Prediction method of reservoir saturation field in high water-cut period using electromagnetic method based on machine learning

Abstract

The present invention relates to the field of oil and gas production engineering technology, and more specifically, to a machine learning-based electromagnetic method for predicting the saturation field of oil reservoirs during the high-water-cut period. The method combines a machine learning algorithm with well-surface electromagnetic saturation field testing. Based on the single-layer saturation field distribution obtained by well-surface electromagnetic testing, it introduces reservoir flow correlation parameters and utilizes various regression algorithms in machine learning to predict the saturation distribution of the entire vertical reservoir, ultimately generating accurate prediction results for the fluid flow in high-water-cut oil reservoirs.

Description

Technical Field

[0001] The present invention relates to the technical field of oil production engineering in petroleum and natural gas, and in particular to a method for predicting the saturation field of an oil reservoir in a high water-cut period using an electromagnetic method based on machine learning. Background Art

[0002] As a non-seismic geophysical technology, well-surface electromagnetic method has been applied to the monitoring of reservoir fluid parameters in recent years. Studies have shown that resistivity is sensitive to parameters such as saturation and permeability of oil and gas reservoirs. The detection of oil, gas and water boundaries and saturation distribution by electromagnetic technology has the advantages of being economical and effective, and having a large detection range. However, for reservoirs with a large number of vertical layers, especially under the influence of factors such as construction costs under today's low oil prices, well-surface electromagnetic technology often only monitors the main target layers. This also makes it difficult to accurately identify the remaining oil in high-water-cut reservoirs, and at the same time restricts the deployment of wells in the later stages of reservoir development. As a currently popular cutting-edge technology, machine learning models are increasingly being used in various industries and are considered to be an effective means to solve complex indicator prediction problems. Lin Botao systematically analyzed the professional applications of artificial intelligence technology from upstream exploration, development, and production to downstream operations, sales, and investment, and the specific requirements put forward for practitioners.

[0003] Focusing on the field of oil exploration and development, WANG used parameters such as well inclination, azimuth, well depth, fluid properties, fracture attributes, and production cycle to establish a sample library and applied neural network algorithms to optimize horizontal well fracturing indicators. Wang Zhizhang selected 12 rock physical parameters sensitive to volcanic rock lithology and pore structure as classification features and used six algorithms: decision tree, support vector machine, logistic regression, AdaBoost-decision tree, AdaBoost-support vector machine, and AdaBoost-logistic regression to identify and classify volcanic rock lithology. Jung analyzed the similarity of data sets using cluster analysis and used random forest (RF), gradient boosting tree (GBM), and support vector machine (SVM) supervised learning models to predict the production capacity of shale reservoirs. Wang Zhiguo used principal component analysis (PCA) for multidimensional data analysis, red, green, and blue (RGB) fusion technology to create sedimentary phase maps, and fuzzy self-organizing maps (FSOM) for unsupervised classification to generate seismic phase classification results for reservoirs.

[0004] Chinese patent application CN108241785A discloses a method for finely characterizing the saturation field of heterogeneous reservoirs, comprising the following steps: 1) establishing a saturation formula; 2) substituting the saturation formula obtained in step 1) into the oilfield porosity and permeability models, taking into account the oil column height distribution, and constructing a saturation field model using geological modeling software. Based on capillary pressure data from actual core samples of heterogeneous reservoirs, this invention comprehensively considers geological reservoir characteristics and conducts an in-depth analysis of the relationship between oil saturation, capillary pressure, and reservoir quality coefficient. First, an empirical saturation formula is established, and then a geological model of the saturation field is constructed, thereby achieving a rational, quantitative, and fine characterization of the saturation field.

[0005] Chinese patent application CN111832227A discloses a method for determining shale gas saturation based on deep learning, the method comprising: obtaining the fracture permeability and target time of a target formation; converting the fracture permeability into an equivalent matrix permeability to obtain equivalent permeability data of the target formation; determining a target equivalent permeability field map of the target formation based on the equivalent permeability data; determining a saturation field map of the target formation at the target time based on the target equivalent permeability field map and the target time using a target deep convolutional decoding network; wherein the target deep convolutional decoding network is used to use an encoder to reduce the resolution of the target equivalent permeability field map to obtain multiple feature maps, and use a decoder to calculate based on the multiple feature maps to obtain the saturation field map at the target time.

