Method for identifying microfacies of carbonate rock reservoirs based on deep learning and related equipment

Abstract

The present invention discloses a method for identifying microfacies of carbonate reservoirs based on deep learning and related devices, belonging to the technical field of deep learning. The method includes the following steps: obtaining the original logging curves of several wells in the same working area, preprocessing the original logging curves and dividing them into a training set and a test set; optimizing the structure of the long short-term memory network to obtain a residual long short-term memory network; training the residual long short-term memory network with training samples to obtain a trained residual long short-term memory network; using the test samples to test the accuracy rate and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network; if the requirements are not met, return to the step of training the residual long short-term memory network with training samples; using the final residual long short-term memory network to identify the microfacies of carbonate reservoirs. The final residual long short-term memory network in the present invention has a higher prediction accuracy rate compared with the traditional long short-term memory network.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep learning, and relates to a method for identifying microfacies of carbonate reservoirs based on deep learning and related equipment. Background Art

[0002] As an important oil-bearing rock, the accurate division of the microfacies of carbonate reservoirs plays an important controlling role in the development and planar distribution of sweet spots. At present, the identification and division of sedimentary microfacies are mainly completed through the petrophysical characteristics (such as crystal structure and size) contained in core and thin-section data. However, core and thin-section data have the characteristics of short vertical sampling distance and sparse horizontal sampling, and the drilling cost of deep carbonate rocks is high, resulting in insufficient quantity of core and thin-section data. In addition, although conventional logging data (such as acoustic time difference curves) have advantages in terms of acquisition cost and quantity, they usually cannot provide effective information at the sedimentary microfacies level. Therefore, establishing the connection between the sedimentary microfacies revealed by core and thin sections and conventional logging data to obtain overall sedimentary microfacies information is an important problem to be solved for the identification of deep carbonate microfacies. Essentially, since the physical properties of rocks, such as mineral composition and fluid type, jointly affect reservoir microfacies, using logging curves for microfacies identification is a non-linear problem, so it is necessary to construct a suitable non-linear mapping. With the excellent performance of machine learning in many remote sensing and seismic data applications, great progress has been made in lithology identification based on machine learning (such as BP neural network (BackPropagation Network), support vector machine (Support Vector Machine, SVM), clustering fuzzy network, KNN algorithm (K Nearest Neighbors), and Bayesian algorithm), which provides a solid foundation for the intelligent identification of reservoir microfacies.

[0003] However, although the BP neural network constructs a non-linear mapping relationship between the input and output, the convergence speed of the model learning process is slow, and the error function is prone to local minima and unstable learning rates, resulting in the incompleteness of the BP network algorithm. The SVM algorithm has strong generalization ability and performs well in dealing with small sample data and non-linear problems, but it is only applicable to binary classification problems and has very poor effects on multi-classification problems. Fuzzy clustering analysis has a high complexity, resulting in a large overhead in time and space, and is affected by data noise, so the results will have large deviations. The KNN method has the problem of too high computational complexity, and when the training data distribution is unbalanced, the model will have a strong tendency, resulting in prediction errors. The Bayesian algorithm requires few parameters, is insensitive to missing data, and has a stable classification efficiency, but it is difficult to meet the prerequisite of independent attributes. Therefore, to a certain extent, these machine learning methods cannot be widely promoted in lithology classification. This is mainly because, on the one hand, such methods have problems of low computational efficiency and overfitting; on the other hand, the premise of applying these methods is that the relationship between lithology and logging input is independent in depth, ignoring the spatial sequence correlation of rocks during sedimentation and diagenesis, resulting in the inability to meet the needs of lithology division, especially fine microfacies division in many cases. Conventional machine learning methods cannot stably and effectively construct a non-linear mapping relationship between logging curves and reservoir microfacies. Summary of the Invention

