A machine learning-based method for optimizing the amount of oil well fracturing proppant

Abstract

The present disclosure relates to a machine learning-based oil well fracturing proppant dosage optimization method, according to the geological, development, process and other basic data of the historical fracturing oil well of the target block, the main control factor of the initial daily oil increment is determined through the correlation algorithm, the proppant dosage is one of the main control factors, a prediction model is established between the initial daily oil increment and the main control factor based on the machine learning algorithm, and then the main control factor of the oil well to be implemented is calculated, the proppant dosage at the maximum initial daily oil increment is predicted as the optimal dosage by setting different proppant dosages; the present application can predict the initial daily oil increment under different proppant dosages, and then the optimal proppant dosage can be determined, so that the fracturing effect is optimal.

Description

Technical Field

[0001] This disclosure relates to the field of oil and gas field development technology, specifically to a method for optimizing the dosage of proppant in oil well fracturing. Background Technology

[0002] The statements in this section provide only background information in connection with this disclosure and do not constitute prior art.

[0003] my country's old oilfields have generally entered the high water-cut development stage. Oil well fracturing is one of the important means of increasing production in old oilfields. The fracturing potential of oil wells is mostly determined by manual analysis, which makes it impossible to achieve quantitative analysis of the fracturing effect.

[0004] The amount of proppant used in oil well fracturing is one of the important process factors affecting the fracturing effect. If the amount of proppant used in oil well fracturing is too small, the fracture will close quickly and the oil production increase effect will not be ideal. However, field production data also shows that the oil production increase is not necessarily greater with the amount of proppant used. In fact, a large amount of proppant used will also increase the fracturing cost and affect the economic benefits. Therefore, determining the appropriate amount of proppant used in the fracturing process is an urgent technical problem that needs to be solved.

[0005] The injection-production relationship and oil-water distribution in old oilfields are complex, and the fracturing effect involves many factors. How to accurately predict the fracturing effect and how to adjust the proppant dosage to improve the fracturing effect are current challenges.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art. Summary of the Invention

[0007] In view of this, this disclosure provides a machine learning-based method for optimizing the dosage of proppant in oil well fracturing, which solves the problem that the dosage of proppant in oil well fracturing is currently determined by manual analysis, while the fracturing effect involves many factors and cannot be accurately predicted.

[0008] To solve the above-mentioned technical problems, the general inventive concept of the method for optimizing the dosage of proppant in oil well fracturing as disclosed in this disclosure is:

[0009] Based on the geological, development, and technological data of historical fracturing wells in the target block, the main controlling factors for daily oil production increase in the initial stage of fracturing are determined through correlation algorithms. Propionate dosage is one of the main controlling factors. A predictive model is established based on machine learning algorithms to establish the relationship between the initial daily oil production increase and the main controlling factors. Then, the main controlling factors are calculated for the wells that are about to be fracturing. By setting different proppant dosages, the proppant dosage at which the initial daily oil production increase is maximized is predicted as the optimal dosage.

[0010] Based on the inventive concept of this invention, the machine learning-based method for optimizing the dosage of proppant in oil well fracturing as disclosed herein includes:

[0011] Identify the key controlling factors affecting daily oil production increase in the initial stage of fracturing in the target block, including proppant dosage.

[0012] Based on machine learning, a predictive model for daily oil production increase in the initial stage of fracturing is established by utilizing the correlation between the daily oil production increase in the initial stage of fracturing and the main controlling factors.

[0013] The main control factor data of the target fractured well are input into the prediction model. By setting different proppant dosages, the daily oil increase in the initial stage of fracture is predicted. The maximum daily oil increase in the initial stage of fracture is taken as the optimal proppant dosage for the target fractured well.

[0014] In this disclosure and possible embodiments, the method for determining the main controlling factors affecting daily oil production increase in the target block, including proppant dosage, includes:

[0015] Based on the geological, development, and technological data of historical fractured wells in the target block, Spearman rank correlation analysis was used to calculate the correlation between the initial daily oil production of the historical fractured wells and various influencing factors, and the main controlling factors were determined.