[0006] Chinese invention patent CN107044277B discloses a method for evaluating the production increase potential of horizontal wells in low-permeability heterogeneous reservoirs through refracturing. The method comprises the following steps: 1) establishing a fracture propagation model using Meyer software and inverting the parameters of the initial hydraulic fracture; 2) establishing a heterogeneous geological model of the reservoir using Eclipse, a reservoir numerical simulation software, and implanting the parameters of the initial hydraulic fracture into the heterogeneous geological model to perform production dynamic history fitting to obtain the residual oil saturation field and formation pressure field distribution; 3) quantitatively evaluating the production increase potential of the horizontal well through refracturing based on the residual oil saturation field and formation pressure field distribution, classifying the initial hydraulic fractures, and proposing targeted refracturing methods.

[0007] However, there are few reports on the prediction of remaining oil saturation field using machine learning, especially the saturation field prediction method that can reflect the identification characteristics of well-surface electromagnetic method. Summary of the Invention

[0008] The main purpose of the present invention is to provide a machine learning-based electromagnetic method for predicting the saturation field of high-water-cut reservoirs. The method combines a machine learning algorithm with the saturation field tested by well-surface electromagnetic method. Based on the single-layer saturation field distribution obtained by well-surface electromagnetic method, reservoir flow correlation parameters are introduced, and different regression algorithms in machine learning are used to predict the saturation distribution of the entire vertical reservoir area, ultimately forming an accurate prediction result for the fluid in the high-water-cut reservoir.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] The present invention provides a method for predicting the saturation field of an oil reservoir in a high water-cut period using electromagnetic method based on machine learning, comprising the following steps: measuring the saturation field using well-surface electromagnetic method; establishing a machine learning prediction model; screening machine learning prediction indicators; and predicting the saturation field.

[0011] Furthermore, the borehole-to-surface electromagnetic method for saturation field determination involves deploying two excitation points at the top and bottom for excitation. After processing the acquired signal to remove noise and eliminate casing effects, the oil-gas-water interface is determined by the spatial distribution of local polarization imbalances. The resistivity field distribution of the reservoir is obtained through field-constrained inversion of well logging and seismic data, and the remaining oil saturation field distribution of the reservoir is finally determined using the Archie formula. The principle of predicting reservoir fluid distribution using borehole-to-surface electromagnetics is to transmit a series of currents of different frequencies through excitation electrodes located in the well. Receiving electrodes are deployed along the surface along the survey line. Through differential processing, the effects of reservoir heterogeneity are eliminated, and the resistivity distribution of oil and gas electrical anomalies is obtained, enabling monitoring of the remaining oil and water saturation distribution within the work area.

[0012] Furthermore, the saturation field prediction of the strong regressor is achieved through the AdaBoost algorithm.

[0013] Furthermore, the specific implementation process of the algorithm is:

[0014] (1) First, a base regression algorithm and training set are given: {(x1, y1), ..., (x N ,y N )}

[0015] where x1, x N ∈X,y1、y N ∈Y, X and Y represent the feature data set respectively;

[0016] (2) Initialize the weight vector of the training data:

[0017]

[0018] in is the weight vector of the first basis regressor, and N is the Nth feature parameter set;

[0019] (3) For the n base regressors set by AdaBoost, when m = 1, 2, ..., n, the weight vector w is used (m) The training data set is used to learn and obtain the regression error rate of the training data set and the discrimination result of the base regressor at the sample:

[0020]

[0021]

[0022] Where, e m is the error rate of the previous step basis regressor; h m (x i ) is the discrimination result of the base regressor; a m is the weight coefficient of the base regressor;

[0023] Update the weight vector distribution of the sample set:

[0024]

[0025] (4) Construct the final strong regressor:

[0026]

[0027] Where C k is the base regressor obtained after the k-th training, and H(x) is the final regressor.

[0028] Furthermore, during the machine learning prediction index screening process, five parameters were selected as geological factors controlling the distribution of remaining oil: formation thickness, permeability, porosity, oil-bearing boundary, and formation dip. The main factors affecting the saturation field distribution can be divided into two categories: geological factors and reservoir factors. Geological factors include complex reservoir lithofacies, strong physical heterogeneity, and large thickness variations, which are the main causes of residual oil dispersion. Reservoir microstructural zones are also important factors controlling the enrichment of residual oil. Therefore, formation thickness, permeability, porosity, oil-bearing boundary, and formation dip were selected as geological factors controlling the distribution of remaining oil.

