A well pattern and well site deployment method for a water injection well in an extra-high water cut stage

Abstract

The present application relates to the technical field of oil exploitation, in particular to a well pattern and well location deployment method for injection wells in an extra-high water cut period. According to the injection well control type in the block in the extra-high water cut period, the present application performs feature screening on the production parameter data of each well in the block, obtains the screened feature production parameter data, constructs a graph network structure based on the screened feature production parameter data and the inter-well connectivity features between the injection wells and the production wells, builds a graph attention network model based on the constructed graph network structure, and predicts the injection well control type in the extra-high water cut period, and deploys the well pattern and well location through the prediction result. The present application significantly improves the accuracy of the injection well control type prediction and improves the effect of the well pattern and well location deployment of the oilfield injection wells.

Description

Technical Field

[0001] This invention relates to the field of oil extraction technology, specifically to a method for deploying well networks and well locations for water injection wells during periods of extremely high water cut. Background Technology

[0002] Water injection is a common measure to improve oil recovery. However, with the deepening of oilfield development, many major oil-bearing layers have entered the ultra-high water-cut stage. Therefore, the dynamic adjustment research of water injection wells is particularly important. Traditional water injection well water lifting or control decisions are usually based on well network correlation. That is, by collecting dynamic production data of oil and water wells to determine the dynamic connectivity of the well network, then studying the impact of production well operations on water injection wells, and finally formulating water injection adjustment plans based on experience.

[0003] However, in practical applications, existing technologies suffer from limited prediction accuracy, high computational complexity, and strong dependence on specific conditions, making them unsuitable for real-time well network and well location deployment decision optimization in large-scale oil reservoir scenarios. While current research has applied graph neural networks to predict dynamic indicators and optimize well locations in oil and water wells—for example, using graph neural network models to simulate outputs and optimize well layout and injection / production strategies—existing research using graph neural network models for well network and well location deployment neglects the impact of injection well control types on well network and well location deployment analysis during ultra-high water cut periods. This results in lower well network and well location deployment effectiveness for injection wells during ultra-high water cut periods, affecting oil production and efficiency. Summary of the Invention

[0004] To address the technical problem of low effectiveness in well network and well location deployment during ultra-high water cut periods due to neglecting the impact of well control type on well network and well location deployment analysis, this invention aims to provide a well network and well location deployment method for water injection wells during ultra-high water cut periods. The specific technical solution adopted is as follows: This invention proposes a method for well network and well location deployment of injection wells during ultra-high water cut periods, the method comprising: Obtain the lifting and control type data of each injection well and the production parameter data of each oil production well in the block to be analyzed at different historical ultra-high water cut periods; The production parameter data is filtered for characteristic production parameters to obtain the characteristic production parameter data of the filtered oil wells; Based on the inter-well connectivity characteristics of water injection wells and oil production wells, as well as the characteristic production parameter data of oil production wells, a graph network structure is constructed for different historical periods of extremely high water cut. The graph network structure includes the connection relationship between water injection wells and oil production wells. The control type prediction model is trained based on the graph network structure and the control type data to obtain the trained control type prediction model. The lifting control type prediction model is used to predict and analyze the lifting control type of water injection wells in the block to be analyzed during the ultra-high water cut period, and the prediction results are obtained. Based on the prediction results, the well network and well locations of water wells in the block to be analyzed are deployed.

[0005] Furthermore, the production parameter data includes at least daily liquid production, daily oil production, daily water production, water cut, formation pressure, and liquid production intensity data.

[0006] Furthermore, the method for filtering the characteristic production parameter data is as follows: Oversample the control type data of the water injection wells, and expand the production parameter data based on the oversampled control type data; Based on the oversampled control type data of the water injection wells, each production parameter data of the oil wells in the block to be analyzed is classified. The effective response value of each production parameter data is determined by the classification results. The effective response value reflects the effectiveness of each production parameter in distinguishing all control types. Based on the effective response value, the characteristic production parameter data in all types of production parameter data are determined.

[0007] Furthermore, the method for constructing the graph network structure is as follows: The water injection wells and oil production wells in the block to be analyzed are respectively regarded as nodes in the graph network structure. Based on the connectivity between the water injection wells and oil production wells, the connection lines and directions between the corresponding nodes of the water injection wells and oil production wells are determined. Based on the connectivity characteristics between water injection wells and oil production wells, and combined with the characteristic production parameter data of oil production wells, the weights of the connections between nodes are determined. The graph network structure is constructed based on the determined nodes, the directions of the connections between nodes, and the weights of the connections between nodes.

[0008] Furthermore, the direction of the connection between the nodes is from the node corresponding to the water injection well to the node corresponding to the oil production well.

[0009] Furthermore, the method for determining the weights is as follows: The connectivity coefficient between water injection wells and oil production wells in the block to be analyzed is determined based on the connectivity strength between them. The connectivity coefficient is corrected based on the correlation between the water injection volume of the injection well and the characteristic production parameter data of the oil well. The corrected connectivity coefficient is used as the weight for the connection between corresponding nodes of water injection wells and oil production wells.

[0010] Furthermore, the process of correcting the connectivity coefficient is as follows: Obtain water injection volume data of injection wells during the historical ultra-high water cut period of the block to be analyzed, and determine the sliding time window based on the periodic variation characteristics of water injection volume of injection wells in the block to be analyzed. For water injection wells and oil production wells that are connected in a graph network structure, the dynamic connectivity strength is determined based on the correlation and change characteristics between the water injection volume of the water injection well and the characteristic production parameter data of the oil production well within the sliding time window. The connectivity coefficient is adjusted based on the dynamic connectivity strength to obtain the corrected connectivity coefficient.

[0011] Furthermore, the method for determining the sliding time window is as follows: For each injection well in the block to be analyzed, the historical injection volume of the injection well is sorted according to the time sequence of collection. The sorted data is used to determine the period length through an autocorrelation function, and a sliding time window is set according to the period length.

[0012] Furthermore, the specific calculation method for the dynamic connectivity strength is as follows: For connected water injection wells and oil production wells, based on the correlation between the water injection volume of the water injection well and the production parameter data of each characteristic of the oil production well within each sliding time window, as well as the degree of fluctuation of the production parameter data of each characteristic of the oil production well, the injection-production dynamic response sensitivity of connected water injection wells and oil production wells for each characteristic production parameter data within each sliding time window is obtained. Based on the variation characteristics of the water injection volume of the injection well within each sliding time window, the dynamic response value of each sliding time window is obtained; Based on the injection-production dynamic response sensitivity and the dynamic response value, the dynamic connectivity strength between the connected water injection wells and oil production wells is determined.

