A neural network-based double casing perforation parameter optimization method and system

Abstract

The present application relates to the technical field of oil and gas well perforation completion operation, and discloses a double-layer casing perforation parameter optimization method based on a neural network, which comprises the following steps: obtaining the correlation between perforation parameters and perforation well productivity by means of a fuzzy comprehensive evaluation method according to perforation parameter sample data and perforation well productivity sample data; screening the perforation parameter sample data and the perforation well productivity sample data according to the correlation between the perforation parameters and the perforation well productivity, and training a perforation completion productivity prediction model by means of a convolutional neural network; and constructing an optimization strategy network of the perforation parameters by means of the perforation completion productivity prediction model, so as to optimize the perforation parameters used for double-layer casing perforation completion. The present application can assist the operating personnel to quickly and accurately obtain the best double-layer casing perforation parameter combination, and improve the efficiency of double-layer casing perforation completion.

Description

Technical Field

[0001] This invention relates to the field of oil and gas well perforation and completion technology, and in particular to a method and system for optimizing perforation parameters of double-casing based on neural networks. Background Technology

[0002] Perforation is a technique that uses energy such as gunpowder or hydraulic pressure to penetrate the casing, cement sheath, and formation near the wellbore, establishing a fluid flow channel between the formation and the wellbore. It is a crucial component of oil and gas field completion engineering and a key step in well testing, generally used in casing wells but also applicable to open-hole wells. As the difficulty of oil and gas exploration and development increases, the requirements for perforation technology are also becoming more demanding. Selecting a reasonable perforation process and optimizing perforation parameters based on factors such as the geological characteristics of the oil and gas reservoir, fluid properties, and well type plays a significant role in increasing production capacity, reducing well workover and re-perforation operations, and improving the efficiency of oil and gas field development and production.

[0003] Casing perforation is currently the most widely used well completion method both domestically and internationally, accounting for over 90%. Compared to other completion methods, shaped charge perforation completion technology has always been a highly efficient perforation completion technology primarily aimed at maximizing oil and gas production capacity. It can prevent crosstalk between formations, maximize the connection of oil and gas production channels, increase production capacity, and thus enable effective production control and monitoring. It is particularly suitable for completing multi-layered systems and oil and gas formations with significant inter-layer differences. With the development of unconventional oil and gas resources, perforation is no longer a direct means of obtaining production capacity, but rather a preliminary step in production enhancement measures such as sand fracturing or acid fracturing. The perforated hole only becomes the final oil and gas production channel after being eroded by fracturing fluids and polished by proppant. Perforation directly affects the implementation and effectiveness of production enhancement measures, and indirectly affects the production capacity of oil and gas wells.

[0004] The focus of oil and gas resource exploration and development in my country is gradually shifting from onshore to offshore, and offshore oil and gas field development technologies are receiving increasing attention. Perforated completion technology, as a crucial component of oil and gas well development, helps to maximize oil and gas production capacity and improve economic efficiency. Casing perforated completion, currently the most widely used completion method in offshore oil and gas fields, has seen rapid development in its technology and processes.

[0005] In actual perforation completion processes, special geological formations exist, such as mudstone interlayers and sandstone reservoirs, which are prone to collapse and sand production. Furthermore, the stability of the wellbore and sand control requirements after completion are very high. However, single-layer casing is insufficient to handle the non-uniform loads applied under complex conditions, necessitating double-layer casing perforation completion to ensure efficient production. To achieve optimal perforation results and maximize economic benefits in actual production, extensive research and analysis are required before construction. However, most existing studies on double-layer casing perforation parameter optimization combine numerical simulation with field tests and theoretical analysis, which are highly subjective and involve cumbersome numerical simulation processes. No effective method exists to optimize double-layer casing perforation parameters and improve perforation quality.

[0006] Therefore, there is an urgent need for a method to optimize the perforation parameters of double-casing casing in order to quickly and accurately obtain the optimal perforation parameters and improve the completion efficiency of double-casing casing perforation. Summary of the Invention

[0007] This invention provides a method and system for optimizing perforation parameters of double-casing wells based on neural networks, which solves the shortcomings of existing technologies that rely on historical perforation and completion experience for experimental research and theoretical analysis of double-casing perforation and completion operations, resulting in high subjectivity, poor accuracy, and low efficiency.

[0008] This invention provides a method for optimizing perforation parameters of a double-layer casing based on a neural network, comprising:

[0009] Obtain perforation parameter sample data and perforated well productivity sample data;

[0010] Based on the perforation parameter sample data and perforated well production capacity sample data, the correlation between perforation parameters and perforated well production capacity is obtained through fuzzy comprehensive evaluation method.

[0011] Based on the correlation between perforation parameters and perforation well productivity, perforation parameter sample data and perforation well productivity sample data are screened. The screened perforation parameter sample data is used as the input variable, and the perforation well productivity sample data corresponding to the screened perforation parameter sample data is used as the target variable. A convolutional neural network is used to train a perforation completion productivity prediction model.

[0012] By using a perforation completion productivity prediction model, an optimization strategy network for perforation parameters is constructed to optimize the perforation parameters used in double-casing perforation completion.

[0013] According to the present invention, a method for optimizing perforation parameters of a double-casing casing based on a neural network is provided, wherein obtaining perforation parameter sample data and perforated well productivity sample data includes:

[0014] Obtain perforation parameter sample data;

[0015] Well breaking simulation was performed based on perforation parameter sample data to obtain production capacity ratio sample data;

[0016] Based on the production ratio sample data, the open-hole well production sample data is obtained through the first expression;

[0017] Based on the production ratio sample data and the open hole well production sample data, the perforated well production sample data is obtained through the second expression.

[0018] Preprocessing is performed on perforation parameter sample data and perforation well productivity sample data, including dimensionless processing and / or outlier processing.

[0019] The first expression is:

[0020]

[0021] In the first expression, Q0 represents the sample data of open-hole well productivity, m 3 / d;K h Indicates penetration rate, um 2 ;P wf P represents the flow pressure, in MPa; e Indicates the bottom hole static pressure, MPa; u g Indicates crude oil viscosity, mP a ·s; T represents temperature, °C; r eh Indicates the oil drain radius, in meters (m); r wh Represents the wellbore radius, in meters (m).

[0022] The second expression is:

[0023] PRI = Q p / Q o ,

[0024] In the second expression, PRI represents the capacity ratio sample data, and Q P Q0 represents the production capacity sample data of perforated wells, while Q0 represents the production capacity sample data of open-hole wells.

[0025] According to the present invention, a method for optimizing perforation parameters of double-layer casing based on neural networks is provided. The method involves obtaining the correlation between perforation parameters and perforated well productivity using a fuzzy comprehensive evaluation method, based on perforation parameter sample data and perforated well productivity sample data. This includes:

[0026] Based on the perforation parameter sample data, determine the evaluation level of each perforation parameter;

[0027] Based on the perforation parameter sample data and the perforation well production capacity sample data, determine the fuzzy weight vector of each perforation parameter;

[0028] A fuzzy evaluation matrix is ​​constructed based on the fuzzy weight vectors of each perforation parameter;

[0029] Based on the evaluation level of each perforation parameter and the fuzzy evaluation matrix, the correlation between perforation parameters and perforated well productivity is obtained.

