A method for predicting breakthrough time of water flooding in carbonate reservoirs based on surrogate model

Abstract

The present invention relates to a method for predicting the breakthrough time of water drive in carbonate oil reservoirs based on an agent model, comprising the following steps: S1: selecting and disclosing a data set of actual production history of a certain oil reservoir; S2: dividing the natural water drive of the oil reservoir into four types of water drive according to the production dynamics of the oil reservoir; S3: using a finite element conceptual model to simulate and calculate the relevant parameters of the oil reservoir to obtain a numerical simulation sample data set; S4: constructing four types of water drive breakthrough time prediction sub-models; S5: constructing a training set of four types of water drive breakthrough time prediction sub-models; S6: training each type of water drive breakthrough time prediction sub-model to finally obtain a trained prediction model; S7: correcting the trained prediction model to obtain a final carbonate oil reservoir water drive breakthrough time prediction model. The use of this technical solution can solve the shortcomings of the existing methods for predicting the breakthrough time of water drive in oil reservoirs, such as long time consumption, poor adaptability and pertinence, and can effectively guide the specific development work of actual oil wells.

Description

Technical Field

[0001] The invention relates to the field of prediction of water flooding breakthrough time in oil reservoirs, and in particular to a method for predicting water flooding breakthrough time in carbonate oil reservoirs based on an agent model. Background Art

[0002] Carbonate oil and gas reservoirs rank first in the world in terms of oil and gas reserves and production, and are a research object of great concern at home and abroad. my country has a large area of ​​carbonate formations, and there is huge room for innovation in the geological research of carbonate oil and gas reservoirs. The proven reserves of some carbonate oil reservoirs are very high, but due to the complex distribution relationship between underground fractures and matrix layers, the reservoir physical properties are extremely different, and the exploitation is extremely difficult. At present, common development problems of fracture-pore carbonate reservoirs include the inability to control the rapid increase of water content, the early water breakthrough of oil wells, which causes the production of some oil wells to decline too early and too fast, and the development effect is poor. Water flooding of oil wells is the dominant factor leading to the decline in production. It costs a lot of money to use conventional numerical simulation methods to simulate the influence of fracture-matrix layer distribution relationship, reservoir characteristic parameters, and fracture attributes (length, width, opening, and direction) on the time of water breakthrough of oil wells.

[0003] At present, there are mainly the following methods for predicting the breakthrough time of natural water flooding in oil reservoirs: (1) Indoor macroscopic three-dimensional physical simulation method. (2) Numerical simulation prediction method for natural water flooding breakthrough time of oil reservoirs. Based on the effective reservoir body carving and three-dimensional physical property geological model of the oil reservoir, according to the equivalent seepage medium theory, the oil-water two-phase mathematical model of the oil reservoir is differentially solved, the oil-water movement law at any time inside the oil reservoir is simulated, and the dynamic evolution process of the water cone formation, uplift and breakthrough of the bottom water in the oil reservoir is simulated, so as to predict the breakthrough time of the bottom water in the oil well production layer. Compared with the simplified analytical method, the advantage of this method is that it can consider the influence of actual parameters such as reservoir geological conditions and oil and water properties. (3) See water map method. This method is based on the development law of sandstone bottom water reservoirs, conventional sandstone bottom water reservoir water cone prediction, division of oil well development stages, establishment of the corresponding relationship between cumulative liquid production and wellhead indicators, and determination of the production stage based on the current cumulative liquid production to determine the breakthrough time of natural water flooding. (4) Historical data analysis and statistical method. The above-mentioned prediction method of natural water flooding breakthrough time of carbonate rock salt reservoirs includes reservoir development engineering, physical modeling, numerical modeling and statistical analogy methods, which can theoretically analyze the influencing factors of natural water flooding breakthrough time of reservoirs and predict the natural water flooding breakthrough time of reservoirs. Unfortunately, the above-mentioned technology is not very applicable and targeted in the actual development of reservoirs, and it is difficult to effectively guide the specific development of oil wells.

[0004] The Chinese patent number CN106150486B, "Method for Identifying Water Breakthrough Laws in Fractured Reservoirs", proposes a method for determining the water flooding time based on the range of the angle Φ formed by different water-containing stages. The patent does not involve a method for predicting the water breakthrough time of oil wells in multi-layer carbonate reservoirs. The water breakthrough time here refers to the water flooding breakthrough time.