[0013] Furthermore, the method for correcting and adjusting the connectivity coefficients is as follows: Pre-set the weight parameters for connectivity coefficients and dynamic connectivity strength; By using the weight parameters of connectivity coefficient and dynamic connectivity strength, the connectivity coefficient and dynamic connectivity strength are weighted and added together to obtain the corrected connectivity coefficient.

[0014] Furthermore, the training method for the control type prediction model includes: Based on the graph network structure of different historical periods of extremely high water content and the aforementioned control type data, a training set, a validation set, and a test set are constructed. Construct a graph attention network model and set the network hyperparameters of the graph attention network model. The network hyperparameters include at least the number of attention heads, hidden layer dimension, dropout rate, learning rate, regularization parameter, number of training epochs, and tolerance for early stopping mechanism. The graph attention network model is trained using the training set, validation set, and test set to obtain a trained graph attention network model.

[0015] Furthermore, the method also includes: The prediction performance of the trained graph attention network model is evaluated using confusion matrix or ROC curve methods. Based on the evaluation results of the prediction performance, it is determined whether the graph attention network model needs to be optimized. If optimization is required, the hyperparameters of the graph attention network model are adjusted and the model is retrained until optimization is no longer necessary.

[0016] Furthermore, the method for deploying the well network and well locations is as follows: Determine the well network and well location deployment strategy based on the prediction results of the water injection well control type in the block to be analyzed; Based on the well network and well location deployment strategy, the optimal well network and well location deployment scheme is determined by combining the genetic optimization algorithm, and the well network and well locations are deployed according to the optimal well network and well location deployment scheme.

[0017] The present invention has the following beneficial effects: In the process of well network and well location deployment for water injection wells during ultra-high water cut periods, graph neural networks are applied to the prediction of dynamic indicators of oil and water wells and well location optimization. Graph neural networks are used to replace the model's simulation output to optimize oil well layout and injection-production strategies. However, traditional methods neglect the impact of well control type on well network and well location deployment analysis during ultra-high water cut periods, resulting in low accuracy in well network and well location deployment during these periods. To address this, a well network and well location deployment method for water injection wells during ultra-high water cut periods is proposed. Based on the lifting control type of water injection wells within a block during ultra-high water cut periods, feature filtering is performed on the production parameter data of each well in the block to obtain the filtered feature production parameter data. A graph network structure is constructed based on the filtered feature production parameter data and the inter-well connectivity features between water injection wells and production wells. This reduces the impact of the lifting control type of water injection wells during ultra-high water cut periods on the judgment of inter-well relationships in the graph network structure. Furthermore, the graph network structure effectively explores the impact of inter-well relationships on water injection decisions. A graph attention network model is built based on the constructed graph network structure to predict the lifting control type of water injection wells during ultra-high water cut periods, significantly improving the accuracy of water injection well lifting control type prediction and enhancing the effectiveness of well network and well location deployment for oilfield water injection wells. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages 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.

[0019] Figure 1 A flowchart illustrating a method for deploying a well network and well locations for injection wells during ultra-high water cut periods, provided in one embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for constructing a graph network structure according to an embodiment of the present invention. Figure 3 A flowchart illustrating a node connection weight correction process according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a graph network structure provided in one embodiment of the present invention; Figure 5 This is a schematic diagram of the prediction results of the water injection well lifting control type provided in one embodiment of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a well network and well location deployment method for injection wells during ultra-high water cut periods proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] The following description, in conjunction with the accompanying drawings, details a specific scheme for the well network and well location deployment method of injection wells during ultra-high water cut periods provided by this invention.

[0023] Please see Figure 1 The diagram illustrates a flowchart of a well network and well location deployment method for injection wells during ultra-high water cut periods, according to an embodiment of the present invention. The method includes: Step S1: Obtain the lifting and control type data of each injection well and the production parameter data of each oil production well in the block to be analyzed at different historical ultra-high water cut periods.

[0024] To analyze the well network and well location deployment strategy for water injection wells during ultra-high water cut periods, and thus achieve the optimal well network and well location deployment during ultra-high water cut periods, it is necessary to obtain the lifting and control type data of water injection wells in the block to be analyzed, as well as the production parameter data of oil wells during historical ultra-high water cut periods. This data will be used for subsequent predictive analysis of the lifting and control type of water injection wells in the block to be analyzed during ultra-high water cut periods.

[0025] In one specific implementation of this invention, the production parameter data includes at least daily liquid production, daily oil production, daily water production, water cut, formation pressure, and liquid production intensity data.

[0026] Due to differences between injection wells and production wells during actual exploration, it may be impossible to collect all types of production parameter data. Therefore, to ensure the consistency of feature dimensions of production wells and to avoid the failure to construct the same type of nodes or the breakage of the graph structure due to inconsistent feature dimensions during subsequent graph network modeling, in a specific implementation of this invention, for each production well in the block to be analyzed, the production parameter data corresponding to each production well is sorted according to the time sequence of collection to form a time-series feature sequence; for the types of production parameter data missing in the production wells, zero-value filling is used to fill the missing production parameter data of that type, so that the feature dimensions of the production parameter data collected by the production wells are consistent; for example, if some wells are shut down, resulting in missing production parameter data such as daily liquid production, daily oil production, and daily water production, zero-value filling is used to fill the missing production parameter data of that well.

[0027] Because the acquisition of production parameter data may be affected by interference, resulting in poor data quality for each production parameter and the potential presence of outliers, in one specific implementation of this invention, the K-means clustering algorithm can be used to identify and remove outliers. Then, linear interpolation can be used to fill in the missing values ​​resulting from the outlier removal. In a specific example: First, max-min normalization is used to eliminate the influence of the original production data dimensions. For the collected daily liquid production, daily oil production, daily water production, water cut, formation pressure, or production intensity data, the dimensions are unified through normalization processing. The specific processing relationship is as follows: ;in, It is the minimum value of each production parameter data in the production parameter data contained in each oil well. It is the maximum value of each production parameter data. It is the normalized result corresponding to each production parameter data; This represents the raw data for each production parameter before normalization.

[0028] Secondly, the elbow method is used to determine the optimal number of clusters k, where the initial number of candidate clusters k is an integer from 4 to 9. Based on the initial number of clusters k, the k-means clustering algorithm is used to cluster the normalized results of each production parameter data of the oil well. After clustering, the WCSS (intra-cluster sum of squares) value is calculated, and the formula for its calculation is as follows: ;in, For clusters The center of mass, This represents the data point corresponding to the normalized result of each production parameter data in this cluster; This represents the squared distance between the data point corresponding to each production parameter and the centroid. The WCSS-k values ​​calculated using the above formula are shown in Table 1. Table 1 k 4 5 6 7 8 9 WCSS 18.77 14.94 12.54 9.49 8.23 7.13 Based on the data in Table 1, a two-dimensional rectangular coordinate system is constructed with the k value as the x-axis and the WCSS value as the y-axis. The data in Table 1 is mapped to the constructed two-dimensional rectangular coordinate system, and curve fitting is performed using the least squares method to draw the WCSS-k curve. By calculating the slope of the curve, the abrupt change point with the largest change in the curve slope is obtained, and then the optimal number of clusters k corresponding to the abrupt change point is determined.

