A quantitative prediction method for potential damage in low-permeability reservoirs based on random forest algorithm

Abstract

This invention discloses a quantitative prediction method for potential damage to low-permeability reservoirs based on the random forest algorithm, mainly including the following steps: S1, based on core clay mineral experimental data and well logging data at different depths, establish a profile of the variation of different types of clay mineral content in the entire reservoir section; S2, based on well logging data, calculate porosity and establish a porosity variation profile in the entire reservoir section; S3, based on experimental data and well logging data, establish a porosity-permeability relationship model; S4, using the random forest algorithm, establish the relationship between macroscopic material parameters and microscopic pore throat characteristic values ​​at different depths in the entire reservoir section; S5, through reservoir sensitivity evaluation experiments, test the potential sensitivity damage index of reservoirs at different depths; S6, using the random forest algorithm, train a dataset composed of clay mineral composition data, macroscopic material property data, microscopic pore throat characteristic data, and potential damage index data of deep-water low-permeability reservoirs, and establish a quantitative prediction model for the potential damage index of deep-water low-permeability reservoirs.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas reservoir protection, specifically a method for quantitatively predicting potential damage to low-permeability reservoirs based on the random forest algorithm. Background Technology

[0002] As the complexity of oil and gas field development and production increases, the pollution level of deep-water low-permeability reservoirs is gradually aggravated, posing a significant challenge to the high-precision evaluation of deep-water low-permeability reservoirs.

[0003] Currently, existing reservoir damage characterization methods mainly have three problems:

[0004] (1) Most oilfield reservoir damage assessment data are mainly based on experimental results obtained from some rock samples, and are not universally applicable to the entire oilfield reservoir.

[0005] (2) Most experiments have problems such as environmental pollution, high test costs, only able to conduct experimental analysis on individual cores, and unable to achieve continuous characterization of potential reservoir damage.

[0006] (3) Current research on reservoir potential damage only focuses on mineral composition, neglecting the influence of macroscopic properties (porosity and permeability) on reservoir damage, and also neglecting the influence of microscopic pore-permeability structure (pore-permeability and throat size distribution) on reservoir damage. For low-permeability reservoirs, the pore-permeability structure is complex and the pore-permeability connectivity is poor. The reservoir potential damage is inseparable from the pore structure of the reservoir rock.

[0007] Therefore, this invention proposes a machine learning-based quantitative prediction method for potential damage to deep-water, low-permeability reservoirs, solving the problem that conventional reservoir damage prediction methods are difficult to apply to the prediction of potential damage to deep-water, low-permeability reservoirs. This method enables rapid prediction and analysis of the potential damage level and damage factors of target reservoirs in deep-water areas in oilfields, providing a basis for the design of subsequent drilling operation parameters. Summary of the Invention

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for quantitative prediction of potential damage to low-permeability reservoirs based on the random forest algorithm, comprising the following steps:

[0009] S1. Based on experimental and logging data of clay minerals in deep-water low-permeability reservoirs at different depths, a random forest algorithm was used to train the logging data and clay mineral experimental dataset; a clay mineral prediction model based on logging data points was established to obtain the content variation profile of different types of clay minerals in the entire reservoir section.

[0010] S2. Based on well logging data, calculate porosity and establish a porosity variation profile for the entire reservoir section;

[0011] S3. Based on experimental and logging data, fit the porosity-permeability relationship and establish a porosity and permeability profile for the entire reservoir section;

[0012] S4. Using the random forest algorithm, a dataset of macroscopic physical property parameters and microscopic pore throat characteristics of rocks at different depths in the entire reservoir section is trained, and a prediction model of microscopic pore throat characteristic values ​​based on macroscopic physical property parameters is established, thereby obtaining the variation profile of microscopic pore throat characteristic values ​​in the entire reservoir section.

[0013] S5. Through reservoir sensitivity evaluation experiments, the sensitivity of rocks at different depths in deep-water low-permeability reservoirs was tested, and the potential damage index of rocks at different depths in deep-water low-permeability reservoirs was calculated.

