A local well pattern injection-production parameter optimization method based on a long short-term memory network
By using a multi-input multi-output structure optimization model based on long short-term memory networks, the problem of neglecting the influence of water injection wells in traditional oilfield modeling is solved, resulting in more accurate water injection schemes and improved production efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HOHAI UNIV
- Filing Date
- 2024-01-15
- Publication Date
- 2026-07-31
AI Technical Summary
Traditional oilfield modeling cannot meet the requirements for precise water injection under complex conditions and ignores the impact of water injection wells on surrounding production wells, resulting in poor model accuracy.
A multi-input multi-output structure based on long short-term memory network is adopted to construct an injection and production parameter optimization model, taking into account the impact of water injection wells on surrounding production wells, and optimizing the water injection scheme through an objective function.
It improves the model's generalization ability, provides more accurate water injection schemes, and enhances the production efficiency and output of production wells.
Smart Images
Figure CN117988795B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for optimizing injection and production parameters of local well networks based on long short-term memory networks, belonging to the field of intelligent oilfield application technology. Background Technology
[0002] Among the various oilfield development methods in my country, water injection is the most effective way to maintain stable oilfield production and improve recovery rates, and water injection remains the primary development method. However, after a long period of water injection, most old oilfields are now in a complex situation with numerous wells, complex stratification, and water breakthroughs in multiple directions. Traditional reservoir modeling cannot meet the precise water injection requirements of oilfields under complex conditions. Therefore, it is imperative to establish an efficient water-drive development system that combines current research hotspots.
[0003] With the development of artificial intelligence and big data, the concept of smart oilfields has gained widespread acceptance. Researchers have successfully applied various artificial intelligence methods to oilfield development, greatly enhancing the level of intelligence in oilfield development. In particular, in the prediction of single-well production, these methods can quickly and accurately characterize water-driven states. However, most current artificial intelligence production prediction methods directly use the dynamic production parameters of multiple injection wells and a single production well as the input and output layers to obtain the injection-production relationship, without considering the multi-directional water breakthrough during the formation diffusion process.
[0004] In conventional machine learning methods, a model is built by constructing a mapping relationship between injection wells and production wells. However, this model predicts the output based on the output of a single production well, ignoring the impact of injection wells on other surrounding production wells, resulting in poor model accuracy. Summary of the Invention
[0005] To overcome the shortcomings of the existing technology, this invention provides a method for optimizing injection and production parameters of a local well network based on long short-term memory networks. The method determines primary connected water injection wells and secondary connected production wells according to the oilfield injection-production connectivity of the research object. A multi-input multi-output (MIMO) structure is used to construct the model. The objective function addresses the problem that previous machine learning methods only studied single production wells, neglecting the impact of water injection wells on secondary connected production wells. This method optimizes the injection and production parameters, enabling the research object to achieve optimal production.
[0006] The technical solution adopted in this invention is as follows.
[0007] On one hand, this invention provides a method for optimizing local well pattern injection and production parameters based on long short-term memory networks, including:
[0008] Identify the research object and determine the injection and production well locations based on the oilfield injection-production connectivity of the research object; the injection and production well locations include primary connected water injection wells and secondary connected production wells;
[0009] Acquire historical production data of injection and production well locations; preprocess the historical production data and establish a database; at the same time, divide the preprocessed historical production data into training set and validation set according to a set ratio.
[0010] A local well network injection and production parameter optimization model based on long short-term memory network is constructed for injection and production well locations;
[0011] The training set was used to train a local well network injection-production parameter optimization model based on a long short-term memory network, resulting in a reservoir simulator.
[0012] The water injection plan was adjusted using historical production data from the database, and the injection and production parameters were optimized using a reservoir simulator and objective function.
[0013] In the process of establishing a local well network injection and production parameter optimization model based on long short-term memory networks, the influence of injection wells on other surrounding production wells is considered, thereby establishing a more reasonable flow process between the primary connected injection wells and the research object, improving the generalization ability of the model, and providing a more accurate water injection scheme for oilfield development of production wells.
[0014] Optionally, depending on the production status of the production well, production wells with high water cut and low production can be selected as the research object.