[0004] An object of the present invention is to solve the technical problem that conventional logging data in the prior art cannot stably and effectively construct a non-linear mapping relationship between logging curves and reservoir microfacies, and to provide a method for identifying carbonate reservoir microfacies based on deep learning and related equipment.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] In the first aspect, the present invention provides a method for identifying carbonate reservoir microfacies based on deep learning, including the following steps:

[0007] Obtain the original logging curves of several wells in the same work area, preprocess the original logging curves and divide them into a training set and a test set. The training set includes training samples, and the test set includes test samples;

[0008] Optimize the structure of the long short-term memory network to obtain a residual long short-term memory network;

[0009] Use the training samples to train the residual long short-term memory network to obtain a trained residual long short-term memory network;

[0010] Use the test samples to test the accuracy and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed and the final residual long short-term memory network is obtained. If the requirements are not met, return to the step of training the residual long short-term memory network using the training samples.

[0011] Use the final residual long short-term memory network to perform microfacies identification on carbonate rock reservoirs.

[0012] A further improvement of the present invention lies in:

[0013] The preprocessing of the original logging curves includes the following steps:

[0014] Select appropriate input curves through input-output sensitivity analysis according to data availability, and perform invalid and abnormal data elimination and data normalization processing.

[0015] The original logging curves include conventional curves, element logging curves, and mineral interpretation result curves.

[0016] The conventional curves include natural gamma ray curve, uranium-free gamma ray curve, photoelectric absorption cross-section index curve; deep resistivity curve, shallow resistivity curve; acoustic travel time curve, neutron porosity curve; thorium content curve, uranium content curve, potassium content curve; the element logging curves include calcium carbonate content curve, calcium magnesium carbonate content curve, and other element content curves; the mineral interpretation result curves include calcite content curve and dolomite content curve.

[0017] The training set is three wells with uniform distribution of various microfacies, and the test set is two wells with uniform distribution of various microfacies.

[0018] The structural optimization of the long short-term memory network is established through the following steps:

[0019] Add a residual structure between the units of every two layers of the long short-term memory network. The extracted features are transformed in vector dimension through a fully connected layer, and then classification is achieved through a Softmax layer to obtain the residual long short-term memory network.

[0020] The specific method of using the test samples to test the accuracy and training error of the trained residual long short-term memory network is as follows:

[0021] The test samples are used as test data, and the trained residual long short-term memory network is used to obtain the mapping relationship, and the accuracy of the trained residual long short-term memory network is tested with the test samples.

[0022] In a second aspect, the present invention provides a carbonate rock reservoir microfacies identification system based on deep learning, which is characterized by including:

[0023] An input module that acquires the original logging curves of several wells in the same work area, preprocesses the original logging curves and then divides them into a training set and a test set. The training set includes training samples, and the test set includes test samples;

[0024] A structure optimization module that optimizes the structure of the long short-term memory network to obtain a residual long short-term memory network;

[0025] A training module that uses training samples to train the residual long short-term memory network to obtain a trained residual long short-term memory network;

[0026] A test module that uses test samples to test the accuracy rate and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network; if the requirements are not met, it returns to the step of using training samples to train the residual long short-term memory network;

[0027] A microfacies identification module that uses the final residual long short-term memory network to identify the microfacies of carbonate rock reservoirs.

[0028] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method for identifying the microfacies of carbonate rock reservoirs based on deep learning are implemented.