[0014] S6. Based on S1-S4, using the random forest algorithm, train different types of clay mineral composition data, macroscopic physical property data, microscopic pore throat characteristic data and reservoir potential damage index data to establish a prediction model for the potential damage index of deep-water low-permeability reservoirs.

[0015] Furthermore, as a preferred embodiment, in S1,

[0016] The elemental content logging data includes K, TH, U, and K / TH logging data;

[0017] The clay mineral experimental data are the contents of montmorillonite, illite, kaolinite, and chlorite in rocks at different depths of the reservoir, obtained by XRD diffraction.

[0018] Furthermore, as a preferred embodiment, in S2,

[0019] The physical logging data includes density logging data and sonic logging data.

[0020] The porosity calculation involves the following formulas, including the formula for calculating porosity using acoustic transit time:

[0021]

[0022] In the formula, Δt0 represents the acoustic transit time of near-surface mudstone, in μs / m;

[0023] Δt f The transit time of sound waves in pore fluids is expressed in μs / m.

[0024] Δt ma The acoustic transit time value of the mudstone skeleton is expressed in μs / m.

[0025] The formula for calculating porosity using density logging data is as follows:

[0026]

[0027] In the formula, ρ maRepresents skeletal density; ρ f ρ represents fluid density. b This indicates the density of the rock.

[0028] Furthermore, as a preferred embodiment, in S3,

[0029] The experimental data consisted of porosity data of rocks at different depths in low-permeability reservoirs obtained by nuclear magnetic resonance (NMR) testing and permeability data of rocks at different depths in low-permeability reservoirs obtained by pressure decay method.

[0030] The porosity-permeability relationship model is obtained by fitting porosity and permeability test data at different depths;

[0031] The porosity and permeability profile of the entire reservoir section is obtained based on the porosity variation profile of the entire reservoir section and the porosity-permeability relationship model established in S2.

[0032] Furthermore, as a preferred embodiment, in S4,

[0033] The macroscopic physical properties include porosity and permeability;

[0034] The microscopic pore throat features include maximum pore throat radius, average pore throat radius, main channel throat radius, maximum connecting channel radius, maximum pore throat radius, median pore radius, average pore throat radius, median saturation pressure, expulsion pressure, maximum mercury saturation, maximum instrument exit efficiency, sorting coefficient, homogeneity coefficient, and structural coefficient.

[0035] The microscopic pore throat characteristics were obtained through high-pressure mercury intrusion and constant-rate mercury intrusion experiments.

[0036] Furthermore, as a preferred embodiment, in S5,

[0037] The reservoir sensitivity evaluation experiments include water sensitivity test, acid sensitivity test, salt sensitivity test, alkali sensitivity test, rate sensitivity test, stress sensitivity test and water lock test; thereby obtaining the reservoir sensitivity index, which includes water sensitivity index, acid sensitivity index, salt sensitivity index, alkali sensitivity index, rate sensitivity index, stress sensitivity index and water lock index;

[0038] The formula for calculating the sensitivity index is as follows:

[0039]

[0040] In the formula: D vn —The rate of change in rock sample permeability at different flow velocities; K n — Rock permeability after damage, in units of (10) -3 ×μm 2 ); K i —Initial permeability, in units of (10) -3×μm 2 ).

[0041] Furthermore, as a preferred embodiment, in S1 and S6,

[0042] The data training steps of the random forest algorithm include: data acquisition, data preprocessing, construction of training / test sets, prediction, comparison, and establishment of the relationship between the true values ​​and the test values.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1) This method can quickly and quantitatively predict the potential damage level and damage factors of deep-water low-permeability reservoirs based on well logging data, providing a basis for drilling parameter design.

