Inter-well dynamic connectivity prediction method and system based on embedded graph neural network

Abstract

The present invention discloses a method and system for predicting dynamic connectivity between wells based on an embedded graph neural network, which relates to the field of oil reservoir technology, including predicting the liquid production of each production well in the target multi-layer well network at the next moment based on the production data of the target multi-layer well network at the current moment and the production well liquid production prediction model; wherein the determination process of the production well liquid production prediction model is: determining the sample production data, sample geometry data and sample formation data of the target multi-layer well network; constructing an embedded graph neural network model; the graph neural network model in the embedded graph neural network model is constructed based on the sample geometry data; and determining the production well liquid production prediction model based on the sample production data, sample formation data and the embedded graph neural network model. The present invention can accurately predict the flow capacity of the fluid in the multi-layer well network, and then accurately evaluate the dynamic connectivity between wells of the multi-layer well network.

Description

Technical Field

[0001] The present invention relates to the field of oil reservoir technology, and in particular to a method and system for predicting dynamic connectivity between wells based on an embedded graph neural network. Background Art

[0002] The oil reservoir is a relatively complex dynamic system. When the water injection volume of the water injection well is different, it will cause the production of the production well to fluctuate. This operation can reflect the connectivity characteristics of the water injection well and the production well in each layer. The better the connectivity, the more consistent the fluctuation of the production of the production well in each layer and the change of the water injection volume in each layer of the water injection well. Inter-well connectivity reflects the ability of the fluid to flow between wells and is an important parameter for reservoir engineering evaluation.

[0003] At present, the models for identifying the flow ability of fluids between wells are all single-layer models, which cannot identify the flow ability of fluids in multi-layer well networks. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for predicting the dynamic connectivity between wells based on an embedded graph neural network, which can accurately predict the flow capacity of fluids in a multi-layer well network, and then accurately evaluate the dynamic connectivity between wells in a multi-layer well network.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for predicting dynamic connectivity between wells based on an embedded graph neural network, comprising:

[0007] Obtain production data of the target multi-layer well network at the current moment;

[0008] Based on the production data of the target multi-layer well network at the current moment and the prediction model of the liquid production of the production wells, predict the liquid production of each production well in the target multi-layer well network at the next moment;

[0009] The determination process of the production well liquid production prediction model is as follows:

[0010] Determine sample data of the target multi-layer well pattern; the sample data includes sample production data, sample geometry data and sample formation data; the sample production data includes sample water injection volume of water injection wells and sample liquid production volume of production wells;

[0011] Constructing an embedded graph neural network model; the embedded graph neural network model includes an embedded neural network model and a graph neural network model connected to an output end of the embedded neural network model; wherein the graph neural network model is constructed based on the sample geometry data;

[0012] A production well fluid production prediction model is determined based on the sample production data, the sample formation data and the embedded graph neural network model.

[0013] Optionally, the determining of the sample data of the target multi-layer well pattern specifically includes:

[0014] Acquire sample geometric data of the target multi-layer well network; the sample geometric data includes the number of water injection wells, the number of production wells, the distance between each production well and each water injection well, the distance between each production well and each production well, the distance between each water injection well and each water injection well, the relative position between each production well and each water injection well, the relative position between each production well and each production well, and the relative position between each water injection well and each water injection well;

[0015] Acquire sample production data of the target multi-layer well pattern; the sample production data includes the liquid production of each production well at different times and the water injection volume of each water injection well at different times;

[0016] Obtain sample formation data of the target multi-layer well network; the sample formation data include sample geological data of each layer of each production well and sample geological data of each layer of each injection well; the geological data include perforation parameters, layer depth, effective thickness, permeability and porosity.

[0017] Optionally, the building of the embedded graph neural network model specifically includes:

[0018] According to the sample formation data, the sample production data is divided to obtain the liquid production of each layer of each production well at different times and the water injection volume of each layer of each water injection well at different times;

[0019] Constructing an embedded neural network model; the input end of the embedded neural network model is used to input sample geological data of each layer of each production well and sample geological data of each layer of each water injection well; the output end of the embedded neural network model is used to output the predicted liquid production of each layer of each production well at different times and the predicted water injection volume of each layer of each water injection well at different times;

[0020] Based on the sample geometric data of the target multi-layer well network, a graph neural network model is established, which uses different nodes to represent each layer of each production well and each layer of each water injection well; wherein, when the node represents a production well and a layer, the feature of the node is the liquid production of the production well and the layer, and when the node represents a water injection well and a layer, the feature of the node is the water injection volume of the water injection well and the layer; the edges of the graph neural network model are used for the distance and relative position between a well point and other well points; a well point is a production well or a water injection well, and the other well points are production wells or water injection wells;

[0021] The input end of the graph neural network model is used to input the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment; the output end of the graph neural network model is used to output the predicted liquid production of each layer of each production well at the second moment; the first moment is the previous moment of the second moment.