[0029] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for identifying the microfacies of carbonate rock reservoirs based on deep learning are implemented.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention discloses a method for identifying microfacies of carbonate reservoirs based on deep learning. For the input, logging curves for carbonate reservoirs are selected through sensitivity analysis, and the original logging curves are industrially preprocessed. Regarding the network structure, the present invention proposes to increase the network depth using a residual structure. The complex internal structure of the long short-term memory network (LSTM) also leads to more parameters and explosive computational complexity, so the long short-term memory network (LSTM) will not exceed 4 layers during training. In addition, the long short-term memory network (LSTM) has limited learning ability for long-term dependencies and cannot effectively utilize shallow formation information when facing ultra-long formation sequences. Due to the addition of the residual structure, the residual long short-term memory network (ResLSTM) can train a deep network with a maximum depth of 8 layers; at the same time, it greatly improves the network's ability to learn long-term dependency relationships, and the addition of the residual structure can help the network better extract sequence features. The model is trained using training samples, and the accuracy of the residual long short-term memory network (ResLSTM) is tested using test samples. The long short-term memory network (LSTM) achieves good classification for major microfacies categories, but performs poorly for rapidly changing thin-layer microfacies; for thin layers with rapidly changing microfacies, the trained final residual long short-term memory network can also make good predictions. Generally speaking, the final residual long short-term memory network in the present invention has a higher prediction accuracy compared to the traditional long short-term memory network (LSTM). Brief Description of the Drawings

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0033] Figure 1 It is a flow chart of the method for identifying microfacies of carbonate reservoirs based on deep learning in the present invention;

[0034] Figure 2 It is a principle flow chart of the method for identifying microfacies of carbonate reservoirs based on deep learning in the present invention;

[0035] Figure 3 It is a training data diagram of a well in the present invention;

[0036] Figure 4 It is a schematic diagram of the distribution of 5 wells in the present invention;

[0037] Figure 5 It is a network structure model for ResLSTM reservoir microfacies identification;

[0038] Figure 6 This is the model prediction result diagram of the final residual long short-term memory network in the present invention;

[0039] Figure 7 This is the system diagram of the carbonate reservoir microfacies identification method based on deep learning in the present invention;

[0040] Figure 8 This is the electronic device module diagram of the carbonate reservoir microfacies identification method based on deep learning in the present invention. Specific Embodiments

[0041] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. Generally, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0042] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0043] It should be noted that: like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0044] The following further describes the present invention in detail with reference to the drawings:

[0045] See Figure 1 , an embodiment of the present invention proposes a carbonate reservoir microfacies identification method based on deep learning, including the following steps:

[0046] S1. Obtain the original logging curves of several wells in the same working area, preprocess the original logging curves and divide them into a training set and a test set. The training set includes training samples, and the test set includes test samples;

[0047] S2. Optimize the structure of the long short-term memory network to obtain a residual long short-term memory network;

[0048] S3. Use the training samples to train the residual long short-term memory network to obtain a trained residual long short-term memory network;

[0049] S4. Use the test samples to test the accuracy and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed, and the final residual long short-term memory network is obtained; if the requirements are not met, return to the step of training the residual long short-term memory network using the training samples.

[0050] S5. Use the final residual long short-term memory network to perform microfacies identification on carbonate rock reservoirs.

[0051] The present invention discloses a method for microfacies identification of carbonate rock reservoirs based on deep learning. For the input, logging curves for carbonate rock reservoirs are selected through sensitivity analysis, and industrial preprocessing is performed on the original logging curves. Regarding the network structure, the present invention proposes to increase the network depth by using the residual structure. The complex internal structure of the long short-term memory network (LSTM) also leads to a larger number of parameters and explosive computational complexity, so the long short-term memory network (LSTM) will not exceed 4 layers during training. In addition, the long short-term memory network (LSTM) has limited learning ability for long-term dependencies and cannot effectively utilize shallow formation information when facing ultra-long formation sequences. Due to the addition of the residual structure, the residual long short-term memory network (ResLSTM) can train a deep network with a maximum depth of 8 layers; at the same time, it greatly improves the network's ability to learn long-term dependency relationships, and the addition of the residual structure helps the network better extract sequence features. Use the training samples to train the model and test samples to test the accuracy of the residual long short-term memory network (ResLSTM). The long short-term memory network (LSTM) has achieved good classification for major microfacies categories, but for rapidly changing thin-layer microfacies, the performance of the long short-term memory network (LSTM) is very poor; while for thin layers with rapidly changing microfacies, the trained final residual long short-term memory network can also make good predictions. Generally speaking, the final residual long short-term memory network in the present invention has a higher prediction accuracy compared to the traditional long short-term memory network (LSTM).