[0029] Furthermore, during the machine learning prediction index screening process, cumulative water flow, instantaneous water flow, and days of fluid flow were used as dynamic indicators for measuring residual oil saturation. From a development dynamics perspective, the amount of water flow in different regions determines the extent of reservoir water injection flushing. Furthermore, the cumulative time that fluids participate in flow in different regions throughout the development process is also a key indicator of saturation changes. Some dead oil zones, where fluids do not participate in flow, retain their original oil saturation. After comprehensive consideration, cumulative water flow, instantaneous water flow, and days of fluid flow are used as dynamic indicators for measuring residual oil saturation.

[0030] Furthermore, in the process of establishing the characteristic data set, 8 indicators, including 5 static parameters and 3 dynamic parameters, were screened to establish the saturation field prediction characteristic data set.

[0031] Furthermore, in the saturation field prediction process, the indicator prediction data set is constructed using dynamic and static indicator data points, and the prediction model is established using the AdaBoost machine learning algorithm. The fit of the machine learning model is evaluated, and the model hyperparameters are optimized using grid search to finally obtain the optimized model and prediction results.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] The method combines machine learning algorithms with saturation field testing using borehole electromagnetics. Based on the single-layer saturation field distribution obtained using borehole electromagnetics, it introduces reservoir flow correlation parameters and uses various regression algorithms in machine learning to predict the saturation distribution of the entire vertical reservoir. This results in more accurate predictions of the fluid flow in the final high-water-cut period. The method can accurately, quickly, and effectively quantitatively predict the saturation field distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0035] Figure 1 This is a residual oil saturation map obtained by using the well-surface electromagnetic method in a specific embodiment of the present invention;

[0036] Figure 2 This is a distribution diagram of static indicator data points according to a specific embodiment of the present invention: a is the permeability data point distribution, b is the porosity data point distribution, c is the oil-bearing boundary data point distribution, d is the formation dip data point distribution, and e is the formation thickness data point distribution.

[0037] Figure 3This is a dynamic indicator data point distribution diagram of a specific embodiment of the present invention: a is the cumulative water flow data point distribution diagram, b is the instantaneous water flow data point distribution diagram, and c is the fluid flow days data point distribution diagram.

[0038] Figure 4 This is a diagram of the Ng2-3-1 sublayer saturation prediction model described in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations and / or combinations thereof.

[0041] In order to enable those skilled in the art to more clearly understand the technical solution of the present invention, the technical solution of the present invention will be described in detail below with reference to specific embodiments.

[0042] Example 1

[0043] The electromagnetic method for predicting reservoir saturation field in the high water-cut period based on machine learning includes the following steps:

[0044] Step 1. Determination of saturation field using the well-surface electromagnetic method: Two excitation points are deployed at the top and bottom for excitation. After denoising the acquired signal and eliminating the influence of casing, the oil-gas-water interface is determined through the spatial distribution of local polarization imbalance. The resistivity field distribution of the reservoir is obtained through field-constrained inversion of logging and seismic data. Finally, the Archie formula is used to determine the remaining oil saturation field distribution of the reservoir.

[0045] Step 2. Machine learning prediction model establishment:

[0046] The saturation field prediction of the strong regressor is achieved through the AdaBoost algorithm. The specific implementation process of the algorithm is as follows:

[0047] (1) First, a base regression algorithm and training set are given: {(x1, y1), ..., (x N ,y N )}

[0048] where x1, x N ∈X,y1、yN ∈Y, X and Y represent the feature data sets respectively.

[0049] (2) Initialize the weight vector of the training data:

[0050]

[0051] in is the weight vector of the first basis regressor, and N is the Nth feature parameter set.

[0052] (3) For the n base regressors set by AdaBoost, when m = 1, 2, ..., n, the weight vector w is used (m) The training data set is used to learn and obtain the regression error rate of the training data set and the discrimination result of the base regressor at the sample:

[0053]

[0054]

[0055] Where, e m is the error rate of the previous step basis regressor; h m (x i ) is the discrimination result of the base regressor; a m is the weight coefficient of the base regressor;

[0056] Update the weight vector distribution of the sample set:

[0057]

[0058] (4) Construct the final strong regressor:

[0059]

[0060] Where C k is the base regressor obtained after the k-th training, and H(x) is the final regressor.

