the type of logging curve to be constructed is unknown, any logging curve can be used as the target logging curve. For example, the logging curves corresponding to the sample well are CAL, CNL, DEN, GR, and AC curves. CAL can be used as the target logging curve, that is, the label data; the CNL, DEN, GR, and AC logging curve data and the small layer division information in the second training set are used for training as feature data, and input into the LSTM model for training. When the LSTM model meets the preset termination condition, the obtained second sub-model is the hard-constrained CAL sub-model. Another example is to use CNL as the target logging curve, that is, the label data; the CAL, DEN, GR, and AC logging curve data and the small layer division information in the second training set are used for training as feature data, and input into the LSTM model for training. When the LSTM model meets the preset termination condition, the obtained second sub-model is the soft-constrained CNL sub-model.

[0076] As a possible implementation, during the process of training the hard-constrained LSTM model, various logging curves corresponding to the sample wells can be separately used for model training to obtain a corresponding number of second sub-models, so as to make full use of the logging curve data of the sample wells and obtain as many hard-constrained sub-models of different types of logging curves as possible when it is not determined which logging curve will be predicted, thereby facilitating subsequent prediction construction for different types of logging curves.

[0077] Specifically, the type of the target prediction curve can be used as the second sub-model discrimination criterion, and corresponding second sub-models can be constructed for different target curves respectively, that is, parameter settings and model training are carried out for different sub-models respectively, and when the sub-model meets the end condition, it is saved as Hard_LSTM_model. Hard_LSTM_model = {Hard_LSTM_model_CNL, Hard_LSTM_model_DEN,...}.

[0078] In some embodiments, the basic parameter settings of the soft-constrained LSTM model or the hard-constrained LSTM model are as follows: the multi-layer neural network is set to three main parts: the input layer, the intermediate hidden layer, and the output layer. The specific data feature scale of each layer can be selected by the Bayesian hyperparameter tuning method, and the Dropout technique is introduced into the hidden layer to prevent overfitting.

[0079] Since there are many hyperparameter combinations in machine learning algorithms and the accuracies of machine learning algorithms under different combinations are different, usually two methods, manual hyperparameter tuning and code hyperparameter tuning, can be adopted. Manual hyperparameter tuning is time-consuming and laborious and highly dependent on experience, so the code automatic hyperparameter tuning method is adopted. Code hyperparameter tuning mainly has three categories. One is grid search hyperparameter tuning, one is random search hyperparameter tuning, and the other is Bayesian hyperparameter tuning method.

[0080] The grid search method is an exhaustive search method for specifying parameter values. The possible values of each parameter are arranged and combined, and all combination results form a "grid". Then, the machine learning model is used to learn the situations under all combinations, and finally the best set of hyperparameters is selected. This method is time-consuming, and when there are many hyperparameters, it faces the curse of dimensionality; the random search hyperparameter tuning method is similar to the grid search hyperparameter tuning method. Compared with the grid search method, it can reduce time consumption, but each search runs independently and cannot make full use of the prior knowledge of previous hyperparameter combinations. The Bayesian hyperparameter tuning method can solve this problem. The main idea of this method is that given the input and output of the machine learning model, the posterior distribution of the objective function is updated by continuously adding sample points, that is, a Gaussian process is performed until the distribution basically fits the true distribution, and the prior knowledge of previous hyperparameter combinations can be considered.

[0081] The specific steps are as follows: First, select the objective value to be minimized. In the LSTM deep neural network, the objective function is the loss of the network model on the validation set using a set of hyperparameters.

[0082] Then, define the value distribution for each hyperparameter. Each time a hyperparameter is searched, the model is trained once.

[0083] Finally, construct a surrogate function and select a method for evaluating the next set of hyperparameter configurations. This part includes two cores, one is the prior function (PF), and the other is the acquisition function (AC). The prior function PF mainly uses the Gaussian process (GP), and the acquisition function AC mainly uses the EI (Expected improvement) method.

[0084] Assume the dataset is Data = {(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x n ,y n )}. If we want to predict y n+1 at x n+1 , the Gaussian process will consider the (n + 1)-th observation point as a sample from a certain (n + 1)-dimensional Gaussian distribution, that is:

[0085]

[0086] where G represents the Gaussian function,

[0087] k = [(x n+1 ,x 1 ),(x n+1 ,x 2 ),...,(x n+1 ,x n )],

[0088] k(x i ,x j ) is the covariance function, and its calculation formula is:

[0089]

[0090] where θ is the smoothness of the model, and the optional values are 0.1, 0.2, and 0.5.

