Oil reservoir profile control and yield increase effect prediction method based on yield relative difference value
By using a method based on the relative difference in production, combined with well, geological, engineering and production parameters, and employing neural networks and ensemble learning algorithms to construct a prediction model, the problems of low prediction accuracy and reliance on expert experience in existing technologies are solved, and high-precision prediction of profile control and production enhancement effects is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-18
- Publication Date
- 2026-03-13
AI Technical Summary
Existing methods for predicting the effects of reservoir profile control on production enhancement suffer from low prediction accuracy, difficulty in distinguishing between the effects of natural decline and profile control interventions, reliance on expert experience, and sensitivity to data noise.
A prediction model based on the relative difference in production is adopted, which combines well, geological, engineering and production parameters. The model is constructed by neural network and ensemble learning algorithm, taking into account the influence of multiple factors. The hyperparameters are optimized using the Optuna framework to improve the generalization and robustness of the model.
It significantly improves the accuracy of predicting the effects of profile regulation on increased production, effectively distinguishes between the effects of natural decline and profile regulation measures, provides scientific basis to guide on-site decision-making, and reduces reliance on expert experience and sensitivity to data noise.
Smart Images
Figure CN121660152A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oilfield development technology, and in particular to a method for predicting the production enhancement effect of reservoir profile modification based on the relative difference in production. Background Technology
[0002] As oil reservoirs enter the high water-cut stage, reservoir heterogeneity increases, extraction difficulty increases significantly, and crude oil recovery rate gradually decreases. In order to improve reservoir heterogeneity, reduce well water production, and increase crude oil recovery rate, chemical profile control measures are widely used in oil reservoir development.
[0003] Predicting the effects of profile control helps optimize process parameters, determine the best profile control scheme, and save economic costs. It serves as an important basis for guiding the decision-making and optimization design of profile control measures in oilfields. Traditional methods for predicting the effects of profile control mainly rely on expert experience, statistical methods, and numerical simulation methods. These methods require researchers to have a high level of knowledge in reservoir engineering, are sensitive to data noise, have long simulation cycles, and limited prediction accuracy, resulting in varying degrees of deviation between predicted and actual oil production after the measures are implemented.
[0004] In recent years, the application of machine learning technology in the oil and gas development field has gradually increased. However, existing machine learning-based methods for predicting profile control effects still have many limitations. First, machine learning-based prediction methods usually refer to unconventional oil and gas production prediction methods, directly transferring or simply applying prediction algorithms to profile control effect prediction. They rely solely on historical profile control production data and absolute production for time-series modeling, making it difficult to effectively distinguish the impact of natural decline and profile control interventions on oil production. Second, machine learning-based prediction methods often ignore the combined effects of multiple factors such as geological parameters, construction parameters, and production dynamic parameters on profile control effects, resulting in weak generalization ability and limited prediction accuracy of machine learning models. Summary of the Invention
[0005] The purpose of this invention is to provide a method for predicting the production enhancement effect of reservoir profile control based on the relative difference in production output, in order to solve the technical problems of low prediction accuracy and difficulty in distinguishing the effects of natural decline and profile control intervention in existing methods for predicting the production enhancement effect of reservoir profile control.
[0006] This invention provides a method for predicting the reservoir profile control and production enhancement effect based on the relative difference in production, comprising:
[0007] Collect profile control well group data for the target reservoir and perform preprocessing;
[0008] Feature selection of factors influencing profile control;
[0009] A neural network was used to construct a daily water drive baseline yield prediction model, and the model's performance was evaluated.
[0010] Calculate the relative difference between the actual cumulative production after profile adjustment and the predicted cumulative water-drive baseline production;
[0011] An ensemble learning algorithm is used to construct a predictive model for the effect of profile regulation on yield increase, and the model effect is evaluated and the prediction results are output.
[0012] In some embodiments, the profile control well group data includes at least well parameters, geological parameters, engineering parameters, and production parameters.
[0013] In some embodiments, the well parameters include at least the well number, well type, and production level; the geological parameters include at least the average permeability of the target layer, the maximum permeability of the target layer, and the breakthrough coefficient; the engineering parameters include at least the profile control injection concentration, profile control injection rate, and profile control reagent dosage; and the production parameters include at least the average daily injection volume before and after the operation, the injection time, and the wellhead pressure.
[0014] In some embodiments, the preprocessing includes outlier detection and repair, missing value imputation, and normalization.