[0016] In this disclosure and possible embodiments, the influencing factors with a correlation greater than 0.3 are determined as the controlling factors.

[0017] In this disclosure and possible embodiments, the geological, development, and process data include the following: the geological data includes the total number of fractured layers, the total thickness of fractured sandstone, the total equivalent thickness of fractured layers, the average permeability of the fractured layers, and the coefficient of variation of the permeability of the original production layers before fracturing; the development data includes the daily oil production, daily fluid production, water cut, flowing pressure, submersion degree, dynamic fluid level, production intensity, the average number of connected directions of the fractured layers, the average oil saturation of the fractured layers, and the minimum injection-production well spacing of the fractured layers; the process data includes the amount of proppant added per unit sandstone thickness for whole-well fracturing, the amount of proppant used for whole-well fracturing, and the amount of fracturing fluid used per unit sandstone thickness for whole-well fracturing.

[0018] In this disclosure and possible embodiments, the method for establishing a predictive model for daily oil production increase in the initial stage of fracturing based on machine learning and utilizing the correlation between the daily oil production increase in the initial stage of fracturing and the main controlling factors includes:

[0019] The correlation is trained using several machine learning algorithms, and the ratio of training to testing historical data is set. The prediction model is determined based on the principle of achieving the highest test accuracy.

[0020] In this disclosure and possible embodiments, the ratio of the number of training to testing historical data entries is 8:2.

[0021] In this disclosure and possible embodiments, the machine learning algorithms include CatBoost, LightGBM, Extremely Random Tree, Gradient Boosting Tree, Random Forest, Theil-Sen Estimation Regression, Bayesian Ridge Regression, Minimum Angle Regression, Linear Regression, Ridge Regression, Stacking Ensemble Algorithm, Multilayer Perceptron Neural Network, Decision Tree, Elastic Regression, and Huber Regression Algorithm.

[0022] The beneficial effects of this invention are as follows:

[0023] This invention discloses a machine learning-based method for optimizing proppant dosage in oil well fracturing. Based on historical geological, development, and technological data of fracturing wells in the target block, it uses a correlation algorithm to determine the main controlling factors for daily oil production increase in the initial stage of fracturing. Propant dosage is one of these main controlling factors. A predictive model is established based on machine learning algorithms to establish the relationship between initial daily oil production increase and the main controlling factors. This model is then used to calculate the main controlling factors for the well to be fracturing. By setting different proppant dosages, the optimal proppant dosage is predicted for the maximum initial daily oil production increase. This invention can predict the initial daily oil production increase under different proppant dosage conditions, thereby determining the optimal proppant dosage to achieve the best fracturing effect. This effectively solves the problem that current oil well fracturing proppant dosage is determined manually, while fracturing effect involves many factors and cannot be accurately predicted. Attached Figure Description

[0024] Embodiments of this disclosure will be described below with reference to the accompanying drawings, in which:

[0025] Figure 1 This is a flowchart of a machine learning-based method for optimizing proppant dosage in oil well fracturing, according to an embodiment of this disclosure.

[0026] Figure 2 A schematic diagram illustrating the degree of influence of various factors on the daily oil production increase in the initial stage of fracturing in this embodiment of the present disclosure;

[0027] Figure 3 This disclosure provides a schematic diagram illustrating the test compliance rates of various machine learning algorithms in embodiments of the present invention. Detailed Implementation

[0028] The present disclosure is described below based on embodiments; however, it is worth noting that the present disclosure is not limited to these embodiments. In the detailed description of the present disclosure below, certain specific details are described in detail. However, those skilled in the art will fully understand the present disclosure for the parts not described in detail.

[0029] Furthermore, unless the context explicitly requires it, the words "comprising," "including," and similar terms throughout the specification and claims should be interpreted as including rather than exclusive or exhaustive; that is, meaning "including but not limited to."

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples of embodiments and applications.