[0091] Due to the different influence degrees of the characteristic curves on the target log curves, the following covariance function is selected:

[0092]

[0093] Among them, diag(θ) is a diagonal matrix.

[0094] Derived from the above formula, the distribution of f n+1 is as follows:

[0095] P(f n+1 | Data, x n+1 ) = G(u(x n+1 ), v 2 (x n+1 ))

[0096] Among them, u(x n+1 ) = kK -1 f 1:n , v 2 (x n+1 ) = k(x n+1 , x n+1 ) - kK -1 k T .

[0097] u(x n+1 ) is the predicted value at x n+1 , and v 2 (x n+1 ) is the noise of the predicted value at x n+1 .

[0098] The calculation method of EI is as follows:

[0099]

[0100]

[0101] Among them, Φ(·) and φ(·) represent the standard normal cumulative distribution and the standard normal distribution respectively, and x max represents the position of the point that maximizes the objective function value among the parameter combinations that have been used.

[0102] When the objective function to be optimized reaches the termination condition, the results of the hyperparameters and the validation set loss are stored.

[0103] Compared with the grid search and random search hyperparameter tuning methods, Bayesian hyperparameter tuning uses a Gaussian process, can consider prior information, has fewer iteration times, a faster update speed, and no dimensional explosion. In addition, this method has strong stability and is more suitable for hyperparameter tuning in machine learning.

[0104] The Dropout technique is introduced into all hidden layers to prevent overfitting, and the Dropout ratio is set to 0.3.

[0105] In some embodiments, the preset termination condition for the model training of the soft-constrained LSTM model or the hard-constrained LSTM model is that the change in the mean absolute percentage error (MAPE) value with the increase in the number of training iterations is less than 10%.

[0106] The MAPE calculation method is as follows:

[0107]

[0108] where y i is the true measured data value, is the data value predicted by the model, and n is the number of sample point groups.

[0109] S700, input the first test set into the first sub-model, input the second test set into the second sub-model, compare the calculation results of the target logging curves obtained by the first sub-model and the second sub-model, and select the sub-model with higher construction accuracy as the logging curve construction model corresponding to the target logging curve.

[0110] In some embodiments, comparing the calculation results of the target logging curves obtained by the first sub-model and the second sub-model, and selecting the sub-model with higher construction accuracy as the logging curve construction model corresponding to the target logging curve includes:

[0111] Calculate the mean absolute percentage error (MAPE) values corresponding to the first sub-model and the second sub-model, and select the model with the lower mean absolute percentage error (MAPE) value as the logging curve construction model corresponding to the target logging curve. That is, the model with the lower MAPE value has a better prediction effect, and its prediction result can be used as the final curve construction result of the test well.

[0112] The method for training a model for constructing an unmeasured logging curve based on small layer information provided by some embodiments of the present disclosure, based on the small layer information, forms a first training set by performing soft constraint processing on the training data, and trains an LSTM model using the first training set to obtain a soft constraint LSTM model; forms a second training set by performing hard constraint processing on the training data, and trains an LSTM model using the second training set to obtain a hard constraint LSTM model. Among them, the characteristic of the soft constraint processing is that the small layer information only has an impact by being added to the logging curve dataset. At this time, the nature of the small layer information is the same as that of other logging curves, and the machine learning model can only find the mapping relationship between the small layer information and other logging curves through training. The hard constraint processing is to put the data of the same small layer together and find the mapping relationship between the data through machine learning. And the formation properties reflected by the data of the same small layer are the same or highly similar. Therefore, the data mapping relationship of the hard constraint is more obvious, and the mapping relationship completely conforms to this small layer. The difference between the soft constraint and the hard constraint is whether the small layer information is added to the dataset or the data is classified according to the small layer information. The data dimension of the soft constraint dataset is M×(N + 1), and the data dimension of the hard constraint dataset is M×N. The method for training a model for constructing an unmeasured logging curve based on small layer information provided by some embodiments of the present disclosure takes into account the advantages of both soft constraint and hard constraint. By comparing the construction results of the soft constraint LSTM model and the hard constraint LSTM model for a target logging curve, the model with higher result accuracy is selected as the logging curve construction model corresponding to the final target logging curve. When the logging curve construction model is applied to predict the target logging curve of the target well in the same block, the target logging curve can be effectively constructed, and its prediction accuracy is high, the method is simple, and the calculation speed is fast.