[0015] In some embodiments, the outlier detection method includes an autoencoder and a Grubbs test; the outlier repair method includes linear interpolation and clustering; and the missing value imputation method includes mean imputation.
[0016] In some embodiments, the normalization target is divided into parameters that are better the larger they are and parameters that are better the smaller they are, wherein the normalization formula for parameters that are better the larger they are is:
[0017]
[0018] The normalization formula for parameters of the "smaller is better" type is:
[0019]
[0020] in, These are sample values; These are the normalized values; The minimum value of the feature data; This represents the maximum value of the feature data.
[0021] In some embodiments, the feature selection method includes Pearson correlation coefficient analysis, Spearman correlation coefficient analysis, random forest feature importance calculation method, principal component analysis method, and SHAP feature importance calculation method.
[0022] In some embodiments, the neural network model includes at least a long short-term memory network; the daily water-driven baseline production prediction model is trained by performing Bayesian optimization to select the optimal hyperparameter combination through the Optuna framework.
[0023] In some embodiments, the ensemble learning model includes random forest, gradient boosting decision tree, and adaptive boosting tree, and the profile adjustment and yield increase prediction model is trained by performing Bayesian optimization to select the optimal hyperparameter combination through the Optuna framework.
[0024] In some embodiments, the formula for calculating the relative difference in output is:
[0025]
[0026] Wherein, PRD represents the relative difference in output; To predict the cumulative oil production from water-driven operations within a given time period; To predict the actual cumulative oil production after profile control within a given time period; and through the formula Calculate the predicted daily output after profile adjustment, where: This represents the daily yield after hull adjustment. The cumulative output from the start time of the diaphragm adjustment to the i-th day (i>1); The cumulative output from the start time of the sectional adjustment to day i-1.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] 1. It fully considers geological parameters, construction parameters and production dynamic parameters, and uses the relative difference in production as an evaluation index to effectively distinguish the impact of natural decline and profile control intervention on oil production, significantly improving prediction accuracy. The prediction results can be directly used to guide the actual application in the oilfield, providing a scientific basis for profile control decision-making and process parameter optimization.
[0029] 2. Based on neural networks and ensemble learning methods, the dependence on model expert experience and sensitivity to data noise are greatly reduced, enhancing the model's generalization and robustness;
[0030] 3. The Optuna framework performs Bayesian optimization, which is more efficient and accurate than traditional grid search and random search methods, thus improving model performance. Attached Figure Description
[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0032] Figure 1 This is a flowchart of the method for predicting the oil reservoir profile control and production enhancement effect of the present invention;
[0033] Figure 2 This is a diagram showing the prediction effect of the Long Short-Term Memory Network of this invention;
[0034] Figure 3 This is a diagram showing the prediction effect of the random forest in this invention;
[0035] Figure 4 This is a comparison chart of the predicted daily output and the actual daily output after the profile adjustment according to the present invention. Detailed Implementation
[0036] The following will be based on embodiments of the present invention. Figures 1-4 The technical solutions in the embodiments of the present invention will be clearly and completely described together. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0037] Example
[0038] This embodiment provides a method for predicting the production enhancement effect of reservoir profile control based on the relative difference in production output. The method includes collecting and preprocessing data of the target reservoir profile control well group; selecting features of profile control influencing factors; constructing a daily water-drive benchmark production prediction model using a neural network and evaluating the model's performance; calculating the relative difference between the actual cumulative production after profile control and the predicted cumulative water-drive benchmark production; constructing a profile control production enhancement effect prediction model using an ensemble learning algorithm, evaluating the model's performance, and outputting the prediction results.
[0039] Specifically,
[0040] First, data on the profile control well group of the target reservoir is collected. This data includes well parameters, geological parameters, engineering parameters, and production parameters. Well parameters include well number, well type, and recovery level; geological parameters include average permeability of the target layer, maximum permeability of the target layer, and breakthrough coefficient; engineering parameters include profile control injection concentration, profile control injection rate, and profile control reagent dosage; and production parameters include average daily injection volume before and after the operation, injection time, and wellhead pressure.
[0041] Next, after data collection, data preprocessing was performed, including outlier detection and repair, missing value imputation, and normalization. Outlier detection used an autoencoder to identify outliers in the data. For detected outliers, linear interpolation was used to repair them, ensuring the continuity and rationality of the data. Statistical analysis showed that the missing data ratio in the dataset was generally below 10%, indicating high data availability. Therefore, mean imputation was used to fill in the missing values. Normalization used the Min-Max normalization method to adjust the dimensions of data with different characteristics to a uniform range.