[0022] Optionally, determining a production well liquid production prediction model based on the sample production data, the sample formation data and the embedded graph neural network model specifically includes:

[0023] Based on the predicted liquid production of each layer of each production well at different times, the predicted water injection of each layer of each water injection well at different times, the liquid production of each layer of each production well at different times, and the water injection of each layer of each water injection well at different times, the model parameters of the embedded neural network model are adjusted to obtain a trained embedded neural network model;

[0024] Based on the sample formation data and the trained embedded neural network model, the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment are obtained; the first moment is any moment in the sample data;

[0025] Based on the predicted liquid production of each layer of each production well at the first moment, the predicted water injection volume of each layer of each water injection well at the first moment and the graph neural network model, a production well liquid production prediction model is determined.

[0026] A well dynamic connectivity prediction system based on embedded graph neural network, comprising:

[0027] A data acquisition module is used to obtain the production data of the target multi-layer well network at the current moment;

[0028] A liquid production prediction module, used to predict the liquid production of each production well in the target multi-layer well network at the next moment based on the production data of the target multi-layer well network at the current moment and the liquid production prediction model of the production wells;

[0029] The determination process of the production well liquid production prediction model is as follows:

[0030] A sample data determination module is used to determine the sample data of the target multi-layer well pattern; the sample data includes sample production data, sample geometry data and sample formation data; the sample production data includes sample water injection volume of water injection wells and sample liquid production volume of production wells;

[0031] A neural network model construction module, used to construct an embedded graph neural network model; the embedded graph neural network model includes an embedded neural network model and a graph neural network model connected to the output end of the embedded neural network model; wherein the graph neural network model is constructed based on the sample geometry data;

[0032] A production well liquid production prediction model determination module is used to determine the production well liquid production prediction model based on the sample production data, the sample formation data and the embedded graph neural network model.

[0033] Optionally, the sample data determination module specifically includes:

[0034] A first acquisition unit is used to acquire sample geometric data of the target multi-layer well network; the sample geometric data includes the number of water injection wells, the number of production wells, the distance between each production well and each water injection well, the distance between each production well and each production well, the distance between each water injection well and each water injection well, the relative position between each production well and each water injection well, the relative position between each production well and each production well, and the relative position between each water injection well and each water injection well;

[0035] A second acquisition unit is used to acquire sample production data of the target multi-layer well pattern; the sample production data includes the liquid production of each production well at different times and the water injection volume of each water injection well at different times;

[0036] The third acquisition unit is used to acquire sample formation data of the target multi-layer well network; the sample formation data includes sample geological data of each layer of each production well and sample geological data of each layer of each injection well; the geological data includes perforation parameters, layer depth, effective thickness, permeability and porosity.

[0037] Optionally, the neural network model construction module specifically includes:

[0038] A division unit, used to divide the sample production data according to the sample formation data, to obtain the liquid production of each layer of each production well at different times and the water injection of each layer of each water injection well at different times;

[0039] The first construction unit is used to construct an embedded neural network model; the input end of the embedded neural network model is used to input sample geological data of each layer of each production well and sample geological data of each layer of each water injection well; the output end of the embedded neural network model is used to output the predicted liquid production of each layer of each production well at different times and the predicted water injection volume of each layer of each water injection well at different times;

[0040] The second construction unit is used to establish a graph neural network model based on the sample geometric data of the target multi-layer well network, using different nodes to represent each layer of each production well and each layer of each water injection well; wherein, when the node represents a production well and a layer, the feature of the node is the liquid production of the production well and the layer, and when the node represents a water injection well and a layer, the feature of the node is the water injection volume of the water injection well and the layer; the edge of the graph neural network model is used for the distance and relative position between a well point and other well points; a well point is a production well or a water injection well, and the other well points are production wells or water injection wells;

[0041] The input end of the graph neural network model is used to input the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment; the output end of the graph neural network model is used to output the predicted liquid production of each layer of each production well at the second moment; the first moment is the previous moment of the second moment.

