Drilling operation cycle prediction method based on random forest regression model

Abstract

The invention discloses a drilling operation cycle prediction method based on a random forest regression model, and relates to the technical field of drilling data processing, and the method comprises the following steps: S1, collecting drilling data, and screening the drilling data to obtain standard drilling data; s2, performing classification processing on the standard drilling data to obtain a classification result; and S3, according to a classification result, obtaining a regression result based on a random forest, and predicting a drilling operation period by using the regression result. According to the method, the requirements of pre-drilling prediction, real-time parameter adjustment and the like can be met according to needs, training samples are continuously supplemented to optimize the model, the adaptability of the model to working characteristics of a research area can be improved, and the random forest regression prediction model has important significance on planning, cost evaluation and the like of a subsequent drilling scheme and has good popularization value.

Description

Technical Field

[0001] The present invention relates to the technical field of drilling data processing, and in particular to a drilling operation cycle prediction method based on a random forest regression model. Background Art

[0002] As a large-scale, high-investment and high-risk production activity, oil and gas drilling occupies an extremely important position in the field of energy exploration and development. The entire drilling operation process involves multiple links such as pre-drilling preparation, combined drilling tools, drilling operations and cementing operations, which require a lot of human, material and financial resources, and must also face many uncertain factors. An accurate prediction of the drilling operation cycle is of great significance to ensure that the overall goals of oil and gas field exploration and development are achieved as planned, and it also helps to reasonably plan and schedule subsequent work.

[0003] With the continuous advancement of technology, technologies such as machine learning and neural networks have made great progress in the fields of drilling machinery drilling speed prediction, well condition identification and time analysis, and their application also shows potential in drilling cycle prediction. At present, the drilling cycle prediction model focuses more on the overall operation process, pays less attention to the characteristics and connections of the well sections, and does not consider the actual operation parameters such as formation level, lithology and drilling fluid, resulting in limited model adaptability. The prediction of the drilling operation cycle should integrate more actual data, deeply explore the relationship between well sections, geological conditions and uncertain factors, and build a model with adaptability and updatability. Machine learning and neural networks are effective means of drilling cycle prediction. They can effectively deal with the multi-dimensional data generated during the drilling operation, are good at capturing complex nonlinear relationships between data, and the model has adaptability, updatability and excellent generalization ability. Therefore, the present invention proposes a machine learning model method based on random forest to describe the timeliness of drilling operations. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a drilling operation cycle prediction method based on a random forest regression model.

[0005] The technical solution of the present invention is: a drilling operation cycle prediction method based on a random forest regression model comprises the following steps:

[0006] S1. Collecting drilling data and screening the drilling data to obtain standard drilling data;

[0007] S2. Classify and process the standard drilling data to obtain classification results;

[0008] S3. According to the classification results, a regression result based on random forest is obtained, and the regression result is used to predict the drilling operation cycle.

[0009] Further, in S1, the drilling data includes well depth, wellbore size, drilling fluid parameters, layer thickness, and lithology thickness;

[0010] In S1, the drilling data is screened using the kernel K-means method to obtain standard drilling data.

[0011] Furthermore, S2 includes the following sub-steps:

[0012] S21. Use standard drilling data to construct a training set;

[0013] S22, randomly sampling the training set to generate a self-service sampling set;

[0014] S23. Classify the self-service sampling set to obtain a classification result.

[0015] Furthermore, in S22, the expression of the self-service sampling set Dt is:

[0016] Dt={(x i ,y i )},i∈{1,2,…,N};

[0017] In the formula, x i represents the drilling operation cycle of the i-th sample of the current node, y i represents the cumulative number of operation days of the i-th sample of the current node, and N represents the total number of samples of the current node.

[0018] Further, S23 includes the following sub-steps:

[0019] S231, calculating the intra-class consistency of each sample according to the automatic sampling set;

[0020] S232, calculating the inter-class difference according to the automatic sampling set;

[0021] S233, generating the best node classification model according to the inter-class differences and the intra-class consistency of each sample;

[0022] S234, using the node optimal classification model to classify the automatic sampling set, generate the best splitting point, and obtain the classification result.