[0029] Finally, based on the determined optimal number of clusters k, k cluster centers are randomly initialized. Using the K-means clustering algorithm, each data point is assigned to the nearest cluster center, obtaining the clustering results for all data. For each cluster, the Euclidean distance from each data point to the cluster center is calculated. All Euclidean distances are sorted in descending order, and the minimum value of the top 5% of Euclidean distances is used as the outlier threshold. Data points in each cluster with an Euclidean distance greater than the outlier threshold are identified as outliers in each cluster, thus determining the outliers for each production parameter. Outliers are removed, and linear interpolation is used to fill in the missing values. The specific filling formula is as follows: ;in, for[ The value within the interval, ( , )and( , The numbers () represent the coordinates of the next and previous adjacent points of the location of the missing value, respectively. The filling value represents the location of the missing value.

[0030] It should be noted that in other embodiments of the present invention, 3 may also be used. The principle or LOF anomaly detection algorithm identifies abnormal data for each production parameter and uses linear interpolation to fill in the missing data.

[0031] In one specific implementation of this invention, the control type data of different water injection wells during the historical exploration process in the block to be analyzed are obtained. The control type data includes three types of data: lifting, control, and stabilization. Each water injection well corresponds to one of the three control types.

[0032] Step S2: Filter the production parameter data for characteristic production parameters to obtain the characteristic production parameter data of the filtered oil wells.

[0033] During the ultra-high water cut period of the block under analysis, the distribution of injection well lifting, control, and stabilization types is naturally uneven, resulting in significant differences in the lifting and control types of different injection wells. The significance of the correlation response of different production parameter data to the lifting and control types of injection wells varies, affecting the accuracy of subsequent analysis of the connectivity characteristics between injection wells and production wells. This reduces the accuracy and stability of subsequent predictions of injection well lifting and control types, impacting the effectiveness of subsequent well network and location optimization. Therefore, feature filtering of production parameter data is performed to obtain filtered characteristic production parameter data. This filtered characteristic production parameter data can effectively analyze the lifting and control types of injection wells within the block under analysis.

[0034] Preferably, in some possible implementations of the embodiments of the present invention, the method for filtering feature production parameter data is as follows: First, the lifting control type data of the water injection wells is oversampled, and the production parameter data is expanded based on the oversampled lifting control type data.

[0035] Due to the uneven distribution of injection well control types during the ultra-high water cut period, which leads to model training bias, the model may favor the majority class samples while ignoring key features of the minority class samples. Therefore, based on the injection well control type data in the block to be analyzed, the injection well control type data is oversampled to obtain oversampled control type data. The production parameter data is then expanded based on the oversampled control type data to balance the distribution of injection well types.

[0036] In a specific implementation of this invention, a single water injection well and an oil production well within the block to be analyzed are taken as independent water injection well samples and oil production well samples, respectively. Each water injection well sample corresponds to a unique lifting and control type (lifting, control, and stabilization), and each oil production well sample corresponds to all types of production parameter data. The SMOTE algorithm (Synthetic Minority Oversampling Technique) is used to oversample the water injection well samples to balance the distribution of the three types of water injection well lifting, control, and stabilization. The nearest neighbor number is set to 5, and the oversampling ratio is determined by rounding the absolute value of the difference between the minority class and the majority class sample numbers to the minority class sample number. For example, if the majority class sample number is 70 and the minority class sample number is 10, then the oversampling ratio is set to 6 to make the minority class sample number close to the majority class sample number. The random number λ is uniformly sampled in the interval (0,1). In a specific example, the K-nearest neighbor algorithm (K = 5) is first used to obtain the K nearest neighbor well samples (those with the closest spatial distance) for each well sample in the minority class. Then, a well sample is randomly selected from these K nearest neighbor samples as the target well sample. The distance between the current well sample and the target well sample is calculated and denoted as . The product of this distance and the random number is calculated as the distance between the new minority class injection well sample and the current injection well sample, denoted as . On the line connecting the current injection well sample and the target injection well sample, select a distance from the current injection well sample. The location is used as a new minority class water injection well sample for insertion; and the production parameter data of the oil well samples connected to the current water injection well sample or the oil well samples connected to the target water injection well sample are copied, and the copied results are used as the production parameter data of the oil well sample corresponding to the new minority class water injection well sample. This process is used to expand the minority class water injection well samples and their corresponding oil well samples, thereby balancing the distribution of different types of water well samples. The oversampling process of the SMOTE algorithm is well known to those skilled in the art and will not be described in detail here.

[0037] Secondly, based on the oversampled control type data of the water injection wells, each production parameter data of the oil wells in the block to be analyzed is classified, and the effective response value of each production parameter data is determined through the classification results.

[0038] There is a clear well-to-well connectivity between water injection wells and production wells in the block to be analyzed. During the ultra-high water cut period, although oversampling balanced the distribution of control type samples for water injection wells in the block to be analyzed, the differences in control types of water injection wells resulted in varying degrees of response of production parameter data of different production wells to changes in the control type of water injection wells, affecting the accuracy of the control type analysis of water injection wells in the block to be analyzed. Therefore, feature analysis is performed on the production parameter data of production wells in the block to be analyzed based on the control type data of oversampled water injection wells. Based on the feature analysis results, characteristic production parameter data are determined. Through the characteristic production parameter data, the control type of water injection wells in the block to be analyzed can be accurately analyzed, improving the accuracy of subsequent prediction analysis of control type of water injection wells in the block to be analyzed based on graph neural network models.

[0039] In a specific implementation of this invention, the effective response value in the production parameter data can be determined using one-way ANOVA. The effective response value is then the F-value and the η² contribution. First, the types of water injection wells include three types: lifting, control, and stabilization. Based on the connection relationship between oil wells and water injection wells, the oversampled water injection well samples and their corresponding connected oil well samples are classified into the same type. For example, a water injection well of type "lifting" is connected to an oil well of type "lifting." After classifying the oil well samples according to the lifting / control type of the oversampled water injection well samples, the F-value and η² contribution between each production parameter data of the oil well and each lifting / control type of the water injection well are calculated. In a specific example, the specific formula for calculating the F-value is: Where SSB is the between-group sum of squares, SSW is the within-group sum of squares, m is the number of control types, N is the total sample data, and the larger the F value, the greater the difference between groups and the smaller the difference within groups.