[0042] It should be noted that the normalization target is divided into two types: parameters that are better the larger they are and parameters that are better the smaller they are. The normalization formula for parameters that are better the larger they are is as follows:
[0043]
[0044] The normalization formula for parameters of the "smaller is better" type is:
[0045]
[0046] in, These are sample values; These are the normalized values; The minimum value of the feature data; This represents the maximum value of the feature data.
[0047] Following data preprocessing, feature selection is performed. Feature selection involves screening out key features that play a dominant role in the profile control and yield enhancement effects from numerous influencing factors. In this embodiment, Spearman correlation coefficient analysis is used to assess the linear correlation between features, and Pearson correlation coefficient analysis is used to assess the linear correlation between features and the target variable. Features with high correlation are initially screened out, and redundant features are eliminated. On this basis, the random forest feature importance calculation method and the SHAP feature importance calculation method are used to assess the importance of features. The results of the two methods are normalized and weighted averaged to obtain a comprehensive feature importance score. Finally, based on expert experience, the main controlling factors of profile control and yield enhancement effects are screened out, including at least the average permeability of the target layer, the thickness of the target layer, porosity, breakthrough coefficient, water content, average apparent water absorption index, cumulative amount of injected reagent, and cumulative water injection volume.
[0048] Then, a neural network model is constructed to predict the daily water-drive baseline production. The daily water-drive baseline production prediction model is implemented using one of a recurrent neural network, a long short-term memory network, or a Transformer neural network. In this embodiment, a long short-term memory network is preferred. It can be understood that this embodiment uses a long short-term memory network as an example. The construction process is as follows:
[0049] Based on the water drive production data collected from the profile control well group, time is set as the input feature vector X and water drive production is set as the output vector Y. The training set and test set are divided in an 8:2 ratio. The sliding window method is used to extract the dynamic change pattern of the time series data and form a supervised learning data format to establish a time series prediction model.
[0050] During the training process, the Optuna framework was used to perform Bayesian optimization, select the optimal combination of hyperparameters, and train the final daily water-driven baseline yield prediction model.
[0051] in, Figure 2The image shows the prediction results of the Long Short-Term Memory Network. It can be seen that the loss value of the Long Short-Term Memory Network on the training set and the test set decreases rapidly and approaches 0 after the iteration begins. The actual oil production is close to the predicted oil production, indicating that the daily water-driven baseline production prediction model has high prediction accuracy.
[0052] It should be noted that the model evaluation metrics include at least the mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), and coefficient of determination (R²). Since the evaluation expressions for the aforementioned metrics are publicly available, they will not be detailed here.
[0053] Next, after predicting the water-drive baseline production, the relative production difference is calculated. This relative production difference is the difference between the actual cumulative oil production after profile control and the predicted cumulative oil production from water-drive, thus directly determining the actual production increase contribution of the profile control measures. Specifically:
[0054] Considering the cumulative impact of water injection on production, the predicted water-drive baseline production is accumulated to obtain the cumulative oil production of water-drive during the prediction period.
[0055] The daily oil production of each well group after the implementation of profile control measures was extracted from the actual production data. Considering the cumulative impact of the amount of profile control agent injected on the production, the actual cumulative oil production after profile control in the predicted time period was obtained by cumulative calculation.
[0056] The actual cumulative oil production after profile control adjustments and the cumulative oil production from waterflooding within the predicted time period are correlated one-to-one in chronological order, and the relative difference (PRD) between the two is calculated. It should be noted that the formula for calculating the relative difference (PRD) is as follows:
[0057]
[0058] Wherein, PRD represents the relative difference in output; To predict the cumulative oil production from water-driven operations within a given time period; To predict the actual cumulative oil production after profile control within a given time period.
[0059] Finally, this embodiment uses random forest to construct a predictive model for the yield increase effect of profile adjustment. The specific construction process is as follows:
[0060] The main control parameter vector of the profile adjustment effect is used as the input vector, denoted as X, and the vector of relative difference in yield is used as the output vector, denoted as Y. The training set and the test set are divided in an 8:2 ratio to establish a prediction model. At the same time, the 10-fold cross-validation method is used to further verify the model's fitting ability and generalization performance on the entire dataset.
[0061] During training, the Optuna framework is used to perform Bayesian optimization, select the optimal combination of hyperparameters, and train the final model for predicting the yield increase effect of profile adjustment.