[0031] To address the current problem that the dosage of proppant used in oil well fracturing is determined manually, while the fracturing effect is influenced by many factors and cannot be accurately predicted, this disclosure provides a machine learning-based method for optimizing the dosage of proppant used in oil well fracturing, which includes the following steps:

[0032] 1. Obtain geological, development, and technological data of historical fracturing wells in the target block:

[0033] In the various embodiments of this disclosure, the geological data includes the total number of fractured sublayers, the total thickness of fractured sandstone, the total equivalent thickness of fractured layers, the average permeability of the fractured layers, and the coefficient of variation of the permeability of the original production layers before fracturing; the development data includes daily oil production, daily fluid production, water cut, flowing pressure, submersion degree, dynamic fluid level, fluid production intensity, the average number of connected directions of the fractured layers, the average oil saturation of the fractured layers, and the minimum injection-production well spacing of the fractured layers; the process data includes the amount of proppant added per unit sandstone thickness for whole-well fracturing, the amount of proppant used for whole-well fracturing, and the amount of fracturing fluid used per unit sandstone thickness for whole-well fracturing.

[0034] 2. Determine the main controlling factors for daily oil production increase in the initial stage of fracturing, including proppant dosage:

[0035] In the various embodiments of this disclosure, the method for determining the main controlling factor is preferably to use the Spearman rank correlation analysis algorithm to calculate the correlation between the initial daily oil increase of historically fractured wells and various influencing factors; in addition, the influencing factors with a correlation greater than 0.3 are preferably regarded as the main controlling factors.

[0036] 3. Establish a predictive model for daily oil production increase during the initial stage of fracturing in the target block:

[0037] In the various embodiments of this disclosure, the prediction model is established by machine learning. The specific establishment process is as follows: based on the main control factor data of historical fractured wells and the daily oil increase data in the early stage of fracture, various machine learning training is carried out, the ratio of training to testing historical data is set, and the prediction model of daily oil increase in the early stage of fracture is determined with the highest test compliance rate as the principle; preferably, the ratio of training to testing historical data is set to 8:2.

[0038] In the various embodiments of this disclosure, the machine learning algorithms used to build the prediction model include CatBoost, LightGBM, Extremely Random Tree, Gradient Boosting Tree, Random Forest, Theil-Sen Estimation Regression, Bayesian Ridge Regression, Minimum Angle Regression, Linear Regression, Ridge Regression, Stacking Ensemble Algorithm, Multilayer Perceptron Neural Network, Decision Tree, Elastic Regression, and Huber Regression Algorithm.

[0039] 4. Calculate the main controlling factors for daily oil production increase in the initial stage of fracturing of the target well:

[0040] The current target well for fracturing is the oil well in the target block that needs to be fracturing. The main controlling factors of the target well for fracturing are the same as the main controlling factors of the target block in step 2 above.

[0041] 5. Set different proppant dosages for the target wells to be fractured. At the same time, input the main control factor data calculated in step 4 into the prediction model. Take the proppant dosage at the time when the daily oil increase is the largest in the initial stage of fracturing as the optimal proppant dosage for the target wells to be fractured.

[0042] The following are preferred embodiments of this disclosure.

[0043] Example

[0044] Figure 1 This is a flowchart of a machine learning-based method for optimizing proppant dosage in oil well fracturing, according to an embodiment of this disclosure; combined with Figure 1 As shown, the steps for optimizing the proppant dosage are as follows:

[0045] Step S1: Collect basic data on the geology, development, and technology of historical fracturing wells in the target block, including:

[0046] Geological data include the total number of fractured sub-layers, the total thickness of fractured sandstone, the total equivalent thickness of fractured layers, the average permeability of fractured layers, and the coefficient of variation of permeability of the original production layers before fracturing.

[0047] Development data includes daily oil production, daily fluid production, water cut, flowing pressure, submersion, dynamic fluid level, fluid production intensity, average number of connected directions of the fractured layer, average oil saturation of the fractured layer, and minimum injection-production well spacing of the fractured layer before fracturing.

[0048] The process data includes the amount of proppant added per unit sandstone thickness for whole-well fracturing, the amount of proppant used for whole-well fracturing, and the amount of fracturing fluid used per unit sandstone thickness for whole-well fracturing.