[0040] Specifically, the formula for the sum of squares between groups of SSB is as follows: ; The specific formula for the sum of squares within the SSW group is: ; The specific formula for SST is: ; in, It is the first The number of oil wells sampled in each type of controlled extraction method. It is the first The average production parameter data of all oil wells corresponding to the type of control. It is the average of the production parameter data of all oil wells corresponding to all types of lifting and control. It is the first The first type of control Production parameter data for each oil well.

[0041] The specific formula for calculating the contribution of η² is as follows: η² ranges between [0,1] and is used to measure the ability of this feature to explain the type of injection well. A higher value indicates that the feature has a stronger ability to identify the three types of lifting, control, and stabilization.

[0042] Finally, characteristic production parameter data are determined from all types of production parameter data based on the valid response values.

[0043] In a specific implementation of this invention, based on the above formula and analysis, the F-values ​​and η² contribution rates of various production parameters of the oil well can be obtained, as shown in Table 2: Table 2 variable F value η² contribution Daily liquid production 7412.00 0.25 Daily oil production 1066.22 0.04 Daily water production 7705.05 0.26 Moisture content 2016879.54 0.98 Formation pressure 1030273.68 0.97 Liquid production intensity 9837.14 0.31 To avoid significant deviations in the predictive analysis process due to insufficient characteristic production parameter data, three characteristic production parameters with relatively large F-values ​​or η² contributions can be selected as the screening results. They can be sorted from largest to smallest according to their F-values ​​or η² contributions, and the three characteristic production parameters with the highest rankings can be selected as the screened characteristic production parameter data. Therefore, based on the calculation results shown in Table 2, water cut, formation pressure, and production intensity can be used as characteristic production parameter data.

[0044] Step S3: Based on the inter-well connectivity characteristics of water injection wells and oil production wells, as well as the characteristic production parameter data of oil production wells, construct a graph network structure for different historical ultra-high water cut periods. The graph network structure includes the connection relationship between water injection wells and oil production wells.

[0045] To accurately analyze the lifting and control types of water injection wells in the analyzed block during the ultra-high water cut period, the production parameter data of oil production wells were filtered based on the differences in lifting and control types of water injection wells during this period, resulting in filtered characteristic production parameter data. Therefore, a graph network structure can be constructed based on the well connectivity characteristics between water injection wells and oil production wells, as well as the characteristic production parameter data of oil production wells. This structure accurately reflects the well connectivity characteristics between water injection wells and oil production wells in the analyzed block, improving the accuracy of subsequent graph neural network model construction and thus enhancing the accuracy of predicting the lifting and control types of water injection wells during the ultra-high water cut period.

[0046] Preferably, in some possible implementations of the embodiments of the present invention, the method for constructing the graph network structure is described in the following reference: Figure 2 The diagram illustrates a flowchart of a method for constructing a graph network structure according to an embodiment of the present invention, the method comprising: Step S100: Treat the water injection wells and oil production wells in the block to be analyzed as nodes in the graph network structure. Based on the connectivity between the water injection wells and oil production wells, determine the connection lines and directions between the corresponding nodes of the water injection wells and oil production wells.

[0047] To accurately reflect the relationship between oil production wells and water injection wells in the block to be analyzed, as well as the response characteristics of changes in oil production well production parameter data to differences in water injection well control types, each water injection well and oil production well in the block to be analyzed is taken as a node, and the nodes are connected according to the connectivity between water injection wells and oil production wells. The selected characteristic production parameter data is used as the data of the corresponding node of the oil production well.

[0048] Preferably, in some possible implementations of the embodiments of the present invention, the direction of the connection between nodes is: from the node corresponding to the water injection well to the node corresponding to the oil production well; since the oil extraction between the connected water injection wells and oil production wells is based on the influence of the pressure change of the water injection well on the oil production well, in the process of constructing the graph network structure, the direction of the connection between the nodes corresponding to the connected water injection wells and oil production wells is from the node corresponding to the water injection well to the node corresponding to the oil production well, reflecting their injection-production relationship.

[0049] Step S200: Based on the connectivity characteristics between water injection wells and oil production wells, and in conjunction with the characteristic production parameter data of oil production wells, determine the weight of the connections between nodes.

[0050] To comprehensively reflect the fluid flow smoothness between water injection wells and oil production wells in the analysis block, the inter-well connectivity characteristics between water injection wells and oil production wells in the analysis block are analyzed. Based on the analysis results, the weights between corresponding nodes of water injection wells and oil production wells are determined to reflect the degree of inter-well connectivity between water injection wells and oil production wells in the analysis block.

[0051] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the weight of the connection between nodes is as follows: First, the connectivity coefficient between the water injection wells and the oil production wells is determined based on the connectivity strength between them within the block to be analyzed.

[0052] Since the connectivity between water injection wells and oil production wells in the block to be analyzed affects the production output of oil production wells, and the significant changes in oil production parameters are significantly affected by the type of water injection well control, in order to accurately reflect the correlation between water injection wells and oil production wells in the constructed graph network structure, the connectivity coefficient between water injection wells and oil production wells in the block to be analyzed is determined based on the connectivity strength between them.

[0053] In one specific implementation of this invention, tracer monitoring technology is used to monitor and analyze the inter-well connectivity in the block to be analyzed, and the connectivity coefficient between the water injection well and the oil production well is obtained. The process of obtaining the connectivity coefficient is a technology well known to those skilled in the art, and will not be described in detail here.

[0054] Next, the connectivity coefficient is corrected based on the correlation between the water injection volume of the injection well and the characteristic production parameter data of the oil well.

[0055] During the construction of the graph network structure, the judgment of inter-well connectivity in the ultra-high water-cut period is affected by a combination of dynamic and static interference factors such as reservoir heterogeneity, high-permeability crossflow channels, local ineffective water circulation, logging data errors, and tracer adsorption loss. This results in low accuracy in judging the strength of inter-well connectivity, which in turn affects the authenticity and reliability of edge weights in the graph network structure. Therefore, in order to enable the graph attention network model to accurately reflect the inter-well dependency and dynamic response characteristics under complex reservoir conditions in the ultra-high water-cut period based on the real and effective inter-well correlation strength, and to provide high-quality graph structure data for subsequent prediction of water injection well control types and optimization of well network and well location deployment, the edge weights that accurately reflect the actual degree of correlation between water injection wells are corrected and adjusted so that the prediction results and optimization adjustment schemes can fit the actual development status of the reservoir.