[0061] Step 3. Machine Learning Prediction Indicator Screening: Five parameters, namely formation thickness, permeability, porosity, oil-bearing boundary, and formation dip, were selected as geological factors controlling the distribution of remaining oil. Cumulative water flow, instantaneous water flow, and the number of days of fluid flow were used as dynamic indicators of remaining oil saturation. During the process of establishing the characteristic dataset, eight indicators, including the five static parameters and three dynamic parameters, were selected to create the saturation field prediction characteristic dataset.

[0062] Step 4. Saturation field prediction: Use dynamic and static indicator data points to construct an indicator prediction data set, use the mentioned AdaBoost machine learning algorithm to establish a prediction model, evaluate the fit of the machine learning model, and use grid search to optimize the model hyperparameters to finally obtain the optimized model and prediction results.

[0063] Example 2

[0064] The method described in Example 1 is used to predict the saturation field of an actual block. The saturation field detection map of the Nm3-4-2 layer of the block is obtained by the well-surface electromagnetic method, as shown in FIG. Figure 1 As shown. Through data screening, an 8-parameter prediction indicator data set is established, as shown Figure 2 、 3 The 8 parameters are predicted by the AdaBoost machine learning model, and the data connection between the 8 parameters and the Nm3-4-2 sub-layer electromagnetic detection saturation model is established to form the Ng2-3-1 sub-layer saturation prediction result, as shown in the figure. Figure 4 shown.

[0065] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. The electromagnetic method for predicting reservoir saturation field in high water-cut period based on machine learning is characterized by: The following steps are involved: Saturation field determination using well-surface electromagnetic method; Machine learning prediction model establishment; machine learning prediction indicator screening; saturation field prediction; The borehole-to-ground electromagnetic method for determining saturation fields uses two excitation points deployed at the top and bottom. After de-noising the acquired signals and eliminating casing effects, the oil-gas-water interface is determined by the spatial distribution of local polarization imbalances. The resistivity field distribution of the reservoir is then obtained through field-constrained inversion of logging and seismic data. Finally, the Archie formula is used to determine the remaining oil saturation field distribution of the reservoir. The saturation field prediction of the strong regressor is achieved through the AdaBoost algorithm; The specific implementation process of the algorithm is: (1) First, a base regression algorithm and training set are given: {(x1, y1), ..., (x N ,y N )}where x1, x N ∈X,y1、y N ∈Y, X and Y represent the feature data set respectively; (2) Initialize the weight vector of the training data: in i=1,2…,N; is the weight vector of the first basis regressor, and N is the Nth feature parameter set; (3) For the n base regressors set by AdaBoost, when m = 1, 2, ..., n, the weight vector w is used (m) The training data set is used to learn and obtain the regression error rate of the training data set and the discrimination result of the base regressor at the sample: Where, e m is the error rate of the previous step basis regressor; h m (x i ) is the discrimination result of the base regressor; a m is the weight coefficient of the base regressor; Update the weight vector distribution of the sample set: (4) Construct the final strong regressor: Where C k is the base regressor obtained after the k-th training, and H(x) is the final regressor.

2. The method according to claim 1, characterized in that In the process of machine learning prediction indicator screening, five static indicators, namely formation thickness, permeability, porosity, oil-bearing boundary and formation dip, were screened as geological factors controlling the distribution of remaining oil; in the process of machine learning prediction indicator screening, cumulative water flow, instantaneous water flow and fluid flow days were used as dynamic indicators to measure the remaining oil saturation.

3. The method according to claim 2, characterized in that In the process of establishing the characteristic data set, 8 indicators, including 5 static parameters and 3 dynamic indicators, were selected to establish the saturation field prediction characteristic data set.

4. The method according to claim 3, characterized in that In the saturation field prediction process, the indicator prediction data set is constructed using dynamic and static indicator data points, and the prediction model is established using the AdaBoost machine learning algorithm. The fit of the machine learning model is evaluated, and the model hyperparameters are optimized using grid search to finally obtain the optimized model and prediction results.

Images