Oil well hole patching effect prediction method based on machine learning

Abstract

The invention relates to the technical field of oil and gas field development and big data, in particular to an oil well hole patching effect prediction method based on machine learning. The method comprises the steps that a target block is determined, and geological data, development data and process data of historical hole patching oil wells of the target block are acquired; according to the obtained geological data, development data and process data, determining main control factors of initial daily oil increase of the refilling well of the target block based on correlation analysis, and establishing a machine learning prediction model of the initial daily oil increase of the refilling well of the target block; determining a current hole patching target well, and calculating main control factors of initial daily oil increase; and substituting the obtained main control factors of the hole patching target well into the prediction model to obtain a prediction result of initial daily oil increase of the hole patching target well, thereby providing a basis for implementation of the hole patching well. According to the method provided by the invention, the main control factors influencing the initial daily oil increase of the patching well are comprehensively considered, the model is established based on the machine learning algorithm, and prediction is carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas field development and big data, and in particular to a method for predicting oil well plugging effects based on machine learning. Background Art

[0002] An oil field refers to the sum of oil and gas reservoirs within the same oil and gas producing area under the control of a single geological structure or stratigraphic factors. An oil and gas field may have one or more oil and gas reservoirs. An area that is mainly composed of oil reservoirs is called an oil field, and an area that is mainly composed of gas reservoirs is called a gas field.

[0003] Among them, onshore oil fields are one of the important oil field resources in my country, and old oil fields refer to oil fields that have entered the late stage of development. At present, old onshore oil fields in my country have generally entered the high water content development stage. Oil well re-perforation is one of the important means of increasing production in old oil fields. Oil field re-perforation mostly uses manual qualitative analysis, and it is impossible to achieve quantitative analysis of the re-perforation effect. However, the injection-production relationship and oil-water distribution of old oil fields are complex, and the re-perforation effect involves many factors. How to accurately predict the re-perforation effect, how to judge the re-perforation potential and improve the re-perforation effect are the current problems. Therefore, in view of the above shortcomings, a method for predicting the effect of oil well re-perforation based on machine learning is proposed. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] The present invention provides a method for predicting the effect of oil well plugging based on machine learning, so as to overcome the problems in the prior art that it is difficult to accurately determine the plugging effect by manual qualitative analysis on site, it is impossible to achieve quantitative analysis of the plugging effect, and it is impossible to further judge the potential of plugging.

[0006] (II) Technical solution

[0007] In order to solve the above problems, the present invention provides a method for predicting the effect of oil well plugging based on machine learning, comprising:

[0008] Step S1: determine the target block, and obtain geological data, development data and process data of the historical plugging oil wells in the target block;

[0009] Step S2: According to the geological data, development data and process data obtained in step S1, the main controlling factors of the initial daily oil production increase of the plugging wells in the target block are determined based on correlation analysis;

[0010] Step S3: establishing a machine learning prediction model for the initial daily oil production increase of the plugging wells in the target block according to the main controlling factors of the target block obtained in step S2;

[0011] Step S4: determining the current target well for perforation repair, and calculating the main controlling factors for the initial daily oil production increase of the target well for perforation repair;

[0012] Step S5: Substitute the main controlling factors of the target well for perforation repair obtained in step S4 into the prediction model obtained in step S3 to obtain the prediction result of the initial daily oil production increase of the target well for perforation repair, so as to provide a basis for the implementation of perforation repair.

[0013] Preferably, in step S1, the geological data include the thickness of the patching, the proportion of the patching thickness to the original production thickness, the number of patching layers, the average permeability of the patching layer and the coefficient of variation of the permeability of the original production layer before patching; the development data include the daily oil production, daily liquid production, water content, flow pressure, submergence, dynamic liquid level, liquid production intensity, the average number of connection directions of the patching layer, the average oil saturation of the patching layer and the minimum injection-production well spacing of the patching layer before patching; the process data include the perforation method, perforation density and patching phase.

[0014] Preferably, in step S2, the main controlling factor of the daily oil production increase in the initial stage of the plugging well in the target block is a factor with a correlation degree greater than 0.3.

