Yield prediction method based on machine learning

Abstract

The invention belongs to the field of oil and gas field development, and particularly relates to a machine learning-based yield prediction method, which comprises the steps of S1, collecting original data, and preprocessing the original data; s2, establishing a random forest liquid production index prediction model, and inputting the preprocessed data into the liquid production index prediction model for training; s3, performing performance analysis on the liquid production index prediction model by utilizing multiple evaluation methods, and adjusting model parameters according to an analysis result until a preset precision standard is met; and S4, carrying out liquid production index prediction on the trained liquid production index prediction model, and judging the productivity of the oil field through the liquid production index. By establishing the random forest liquid production index prediction model, the liquid production index is predicted through the liquid production index model, so that the productivity of the oil field is predicted, the influence of interference factors on the prediction result is small, and the stability and precision of the prediction system are high.

Description

Technical Field

[0001] The present invention belongs to the field of oil and gas field development, and in particular relates to a production prediction method based on machine learning. Background Art

[0002] Oilfield development is characterized by high technology and high investment, so accurate capacity evaluation is of great significance. In reservoir management, capacity evaluation can help operators evaluate the results of mining, thereby providing adequate preparation for corporate decision-making. In addition, capacity evaluation can also help companies take proactive measures to improve development processes and thus increase production. Through capacity evaluation, companies can better analyze their benefits, grasp future development trends, avoid potential risks, and avoid investment losses.

[0003] Capacity evaluation is the key to the study of low permeability reservoir reserves and preliminary research. Its evaluation methods mainly include capacity testing and capacity prediction. Among them, capacity testing is currently the most effective method to accurately evaluate the initial capacity of low permeability reservoirs. It can not only realize the capacity evaluation of test wells / reservoirs, but also realize the understanding of regional capacity laws and guide the capacity prediction of untested wells / reservoirs. Although a lot of research has been done at home and abroad on the seepage mechanism of low permeability reservoirs, the establishment of relevant capacity equations, numerical simulation and capacity prediction, which has a certain reference value for capacity prediction, due to the complex spatial distribution and strong heterogeneity of low permeability oil and gas reservoirs, the relevant results are not universal, and there are still problems such as low accuracy of oilfield capacity prediction and inaccurate prediction results, which affect production prediction. Summary of the invention

[0004] The technical problem solved by the present invention is to provide a production prediction method based on machine learning, which predicts the liquid production index of an oil field through machine learning, thereby predicting the oil field production capacity, solving the problems of low accuracy in oil field production capacity prediction and inaccurate prediction results, and thereby improving the accuracy of production prediction.

[0005] The basic solution provided by the present invention is a yield prediction method based on machine learning, which specifically includes the following steps:

[0006] S1: Collect and preprocess raw data;

[0007] S2: Establish a random forest liquid production index prediction model, and input the preprocessed data into the liquid production index prediction model for training;

[0008] S3: Analyze the performance of the liquid production index prediction model using a variety of evaluation methods, and adjust the model parameters according to the analysis results until the predetermined accuracy standards are met;

[0009] S4: Use the trained liquid production index prediction model to predict the liquid production index, and judge the oil field production capacity through the liquid production index.

[0010] Preferably, the original data include engineering parameters and geological parameters. The engineering parameters include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, and liquid production index. The geological parameters include formation crude oil viscosity, mud content, porosity, water saturation, logging permeability, and original formation pressure.

[0011] Preferably, in step S1, data preprocessing includes abnormal data processing, missing value processing, and data standardization processing. The abnormal data processing and missing value processing are performed on the data through data cleaning, and the data is standardized through normalization.

[0012] Preferably, in step S2, the steps of establishing a random forest liquid production index prediction model are as follows:

[0013] S2-1) defining input parameters and output parameters of the liquid production index prediction model, wherein the input parameters include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, liquid production index, formation crude oil viscosity, mud content, porosity, water saturation, logging permeability, and original formation pressure, and the output parameter is the liquid production index;

[0014] S2-2) Calculate the Gini coefficient of each parameter and select characteristic parameters according to the Gini coefficient;

[0015] S2-3) constructing a random forest regression tree model based on the selected characteristic parameters, wherein the regression tree model is expressed as:

[0016]

[0017] In the formula, m is the number of units in the feature space, M is the total number of feature parameters, and c m is the optimal output weight, I is the indicator function (if x belongs to the mth region R m , then the indicator function value is 1; if x does not belong to the mth region R m , then the value of the indicator function is 0);

[0018] S2-4) perform hyperparameter optimization to determine the number of regression trees;

[0019] S2-5) According to the number of regression trees and the regression tree model, a random forest liquid production index prediction model is built.