[0005] The Chinese patent "Weathering crust karst fracture-cave oil reservoir water breakthrough time prediction method and equipment" with patent number CN110020495A proposes a weathering crust karst fracture-cave oil reservoir water breakthrough time prediction method and equipment based on data mining algorithm, by determining the karst background and reservoir type of fracture-cave oil reservoir, screening effective samples, establishing a corresponding relationship between the fracture-cave oil reservoir drive stage and oil pressure characteristics, and establishing a neural network prediction model that takes into account geological reserves, bottom water development, reservoir type, whether it is a heavy oil well, depth into the mountain, whether it is acid fracturing, production degree, whether it is sporadic water breakthrough, oil well energy level, and instantaneous production. This patent does not involve the influence of the fracture-matrix layer distribution relationship on the water breakthrough time. The water breakthrough time here refers to the water drive breakthrough time. Summary of the invention

[0006] In view of the above problems existing in the prior art, the technical problem to be solved by the present invention is: how to accurately predict the breakthrough time of natural water flooding in oil reservoirs.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solution: a method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model, comprising the following steps:

[0008] A method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model comprises the following steps:

[0009] S100: selecting a public actual production history data set of a certain oil reservoir, wherein the actual production history data set includes oil reservoir production dynamics, oil reservoir recovery rate, oil reservoir reservoir parameters and oil reservoir fluid parameters;

[0010] S200: dividing the natural water drive of the oil reservoir into four water drive types according to the production performance of the oil reservoir, and dividing all the data in the actual production history data set into four types of original data corresponding to the four water drive types according to the production performance of the oil reservoir, wherein the four water drive types include: type I, type II, type III and type IV;

[0011] S300: Use the finite element conceptual model of oil-water two-phase flow in fracture-pore carbonate reservoirs to simulate and calculate the reservoir parameters and reservoir fluid parameters to obtain a numerical simulation sample data set;

[0012] S400: constructing a carbonate reservoir water flooding breakthrough time prediction model, wherein the carbonate reservoir water flooding breakthrough time prediction model includes four water flooding breakthrough time prediction sub-models, the four water flooding breakthrough time prediction sub-models correspond to the four water flooding types one by one, and each water flooding breakthrough time prediction sub-model is a multi-layer perceptron neural network;

[0013] S500: constructing training sets of four water flooding breakthrough time prediction sub-models respectively, combining a class of original data and a numerical simulation sample data set to obtain a training set of the water flooding breakthrough time prediction sub-model corresponding to the class of original data;

[0014] S600: setting the maximum number of training iterations, taking the training set corresponding to each water flooding breakthrough time prediction sub-model as input, using the training module in MATLAB to train each water flooding breakthrough time prediction sub-model, stopping the training when the training reaches the maximum number of iterations, and obtaining four trained water flooding breakthrough time prediction sub-models, that is, obtaining a trained carbonate reservoir water flooding breakthrough time prediction model;

[0015] S700: using the oil reservoir recovery factor to correct the trained carbonate reservoir water flooding breakthrough time prediction model to obtain a final carbonate reservoir water flooding breakthrough time prediction model;

[0016] S800: Select a fracture-pore carbonate reservoir to be predicted, and determine the water drive type of the reservoir by using the reservoir production performance of the fracture-pore carbonate reservoir to be predicted;

[0017] S900: taking the reservoir parameters and reservoir fluid parameters of the fracture-pore carbonate reservoir to be predicted as inputs of a final carbonate reservoir water flooding breakthrough time prediction model, and outputting a water flooding breakthrough time prediction result of the fracture-pore carbonate reservoir to be predicted.

[0018] Preferably, the reservoir production dynamics in S100 include oil well pressure, water content, oil-water front, and reservoir recovery rate.

[0019] Oil well pressure and water content are data that are easy to collect in oil fields. The oil-water front is the calculable parameter that most directly reflects the natural water drive breakthrough rate. There are large differences in these three parameters for different reservoir types. Using oil well pressure, water content and oil-water front, the water breakthrough types of reservoirs can be divided into four categories in a more detailed manner.