[0015] Preferably, the main controlling factors of the target block are calculated by the Spearman rank correlation analysis algorithm to calculate the correlation between the initial daily oil increase of the historical plugging wells and various influencing factors, and the calculation formula is:

[0016]

[0017] In the formula, r is the correlation degree; d i is the order difference of two data; n is the number of data of historical wells with supplementary holes.

[0018] Preferably, in step S3, the machine learning prediction model is based on the main controlling factor data and initial daily oil increase data of historical re-perforation wells, is trained using multiple machine learning algorithms, and the ratio of the number of training and test historical data is set. The initial daily oil increase prediction model for re-perforation is preferably determined based on the principle of the highest test compliance rate.

[0019] Preferably, the ratio of the number of historical data items of training and testing is set to 8:2.

[0020] Preferably, the compliance rate is that the error rate of the absolute value of the initial daily oil increase is within 10%, the limit value of the absolute value of the initial daily oil increase is 3 tons, when the absolute value of the initial daily oil increase is greater than 3 tons, the relative error is within 10% and is marked as compliance; when the absolute value of the initial daily oil increase is less than or equal to 3 tons, the absolute error is within 0.3 tons and is marked as compliance.

[0021] Preferably, the calculation method of the machine learning prediction model includes designing decision trees, random forests, support vector machines, multi-layer perceptron neural networks, integrated boosting tree regression, gradient descent regression, bagging and stacking integrated regression algorithms.

[0022] Preferably, in step S4, the current re-perforation target well is an oil well in the target block that needs to be re-perforated, and the main controlling factors of the re-perforation target well are consistent with the main controlling factors of the target block.

[0023] (III) Beneficial effects

[0024] The method for predicting the effect of oil well plugging based on machine learning provided by the present invention comprehensively considers the main controlling factors affecting the initial daily oil increase of the plugging well, establishes a prediction model for the initial daily oil increase of the plugging well based on machine learning, and predicts the initial daily oil increase of the plugging target well. This can not only solve the problem that manual qualitative analysis is difficult to accurately determine the plugging effect, but also realize quantitative analysis of the plugging effect, providing a new way to further optimize the plugging potential and improve the plugging effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a method for predicting the effect of oil well plugging based on machine learning according to an embodiment of the present invention;

[0026] Figure 2 This is a diagram showing the influence of various factors on the initial daily oil production increase of a plugging well according to an embodiment of the present invention;

[0027] Figure 3 This is a graph of the compliance rates of various machine learning algorithms tested in the embodiments of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Figure 1 FIG. 1 is a flow chart of a method for predicting the effect of oil well plugging based on machine learning according to an embodiment of the present invention. Figure 1 As shown, the present invention provides a method for predicting the effect of oil well plugging based on machine learning, comprising:

[0030] Step S1: determine the target block, and obtain geological data, development data and process data of the historical plugging oil wells in the target block;

[0031] Step S2: According to the geological data, development data and process data obtained in step S1, the main controlling factors of the initial daily oil production increase of the plugging wells in the target block are determined based on correlation analysis;

[0032] Step S3: establishing a machine learning prediction model for the initial daily oil production increase of the plugging wells in the target block according to the main controlling factors of the target block obtained in step S2;

[0033] Step S4: determining the current target well for perforation repair, and calculating the main controlling factors for the initial daily oil production increase of the target well for perforation repair;

[0034] Step S5: Substitute the main controlling factors of the target well for perforation repair obtained in step S4 into the prediction model obtained in step S3 to obtain the prediction result of the initial daily oil production increase of the target well for perforation repair, so as to provide a basis for the implementation of perforation repair.

[0035] In practical applications, in step S1, geological data include the thickness of the perforation patching, the proportion of the perforation patching thickness to the original production thickness, the number of perforation patching layers, the average permeability of the perforation patching layer, and the coefficient of variation of the permeability of the original production layer before perforation patching; development data include daily oil production, daily liquid production, water content, flow pressure, subsidence, dynamic liquid level, liquid production intensity, the average number of connected directions of the perforation patching layer, the average oil saturation of the perforation patching layer, and the minimum injection-production well spacing of the perforation patching layer before perforation patching; process data include perforation mode, perforation density, and perforation patching phase.