[0023] Further, in S231, the intra-class consistency WCV of the k-th class of samples in the automatic sampling set is Relative The calculation formula of (k) is:

[0024]

[0025] Where VCV(k) represents the intra-class consistency of the k-th class of samples in the automatic sampling set, and Var(X) represents the variance of the automatic sampling set.

[0026] Furthermore, in S232, the calculation formula of the inter-class difference BCV is:

[0027]

[0028] In the formula, μ k represents the mean of the k-th sample in the automatic sampling set, μ represents the overall mean of the automatic sampling set, σ μ represents the standard deviation of the automatic sampling set, and K represents the total number of samples in the automatic sampling set.

[0029] Furthermore, in S233, the expression of the node optimal classification model CCI is:

[0030]

[0031] In the formula, K represents the total number of samples in the automatic sampling set, n k represents the number of samples in the kth class, N represents the total number of samples in the current node, BCV represents the difference between classes, WCV Relative (k) represents the intra-class consistency of the k-th class of samples in the automatic sampling set, and λ represents the weight of the penalty term.

[0032] Furthermore, in S234, the optimal split point G split The calculation formula is:

[0033]

[0034] Where N L Indicates the number of samples of the left child node, N R represents the number of samples of the right child node, N represents the total number of samples of the current node, G(t L ) represents the calculation result of the purity metric function of the left child node, G(t R ) represents the calculation result of the purity measurement function of the right child node, t L Represents the left child node, which is the left child node generated after the current node is split, and contains samples that are classified to the left according to the splitting condition of a certain feature. R Represents the right child node, which is the right child node generated after the current node is split, and contains samples that are classified to the right according to the splitting condition of a certain feature.

[0035] Furthermore, in S3, the regression result R final The calculation formula is:

[0036]

[0037] In the formula, R j represents the regression result of the jth decision tree, and J represents the number of decision trees.

[0038] The beneficial effects of the present invention are as follows: the present invention optimizes the factors that affect the timeliness of drilling operations, and adopts a random forest regression model to predict the drilling operation cycle for multiple drilling operations, thereby improving the problem of insufficient prediction of the drilling operation cycle of previous researchers. The prediction accuracy of each drilling model is high and can meet actual needs. The present invention can realize pre-drilling prediction and real-time parameter adjustment as needed, continuously supplement training samples to optimize the model, and improve the adaptability of the model to the operation characteristics of the study area. The random forest regression prediction model is of great significance to the planning of subsequent drilling plans, cost evaluation, etc., and has good promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Flow chart of the drilling operation cycle prediction method based on random forest regression model. DETAILED DESCRIPTION

[0040] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0041] like Figure 1 As shown, the present invention provides a drilling operation cycle prediction method based on a random forest regression model, comprising the following steps:

[0042] S1. Collecting drilling data and screening the drilling data to obtain standard drilling data;

[0043] S2. Classify and process the standard drilling data to obtain classification results;

[0044] S3. According to the classification results, a regression result based on random forest is obtained, and the regression result is used to predict the drilling operation cycle.

[0045] The random forest algorithm is an integrated machine learning method with a basic unit of CART decision tree, which can be used for classification and regression problems. In random forest regression, each decision tree is independently constructed, and samples are usually extracted from the original data set by self-service sampling. In addition, the feature selection during node splitting is also based on a random subset of features, thereby increasing the diversity of the forest and further reducing the risk of overfitting. The output of the model is obtained by summarizing the results of all decision trees (such as averaging). The random forest regression algorithm has good robustness to noisy data and missing outliers. In addition, the random forest algorithm reduces the dependence on data preprocessing during the training phase, and its implementation process is standardized and has high learning efficiency. Since the characteristic parameters of the drilling operation cycle prediction contain complex and interdependent classification features, when facing more complex situations, the model needs to have a strong tolerance for random errors and missing values. Based on the advantages of the random forest regression model algorithm, the present invention selects the random forest regression model for research.