[0015] Optionally, the injection and production well locations can be determined based on the injection-production connectivity of the oilfield under study, including:
[0016] The primary interconnected water injection wells to be studied are determined based on the oilfield injection-production connectivity of the research object. The specific screening method is as follows:
[0017]
[0018] Where P represents a production well and I represents a water injection well. The research object is specifically the k-th production well with high water cut and low production; ω ik This represents the oilfield injection-production connectivity between the i-th injection well and the k-th production well, ω. ik The value is either 0 or 1, where 0 indicates no oilfield injection-production connectivity and 1 indicates an oilfield injection-production connectivity. Finally, the primary connected water injection wells that have an oilfield injection-production connectivity with the research object are selected.
[0019] Next, based on the primary interconnected water injection well The injection-production connectivity of the oilfield determines the secondary interconnected production wells to be studied. The specific screening method is as follows:
[0020]
[0021] Where P represents a production well, I represents a water injection well, and ωij This indicates that the j-th production well is connected to the first-level injection well. The oilfield injection-production connection relationship, ω ij The value is either 0 or 1, where 0 indicates no oilfield injection-production connectivity and 1 indicates an oilfield injection-production connectivity; finally, injection wells connected to the primary level are selected. Secondary interconnected production wells with oilfield injection-production connectivity This represents the union of two-level connected production wells.
[0022] Optionally, the historical production data of the injection and production well locations include the water injection volume of the first-level connected water injection wells, the fluid production and oil production of the research object, and the fluid production and oil production of the second-level connected production wells. Among them, the water injection volume of the first-level connected water injection wells is a feature of the dataset, and the fluid production and oil production are the labels of the dataset.
[0023] Optionally, the local well network injection-production parameter optimization model based on long short-term memory network adopts a multi-input multi-output structure. The water injection volume of the first-level connected injection wells is used as the input layer, the production volume and oil production of the research object are used as the main output, and the production volume and oil production of the second-level connected production wells are used as the auxiliary output. Considering the multi-directional water breakthrough situation of the injection wells in the reservoir, the water injection information of the first-level connected injection wells is constrained by the second-level connected production wells, thereby establishing a more accurate model.
[0024] Optionally, when training the local well pattern injection-production parameter optimization model based on long short-term memory network, a threshold is set according to the accuracy of the validation set in the local well pattern injection-production parameter optimization model based on long short-term memory network. When the training set reaches the threshold during the model training process, the validation set is used as the training sample to continue training, and the finally trained local well pattern injection-production parameter optimization model based on long short-term memory network is used as a reservoir simulator.
[0025] Optionally, the water injection scheme is an adjustment to the water injection volume of the injection well in the latest period. This adjustment process usually generates a large number of water injection schemes.
[0026] The production rates of the research object and the secondary connected production well under each water injection scheme were determined using a reservoir simulator, and the water cut was calculated using the following formula:
[0027]
[0028] Among them, f w The water cut represents the overall water cut of the research object and the secondary connected production wells. The oil production of the research object is the sum of the oil production of the secondary connected production wells. The sum of the production volume of the research object and the production volume of the secondary connected production well.
[0029] Optionally, the objective function is calculated by combining the production rate of the research object and the sum of the production rates of the secondary interconnected production wells, along with the water cut. The values of these objective functions are then compared, and the water injection scheme corresponding to the optimal objective function is selected as the subsequent water injection scheme. The formula for calculating the objective function is as follows:
[0030]
[0031] Where J is the objective function value; α is the importance factor for the yield; f w The water content corresponding to each water injection scheme; This is the sum of the production volume of the research object and the production volume of the secondary interconnected production well in each water injection scheme; They are respectively The maximum and minimum values of the objective function are considered. When the objective function value is minimized, it is the optimal objective function, and the water injection scheme corresponding to the research object is the optimal water injection scheme. In this process, the production volume of the research object and the production volume of the secondary interconnected production well are standardized and then the reciprocal is taken so that the production volume of the research object and the production volume of the secondary interconnected production well are in the same dimension as the water cut.
[0032] Optional data preprocessing includes outlier handling and data standardization.
[0033] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:
[0034] This invention determines the primary connected water injection wells and secondary connected production wells based on the oilfield injection-production connectivity of the research object. In the process of model building, a multi-input multi-output structure is adopted. The water injection volume of the primary connected water injection well is used as the input layer, the production volume and oil production of the research object are used as the main outputs, and the production volume and oil production of the secondary connected production wells are used as auxiliary outputs. This solves the problem that previous machine learning studies only focused on a single production well and ignored the impact of water injection wells on secondary connected production wells, thus improving the generalization ability of the model.