[0030] The expression for the fuzzy evaluation matrix is ​​as follows:

[0031]

[0032] In the expression for the fuzzy evaluation matrix, R represents the fuzzy evaluation matrix. nm This represents the subordinate element relationship between the evaluation levels corresponding to the elements in the X domain, where the element refers to the perforation parameter and the evaluation object refers to the perforation well productivity.

[0033] Furthermore, the process of obtaining the correlation between perforation parameters and perforated well productivity based on the evaluation level and fuzzy evaluation matrix of each perforation parameter includes:

[0034] Based on the fuzzy evaluation matrix, the fuzzy comprehensive evaluation result vector at the evaluation index level is obtained through the third expression;

[0035] Based on the fuzzy comprehensive evaluation result vector at the evaluation index level and the evaluation level of each perforation parameter, the correlation between perforation parameters and perforation well productivity is obtained through the fourth expression.

[0036] The third expression is:

[0037] B = W * R,

[0038] In the third expression, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, W represents the fuzzy subset of the membership relationship between each element in the X domain and the evaluated object, and R represents the fuzzy evaluation matrix.

[0039] The fourth expression is:

[0040] N = BE T ,

[0041] In the fourth expression, N represents the correlation between perforation parameters and perforated well productivity, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, and E represents the score of the evaluation level of each perforation parameter.

[0042] According to the present invention, a method for optimizing perforation parameters of double-casing casing based on neural networks is provided. The method involves using selected perforation parameter sample data as input variables, and the corresponding perforated well productivity sample data as target variables. A convolutional neural network is then used to train a perforated well completion productivity prediction model. The method includes:

[0043] The root mean square error method is used to obtain the model accuracy for each training of the perforation completion production capacity prediction model;

[0044] When the model accuracy meets the pre-approval requirements, the perforation completion production capacity prediction model training is deemed complete.

[0045] The expression for the root mean square error method is:

[0046]

[0047] In the expression for the root mean square error method, RMSE represents the model accuracy each time the perforation completion production capacity prediction model is trained, Y represents the actual perforation completion production capacity, Y' represents the predicted perforation completion production capacity, and K represents the number of samples in the validation data sample set used to validate the perforation completion production capacity prediction model.

[0048] According to the present invention, a method for optimizing perforation parameters in double-casing wells based on a neural network is provided. The method involves constructing an optimization strategy network for perforation parameters using a perforation completion productivity prediction model to optimize the perforation parameters for double-casing perforation completion. The method includes:

[0049] Based on the perforation completion productivity model obtained through training, a multi-dimensional relationship diagram of perforation parameters is constructed. The multi-dimensional relationship diagram includes a first relationship diagram related to productivity, a second relationship diagram related to casing strength, and a third relationship diagram related to cost.

[0050] By using the original graph attention network, feature extraction, feature concatenation, and feature fusion are performed on the multi-dimensional relationship graph of the perforation parameters to obtain the fused relationship features.

[0051] Obtain real relation feature annotations, combine them with the fused relation features, and obtain an optimized graph attention network through a preset optimizer;

[0052] Based on the fused relational features and the optimized graph attention network, a strategy network for the perforation parameters of the double-layer casing is constructed. The strategy network includes a state factor for representing the state, an action factor for representing the action, and a reward factor for representing the reward.

[0053] Based on the fused relational characteristics, state factors, action factors, and reward factors, the strategy network parameters are updated using a strategy gradient to optimize the decision-making performance of the strategy network, resulting in an optimized strategy network for optimizing perforation parameters used in double-casing perforation completion.

[0054] According to the present invention, a method for optimizing perforation parameters of a double-layer casing based on a neural network is provided. The method involves extracting, concatenating, and fusing features from a multi-dimensional relationship graph of the perforation parameters using an original graph attention network to obtain fused relationship features, including:

[0055] By using the original graph attention network, features are extracted from the first, second, and third relationship graphs to obtain production capacity relationship features, casing strength relationship features, and cost relationship features.

[0056] The features of capacity relationship, casing strength relationship, and cost relationship are concatenated to obtain a feature vector;

[0057] Based on feature vectors, feature fusion is performed on capacity relationship features, casing strength relationship features, and cost relationship features through linear mapping to obtain fused relationship features;

[0058] The expression for the fused relational features is as follows:

[0059] F = Linear(Concat(F) c ,F q ,F m )),

[0060] In the expression of the merged relational features, F represents the merged relational features. c F represents the characteristics of the production capacity relationship. q F represents the characteristic of the casing strength relationship. m The `concat()` function represents the cost relationship characteristics, and `linear()` represents the vector concatenation operator.

[0061] Furthermore, the process of obtaining real relation feature annotations, combining them with the fused relation features, and then using a preset optimizer to obtain an optimized graph attention network specifically involves:

[0062] Obtain the true relation feature annotations, combine them with the fused relation features, and use the mean squared error loss method through a preset optimizer to obtain an optimized graph attention network;

[0063] The expression for the mean squared error loss method is as follows:

[0064] Loss(θ) = MSE(F, F gt ),

[0065] In the expression for the mean squared error loss method, θ represents the original graph attention network, and F... gt The true relational feature is represented by F, the fused relational feature is represented by F, and the mean squared error loss function is represented by MSE().

[0066] According to the present invention, a method for optimizing the perforation parameters of a double-layer casing based on a neural network is provided. The method involves updating the policy network parameters using policy gradients based on fused relational features, state factors, action factors, and reward factors to optimize the decision-making performance of the policy network and obtain an optimized policy network. The method includes:

[0067] Initialize the policy network and the valuation network, taking the fused relational features as input, and output the action probability distribution vector and state value respectively. The expressions for the action probability distribution vector and state value are as follows:

[0068] P = Policy(F), V = Value(F)

[0069] In the formula, Policy(F) and V=Value(F) represent the policy network and the value network, respectively, and F represents the input;

[0070] The policy gradient is obtained based on the action probability distribution vector and the state value;

[0071] By maximizing the policy gradient using the gradient ascent method and updating the policy network parameters, the decision performance of the policy network is optimized, and an optimized policy network is adaptively formed.

[0072] The expression for the policy gradient is:

[0073]

[0074] In the expression for the policy gradient, φ represents the policy network parameters, T represents the total number of iterations, t represents the current round, and R... t V represents the reward value in the current round, which is calculated based on the action probability distribution vector. t This represents the state value in the current round.

[0075] This invention also provides a neural network-based system for optimizing perforation parameters in a double-layer casing, comprising:

[0076] The sample data acquisition module is used to obtain perforation parameter sample data and perforated well productivity sample data.