[0049] Step S2: Based on the geological, development, and technological data obtained in Step S1, determine the main controlling factors for daily oil production increase in the initial stage of fracturing using a correlation algorithm. Propionate dosage is one of the main controlling factors. The specific determination process is as follows:

[0050] In this embodiment, as Figure 2 As shown, Spearman rank correlation calculations were used to determine the correlation between the initial daily oil production of historically fractured wells and various influencing factors. The figure lists the correlations between 11 influencing factors and the initial daily oil production of fractured wells, including total fractured thickness, average daily oil production per well in the surrounding well network, daily oil production before fracturing, average number of connected directions of the fractured layer, average oil saturation of the fractured layer, proppant usage, minimum injection-production well spacing of the fractured layer, daily fluid production before fracturing, fracturing fluid usage, the ratio of fractured thickness to original production thickness, and the coefficient of variation of permeability of the production layer before fracturing.

[0051] exist Figure 2 In the process, based on the degree of relevance, the eight main influencing factors for daily oil production during the initial stage of fracturing were determined as follows: total fracturing equivalent thickness, average daily oil production per well in the surrounding well network, daily oil production before fracturing, average number of connected directions of the fracturing layer, average oil saturation of the fracturing layer, proppant usage, minimum injection-production well spacing of the fracturing layer, and daily fluid production before fracturing.

[0052] Step S3: Based on the main controlling factors of the target block obtained in Step S1, establish a prediction model for the daily oil production increase in the initial stage of fracturing of the target block. The specific establishment process is as follows:

[0053] A sample of data was established, including the initial daily oil production of historical fractured wells and the eight main control factors obtained through step S2. The ratio of training data to test historical data was set to 8:2, resulting in 879 training data and 220 test data.

[0054] In this embodiment, algorithms such as random forest, gradient boosting tree, and stacking ensemble regression are designed to perform machine learning tests on the consistency rate between the initial daily oil production of historical fractured wells and the main controlling factors. Figure 3 As shown in the figure, the test compliance rates of 14 machine learning algorithms, including Stacking, LightGBM, and ensemble boosting tree, are listed. Based on the principle of the highest test compliance rate, the CatBoost prediction model for daily oil production in the early stage of fracturing is selected, and the test data compliance rate reaches 78.62%.

[0055] In this embodiment, when the absolute value of the initial daily increase in oil production is greater than 5 tons, a relative error within 10% is marked as compliant; when the absolute value of the initial daily increase in oil production is less than or equal to 5 tons, an absolute error within 1 ton is marked as compliant.

[0056] Step S4: Determine the target well for fracturing and calculate the main controlling factors for daily oil production increase in the initial stage of fracturing. The specific calculation results are as follows:

[0057] Eight key control factors for the five oil wells currently planned for fracturing in the target block were calculated, and the results are shown in Table 1.

[0058] Table 1. Seven main controlling factors for fracturing effect in 5 wells

[0059]

[0060] Step S5: Substitute the main control factor data of the 5 target wells in Table 1 into the prediction model obtained in Step S3, set different proppant dosages for the current target wells, and predict the daily oil increase in the initial stage of fracturing. The proppant dosage with the maximum daily oil increase in the initial stage of fracturing in the prediction results is taken as the optimal proppant dosage for the current target well. The specific prediction process is as follows:

[0061] Referring to the historical proppant usage range of fractured wells in this block, different proppant usages were set for each target fractured well, starting with 2, ending with 10, and with a step size of 0.5. The main control factor data of the target fractured wells in Table 1 were substituted into the prediction model obtained in step 3, and the optimal proppant usage for the target fractured well was set to maximize the daily oil increase in the initial stage of fracture.

[0062] Taking the fractured well X431 as an example, as shown in Table 2, setting the proppant dosage to 5, 5.5, 6, and 6.5 results in the highest daily oil increase during the initial stage of fracturing, achieving the best fracturing effect. However, considering the premise of using as little proppant as possible, the optimal proppant dosage for fractured well X431 is determined to be 5m. 3 / m.