[0035] Furthermore, this invention constructs an original objective function, which is calculated using the sum of the production volume of the research object and the production volume of the secondary interconnected production well, as well as the water cut. Based on the objective function, the optimal water injection scheme is determined, thereby optimizing the injection and production parameters and enabling the research object to achieve the optimal production.
[0036] Furthermore, conventional single-well prediction models only consider the impact on the research object (the central production well) when optimizing injection and production parameters, thus neglecting the comprehensive effect of the injection well on surrounding production wells. Therefore, this method, by considering the comprehensive impact of the injection well, provides a scientific basis for obtaining accurate and efficient water injection schemes and effectively improving the production efficiency of production wells, which is of great significance. Attached Figure Description
[0037] Figure 1 This is a flowchart illustrating the specific process for optimizing injection and production parameters in a local well network based on long short-term memory networks.
[0038] Figure 2 A schematic diagram showing water seepage from multiple directions during water injection;
[0039] Figure 3 It is a Long Short-Term Memory (LSTM) network structure. Detailed Implementation
[0040] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of this application.
[0041] Example 1
[0042] This embodiment introduces a method for optimizing local well network injection and production parameters based on long short-term memory networks. The specific implementation steps include:
[0043] The research object is identified, and the injection and production well locations are determined based on the oilfield injection-production connectivity of the research object; the injection and production well locations include primary connected water injection wells and secondary connected production wells;
[0044] Acquire historical production data of the injection and production well locations; perform data preprocessing on the historical production data and establish a database; simultaneously, divide the preprocessed historical production data into a training set and a validation set according to a set ratio.
[0045] A local well network injection and production parameter optimization model based on a long short-term memory network is constructed for the aforementioned injection and production well locations;
[0046] The training set is used to train the local well network injection and production parameter optimization model based on long short-term memory network, resulting in a reservoir simulator;
[0047] The water injection plan is adjusted using historical production data in the database, and the injection and production parameters are optimized using the reservoir simulator and objective function.
[0048] In the process of establishing the local well network injection-production parameter optimization model based on long short-term memory network, the influence of injection wells on other surrounding production wells is considered, thereby establishing a more reasonable flow process between the first-order connected injection wells and the research object. The local well network injection-production parameter optimization model based on long short-term memory network is proposed to improve the generalization ability of the model and provide a more accurate water injection scheme for oilfield development of production wells, so that the research object can achieve the optimal production.
[0049] Example 2
[0050] Based on Example 1, this example further introduces and explains the method for optimizing injection and production parameters of local well networks based on long short-term memory networks.
[0051] Local well pattern injection-production parameter optimization method based on long short-term memory network, such as Figure 1 Specifically, the implementation steps are as follows:
[0052] Step 1: Determine the research object and injection / production well locations, such as... Figure 2 As shown, the research object refers to a production well selected based on its production status, which is in a high water cut and low production state; the injection-production well location includes a primary connected water injection well and a secondary connected production well; specifically, the injection-production well location is determined based on the oilfield injection-production connectivity of the research object.
[0053] First, based on the oilfield injection-production connectivity of the research object, the primary interconnected water injection wells to be studied are determined. The specific screening method is as follows:
[0054]
[0055] Where P represents a production well and I represents a water injection well. The research object refers specifically to the k-th high water-cut, low-yield production well; ω ik This represents the oilfield injection-production connectivity between the i-th injection well and the k-th production well, ω. ik The value is either 0 or 1, where 0 indicates no oilfield injection-production connectivity and 1 indicates an oilfield injection-production connectivity. Finally, the primary connected water injection wells that have an oilfield injection-production connectivity with the research object are selected.
[0056] Next, according to the aforementioned primary interconnected water injection well The specific screening method for determining the secondary interconnected production wells to be studied is as follows: (Based on the oilfield injection-production connectivity relationship)
[0057]
[0058] Where P represents a production well, I represents a water injection well, and ω ij This indicates that the j-th production well is connected to the primary injection well. The oilfield injection-production connection relationship, ω ij The value is either 0 or 1, where 0 indicates no oilfield injection-production connectivity and 1 indicates an oilfield injection-production connectivity; finally, the injection wells connected to the first-level connectivity are selected. Secondary interconnected production wells with oilfield injection-production connectivity This represents the union of two-level connected production wells.