[0077] The correlation calculation module is used to: obtain the correlation between perforation parameters and perforation well productivity based on perforation parameter sample data and perforation well productivity sample data through fuzzy comprehensive evaluation method;

[0078] The perforation completion production capacity prediction model training module is used to: filter perforation parameter sample data and perforation production capacity sample data according to the correlation between perforation parameters and perforation production capacity, use the filtered perforation parameter sample data as input variables, use the perforation production capacity sample data corresponding to the filtered perforation parameter sample data as target variables, and use a convolutional neural network to train the perforation completion production capacity prediction model.

[0079] The double-casing perforation parameter optimization module is used to: construct an optimization strategy network for perforation parameters through a perforation completion productivity prediction model, so as to optimize the perforation parameters used for double-casing perforation completion.

[0080] The present invention also provides an electronic device, including a processor and a memory storing a computer program, characterized in that the processor executes the computer program to implement any of the above-described neural network-based double-layer casing perforation parameter optimization methods.

[0081] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described neural network-based double-casing perforation parameter optimization methods.

[0082] The present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute any of the above-described neural network-based double-layer casing perforation parameter optimization methods.

[0083] This invention provides a neural network-based method and system for optimizing perforation parameters in double-casing wells. Based on perforation parameter sample data, corresponding perforated well productivity sample data is obtained using fluid simulation software. Then, a fuzzy comprehensive evaluation method is used to determine the correlation between perforation parameters and perforated well productivity. Based on this correlation, the perforation parameter sample data and perforated well productivity sample data are filtered, and the perforation parameter sample data and perforated well productivity sample data with high correlation are used to train a perforation completion productivity prediction model. Finally, an optimization strategy network for perforation parameters is constructed to optimize the perforation parameters used in double-casing perforation completion. This method can assist operators in quickly and accurately obtaining the optimal combination of double-casing perforation parameters, thereby improving the efficiency of double-casing perforation completion. Attached Figure Description

[0084] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0085] Figure 1 This is one of the flowcharts illustrating a neural network-based method for optimizing perforation parameters in a double-layer casing, as provided by the present invention.

[0086] Figure 2 This is the second flowchart illustrating a method for optimizing perforation parameters of a double-layer casing based on a neural network, as provided by the present invention.

[0087] Figure 3 This is a schematic diagram of outlier detection in a line graph box.

[0088] Figure 4 This is a schematic diagram of the training process for a perforation completion productivity prediction model.

[0089] Figure 5 A flowchart illustrating the process for optimizing perforation parameters in double-casing perforated well completion.

[0090] Figure 6 This is a schematic diagram of a neural network-based double-layer casing perforation parameter optimization system provided by the present invention.

[0091] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0092] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0093] The following is combined with Figures 1-7 This invention describes a neural network-based method and system for optimizing perforation parameters in double-layer casing.

[0094] Figures 1-2 This is a flowchart illustrating the neural network-based perforation parameter optimization method for double-layer casing provided by the present invention. (Refer to...) Figure 1 The present invention provides a method for optimizing perforation parameters of a double-layer casing based on a neural network, which may include:

[0095] Step S110: Obtain perforation parameter sample data and perforation well productivity sample data. The perforation parameters include any one of the following or any combination thereof: hole depth, hole density, hole diameter, phase angle, etc.

[0096] Step S120: Based on the perforation parameter sample data and the perforation well production capacity sample data, obtain the correlation degree between the perforation parameters and the perforation well production capacity through the fuzzy comprehensive evaluation method;

[0097] Step S130: Based on the correlation between perforation parameters and perforation well productivity, the perforation parameter sample data and perforation well productivity sample data are screened. The screened perforation parameter sample data is used as the input variable, and the perforation well productivity sample data corresponding to the screened perforation parameter sample data is used as the target variable. A convolutional neural network is used to train the perforation completion productivity prediction model.

[0098] Step S140: Construct an optimization strategy network for perforation parameters using the perforation completion productivity prediction model to optimize the perforation parameters for double-casing perforation completion.

[0099] It should be noted that the execution subject of the neural network-based double-layer casing perforation parameter optimization method provided by the present invention can be any network-side device / terminal-side device that meets the technical requirements, such as a neural network-based double-layer casing perforation parameter optimization device, etc.

[0100] In one embodiment, reference is made to... Figure 2 Step S110 can be achieved through the following steps:

[0101] The perforation parameters that mainly affect the productivity of double-casing perforated well completion are obtained, including hole depth, hole density, hole diameter, and phase angle. Perforation parameter sample data are formed, and numerical simulation schemes for the productivity of double-casing perforated well completion under different combinations of perforation parameters are formulated. In actual perforation operations, the impact of multiple perforation parameters on productivity must be fully considered. Perforation parameters include hole density, hole diameter, hole depth, and phase angle. If the perforation parameters are not selected reasonably, it will cause a decrease in perforation operation efficiency, oil and gas flow efficiency, and oil and gas well productivity.

[0102] Well breaking simulations are performed based on perforation parameter sample data to obtain production ratio sample data. For example, the well breaking simulation model can be established using the Design Modeler module built into the fluid simulation software ANSYS Fluent. Then, the well breaking simulation model established in the Design Modeler module of ANSYS Fluent is imported into the Mesh module of ANSYS Fluent to complete the mesh generation. The discrete mesh is then imported into the Fluent solver. After entering the Fluent solver, relevant parameters are set as required, and a wellhead flow detector is set to detect changes in wellhead flow. According to the pre-defined experimental plan, production ratio sample data is obtained through simulation using ANSYS Fluent software.

[0103] Based on the production ratio sample data, the open hole well production sample data is obtained through the first expression. Then, based on the production ratio sample data and the open hole well production sample data, the perforated well production sample data is obtained through the second expression.

[0104] The first expression is:

[0105]

[0106] In the first expression, Q0 represents the sample data of open-hole well productivity, m 3 / d;K h Indicates penetration rate, um 2 ;P wf P represents the flow pressure, in MPa; e Indicates the bottom hole static pressure, MPa; u g Indicates crude oil viscosity, mP a ·s; T represents temperature, °C; r eh Indicates the oil drain radius, in meters (m); r wh Represents the wellbore radius, in meters (m).

[0107] The second expression is:

[0108] PRI = Q p / Q o ,

[0109] In the second expression, PRI represents the capacity ratio sample data, and Q P Q0 represents the production capacity sample data of perforated wells, while Q0 represents the production capacity sample data of open-hole wells.

[0110] Then, the perforation parameter sample data and perforation well productivity sample data are preprocessed, including dimensionless processing and / or outlier processing.

[0111] Because the physical meanings and dimensions of the various factors differ, the results obtained from numerical observations are likely to vary significantly. Ignoring this and directly performing analysis and calculations could very likely lead to errors that exaggerate or underestimate the effect of one or more factors. Therefore, it is essential to perform dimensionless preprocessing on all the data involved. Methods for performing dimensionless preprocessing on all data include: the mean method, the initial value method, etc.

[0112] The expression for the mean method is:

[0113]

[0114]

[0115] In the expression for the mean method, x i (k) represents each data point in the corresponding original sequence. y represents the average values ​​of each sequence. i(k) represents the new sequence obtained after the mean transformation, where n represents the total number of data, k represents the original sequence, and i represents the nth sequence.