[0052] Data-driven deep learning strategies perform very well in constructing non-linear mapping relationships of data. The Long Short-Term Memory (LSTM) network has improved the problems of gradient explosion and gradient disappearance to a certain extent; in addition, the LSTM network can naturally learn long-term dependencies. When using logging data for lithology prediction, the LSTM network is also superior to other point-to-point models; moreover, the LSTM network also has good results in lithology identification in complex formations, such as volcanic rocks and carbonate rocks. However, the complex internal structure of the LSTM network also leads to more parameters and explosive computational complexity, so the LSTM network will not exceed 4 layers during training. In addition, the learning of long-term dependencies by the LSTM network is limited, and the LSTM network cannot effectively utilize shallow formation information when facing ultra-long formation sequences. The Residual Long Short-Term Memory (ResLSTM) network can train an 8-layer deep network due to the addition of the residual structure; at the same time, it has well improved the network's ability to learn long-term dependencies. This provides a new way of thinking for the identification and interpretation of sedimentary microfacies.

[0053] The following details the content of the present invention in conjunction with specific embodiments:

[0054] See Figure 2 , which is the principle flow chart of the carbonate reservoir microfacies identification method based on deep learning in the present invention.

[0055] Step 1: Selection of original logging curves:

[0056] According to the data availability and input-output sensitivity analysis, after eliminating invalid and abnormal data and normalizing the data, 15 logging curves are selected as inputs, which are divided into conventional curves, element logging curves, and mineral interpretation result curves. The conventional curves include the natural gamma ray curve (GR), the gamma ray curve without uranium (CGR), and the photoelectric absorption cross-section index curve (PE); the deep resistivity curve (RD), the shallow resistivity curve (RS); the acoustic travel time curve (DT), the neutron porosity curve (CNL); the thorium content curve (TH), the uranium content curve (U), and the potassium content curve (K). The element logging curves include: the calcium carbonate content curve (CALC_VOL), the calcium magnesium carbonate content curve (DOLO_VOL), and the other element content curve (Other_Mud). The mineral interpretation result curves include: the calcite content curve (CALC_Mud) and the dolomite content curve (DOLO_Mud). The conventional curves include lithology curves (natural gamma ray (GR), gamma ray curve without uranium (CGR), photoelectric absorption cross-section index curve (PE)), electrical property curves (deep resistivity (RD), shallow resistivity curve (RS)), porosity curves (acoustic travel time curve (DT), neutron porosity curve (CNL)), and radioactive logging curves (thorium content (TH), uranium content (U), potassium content (K)).

[0057] Among them, the natural gamma ray curve (GR) and the gamma ray curve without uranium (CGR) are often used to identify the clay mineral content inside the reservoir, and the photoelectric absorption cross-section index is sensitive to the difference in dolomite content and calcite content inside the reservoir. Therefore, the lithology curve has strong sensitivity and applicability for identifying the basic structure of lithofacies. The electrical property curve and the porosity curve have strong identification ability for pore structure and pore size. Therefore, when there are physical property differences between different lithologies, there will be certain differences in the responses of these two groups of curves. The radioactive curve has a certain indication effect on reflecting the formation sedimentary environment and vertical sedimentary differences. Therefore, it becomes an important input curve as a sensitive curve for microfacies identification. In addition, to address the problem of complex predicted target microfacies, in this experiment, the element logging curves (calcium carbonate content (CALC_VOL), calcium magnesium carbonate content (DOLO_VOL), other element content (Other_Mud)) and the mineral interpretation result curves (calcite content (CALC_Mud), dolomite content (DOLO_Mud)) are added to the input data to improve the diversity of the data and the stability of the prediction. The vertical sampling rate of the logging data is 0.125 m, which can meet the resolution requirements for microfacies identification in the target interval of the study area.