[0056] Preferably, in some possible implementations of the embodiments of the present invention, the process of correcting the connection weights between nodes is described in [reference needed]. Figure 3 The diagram illustrates a flowchart of a node connection weight correction process according to an embodiment of the present invention, including: Step S201: Obtain the water injection volume data of the injection wells during the historical ultra-high water cut period of the block to be analyzed, and determine the sliding time window based on the periodic variation characteristics of the water injection volume of the injection wells in the block to be analyzed.

[0057] If multiple factors interfere with the analysis block, it may cause phased changes in the water injection volume of the injection wells within the analysis block. These changes in water injection volume will affect the production of the oil wells. Therefore, in order to accurately reflect the connectivity characteristics between the injection wells and the oil wells under the influence of interference factors, the water injection volume data of the injection wells during the historical period of extremely high water cut in the analysis block are obtained. The sliding time window is determined by combining the periodic change characteristics of the water injection volume of the injection wells, which is used for the accurate analysis of the connectivity characteristics between the injection wells and the oil wells under the phased changes in water injection volume.

[0058] Preferably, in some possible implementations of the embodiments of the present invention, the method for determining the sliding time window is as follows: For each injection well in the block to be analyzed, the historical injection volume of the injection well is sorted according to the time sequence of collection. The sorted data is used to determine the period length through an autocorrelation function, and a sliding time window is set according to the period length. The process of determining the period length through the autocorrelation function is well known to those skilled in the art and will not be described in detail here. The period length is used as the length of the sliding window, and the overlap rate between sliding time windows is 0 (that is, the step size for this sliding is the length of the sliding window). The sliding time window slides from left to right in the sorted data.

[0059] Step S202: For water injection wells and oil production wells that are connected in the graph network structure, determine the dynamic connectivity strength based on the correlation and change characteristics between the water injection volume of the water injection well and the characteristic production parameter data of the oil production well within the sliding time window.

[0060] The higher the effective connectivity between connected water injection wells and oil production wells in the analysis block, the more significant the phased correlation between the water injection volume of the water injection wells and the production parameter data of the oil production wells in the analysis block. Therefore, based on the correlation between the water injection volume and characteristic production parameter data of connected water injection wells and oil production wells within the sliding time window, the dynamic connectivity strength is obtained to reflect the effective dynamic connectivity characteristics between connected water injection wells and oil production wells under the influence of dynamic disturbances.

[0061] Preferably, in some possible implementations of the embodiments of the present invention, the specific method for calculating dynamic connectivity strength is as follows: First, for connected water injection wells and oil production wells, based on the correlation between the water injection volume of the water injection well and the production parameter data of each characteristic of the oil production well within each sliding time window, as well as the degree of fluctuation of the production parameter data of each characteristic of the oil production well, the injection-production dynamic response sensitivity of connected water injection wells and oil production wells for each characteristic production parameter data within each sliding time window is obtained.

[0062] Since the greater the decrease in water cut and increase in oil production in the production wells after water injection in the block under analysis, the stronger the connectivity between the injection wells and the production wells, the more accurate the analysis of the dynamic disturbance transmission capability of water injection fluctuations on the production end parameters, and the more precise the identification of sensitive features that respond significantly to water injection adjustments, the correlation between the water injection volume and each characteristic production parameter data between connected injection wells and production wells in each sliding time window is analyzed. Combined with the degree of fluctuation change of each characteristic production parameter data of the production wells, the sensitivity of injection-production dynamic response is obtained, reflecting the dynamic change difference of the connectivity between connected injection wells and production wells in different time periods. The greater the sensitivity of injection-production dynamic response, the more significant the impact of the change in water injection volume of the injection wells on the water cut, formation pressure and production intensity of the production wells in the current sliding time window, the stronger the correlation of dynamic response between wells, and the more significant the dynamic connectivity characteristics between wells.

[0063] In a specific implementation of this invention, for each sliding time window, the absolute value of the Pearson correlation coefficient between the water injection volume of the injection well and the production parameters (water cut, formation pressure, and production intensity) of the oil well within the corresponding time range of the sliding time window is calculated and denoted as . ; and calculate the range and mean of each characteristic production parameter data of the oil well within the corresponding time range of the sliding time window, and use the ratio of the range to the mean as the characteristic value of the local dynamic fluctuation amplitude between wells, denoted as . If the calculated mean is 0, then... Set to 0; and The product of these values ​​represents the sensitivity of the injection-production dynamic response data of each characteristic production parameter between the injection volume of the injection well and the oil production well connected to the injection well within each sliding time window.

[0064] Secondly, based on the variation characteristics of the water injection volume of the injection well within each sliding time window, the dynamic response value of each sliding time window is obtained.

[0065] Since fluctuations in the injection volume of water injection wells have a significant impact on changes in the production parameter data of oil production wells, if the magnitude of the change in the injection volume over a period of time is greater, and the parameters between water injection wells and oil production wells still show significant correlation characteristics during that period, then the dynamic connectivity characteristics analyzed based on the correlation characteristics between water injection wells and oil production wells will be more accurate. Therefore, based on the magnitude of the change in the injection volume of water injection wells within each sliding time window, the dynamic response value of each sliding time window is obtained. The larger the dynamic response value, the more accurately the correlation change characteristics between water injection wells and oil production wells within the corresponding sliding time window can reflect the dynamic connectivity characteristics between wells.

[0066] In a specific implementation of this invention, the extreme value of water injection volume within each sliding window is calculated. The extreme values ​​corresponding to all sliding time windows are used as input, and the normalization result of all extreme values ​​is obtained using the Softmax function. The normalization result of the extreme value corresponding to each sliding time window is recorded as the dynamic response value.

[0067] Finally, based on the injection-production dynamic response sensitivity and dynamic response value, the dynamic connectivity strength between connected water injection wells and oil production wells is determined.

[0068] When the dynamic fluctuation of water injection volume is more significant within the time period corresponding to different sliding time windows, the stronger the overall dynamic correlation between water injection wells and oil production wells, the more stable the dynamic connectivity characteristics between connected water injection wells and oil production wells, and the higher the connectivity, the less affected by dynamic interference factors during the ultra-high water cut period. Therefore, in order to reduce the impact of various dynamic interferences during the ultra-high water cut period on the judgment of connectivity, the dynamic response values ​​within all sliding time windows and the sensitivity of injection and production dynamic response between connected water injection wells and oil production wells are combined to obtain the dynamic connectivity strength between connected water injection wells and oil production wells, reflecting the effective dynamic connectivity characteristics between connected water injection wells and oil production wells under the influence of dynamic interference.