[0116] Dividing each original data point in the same sequence by the first data point initializes the sequence, resulting in a new sequence of multiples. The sequence has a dimension of one, all values ​​are greater than zero, and they share a common starting point. Most of the data points in the sequence are close to the dimension of one, and the curves of the sequences intersect each other.

[0117]

[0118] In the formula, x i (k) represents each data point in the corresponding original sequence, x i (0) represents the first data in the corresponding original sequence, y i (k) represents the new sequence obtained after the mean transformation.

[0119] Outliers are data that deviates from the expected pattern. In some modeling scenarios, ignoring outliers can lead to seriously biased conclusions. Therefore, identifying and handling outliers is crucial for subsequent processes. Outlier detection in data cleaning typically employs three methods: simple statistical analysis, the 3σ rule, and box plots. Box plots, in particular, are a method for visualizing outliers, using a graphical representation of the dataset's quartiles, and are both simple and efficient.

[0120] See Figure 3 A box plot consists of the upper quartile, upper quartile, mean, median, lower quartile, and lower quartile. The lower quartile corresponds to the 25th percentile (Q1); the median to the 50th percentile (Q2); the upper quartile corresponds to the 75th percentile (Q3); the upper quartile is calculated as Q3 + 1.5(Q3 - Q1); and the lower quartile is calculated as Q1 - 1.5(Q3 - Q1). Here, Q3 - Q1 represents the interquartile range. The upper and lower quartiles mark the boundaries of the data distribution; values ​​below the lower quartile or above the upper quartile are considered outliers. Data cleaning typically employs various methods to handle outliers, including: deleting outliers, treating them as missing values, correcting the mean, using the capping method, binning, regression imputation, and multiple imputation.

[0121] The preprocessed perforation parameter sample data and perforated well productivity sample data have high integrity and better accuracy, which helps to improve the efficiency of subsequent training of perforated well completion productivity prediction models.

[0122] In one embodiment, step S120 can be implemented through the following steps:

[0123] Based on the perforation parameter sample data, determine the evaluation level of each perforation parameter. For example, first determine the domain of discourse for each evaluation factor, and then establish the perforation parameter set, such as X = {X1, X2, X3, ..., X...}. n In the above set of perforation parameters, X1, X2, X3, ..., X... n Let V represent the first perforation parameter, the second perforation parameter, the third perforation parameter, ..., the nth perforation parameter (including orifice density, orifice diameter, orifice depth, and phase angle, etc.). A set of perforation parameters is then established. Next, the rating level is determined, and the evaluation set is defined as V = {V1, V2, V3, ..., V...}. n Furthermore, the scores corresponding to each item in the evaluation set are determined as E = {E1, E2, E3, ..., E}. n}, in which E1, E2, E3, ..., E n These represent the scores for the first evaluation level, the second evaluation level, the third evaluation level, ..., the nth evaluation level.

[0124] Based on perforation parameter sample data and perforated well productivity sample data, a fuzzy weight vector for each perforation parameter is determined. For example, based on the universe of discourse X, W represents the membership relationship between each element and the evaluated object. The fuzzy method can assign different weights to rating objects with different importance in the universe of discourse X. Therefore, W can be represented as a fuzzy subset on the universe of discourse X, expressed as: W = W1, W2, W3, ..., W n , where n is the number of rating indicators, and

[0125] Based on the fuzzy weight vectors of each perforation parameter, a fuzzy evaluation matrix is ​​constructed. For example, a fuzzy evaluation matrix R is established, and the subordinate relationship between X and V is characterized using R. The elements in the matrix are represented as the evaluation level values ​​of the evaluation index items, and the elements X in X are... n The corresponding rating level V m The hierarchical relationship between them is characterized by the element Rnm. The expression for the fuzzy evaluation matrix is:

[0126]

[0127] In the expression for the fuzzy evaluation matrix, R represents the fuzzy evaluation matrix. nm This represents the subordinate element relationship between the evaluation levels corresponding to the elements in the X domain. The elements in the X domain refer to the perforation parameters, and the evaluated object refers to the perforation well productivity.

[0128] Based on the evaluation levels of each perforation parameter and the fuzzy evaluation matrix, the correlation between perforation parameters and perforated well productivity is obtained. For example, this can be achieved through the following steps: Based on the fuzzy evaluation matrix, the fuzzy comprehensive evaluation result vector at the evaluation index level is obtained through the third expression, where the third expression is:

[0129] B = W * R,

[0130] In the third expression, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, W represents the fuzzy subset of the membership relationship between each element in the X domain and the evaluated object, and R represents the fuzzy evaluation matrix.

[0131] Then, based on the fuzzy comprehensive evaluation result vector at the evaluation index level and the evaluation level of each perforation parameter, the correlation between perforation parameters and perforation well productivity is obtained through the fourth expression;

[0132] The fourth expression is:

[0133] N = BE T ,

[0134] In the fourth expression, N represents the correlation between perforation parameters and perforated well productivity, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, and E represents the score of the evaluation level of each perforation parameter.

[0135] After calculating the correlation between perforation parameters and perforated well productivity, step S130 can filter the perforation parameter sample data and perforated well productivity sample data according to actual needs. That is, filtering the perforation parameter sample data and perforated well productivity sample data for secondary preprocessing can greatly improve the prediction performance of the subsequent training of the perforated well completion productivity prediction model, while reducing training costs and improving training efficiency.

[0136] In one embodiment, reference is made to... Figure 4 Step S130 can be achieved through the following steps:

[0137] The selected perforation parameter sample data with high correlation to well productivity are used as the feature dataset X, where X = {X_hole depth, X_hole density, X_hole diameter, X_phase angle, ...}, and the corresponding perforated well productivity sample data are used as the label dataset Y. Using the feature dataset and the label dataset, a perforation completion productivity prediction model based on a convolutional neural network is trained. For example, 1000 sets of X and Y data can be obtained, and the feature dataset and its corresponding label dataset can be divided into three parts: 70% of the data is used to train the model, 20% of the data is used to validate the model training effect, and 10% of the data is used to test the model. Then, using the divided 70% training set, 20% validation set, and 10% test set, a CNN (convolutional neural network)-based perforation completion productivity prediction model for double-casing perforated completion is established. The established CNN includes:

[0138] There is one input layer, which converts the input influencing factor data (i.e., X, perforation parameter sample data) into data in n×1×9 format, where n represents the number of samples in the training set and the validation set, and 9 is the feature dimension;

[0139] There is one output layer, in which there is only one neuron, representing the target variable (perforated well productivity sample data) of the output;

[0140] There are 3 convolutional layers and 3 pooling layers, with the convolutional and pooling layers alternating. Each convolutional layer has a 3x3 kernel and uses max pooling. Finally, there are two fully connected layers, and the number of neurons in each layer can be set as needed.