[0020] Further preferably, in step S2-2), the screened features include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, liquid production index, formation crude oil viscosity, mud content, porosity, water saturation, and logging permeability.

[0021] Further preferably, in step S2-4), the number of regression trees is 300.

[0022] Preferably, in step S3, R 2 The liquid production index prediction model is analyzed by two or more evaluation methods including MAE, MSE and RMSE.

[0023] Further preferably, the predetermined accuracy standard includes a first preset accuracy and a second preset accuracy, wherein the first preset accuracy is used to compare with the analysis results of the MAE, MSE, and RMSE evaluation methods, and the second preset accuracy is used to compare with the R 2 The analysis results of the two evaluation methods are compared. If both comparison results meet the accuracy standard requirements, the liquid production index prediction model meets the accuracy standard requirements.

[0024] More preferably, R 2 The evaluation method is expressed as

[0025]

[0026] In the formula, y i is the true value of the liquid production index, is the liquid production index value predicted by the model, It is the average value of the true values ​​of all liquid production indices.

[0027] Preferably, in step S4, the production capacity of the oil well is determined based on the liquid production index combined with the production pressure difference, and the liquid production index and the production capacity have the following relationship:

[0028] Q=J×ΔP

[0029] Where J is the liquid production index, Q is the liquid production, and ΔP is the production pressure difference.

[0030] The principles and advantages of the present invention are:

[0031] 1. A liquid production index prediction model is established based on machine learning. The liquid production index is predicted through the liquid production index model, thereby predicting the oil field production capacity. The machine learning model can process a large amount of historical data and learn complex patterns and trends from it. It can more accurately predict the future production capacity of the oil field. Accurate production capacity prediction helps companies better plan and allocate human and material resources and improve production efficiency. It can continuously update and adjust its own parameters as new data is introduced, and can adapt to changing market conditions and production environments, with a wide range of applications. Machine learning reduces the need for human intervention, thereby reducing the possibility of human error.

[0032] 2. Predicting production capacity by predicting the liquid production index. Compared with the direct prediction of production capacity, direct prediction of production capacity does require a richer data set to cover all factors that may affect production, while predicting the liquid production index requires less data and the prediction result is less affected by interference factors, thus ensuring the stability and accuracy of the prediction system. In contrast, the liquid production index is relatively stable in definition, making its data easier to obtain and of higher quality. Each step can be verified, making it easier to adjust and optimize.

[0033] The prediction model based on the liquid production index has fewer feature requirements, can be trained faster, and can better capture the changing trend of the index. The model can achieve a satisfactory performance level in a shorter time. Compared with the existing technology that directly predicts production capacity, it does not need to consider data in multiple dimensions, has low requirements on the model's learning ability and computing resource consumption, and reduces the difficulty of training.

[0034] As a universal parameter, the liquid production index can adapt to different types of oil wells and has a certain degree of applicability even when the process changes. This provides a certain degree of generalization ability for the model. However, the model that directly predicts production capacity is sensitive to specific conditions: the method of directly predicting production capacity tends to rely more on specific training data sets. If the key factors in the production process change (for example, the adoption of new drilling technology), the original model may lose its effectiveness and need to be redesigned or adjusted.

[0035] 3. The liquid production index prediction model established by random forest (RF) is composed of multiple regression trees. Each regression tree is trained on a random subset of the data, and the random subset of features is considered when the node is split, which helps to reduce the overfitting problem that may be caused by a single regression tree and make the model more robust; random forest can effectively select the most important features for prediction, and only consider a part of the features for each split, which has high computational efficiency; random forest has a good tolerance for missing data. In the process of building each tree, if a sample has missing values, other non-missing features can be used for classification or regression.

[0036] 4. Use a variety of evaluation methods to comprehensively evaluate the prediction model, capture different types of errors, and enhance the robustness of the model. Selecting the appropriate evaluation combination according to specific needs can make the model better serve specific tasks and improve the quality and reliability of the model.