[0069] In a specific implementation of this invention, the formula for calculating dynamic connectivity strength can be: ;in, This indicates the dynamic connectivity strength between connected water injection wells and oil production wells; Indicates the first The water injection volume of the injection well in the sliding time window and the water injection volume of the oil well in the first sliding time window are related to the water injection volume of the oil well in the first sliding time window. Sensitivity of injection-collection dynamic response between various characteristic production parameter data; Indicates the first The dynamic response value of a sliding time window; This indicates the number of types of characteristic production parameter data corresponding to the oil well, including water cut, formation pressure, and production intensity. This indicates the number of sliding windows.

[0070] Step S203: Adjust the connectivity coefficients according to the dynamic connectivity strength to obtain the corrected connectivity coefficients.

[0071] During the ultra-high water-cut period, the analysis of connectivity strength between connected injection wells and production wells within the analyzed block is subject to interference from multiple factors, including increased reservoir heterogeneity, development of high-permeability crossflow channels, local ineffective water circulation, logging data errors, and tracer adsorption losses. This can lead to significant deviations in the obtained connectivity coefficients. Conversely, the more significant the dynamic correlation changes in parameters between connected injection wells and production wells during the ultra-high water-cut period, the more stable the effective connectivity between them, and the less affected by dynamic interference factors. Therefore, the connectivity coefficients are corrected and adjusted based on dynamic connectivity strength to obtain corrected connectivity coefficients, thereby reducing the impact of interference factors on the connectivity analysis between connected injection wells and production wells during the ultra-high water-cut period.

[0072] Preferably, in some possible implementations of the embodiments of the present invention, the method for correcting and adjusting the connectivity coefficient is as follows: Pre-set the weight parameters for connectivity coefficient and dynamic connectivity strength; use the weight parameters for connectivity coefficient and dynamic connectivity strength to perform weighted summation and fusion to obtain the corrected connectivity coefficient.

[0073] To balance the results of tracer monitoring technology on connectivity analysis and dynamic connectivity analysis during ultra-high water cut periods, and to improve the accuracy of connectivity assessment between connected injection wells and production wells during ultra-high water cut periods, weighted parameters for connectivity coefficient and dynamic connectivity intensity are pre-set. A corrected connectivity coefficient is obtained by weighting the dynamic connectivity intensity and connectivity coefficient with the corresponding weighted parameters. This reduces the impact of dynamic interference factors on the accuracy of connectivity assessment between injection wells and production wells during ultra-high water cut periods by assessing the dynamic correlation between them.

[0074] In one specific implementation of this invention, a maximum-minimum normalization algorithm is used to measure dynamic connectivity strength. After normalization, denoted as V, the relationship for adjusting the connectivity coefficients based on the normalization result can be: ;in This represents the corrected connectivity coefficient. This represents the connectivity coefficient obtained by monitoring inter-well connectivity using tracer monitoring technology. This represents the normalized result of the dynamic connectivity strength between connected water injection wells and oil production wells; and These are the weight parameters for connectivity coefficient and dynamic connectivity strength, respectively.

[0075] It should be understood that, and The settings are used to balance the results of tracer monitoring technology on connectivity analysis and dynamic connectivity analysis during ultra-high water cut periods. Implementers can consider the significance of dynamic interference characteristics during ultra-high water cut periods, such as increased reservoir heterogeneity, development of high-permeability crossflow channels, localized ineffective water circulation, logging data errors, and tracer adsorption losses. For example, when the impact of dynamic interference characteristics is relatively small, the settings can be adjusted accordingly. It is 0.7.

[0076] Step S300: Construct the graph network structure based on the determined nodes, the direction of the connections between nodes, and the weights of the connections between nodes.

[0077] In a specific implementation of this invention, all nodes are connected based on the determined nodes, the direction of the connections between nodes, and the weights of the connections. A weight threshold of 0.7 is set for each node and its connections. If the weight is greater than or equal to the weight threshold, it indicates a high degree of correlation between the nodes, meaning a higher connectivity between the corresponding water injection wells and oil production wells. If the weight is less than the weight threshold, it indicates a moderate degree of correlation between the nodes, meaning a relatively low connectivity between the corresponding water injection wells and oil production wells. The type of connections between nodes is marked based on the weight determination results, resulting in a graph network structure. A schematic diagram of the connected graph network structure is shown below. Figure 4 As shown.

[0078] Step S4: Train the control type prediction model based on the graph network structure and control type data to obtain the trained control type prediction model.

[0079] To further predict and analyze the lifting control type of water injection wells based on the graph network structure of all wells in the constructed block, a graph attention network model needs to be built using the graph network structure to accurately predict the lifting control type of water injection wells.

[0080] Preferably, in some possible implementations of the embodiments of the present invention, the training method of the control type prediction model is as follows: Based on the graph network structure of different historical periods of exceptionally high water content and the data of different control types, training sets, validation sets, and test sets are constructed. A graph attention network model is constructed, and the network hyperparameters of the graph attention network model are set. The network hyperparameters include at least the number of attention heads, the dimension of hidden layers, the dropout rate, the learning rate, the regularization parameter, the number of training epochs, and the tolerance of the early stopping mechanism. The graph attention network model is trained using the training set, validation set, and test set to obtain the trained graph attention network model.

[0081] In a specific implementation of this invention, multiple graph network structures representing different historical periods of extremely high water cut are constructed. The control types of injection wells in the graph network structures are labeled according to their control types during these periods. The dataset, composed of all labeled graph network structures representing different historical periods of extremely high water cut, is divided into training, validation, and test sets in a 6:2:2 ratio. Network hyperparameters are set, including the number of attention heads, hidden layer dimension, dropout rate, learning rate, regularization parameter, number of training epochs, and early stopping tolerance. Specifically, there are 8 attention heads, 16 hidden layer dimensions, ReLU activation function for the hidden layers, a dropout rate of 0.5, a learning rate of 0.01, a regularization parameter of 5e-4, 500 training epochs, and an early stopping strategy where training stops when the loss on the validation set does not decrease for 50 consecutive epochs to prevent overfitting. The Softmax function is used to process the output layer vectors, outputting the probability value of each injection well belonging to each control type.

[0082] Preferably, in some possible implementations of the embodiments of the present invention, a process of optimizing the graph attention network model is also included: First, the predictive performance of the trained graph attention network model is evaluated using confusion matrix or ROC curve methods.

[0083] To improve the accuracy of graph attention network model predictions and avoid large deviations in the prediction results output by the graph attention model, which could affect the well network and well location deployment during periods of extremely high water cut, the prediction performance of the graph attention network model can be evaluated using confusion matrix or ROC curve assessment methods. This verifies the model's generalization ability, allows for optimization and adjustment, prevents overfitting, ensures prediction accuracy, and improves the reliability of subsequent well location deployment strategies.