[0063] Table 2. Predicted Daily Oil Increase Results for Different Propionate Doses in X431 Fracturing Wells

[0064]

[0065] In summary, the machine learning-based proppant dosage optimization method for oil well fracturing provided in this disclosure comprehensively considers the main controlling factors affecting the initial daily oil production increase of fracturing wells, establishes a machine learning-based prediction model for the initial daily oil production increase of fracturing wells, sets different proppant dosages, and predicts the initial daily oil production increase of target fracturing wells. This not only solves the problem that manual qualitative analysis is difficult to accurately determine the fracturing effect, but also determines the optimal proppant dosage, further enhancing the fracturing potential.

[0066] The embodiments described above are merely illustrative of implementation methods of this disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent disclosure. It should be noted that those skilled in the art can make various modifications, equivalent substitutions, and improvements without departing from the concept of this disclosure, and these all fall within the protection scope of this disclosure. Therefore, the protection scope of this patent disclosure should be determined by the appended claims.

Claims

1. A method for optimizing proppant dosage in oil well fracturing based on machine learning, characterized in that, include: Identify the key controlling factors affecting daily oil production increase in the initial stage of fracturing in the target block, including proppant dosage. Based on machine learning, a quantitative prediction model for daily oil production during the initial stage of fracturing is established by utilizing the correlation between the daily oil production increase during the initial stage of fracturing and the main controlling factors. The main control factor data of the target fractured well are input into the prediction model. The daily oil increase in the initial stage of fracturing is predicted by setting different proppant dosages. The maximum daily oil increase in the initial stage of fracturing is taken as the optimal proppant dosage for the target fractured well.

2. The method for optimizing proppant dosage in oil well fracturing based on machine learning according to claim 1, characterized in that, The method for determining the main controlling factors affecting daily oil production increase in the initial stage of fracturing in the target block, including proppant dosage, includes: Based on the geological, development, and technological data of historical fractured wells in the target block, Spearman rank correlation analysis was used to calculate the correlation between the initial daily oil production of the historical fractured wells and various influencing factors, and the main controlling factors were determined.

3. The method for optimizing proppant dosage in oil well fracturing based on machine learning according to claim 2, characterized in that: Factors with a correlation greater than 0.3 are identified as the controlling factors.

4. The method for optimizing proppant dosage in oil well fracturing based on machine learning according to claim 2, characterized in that: The geological, development, and process data include the following: the geological data includes the total number of fracturing layers, the total thickness of fracturing sandstone, the total equivalent thickness of fracturing, the average permeability of the fracturing layer, and the coefficient of variation of the permeability of the original production layer before fracturing; the development data includes the daily oil production, daily fluid production, water cut, flowing pressure, submersion degree, dynamic fluid level, production intensity, the average number of connected directions of the fracturing layer, the average oil saturation of the fracturing layer, and the minimum injection-production well spacing of the fracturing layer; and the process data includes the amount of proppant added per unit sandstone thickness for whole-well fracturing, the amount of proppant used for whole-well fracturing, and the amount of fracturing fluid used per unit sandstone thickness for whole-well fracturing.

5. The method for optimizing proppant dosage in oil well fracturing based on machine learning according to any one of claims 1-4, characterized in that, The method for establishing a predictive model for daily oil production increase in the initial stage of fracturing based on machine learning and utilizing the correlation between the daily oil production increase in the initial stage of fracturing and the main controlling factors includes: The correlation is trained using several machine learning algorithms, and the ratio of training to testing historical data is set. The prediction model is determined based on the principle of achieving the highest test accuracy.

6. The method for optimizing proppant dosage in oil well fracturing based on machine learning according to claim 5, characterized in that: The ratio of training to testing historical data entries is 8:

2.

7. The method for optimizing proppant dosage in oil well fracturing based on machine learning according to claim 6, characterized in that: The machine learning algorithms include CatBoost, LightGBM, Extreme Random Tree, Gradient Boosting Tree, Random Forest, Theil-Sen Estimation Regression, Bayesian Ridge Regression, Minimum Angle Regression, Linear Regression, Ridge Regression, Stacking Ensemble Algorithm, Multilayer Perceptron Neural Network, Decision Tree, Elastic Regression, and Huber Regression Algorithm.

Images