[0058] Step 2: Division of the training set and the test set:

[0059] According to the distribution of the data obtained in Step 1, as Figure 3As shown, it can be seen that the logging curves of the three wells approximately exhibit a mixed Gaussian distribution, covering all target rock types. That is, using the three wells as a training set can statistically meet the needs of modeling training. Three wells with uniform microfacies distribution are used as the training set, and the other two wells are used as the test set. As Figure 4 shown, the logging curves of one of the training wells are presented. The well data is rich, but there are many thin interbeds and the sample distribution is uneven.

[0060] Step 3: Optimize the structure of the long short-term memory network to obtain a residual long short-term memory network:

[0061] As the network depth increases, the errors generated during training will first decrease and then increase. The increase in errors is not due to overfitting, but rather the increase in training difficulty caused by the deepening of the network layers. Theoretically, a deep network should have better performance than a shallow network. However, in fact, the increase in the number of network layers results in an increase in training difficulty and a consequent decrease in network performance, which is known as network degradation.

[0062] The residual structure effectively solves the degradation problem in deep networks. Compared with ordinary neural networks, neural networks with a residual structure can also achieve good results when training deeper models. This is because the residual structure improves the correlation between gradients and losses, thereby enhancing the learning ability of the network; at the same time, it also alleviates the problems of gradient vanishing and gradient explosion. Its structure diagram is as Figure 2 shown. The core units of the residual structure are two: the shortcut connection and the identity mapping. The shortcut connection makes the residual possible, and the identity mapping can deepen the network. There are mainly two identity mappings: the skip connection and the activation function (ReLu). Since the identity mapping is always on, the output of the function only needs to learn the residual mapping. This relationship is expressed as formula (1):

[0063]

[0064] Among them, y is the output layer, x is the input layer, and F( x , W) is a mapping function containing the internal core parameter W. Without the shortcut connection, F( x , W) should represent the result x produced by the input y passing through two layers of the network. However, for the identity mapping x , F( x , W) only needs to learn the residual mapping y – x . As the number of network layers is stacked and increased, if no new residual mapping is required, the network can directly bypass the identity mapping without training, which can greatly simplify the training of the network.

[0065] Inspired by the idea of modeling the difference between the intermediate layer and the output target, this paper introduces residual connections between the long short-term memory (LSTM) layers. LSTMi and LSTMi+1 represent the ith and (i + 1)th LSTM layers in the network structure, with parameters Wi and Wi+1 respectively. At the nth time step, for the long short-term memory network (LSTM) without residual connections, we have:

[0066]

[0067]

[0068]

[0069] where, is the input of LSTMi at time step t, is the hidden state of LSTMi+1 at time step t, is the memory state of LSTMi at time step t.

[0070] If a residual connection is added between the LSTMi layer and the LSTMi+1 layer, the above equation becomes:

[0071]

[0072]

[0073]

[0074] The addition of the residual connection greatly improves the gradient flow of the network during backpropagation, which enables the residual long short-term memory network (ResLSTM) to train deeper networks than the long short-term memory network (LSTM).

[0075] The traditional long short-term memory network (LSTM) has a total of 6 hidden layers. A residual connection is added every two layers. The extracted features are passed through a fully connected layer to convert the vector dimension, and then classification is achieved through a Softmax layer. The experimental results show that the addition of the residual structure enables the training of deeper networks in model training.

[0076] Step 4: Use the training samples to train the residual long short-term memory network (ResLSTM) to obtain the trained residual long short-term memory network:

[0077] Input the training samples into the residual long short-term memory network (ResLSTM) to train the residual long short-term memory network (ResLSTM), as Figure 5As shown, 15 preprocessed original logging curves are input into the Residual Long Short-Term Memory Network (ResLSTM) for training. At the same time, the training samples are input into the Long Short-Term Memory Network (LSTM) to train the LSTM, which is used to compare the prediction results of the trained ResLSTM and LSTM.

