Variable cycle combination method for multiple oil displacement systems based on gradient boosting tree algorithm

Abstract

The invention discloses a variable cycle combination method for multiple oil displacement systems based on a gradient boosting tree algorithm. The method comprises the following steps: S1, constructing a variable cycle combination large-scale learning sample of the oil displacement system; and S2, establishing a gradient boosting tree prediction model. According to the gradient boosting tree algorithm, a plurality of weak decision trees are integrated to gradually optimize model errors, flexibility and interpretability are kept, meanwhile, high robustness and over-fitting resistance are achieved, a plurality of weak learners can be weighted and combined in each step, and high prediction capacity and precision are achieved. According to the invention, on the basis of the gradient boosting tree algorithm, mining learning is carried out on large-scale learning samples by designing an appropriate deep neural network, the internal law of the numerical calculation process can be captured, and then time-consuming large-scale matrix iteration in numerical calculation is replaced; accurate prediction solving of the variable circulation oil displacement process of multiple oil displacement systems is achieved while calculation consumption is remarkably reduced.

Description

Technical Field

[0001] The invention discloses a variable cycle combination method of multiple oil displacement systems based on a gradient boosting tree algorithm, belonging to the technical field of oil reservoir development. Background Art

[0002] Compared with onshore oil fields, offshore oil fields are mostly loose sandstone with strong heterogeneity, and have the characteristics of low well network perfection, large well spacing, large-section multi-layer joint injection and production, strong injection and strong production, etc. The heterogeneity is further aggravated during long-term water drive to high water content stage. Chemical flooding technology based on polymer flooding is an effective technical means to improve the recovery rate of offshore oil fields, but profile reversal often occurs in the later stage of polymer flooding, which gradually reduces the controlled reserves in the later stage of polymer flooding, resulting in about 50% of crude oil remaining in the reservoir after polymer flooding. In response to the problems existing in the chemical flooding process, the common improvement methods at present are to change the concentration of chemical flooding agents or add other agents to the segment plugs and improve the performance of the agents. However, this combined use generally achieves phased and local improvements. Few researchers have reasonably combined these systems at different development stages or even the entire life cycle of the oil field to form a more effective and long-term global system with the coordinated application of multiple technologies to greatly improve the recovery rate. When multiple oil recovery systems are applied in combination with variable circulation, the performance improvement of different agents and the improvement of injection methods have different action rules and migration characteristics in the reservoir, and each has different applicable conditions and pertinence in solving problems. A large number of physical simulation experiments are needed to optimize the injection parameters and analyze the influence of parameters, and at the same time, quantitatively predict the extent of the variable circulation combination to improve the recovery rate. At present, there are not many research papers on related issues, which makes it difficult to provide theoretical guidance for the application technology of variable circulation combination of multiple oil recovery systems. Summary of the invention

[0003] The purpose of the present invention is to provide a method for combining multiple oil recovery systems by variable cycles based on a gradient boosting tree algorithm, which can provide important technical support for improving crude oil recovery by variable cycle combinations of multiple oil recovery systems. It is an analysis method with a simple process and very obvious technical and economic effects.

[0004] The present invention provides a method for combining multiple oil displacement systems with variable cycles based on a gradient boosting tree algorithm, comprising the following steps:

[0005] S1. Construction of large-scale learning samples of variable cycle combination of oil displacement system;

[0006] S2. Gradient boosting tree prediction model is established.

[0007] Preferably, step S1 includes the following two steps:

[0008] Physical simulation of injection parameter optimization of variable cycle combination of various oil recovery systems; numerical simulation of injection parameter optimization of variable cycle combination of various oil recovery systems.

[0009] Preferably, the physical simulation of injection parameter optimization of multiple oil displacement systems with variable cycles comprises the following steps:

[0010] Carry out variable cycle combined flooding experiments with different flooding systems for target oil fields and optimize injection parameters.