[0113] Some embodiments of the present disclosure also provide an intelligent method for constructing an unmeasured logging curve based on small layer information. As Figure 2 shown, the intelligent method for constructing an unmeasured logging curve based on small layer information includes:

[0114] S1, obtaining the logging curve data of the target well in the target block;

[0115] S2, according to the type of the logging curve construction model, performing soft constraint processing or hard constraint processing on the logging curve data of the target well, and inputting the processed data into the logging curve construction model to obtain the target logging curve corresponding to the target well;

[0116] wherein, the logging curve construction model is the logging curve construction model of any of the above embodiments.

[0117] It can be understood that when the well logging curve construction model selected by the model training method for constructing unmeasured well logging curves based on small layer information is a soft constraint LSTM model, the well logging curve data of the target well is subjected to soft constraint processing; when the well logging curve construction model selected by the model training method for constructing unmeasured well logging curves based on small layer information is a hard constraint LSTM model, the well logging curve data of the target well is subjected to hard constraint processing.

[0118] The intelligent construction method for unmeasured well logging curves based on small layer information provided by some embodiments of the present disclosure has the same beneficial effects as the model training method for constructing unmeasured well logging curves based on small layer information provided by some embodiments of the present disclosure, and will not be elaborated here.

[0119] The following further illustrates the intelligent construction method for unmeasured well logging curves based on small layer information and the model training method for constructing unmeasured well logging curves based on small layer information provided by some embodiments of the present disclosure by taking the well logging curves of 12 horizontal wells in a certain block as an example.

[0120] First, 10 horizontal wells (well numbers marked as W-1, W-2, …, W-10) are selected from 12 horizontal well sample wells as training wells. Each horizontal well corresponds to five well logging curves, namely the caliper CAL curve, neutron CNL curve, density DEN curve, acoustic travel time AC curve, and natural gamma GR curve. The number of sampling point groups is respectively number={15248, 12000, 11744, 15152, 15368, 11656, 16832, 15544, 11504, 13336}. The same type of well logging curves of all wells are subjected to row expansion operation, that is, data concatenation. After concatenation, there are a total of 138384 groups of sampling points. The data body is denoted as Data_log={log_CAL, log_DEN, log_AC, log_GR, log_CNL}, which is also the training data, and the data dimension is 138384×5. At the same time, the small layer division information corresponding to these 10 wells is obtained. After row expansion of the small layer division information, Data_zone is obtained, and the data dimension is 138384×1. The Data_log and Data_zone data together constitute the data set required for training the neural network model.

[0121] The small layer division information Data_zone is coupled with the training data Data_log as prior information in a column expansion manner to obtain the first training set Data_log_zone={Data_log, Data_zone}. The dimension size of the input data is 138384×6, and the Data_log_zone data is the data set required for the soft constraint model.

[0122] On the other hand, prepare a training dataset for the hard constraint model, that is, screen the corresponding logging curve data according to the small layer division information Data_zone and save it into the data volume corresponding to each small layer. There are 8 small layers in the reservoir of this block, that is, n_zone = 8, and each small layer is respectively marked as zone_1, zone_2, zone_3, zone_4, zone_5, zone_6, zone_7, and zone_8. The dimensional sizes of each data volume are different, but the total number of sampling points of the logging curves for each well is the corresponding total number in number. Specifically, starting from the first sampling point group, traverse all sampling point groups of the logging curves and make judgments. The dimensional size of each sampling point group is 1×5. If the small layer corresponding to this sampling point group is the zone_1 layer, then save this group of data into the Data_zone_1 data volume, and so on, and the corresponding data for other small layers can be obtained.

[0123] Construct soft constraint and hard constraint LSTM models respectively.