[0036] In this prediction method, in step S2, the main controlling factors of the target block are determined by calculating the correlation between the initial daily oil production increase of the historical plugging wells and various influencing factors according to the Spearman rank correlation analysis algorithm.

[0037]

[0038] In the formula, r is the correlation degree; d i is the order difference of two data; n is the number of data of historical wells with supplementary holes.

[0039] In practical applications, the calculated correlation levels are sorted, and factors with a correlation level greater than 0.3 are identified as the main factors affecting daily oil production in the initial stage of perforation repair.

[0040] In practical applications, in step S3, the machine learning prediction model is trained through a variety of machine learning algorithms. The data set uses the initial daily oil increase of historical re-perforation wells as the label value, the main control factor data as the training feature, and the ratio of the number of training data to the number of test data is set to 8:2. The algorithm with the highest test compliance rate is selected to establish the final machine learning model. Figure 3 This is a graph showing the compliance rate of various machine learning algorithms tested in the embodiments of the present invention, such as Figure 3 As shown in the figure, the calculation methods of machine learning prediction models include designing decision trees, random forests, support vector machines, multi-layer perceptron neural networks, integrated boosting tree regression, gradient descent regression, bagging and stacking integrated regression algorithms.

[0041] Among them, the compliance rate is that the error rate of the absolute value of the initial daily oil increase is within 10%, and the limit value of the absolute value of the initial daily oil increase is 3 tons. When the absolute value of the initial daily oil increase is greater than 3 tons, the relative error is within 10% and is marked as compliant; when the absolute value of the initial daily oil increase is less than or equal to 3 tons, the absolute error is within 0.3 tons and is marked as compliant.

[0042] In the prediction method, in step S4, the main controlling factors of the target well for supplementary drilling are consistent with the main controlling factors of the target block.

[0043] In addition, the plugging target well is an oil well that needs to be plugged. In practical applications, it is necessary to obtain dynamic and static data of the plugging target well, and calculate the main controlling factors of the plugging target well based on the dynamic and static data.

[0044] In this prediction method, in step S5, the main controlling factors of the target well for re-perforation that need to be obtained are input into a machine learning prediction model established according to the main controlling factors of the target block to predict the initial daily oil increase of the target well for re-perforation, further providing a basis for the implementation of re-perforation.

[0045] The method for predicting the effect of oil well plugging based on machine learning provided by the present invention comprehensively considers the main influencing factors affecting the initial daily oil increase of oil well plugging wells, establishes a prediction model for the initial daily oil increase of plugging wells based on machine learning, solves the problem that manual qualitative analysis is difficult to accurately determine the plugging effect, and can provide guidance for optimizing the plugging potential and improving the plugging effect.

[0046] Taking a development block of Daqing Oilfield as an example, based on the geological, development and process data information of the historical plugging wells in the block, the main influencing factors affecting the initial daily oil increase of the plugging wells are determined through correlation analysis. Machine learning is performed between the initial daily oil increase of the plugging wells and the influencing factors, and a quantitative prediction model for the initial daily oil increase of the plugging wells is established. Then, the initial daily oil increase of the current plugging target well is predicted. The implementation process of this oil well plugging effect prediction method based on machine learning is described in detail:

[0047] Step 1: Determine the target block and obtain the geological data, development data and process data of the historical re-perforation oil wells in the target block.

[0048] In this embodiment, the geological data include the thickness of the patched hole, the ratio of the patched hole thickness to the original production thickness, the number of patched hole layers, the average permeability of the patched hole layer, and the permeability variation coefficient of the original production layer before the patched hole;

[0049] Development data include daily oil production, daily liquid production, water content, flow pressure, subsidence, dynamic liquid level, liquid production intensity, average number of connected directions of the perforated layer, average oil saturation of the perforated layer, and minimum injection-production well spacing of the perforated layer before perforation repair;

[0050] The process data include perforation mode, perforation density and perforation filling phase.

[0051] Step 2: According to the acquired geological data, development data and process data, the main controlling factors of the initial daily oil production increase of the plugging wells in the target block are determined based on correlation analysis.