[0011] Preferably, the numerical simulation of injection parameter optimization of the variable cycle combination of multiple oil displacement systems comprises the following steps:

[0012] Through numerical simulation technology, the core displacement experiment results are scaled up to the reservoir scale to obtain the effect prediction and scheme optimization results of the oil displacement system under different variable circulation combinations.

[0013] Preferably, in step S2, the gradient boosting tree prediction model establishment comprises the following steps:

[0014] By initializing the model to a constant value, the residual (the error between the target value and the current model prediction value) is calculated step by step, and a new decision tree is trained in each iteration to fit the residual; the prediction results of each tree are weighted and updated to the existing model, and the error is gradually reduced through multiple iterations, and finally combined to form a prediction model.

[0015] Preferably, the loss function is minimized by gradient descent, the model performance is optimized, and the variable cycle combination effects and optimization schemes of various oil recovery systems are obtained to form a comprehensive efficiency index chart for improving oil recovery.

[0016] More preferably, the gradient boosting tree prediction model establishment comprises the following steps:

[0017] The single-factor sensitivity analysis method was used to study the impact of different factors on enhanced oil recovery, and production data samples were obtained. The training set data was learned using a machine learning model. The important features in the data were filtered and extracted through the neural network layer, and the production sample data was trained and fitted. A gradient boosting tree model was established to predict the displacement effects of variable cycle combinations of various oil recovery systems.

[0018] The important characteristics refer to injection parameters such as permeability difference, injection timing, heterogeneous system concentration, injection PV number, alternating cycle, etc.

[0019] The present invention has the following beneficial technical effects:

[0020] 1. The gradient boosting tree algorithm gradually optimizes the model error by integrating multiple weak decision trees. While maintaining flexibility and interpretability, it has strong robustness and resistance to overfitting. It can weightedly combine multiple weak learners at each step, giving it high predictive ability and accuracy.

[0021] 2. Based on the gradient boosting tree algorithm, by designing appropriate deep neural networks to mine and learn large-scale learning samples, the inherent laws of the numerical calculation process can be captured, thereby replacing the time-consuming large-scale matrix iteration in numerical calculations, significantly reducing the computing cost while achieving accurate prediction and solution of variable cycle flooding processes in various flooding systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a gradient boosted tree prediction model.

[0023] Figure 2 It is the moisture content of the numerical model and physical model of the fitting inversion.

[0024] Figure 3 It is the degree of extraction of numerical and physical models of fitting inversion.

[0025] Figure 4 It is the relationship diagram of the determination coefficient of the gradient boosting tree to predict the enhanced oil recovery.

[0026] Figure 5 It is a comparison chart of the recovery factor predicted by the gradient boosting tree and the true value.

[0027] Figure 6 It is a graph of the importance of different influencing factors on improving oil recovery predicted by the gradient boosting tree.

[0028] Figure 7 It is a chart of comprehensive efficiency index for improving oil recovery. DETAILED DESCRIPTION

[0029] Unless otherwise specified, the experimental methods used in the following examples are conventional methods.

[0030] Unless otherwise specified, the materials and reagents used in the following examples can be obtained from commercial sources.

[0031] Example 1. Experimental conditions

[0032] 1. The heterogeneous system is a flexible microgel particle water dispersion system with an effective content of 100%; the polymer is partially hydrolyzed polyacrylamide powder (HPAM) with a relative molecular mass of 20 to 22 million and a solid content of 88%; the phenolic crosslinker is provided by CNOOC Research Institute Co., Ltd. The experimental oil is a mixture of degassed crude oil and kerosene from J Oilfield in a certain proportion, with a viscosity of 17mPa·s at 57°C. The experimental water uses simulated water prepared according to the mineralization of J Oilfield, and the water quality analysis is shown in Table 1.

[0033] Table 1 Water quality analysis

[0034]

[0035] When conducting indoor experiments, the experimental temperature should simulate the actual temperature of the oil reservoir in the target block of the offshore oil field, which is 57°C.