[0059] Step 2: Obtain historical production data for the injection and production well locations, specifically including the water injection volume of the primary connected water injection well, the fluid production and oil production of the research object, and the fluid production and oil production of the secondary connected production well. The water injection volume of the primary connected water injection well is a feature of the dataset, and the fluid production and oil production are labels of the dataset.
[0060] The historical production data is preprocessed and a database is established. The data preprocessing includes outlier handling and data standardization. The first 80% of the historical production data is used as the training set, and the last 20% of the historical production data is used as the validation set.
[0061] Step 3: Construct a local well network injection and production parameter optimization model based on a long short-term memory network for the injection and production well locations mentioned in Step 1, such as... Figure 2 and Figure 3 As shown, the local well network injection and production parameter optimization model based on long short-term memory network adopts a multi-input multi-output structure. The injection volume of the first-level connected injection well is used as the input layer, the production volume and oil production of the research object are used as the main output, and the production volume and oil production of the second-level connected production well are used as the auxiliary output.
[0062] When adjusting the production status of a single production well, conventional machine learning methods establish a mapping relationship between the research object and the primary connected injection wells, neglecting the impact of the injection wells on the secondary connected production wells during the production process. This leads to poor accuracy of the local well network injection-production parameter optimization model based on long short-term memory networks. Therefore, this invention considers the multi-directional water breakthrough problem of injection wells in the reservoir, using the secondary connected production wells to constrain the water injection information of the primary connected injection wells, thereby establishing a more accurate model.
[0063] Step 4: Use the training set to train the local well network injection and production parameter optimization model based on long short-term memory network described in Step 3, and obtain the reservoir simulator.
[0064] Specifically, a threshold is set based on the accuracy of the validation set in the local well pattern injection-production parameter optimization model based on long short-term memory network. When the training set reaches the threshold during the training process of the local well pattern injection-production parameter optimization model based on long short-term memory network, the validation set is used as a training sample to continue training, and the finally trained local well pattern injection-production parameter optimization model based on long short-term memory network is used as the reservoir simulator.
[0065] The water injection plan is an adjustment of the water injection volume of the injection well in the latest period. This adjustment process usually generates a large number of water injection plans.
[0066] The reservoir simulator is used to determine the fluid production and oil production of the research object and the secondary connected production well under each water injection scheme, and the water cut is calculated using the following formula:
[0067]
[0068] Among them, f w The water cut refers to the overall water cut of the research object and the secondary interconnected production well. The sum of the oil production of the research object and the secondary connected production wells. The sum of the production volume of the research object and the production volume of the secondary connected production well.
[0069] The objective function is calculated by combining the production rate of the research object and the production rate of the secondary interconnected production well, along with the water cut, and the values are compared. The water injection scheme corresponding to the optimal objective function is selected as the subsequent water injection scheme. The formula for calculating the objective function is as follows:
[0070]
[0071] Where J is the objective function value; α is the importance factor for the yield; f w The water content corresponding to each water injection scheme; This is the sum of the production volume of the research object described in each water injection scheme and the production volume of the secondary interconnected production well; The respective The maximum and minimum values of the objective function are considered. When the objective function value is minimized, the water injection scheme corresponding to the research object is considered optimal. This process standardizes the liquid production rate and takes its reciprocal to bring it to the same dimension as the water content.
[0072] Step 5: Adjust the water injection plan using the historical production data in the database, and optimize the injection and production parameters using the reservoir simulator and objective function.
[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications and equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.