[0046] In the embodiment of the present invention, in S1, the drilling data includes well depth, wellbore size, drilling fluid parameters, layer thickness and lithology thickness;

[0047] In S1, the drilling data is screened using the kernel K-means method to obtain standard drilling data.

[0048] In this embodiment of the present invention, S2 includes the following sub-steps:

[0049] S21. Use standard drilling data to construct a training set;

[0050] S22, randomly sampling the training set to generate a self-service sampling set;

[0051] S23. Classify the self-service sampling set to obtain a classification result.

[0052] In the embodiment of the present invention, in S22, the expression of the self-service sampling set Dt is:

[0053] Dt={(x i ,y i )},i∈{1,2,…,N};

[0054] In the formula, x i represents the drilling operation cycle of the i-th sample of the current node, y i represents the cumulative number of operation days of the i-th sample of the current node, and N represents the total number of samples of the current node.

[0055] In this embodiment of the present invention, S23 includes the following sub-steps:

[0056] S231, calculating the intra-class consistency of each sample according to the automatic sampling set;

[0057] S232, calculating the inter-class difference according to the automatic sampling set;

[0058] S233, generating the best node classification model according to the inter-class differences and the intra-class consistency of each sample;

[0059] S234, using the node optimal classification model to classify the automatic sampling set, generate the best splitting point, and obtain the classification result.

[0060] In the embodiment of the present invention, in S231, the intra-class consistency WCV of the k-th class of samples in the automatic sampling set is Relative The calculation formula of (k) is:

[0061]

[0062] Where VCV(k) represents the intra-class consistency of the k-th class of samples in the automatic sampling set, and Var(X) represents the variance of the automatic sampling set.

[0063] In the embodiment of the present invention, in S232, the calculation formula of the inter-class difference BCV is:

[0064]

[0065] In the formula, μ k represents the mean of the k-th sample in the automatic sampling set, μ represents the overall mean of the automatic sampling set, σ μ represents the standard deviation of the automatic sampling set, and K represents the total number of samples in the automatic sampling set.

[0066] In the embodiment of the present invention, in S233, the expression of the node optimal classification model CCI is:

[0067]

[0068] In the formula, K represents the total number of samples in the automatic sampling set, n k represents the number of samples in the kth class, N represents the total number of samples in the current node, BCV represents the difference between classes, WCV Relative (k) represents the intra-class consistency of the k-th class of samples in the automatic sampling set, and λ represents the weight of the penalty term.

[0069] In the embodiment of the present invention, in S234, the optimal splitting point G split The calculation formula is:

[0070]

[0071] Where N L Indicates the number of samples of the left child node, N R represents the number of samples of the right child node, N represents the total number of samples of the current node, G(t L ) represents the calculation result of the purity metric function of the left child node, G(t R ) represents the calculation result of the purity measurement function of the right child node, t L Represents the left child node, which is the left child node generated after the current node is split, and contains samples that are classified to the left according to the splitting condition of a certain feature. R Represents the right child node, which is the right child node generated after the current node is split, and contains samples that are classified to the right according to the splitting condition of a certain feature.

[0072] In the embodiment of the present invention, in S3, the regression result R final The calculation formula is:

[0073]

[0074] In the formula, R jrepresents the regression result of the jth decision tree, and J represents the number of decision trees.

[0075] The following is an explanation in conjunction with a specific embodiment. Take Well S15 as an example. The well is deployed at the front edge of the inversion structural belt in the eastern area of ​​the Wushi Sag in the Beibu Gulf Basin. It is an evaluation well of the three-opening directional type. Its data did not participate in the model training. The well parameter data are shown in Table 1. The model pre-drilling prediction results are shown in Table 2. The cumulative time from one to three openings is 3.46d, 8.43d and 18.34d, respectively, and the difference in days from the actual time is 0.02d, 0.57d and 1.25d, respectively. Comparing the model prediction results with the original design cumulative days calculated by the quota, it can be found that the model results are closer to the actual cumulative days, which greatly improves the efficiency and accuracy of the drilling operation cycle prediction, and can provide a reference for subsequent drilling planning and cost evaluation.