[0078] Step Five, Accuracy Test and Microfacies Prediction:

[0079] Use the test samples in Step 2. The ResLSTM trained in Step Four is used to obtain the mapping relationship and compared with the traditional Long Short-Term Memory Network (LSTM) to verify the accuracy and superiority of the trained ResLSTM in reservoir microfacies identification. If the requirements are met, the training is completed and the final ResLSTM is obtained; if the requirements are not met, return to the step of training the ResLSTM using the training samples; use the final ResLSTM to identify the microfacies of carbonate reservoirs.

[0080] See Figure 6 , the microfacies in the figure show that the microfacies distribution of this well is quite complex and there are certain thin layers with rapid microfacies changes. Such as Figure 5 shown, which shows the prediction results and the actual situation of the two networks in this well. Specifically, it can be seen from the prediction results that the predicted microfacies and the actual situation are basically corresponding, which indicates that the model has extracted and learned the data features. Compared with the Long Short-Term Memory Network (LSTM), the prediction results of the final ResLSTM are more accurate (as shown in Table 1). Specifically, the LSTM has achieved good classification for large categories of microfacies, but for thin layers of microfacies with rapid changes, the LSTM performs poorly; while even for thin layers of microfacies with rapid changes, the final ResLSTM in the present invention can still make relatively good predictions. Generally speaking, the final ResLSTM has a higher prediction accuracy than the traditional LSTM.

[0081] Table 1 Microfacies Prediction Accuracy of Network Models on Two Test Wells

[0082]

[0083] See Figure 7 , an embodiment of the present invention proposes a carbonate reservoir microfacies identification system based on deep learning, including:

[0084] An input module that acquires the original logging curves of several wells in the same working area, preprocesses the original logging curves, and divides them into a training set and a test set. The training set includes training samples, and the test set includes test samples;

[0085] A structure optimization module that optimizes the structure of the long short-term memory network to obtain a residual long short-term memory network;

[0086] A training module that trains the residual long short-term memory network using training samples to obtain a trained residual long short-term memory network;

[0087] A test module that uses test samples to test the accuracy rate and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network; if the requirements are not met, it returns to the step of training the residual long short-term memory network using training samples;

[0088] A microfacies identification module that uses the final residual long short-term memory network to identify the microfacies of carbonate rock reservoirs.

[0089] See Figure 8 , the third object of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for identifying the microfacies of carbonate rock reservoirs based on deep learning.

[0090] Acquire the original logging curves of several wells in the same working area, preprocess the original logging curves, and divide them into a training set and a test set. The training set includes training samples, and the test set includes test samples;

[0091] Optimize the structure of the long short-term memory network to obtain a residual long short-term memory network;

[0092] Train the residual long short-term memory network using training samples to obtain a trained residual long short-term memory network;

[0093] Use test samples to test the accuracy rate and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network; if the requirements are not met, it returns to the step of training the residual long short-term memory network using training samples;

[0094] Use the final residual long short-term memory network to identify the microfacies of carbonate rock reservoirs.

[0095] The fourth object of the present invention is to provide a computer-readable storage medium storing a computer program, which when executed by a processor, implements the steps of the method for identifying microfacies of carbonate reservoirs based on deep learning.

[0096] Obtain the original logging curves of several wells in the same working area, preprocess the original logging curves and divide them into a training set and a test set. The training set includes training samples, and the test set includes test samples;

[0097] Optimize the structure of the long short-term memory network to obtain a residual long short-term memory network;

[0098] Use the training samples to train the residual long short-term memory network to obtain a trained residual long short-term memory network;

[0099] Use the test samples to test the accuracy and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network; if the requirements are not met, return to the step of using the training samples to train the residual long short-term memory network;

[0100] Use the final residual long short-term memory network to identify the microfacies of carbonate reservoirs.

[0101] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the specified functions in one Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0103] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in the process(es) Figure 1 step(s) or multiple step(s) and / or block(s) Figure 1 block(s) or multiple block(s).

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the process(es) Figure 1 step(s) or multiple step(s) and / or block(s) Figure 1 block(s) or multiple block(s), as Figure 8 illustrated.