[0141] Furthermore, step S130 can also utilize the validation dataset and the root mean square error method to obtain the model accuracy for each training of the perforation completion production capacity prediction model; when the model accuracy meets the pre-approval requirements, the training of the perforation completion production capacity prediction model is deemed complete. The expression for the root mean square error method is as follows:

[0142]

[0143] In the expression for the root mean square error method, RMSE represents the model accuracy each time the perforation completion production capacity prediction model is trained, Y represents the actual perforation completion production capacity, Y' represents the predicted perforation completion production capacity, and K represents the number of samples in the validation data sample set used to validate the perforation completion production capacity prediction model.

[0144] The perforation completion productivity prediction model obtained from step S130 can accurately predict the perforation completion productivity of double-casing based on different combinations of perforation parameters. This is helpful for the research on double-casing perforation completion operations and improves the efficiency of double-casing perforation completion operations.

[0145] In one embodiment, reference is made to... Figure 5 Step S140 can be achieved through the following steps:

[0146] Based on the perforation completion productivity model obtained through training, a multi-dimensional relationship graph of perforation parameters is constructed. This multi-dimensional relationship graph includes a first relationship graph G related to productivity. c The second relationship graph G related to casing strength q And the third relationship graph G related to costs m For example, by using the original graph attention network, features are extracted from the first, second, and third relationship graphs to obtain the production capacity relationship feature F. c Characteristics of the relationship between casing strength and F q and cost relationship characteristics F m F c ,F q ,F m =GAT(G c G q G m GAT() represents the original graph attention network, which contains 3 graph convolutional layers, has a node embedding dimension of 64, and uses ReLU as the activation function.

[0147] The capacity relationship features, casing strength relationship features, and cost relationship features are then concatenated into a single feature vector. A linear mapping layer is then used to fuse these features, resulting in the fused relationship features. The expression for the fused relationship features is as follows:

[0148] F = Linear(Concat(F) c ,F q ,F m )),

[0149] In the expression of the merged relational features, F represents the merged relational features. c F represents the characteristics of the production capacity relationship. q F represents the characteristic of the casing strength relationship. m The `concat()` function represents the cost relationship feature, `linear()` represents the vector concatenation operator, and `linear()` represents the linear mapping.

[0150] Then, the true relation feature annotations are obtained, and combined with the fused relation features, an optimized graph attention network is obtained through a preset optimizer. For example, the true relation feature annotations (a 64-dimensional feature vector) are obtained, combined with the fused relation features, and an optimized graph attention network is obtained through a preset optimizer using the mean squared error loss method. The expression for the mean squared error loss method is:

[0151] Loss(θ) = MSE(F, F gt ),

[0152] In the expression for the mean squared error loss method, θ represents the original graph attention network, and F... gt The true relational feature is represented by F, the fused relational feature is represented by F, MSE() represents the mean squared error loss function, the Adam optimizer with a preset learning rate of 0.001 is used, and the loss function is minimized based on the gradient descent backpropagation algorithm to optimize the parameters of the original graph attention network.

[0153] Based on the fused relational features and the optimized graph attention network, a policy network for the perforation parameters of the double-layer casing is constructed. This policy network includes state factors representing states, action factors representing actions, and reward factors representing rewards. For example, the fused relational features are used as state factors representing states; different choices of perforation parameters are divided into multiple discrete action categories; and based on each state and its corresponding state features, an action is determined from these discrete action categories and executed as the current action. For the current action, a corresponding reward value is obtained based on a preset reward value calculation model.

[0154] Based on the fused relational features, state factors, action factors, and reward factors, the policy network parameters are updated using policy gradients to optimize the decision-making performance of the policy network, resulting in an optimized policy network for optimizing perforation parameters used in double-casing perforation completion. For example, the policy network and valuation network are first initialized using a multilayer perceptron, with the fused relational features as input and the probability distribution vector and state value as outputs, as shown in the following expression:

[0155] P = Policy(F), V = Value(F),

[0156] In the formula, Policy(F) represents the policy network, and V = Value(F) represents the value network. The policy network is a multilayer perceptron containing two neural network layers with 64 and n neurons respectively, where n is the number of decisions for the aperture parameter. The value network is also a multilayer perceptron, containing two neural network layers with 64 and 1 neurons respectively. P is a probability distribution vector of dimension n, and V is the state value of dimension 1.

[0157] Based on the action probability distribution vector P, an action policy is obtained and executed, and the corresponding reward value is calculated. The policy gradient is then calculated based on the reward value and the estimated value from the evaluation network, where the expression for the policy gradient is:

[0158]

[0159] In the expression for the policy gradient, φ represents the policy network parameters, T represents the total number of iterations, t represents the current round, and R... t V represents the reward value in the current round, which is calculated based on the action probability distribution vector. t This represents the state value in the current round.

[0160] The gradient ascent algorithm is employed to maximize the policy gradient value and update the policy network parameters. This process is iterated repeatedly. An optimized graph attention network is used to extract the relational features of the current state before each determined action, and then selects the corresponding action based on these features. This optimizes the decision-making performance of the policy network, adaptively forming the optimized policy network. This approach enables enterprises to quickly extract appropriate perforation parameters from massive amounts of data, determining perforation parameters that balance capacity, casing strength, and cost.

[0161] Based on perforation parameter sample data, corresponding perforated well productivity sample data are obtained through fluid simulation software. Then, the correlation between perforation parameters and perforated well productivity is obtained using the fuzzy comprehensive evaluation method. Based on the correlation, the perforation parameter sample data and perforated well productivity sample data are screened, and the perforation parameter sample data and perforated well productivity sample data with high correlation are used to train the perforation completion productivity prediction model. Then, an optimization strategy network for perforation parameters is constructed to optimize the perforation parameters used for double-casing perforation completion. This can help operators quickly and accurately obtain the best combination of double-casing perforation parameters and improve the efficiency of double-casing perforation completion.

[0162] The following describes the neural network-based double-layer casing perforation parameter optimization system provided by the present invention. The neural network-based double-layer casing perforation parameter optimization system described below can be referred to in correspondence with the neural network-based double-layer casing perforation parameter optimization method described above.

[0163] Reference Figure 6 The present invention provides a neural network-based double-layer casing perforation parameter optimization system, which may include:

[0164] The sample data acquisition module is used to obtain perforation parameter sample data and perforated well productivity sample data.

[0165] The correlation calculation module is used to: obtain the correlation between perforation parameters and perforation well productivity based on perforation parameter sample data and perforation well productivity sample data through fuzzy comprehensive evaluation method;

[0166] The perforation completion production capacity prediction model training module is used to: filter perforation parameter sample data and perforation production capacity sample data according to the correlation between perforation parameters and perforation production capacity, use the filtered perforation parameter sample data as input variables, use the perforation production capacity sample data corresponding to the filtered perforation parameter sample data as target variables, and use a convolutional neural network to train the perforation completion production capacity prediction model.

[0167] The double-casing perforation parameter optimization module is used to: construct an optimization strategy network for perforation parameters through a perforation completion productivity prediction model, so as to optimize the perforation parameters used for double-casing perforation completion.

