A carbonate reservoir reservoir body pattern automatic identification method

Abstract

This invention discloses an automatic identification method for carbonate reservoir patterns, comprising the following steps: establishing typical reservoir patterns; establishing a numerical simulation conceptual model based on the typical reservoir patterns; simulating the development process of different conceptual models and extracting production dynamic information; and constructing a fractured-vuggy reservoir pattern inversion model using a deep learning algorithm. This invention employs the aforementioned automatic identification method for carbonate reservoir patterns, using a deep learning algorithm to quickly and accurately identify fractured-vuggy reservoir patterns. It overcomes the shortcomings of time-consuming, costly, and highly specialized geostatistical inversion methods, avoids the influence of subjective human factors, significantly improves economic efficiency, and greatly reduces testing time costs.

Description

Technical Field

[0001] This invention relates to the field of unconventional oil reservoir development technology, and in particular to an automatic identification method for carbonate reservoir patterns. Background Technology

[0002] Currently, the main methods for identifying fractured / fractured-vuggy reservoir models include inversion combining seismic and well logging interpretation, and manual identification through production dynamic curves. Seismic and well logging interpretation is highly specialized, complex, and extremely costly, with limited accuracy. Manual identification of reservoir models through production dynamic curves is heavily influenced by individual subjectivity; each person's understanding of the curve differs, leading to different results and making it difficult to establish a unified standard. Furthermore, numerous influencing factors exist in actual field conditions, making it difficult for humans to judge subtle changes in the production dynamic curves. In addition, manual identification involves a huge workload and high time costs.

[0003] Geostatistical inversion methods combining seismic and well logging data mainly include steps such as collecting seismic interpretation data, well logging interpretation data, and geostatistical inversion. Seismic data collection requires artificial methods such as explosive detonation and controlled-source vibration to induce crustal vibrations, and collecting information from each receiving point after the explosion. The recorded data is then processed to infer the reservoir model of fractured-vuggy reservoirs. Well logging requires drilling exploratory wells in the early stages and using the geophysical properties of the rock formations, such as electrochemical, electrical, acoustic, and radioactive characteristics, to measure geophysical parameters. Geostatistical inversion first uses seismic data to establish probability distribution functions and variograms under different conditions based on lithological differences, obtaining an initial geological model of the reservoir lithology. Forward modeling is then performed using this initial geological model, iterating repeatedly with the original seismic data to achieve the required signal-to-noise ratio. Finally, single-well logging results are used to verify the inversion results, ultimately yielding the reservoir model. Therefore, it is evident that determining reservoir models using seismic logging data is only applicable before reservoir development. It is not suitable for adjusting production parameters of already developed wells by determining reservoir models. Furthermore, the data interpretation requires a high level of expertise, involves a complex process, and is extremely costly, making it unsuitable for all oilfields.

[0004] The manual identification method for production dynamic curves involves manually analyzing the dynamic curves of flowing pressure and production in reservoirs to obtain the reservoir pattern for the region. This method mainly includes collecting production dynamic curves of the target block, plotting energy indicator curves, and analyzing and comparing the energy indicator curves. However, the actual production dynamics in the field are complex, and the different influences of different production systems and geological conditions make it difficult to distinguish the energy indicator curves in some areas. Furthermore, due to the large number of wells and blocks in the field, plotting the energy indicator curves of all wells and analyzing and comparing them one by one is time-consuming and labor-intensive, requiring a significant amount of time and effort, and the accuracy cannot be guaranteed, resulting in shortcomings in both time cost and accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic identification method for carbonate oil reservoir patterns. Through deep learning algorithms, it can quickly and accurately identify fracture-vuggy reservoir patterns, overcoming the shortcomings of geostatistical inversion, such as long time consumption, high cost, and high degree of specialization. It also avoids the influence of human subjective factors, significantly improves economic efficiency, and greatly reduces the time cost of testing.