[0037] 5. By predicting the liquid production index, we can accurately predict production capacity and output, which can guide the development of oil and gas fields and has important application value and promotion prospects in oil and gas field exploitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of the present invention;

[0039] Figure 2 This is a structural diagram of the liquid production index prediction model of the present invention;

[0040] Figure 3 A comparison chart of the actual value and the predicted value of the liquid production index of the present invention;

[0041] Figure 4 This is an analysis diagram of the evaluation results of the present invention;

[0042] Figure 5 The random forest model of the present invention predicts the liquid production index of wells 13 and 14 and compares the error with the true value;

[0043] Figure 6 It is the result of the Spearman correlation coefficient of the present invention. DETAILED DESCRIPTION

[0044] The following is further described in detail through specific implementation methods:

[0045] The specific implementation process is as follows:

[0046] See also Figures 1 to 6 , a yield prediction method based on machine learning, specifically comprising the following steps,

[0047] S1: Collect and preprocess raw data;

[0048] The original data include engineering parameters and geological parameters. The engineering parameters include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, and liquid production index. The geological parameters include formation crude oil viscosity, mud content (VSH), porosity, water saturation, logging permeability, and original formation pressure.

[0049] In step S1, data preprocessing includes abnormal data processing, missing value processing, and data standardization processing. The data is processed for abnormal data and missing values ​​through data cleaning, and the data is standardized through normalization.

[0050] Through abnormal data processing, excessively large or small values ​​in the original data are removed to prevent such data from affecting the analysis results; missing values ​​are completed through interpolation or data filling to complete the processing of missing values ​​and improve the integrity and feasibility of the data; the original data is processed by using Z-score standardization or Min-Max normalization data standardization processing methods to effectively improve the efficiency and accuracy of model training.

[0051] As shown in Table 1, the original data of this embodiment is preprocessed:

[0052] Table 1 Engineering parameters and geological parameters after preprocessing

[0053]

[0054] S2: Establish a random forest liquid production index prediction model, and input the preprocessed data of wells 1 to 12 into the liquid production index prediction model for training;

[0055] In step S2, the steps of establishing the random forest liquid production index prediction model are as follows:

[0056] S2-1) defining input parameters and output parameters of the liquid production index prediction model, wherein the input parameters include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, liquid production index, formation crude oil viscosity, mud content, porosity, water saturation, logging permeability, and original formation pressure, and the output parameter is the liquid production index;

[0057] S2-2) using the random forest algorithm to calculate the Gini coefficient of each parameter in the data set, ranking the parameters according to the Gini coefficient, determining the Spearman correlation coefficients of the top 12 parameters in the importance ranking, selecting high-correlation parameters whose Spearman correlation coefficients are greater than a set threshold, and excluding parameters whose Spearman correlation coefficients are lower than the set threshold to obtain characteristic parameters;

[0058] like Figure 6 As shown in the figure, it is a heat map of the Spearman correlation coefficient of each parameter. It can be seen from the figure that among the parameters such as drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, liquid production index, formation crude oil viscosity, mud content, porosity, water saturation, logging permeability, and original formation pressure, the Spearman correlation coefficient of the original formation pressure is the lowest, so it is believed that the original formation pressure has little effect on the liquid production index. Therefore, the original formation pressure as a geological parameter is excluded, and finally the drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, liquid production index, formation crude oil viscosity, mud content, porosity, water saturation, and logging permeability are selected as characteristic parameters.

[0059] S2-3) constructing a random forest regression tree model according to the characteristic parameters, wherein the regression tree model is expressed as:

[0060]

[0061] In the formula, m is the number of units in the feature space, M is the total number of feature parameters, and c m is the optimal output weight, I is the indicator function (if x belongs to the mth region R m , then the indicator function value is 1; if x does not belong to the mth region R m , then the value of the indicator function is 0);

[0062] In this embodiment, M is 12.

[0063] S2-4) Perform hyperparameter optimization to determine the number of regression trees:

[0064] There are many important hyperparameters in the establishment of the prediction model of random forest, such as the number of regression trees (n_estimators), splitting criteria (criterion) and maximum depth (max_depth). n_estimators defines the number of regression trees in the random forest prediction model. Generally, the higher this value, the more accurate the model. In this embodiment, n_estimators is 300, that is, 300 independent regression trees are constructed, and the average value of these regression trees will be used as the final prediction value; the splitting criteria and the most suitable nodes are determined by the MSE method; the maximum depth (max_depth) of the regression tree is determined according to the amount of data collected. The maximum depth of the regression tree in this embodiment is 3.