[0020] Preferably, the oil reservoir parameters in S100 include fracture-matrix layer distribution relationship characteristic parameters, reservoir characteristic parameters and fracture attribute parameters.

[0021] Compared with other water flooding breakthrough time prediction methods, this method takes into account the influence of fracture-matrix layer distribution relationship and fracture properties on water flooding breakthrough time. In the production process, fractures are often seepage channels. The distribution pattern and distribution law of fractures have a great influence on water breakthrough in oil reservoirs and water breakthrough in oil wells. Therefore, the fracture-matrix distribution relationship and fracture properties need to be considered.

[0022] Preferably, the characteristic parameters of the fracture-matrix layer distribution relationship include the number of fracture matrix layers, the number of bedrock layers, and layer thickness; the reservoir characteristic parameters include porosity and permeability; and the fracture attribute parameters include fracture aperture and fracture density.

[0023] Compared with other water flooding breakthrough time prediction methods, this method combines more factors and believes that most oil reservoirs are complex superimposed reservoirs, and the superposition pattern of fractures and matrix is ​​complex, which greatly affects the prediction of oil reservoir water flooding breakthrough time. Therefore, the qualitative analysis of fracture-matrix related parameters is particularly important. This method is based on the MLP model, fully considers the reservoir superposition and reservoir body physical properties, and has high accuracy.

[0024] Preferably, the reservoir fluid parameters in S100 include oil-water viscosity, density and relative permeability data.

[0025] Oil-water viscosity, density and relative permeability can accurately reflect the flow state of underground fluids under reservoir conditions.

[0026] Preferably, the specific method for obtaining the numerical simulation sample data set in S300 is as follows:

[0027] S310: Establishment of a conceptual finite element model of oil-water two-phase flow in fracture-pore carbonate reservoirs using COMSOL software;

[0028] S320: Calculate the mass conservation equation of oil and water phases in the matrix of fracture-pore carbonate reservoirs. The calculation expression is as follows:

[0029]

[0030] Among them, φ m is the matrix porosity; S mo is the matrix oil saturation; S mw is the matrix water saturation; t is the time; v mo is the velocity vector of the matrix oil phase, in m·s -1 ;v mw is the matrix water phase velocity vector, unit: m·s -1 ;q mo is the source and sink term of the matrix oil phase; q mw is the source and sink term of the matrix water phase;

[0031] Among them, v mo 、vmw The equation of motion of the oil-water two phases in the matrix is ​​as follows:

[0032]

[0033] Where K is the permeability tensor; k ro is the relative permeability of the oil phase; k rw is the relative permeability of water phase; μ o is the viscosity of the oil phase; μ w is the viscosity of the water phase; p mo is the oil phase fluid pressure in the matrix; p mw is the water phase fluid pressure in the matrix;

[0034] S330: Substituting formula (2) into formula (1) obtains the control equation of oil-water two phases in the matrix of fracture-pore carbonate reservoirs. The calculation expression is as follows:

[0035]

[0036] S340: Calculate the mass conservation equation and the motion equation of the oil-water two-phase in the fracture. The calculation expressions are as follows:

[0037]

[0038]

[0039] In the formula, φ f is the fracture porosity; p fo is the oil phase fluid pressure in the fracture; p fw is the water phase fluid pressure in the fracture; S fo is the fracture oil saturation; S fw is the water saturation of the fracture; t is the time; v fo is the fracture oil phase velocity vector, m·s -1 ;v fw is the fracture water phase velocity vector, m·s -1 ;q fo is the source and sink term of fracture oil phase; q fw is the source and sink term of fracture water phase; p fo is the oil phase fluid pressure in the fracture; p fw is the water phase fluid pressure in the fracture;

[0040] S350: Substituting formula (5) into formula (4) obtains the control equation of oil-water phase in fractures of fracture-pore carbonate reservoirs. The calculation expression is as follows:

[0041]

[0042] S360: The comprehensive flow equation of water phase and oil phase in fracture-pore carbonate reservoir is obtained from formula (3) and formula (6), which is as follows:

[0043]

[0044] Among them, S w Represents the water saturation of the reservoir, S o represents the oil saturation of the reservoir, q w represents the source and sink of water phase in the reservoir, q o represents the oil source and sink term of the reservoir;