[0052] In this embodiment, the correlation between the initial daily oil increase of the historical re-perforation wells and various influencing factors is calculated by Spearman rank correlation, and the eight main influencing factors of the initial daily oil increase of the re-perforation wells are determined according to the correlation degree. Figure 2 Graph showing the influence of various factors on the initial daily oil increase of the plugging well according to the embodiment of the present invention. Figure 2 As shown in the figure, the eight main influencing factors are: daily liquid production before perforation patching, daily oil production before perforation patching, submergence before perforation patching, perforation patching thickness, proportion of perforation patching thickness to original production thickness, average number of connected directions of perforation patching layer, average oil saturation of perforation patching layer and minimum injection-production well spacing of perforation patching layer.

[0053] Step 3: Based on the main controlling factors of the target block, a machine learning prediction model for the initial daily oil production increase of the plugging wells in the target block is established.

[0054] In this embodiment, sample data including the initial daily oil production increase of historical plugging wells and the eight main control factors obtained in step 2 are established, and the ratio of the number of training data to the number of test historical data is set to 8:2, corresponding to 614 training data and 153 test data;

[0055] At the same time, a series of machine learning algorithms, including decision tree design, random forest, support vector machine, multi-layer perceptron neural network, integrated boosting tree regression, gradient descent regression, bagging and stacking integrated regression, were used to conduct machine learning on the relationship between the initial daily oil increase of historical re-perforation wells and the main controlling factors. The principle of highest test compliance rate was adopted. In actual application, after comparison, it was preferred to establish an initial daily oil increase prediction model for re-perforation wells based on the Stacking integrated regression algorithm, and the test data compliance rate reached 76.6%.

[0056] Step 4: Determine the current target well for perforation repair and calculate the main controlling factors for the initial daily oil increase of the target well for perforation repair.

[0057] In this embodiment, 5 oil wells that are currently planned to be re-drilled in the target block are selected as re-drilling target wells, and the static data of the 5 re-drilling target wells are respectively obtained, and the main controlling factors of each re-drilling target well are calculated according to their respective static data. It should be noted that the types of the main controlling factors of the re-drilling target wells are consistent with the types of the main controlling factors of the target block, both of which are 8 items. Table 1 shows the 8 main controlling factors of the re-drilling effect of the 5 wells, as shown in Table 1, which is the summary result of the main controlling factors of the 5 re-drilling target wells in this embodiment.

[0058] Table 1 Eight main controlling factors of the plugging effect in five wells

[0059]

[0060]

[0061] Step 5: Substitute the main controlling factors of the target well for perforation repair into the prediction model to obtain the prediction results of the initial daily oil production increase of the target well for perforation repair, so as to provide a basis for the implementation of perforation repair.

[0062] In this embodiment, the main controlling factor data of the five re-perforating target wells obtained by the calculation in step 4, that is, the data in Table 1, are substituted into the machine learning prediction model of the target block established in step 3, that is, the prediction model of the initial daily oil increase of the re-perforating wells based on the Stacking integrated regression algorithm. After calculation by the prediction model, the prediction results of the initial daily oil increase of the re-perforating wells of the re-perforating target wells are obtained. Table 2 is the prediction results of the initial daily oil increase of the re-perforating wells of the five wells. As shown in Table 2, the actual initial daily oil increase of the re-perforating wells of the five wells is slightly different from the predicted initial daily oil increase of the re-perforating wells, which has practical significance and further verifies the rationality of the prediction method. At the same time, according to the obtained calculation results, it can provide guidance for optimizing the re-perforating potential and improving the re-perforating effect.

[0063] Table 2 Prediction results of initial daily oil production increase of five wells

[0064] hashtag Actual initial daily increase in oil (tons) Forecasted initial daily oil production (tons) Absolute error (tons) X628 0.53 0.63 0.09 X721 0.41 0.59 0.18 X632 0.90 0.81 0.09 X427 0.33 0.47 0.13 X615 0.96 0.80 0.16

[0065] The method for predicting the effect of oil well plugging based on machine learning provided by the present invention comprehensively considers the main controlling factors affecting the initial daily oil increase of the plugging well, establishes a prediction model for the initial daily oil increase of the plugging well based on machine learning, and predicts the initial daily oil increase of the plugging target well. This can not only solve the problem that manual qualitative analysis is difficult to accurately determine the plugging effect, but also realize quantitative analysis of the plugging effect, providing a new way to further optimize the plugging potential and improve the plugging effect.