[0065] In order to let this prediction model select the bootstrap aggregation in question, set bootstrap to True and perform sampling with replacement; set n_jobs to -1 (consistent with the number of CPU cores).

[0066] S2-5) According to the calculated number of regression trees and regression tree models, a random forest liquid production index prediction model is built.

[0067] S3: Analyze the performance of the liquid production index prediction model using a variety of evaluation methods, and adjust the model parameters according to the analysis results until the predetermined accuracy standards are met;

[0068] In step S3, R 2 The liquid production index prediction model is analyzed by two or more evaluation methods including MAE, MSE and RMSE.

[0069] The predetermined accuracy standard includes a first preset accuracy and a second preset accuracy. The first preset accuracy is used to compare with the analysis results of the MAE, MSE, and RMSE evaluation methods, and the second preset accuracy is used to compare with the R 2 The analysis results of the two evaluation methods are compared. If both comparison results meet the accuracy standard requirements, the liquid production index prediction model meets the accuracy standard requirements.

[0070] Specifically, the analysis results of the MAE, MSE, and RMSE evaluation methods are less than the first preset accuracy, and R 2 If the analysis result of the evaluation method is greater than the second preset accuracy, the liquid production index prediction model meets the accuracy standard requirements.

[0071] R 2 The evaluation method is expressed as

[0072]

[0073] In the formula, y i is the true value of the liquid production index, is the liquid production index value predicted by the model, It is the average value of the true values ​​of all liquid production indices.

[0074] The MAE evaluation method is expressed as

[0075]

[0076] In the formula, y i is the true value of the liquid production index, is the liquid production index value predicted by the model, and N is the number of data in the training set.

[0077] In this embodiment, wells 1 to 12 participate in the training, and N is 12.

[0078] The MAE value of the data is equivalent to the L1 norm. Generally, the smaller the value, the better the model fitting effect.

[0079] The MSE evaluation method is expressed as

[0080]

[0081] In the formula, y i is the true value of the liquid production index, is the liquid production index value predicted by the model, and N is the number of data in the training set.

[0082] The RMSE evaluation method is expressed as

[0083]

[0084] In the formula, y i is the true value of the liquid production index, is the liquid production index value predicted by the model, and N is the number of data in the training set.

[0085] In this embodiment, four evaluation methods are used to analyze the model, the first preset accuracy is 0.1, and the second preset accuracy is 0.85.

[0086] See also Figure 4 , are the four evaluation results of the liquid production index prediction model based on random forest. It can be seen that R 2 The analysis results of MAE, MSE, and RMSE are all less than 0.1, and R 2The analysis results are greater than 0.85, which meets the preset accuracy standard requirements. The prediction meets the preset accuracy and meets the design requirements.

[0087] S4: Use the trained liquid production index prediction model to predict the liquid production index, and judge the oil field production capacity through the liquid production index.

[0088] As shown in Table 2, the liquid production index is predicted based on the data of wells 13 and 14. The actual liquid production index of well 13 is 1.58, the predicted liquid production index is 1.49, and the prediction error is 5.4%; the actual liquid production index of well 14 is 1.19, the predicted liquid production index is 1.36, and the prediction error is -14.45%. Figure 5 , which is a comparison chart of the predicted results of the liquid production index prediction for wells 13 and 14 and the actual liquid production index values.

[0089] Table 2 Verification data, prediction results and prediction errors

[0090]

[0091] According to the prediction results of the liquid production index, the prediction errors of the two wells are within 20%. The liquid production index predicted by the random forest model is credible, and the production capacity of the oil well can be judged by the daily oil production under unit pressure difference, that is, the liquid production index.

[0092] The liquid production index and the production capacity have the following relationship:

[0093] Q=J×ΔP

[0094] Where J is the liquid production index, Q is the liquid production, and ΔP is the production pressure difference.

[0095] In this embodiment, the liquid production index prediction model is used to obtain the liquid production indexes of wells 13 and 14, which are 1.49 and 1.36 respectively. The predicted daily oil production under unit pressure difference is 1.49 m3 / d and 1.36 m3 / d respectively.