[0036] 2. The experimental core is a man-made homogeneous core bonded with quartz sand and epoxy resin. The experimental model is made up of two cores connected in parallel. The external geometric dimensions are length × width × height = 30 cm × 4.5 cm × 4.5 cm. The gas permeability is K g =400, 800, 1200, 1600, 2000, 2400×10 -3 μm 2 The core is sealed by epoxy resin casting.

[0037] 3. Solution preparation

[0038] (1) Polymer solution

[0039] In order to simulate the shearing effect on the polymer solution during preparation and injection in the mine, it was pre-sheared before the experiment so that its viscosity retention rate was 60%.

[0040] (2) Heterogeneous system

[0041] 1) Before preparing the heterogeneous system solution, shake the sample bottle vigorously manually or stir with a glass rod to make the heterogeneous system evenly dispersed.

[0042] 2) Extract the stock solution of the heterogeneous system according to the designed concentration, mix it with a certain amount of simulated water, and then place it on a magnetic stirrer and stir for 15 minutes.

[0043] 3) Transfer the heterogeneous system solution into a piston container with a stirring device and start the stirrer to avoid particle aggregation and sedimentation and thus adverse effects on the experimental results.

[0044] 4. The core displacement experimental device mainly includes a horizontal flow pump, a pressure sensor and an intermediate container. Except for the horizontal flow pump, the other parts are placed in a 60℃ constant temperature box. The experimental steps are as follows:

[0045] (1) Pump out saturated water from the core at room temperature and calculate the porosity;

[0046] (2) The core is saturated with oil at reservoir temperature and the oil saturation is calculated;

[0047] (3) The core composition parallel model, water flooding to 98%, calculate the water flooding recovery factor;

[0048] (4) Inject the chemical solution with designed PV number into the core and calculate the water content and recovery factor;

[0049] (5) The core is subsequently water-flooded to a water content of 98% and the water content and recovery factor are calculated.

[0050] In the experiment of the present invention, the injection rate of the core flooding experiment was 0.6 mL / min.

[0051] Example 2: Multiple oil displacement system variable cycle combinations based on gradient boosting tree algorithm

[0052] 1. Physical simulation of injection parameter optimization for variable cycle combination of various oil recovery systems

[0053] The physical simulation experimental scheme for optimizing injection parameters of variable cycle combination of various oil displacement systems is shown in Table 2.

[0054] Table 2 Scheme and process conditions

[0055]

[0056] (Note: When the permeability difference is greater than 4, it is necessary to use the gel system to adjust the liquid absorption profile before performing a combination of multiple oil displacement systems and variable cycles.)

[0057] Experiments were carried out under the conditions of permeability differences of 2, 3, 4, 5, and 6. The pressure during each experiment was recorded, and the water cut and recovery factor were calculated. The experimental phenomena and principles of variable circulation combinations of various oil recovery systems were analyzed.

[0058] 2. Numerical simulation of injection parameter optimization for variable cycle combination of various oil displacement systems

[0059] According to the above physical simulation experimental results, the STARS module of CMG software was used to conduct core experimental inversion research and establish a numerical inversion model, in which the numerical model parameters were as follows: number of grids: 30×1×2; grid size: 1cm×4.5cm×4.5cm; adsorption parameters, polymer concentration changes, residual resistance coefficient and other polymer flooding characteristic parameters were set according to the indoor experimental results. The parameters were adjusted comprehensively for fitting, and the physical simulation experimental results were analyzed with the numerical simulation results.

[0060] The numerical simulation results at the core scale are effectively enlarged to the reservoir scale. According to the characteristics of typical blocks in the S oilfield, a three-layer heterogeneous injection-production mechanism model is established. It is divided into three small layers: upper, middle and lower. The number of grids in the X, Y and Z directions is 100×18×3, the grid step is 3×10×5m, the total thickness is 15m, the top surface depth is 800m, the horizontal permeabilities of each layer are 700mD, 1400mD and 3500mD respectively, the vertical permeability is 1 / 10 of the horizontal permeability, the porosity is 30%, and the well pattern adopts one injection and one production.