[0006] To achieve the above objectives, the present invention provides an automatic identification method for carbonate oil reservoir patterns, comprising the following steps:

[0007] S1. Establish a typical collective model for storage;

[0008] S2. Establish a numerical simulation conceptual model based on typical reservoir patterns;

[0009] S3. Simulate the development process of different conceptual models and extract dynamic production information;

[0010] S4. A deep learning algorithm is used to construct a reservoir pattern inversion model for fractured-vuggy oil reservoirs.

[0011] Preferably, the establishment of a typical reservoir model in step S1 specifically involves classifying fracture / fracture-vuggy carbonate oil reservoirs into basic reservoir models based on the spatial characteristics of currently developed fracture / fracture-vuggy carbonate oil reservoirs.

[0012] Preferably, the basic reservoir type includes a fracture type, a cave type, a fracture + cave type, a bottom water fracture type, a bottom water cave type, and a bottom water fracture + cave type.

[0013] Preferably, in step S2, establishing a numerical simulation conceptual model based on a typical reservoir model specifically involves classifying and establishing a large number of numerical simulation conceptual models for different basic reservoir models. The parameter settings of the conceptual model include geological parameters, fluid parameters, and development parameters. Geological parameters include dip angle, thickness, permeability, porosity, oil-water interface depth, and edge-bottom water intensity. Fluid parameters include fluid viscosity, density, dissolved gas-oil ratio (PVT) properties, and relative permeability curves. Development parameters include well location and operating regime.

[0014] Preferably, in step S3, simulating the development process of different conceptual models and extracting production dynamic information specifically involves using a numerical simulator to simulate the development dynamics of different conceptual models, thereby obtaining dynamic information during the depletion development process of different conceptual models. The dynamic information includes cumulative oil production, cumulative water production, water cut, and bottom hole flowing pressure production dynamics.

[0015] Preferably, the numerical simulator includes a black oil numerical simulator, a component numerical simulator, and a thermal recovery numerical simulator.

[0016] Preferably, the step S4, which uses a deep learning algorithm to construct a fractured-vuggy reservoir model inversion model, includes the following steps:

[0017] (1) The deep learning model is trained by taking the four characteristic curves of cumulative oil production, cumulative water production, water cut and bottom hole flowing pressure as inputs and the reservoir pattern corresponding to the production characteristic as output.

[0018] (2) The mainstream deep learning classification algorithms include Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). A CNN consists of an input layer, a hidden layer, and an output layer. The input layer receives and processes the training data and passes it to the hidden layer. The hidden layer includes convolutional layers, pooling layers, and fully connected layers, which are used to train the data and extract features. Finally, the output layer is used to output the classification label. A Recurrent Neural Network (RNN) consists of recurrent units and is suitable for solving sequence problems. LSTM (Long Short-Term Memory Artificial Neural Network) and GRU (Gated Recurrent Unit) are two variants of Recurrent Neural Networks (RNN), which add Dropout layers and L1 and L2 regularization. Time series classification algorithm models include one-dimensional convolutional neural networks (1DCNN), CNN-LSTM, and ConvLSTM models.

[0019] (3) Randomly select 80% of the samples in the concept model sample library as the training set and 20% of the samples as the validation set. Use the training set data to train the deep learning model. After training, use the validation set to test the model recognition effect. If the recognition accuracy is higher than 90%, the model training effect is considered to be good and can be used for rapid recognition of stored patterns. If the recognition accuracy is lower than 90%, the model hyperparameters need to be adjusted and retrained until the recognition accuracy is higher than 90%. The adjustable hyperparameters of the model include batch size, learning rate, number of training steps, number of convolution kernels, number of layers, number of recurrent neural units, and type of optimizer.