[0084] In a specific implementation of this invention, ROC curves are used to evaluate the model's predictive performance, and optimization is performed based on the evaluation results: First, the True Positive Rate (TPR) and False Positive Rate (FPR) are calculated, using the following formulas:

[0085]

[0086] In the formula, This indicates one of the three control types: raising, controlling, and stabilizing. It is the correct prediction as the number Number of samples of each type of control Other types of control samples were incorrectly predicted as the first. Number of samples of each type of control It is a correct prediction as non-first Number of samples of each type of control It is the first The number of samples of a certain control type that are incorrectly predicted as other classes. A two-dimensional Cartesian coordinate system is constructed with FPR values ​​as the x-axis and TPR values ​​as the y-axis. A ROC curve is then constructed in this system using curve fitting methods to obtain the first... The ROC curves corresponding to each control type are described below. The detailed construction process of the ROC curves is well known to those skilled in the art and will not be elaborated further. The overall predictive performance is quantified by the mean of the AUC (Area Under Curve) values ​​of the ROC curves corresponding to all control types. The closer the mean AUC value is to 1, the stronger the model's discriminative ability. When the mean AUC value is ≥0.9, the model is considered to have excellent predictive performance; conversely, it indicates poor predictive performance.

[0087] Next, based on the evaluation results of the prediction performance, it is determined whether the graph attention network model needs to be optimized. If optimization is required, the hyperparameters of the graph attention network model are adjusted and the model is retrained until optimization is no longer needed.

[0088] If the evaluation result indicates poor model prediction performance, the model needs to be optimized by fine-tuning the network hyperparameters until the prediction performance evaluation of the graph attention network model satisfies an AUC value ≥ 0.9. This completes the optimized training of the graph attention network model. In a specific example, if the AUC is below 0.9 and the model exhibits underfitting, the prediction performance can be improved by reducing the regularization coefficient until the prediction performance evaluation of the graph attention network model satisfies an AUC value ≥ 0.9. The process of adjusting the performance of the graph attention network model by fine-tuning the network hyperparameters is well-known to those skilled in the art and will not be elaborated further.

[0089] Step S5: Use the lifting control type prediction model to predict and analyze the lifting control type of the water injection wells in the block to be analyzed during the ultra-high water cut period, obtain the prediction results, and deploy the well network and well locations in the block to be analyzed based on the prediction results.

[0090] If the prediction performance of the graph attention network model is poor, it may lead to a large prediction deviation in the control type of water injection wells within the block, thereby affecting the accuracy of the well network and well location deployment strategy within the block. Therefore, the optimized graph attention network model is used to predict and analyze the control type of water injection wells within the block to obtain the prediction results. In a specific implementation of this invention, a schematic diagram of the prediction results of the control type of water injection wells within the block is shown below. Figure 5 As shown.

[0091] During the deployment of injection wells in the ultra-high water cut period, targeted deployment is required based on the lifting and control type of the injection wells. For example, the deployment in the water lifting zone is to enhance the displacement of residual oil, while the deployment in the water control zone is to carry out plugging to redirect the injected water to the medium-low permeability residual oil zone. Therefore, the deployment of wells in the well network and well locations is based on the prediction results of the lifting and control type of the injection wells in the block to be analyzed.

[0092] First, the well network and well location deployment strategy is determined based on the prediction results of the water injection well control type in the block to be analyzed.

[0093] In a specific implementation of this invention, the process of determining the well network deployment strategy is as follows: the predicted control type of water injection wells and the correlation between different types of water injection wells and oil production wells can be obtained through the prediction results. The correlation refers to the type of "high correlation" and "general correlation" between water injection wells and oil production wells. Then, the distribution range of high permeability channeling and the spatial matching degree between the predicted high permeability channeling and the remaining oil enrichment area and the predicted type of water injection wells in the graph network structure model are analyzed and processed by Petrel software to obtain the high potential remaining oil enrichment area corresponding to the water-lifting type water injection wells, the distribution range of high permeability channeling corresponding to the water-controlling type water injection wells, and the contiguous remaining oil development area corresponding to the water-stabilizing type water injection wells. For areas predicted to be water-lifting zones, 1-2 directional water injection wells will be deployed centered on the remaining oil-rich area, along with 2-3 production wells, forming point-like targeted injection-production units with a well spacing of 100-200m to enhance the displacement of remaining oil. For areas predicted to be water-controlling zones, water injection wells will be deployed densely on both sides of the high-permeability crossflow channel to control the water injection volume of a single well. At the same time, profile-adjusting wells will be deployed upstream of the channel to block ineffective water circulation channels, directing the injected water towards the medium- and low-permeability remaining oil area. For areas predicted to be water-stabilizing zones, the original regular well network framework will be retained, and differentiated injection will be implemented for different layers using stratified water injection technology. Priority will be given to stabilizing water injection in layers with high remaining oil saturation, ensuring that the newly deployed water injection wells have optimal connectivity with surrounding oil wells, avoiding areas with severe inter-well interference, and achieving well network and well location deployment during the ultra-high water-cut period.

[0094] Next, based on the well network and well location deployment strategy, the optimal well network and well location deployment scheme is determined by combining the genetic optimization algorithm, and the well network and well locations are deployed according to the optimal well network and well location deployment scheme.

[0095] Since the actual well network and well location deployment process needs to consider maximizing cumulative oil production or economic net present value (NPV), the determined well network and well location deployment strategy needs to be optimized to determine the optimal deployment scheme.

[0096] In a specific implementation of this invention, a genetic optimization algorithm is used with the goal of maximizing cumulative oil production or net present value (NPV). Multiple preliminary well network and well location deployment schemes are determined based on the well network deployment strategy. These schemes are then used as input to obtain the optimal well network and well location deployment scheme using the genetic optimization algorithm. Specifically, the parameters are set as follows: population size is 100-150, number of generations is 200-300, crossover probability is 0.7-0.8, and mutation probability is 0.05-0.1. The fitness function uses a weighted summation. With NPV as the core objective, cumulative oil production is weighted at 0.4, investment cost at 0.3, operating cost at 0.2, and energy cost at 0.1. NPV is correlated with cumulative oil production... The cumulative oil production is directly proportional to the economic net present value, while it is inversely proportional to the investment cost, operating cost, and energy cost. Before weighted summation, the Z-score algorithm is used to normalize the cumulative oil production, investment cost, operating cost, and energy cost data to avoid the influence of dimensional differences on the calculation results. At the same time, constraint conditions and judgment thresholds are set, including a minimum well spacing ≥ 50m, a distance of ≥ 30m between new injection wells and high-permeability channel, and a remaining oil-rich area coverage ratio ≥ 85%. Based on the above parameter settings, a genetic optimization algorithm is used to efficiently search for the globally optimal well network and well location deployment scheme based on the deployment strategy, and the corresponding deployment is carried out according to the optimal well network and well location deployment scheme. The specific process of using the genetic optimization algorithm to determine the optimal scheme is well-known to those skilled in the art and will not be described in detail here.

