A method for optimizing the production of a shale gas well
By employing ensemble learning and parameter search techniques, an ensemble model is constructed to optimize the parameter combination of shale gas wells. This solves the problems of inaccurate shale gas well production capacity prediction and low optimization efficiency of key control factors, thereby maximizing production capacity and reducing costs, and providing a scientific development solution.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA NAT PETROLEUM CORP
- Filing Date
- 2024-12-18
- Publication Date
- 2026-06-19
AI Technical Summary
Existing technologies for predicting shale gas well production capacity neglect the complex nonlinear relationship between geological and engineering parameters, resulting in inaccurate predictions. Furthermore, the optimization of key control factors lacks a systematic approach, making it difficult to maximize production capacity.
By employing ensemble learning and parameter search techniques, the advantages of multiple machine learning models are combined to construct an ensemble model, optimize the combination of gas well parameters, and improve the accuracy and generalization ability of production capacity prediction. Combined with parameter search technology, the optimal combination of controlling factors is automatically found.
It improves the accuracy and generalization ability of production capacity forecasting, maximizes gas well production capacity, provides scientific suggestions for the combination of key control factors, reduces production costs, and provides strong support for shale gas well development.
Smart Images
Figure CN122242904A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of shale gas well development technology, and particularly to production capacity prediction and optimization of key control factors. It aims to achieve dynamic control and maximization of shale gas well production capacity through integrated learning and parameter search technologies. Specifically, it relates to a method for optimizing shale gas well production capacity. Background Technology
[0002] Shale gas well productivity prediction and optimization of key control factors are two core issues in shale gas field development. Optimization of key control factors refers to controlling the main variables or important influencing factors affecting gas well productivity, i.e., adjusting key engineering parameters related to the gas well to achieve better productivity efficiency. With the deepening of shale gas development, how to accurately predict gas well productivity and how to optimize key control factors to improve productivity have become key research areas in the industry.
[0003] In predicting shale gas well productivity, traditional methods mainly rely on empirical formulas and single prediction models. These methods often overlook the complex nonlinear relationships between geological and engineering parameters and productivity, leading to inaccurate prediction results. Furthermore, due to the complexity and uncertainty of shale gas reservoirs, a single prediction model is difficult to adapt to the characteristics of different gas wells and regions.
[0004] In optimizing key control factors, traditional methods are mostly trial-and-error or based on experience-based rules, lacking systematic parameter search and optimization strategies. This approach is not only inefficient but also struggles to maximize production capacity. As shale gas field development deepens, the complexity and diversity of key control factors continue to increase, making traditional optimization methods inadequate for practical needs.
[0005] Machine learning models can learn patterns from vast amounts of historical data to uncover the complex relationships between geological and engineering parameters and production capacity, thereby achieving more accurate production capacity predictions. Simultaneously, parameter search techniques, combined with machine learning models, can automatically search for optimal parameter combinations to maximize production capacity. Therefore, this invention proposes a shale gas well production capacity optimization method based on ensemble learning and parameter search. It aims to improve the accuracy and generalization ability of production capacity prediction by integrating the advantages of multiple machine learning models; and simultaneously, by combining parameter search techniques, optimize the combination of key controlling factors to achieve the goal of maximizing production capacity. This proposed method will provide new technical support and solutions for shale gas well development. Summary of the Invention
[0006] The purpose of this invention is to address at least one of the aforementioned shortcomings of the prior art. For example, the purpose of this invention is to provide a method for optimizing the productivity of shale gas wells.
[0007] To achieve the above objectives, the present invention provides a method for optimizing the production capacity of shale gas wells, the method comprising:
[0008] Acquire existing well data, perform preprocessing, and obtain basic data;
[0009] Multiple base learners are established, and each base learner fits the nonlinear relationship between gas well parameters and production capacity based on basic data;
[0010] Optimize the hyperparameters of each base learner;
[0011] By combining the multiple base learners using ensemble learning, an ensemble model containing a meta-model is constructed. The meta-model can obtain the final prediction result based on the prediction results generated by each base learner.
[0012] Based on the meta-model, gas well productivity optimization is performed through parameter combination search.
[0013] Optionally, the gas well data includes gas well lifecycle production data, gas well geological parameters, and gas well engineering data.
[0014] Alternatively, the plurality of base learners may include random forest regression, support vector regression, and lightweight gradient boosting machine.
[0015] Alternatively, in the process of optimizing the hyperparameters of each base learner: the hyperparameter optimization problem of the base learner is defined as: argmin x∈X f(x), where x is a set of values for the hyperparameters, and X is the search space for the parameters; f(x) is the objective function in hyperparameter optimization, and the goal of f(x) is to find the global optimum as much as possible, such that x * =argmin x∈X f(x); where x * This represents the optimal parameter combination.