[0020] Therefore, the present invention adopts the above-mentioned automatic identification method for carbonate oil reservoir patterns. Through deep learning algorithms, it can quickly and accurately identify the patterns of fracture-vuggy reservoirs, which solves the shortcomings of geostatistical inversion, such as long time consumption, high cost and high professional requirements. It avoids the influence of human subjective factors, significantly improves economic efficiency and greatly reduces the time cost of testing.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an embodiment of an automatic identification method for carbonate oil reservoir collective patterns according to the present invention;

[0023] Figure 2This is a conceptual model diagram of an embodiment of the automatic identification method for carbonate oil reservoir collective patterns according to the present invention;

[0024] Figure 3 This is a flowchart of the deep learning algorithm for an embodiment of the automatic identification method for collective patterns in carbonate oil reservoirs according to the present invention.

[0025] Figure 4 This is a diagram of the FCN-GRU algorithm model, representing an embodiment of an automatic identification method for carbonate oil reservoir patterns according to the present invention. Detailed Implementation

[0026] This invention provides an automatic identification method for carbonate oil reservoir patterns, comprising the following steps:

[0027] S1. Establish typical reservoir models. Specifically, based on the spatial characteristics of the currently developed fracture / fracture-vuggy carbonate oil reservoirs, the fracture / fracture-vuggy carbonate oil reservoirs are divided into basic reservoir models. The basic reservoir models include fracture model, cave model, fracture + cave model, bottom water fracture model, bottom water cave model, and bottom water fracture + cave model.

[0028] S2. Establish numerical simulation conceptual models based on typical reservoir models. Specifically, for different basic reservoir models, establish a large number of numerical simulation conceptual models. The parameter settings of the conceptual models include geological parameters, fluid parameters, and development parameters. Geological parameters include dip angle, thickness, permeability, porosity, oil-water interface depth, and edge and bottom water intensity. Fluid parameters include fluid viscosity, density, dissolved gas-oil ratio (PVT) properties, and relative permeability curves. Development parameters include well location and operating regime (the operating regime is not a fixed value, but rather a dynamic change over time in terms of bottom hole pressure, daily fluid production, or daily oil production). To ensure the accuracy of subsequent inversion, the number of conceptual models for each reservoir model should be increased as much as possible.

[0029] S3. Simulate and calculate the development process of different conceptual models to extract production dynamic information. Specifically, use numerical simulators to simulate and calculate the development dynamics of different conceptual models to obtain dynamic information of the depletion development process of different conceptual models. The dynamic information includes cumulative oil production, cumulative water production, water cut, and bottom hole flowing pressure production dynamics. The numerical simulators include black oil numerical simulators, component numerical simulators, and thermal recovery numerical simulators.

[0030] S4. Construct a fracture-vuggy reservoir model inversion model using deep learning algorithms, including the following steps:

[0031] (1) The deep learning model is trained by taking the four characteristic curves of cumulative oil production, cumulative water production, water cut and bottom hole flowing pressure as inputs and the reservoir pattern corresponding to the production characteristic as output.

[0032] (2) The mainstream deep learning classification algorithms include Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). A CNN consists of an input layer, a hidden layer, and an output layer. The input layer receives and processes the training data and passes it to the hidden layer. The hidden layer includes convolutional layers, pooling layers, and fully connected layers, which are used to train the data and extract features. Finally, the output layer is used to output the classification label. A Recurrent Neural Network (RNN) consists of recurrent units and is suitable for solving sequence problems. LSTM (Long Short-Term Memory Artificial Neural Network) and GRU (Gated Recurrent Unit) are two variants of Recurrent Neural Networks (RNN), which add Dropout layers and L1 and L2 regularization. Time series classification algorithm models include one-dimensional convolutional neural networks (1DCNN), CNN-LSTM, and ConvLSTM models.

[0033] (3) Randomly select 80% of the samples in the concept model sample library as the training set and 20% of the samples as the validation set. Use the training set data to train the deep learning model. After training, use the validation set to test the model recognition effect. If the recognition accuracy is higher than 90%, the model training effect is considered to be good and can be used for rapid recognition of stored patterns. If the recognition accuracy is lower than 90%, the model hyperparameters need to be adjusted and retrained until the recognition accuracy is higher than 90%. The adjustable hyperparameters of the model include batch size, learning rate, number of training steps, number of convolution kernels, number of layers, number of recurrent neural units, and type of optimizer.