[0045] S370: Calculate the pressure equation and saturation matrix equation of the oil-water phase in the fracture-pore carbonate reservoir. The expressions are as follows:

[0046]

[0047] Among them, P o represents the oil phase fluid pressure of the reservoir, φ represents the overall porosity of the reservoir, and p c ' indicates p c The first derivative of c Indicates capillary force, unit Pa, p c ' and p c The expression is as follows:

[0048] p c (S w )=p o -p w ; (9)

[0049]

[0050] Among them, p o represents the oil phase fluid pressure of the reservoir, p w It represents the water phase fluid pressure of the reservoir;

[0051] S380: Use formula (8) to obtain the saturation field of fracture-pore carbonate reservoirs at different times. When the water saturation of the oil wells in the reservoir is not 0, record the time as the water breakthrough time of the oil wells under the reservoir conditions, and use the collection of all water breakthrough time data as the numerical simulation sample data set.

[0052] Preferably, the specific steps of correcting the trained carbonate reservoir water flooding breakthrough time prediction model in S700 are as follows:

[0053] S710: Calculate the trained carbonate reservoir water flooding breakthrough time prediction model using COMSOL software to obtain the predicted recovery factor corresponding to the model;

[0054] S720: Set the characterization deviation value Z(e) of the reservoir using a probability function;

[0055] S720: Calculate the difference between the predicted recovery factor and the actual recovery factor. If the difference is less than Z(e), then obtain the final prediction model for the water drive breakthrough time of the carbonate reservoir; if the difference is greater than Z(e), then use a random number function to correct the trained prediction model for the water drive breakthrough time of the carbonate reservoir. The specific steps are as follows:

[0056] S721: Use the srand((unsigned)time(NULL)) function to process the recovery factor variable in the trained prediction model for the water drive breakthrough time of the carbonate reservoir to obtain different recovery factor seed values; where unsigned is the variable type, and here the recovery factor variable is selected; time(NULL) is the time value;

[0057] S721: Use the rand() function to "randomize" the different recovery factor seed values. The expression is as follows:

[0058] X i ′=kX i ; (11)

[0059] k = rand() / 16383.5; (12)

[0060] where k represents the correction coefficient, X i ′ represents the i-th corrected recovery factor, X i represents the i-th recovery factor seed value, i = 1,..., n;

[0061] S722: Calculate the deviation value S. The specific expression is as follows:

[0062]

[0063] where, represents the average recovery factor of all prediction results, and n represents the number of predictions;

[0064] S723: When S < Z(e), obtain the final prediction model for the water drive breakthrough time of the carbonate reservoir; otherwise, return to S721.

[0065] Compared with the conventional water flooding breakthrough time prediction method, this method uses COMSOL commercial software to build a model, which fully realizes the construction of a multi-fracture-bedrock distribution model; based on the proxy model method, the water flooding breakthrough time prediction and model feature link analysis are carried out under the conditions of typical characteristic reservoirs considering fracture-bedrock distribution patterns, physical properties and other conditions, and the corresponding typical characteristic proxy model is generated through training and learning. Based on the proxy model and the real physical properties of the reservoir, the water breakthrough time is accurately predicted, thereby improving the prediction accuracy; then the results are pre-judged based on the deviation value theory, realizing high-precision water breakthrough time prediction simulation at the reservoir scale.

[0066] Compared with the prior art, the present invention has at least the following advantages:

[0067] 1. The current existing methods for predicting the breakthrough time of water flooding in oil reservoirs consume a lot of computing costs. The use of the technical solution of the present invention can greatly reduce the operation time and improve the prediction accuracy. This solution divides the water flooding type through the existing measured data of the oil reservoir, and then uses the existing data simulation technology to process the actual data to obtain a data simulation sample set; then the data corresponding to different water flooding types are combined with the numerical simulation sample set as the training set for the overall model prediction; the MLP multi-layer perceptron neural network is used as the baseline model to construct a fracture-pore carbonate reservoir water flooding breakthrough time prediction model, and the corresponding training set is used for model training, and then the trained model is conditionally verified. When it does not meet the verification conditions, further model correction is performed to obtain the final prediction model. The model correction based on the deviation value theory is, in principle, to pre-judge the results and achieve high-precision prediction at the reservoir scale.