[0061] The values ​​of reservoir and fluid composition parameters are shown in Table 3.

[0062] Table 3 Reservoir and fluid composition parameters

[0063]

[0064]

[0065] The optimization process of injection parameters of variable circulation combination of various oil recovery systems involves the evaluation of 5 factors, as shown in Table 4. In combination with the characteristics of orthogonal experimental design, L16 (5 factors and 4 levels) orthogonal design was adopted. According to the orthogonal design results, the STARS module of CMG software was used to complete the calculation, and the recovery rate of variable circulation combination of various oil recovery systems was obtained under different experimental conditions compared with continuous polymer injection.

[0066] Table 4 Injection parameters and values

[0067]

[0068] 3. Establish a gradient boosting tree prediction model

[0069] Gradient Boosted Decision Tree (GBDT), see Figure 1 , gradient boosting regression tree regression was used to predict the displacement effects of variable cycle combinations of various oil displacement systems.

[0070] The gradient regression tree performs regression prediction, and the loss function is generally square loss: Pseudo response variable is the current residual, let the current predicted value The steps of gradient boosting regression tree regression prediction are as follows:

[0071] (1) Calculate pseudo response variables:

[0072] (2) Solving the parameter model:

[0073] (3) The updated predicted value is expressed as:

[0074]

[0075] Among them, h b (X; α b ) is a regression tree with J leaf nodes; R jb The regional division obtained by building a regression tree based on the pseudo response variable for the bth iteration; For the predicted value to fall within R jb Sample observation points of the region The mean value of b Obtained through line search.

[0076] (4) When analyzing the factors that affect enhanced oil recovery, the data should be normalized and the dimensions of each data should be normalized to make the results comparable.

[0077] (5) Based on the analysis of the influence of various factors on enhanced oil recovery through machine learning, in order to more clearly reflect the relationship between enhanced oil recovery and various factors, a comprehensive efficiency index is defined to judge the oil increase effect under specific working conditions. The specific calculation model is as follows:

[0078]

[0079] In the formula, Z i -Comprehensive oil displacement index, dimensionless

[0080] N-the number of key influencing factors

[0081] λ i -The weight coefficient of the i-th key parameter

[0082] x i -The value parameter of the i-th key parameter after standardization

[0083] By integrating various factors, a comprehensive efficiency index chart for enhanced oil recovery is formed, which can more intuitively and accurately determine the pros and cons of the proposed oil recovery plan.

[0084] Example 3, analysis of experimental results:

[0085] 1. Physical simulation experimental results of injection parameter optimization for variable cycle combination of various oil recovery systems

[0086] The results of physical simulation experiments on injection parameter optimization of variable cycle combination of various oil displacement systems are shown in Table 5.

[0087] Table 5 Experimental results

[0088]

[0089] It can be seen from Table 5 that, under the conditions of reservoir permeability difference of 2 to 6 and the same reagent cost, compared with continuous polymer injection, the periodic injection of polymer solution and heterogeneous system can increase the recovery rate by 3.3% to 6.1%. This is because the variable cycle combined injection method of multiple oil displacement systems can produce a better liquid flow diversion effect, slow down the process of liquid absorption profile reversal, and expand the swept volume, thereby increasing the oil recovery rate.

[0090] At the same time, through this group of experimental results, the alternating cycle under different permeability differences can be optimized. A reasonable alternating cycle can effectively increase the relative liquid absorption of the low permeability layer. If the alternating rounds are too few, the length of a single plug is large, and the effect of improving the seepage resistance is not good; if the alternating rounds are too frequent, the high-viscosity plug cannot form an effective sealing and shielding effect in the high permeability layer, and the subsequent oil recovery system will advance along the high permeability layer, weakening the liquid flow diversion effect, resulting in a decrease in the recovery rate increase. In this group of experiments, when the reservoir permeability difference is 2 to 6, the optimal alternating cycle increases with the increase in the difference, but the liquid flow diversion effect gradually deteriorates after the alternating cycle exceeds 4.