[0034] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] Example

[0036] Taking the pattern recognition of a fractured-vuggy reservoir as an example, the method process is explained using data from Cheng Hong's paper "Research on the Indicative Significance of Production Dynamic Curves in Fractured-Vuggy Carbonate Reservoirs". The accuracy of the method is verified using actual field data from Cheng Hong's paper "Research on the Indicative Significance of Production Dynamic Curves in Fractured-Vuggy Carbonate Reservoirs".

[0037] (1) Six types of fissure-cavity reservoir models were established, namely fissure type model, karst cave type model, fissure-karst cave type model, bottom water fissure type model, bottom water karst cave type model, and bottom water fissure-karst cave type model.

[0038] (2) Establish conceptual models of fractured-vuggy reservoirs under different modes (e.g.) Figure 2 Design reasonable static parameters (as shown in Table 1).

[0039] Table 1 Static Parameter Settings

[0040]

[0041] (3) Calculate the numerical simulation models of different modes, compare the production dynamic curves of different modes, and select four characteristic curves: cumulative oil production, cumulative water production, water cut and bottom hole flowing pressure.

[0042] (4) Design reasonable production dynamic parameters (as shown in Table 2), calculate a large number of numerical simulation models, and construct a sample library for reservoir pattern inversion and identification.

[0043] In the established conceptual model, reasonable dynamic and static parameters are designed. Conceptual models with different parameter combinations are randomly established for each mode, and numerical simulations are performed. Simultaneously, four production dynamics—cumulative oil production, cumulative water production, water cut, and bottomhole flowing pressure—are extracted from the calculation results, and the corresponding reservoir pattern is recorded. This process is repeated multiple times to construct a large sample library for reservoir pattern recognition.

[0044] Table 2 Production Dynamic Parameter Settings

[0045] Water-oil volume ratio 0～50 Bottom hole flowing pressure / KPa 8000～30000 Oil-water interface depth / m 3100～3400 Well location (distance from oil-water interface) / m -400～1900 <![CDATA[Production liquid volume / m 3 / d]]> 50～500

[0046] (5) Use deep learning algorithms (such as...) Figure 3 (As shown) Construct a fracture-vuggy reservoir pattern inversion and automatic identification model to achieve rapid and accurate identification of fracture-vuggy reservoir patterns.

[0047] Using the gated recurrent unit fully convolutional neural network (FCN-GRU) algorithm (e.g.) Figure 4 A reservoir pattern inversion and automatic identification model was constructed for fractured-vuggy reservoirs (as shown). The optimal hyperparameter combination was adjusted to achieve the best accuracy. The model settings included a batch size of 32, a learning rate of 0.1, 32, 64, and 32 kernels per convolutional layer, 8 GRU nodes, ReLU activation function, categorical crossentropy loss function to handle multi-class classification, and Adam optimizer. After model construction, the input variables were cumulative oil production, cumulative water production, bottom hole flowing pressure, and water cut changes (production dynamics), with the target label being reservoir pattern. 80% of the sample library was used as the training set (accuracy 0.91), and 20% was used as the test set (accuracy 0.88). After model training, inputting production dynamics enabled automatic, rapid, and accurate reservoir pattern identification. The accuracy of the model was verified using actual field data from Cheng Hong's paper, "Research on the Indicative Significance of Production Dynamic Curves in Fractured-Vuggy Carbonate Reservoirs" (Table 3).

[0048] As can be seen from Table 3 below, the model for reservoir pattern inversion and automatic identification of fracture / fractured reservoirs constructed using the gated recurrent unit fully convolutional neural network (FCN-GRU) algorithm has a high accuracy rate. It accurately predicts the actual field data in the literature "Research on the Indicative Significance of Production Dynamic Curves of Fractured Carbonate Reservoirs", indicating that the method has the characteristics of high efficiency and accuracy and can be used for reservoir pattern identification in actual field.