[0076] Table 1

[0077]

[0078]

[0079] Table 2

[0080]

[0081] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A drilling operation cycle prediction method based on a random forest regression model, characterized in that: The following steps are involved: S1. Collecting drilling data and screening the drilling data to obtain standard drilling data; S2. Classify and process the standard drilling data to obtain classification results; S3. According to the classification results, a regression result based on random forest is obtained, and the regression result is used to predict the drilling operation cycle.

2. The drilling operation cycle prediction method based on random forest regression model according to claim 1 is characterized in that: In S1, the drilling data includes well depth, wellbore size, drilling fluid parameters, layer thickness and lithology thickness; In S1, the drilling data is screened using the kernel K-means method to obtain standard drilling data.

3. The drilling operation cycle prediction method based on random forest regression model according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21. Use standard drilling data to construct a training set; S22, randomly sampling the training set to generate a self-service sampling set; S23. Classify the self-service sampling set to obtain a classification result.

4. The drilling operation cycle prediction method based on random forest regression model according to claim 3 is characterized in that: In S22, the expression of the self-service sampling set Dt is: Dt={(x i ,y i )},i∈{1,2,…,N}; In the formula, x i represents the drilling operation cycle of the i-th sample of the current node, y i represents the cumulative number of operation days of the i-th sample of the current node, and N represents the total number of samples of the current node.

5. The drilling operation cycle prediction method based on random forest regression model according to claim 3 is characterized in that: The S23 comprises the following sub-steps: S231, calculating the intra-class consistency of each sample according to the automatic sampling set; S232, calculating the inter-class difference according to the automatic sampling set; S233, generating the best node classification model according to the inter-class differences and the intra-class consistency of each sample; S234, using the node optimal classification model to classify the automatic sampling set, generate the best splitting point, and obtain the classification result.

6. The method for predicting drilling operation cycle based on random forest regression model according to claim 5, characterized in that: In S231, the intra-class consistency WCV of the k-th class of samples in the automatic sampling set Relative The calculation formula of (k) is: Where WCV(k) represents the intra-class consistency of the k-th class of samples in the automatic sampling set, and Var(X) represents the variance of the automatic sampling set.

7. The method for predicting drilling operation cycle based on random forest regression model according to claim 5, characterized in that: In S232, the calculation formula of the inter-class difference BCV is: In the formula, μ k represents the mean of the k-th sample in the automatic sampling set, μ represents the overall mean of the automatic sampling set, σ μ represents the standard deviation of the automatic sampling set, and K represents the total number of samples in the automatic sampling set.

8. The method for predicting drilling operation cycle based on random forest regression model according to claim 5, characterized in that: In S233, the expression of the node optimal classification model CCI is: In the formula, K represents the total number of samples in the automatic sampling set, n k represents the number of samples in the kth class, N represents the total number of samples in the current node, BCV represents the difference between classes, WCV Relative (k) represents the intra-class consistency of the k-th class of samples in the automatic sampling set, and λ represents the weight of the penalty term.

9. The method for predicting drilling operation cycle based on random forest regression model according to claim 5, characterized in that: In S234, the optimal splitting point G split The calculation formula is: Where N L Indicates the number of samples of the left child node, N R represents the number of samples of the right child node, N represents the total number of samples of the current node, G(t L ) represents the calculation result of the purity metric function of the left child node, G(t R ) represents the calculation result of the purity measurement function of the right child node, t L represents the left child node, t R Represents the right child node.

10. The method for predicting drilling operation cycle based on random forest regression model according to claim 1, characterized in that: In S3, the regression result R final The calculation formula is: In the formula, R j represents the regression result of the jth decision tree, and J represents the number of decision trees.

Images