[0045] 2) This method takes into account the influence of rock mineral composition, macroscopic physical properties (porosity and permeability) and microscopic pore throat structure on reservoir damage, and can achieve accurate and comprehensive prediction of the potential damage level and damage factors of deep-water low-permeability reservoirs. Attached Figure Description

[0046] Figure 1 Flowchart of a quantitative prediction method for the potential damage index of deep-water low-permeability reservoirs;

[0047] Figure 2 A comparison of prediction results for different types of clay minerals based on the random forest algorithm;

[0048] Figure 3 A profile showing the variation of different types of clay minerals throughout the entire reservoir section;

[0049] Figure 4 A profile of porosity and permeability changes across the entire reservoir section;

[0050] Figure 5 This is a graph showing the fitted relationship between porosity and permeability in the reservoir section.

[0051] Figure 6 A comparison chart showing the prediction results of microscopic pore throat eigenvalues ​​based on the random forest algorithm;

[0052] Figure 7 This is a profile showing the variation of microscopic pore-throat characteristics across the entire reservoir section;

[0053] Figure 8 A comparison chart of the prediction results for the sensitivity index;

[0054] Figure 9 A profile for predicting changes in the sensitivity index of the entire reservoir section; Detailed Implementation

[0055] Please see Figures 1-5In this embodiment of the invention, a method for quantitatively predicting potential damage to low-permeability reservoirs based on the random forest algorithm includes the following steps:

[0056] S1. Based on experimental and logging data of clay minerals in deep-water low-permeability reservoirs at different depths, a random forest algorithm was used to train the logging data and clay mineral experimental dataset; a clay mineral prediction model based on logging data points was established to obtain the content variation profile of different types of clay minerals in the entire reservoir section.

[0057] S2. Based on well logging data, calculate porosity and establish a porosity variation profile for the entire reservoir section;

[0058] S3. Based on experimental and logging data, fit the porosity-permeability relationship and establish a porosity and permeability profile for the entire reservoir section;

[0059] S4. Using the random forest algorithm, a dataset of macroscopic physical property parameters and microscopic pore throat characteristics of rocks at different depths in the entire reservoir section is trained, and a prediction model of microscopic pore throat characteristic values ​​based on macroscopic physical property parameters is established, thereby obtaining the variation profile of microscopic pore throat characteristic values ​​in the entire reservoir section.

[0060] S5. Through reservoir sensitivity evaluation experiments, the sensitivity of rocks at different depths in deep-water low-permeability reservoirs was tested, and the potential damage index of rocks at different depths in deep-water low-permeability reservoirs was calculated.

[0061] S6. Based on S1-S4, using the random forest algorithm, train different types of clay mineral composition data, macroscopic physical property data, microscopic pore throat characteristic data and reservoir potential damage index data to establish a prediction model for the potential damage index of deep-water low-permeability reservoirs.

[0062] In this embodiment, in S1,

[0063] The elemental content logging data includes K, TH, U, and K / TH logging data;

[0064] The clay mineral experimental data are the contents of montmorillonite, illite, kaolinite, and chlorite in rocks at different depths of the reservoir, obtained by XRD diffraction.

[0065] In a preferred embodiment, in step S2,

[0066] The physical logging data includes density logging data and sonic logging data.

[0067] The porosity calculation involves the following formulas, including the formula for calculating porosity using acoustic transit time:

[0068]

[0069] In the formula, Δt0 represents the acoustic transit time of near-surface mudstone, in μs / m;

[0070] Δt f The transit time of sound waves in pore fluids is expressed in μs / m.

[0071] Δt ma The acoustic transit time value of the mudstone skeleton is expressed in μs / m.

[0072] The formula for calculating porosity using density logging data is as follows:

[0073]

[0074] In the formula, ρ ma Represents skeletal density; ρ f ρ represents fluid density. b This indicates the density of the rock.

[0075] In this embodiment, in step S3,

[0076] The experimental data consisted of porosity data of rocks at different depths in low-permeability reservoirs obtained by nuclear magnetic resonance (NMR) testing and permeability data of rocks at different depths in low-permeability reservoirs obtained by pressure decay method.