[0168] According to the present invention, a neural network-based double-layer casing perforation parameter optimization system is provided, wherein the sample data acquisition module may include:

[0169] The first data acquisition submodule is used to: obtain perforation parameter sample data;

[0170] The second data acquisition submodule is used to: perform well breaking simulation based on perforation parameter sample data to obtain production ratio sample data;

[0171] The third data acquisition submodule is used to: obtain open-hole well production sample data based on the production ratio sample data and through the first expression;

[0172] The fourth data acquisition submodule is used to: obtain perforated well production sample data based on the production ratio sample data and the open hole well production sample data through the second expression;

[0173] The data preprocessing submodule is used to preprocess perforation parameter sample data and perforation well productivity sample data, wherein the preprocessing includes dimensionless processing and / or outlier processing.

[0174] The first expression is:

[0175]

[0176] In the first expression, Q0 represents the sample data of open-hole well productivity, m 3 / d;K h Indicates penetration rate, um 2 ;P wf P represents the flow pressure, in MPa; e Indicates the bottom hole static pressure, MPa; u g Indicates crude oil viscosity, mP a ·s; T represents temperature, °C; r eh Indicates the oil drain radius, in meters (m); r wh Represents the wellbore radius, in meters (m).

[0177] The second expression is:

[0178] PRI = Q p / Q o ,

[0179] In the second expression, PRI represents the capacity ratio sample data, and Q P Q0 represents the production capacity sample data of perforated wells, while Q0 represents the production capacity sample data of open-hole wells.

[0180] According to the present invention, a neural network-based double-layer casing perforation parameter optimization system includes a correlation calculation module that may include:

[0181] The rating determination submodule is used to: determine the evaluation level of each perforation parameter based on the sample data of the perforation parameters;

[0182] The vector determination submodule is used to: determine the fuzzy weight vector of each perforation parameter based on the perforation parameter sample data and the perforation well productivity sample data;

[0183] The matrix construction submodule is used to construct a fuzzy evaluation matrix based on the fuzzy weight vector of each perforation parameter.

[0184] The correlation calculation submodule is used to: obtain the correlation between perforation parameters and perforated well productivity based on the evaluation level and fuzzy evaluation matrix of each perforation parameter;

[0185] The expression for the fuzzy evaluation matrix is ​​as follows:

[0186]

[0187] In the expression for the fuzzy evaluation matrix, R represents the fuzzy evaluation matrix. nm This represents the subordinate element relationship between the evaluation levels corresponding to the elements in the X domain. The elements in the X domain refer to the perforation parameters, and the evaluated object refers to the perforation well productivity.

[0188] Furthermore, the process of obtaining the correlation between perforation parameters and perforated well productivity based on the evaluation level and fuzzy evaluation matrix of each perforation parameter includes:

[0189] Based on the fuzzy evaluation matrix, the fuzzy comprehensive evaluation result vector at the evaluation index level is obtained through the third expression;

[0190] Based on the fuzzy comprehensive evaluation result vector at the evaluation index level and the evaluation level of each perforation parameter, the correlation between perforation parameters and perforation well productivity is obtained through the fourth expression.

[0191] The third expression is:

[0192] B = W * R,

[0193] In the third expression, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, W represents the fuzzy subset of the membership relationship between each element in the X domain and the evaluated object, and R represents the fuzzy evaluation matrix.

[0194] The fourth expression is:

[0195] N = BE T ,

[0196] In the fourth expression, N represents the correlation between perforation parameters and perforated well productivity, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, and E represents the score of the evaluation level of each perforation parameter.

[0197] According to the present invention, a neural network-based double-casing perforation parameter optimization system is provided, wherein the perforation completion productivity prediction model training module may include:

[0198] The accuracy determination submodule is used to: obtain the model accuracy of the perforation completion production capacity prediction model for each training session using the root mean square error method; and determine that the perforation completion production capacity prediction model training is complete when the model accuracy meets the pre-approval requirements.

[0199] The expression for the root mean square error method is:

[0200]

[0201] In the expression for the root mean square error method, RMSE represents the model accuracy each time the perforation completion production capacity prediction model is trained, Y represents the actual perforation completion production capacity, Y' represents the predicted perforation completion production capacity, and K represents the number of samples in the validation data sample set used to validate the perforation completion production capacity prediction model.

[0202] According to the present invention, a neural network-based double-layer casing perforation parameter optimization system is provided, wherein the double-layer casing perforation parameter optimization module may include:

[0203] The multi-dimensional relationship graph construction submodule is used to: construct a multi-dimensional relationship graph of perforation parameters based on the perforation completion productivity model obtained through training. The multi-dimensional relationship graph includes a first relationship graph related to productivity, a second relationship graph related to casing strength, and a third relationship graph related to cost.

[0204] The feature processing submodule is used to: extract, concatenate, and fuse features from the multi-dimensional relationship graph of the perforation parameters using the original graph attention network to obtain the fused relationship features;

[0205] The graph attention network optimization submodule is used to: obtain real relation feature annotations, combine them with the fused relation features, and obtain an optimized graph attention network through a preset optimizer;

[0206] The strategy network construction submodule is used to: construct a strategy network for the perforation parameters of the double-layer casing based on the fused relational features and the optimized graph attention network. The strategy network includes a state factor for representing the state, an action factor for representing the action, and a reward factor for representing the reward.

[0207] The strategy network optimization submodule is used to: update the strategy network parameters using strategy gradient based on the fused relational features, state factors, action factors, and reward factors, optimize the decision performance of the strategy network, and obtain an optimized strategy network to optimize the perforation parameters used for double-casing perforation completion.

[0208] According to the present invention, a neural network-based double-layer casing perforation parameter optimization system may include a feature processing submodule comprising:

[0209] The feature extraction submodule is used to: extract features from the first relationship graph, the second relationship graph, and the third relationship graph through the original graph attention network to obtain production capacity relationship features, casing strength relationship features, and cost relationship features;

[0210] The feature splicing submodule is used to splice the features of production capacity relationship, casing strength relationship, and cost relationship to obtain a feature vector;

[0211] The feature fusion submodule is used to: fuse the capacity relationship features, casing strength relationship features, and cost relationship features based on feature vectors through linear mapping to obtain the fused relationship features;

[0212] The expression for the fused relational features is as follows:

[0213] F = Linear(Concat(F) c ,F q ,F m )),

[0214] In the expression of the merged relational features, F represents the merged relational features. c F represents the characteristics of the production capacity relationship. q F represents the characteristic of the casing strength relationship. m The `concat()` function represents the cost relationship characteristics, and `linear()` represents the vector concatenation operator.

[0215] Furthermore, the graph attention network optimization submodule is specifically used for:

[0216] Obtain the true relation feature annotations, combine them with the fused relation features, and use the mean squared error loss method through a preset optimizer to obtain an optimized graph attention network;

[0217] The expression for the mean squared error loss method is as follows:

[0218] Loss(θ) = MSE(F, F gt ),

[0219] In the expression for the mean squared error loss method, θ represents the original graph attention network, and F... gt The true relational feature is represented by F, the fused relational feature is represented by F, and the mean squared error loss function is represented by MSE().