[0097] It should be understood that multiple preliminary well network and well location deployment schemes can be determined based on the well network and well location deployment strategy. The determined well network and well location deployment schemes include the coordinates of new injection wells, well type, upper limit of single-well injection volume, ratio of injection wells, minimum well spacing threshold, boundary of drillable area, avoidance range of high-permeability flow channels, priority of coverage of remaining oil-rich areas, and constraints on surface facilities. The process of formulating well network and well location deployment schemes based on the well network and well location deployment strategy is well known to those skilled in the art and will not be described in detail here.

[0098] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0099] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for deploying a well network and well locations for injection wells during ultra-high water cut periods, characterized in that, The method includes: Obtain the lifting and control type data of each injection well and the production parameter data of each oil production well in the block to be analyzed at different historical ultra-high water cut periods; The production parameter data is filtered for characteristic production parameters to obtain the characteristic production parameter data of the filtered oil wells; Based on the inter-well connectivity characteristics of water injection wells and oil production wells, as well as the characteristic production parameter data of oil production wells, a graph network structure is constructed for different historical periods of extremely high water cut. The graph network structure includes the connection relationship between water injection wells and oil production wells. The control type prediction model is trained based on the graph network structure and the control type data to obtain the trained control type prediction model. The lifting control type prediction model is used to predict and analyze the lifting control type of water injection wells in the block to be analyzed during the ultra-high water cut period, and the prediction results are obtained. Based on the prediction results, the well network and well locations of water wells in the block to be analyzed are deployed.

2. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 1, characterized in that, The production parameter data includes at least the daily liquid production, daily oil production, daily water production, water content, formation pressure, and liquid production intensity data.

3. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 1, characterized in that, The method for filtering the characteristic production parameter data is as follows: Oversample the control type data of the water injection wells, and expand the production parameter data based on the oversampled control type data; Based on the oversampled control type data of the water injection wells, each production parameter data of the oil wells in the block to be analyzed is classified. The effective response value of each production parameter data is determined by the classification results. The effective response value reflects the effectiveness of each production parameter in distinguishing all control types. Based on the effective response value, the characteristic production parameter data in all types of production parameter data are determined.

4. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 1, characterized in that, The process of constructing the graph network structure is as follows: The water injection wells and oil production wells in the block to be analyzed are respectively regarded as nodes in the graph network structure. Based on the connectivity between the water injection wells and oil production wells, the connection lines and directions between the corresponding nodes of the water injection wells and oil production wells are determined. Based on the connectivity characteristics between water injection wells and oil production wells, and combined with the characteristic production parameter data of oil production wells, the weights of the connections between nodes are determined. The graph network structure is constructed based on the determined nodes, the directions of the connections between nodes, and the weights of the connections between nodes.

5. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 4, characterized in that, The direction of the connection between the nodes is from the node corresponding to the water injection well to the node corresponding to the oil production well.

6. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 4, characterized in that, The method for determining the weights is as follows: The connectivity coefficient between water injection wells and oil production wells in the block to be analyzed is determined based on the connectivity strength between them. The connectivity coefficient is corrected based on the correlation between the water injection volume of the injection well and the characteristic production parameter data of the oil well. The corrected connectivity coefficient is used as the weight for the connection between corresponding nodes of water injection wells and oil production wells.

7. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 6, characterized in that, The process of correcting the connectivity coefficient is as follows: Obtain water injection volume data of injection wells during the historical ultra-high water cut period of the block to be analyzed, and determine the sliding time window based on the periodic variation characteristics of water injection volume of injection wells in the block to be analyzed. For water injection wells and oil production wells that are connected in a graph network structure, the dynamic connectivity strength is determined based on the correlation and change characteristics between the water injection volume of the water injection well and the characteristic production parameter data of the oil production well within the sliding time window. The connectivity coefficient is adjusted based on the dynamic connectivity strength to obtain the corrected connectivity coefficient.

8. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 7, characterized in that, The method for determining the sliding time window is as follows: For each injection well in the block to be analyzed, the historical injection volume of the injection well is sorted according to the time sequence of collection. The sorted data is used to determine the period length through an autocorrelation function, and a sliding time window is set according to the period length.

9. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 7, characterized in that, The specific calculation method for the dynamic connectivity strength is as follows: For connected water injection wells and oil production wells, based on the correlation between the water injection volume of the water injection well and the production parameter data of each characteristic of the oil production well within each sliding time window, as well as the degree of fluctuation of the production parameter data of each characteristic of the oil production well, the injection-production dynamic response sensitivity of connected water injection wells and oil production wells for each characteristic production parameter data within each sliding time window is obtained. Based on the variation characteristics of the water injection volume of the injection well within each sliding time window, the dynamic response value of each sliding time window is obtained; Based on the injection-production dynamic response sensitivity and the dynamic response value, the dynamic connectivity strength between the connected water injection wells and oil production wells is determined.

10. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 7, characterized in that, The method for correcting and adjusting the connectivity coefficient is as follows: Pre-set the weight parameters for connectivity coefficients and dynamic connectivity strength; By using the weight parameters of connectivity coefficient and dynamic connectivity strength, the connectivity coefficient and dynamic connectivity strength are weighted and added together to obtain the corrected connectivity coefficient.

11. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 1, characterized in that, The training method for the lifting control type prediction model includes: Based on the graph network structure of different historical periods of extremely high water content and the aforementioned control type data, a training set, a validation set, and a test set are constructed. Construct a graph attention network model and set the network hyperparameters of the graph attention network model. The network hyperparameters include at least the number of attention heads, hidden layer dimension, dropout rate, learning rate, regularization parameter, number of training epochs, and tolerance for early stopping mechanism. The graph attention network model is trained using the training set, validation set, and test set to obtain a trained graph attention network model.

12. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 11, characterized in that, The method further includes: The prediction performance of the trained graph attention network model is evaluated using confusion matrix or ROC curve methods. Based on the evaluation results of the prediction performance, it is determined whether the graph attention network model needs to be optimized. If optimization is required, the hyperparameters of the graph attention network model are adjusted and the model is retrained until optimization is no longer necessary.

13. The well network and well location deployment method for injection wells during ultra-high water cut periods according to claim 1, characterized in that, The method for deploying the well network and well locations is as follows: Determine the well network and well location deployment strategy based on the prediction results of the water injection well control type in the block to be analyzed; Based on the well network and well location deployment strategy, the optimal well network and well location deployment scheme is determined by combining the genetic optimization algorithm, and the well network and well locations are deployed according to the optimal well network and well location deployment scheme.

Images