[0077] The porosity-permeability relationship model is obtained by fitting porosity and permeability test data at different depths;

[0078] The porosity and permeability profile of the entire reservoir section is obtained based on the porosity variation profile of the entire reservoir section and the porosity-permeability relationship model established in S2.

[0079] In this embodiment, in step S4,

[0080] The macroscopic physical properties include porosity and permeability;

[0081] The microscopic pore throat features include maximum pore throat radius, average pore throat radius, main channel throat radius, maximum connecting channel radius, maximum pore throat radius, median pore radius, average pore throat radius, median saturation pressure, expulsion pressure, maximum mercury saturation, maximum instrument exit efficiency, sorting coefficient, homogeneity coefficient, and structural coefficient.

[0082] The microscopic pore throat characteristics were obtained through high-pressure mercury intrusion and constant-rate mercury intrusion experiments.

[0083] In this embodiment, in step S5,

[0084] The reservoir sensitivity evaluation experiments include water sensitivity test, acid sensitivity test, salt sensitivity test, alkali sensitivity test, rate sensitivity test, stress sensitivity test and water lock test; thereby obtaining the reservoir sensitivity index, which includes water sensitivity index, acid sensitivity index, salt sensitivity index, alkali sensitivity index, rate sensitivity index, stress sensitivity index and water lock index;

[0085] The formula for calculating the sensitivity index is as follows:

[0086]

[0087] In the formula: D vn —The rate of change in rock sample permeability at different flow velocities; K n — Rock permeability after damage, in units of (10) -3 ×μm 2 ); K i —Initial permeability, in units of (10) -3 ×μm 2 ).

[0088] In a preferred embodiment, in S1 and S6,

[0089] The data training steps of the random forest algorithm include: data acquisition, data preprocessing, construction of training / test sets, prediction, comparison, and establishment of the relationship between the true values ​​and the test values.

[0090] The specific implementation process is as follows:

[0091] Step 1: Based on experimental and logging data of clay minerals at different depths in deep-water, low-permeability reservoirs, establish a profile of the variation in the content of different types of clay minerals throughout the entire reservoir section. The specific steps are as follows:

[0092] 1. Select a deep-water, low-permeability reservoir well and take more than 100 rock samples at different depths. Use XRD diffraction to analyze the content of different types of clay minerals in the different samples. The different types of clay minerals include montmorillonite, illite, kaolinite, and chlorite.

[0093] 2. Query the K, TH, U, and K / TH logging data for the corresponding depth.

[0094] 3. Establish a dataset containing experimental data and well logging data of clay minerals at different depths.

[0095] 4. Normalize the constructed dataset using the following formula:

[0096]

[0097] 5. The random forest algorithm was used to train the constructed data (training results are shown in...). Figure 2 This study establishes the relationship between the content of different types of clay minerals and four types of well logging data. The well logging data includes K, TH, U, and K / TH logging data.

[0098] 6. Based on the K, TH, U, and K / TH logging data of the entire reservoir section, establish a profile of the variation of different types of clay mineral content in the entire reservoir section. Figure 3 ).

[0099] Step 2: Based on well logging data, calculate porosity and establish a porosity variation profile for the entire reservoir section. The specific steps are as follows:

[0100] 1. For the same well, select logging data for calculating porosity; the logging data refers to density logging data and sonic transit time logging data.

[0101] 2. Calculate the reservoir porosity at different depths based on the porosity calculation formula; the porosity calculation formula is as follows:

[0102] Calculating porosity using acoustic transit time:

[0103]

[0104] Calculate porosity using density logging data:

[0105]

[0106] 3. Based on density logging data and sonic transit time logging data for the entire reservoir section, establish a porosity variation profile for the entire reservoir section, such as... Figure 4 .

[0107] Step 3: Based on experimental and logging data, fit a porosity-permeability relationship model and establish a permeability variation profile for the entire reservoir section. The specific steps are as follows:

[0108] 1. For the same well, take more than 50 sets of standard rock samples at different depths; the diameter of the rock samples is 2.5cm and the height is 4cm.