[0091] 2. Numerical simulation results of injection parameter optimization for variable cycle combination of various oil displacement systems

[0092] The fitting curve of the core flooding experiment results is as follows Figure 2 and Figure 3 As shown. Figure 2 and Figure 3 It can be seen that the numerical simulation results are highly consistent with the core displacement experiment results, indicating that the parameter settings are relatively reasonable. The established mechanism model can better simulate the variable cycle combined injection experiment process of various oil displacement systems. At the same time, it also verifies the reliability of the indoor experiment. The parameters can be used in subsequent numerical simulation research to provide support for regularity research and parameter optimization.

[0093] Table 6 Orthogonal design and calculation results

[0094]

[0095] (Note: When the permeability difference is greater than 4, it is necessary to use the gel system to adjust the liquid absorption profile before performing a combination of multiple oil displacement systems and variable cycles.)

[0096] As can be seen from Table 6, the orthogonal test results show that the recovery rate of various oil displacement system variable cycle combinations is between 3.28% and 7.02%, which is consistent with the experimental results of various oil displacement system variable cycle combinations. Therefore, the physical simulation experimental results and numerical simulation calculation results can be used for the quantitative prediction of EOR of various oil displacement system variable cycle combinations in the future.

[0097] 3. Establish a gradient boosting tree prediction model

[0098] The present invention is based on CMG simulation of variable cycle combination flooding of various oil recovery systems, and adopts a single factor sensitivity analysis method to study the relationship between permeability difference, injection PV number, alternating cycle, heterogeneous system concentration, injection timing and other factors on improving oil recovery. 1024 groups of production data samples are obtained, and the training set data are learned by a machine learning model. The important features in the data are filtered and extracted through numerous neural network layers, and the production sample data are trained and fitted. Finally, a prediction model is established to predict the displacement effects of variable cycle combination of various oil recovery systems.

[0099] (1) Correlation analysis

[0100] The correlation analysis between enhanced oil recovery and different influencing factors is shown in Table 7.

[0101] Table 7 Correlation analysis results between enhanced oil recovery and different influencing factors

[0102]

[0103] It can be seen from Table 7 that the enhanced oil recovery is positively correlated with the permeability difference, heterogeneous system concentration, injection PV number and alternating cycle, and negatively correlated with the injection timing.

[0104] (2) Gradient Boosting Tree Algorithm Analysis

[0105] 1) Enhanced oil recovery prediction

[0106] The results of determination coefficient of gradient boosting tree for enhanced oil recovery prediction are shown in Figure 4 .

[0107] from Figure 4 It can be seen that the determination coefficient is 0.988, the relative absolute error (RAE) is 0.078, and the determination coefficient of the training set is almost the same as that of the test set, indicating that the gradient boosting tree has a good prediction effect on the enhanced oil recovery of variable cycle combination flooding in various oil flooding systems.

[0108] Comparison of the prediction results of EOR by gradient boosting tree with the true value is shown in Figure 5 .

[0109] from Figure 5 It can be seen that most of the data points are very close to the straight line x=y, indicating that the EOR predicted by the gradient boosting tree is close to the actual EOR value, the model training fitting accuracy is high, and it can accurately predict the displacement effect of variable cycle combinations of various oil recovery systems.

[0110] 2) Analysis of the importance of factors affecting oil recovery

[0111] The importance of different factors in predicting the recovery efficiency of the gradient boosting tree is shown in Table 8 and Figure 6 .

[0112] Table 8 Importance of gradient boosting tree prediction parameter variables for improving oil recovery

[0113]

[0114] From Table 8 and Figure 6 It can be seen that the importance of various factors on improving oil recovery is ranked as follows: number of injected PVs > concentration of injected heterogeneous system > injection timing > permeability difference > alternating cycle, among which the number of injected PVs is the most important influencing factor, followed by the concentration of injected heterogeneous system.