[0016] The detailed algorithm is as follows:
[0017]
[0018]
[0019] Alternatively, the objective function may be a loss function.
[0020] Alternatively, the optimization may include: learning a proxy model from training data; and determining the next data collection point using a collection function.
[0021] Optionally, the optimization of gas well production capacity under parameter combination search includes: inputting different parameter combinations of the gas well to be developed into an integrated model to obtain the life cycle production capacity estimate of the gas well to be developed; and searching for the optimal combination of main control factors with the ultimate production capacity of the gas well to be developed as the optimization objective.
[0022] In other words, the method includes: using the maximum value of the main controlling factor of the existing gas well as the upper bound of the search, using the minimum value of the main controlling factor as the lower bound of the search, and using the cumulative production capacity of a single gas well as the objective, to establish a model and solve for the optimal extraction scheme.
[0023] Alternatively, the engineering parameter combination in the gas well construction plan can be set as X = (x1, x2, ..., x...). p The cumulative production capacity function is P(X), and the established model is as follows:
[0024] maxP(X)
[0025] x min ≤x i ≤x max i = 1, 2, ..., p
[0026] choose As the initial search point, its X min =(x 1,min ,x 2,min ,…,x p,min ), X max =(x 1,max ,x 2,max ,…,x p,max ).
[0027] The specific algorithm is as follows:
[0028] Algorithm 2:
[0029] Input: Search space X min =(x 1,min ,x 2,min ,…,x p,min ), X max =(x 1,max ,x 2,max ,…,x p,max ).
[0030] Cumulative capacity function P(X0)
[0031] Output: Optimal combination of engineering parameters X optimal
[0032] Initialization: Initial search parameter combination
[0033] 1. Calculate the gas well production capacity P(X0) corresponding to the current parameters.
[0034] 2. Based on the parameter search space X min ,X max Solve for the objective function Max P(X).
[0035] 3. Output the optimal combination of parameters that satisfies the constraints.
[0036] 4 Return X optimal
[0037] Compared with the prior art, the beneficial effects of the present invention include at least one of the following:
[0038] (1) This invention effectively integrates the advantages of multiple machine learning algorithms through ensemble learning technology, thereby improving the accuracy and generalization ability of capacity prediction.
[0039] (2) Combining the parameter search strategy, this invention can automatically find the optimal combination of main control factors, providing strong support for actual production operations.
[0040] (3) This invention can be used to predict the future production dynamics of gas wells, provide theoretical support for the development of new gas wells, and provide scientific suggestions on the combination of main control factors for the production optimization of existing gas wells.
[0041] (4) The method provided by the present invention can effectively reduce production costs, increase gas well productivity, and provide strong support for the efficient development of shale gas wells. Attached Figure Description
[0042] The above and other objects and / or features of the present invention will become clearer from the following description taken in conjunction with the accompanying drawings, in which:
[0043] Figure 1 A schematic diagram of the integrated framework of the capacity optimization method of the present invention is shown.
[0044] Figure 2 A schematic diagram illustrating the impact of parameter optimization of the production capacity optimization method of the present invention on the first-year production capacity of a simulated gas well is shown. Detailed Implementation
[0045] In the following sections, a method for optimizing the productivity of shale gas wells according to the present invention will be described in detail with reference to exemplary embodiments.
[0046] Exemplary embodiments
[0047] This exemplary embodiment provides a method for optimizing the productivity of shale gas wells.
[0048] The method may include the following steps:
[0049] S1. Obtain existing well data, perform preprocessing, and obtain basic data.
[0050] In this embodiment, the gas well data includes gas well lifecycle production capacity data, gas well geological parameters, and gas well engineering data.
[0051] S2. Establish multiple base learners, each of which fits the nonlinear relationship between gas well parameters and production capacity based on basic data.
[0052] In this embodiment, the plurality of base learners include Random Forest Regression (RFR), Support Vector Regression (SVR), and Lightweight Gradient Boosting Machine (LGBM).
[0053] S3. Optimize the hyperparameters of each base learner.
[0054] In this embodiment, during the optimization of the hyperparameters of each base learner: the hyperparameter optimization problem of the base learner is defined as: argmin x∈X f(x), where x is a set of values for the hyperparameters, and X is the search space for the parameters; f(x) is the objective function in hyperparameter optimization, and the goal of f(x) is to find the global optimum as much as possible, such that x * =argmin x∈ X f(x); where x * This represents the optimal parameter combination.