[0220] According to the present invention, a neural network-based double-layer casing perforation parameter optimization system may include a strategy network optimization submodule comprising:

[0221] The initialization submodule is used to: initialize the policy network and the valuation network, take the fused relation features as input, and output and obtain the action probability distribution vector (action factor) and state value (state factor) respectively, as shown in the following expressions:

[0222] P = Policy(F), V = Value(F)

[0223] In the formula, Policy(F) and V = Value(F) are the policy network and the value network, respectively, and F represents the network input;

[0224] Obtain the action probability distribution vector and state value;

[0225] The policy gradient calculation submodule is used to: obtain the policy gradient based on the action probability distribution vector and the state value;

[0226] The policy gradient maximization submodule is used to: maximize the policy gradient using the gradient ascent method and update the policy network parameters to optimize the decision performance of the policy network and adaptively form an optimized policy network.

[0227] The expression for the policy gradient is:

[0228]

[0229] In the expression for the policy gradient, φ represents the policy network parameters, T represents the total number of iterations, t represents the current round, and R... t V represents the reward value in the current round, which is calculated based on the action probability distribution vector. t This represents the state value in the current round.

[0230] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a neural network-based double-sleeve perforation parameter optimization method, which includes:

[0231] Obtain perforation parameter sample data and perforated well productivity sample data;

[0232] Based on the perforation parameter sample data and perforated well production capacity sample data, the correlation between perforation parameters and perforated well production capacity is obtained through fuzzy comprehensive evaluation method.

[0233] Based on the correlation between perforation parameters and perforation well productivity, perforation parameter sample data and perforation well productivity sample data are screened. The screened perforation parameter sample data is used as the input variable, and the perforation well productivity sample data corresponding to the screened perforation parameter sample data is used as the target variable. A convolutional neural network is used to train a perforation completion productivity prediction model.

[0234] By using a perforation completion productivity prediction model, an optimization strategy network for perforation parameters is constructed to optimize the perforation parameters used in double-casing perforation completion.

[0235] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0236] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the neural network-based double-casing perforation parameter optimization method provided by the above methods, the method comprising:

[0237] Obtain perforation parameter sample data and perforated well productivity sample data;

[0238] Based on the perforation parameter sample data and perforated well production capacity sample data, the correlation between perforation parameters and perforated well production capacity is obtained through fuzzy comprehensive evaluation method.

[0239] Based on the correlation between perforation parameters and perforation well productivity, perforation parameter sample data and perforation well productivity sample data are screened. The screened perforation parameter sample data is used as the input variable, and the perforation well productivity sample data corresponding to the screened perforation parameter sample data is used as the target variable. A convolutional neural network is used to train a perforation completion productivity prediction model.

[0240] By using a perforation completion productivity prediction model, an optimization strategy network for perforation parameters is constructed to optimize the perforation parameters used in double-casing perforation completion.

[0241] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the neural network-based double-casing perforation parameter optimization method provided by the above methods, the method comprising:

[0242] Obtain perforation parameter sample data and perforated well productivity sample data;

[0243] Based on the perforation parameter sample data and perforated well production capacity sample data, the correlation between perforation parameters and perforated well production capacity is obtained through fuzzy comprehensive evaluation method.

[0244] Based on the correlation between perforation parameters and perforation well productivity, perforation parameter sample data and perforation well productivity sample data are screened. The screened perforation parameter sample data is used as the input variable, and the perforation well productivity sample data corresponding to the screened perforation parameter sample data is used as the target variable. A convolutional neural network is used to train a perforation completion productivity prediction model.

[0245] By using a perforation completion productivity prediction model, an optimization strategy network for perforation parameters is constructed to optimize the perforation parameters used in double-casing perforation completion.

[0246] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0247] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0248] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing perforation parameters of a double-layer casing based on a neural network, characterized in that, include: Obtain perforation parameter sample data and perforated well productivity sample data; Based on the perforation parameter sample data and perforated well production capacity sample data, the correlation between perforation parameters and perforated well production capacity is obtained through fuzzy comprehensive evaluation method. Based on the correlation between perforation parameters and perforation well productivity, perforation parameter sample data and perforation well productivity sample data are screened. The screened perforation parameter sample data is used as the input variable, and the perforation well productivity sample data corresponding to the screened perforation parameter sample data is used as the target variable. A convolutional neural network is used to train a perforation completion productivity prediction model. Based on the perforation completion productivity model obtained through training, a multi-dimensional relationship diagram of perforation parameters is constructed. The multi-dimensional relationship diagram includes a first relationship diagram related to productivity, a second relationship diagram related to casing strength, and a third relationship diagram related to cost. By using the original graph attention network, feature extraction, feature concatenation, and feature fusion are performed on the multi-dimensional relationship graph of the perforation parameters to obtain the fused relationship features. Obtain real relation feature annotations, combine them with the fused relation features, and obtain an optimized graph attention network through a preset optimizer; Based on the fused relational features and the optimized graph attention network, a strategy network for the perforation parameters of the double-layer casing is constructed. The strategy network includes a state factor for representing the state, an action factor for representing the action, and a reward factor for representing the reward. Based on the fused relational characteristics, state factors, action factors, and reward factors, the strategy network parameters are updated using a strategy gradient to optimize the decision-making performance of the strategy network, resulting in an optimized strategy network for optimizing perforation parameters used in double-casing perforation completion.

2. The method for optimizing perforation parameters of a double-layer casing based on a neural network according to claim 1, characterized in that, The obtained perforation parameter sample data and perforated well productivity sample data include: Obtain perforation parameter sample data; Numerical simulations of well breaking were performed based on perforation parameter sample data to obtain production capacity ratio sample data; Based on the production ratio sample data, the open-hole well production sample data is obtained through the first expression; Based on the production ratio sample data and the open hole well production sample data, the perforated well production sample data is obtained through the second expression. Preprocessing is performed on perforation parameter sample data and perforation well productivity sample data, including dimensionless processing and / or outlier processing. The first expression is: ， In the first expression, Q0 represents the sample data of open-hole well productivity, and K h P represents penetration rate. wf P represents the flow pressure. e Indicates the bottom static pressure, u g The value represents the viscosity of the crude oil, T represents the temperature, and r represents the temperature. eh Indicates the oil drain radius, r wh Indicates the wellbore radius; The second expression is: ， In the second expression, PRI represents the capacity ratio sample data, and Q P Q0 represents the production capacity sample data of perforated wells, while Q0 represents the production capacity sample data of open-hole wells.