[0109] 2. The porosity of standard rock samples at different depths was tested using nuclear magnetic resonance (NMR) technology.

[0110] 3. The pressure decay method was used to test the permeability of standard rock samples at different depths.

[0111] 4. By fitting the permeability and permeability data obtained from experimental tests using an exponential model, a pore-permeability relationship model is established.

[0112] 5. Based on the porosity variation profile data of the entire reservoir section established in step 2, input the data into the porosity-permeability relationship model to establish a permeability variation profile of the entire reservoir section, such as... Figure 4 .

[0113] Step four: Using the random forest algorithm, train the relationship between macroscopic material parameters and microscopic pore-throat characteristic values ​​at different depths throughout the entire reservoir section to construct a prediction model for the microscopic pore-throat characteristics of deep-water low-permeability reservoirs. Establish a profile of pore-throat characteristic value changes across the entire deep-water low-permeability reservoir section. Specific steps are as follows:

[0114] 1. For the rock sample in step 3, according to the standard GB / T 29171-2012 "Determination of capillary pressure curve of rock", high pressure mercury injection and constant rate mercury injection experiments were carried out to test the microscopic pore throat characteristics of rocks at different depths, including the maximum pore throat radius, average pore throat radius, main channel throat radius, and maximum connecting throat radius.

[0115] 2. Combining steps three and four, a dataset containing macroscopic physical property parameters and microscopic pore throat characteristic values ​​is constructed.

[0116] 3. The random forest algorithm was used to train the constructed data (training results are shown in...). Figure 6 Establish the relationship between macroscopic physical properties (porosity, permeability) and microscopic pore throat structural characteristics (maximum pore throat radius, average pore throat radius, main channel throat radius, maximum connecting throat radius).

[0117] 4. Based on the porosity and permeability variation profiles established in steps 2 and 3, substitute the above relationships to establish a microscopic pore throat characteristic variation profile for the entire reservoir section, such as... Figure 7 .

[0118] Step 5: Through reservoir sensitivity evaluation experiments, test the potential damage index of deep-water low-permeability reservoirs at different depths. The specific steps are as follows:

[0119] 1. For rock samples at different depths in the deep-water low-permeability reservoir in step 3, sensitivity experiments were conducted in accordance with the standard 5358-2010 "Evaluation Method of Reservoir Sensitivity Flow Experiment".

[0120] 2. For rock samples at different depths in the deep-water low-permeability reservoir in step 3, water-locking experiments were conducted in accordance with the QSY 1832-2015 standard "Experimental Evaluation Method for Water Locking Damage in Tight Gas Reservoirs".

[0121] 3. Calculate the sensitivity index. The formula is as follows:

[0122]

[0123] In the formula: D vn —The rate of change in rock sample permeability at different flow velocities; K n — Rock permeability after damage, in units of (10) -3 ×μm 2 ); K i —Initial permeability, in units of (10) -3×μm 2 ).

[0124] 4. Combining steps 1, 2, 3, and 4, construct a dataset including clay mineral composition data (montmorillonite, illite, chlorite, and kaolinite), macroscopic physical property data (porosity and permeability), microscopic pore structure data (maximum pore throat radius, average pore throat radius, main channel throat radius, and maximum connected channel throat radius), and sensitivity index data (water sensitivity, acid sensitivity, salt sensitivity, alkali sensitivity, rate sensitivity, stress sensitivity, and water lock).

[0125] 5. The Random Forest algorithm was used for training data (training results are shown in...). Figure 8 A model was established to establish the relationship between the composition of clay minerals, macroscopic physical properties, microscopic pore structure characteristics, and sensitivity index of deep-water low-permeability reservoir rocks.

[0126] 6. Based on all the previous results, a profile of the potential damage index variation across the entire reservoir section of the deep-water low-permeability reservoir was established, see [link to relevant documentation]. Figure 9 .