[0124] Divide the first training set, that is, the dataset Data_log_zone, into two parts: the feature data and the label data required by the neural network model framework. Among them, the label data is the target logging curve. In this embodiment, the data of the DEN logging curve is the label data, that is, the label data is Label = {DEN}. The feature data Feature is CNL, GR, CAL, AC, and the small layer information. Then input the Feature data and the Label data into the LSTM model for training, and adjust the parameters by the Bayesian method. For the density curve DEN, the optimal parameters of the LSTM model are: the feature scale of the input layer is set to 100, and the feature scales of the subsequent layers are set to 80, 60, 40, 20, 10 in sequence. Dropout layers are introduced in the first three layers, the ratio is 0.3, the keep_prob parameter value is 0.7, and the output layer is composed of a Softmax network layer.

[0125] Construct a hard constraint LSTM model in a similar way. Specifically, divide the second training set, that is, the dataset Data_zone_i, into two parts: the feature data and the label data required by the neural network model framework. Among them, the label data is the data of the target logging curve DEN, and the feature data is the other logging curves CNL, GR, CAL, and AC. Also use the LSTM model for learning.

[0126] When the training of the hard-constrained LSTM model and the soft-constrained LSTM model meets the termination condition, that is, the change in the mean absolute percentage error (MAPE) value with the increase in the number of training iterations is less than 10%, the model training ends. In this embodiment, the model training ends when the MAPE is 7%. At this time, the soft-constrained LSTM model and the hard-constrained LSTM model are respectively saved as the first sub-model Soft_LSTM_model_DEN and the second sub-model Hard_LSTM_model_DEN.

[0127] Similarly, according to the above steps, the first sub-model Soft_LSTM_model_CNL, the first sub-model Soft_LSTM_model_GR, the first sub-model Soft_LSTM_model_CAL, and the first sub-model Soft_LSTM_model_CNL can also be obtained.

[0128] Select well W-11 and well W-12 in the 12-well sample wells as the blind well test data bodies, that is, the test wells. Perform soft-constraint processing and hard-constraint processing on the data of well W-11 and well W-12 respectively, and select the corresponding Soft_LSTM_model sub-model and Hard_LSTM_model sub-model to construct the DEN logging curves of well W-11 and well W-12. Here, the first sub-model Soft_LSTM_model_DEN and the second sub-model Hard_LSTM_model_DEN are selected.

[0129] For the prediction results of the DEN logging curves of well W-11 using the first sub-model Soft_LSTM_model_DEN and the second sub-model Hard_LSTM_model_DEN, please refer to Figure 3 and Figure 4 .

[0130] Calculate the construction accuracy MAPE of the first sub-model Soft_LSTM_model_DEN and the second sub-model Hard_LSTM_model_DEN for well W-11 and well W-12 respectively and compare them. The MAPE value of Soft_LSTM_model_DEN is 11%, and the MAPE value of Hard_LSTM_model_DEN is 7%. Therefore, the prediction results of Hard_LSTM_model_DEN are selected as the final curve construction results for well W-11 and well W-12.

[0131] There is a target well W-13 in this block. Its DEN is an unknown logging curve, and the data of other logging curves CNL, GR, CAL, and AC are known.

[0132] According to the calculation results of the above wells W-11 and W-12, when it is necessary to construct the DEN logging curve of the target well W-13, the data of the logging curves CNL, GR, CAL, and AC of the well W-13 can be subjected to hard constraint processing, and the processed data is input into the Hard_LSTM_model_DEN model optimized through the calculation results of the wells W-11 and W-12, and then the target logging curve, i.e., the DEN logging curve, corresponding to the target well W-13 can be obtained.

[0133] The embodiment of the present disclosure further provides a device for training a logging curve construction model. The device includes a processor and a memory. Among them, computer program instructions suitable for being executed by the processor are stored in the memory. When the computer program instructions are run by the processor, the processor executes the method for training a construction model of an unmeasured logging curve based on small layer information provided in any of the above embodiments.

[0134] The embodiment of the present disclosure further provides a device for constructing a logging curve. The device includes a processor and a memory. Among them, computer program instructions suitable for being executed by the processor are stored in the memory. When the computer program instructions are run by the processor, the processor executes the intelligent construction method of an unmeasured logging curve based on small layer information provided in any of the above embodiments.