[0115] 3) Comparative analysis of different machine learning algorithms

[0116] The comparison of the results of random forest and gradient boosting tree is shown in Table 9.

[0117] Table 9 Comparison of random forest and gradient boosting tree results

[0118]

[0119] From the data comparison in Table 9, it can be seen that the gradient boosting regression tree has a smaller error and a higher determination coefficient, so the prediction accuracy is better than that of the random forest.

[0120] Random forest is suitable for situations that require fast training, high robustness, and low requirements for parameter tuning, especially when the data set is large. It can handle high-dimensional data and noisy data well, and the training process is easy to parallelize. Gradient boosting tree performs better in situations where high precision is sought, especially for those that need to handle unbalanced data or have special requirements for optimizing loss. However, it takes a long time to train and is more difficult to tune parameters, so it is more suitable for conditions where resources permit and high model accuracy is required.

[0121] 4) Comprehensive efficiency index for enhanced oil recovery

[0122] After integrating various factors, the comprehensive efficiency index chart of enhanced oil recovery is formed. Figure 7 , the advantages and disadvantages of the proposed oil recovery plan can be determined more intuitively and accurately.

[0123] from Figure 7 It can be seen that there is a good correlation between the enhanced oil recovery and the efficiency index in the two intervals of less than 0.75 and greater than 0.75. In practical applications, the permeability difference, injection PV number, alternating cycle, heterogeneous system concentration, and injection timing can be used to calculate the response comprehensive efficiency index, and the effect of the scheme can be judged in combination with the chart.

Claims

1. A variable cycle combination method of multiple oil displacement systems based on a gradient boosting tree algorithm, comprising the following steps: S1. Construction of large-scale learning samples of variable cycle combination of oil displacement system; S2. Gradient boosting tree prediction model is established.

2. The multiple oil displacement system variable cycle combination method according to claim 1, characterized in that: Step S1 includes the following two steps: Physical simulation of injection parameter optimization of variable cycle combination of various oil recovery systems; numerical simulation of injection parameter optimization of variable cycle combination of various oil recovery systems.

3. The multiple oil displacement system variable cycle combination method according to claim 2, characterized in that: The physical simulation of injection parameter optimization of multiple oil displacement systems with variable cycles comprises the following steps: Carry out variable cycle combined flooding experiments with different flooding systems for target oil fields and optimize injection parameters.

4. The multiple oil displacement system variable cycle combination method according to claim 3, characterized in that: The numerical simulation of injection parameter optimization of the variable cycle combination of multiple oil displacement systems includes the following steps: Through numerical simulation technology, the core displacement experiment results are scaled up to the reservoir scale to obtain the effect prediction and scheme optimization results of the oil displacement system under different variable circulation combinations.

5. The method for combining multiple oil displacement systems according to any one of claims 1 to 4, characterized in that: In step S2, the gradient boosting tree prediction model is established including the following steps: The model is initialized to a constant value, the residual is calculated step by step, and a new decision tree is trained in each iteration to fit the residual; the prediction results of each tree are weighted and updated to the existing model, and the error is gradually reduced through multiple iterations, and finally combined to form a prediction model.

6. The multiple oil displacement system variable cycle combination method according to claim 5, characterized in that: By minimizing the loss function through gradient descent and optimizing the model performance, we can obtain the variable cycle combination effects and optimization schemes of various oil recovery systems, and form a comprehensive efficiency index chart for improving oil recovery.

7. The multiple oil displacement system variable cycle combination method according to claim 5, characterized in that: Preferably, the gradient boosting tree prediction model establishment comprises the following steps: The single-factor sensitivity analysis method was used to study the impact of different factors on enhanced oil recovery, and production data samples were obtained. The machine learning model was used to learn the training set data. The important features in the data were filtered and extracted through the neural network layer, and the production sample data was trained and fitted. A gradient boosting tree model was established to predict the displacement effects of variable cycle combinations of various oil recovery systems.

Images