[0105] The foregoing are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying microfacies of carbonate rock reservoirs based on deep learning, characterized in that, It includes the following steps: Obtain the original logging curves of several wells in the same working area, preprocess the original logging curves and divide them into a training set and a test set. The training set includes training samples, and the test set includes test samples. The preprocessing of the original logging curves includes the following steps: Select appropriate input curves through input-output sensitivity analysis according to data availability, and perform invalid and abnormal data elimination and data normalization processing; Optimize the structure of the long short-term memory network to obtain a residual long short-term memory network. The structure optimization of the long short-term memory network is established through the following steps: Add a residual structure between the units of every two layers of the long short-term memory network. The extracted features are transformed in vector dimension through a fully connected layer, and then classification is achieved through a Softmax layer to obtain a residual long short-term memory network; Use the training samples to train the residual long short-term memory network to obtain the trained residual long short-term memory network; Use the test samples to test the accuracy and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network. If the requirements are not met, return to the step of using the training samples to train the residual long short-term memory network; Use the final residual long short-term memory network to perform microfacies identification on carbonate rock reservoirs.

2. The method for identifying microfacies of carbonate rock reservoirs based on deep learning according to claim 1, wherein The original logging curves include conventional curves, element logging curves, and mineral interpretation result curves.

3. The method for identifying the microfacies of carbonate rock reservoirs based on deep learning according to claim 2, wherein The conventional curves include natural gamma ray curve, uranium-free gamma ray curve, photoelectric absorption cross-section index curve; deep resistivity curve, shallow resistivity curve; acoustic travel time curve, neutron porosity curve; thorium content curve, uranium content curve, potassium content curve; the element logging curves include calcium carbonate content curve, calcium magnesium carbonate content curve, other element content curves; the mineral interpretation result curves include calcite content curve and dolomite content curve.

4. The method for identifying microfacies of carbonate reservoirs based on deep learning according to claim 1, wherein The training set is three wells with uniform distribution of various microfacies, and the test set is two wells with uniform distribution of various microfacies.

5. The method for identifying microfacies of carbonate reservoirs based on deep learning according to claim 4, wherein The specific steps of using the test samples to test the accuracy and training error of the trained residual long short-term memory network are as follows: Use the test samples as test data, and use the trained residual long short-term memory network to obtain the mapping relationship, and test the accuracy of the trained residual long short-term memory network with the test samples.

6. A carbonate reservoir microfacies identification system based on deep learning, characterized in that, It includes: An input module that obtains the original logging curves of several wells in the same working area, preprocesses the original logging curves and divides them into a training set and a test set. The training set includes training samples, and the test set includes test samples. The preprocessing of the original logging curves includes the following steps: Select appropriate input curves through input-output sensitivity analysis according to data availability, and perform invalid and abnormal data elimination and data normalization processing; The structure optimization module optimizes the structure of the long short-term memory network to obtain a residual long short-term memory network; the structure optimization of the long short-term memory network is established through the following steps: adding a residual structure between the units of every two layers of the long short-term memory network, converting the dimension of the extracted features through a fully connected layer, and then realizing classification through a Softmax layer to obtain a residual long short-term memory network; The training module uses training samples to train the residual long short-term memory network to obtain a trained residual long short-term memory network; The testing module uses test samples to test the accuracy rate and training error of the trained residual long short-term memory network. If the requirements are met, the training is completed to obtain the final residual long short-term memory network; if the requirements are not met, it returns to the step of using training samples to train the residual long short-term memory network; The microfacies identification module uses the final residual long short-term memory network to identify the microfacies of carbonate rock reservoirs.

7. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for identifying the microfacies of carbonate rock reservoirs based on deep learning according to any one of claims 1-5 are implemented.

8. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the steps of the method for identifying the microfacies of carbonate rock reservoirs based on deep learning according to any one of claims 1-5 are implemented.

Images