3. The method for optimizing perforation parameters of a double-layer casing based on a neural network according to claim 2, characterized in that, The correlation between perforation parameters and perforated well productivity is obtained using a fuzzy comprehensive evaluation method based on perforation parameter sample data and perforated well productivity sample data, including: Based on the perforation parameter sample data, determine the evaluation level of each perforation parameter; Based on the perforation parameter sample data and the perforation well production capacity sample data, determine the fuzzy weight vector of each perforation parameter; A fuzzy evaluation matrix is ​​constructed based on the fuzzy weight vectors of each perforation parameter; Based on the evaluation level of each perforation parameter and the fuzzy evaluation matrix, the correlation between perforation parameters and perforated well productivity is obtained. The expression for the fuzzy evaluation matrix is ​​as follows: ， In the expression for the fuzzy evaluation matrix, R represents the fuzzy evaluation matrix. nm This represents the subordinate element relationship between the evaluation levels corresponding to the elements in the X domain. The elements in the X domain refer to the perforation parameters, and the evaluated object refers to the perforation well productivity. Furthermore, the process of obtaining the correlation between perforation parameters and perforated well productivity based on the evaluation level and fuzzy evaluation matrix of each perforation parameter includes: Based on the fuzzy evaluation matrix, the fuzzy comprehensive evaluation result vector at the evaluation index level is obtained through the third expression; Based on the fuzzy comprehensive evaluation result vector at the evaluation index level and the evaluation level of each perforation parameter, the correlation between perforation parameters and perforation well productivity is obtained through the fourth expression. The third expression is: ， In the third expression, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, W represents the fuzzy subset of the membership relationship between each element in the X domain and the evaluated object, and R represents the fuzzy evaluation matrix. The fourth expression is: ， In the fourth expression, N represents the correlation between perforation parameters and perforated well productivity, B represents the fuzzy comprehensive evaluation result vector at the evaluation index level, and E represents the score of the evaluation level of each perforation parameter.

4. The method for optimizing perforation parameters of a double-layer casing based on a neural network according to claim 3, characterized in that, The process involves using the selected perforation parameter sample data as input variables, the perforated well productivity sample data corresponding to the selected perforation parameter sample data as target variables, and training a perforated well completion productivity prediction model using a convolutional neural network, including: The root mean square error method is used to obtain the model accuracy for each training of the perforation completion production capacity prediction model; When the model accuracy meets the pre-approval requirements, the perforation completion production capacity prediction model training is deemed complete. The expression for the root mean square error method is: ， In the expression for the root mean square error method, RMSE represents the model accuracy each time the perforation completion production capacity prediction model is trained, Y represents the actual perforation completion production capacity, Y' represents the predicted perforation completion production capacity, and K represents the number of samples in the validation data sample set used to validate the perforation completion production capacity prediction model.

5. The method for optimizing perforation parameters of a double-layer casing based on a neural network according to claim 1, characterized in that, The process involves using the original graph attention network to extract, concatenate, and fuse features from the multi-dimensional relationship graph of the perforation parameters, resulting in fused relationship features, including: By using the original graph attention network, features are extracted from the first, second, and third relationship graphs to obtain production capacity relationship features, casing strength relationship features, and cost relationship features. The features of capacity relationship, casing strength relationship, and cost relationship are concatenated to obtain a feature vector; Based on feature vectors, feature fusion is performed on capacity relationship features, casing strength relationship features, and cost relationship features through linear mapping to obtain fused relationship features; The expression for the fused relational features is as follows: ， In the expression of the merged relational features, F represents the merged relational features. c F represents the characteristics of the production capacity relationship. q F represents the characteristic of the casing strength relationship. m The functions represent cost relationship characteristics, concat() represents the vector concatenation operator, and linear() represents the linear mapping. Furthermore, the process of obtaining real relation feature annotations, combining them with the fused relation features, and then using a preset optimizer to obtain an optimized graph attention network specifically involves: Obtain the true relation feature annotations, combine them with the fused relation features, and use the mean squared error loss method through a preset optimizer to obtain an optimized graph attention network; The expression for the mean squared error loss method is as follows: ， In the expression for the mean squared error loss method, F represents the original graph attention network. gt The label represents the true relational features, F represents the fused relational features, and MSE() represents the mean squared error loss function.

6. The method for optimizing perforation parameters of a double-layer casing based on a neural network according to claim 5, characterized in that, The process of updating the policy network parameters using policy gradient based on the fused relational features, state factors, action factors, and reward factors to optimize the decision-making performance of the policy network and obtain an optimized policy network includes: Initialize the policy network and the valuation network to obtain the action probability distribution vector and the state value; The policy gradient is obtained based on the action probability distribution vector and the state value; By maximizing the policy gradient using the gradient ascent method and updating the policy network parameters, the decision performance of the policy network is optimized, and an optimized policy network is adaptively formed. The expression for the policy gradient is: ， In the expression for the policy gradient, Let R represent the policy network parameters, T represent the total number of iterations, t represent the current round, and R represent the network parameters. t V represents the reward value in the current round, which is calculated based on the action probability distribution vector. t This represents the state value in the current round.

7. A neural network-based system for optimizing perforation parameters in a double-layer casing, characterized in that, include: The sample data acquisition module is used to obtain perforation parameter sample data and perforated well productivity sample data. The correlation calculation module is used to: obtain the correlation between perforation parameters and perforation well productivity based on perforation parameter sample data and perforation well productivity sample data through fuzzy comprehensive evaluation method; The perforation completion production capacity prediction model training module is used to: filter perforation parameter sample data and perforation production capacity sample data according to the correlation between perforation parameters and perforation production capacity, use the filtered perforation parameter sample data as input variables, use the perforation production capacity sample data corresponding to the filtered perforation parameter sample data as target variables, and use a convolutional neural network to train the perforation completion production capacity prediction model. The double-casing perforation parameter optimization module is used to: construct an optimization strategy network for perforation parameters through a perforation completion productivity prediction model, so as to optimize the perforation parameters used for double-casing perforation completion; The step of constructing an optimization strategy network for perforation parameters using a perforation completion productivity prediction model to optimize perforation parameters for double-casing perforation completion includes: Based on the perforation completion productivity model obtained through training, a multi-dimensional relationship diagram of perforation parameters is constructed. The multi-dimensional relationship diagram includes a first relationship diagram related to productivity, a second relationship diagram related to casing strength, and a third relationship diagram related to cost. By using the original graph attention network, feature extraction, feature concatenation, and feature fusion are performed on the multi-dimensional relationship graph of the perforation parameters to obtain the fused relationship features. Obtain real relation feature annotations, combine them with the fused relation features, and obtain an optimized graph attention network through a preset optimizer; Based on the fused relational features and the optimized graph attention network, a strategy network for the perforation parameters of the double-layer casing is constructed. The strategy network includes a state factor for representing the state, an action factor for representing the action, and a reward factor for representing the reward. Based on the fused relational characteristics, state factors, action factors, and reward factors, the strategy network parameters are updated using a strategy gradient to optimize the decision-making performance of the strategy network, resulting in an optimized strategy network for optimizing perforation parameters used in double-casing perforation completion.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the neural network-based double-casing perforation parameter optimization method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the neural network-based double-casing perforation parameter optimization method as described in any one of claims 1 to 6.

Images