[0068] 2. This scheme takes into account for the first time the influence of the fracture-bedrock layer distribution relationship on the natural water drive breakthrough time of the reservoir. In step 3, the characteristic parameters of the fracture-bedrock layer distribution relationship, including the number of fracture matrix layers, the number of bedrock layers, and the layer thickness, are taken into consideration and reflected in the Comsol numerical simulation modeling process and the data collection process of the actual oil field as the input data for training the MLP to improve the prediction accuracy of the proxy model.

[0069] 3. Compared with the conventional water drive breakthrough time prediction method, this method is based on COMSOL commercial software for model construction, which fully realizes the construction of multi-fracture-bedrock distribution model; based on the proxy model method, the water drive breakthrough time prediction and model feature link analysis are carried out under the conditions of typical characteristic reservoirs considering fracture-bedrock distribution pattern, physical properties and other conditions, and the corresponding typical characteristic proxy model is generated by training and learning. Based on the proxy model and the real physical properties of the reservoir, the water drive breakthrough time is accurately predicted, and the prediction accuracy is improved; then the results are pre-judged based on the deviation value theory, and the prediction simulation of water drive breakthrough time at reservoir scale with high precision is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0071] Figure 2 Schematic diagram of the MLP framework.

[0072] Figure 3 Create a flow chart for the digital model.

[0073] Figure 4 This is a two-dimensional diagram of the Comsol model.

[0074] Figure 5 This is the oil-water saturation diagram of the Comsol model.

[0075] Figure 6 Correction flow chart for recovery factor. DETAILED DESCRIPTION

[0076] The present invention is described in further detail below.

[0077] The invention analyzes the oil-water two-phase flow characteristics of underground fracture-porous carbonate reservoirs, simplifies and develops the general water breakthrough time prediction model, constructs a fracture-porous carbonate reservoir proxy model, uses a finite element model to simulate orthogonal experiments to obtain numerical simulation samples, and then solves the parameters in the proxy model. Taking into account the influence of the edge and bottom water development degree and the production degree on the water breakthrough time, the model is corrected by using the recovery factor to make it more in line with practical application requirements, and a fracture-porous carbonate reservoir water drive breakthrough time prediction model that meets the real-time simulation requirements is obtained. The natural water drive breakthrough time can be simulated and calculated in real time, and the water drive breakthrough time can be predicted in combination with the characteristic parameters of the fracture-porous carbonate reservoir.

[0078] Example 1: Taking the T unit of the Tahe oil reservoir as an example to further illustrate the effectiveness of the technical method, the implementation mode of the present invention is described as follows:

[0079] See also Figure 1-Figure 6 , a method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model, comprising the following steps:

[0080] S01: According to production dynamics, such as oil well pressure, water content, oil-water front, etc., water breakthrough types are divided into four types. Among them, the four water flooding types are described as follows: Type I: The water content rises sharply to 40%-60% within two years after well opening, and the expression is t 1 =A 1 x f +B 1 P+C 1 f w; Type II: There is a period of waterless recovery two years after the well is opened, with a water content of 0, and then the water content rises slowly, and the oil well pressure remains at 70%-90% of the original formation pressure, expressed as t 2 =A 2 P+B 2 x f +C 2 f w ; Type III: There is a period of water-free recovery two years after the well is opened, and then the water content rises sharply to 40%-60%, expressed as t 3 =A 3 x f +B 3 P+C 3 f w ; Type IV: Two years after the well was opened, the oil well pressure slowly dropped to 70%-90% of the original formation pressure, but the water content remained stable at 20%-40%, expressed as t 4 =A 4 P+B 4 x f +C 4 f w . They correspond to the following types of edge and bottom water: edge and bottom water is close to the wellhead and connected by fractures and other advantageous channels; there is a certain water avoidance height and the reservoir fractures are not well developed; there is a certain water avoidance height and the fractures are well developed; edge and bottom water is mainly local sealed water. The actual production data of the reservoir is divided into four types and preprocessed: including outlier processing and data standardization.