[0127] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for quantitatively predicting potential damage in low permeability reservoirs based on random forest algorithm, characterized in that, It comprises the following steps: S1. Based on the experimental data and logging data of clay minerals of deep water low permeability reservoir rocks at different depths, the logging data and clay mineral experimental data set is trained by using the random forest algorithm; a clay mineral prediction model based on logging data points is established, and the content variation profile of different types of clay minerals in the whole reservoir section is obtained; S2. Based on the logging data, the porosity is calculated, and the porosity variation profile of the whole reservoir section is established; S3. Based on the experimental data and logging data, the porosity-permeability relationship is fitted, and the porosity and permeability profiles of the whole reservoir section are established; S4. Through the random forest algorithm, the rock macroscopic physical property parameter and micro pore throat characteristic data set at different depths of the whole reservoir section is trained, and a micro pore throat characteristic value prediction model based on macroscopic physical property parameters is established, so as to obtain the variation profile of the micro pore throat characteristic value of the whole reservoir section; S5. Through reservoir sensitivity evaluation experiment, the sensitivity of rock at different depths of deep water low permeability reservoir is tested, and the potential damage index of rock at different depths of deep water low permeability reservoir is calculated and obtained; S6. Based on S1-S4, the random forest algorithm is used to train different types of clay mineral component data, macroscopic physical property data and micro pore throat characteristic data and reservoir potential damage index data, and a deep water low permeability reservoir potential damage index prediction model is established.

2. The method of claim 1, wherein the method is characterized by: In S1, The logging data is K, TH, U, K / TH logging data; The clay mineral experimental data is the content of montmorillonite, illite, kaolinite and chlorite in the rock at different depths of the reservoir obtained by XRD diffractometer test.

3. The method of claim 1, wherein the method is characterized by: In S2, The logging data is density logging data and acoustic logging data; The porosity calculation involves the following calculation formula: wherein, represents the acoustic traveltime value of the near-surface mudstone, unit ; represents the acoustic travel time value of the pore fluid, in units of ; sonic travel time value indicative of the shale matrix, in units of ; The porosity calculation formula using density logging data is as follows: wherein, represents the matrix density; represents the fluid density; represents the rock density.

4. The method of claim 1, wherein the method is characterized by: In S3, The experimental data is the porosity data of low permeability reservoir rock at different depths obtained by nuclear magnetic resonance technology and the permeability data of low permeability reservoir rock at different depths obtained by pressure decay method; The fitting of porosity and permeability relationship is through fitting the porosity and permeability test data at different depths; The porosity and permeability profiles of the whole reservoir section are obtained according to the porosity variation profile of the whole reservoir section established in S2 and the porosity-permeability relationship model.

5. The method of claim 1, wherein the method is characterized by: In S4, The macroscopic physical property parameters include porosity and permeability; The micro pore throat characteristics include maximum pore throat radius, average pore throat radius, main flow throat radius, maximum connected throat radius, pore median radius, saturation median pressure, displacement pressure, maximum mercury saturation, instrument maximum exit efficiency, sorting coefficient, homogeneity coefficient, structure coefficient; The micro pore throat characteristic values are obtained by high pressure mercury injection and constant speed mercury injection experiment.

6. The method of claim 1, wherein the method is characterized by: In S5, The reservoir sensitivity evaluation experiment includes water sensitivity experiment, acid sensitivity experiment, salt sensitivity experiment, alkali sensitivity experiment, velocity sensitivity experiment, stress sensitivity experiment and water lock experiment; so as to obtain the reservoir sensitivity index, which includes water sensitivity index, acid sensitivity index, salt sensitivity index, alkali sensitivity index, velocity sensitivity index, stress sensitivity index and water lock index; The sensitivity index calculation formula is: wherein: - the rate of change of permeability of the rock sample corresponding to different flow rates; - the permeability of the rock sample after damage, in Darcy; ; - the initial permeability, in Darcy; .

Images