[0066] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The patent protection scope of the present invention should be defined by the claims.

Claims

1. A method for predicting the effect of oil well plugging based on machine learning, characterized in that: include: Step S1: determine the target block, and obtain geological data, development data and process data of the historical plugging oil wells in the target block; Step S2: According to the geological data, development data and process data obtained in step S1, the main controlling factors of the initial daily oil production increase of the plugging wells in the target block are determined based on correlation analysis; Step S3: establishing a machine learning prediction model for the initial daily oil production increase of the plugging wells in the target block according to the main controlling factors of the target block obtained in step S2; Step S4: determining the current target well for perforation repair, and calculating the main controlling factors for the initial daily oil production increase of the target well for perforation repair; Step S5: Substitute the main controlling factors of the target well for perforation repair obtained in step S4 into the prediction model obtained in step S3 to obtain the prediction result of the initial daily oil production increase of the target well for perforation repair, so as to provide a basis for the implementation of the perforation repair well.

2. The method for predicting the effect of oil well plugging based on machine learning according to claim 1, characterized in that: In step S1, the geological data include the thickness of the patching, the proportion of the patching thickness to the original production thickness, the number of patching layers, the average permeability of the patching layer and the coefficient of variation of the permeability of the original production layer before patching; the development data include the daily oil production, daily liquid production, water content, flow pressure, submergence, dynamic liquid level, liquid production intensity, the average number of connected directions of the patching layer, the average oil saturation of the patching layer and the minimum injection-production well spacing of the patching layer before patching; the process data include the perforation method, perforation density and patching phase.

3. The method for predicting the effect of oil well plugging based on machine learning according to claim 1, characterized in that: In step S2, the main controlling factors of the daily oil production increase in the initial stage of the perforation repair of the target block are factors with a correlation degree greater than 0.

3.

4. The method for predicting the effect of oil well plugging based on machine learning according to claim 1, characterized in that: The main controlling factors of the target block are calculated by the Spearman rank correlation analysis algorithm to calculate the correlation between the initial daily oil increase of the historical plugging wells and various influencing factors. The calculation formula is: In the formula, r is the correlation degree; d i is the order difference of two data; n is the number of data of historical wells with supplementary holes.

5. The method for predicting the effect of oil well plugging based on machine learning according to claim 1, characterized in that: In step S3, the machine learning prediction model is based on the main controlling factor data and initial daily oil increase data of historical plugging wells, is trained using multiple machine learning algorithms, sets the ratio of training to test historical data, and preferably determines the initial daily oil increase prediction model for plugging based on the principle of the highest test compliance rate.

6. The method for predicting the effect of oil well plugging based on machine learning according to claim 5, characterized in that: The algorithms of the machine learning prediction model include designing decision trees, random forests, support vector machines, multi-layer perceptron neural networks, integrated boosting tree regression, gradient descent regression, bagging and stacking integrated regression algorithms.

7. The method for predicting the effect of oil well plugging based on machine learning according to claim 5, characterized in that: The ratio of the number of historical data items for training and testing is set to 8:

2.

8. The method for predicting the effect of oil well plugging based on machine learning according to claim 5, characterized in that: The compliance rate is that the error rate of the absolute value of the initial daily oil increase is within 10%, and the limit value of the absolute value of the initial daily oil increase is 3 tons. When the absolute value of the initial daily oil increase is greater than 3 tons, the relative error is within 10% and is marked as compliant; when the absolute value of the initial daily oil increase is less than or equal to 3 tons, the absolute error is within 0.3 tons and is marked as compliant.

9. The method for predicting the effect of oil well plugging based on machine learning according to claim 1, characterized in that: In step S4, the current re-perforation target well is an oil well in the target block that needs to be re-perforated, and the main controlling factors of the re-perforation target well are consistent with the main controlling factors of the target block.

Images