[0049] Table 3. Verification results of the collective pattern recognition model

[0050]

[0051] Therefore, the present invention adopts the above-mentioned automatic identification method for carbonate oil reservoir patterns. Through deep learning algorithms, it can quickly and accurately identify the patterns of fracture-vuggy reservoirs, which solves the shortcomings of geostatistical inversion, such as long time consumption, high cost and high professional requirements. It avoids the influence of human subjective factors, significantly improves economic efficiency and greatly reduces the time cost of testing.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for automatic identification of carbonate oil reservoir patterns, characterized in that, Includes the following steps: S1. Establish a typical collective storage model; S2. Establish numerical simulation conceptual models based on typical reservoir models; specifically, for different basic reservoir models, establish a large number of numerical simulation conceptual models. The parameter settings of the conceptual models include geological parameters, fluid parameters and development parameters. Geological parameters include dip angle, thickness, permeability, porosity, oil-water interface depth, and edge and bottom water intensity. Fluid parameters include fluid viscosity, density, dissolved gas-oil ratio (PVT) properties and relative permeability curve. Development parameters include well location and working conditions. S3. Simulate and calculate the development process of different conceptual models to extract production dynamic information; specifically, use a numerical simulator to simulate and calculate the development dynamics of different conceptual models to obtain dynamic information during the depletion development process of different conceptual models. The dynamic information includes cumulative oil production, cumulative water production, water cut, and bottom hole flowing pressure. S4. A deep learning algorithm is used to construct a reservoir collective model inversion model for fractured-vuggy oil reservoirs. (1) The deep learning model is trained by taking four production dynamic features, namely cumulative oil production, cumulative water production, water cut and bottom hole flowing pressure, as inputs and the reservoir pattern corresponding to the production dynamic features as outputs. (2) The mainstream deep learning classification algorithms include Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). A CNN consists of an input layer, a hidden layer, and an output layer. The input layer receives and processes the training data and passes it to the hidden layer. The hidden layer includes convolutional layers, pooling layers, and fully connected layers, which are used to train the data and extract features. Finally, the output layer is used to output the classification label. A Recurrent Neural Network (RNN) consists of recurrent units and is suitable for solving sequence problems. Long Short-Term Memory Artificial Neural Network (LSTM) and Gated Recurrent Unit (GRU) are two variants of Recurrent Neural Networks (RNN), which add Dropout layers and L1 and L2 regularization. Time series classification algorithm models include one-dimensional convolutional neural networks (1DCNN), CNN-LSTM, and ConvLSTM models. (3) Randomly select 80% of the samples in the concept model sample library as the training set and 20% of the samples as the validation set. Use the training set data to train the deep learning model. After training, use the validation set to test the model recognition effect. If the recognition accuracy is higher than 90%, the model training effect is considered to be good and can be used for rapid recognition of stored patterns. If the recognition accuracy is lower than 90%, the model hyperparameters need to be adjusted and retrained until the recognition accuracy is higher than 90%. The adjustable hyperparameters of the model include batch size, learning rate, number of training steps, number of convolution kernels, number of layers, number of recurrent neural units, and type of optimizer.

2. The method for automatic identification of carbonate oil reservoir patterns according to claim 1, characterized in that: The establishment of typical reservoir models in step S1 specifically involves classifying fracture / fracture-vuggy carbonate oil reservoirs into basic reservoir models based on the spatial characteristics of currently developed fracture / fracture-vuggy carbonate oil reservoirs.

3. The method for automatic identification of carbonate oil reservoir patterns according to claim 2, characterized in that: The basic reservoir models include fracture type, cave type, fracture + cave type, bottom water fracture type, bottom water cave type, and bottom water fracture + cave type.

4. The method for automatic identification of carbonate oil reservoir patterns according to claim 1, characterized in that: The numerical simulators include a black oil numerical simulator, a component numerical simulator, and a thermal recovery numerical simulator.

Images