[0135] The embodiment of the present disclosure further provides a computer-readable storage medium. Computer program instructions are stored in the storage medium. When the computer program instructions are executed by the processor of the user device, the user device is enabled to execute the method for training a construction model of an unmeasured logging curve based on small layer information disclosed in any of the above embodiments, and / or the intelligent construction method of an unmeasured logging curve based on small layer information.

[0136] The computer-readable storage medium provided in any embodiment of the present disclosure includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of the computer storage medium 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 technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0137] An embodiment of the present disclosure also provides an electronic device, including a processor and a memory. Computer program instructions suitable for execution by the processor are stored in the memory. When the computer program instructions are run by the processor, the methods disclosed in any of the above embodiments are executed.

[0138] The electronic device provided in any embodiment of the present disclosure may be a mobile phone, a computer, a tablet computer, a server, a network device, etc., or may also be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disc, etc.

[0139] For example, the electronic device may include: a processor, a memory, an input / output interface, a communication interface, and a bus. Among them, the processor, the memory, the input / output interface, and the communication interface are communicatively connected to each other inside the device through the bus.

[0140] The processor may be implemented in a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0141] The memory may be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory may store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory and are called and executed by the processor.

[0142] The input / output interface is used to connect to an input / output module to implement information input and output. The input / output module may be configured as a component in the device or externally connected to the device to provide corresponding functions. Among them, the input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0143] The communication interface is used to connect to a communication module to implement communication interaction between this device and other devices. Among them, the communication module may implement communication in a wired manner (such as USB, network cable, etc.) or in a wireless manner (such as mobile network, WIFI, Bluetooth, etc.).

[0144] The bus includes a path for transmitting information between various components of the device, such as a processor, a memory, an input / output interface, and a communication interface.

[0145] It should be noted that although the above device only shows a processor, a memory, an input / output interface, a communication interface, and a bus, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device may also only include the components necessary for implementing the solution of the embodiments of this specification, and does not necessarily include all the components described above.

[0146] From the description of the above embodiments, those skilled in the art can clearly understand that the embodiments of this specification can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the embodiments of this specification, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this specification.

[0147] The method or device illustrated in the above embodiments can be specifically implemented by a computer chip or an entity, or by a product with a certain function. A typical implementation device is a computer, and the specific form of the computer can be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver device, a game console, a tablet computer, a wearable device, or a combination of any several of these devices.

[0148] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. The method embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated. When implementing the solution of the embodiments of this specification, the functions of each module can be implemented in the same or multiple software and / or hardware. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0149] In the description of this specification, the descriptions referring to terms such as "one embodiment / way", "some embodiments / ways", "example", "specific example", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment / way or example are included in at least one embodiment / way or example of this application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment / way or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments / ways or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments / ways or examples described in this specification and the features of different embodiments / ways or examples.

[0150] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of such features. In the description of this disclosure, "a plurality" means at least two, such as two, three, etc., unless otherwise specifically and clearly defined. "And / or" is only used to describe the association relationship of associated objects, indicating three relationships. For example, A and / or B means: A exists alone, A and B exist simultaneously, and B exists alone these three situations. At the same time, in the description of this disclosure, unless otherwise clearly stipulated and limited, the terms "connected" and "connected to" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in this disclosure can be understood according to specific circumstances.

[0151] Those skilled in the art should understand that the above embodiments are only for clearly explaining this disclosure and not for limiting the scope of this disclosure. For those skilled in the art, other changes or modifications can be made based on the above disclosure, and these changes or modifications are still within the scope of this disclosure.

Claims

1. A method for training a model for constructing an unmeasured logging curve based on small layer information, characterized in that, the target block has multiple sample wells, and each sample well corresponds to data of multiple logging curves; the sample wells include multiple training wells and at least one test well; the method includes: Obtaining the logging curve data corresponding to multiple training wells as training data, and obtaining the logging curve data corresponding to the at least one test well as test data; Obtaining the small layer division information of multiple small layers in the reservoir depth direction of the target block; Performing soft constraint processing on the training data and the test data respectively, and correspondingly forming a first training set and a first test set; the soft constraint processing is to use the small layer division information as prior information and couple it with the data to be processed in a column expansion manner; Performing hard constraint processing on the training data and the test data respectively, and correspondingly forming a second training set and a second test set; the hard constraint processing is to save the data to be processed corresponding to each small layer into the data body corresponding to the small layer; Training an LSTM model with the first training set to obtain a soft constraint LSTM model; the soft constraint LSTM model includes at least a first sub-model corresponding to the type of the target logging curve; Training an LSTM model with the second training set to obtain a hard constraint LSTM model; the hard constraint LSTM model includes at least a second sub-model corresponding to the type of the target logging curve; Inputting the first test set into the first sub-model, inputting the second test set into the second sub-model, comparing the calculation results of the target logging curves obtained by the first sub-model and the second sub-model, and selecting the sub-model with higher construction accuracy as the logging curve construction model corresponding to the target logging curve.