[0055] The detailed algorithm is as follows:
[0056] Algorithm 1: Base Learner Hyperparameter Optimization
[0057] Input: Initialize candidate parameters n0, maximum number of iterations N, surrogate model g(x), acquisition function a(x|D)
[0058] Output: Optimal parameter combination x *
[0059] Begin
[0060] Step 1: Randomly initialize n0, let
[0061] Step 2: Substitute the parameter combination into the objective function to obtain f(X) init ), initial point set D0 = {X init ,f(X init )}
[0062] Let t = n0, D t-1 =D0
[0063] While t <N do:
[0064] Step 3: Based on the currently obtained point set D t-1 Construct the proxy model g(x)
[0065] Step 4: Based on the surrogate model g(x), maximize the acquisition function a(x|D) t-1 ), to obtain the next evaluation point x * =argmin x∈X a(x|D t-1 )
[0066] Step 5: Obtain the evaluation point x t The function value f(x) t Add it to the current set of evaluation points D. t =D t-1 U{x t ,f(x t Return to step 3.
[0067] End
[0068] Output: Optimal candidate parameter combination x *
[0069] End
[0070] In this embodiment, the objective function is a loss function.
[0071] In this embodiment, the optimization includes: learning a proxy model through training data; and determining the next data collection point through a collection function.
[0072] S4. Using ensemble learning to combine the multiple base learners, an ensemble model containing a meta-model is constructed. The meta-model can obtain the final prediction result based on the prediction results generated by each base learner.
[0073] Among them, the integration framework is as follows Figure 1 As shown, the optimized base learner is used for testing to evaluate its predictive performance. The prediction results of the base learner are then used as input to train the meta-model. The trained meta-model is then used to predict the data. First, predictions are obtained through the base learner, and then these results are input into the meta-model, which generates the final prediction. For the i-th sample in dataset X, the predicted value on the meta-model is... It can be represented as Where x ij It is the prediction result of the i-th sample on the j-th model.
[0074] S5. Based on the meta-model, optimize gas well productivity through parameter combination search.
[0075] In this embodiment, the gas well production capacity optimization under parameter combination search includes: inputting different parameter combinations of the gas well to be developed into an integrated model to obtain the life cycle production capacity estimate of the gas well to be developed; and searching for the optimal combination of main control factors with the ultimate production capacity of the gas well to be developed as the optimization objective. The optimal combination of main control factors is the optimal exploitation scheme corresponding to the gas well.
[0076] In other words, the method includes: using the maximum value of the main controlling factor of the existing gas well as the upper bound of the search, using the minimum value of the main controlling factor as the lower bound of the search, and using the cumulative production capacity of a single gas well as the objective, to establish a model and solve for the optimal extraction scheme.
[0077] Let the combination of engineering parameters in the gas well construction plan be X = (x1, x2, ..., x p The cumulative production capacity (i.e., cumulative gas production) function is P(X), and the established model is as follows:
[0078] maxP(X)
[0079] x min ≤x i ≤x max i = 1, 2, ..., p
[0080] choose As the initial search point, its X min =(x 1,min ,x 2,min ,…,x p,min ), X max =(x 1,max ,x 2,max ,…,x p,max ).
[0081] The specific algorithm is as follows:
[0082] Algorithm 2:
[0083] Input: Search space X min =(x 1,min ,x 2,min ,…,x p,min ), X max =(x 1,max ,x 2,max ,…,x p,max ).
[0084] Cumulative capacity function P(X0)
[0085] Output: Optimal combination of engineering parameters X optimal
[0086] Initialization: Initial search parameter combination
[0087] 1. Calculate the gas well production capacity P(X0) corresponding to the current parameters.
[0088] 2. Based on the parameter search space X min ,X max Solve for the objective function Max P(X).
[0089] 3. Output the optimal combination of parameters that satisfies the constraints.
[0090] 4 Return X optimal
[0091] To better understand the exemplary embodiments described above, further explanation is provided below with specific examples.
[0092] Example
[0093] Take all the gas wells in the entire development area as an example.
[0094] 1. Integrated model for capacity forecasting
[0095] (1) Data preparation, obtaining existing gas well data;
[0096] (2) Establish a base learner to fit the relationship between gas well parameters and production capacity respectively;
[0097] (3) Optimize the hyperparameters of each base learner using Algorithm 1;
[0098] (4) Use ensemble learning to combine multiple base learners to construct a meta-model.
[0099] Based on the integrated model constructed using the above method, all gas wells were modeled and their performance evaluated. The evaluation results are shown in Table 1.
[0100] Table 1 Performance evaluation of integrated model for production capacity prediction (data from 242 gas wells)
[0101]
[0102] In Table 1, RMSE, MAE, and R2 represent the root mean square error, mean absolute error, and coefficient of determination, respectively.