Claims
1. A local well pattern injection-production parameter optimization method based on a long short-term memory network, characterized in that, include: Identify the research object and determine the injection and production well locations based on the oilfield injection-production connectivity of the research object; The injection and production well locations include primary interconnected water injection wells and secondary interconnected production wells; Acquire historical production data of the injection and production well locations; perform data preprocessing on the historical production data and establish a database; simultaneously, divide the preprocessed historical production data into a training set and a validation set according to a set ratio. A local well network injection and production parameter optimization model based on a long short-term memory network is constructed for the aforementioned injection and production well locations; The training set is used to train the local well network injection and production parameter optimization model based on long short-term memory network, resulting in a reservoir simulator; The water injection plan is adjusted using historical production data in the database, and the injection and production parameters are optimized using the reservoir simulator and objective function. Determining the injection-production well locations based on the oilfield injection-production connectivity of the research object includes: The primary interconnected water injection wells to be studied are determined based on the oilfield injection-production connectivity of the research object. The specific screening method is as follows: Where P represents a production well and I represents a water injection well. The research object is specifically the k-th production well with high water cut and low production. This represents the oilfield injection-production connectivity between the i-th injection well and the k-th production well. The value is either 0 or 1, where 0 indicates no oilfield injection-production connectivity and 1 indicates an oilfield injection-production connectivity. Finally, the primary connected water injection wells that have an oilfield injection-production connectivity with the research object are selected. ; Then, according to the oilfield injection-production communication relationship of the primary communicating injection well The secondary communicating production well to be studied is determined, and the specific screening method is as follows: ;in, This indicates that the j-th production well is connected to the primary injection well. The oilfield injection-production connection relationship, The value is either 0 or 1, where 0 indicates no oilfield injection-production connectivity and 1 indicates an oilfield injection-production connectivity; finally, the injection wells connected to the first-level connectivity are selected. The secondary interconnected production well with oilfield injection-production connectivity , This represents the union of the two-level connected production wells; The objective function is calculated by combining the production rate of the research object and the production rate of the secondary interconnected production well, along with the water cut. The values of these values are compared, and the water injection scheme corresponding to the optimal objective function is selected as the subsequent water injection scheme. The formula for calculating the objective function is as follows: ;in, The objective function value; It is an important factor for liquid production; The water content corresponding to each water injection scheme; This is the sum of the production volume of the research object described in each water injection scheme and the production volume of the secondary interconnected production well; , The respective The maximum and minimum values of the objective function are considered, and the objective function is considered optimal when the objective function value is minimized. The water injection scheme corresponding to the research object is the optimal water injection scheme.
2. The long short-term memory network-based local well pattern injection-production parameter optimization method according to claim 1, characterized in that, Based on the production status of the production wells, production wells with high water cut and low production are selected as the research objects.
3. The long short-term memory network-based local well pattern injection-production parameter optimization method according to claim 2, characterized in that, The historical production data of the injection and production well locations includes the water injection volume of the primary connected water injection well, the fluid production and oil production of the research object, and the fluid production and oil production of the secondary connected production well. The water injection volume of the primary connected water injection well is a feature of the dataset, and the fluid production and oil production are labels of the dataset.
4. The method for optimizing local well pattern injection and production parameters based on long short-term memory networks according to claim 3, characterized in that, The local well network injection and production parameter optimization model based on long short-term memory network adopts a multi-input multi-output structure. The injection volume of the first-level connected injection well is used as the input layer, the production volume and oil production of the research object are used as the main output, and the production volume and oil production of the second-level connected production well are used as the auxiliary output.
5. The long short-term memory network-based local well pattern injection-production parameter optimization method according to claim 4, characterized in that, When training the local well pattern injection-production parameter optimization model based on the long short-term memory network, a threshold is set according to the accuracy of the validation set in the local well pattern injection-production parameter optimization model based on the long short-term memory network. When the training set reaches the threshold during the model training process, the validation set is used as the training sample to continue training, and the finally trained local well pattern injection-production parameter optimization model based on the long short-term memory network is used as the reservoir simulator.
6. The long short-term memory network-based local well pattern injection-production parameter optimization method according to claim 5, characterized in that, The water injection plan is an adjustment of the water injection volume of the injection well in the latest period. The adjustment process usually generates a large number of water injection plans. The reservoir simulator is used to determine the fluid production and oil production of the research object under each water injection scheme, as well as the fluid production and oil production of the secondary interconnected production well, and the water cut is calculated using the following formula: ; wherein, is the total water cut of the subject and the secondary connected production well, is the sum of the oil production of the subject and the oil production of the secondary connected production well, is the sum of the fluid production of the subject and the fluid production of the secondary connected production well.
7. The long short-term memory network-based local well pattern injection-production parameter optimization method according to claim 1, characterized in that, The data preprocessing includes outlier handling and data standardization.