[0081] S02: Input parameters include: reservoir parameters, such as fracture-matrix layer distribution relationship characteristic parameters, reservoir characteristic parameters, fracture attribute parameters, including fracture matrix layer number, sandstone layer number, layer thickness, porosity 0.2, permeability 100mD, fracture aperture, fracture density, etc.; fluid parameters, crude oil viscosity 7.75mPa / s, density 0.84g / cm 3 , injection water viscosity 0.48mPa / s, density 1.02g / cm 3 Phase permeability data, etc. It mainly includes two parts: actual production history data and numerical simulation sample data. The numerical simulation sample uses COMSOL software to construct a fracture-pore carbonate reservoir conceptual model. Figure 4 , COMSOL software is the existing technology, the finite element conceptual model of oil-water two-phase flow in fracture-pore carbonate reservoirs is a known and trained existing model, the discrete fracture model is used for fractured reservoirs, the PDE module is used to calculate the saturation of each phase, the lower boundary is the inflow interface, the upper boundary is the outflow interface, and the maximum water saturation at each point on the outflow interface is equal to 1 as the termination condition. The operation time at this time is recorded as the water breakthrough time, which is used as a natural water flooding breakthrough time sample to solve the natural water flooding breakthrough time prediction model. The results are shown in Figure 5 .

[0082] Combining the mass conservation equation and motion equation of the oil-water two-phase in the matrix and in the fracture, ordinary differential equations are derived to obtain the control equations of the oil-water two-phase only about saturation and pressure. The control equations are organized into matrix form to realize the calculation of the oil-water two-phase flow considering discrete fractures in Comsol software.

[0083] S03: In order to realize the construction of natural water flooding breakthrough time prediction model for fracture-porous carbonate reservoirs with high-dimensional variables, the characteristics of oil-water two-phase flow in fracture-porous carbonate reservoirs are analyzed, based on the developed mathematical model, and MLP (multi-layer perceptron) neural network is used. The finite element conceptual model of oil-water two-phase flow in fracture-porous carbonate reservoirs is an existing and trained model; the multi-layer perceptron neural network is an existing technology;

[0084] S04: First, input reservoir physical parameters such as fracture-bedrock layer distribution relationship characteristic parameters, reservoir characteristic parameters, fracture attribute parameters and fluid parameters, oil-water viscosity, density, relative permeability data, etc., and output water content (f w ), oil well pressure (P), oil-water front (x f ), based on the front distance from the well, the pressure maintenance level, and the water content change trend, it is judged that the T unit of the Tahe reservoir belongs to type II; the mapping relationship expression of the corresponding type of water breakthrough time prediction model is solved as t 2 =A 2 P+B 2 x f +C 2 f w .

[0085] S05: Using the actual production data of the reservoir and the water breakthrough time samples of different conceptual models obtained by numerical simulation methods, 80% of them are used as the initial data set for training the proxy model, and 20% are used as the test set. MATLAB is used to solve the mapping relationship expression of the natural water flooding breakthrough time prediction proxy model for fracture-pore carbonate reservoirs constructed in S03. The parameters to be solved mainly include A 2 , B 2 , C 2 ; The training module in MATLAB is the prior art.

[0086] S06: Modify various types of proxy models based on secondary control factors such as recovery factor. For the specific process, see Figure 6As shown, a probability function is used to characterize the deviation value Z(e), which is an allowable error value proposed based on experience. If the deviation between the model recovery factor and the actual reservoir recovery factor is less than Z(e), the proxy model meets the calculation requirements. If not, the proxy model is corrected using a random number function, and then the above determination is repeated until the deviation value requirement is met.

[0087] The random number function mainly uses the srand() function and rand() function in C language. The seed value is changed by the srand() function, usually srand((unsigned)time(NULL)), so that the rand() function can achieve further "randomization" according to different seed values. Because time is constantly changing, the seed value is constantly changing. This "randomization" is that after using the rand() function, it will return corresponding random numbers corresponding to different seed values. The continuous change of the seed makes the returned rand() function value more random. The range of the obtained random number is between 0 and 32767, and the range of k is between 0-2. The programming of the random number function is k=rand() / 16383.5. Finally, it can be known that the water-seeking time is 1.5a, and the agent model deviation value is less than 5%.