[0062] in, Figure 3 The image shows the prediction results of Random Forest. It can be seen that the relative difference between the actual output and the relative difference between the predicted output are very close in both the test set and the cross-validation test set. The residuals are all distributed around y=0, which shows high prediction accuracy and robustness. The prediction result is the cumulative output relative difference. The predicted cumulative output after the split can be obtained by substituting it into the PRD formula of the relative difference, and the daily output after the split can be obtained by difference calculation.
[0063] It should be noted that the formula for calculating the difference is as follows:
[0064]
[0065] in: This represents the daily yield after hull adjustment. The cumulative output from the start time of the diaphragm adjustment to the i-th day (i>1); The cumulative output from the start time of the sectional adjustment to day i-1.
[0066] To verify the practical application effect of this method, the inventors selected data from a profile control well group in the target reservoir that was not involved in the training to predict the profile control and production enhancement effect. The prediction results were substituted into the relative difference PRD formula to obtain the predicted cumulative production after profile control, and the daily production after profile control was obtained through differential processing. Figure 4 The figure shows a comparison between the predicted daily production and the actual daily production after profile adjustment. It can be seen that the daily oil production after profile adjustment calculated by the relative difference of production predicted by random forest has a high degree of consistency with the actual daily oil production. This indicates that the profile adjustment production increase prediction method proposed in this invention has a good effect and strong engineering applicability.
[0067] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0068] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A method for predicting the production enhancement effect of reservoir profile control based on relative production differences, characterized in that, include: Collect profile control well group data for the target reservoir and perform preprocessing; Feature selection of factors influencing profile control; A neural network was used to construct a daily water drive baseline yield prediction model, and the model's performance was evaluated. Calculate the relative difference between the actual cumulative production after profile adjustment and the predicted cumulative water-drive baseline production; An ensemble learning algorithm is used to construct a predictive model for the effect of profile regulation on yield increase, and the model effect is evaluated and the prediction results are output.
2. The method according to claim 1, characterized in that, The profile control well group data includes at least well parameters, geological parameters, engineering parameters, and production parameters.
3. The method according to claim 2, characterized in that, The well parameters include at least the well number, well type, and production level; the geological parameters include at least the average permeability of the target layer, the maximum permeability of the target layer, and the breakthrough coefficient; the engineering parameters include at least the profile control injection concentration, profile control injection rate, and profile control reagent dosage; and the production parameters include at least the average daily injection volume before and after the operation, the injection time, and the wellhead pressure.
4. The method according to claim 1, characterized in that, The preprocessing includes outlier detection and repair, missing value imputation, and normalization.
5. The method according to claim 4, characterized in that, The outlier detection method includes an autoencoder and the Grubbs test; the outlier repair method includes linear interpolation and clustering; and the missing value imputation method includes mean imputation.
6. The method according to claim 4, characterized in that, The normalization targets are divided into parameters that are better the larger they are and parameters that are better the smaller they are. The normalization formula for parameters that are better the larger they are is as follows: ， The normalization formula for parameters of the "smaller is better" type is: ， in, These are sample values; These are the normalized values; The minimum value of the feature data; This represents the maximum value of the feature data.
7. The method according to claim 1, characterized in that, The feature selection methods include Pearson correlation coefficient analysis, Spearman correlation coefficient analysis, random forest feature importance calculation method, principal component analysis method, and SHAP feature importance calculation method.
8. The method according to claim 1, characterized in that, The neural network model includes at least a long short-term memory network; the daily water-driven baseline yield prediction model is trained by performing Bayesian optimization to select the optimal hyperparameter combination through the Optuna framework.
9. The method according to claim 1, characterized in that, The ensemble learning model includes random forest, gradient boosting decision tree, and adaptive boosting tree. The profile adjustment and yield increase prediction model is trained by performing Bayesian optimization to select the optimal hyperparameter combination through the Optuna framework.
10. The method according to claim 1, characterized in that, The formula for calculating the relative difference in output is: ， Wherein, PRD represents the relative difference in output; To predict the cumulative oil production from water-driven operations within a given time period; To predict the actual cumulative oil production after profile control within a given time period; and through the formula Calculate the predicted daily output after profile adjustment, where: This represents the daily output after hull adjustment. The cumulative output from the start time of the diaphragm adjustment to the i-th day (i>1); The cumulative output from the start time of the sectional adjustment to day i-1.