[0103] 2. Gas well production capacity optimization
[0104] First, geological and engineering parameters are selected from existing gas wells to construct a simulated gas well. Then, the parameters of the simulated gas well are optimized, and the effect of parameter optimization on improving gas well productivity is evaluated. The search space of the main control factors of the gas well, the initial values of the main control factors of the simulated gas well, and the optimized values of various parameters are shown in Tables 2, 3, and 4, respectively.
[0105] Next, using an integrated production capacity prediction model, the daily gas production of the simulated gas wells was predicted over six years, yielding the cumulative production capacity over those six years. Then, the main controlling factors of the simulated gas wells were optimized, and the production capacity of the optimized simulated gas wells was predicted again, comparing the changes in production capacity before and after optimization. The changes in simulated gas well production capacity and the first-year daily production curve are shown in Tables 5 and 5, respectively. Figure 2 .
[0106] Finally, 10 gas wells were selected as case studies from 34 gas wells with actual production cycles of more than 6 years. The main control factors of the 10 gas wells were optimized, and the cumulative production capacity changes of the gas wells before and after optimization were compared. See Table 6 for details.
[0107] Table 2 Search Space of Key Controlling Factors for Gas Wells
[0108]
[0109]
[0110] Table 3 Initial Parameter Values for Simulated Gas Wells
[0111]
[0112] Table 4. Parameter values after optimization of simulated gas wells
[0113]
[0114] Table 5 Simulated Gas Well Production Changes
[0115]
[0116] Table 6 shows the effect of parameter optimization on improving the production capacity of developed gas wells.
[0117]
[0118]
[0119] Although the present invention has been described above in conjunction with exemplary embodiments and accompanying drawings, those skilled in the art should understand that various modifications can be made to the above embodiments without departing from the spirit and scope of the claims.
Claims
1. A method for optimizing the productivity of shale gas wells, characterized in that, The method includes: Acquire existing well data, perform preprocessing, and obtain basic data; Multiple base learners are established, and each base learner fits the nonlinear relationship between gas well parameters and production capacity based on basic data; Optimize the hyperparameters of each base learner; By combining the multiple base learners using ensemble learning, an ensemble model containing a meta-model is constructed. The meta-model can obtain the final prediction result based on the prediction results generated by each base learner. Based on the meta-model, gas well productivity optimization is performed through parameter combination search.
2. The shale gas well productivity optimization method according to claim 1, characterized in that, The gas well data includes gas well lifecycle production data, gas well geological parameters, and gas well engineering data.
3. The shale gas well productivity optimization method according to claim 1, characterized in that, The base learners include random forest regression, support vector regression, and lightweight gradient boosting machine.
4. The shale gas well productivity optimization method according to claim 1, characterized in that, In the process of optimizing the hyperparameters of each base learner: the hyperparameter optimization problem of the base learner is defined as: argmin x∈X f(x), where x is a set of values for the hyperparameters, and X is the search space for the parameters; f(x) is the objective function in hyperparameter optimization, and the goal of f(x) is to find the global optimum as much as possible, such that x * =argmin x∈X f(x); where x * This represents the optimal parameter combination.
5. The shale gas well productivity optimization method according to claim 4, characterized in that, The objective function is the loss function.
6. The shale gas well productivity optimization method according to claim 4, characterized in that, The optimization includes: learning a proxy model using training data; and determining the next data collection point using a collection function.
7. The shale gas well productivity optimization method according to claim 1, characterized in that, The optimization of gas well production capacity under parameter combination search includes: inputting different parameter combinations of the gas well to be developed into an integrated model to obtain the life cycle production capacity estimate of the gas well to be developed; and searching for the optimal combination of main control factors with the ultimate production capacity of the gas well to be developed as the optimization objective.
8. The shale gas well productivity optimization method according to claim 1, characterized in that, The method further includes: using the maximum value of the main controlling factor of the existing gas well as the upper limit of the search, using the minimum value of the main controlling factor as the lower limit of the search, and using the cumulative production capacity of a single gas well as the objective, to establish a model and solve for the optimal extraction scheme.
9. The shale gas well productivity optimization method according to claim 8, characterized in that, Let the combination of engineering parameters in the gas well construction plan be X = (x1, x2, ..., x p The cumulative production capacity function is P(X), and the established model is as follows: maxP(X) x min ≤x i ≤x max ,i=1,2,…,p choose As the initial search point, its X min =(x 1,min ,x 2,min ,…,x p,min ), X max =(x 1,max ,x 2,max ,…,x p,max ).