[0088] The method disclosed in the present invention deepens the research on fracture-pore carbonate reservoirs. The natural water drive breakthrough time prediction method for fracture-pore carbonate reservoirs based on a neural network model makes up for the shortcomings of the existing natural water drive breakthrough time prediction method, which is time-consuming, not highly applicable and not targeted in the actual development of oil fields, and can effectively guide the specific development of oil wells.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model, characterized in that: The steps include: S100: selecting a public actual production history data set of a certain oil reservoir, wherein the actual production history data set includes oil reservoir production dynamics, oil reservoir recovery rate, oil reservoir reservoir parameters and oil reservoir fluid parameters; The reservoir parameters include fracture-matrix layer distribution relationship characteristic parameters, reservoir characteristic parameters and fracture attribute parameters; S200: dividing the natural water drive of the oil reservoir into four water drive types according to the production performance of the oil reservoir, and dividing all the data in the actual production history data set into four types of original data corresponding to the four water drive types according to the production performance of the oil reservoir, wherein the four water drive types include: type I, type II, type III and type IV; S300: Use the finite element conceptual model of oil-water two-phase flow in fracture-pore carbonate reservoirs to simulate and calculate the reservoir parameters and reservoir fluid parameters to obtain a numerical simulation sample data set; S400: constructing a carbonate reservoir water flooding breakthrough time prediction model, wherein the carbonate reservoir water flooding breakthrough time prediction model includes four water flooding breakthrough time prediction sub-models, the four water flooding breakthrough time prediction sub-models correspond to the four water flooding types one by one, and each water flooding breakthrough time prediction sub-model is a multi-layer perceptron neural network; S500: constructing training sets of four water flooding breakthrough time prediction sub-models respectively, combining a class of original data and a numerical simulation sample data set to obtain a training set of the water flooding breakthrough time prediction sub-model corresponding to the class of original data; S600: setting the maximum number of training iterations, taking the training set corresponding to each water flooding breakthrough time prediction sub-model as input, using the training module in MATLAB to train each water flooding breakthrough time prediction sub-model, stopping the training when the training reaches the maximum number of iterations, and obtaining four trained water flooding breakthrough time prediction sub-models, that is, obtaining a trained carbonate reservoir water flooding breakthrough time prediction model; S700: using the oil reservoir recovery factor to correct the trained carbonate reservoir water flooding breakthrough time prediction model to obtain a final carbonate reservoir water flooding breakthrough time prediction model; S800: selecting a fracture-pore carbonate reservoir to be predicted, and determining the water drive type of the reservoir by using the reservoir production performance of the fracture-pore carbonate reservoir to be predicted; S900: taking the reservoir parameters and reservoir fluid parameters of the fracture-pore carbonate reservoir to be predicted as inputs of a final carbonate reservoir water flooding breakthrough time prediction model, and outputting a water flooding breakthrough time prediction result of the fracture-pore carbonate reservoir to be predicted.

2. A method for predicting breakthrough time of carbonate reservoir water flooding based on an agent model as claimed in claim 1, characterized in that: The oil reservoir production dynamics in S100 include oil well pressure, water content, oil-water front, and oil reservoir recovery rate.

3. A method for predicting breakthrough time of carbonate reservoir water flooding based on proxy model according to claim 2, characterized in that: The characteristic parameters of the fracture-matrix layer distribution relationship include the number of fracture matrix layers, the number of bedrock layers, and layer thickness; the reservoir characteristic parameters include porosity and permeability; and the fracture attribute parameters include fracture aperture and fracture density.

4. A method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model as claimed in claim 3, characterized in that: The reservoir fluid parameters in S100 include oil-water viscosity, density and relative permeability data.