[0042] Optionally, the production well liquid production prediction model determination module specifically includes:

[0043] A first training unit is used to adjust the model parameters of the embedded neural network model based on the predicted liquid production of each layer of each production well at different times, the predicted water injection of each layer of each water injection well at different times, the liquid production of each layer of each production well at different times, and the water injection of each layer of each water injection well at different times, so as to obtain a trained embedded neural network model;

[0044] The second training unit is used to:

[0045] Based on the sample formation data and the trained embedded neural network model, the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment are obtained; the first moment is any moment in the sample data;

[0046] Based on the predicted liquid production of each layer of each production well at the first moment, the predicted water injection volume of each layer of each water injection well at the first moment and the graph neural network model, a production well liquid production prediction model is determined.

[0047] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0048] The present invention makes full use of the historical production data and reservoir geological information of the multi-layer well network, and studies the dynamic connectivity of the reservoir through an embedded graph neural network, thereby being able to accurately predict the flow capacity of the fluid in the multi-layer well network and accurately evaluate the dynamic connectivity between wells in the multi-layer well network. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0050] Figure 1 It is a flow chart of the method for predicting dynamic connectivity between wells based on embedded graph neural network of the present invention;

[0051] Figure 2 It is a structural diagram of the well-to-well dynamic connectivity prediction system based on the embedded graph neural network of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] The present invention provides a method and system for predicting dynamic connectivity between wells based on an embedded graph neural network. The purpose is to identify the dynamic connectivity status between different wells in a multi-layer well network by using the geometric information, perforation parameters, geological parameters and historical production data of the well network through machine learning of the embedded graph neural network.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Embodiment 1

[0056] The embodiment of the present invention takes into account the conditions such as incomplete monitoring data and shut-in wells, and based on the material balance principle, makes full use of production data and geological data to establish a multi-layer well-to-well dynamic connectivity model. The embodiment of the present invention provides a method for predicting the dynamic connectivity between wells in the high water cut period based on an embedded graph neural network, which specifically includes:

[0057] Step 1: Obtain diverse information of the target multi-layer well pattern, including:

[0058] Determine the target multi-layer well pattern that needs dynamic connectivity analysis;

[0059] Acquire geometric data of the target multi-layer well network, the geometric data including the number of injection wells, the number of production wells, the distance between each well point and the relative position between each well point;

[0060] Obtain the start time of each production well and each water injection well in the target multi-layer well network, as well as the shut-in operations and shut-in durations performed at different times;

[0061] Acquire production dynamic data of the target multi-layer well network, the production dynamic data including the liquid production of each production well at different times and the water injection volume of each water injection well at different times;

[0062] Obtain the reservoir geological data of each production well and each water injection well in each layer of the target multi-layer well network, and the reservoir geological data includes perforation parameters, layer depth, effective thickness, permeability, porosity, etc.

[0063] Step 2: Before establishing the neural network model, the production dynamic data and reservoir geological data of the target multi-layer well network are preprocessed, including:

[0064] Delete abnormal data;

[0065] Record missing data points;

[0066] Use reasonable time intervals to reduce the impact of missing and abnormal data;

[0067] The production dynamic data of the high water content stage (water content greater than 60%) are selected and standardized. The standardized formula is shown as follows:

[0068] Step 3: Establish an embedded neural network model to obtain production data from each layer, including:

[0069] According to the reservoir geological data of each production well and each water injection well, it is divided into different strata; the division standard is: the oil-bearing strata separated by rocks are divided into strata according to the perforation parameters.

[0070] A neural network model is established based on the reservoir geological data and production dynamic data of different layers of each well. The input is the stratigraphic data of each layer (including perforation parameters, layer depth, effective thickness, permeability, porosity) and production data (including the liquid production or water injection volume of each production well and each water injection well at different times). After 3 to 5 fully connected layers, the output is the production data value of each layer.