2. The method for training a model for constructing an unmeasured logging curve based on small layer information according to claim 1, characterized in that, the training the LSTM model with the first training set to obtain a soft constraint LSTM model includes: Using the logging curve data corresponding to the type of the target logging curve in the first training set as label data, and using the other data in the first training set as feature data, and inputting them into the LSTM model for training. When the LSTM model meets the preset termination condition, a first sub-model corresponding to the type of the target logging curve is obtained.

3. The method for training a model for constructing an unmeasured logging curve based on small layer information according to claim 1, characterized in that, The data of multiple logging curves corresponding to the sampling points at the same depth is a set of sampling point group data, and each small layer corresponds to multiple sampling point groups; The saving the data to be processed corresponding to each small layer into the data body corresponding to the small layer includes: Starting from the first sampling point group, traversing all sampling point groups, judging the small layer corresponding to the current sampling point group, and saving the data of the current sampling point group into the data body corresponding to the small layer.

4. The method for training a model for constructing an unmeasured logging curve based on small layer information according to claim 1, characterized in that, the training the LSTM model with the second training set to obtain a hard constraint LSTM model includes: Use the well logging curve data corresponding to the type of the target well logging curve in the second training set as label data, and use the other data in the second training set as feature data, and input them into the LSTM model for training. When the LSTM model meets the preset termination condition, obtain a second sub-model corresponding to the type of the target well logging curve.

5. The method for training a model for constructing an unmeasured well logging curve based on sub-layer information according to claim 2 or 4, wherein, the preset termination condition for model training is that the change of the mean absolute percentage error (MAPE) value with the increase of the training iteration times is less than 10%.

6. The method for training a model for constructing an unmeasured well logging curve based on sub-layer information according to claim 1, wherein, comparing the calculation results of the target well logging curve obtained by the first sub-model and the second sub-model, and selecting the sub-model with higher construction accuracy as the well logging curve construction model corresponding to the target well logging curve, includes: calculating the mean absolute percentage error (MAPE) values corresponding to the first sub-model and the second sub-model, and selecting the model with the lower mean absolute percentage error (MAPE) value as the well logging curve construction model corresponding to the target well logging curve.

7. A device for training a well logging curve construction model, wherein, the device includes a processor and a memory, and the memory stores computer program instructions suitable for being executed by the processor. When the computer program instructions are run by the processor, the steps in the method for training a model for constructing an unmeasured well logging curve based on sub-layer information according to any one of claims 1 to 6 are executed.

8. An intelligent method for constructing an unmeasured well logging curve based on sub-layer information, wherein, the method includes: obtaining the well logging curve data of a target well in a target block; performing soft constraint processing or hard constraint processing on the well logging curve data of the target well according to the type of the well logging curve construction model, and inputting the processed data into the well logging curve construction model to obtain the target well logging curve corresponding to the target well; wherein, the well logging curve construction model is the well logging curve construction model according to any one of claims 1 to 6.

9. A device for constructing a well logging curve, wherein, the device includes a processor and a memory, and the memory stores computer program instructions suitable for being executed by the processor. When the computer program instructions are run by the processor, the steps in the intelligent method for constructing an unmeasured well logging curve based on sub-layer information according to claim 8 are executed.

10. A computer-readable storage medium, wherein, the storage medium stores computer program instructions. When the computer program instructions are executed by the processor of a user device, the user device is enabled to execute the method for training a model for constructing an unmeasured well logging curve based on sub-layer information according to any one of claims 1 to 6, and / or, the intelligent method for constructing an unmeasured well logging curve based on sub-layer information according to claim 8.

Images