5. A method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model as claimed in claim 4, characterized in that: The specific method of obtaining the numerical simulation sample data set in S300 is as follows: S310: Establish a finite element conceptual model of two-phase oil-water flow in a fractured-porous carbonate reservoir using COMSOL software; S320: Calculate the mass conservation equations of two-phase oil and water in the matrix of a fractured-porous carbonate reservoir. The calculation expressions are as follows: Among them, φ m is the matrix porosity; S mo is the matrix oil saturation; S mw is the matrix water saturation; t is the time; v mo is the velocity vector of the matrix oil phase, in m·s -1 ;v mw is the matrix water phase velocity vector, unit: m·s -1 ;q mo is the source and sink term of the matrix oil phase; q mw is the source and sink term of the matrix water phase; Among them, v mo 、v mw The equation of motion of the oil-water two phases in the matrix is ​​as follows: Where K is the permeability tensor; k ro is the relative permeability of the oil phase; k rw is the relative permeability of water phase; μ o is the viscosity of the oil phase; μ w is the viscosity of the water phase; p mo is the oil phase fluid pressure in the matrix; p mw is the water phase fluid pressure in the matrix; S330: Substitute formula (2) into formula (1) to obtain the control equations of two-phase oil and water in the matrix of a fractured-porous carbonate reservoir. The calculation expressions are as follows: S340: Calculate the mass conservation equations and the equations of motion of two-phase oil and water in the fractures. The calculation expressions are as follows: In the formula, φ f is the fracture porosity; p fo is the oil phase fluid pressure in the fracture; p fw is the water phase fluid pressure in the fracture; S fo is the fracture oil saturation; S fw is the water saturation of the fracture; t is the time; v fo is the fracture oil phase velocity vector, m·s -1 ;v fw is the fracture water phase velocity vector, m·s -1 ;q fo is the source and sink term of fracture oil phase; q fw is the source and sink term of fracture water phase; p fo is the oil phase fluid pressure in the fracture; p fw is the water phase fluid pressure in the fracture; S350: Substitute formula (5) into formula (4) to obtain the control equations of two-phase oil and water in the fractures of a fractured-porous carbonate reservoir. The calculation expressions are as follows: S360: Obtain the comprehensive flow equations of the water phase and the oil phase in a fractured-porous carbonate reservoir from formula (3) and formula (6). The expressions are as follows: Among them, S w Represents the water saturation of the reservoir, S o represents the oil saturation of the reservoir, q w represents the source and sink of water phase in the reservoir, q o represents the oil source and sink term of the reservoir; S370: Calculate the pressure equation and the saturation matrix equation of two-phase oil and water in a fractured-porous carbonate reservoir. The expressions are as follows: Among them, P o represents the oil phase fluid pressure of the reservoir, φ represents the overall porosity of the reservoir, and p c ' indicates p c The first derivative of c Indicates capillary force, unit Pa, p c ' and p c The expression is as follows: p c (S w )=p o -p w ; (9) Among them, p o represents the oil phase fluid pressure of the reservoir, p w represents the water phase fluid pressure of the reservoir; S380: Use formula (8) to obtain the saturation fields at different times in a fractured-porous carbonate reservoir. When the water cut saturation of the oil well in this reservoir is not 0, record this time as the water breakthrough time of the oil well under the conditions of this reservoir. Take the set of all water breakthrough time data as the numerical simulation sample data set.

6. A method for predicting breakthrough time of water flooding in carbonate reservoirs based on an agent model as claimed in claim 5, characterized in that: The specific steps for modifying the trained prediction model of the water drive breakthrough time of a carbonate reservoir in S700 are as follows: S710: Use COMSOL software to calculate the trained prediction model of the water drive breakthrough time of a carbonate reservoir to obtain the predicted recovery factor corresponding to this model; S720: Set the characterization deviation value Z(e) of the reservoir using a probability function; S720: Calculate the difference between the predicted recovery factor and the actual recovery factor. If the difference is less than Z(e), then obtain the final prediction model of the water drive breakthrough time of a carbonate reservoir; if the difference is greater than Z(e), then use a random number function to modify the trained prediction model of the water drive breakthrough time of a carbonate reservoir. The specific steps are as follows: S721: Use the srand((unsigned)time(NULL)) function to process the recovery factor variable in the trained prediction model of the water drive breakthrough time of a carbonate reservoir to obtain different recovery factor seed values; where, unsigned is the variable type, and here the recovery factor variable is selected; time(NULL) is the time value; S721: Use the rand() function to perform "randomization" processing on different recovery factor seed values. The expression is as follows: X i ′=kX i (11) k = rand() / 16383.5; (12) Where k is the correction factor, X i ′ represents the i-th corrected recovery factor, X i represents the i-th recovery factor seed value, i=1,…,n; S722: Calculate the deviation value S. The specific expression is as follows: in, represents the average recovery factor of all prediction results, and n represents the number of predictions; S723: When S < Z(e), obtain the final prediction model of the water drive breakthrough time of a carbonate reservoir; otherwise, return to S721.

Images