[0071] Step 4: Establish a multi-layer graph neural network model based on the diverse information of the target multi-layer well network, including:

[0072] Different nodes are established through the production data and geometric information of each layer of the target multi-layer well network to represent the points of each layer of the production well and the points of each layer of the water injection well;

[0073] xv Represents the liquid production or water injection volume as the node feature of a certain node;

[0074] x co[v] Represent the distance and relative position between different well points as edge attributes;

[0075] x ne[v] Represents the node characteristics of a node's neighbor nodes;

[0076] Take f as the state update parameter that changes with time interval;

[0077] The state update function is: h v t+1 =f(x v ,x co[v] ,x ne[v] );

[0078] The input is the output in step 3, that is, the production data value of each layer, and the output is the production data value of the production well.

[0079] This process includes two processes in graph neural networks.

[0080] 1. From each layer of the injection well to each layer of the production well, the weight values ​​between the layers are obtained based on the attention mechanism through the assigned attributes, and the data of each layer of the injection well is transferred to each layer of the production well through the edge.

[0081] 2. Based on the attributes of each layer section of the production well in step 3, the weight value from the production well to each layer section is obtained through MLP, and the data of each layer section of the production well is transferred to the production well.

[0082] Step 5: Train the graph neural network model to obtain dynamic connectivity, including:

[0083] The input data used for the overall training is the formation data and production data of the current and previous multiple moments (50-100). The output is the production data of the production well at the next moment.

[0084] Eighty percent of the data is used as training data;

[0085] Ten percent of the data is used as test data;

[0086] Ten percent of the data was used as validation data;

[0087] Call the state update parameter f several times until convergence;

[0088] The output parameter at this time is represented by g;

[0089] The true value of the predicted data is represented by t;

[0090] Back propagation by calculating the error Loss = ∑|gt|;

[0091] Update the state update parameter f of the model until the error Loss is close to zero. At this time, the model converges and the training of the embedded neural network and graph neural network model ends;

[0092] Call the state update parameter f about the node feature x v and neighbor node features x ne[v] Parameters;

[0093] The calling parameters are normalized using softmax, and the processed parameters are the weight parameters of the connectivity between wells in the multi-layer system.

[0094] The reservoir, well network and dynamic change parameters are used as node attributes, and the layers of water injection wells and production wells in the well network are used as data structure vertices. The edges of the data structure reflect the connectivity between wells. Considering the influence of space and time on the data, the data is pre-processed to train an embedded graph neural network to obtain dynamic connectivity changes between multi-layer systems. At present, most of the inversion dynamic connectivity models between wells are single-layer models, which cannot meet the needs of inverting the dynamic connectivity between wells in multi-layer oil reservoirs. The advantage of the present invention is that the dynamic changes of multi-layer system well connectivity are obtained based on historical data training.

[0095] Embodiment 2

[0096] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting dynamic connectivity between wells based on an embedded graph neural network, comprising:

[0097] Step 101: Obtain production data of the target multi-layer well network at the current moment.

[0098] Step 102: Based on the production data of the target multi-layer well network at the current moment and the production well liquid production prediction model, predict the liquid production of each production well in the target multi-layer well network at the next moment.

[0099] The determination process of the production well liquid production prediction model is as follows:

[0100] Determine sample data of the target multi-layer well network; the sample data includes sample production data, sample geometry data and sample formation data; the sample production data includes sample water injection volume of injection wells and sample liquid production volume of production wells.

[0101] Constructing an embedded graph neural network model; the embedded graph neural network model includes an embedded neural network model and a graph neural network model connected to the output end of the embedded neural network model; wherein the graph neural network model is constructed based on the sample geometry data.

[0102] A production well fluid production prediction model is determined based on the sample production data, the sample formation data and the embedded graph neural network model.

[0103] Furthermore, the determining of the sample data of the target multi-layer well pattern specifically includes:

[0104] Acquire sample geometric data of the target multi-layer well network; the sample geometric data includes the number of water injection wells, the number of production wells, the distance between each production well and each water injection well, the distance between each production well and each production well, the distance between each water injection well and each water injection well, the relative position between each production well and each water injection well, the relative position between each production well and each production well, and the relative position between each water injection well and each water injection well;

[0105] Acquire sample production data of the target multi-layer well pattern; the sample production data includes the liquid production of each production well at different times and the water injection volume of each water injection well at different times;

[0106] Obtain sample formation data of the target multi-layer well network; the sample formation data include sample geological data of each layer of each production well and sample geological data of each layer of each injection well; the geological data include perforation parameters, layer depth, effective thickness, permeability and porosity.