[0096] The liquid production index is an important indicator of the oil well productivity. This indicator comprehensively reflects the relationship between engineering parameters and geological parameters and liquid production. Therefore, the oil well productivity can be judged by the change of the liquid production index. The daily oil production under unit pressure difference obtained by dividing the liquid production by the production pressure difference is the liquid production index. The higher the liquid production index, the more daily oil production under unit pressure difference, indicating that the well productivity is higher. Studying the changing law of the liquid production index can more accurately reflect the changing characteristics of the reservoir liquid production.

[0097] By predicting the liquid production index and production capacity, we can accurately determine the output of the oil field, providing basic data support for the entire oil field management and operation, production optimization and economic analysis.

[0098] The above are only embodiments of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the relevant field know all the common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A yield prediction method based on machine learning, characterized in that: The specific steps include: S1: Collect and preprocess raw data; S2: Establish a random forest liquid production index prediction model, and input the preprocessed data into the liquid production index prediction model for training; S3: Analyze the performance of the liquid production index prediction model using a variety of evaluation methods, and adjust the model parameters according to the analysis results until the predetermined accuracy standard requirements are met; S4: Use the trained liquid production index prediction model to predict the liquid production index, and judge the oil field production capacity through the liquid production index.

2. The yield prediction method based on machine learning according to claim 1, characterized in that: The original data include engineering parameters and geological parameters. The engineering parameters include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, and liquid production index. The geological parameters include formation crude oil viscosity, mud content, porosity, water saturation, logging permeability, and original formation pressure.

3. The yield prediction method based on machine learning according to claim 1, characterized in that: In step S1, data preprocessing includes abnormal data processing, missing value processing, and data standardization processing. The data is processed for abnormal data and missing values ​​through data cleaning, and the data is standardized through normalization.

4. The yield prediction method based on machine learning according to claim 1, characterized in that: In step S2, the steps of establishing the random forest liquid production index prediction model are as follows: S2-1) defining input parameters and output parameters of the liquid production index prediction model, wherein the input parameters include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, liquid production index, formation crude oil viscosity, mud content, porosity, water saturation, logging permeability, and original formation pressure, and the output parameter is the liquid production index; S2-2) Calculate the Gini coefficient of each parameter and select characteristic parameters according to the Gini coefficient; S2-3) constructing a random forest regression tree model based on the selected characteristic parameters, wherein the regression tree model is expressed as: In the formula, m is the number of units in the feature space, M is the total number of feature parameters, and c m is the optimal output weight, I is the indicator function (if x belongs to the mth region R m , then the value of the indicator function is 1; If x does not belong to the mth region R m , then the value of the indicator function is 0); S2-4) perform hyperparameter optimization to determine the number of regression trees; S2-5) According to the number of regression trees and the regression tree model, a random forest liquid production index prediction model is built.

5. The yield prediction method based on machine learning according to claim 4, characterized in that: In step S2-2), the screened features include drilling fluid density, drilling fluid viscosity, filtration loss, pore density, perforation thickness, production pressure difference, production fluid index, formation crude oil viscosity, mud content, porosity, water saturation, and logging permeability.

6. The yield prediction method based on machine learning according to claim 4, characterized in that: In step S2-4), the number of regression trees is 300.

7. The yield prediction method based on machine learning according to claim 1, characterized in that: In step S3, R 2 The liquid production index prediction model is analyzed by two or more evaluation methods including MAE, MSE and RMSE.

8. The yield prediction method based on machine learning according to claim 7, characterized in that: The predetermined accuracy standard includes a first preset accuracy and a second preset accuracy. The first preset accuracy is used to compare with the analysis results of the MAE, MSE, and RMSE evaluation methods, and the second preset accuracy is used to compare with the R 2 The analysis results of the two evaluation methods are compared. If both comparison results meet the accuracy standard requirements, the liquid production index prediction model meets the accuracy standard requirements.

9. The yield prediction method based on machine learning according to claim 7, characterized in that: R 2 The evaluation method is expressed as In the formula, y i is the true value of the liquid production index, is the liquid production index value predicted by the model, It is the average value of the true values ​​of all liquid production indices.

10. The yield prediction method based on machine learning according to claim 1, characterized in that: In step S4, the production capacity of the oil well is determined based on the liquid production index combined with the production pressure difference. The liquid production index and the production capacity have the following relationship: Q=J×ΔP Where J is the liquid production index, Q is the liquid production, and ΔP is the production pressure difference.

Images