[0107] The construction of the embedded graph neural network model specifically includes:

[0108] According to the sample formation data, the sample production data is divided to obtain the liquid production of each layer of each production well at different times and the water injection volume of each layer of each water injection well at different times;

[0109] Constructing an embedded neural network model; the input end of the embedded neural network model is used to input sample geological data of each layer of each production well and sample geological data of each layer of each water injection well; the output end of the embedded neural network model is used to output the predicted liquid production of each layer of each production well at different times and the predicted water injection volume of each layer of each water injection well at different times;

[0110] Based on the sample geometric data of the target multi-layer well network, a graph neural network model is established, which uses different nodes to represent each layer of each production well and each layer of each water injection well; wherein, when the node represents a production well and a layer, the feature of the node is the liquid production of the production well and the layer, and when the node represents a water injection well and a layer, the feature of the node is the water injection volume of the water injection well and the layer; the edges of the graph neural network model are used for the distance and relative position between a well point and other well points; a well point is a production well or a water injection well, and the other well points are production wells or water injection wells;

[0111] The input end of the graph neural network model is used to input the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment; the output end of the graph neural network model is used to output the predicted liquid production of each layer of each production well at the second moment; the first moment is the previous moment of the second moment.

[0112] The step of determining a production well liquid production prediction model based on the sample production data, the sample formation data and the embedded graph neural network model specifically includes:

[0113] Based on the predicted liquid production of each layer of each production well at different times, the predicted water injection of each layer of each water injection well at different times, the liquid production of each layer of each production well at different times, and the water injection of each layer of each water injection well at different times, the model parameters of the embedded neural network model are adjusted to obtain a trained embedded neural network model;

[0114] Based on the sample formation data and the trained embedded neural network model, the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment are obtained; the first moment is any moment in the sample data;

[0115] Based on the predicted liquid production of each layer of each production well at the first moment, the predicted water injection volume of each layer of each water injection well at the first moment and the graph neural network model, a production well liquid production prediction model is determined.

[0116] Embodiment 3

[0117] like Figure 2 As shown, this embodiment provides a well-to-well dynamic connectivity prediction system based on an embedded graph neural network, including:

[0118] The data acquisition module 201 is used to acquire the production data of the target multi-layer well pattern at the current moment.

[0119] The liquid production prediction module 202 is used to predict the liquid production of each production well in the target multi-layer well network at the next moment based on the production data of the target multi-layer well network at the current moment and the liquid production prediction model of the production wells.

[0120] The determination process of the production well liquid production prediction model is as follows:

[0121] A sample data determination module is used to determine the sample data of the target multi-layer well pattern; the sample data includes sample production data, sample geometry data and sample formation data; the sample production data includes sample water injection volume of water injection wells and sample liquid production volume of production wells;

[0122] A neural network model construction module, used to construct an embedded graph neural network model; the embedded graph neural network model includes an embedded neural network model and a graph neural network model connected to the output end of the embedded neural network model; wherein the graph neural network model is constructed based on the sample geometry data;

[0123] A production well liquid production prediction model determination module is used to determine the production well liquid production prediction model based on the sample production data, the sample formation data and the embedded graph neural network model.

[0124] Furthermore, the sample data determination module specifically includes:

[0125] A first acquisition unit is used to acquire sample geometric data of the target multi-layer well network; the sample geometric data includes the number of water injection wells, the number of production wells, the distance between each production well and each water injection well, the distance between each production well and each production well, the distance between each water injection well and each water injection well, the relative position between each production well and each water injection well, the relative position between each production well and each production well, and the relative position between each water injection well and each water injection well;

[0126] A second acquisition unit is used to acquire sample production data of the target multi-layer well pattern; the sample production data includes the liquid production of each production well at different times and the water injection volume of each water injection well at different times;

[0127] The third acquisition unit is used to acquire sample formation data of the target multi-layer well network; the sample formation data includes sample geological data of each layer of each production well and sample geological data of each layer of each injection well; the geological data includes perforation parameters, layer depth, effective thickness, permeability and porosity.

[0128] The neural network model building module specifically includes:

[0129] A division unit, used to divide the sample production data according to the sample formation data, to obtain the liquid production of each layer of each production well at different times and the water injection of each layer of each water injection well at different times;

[0130] The first construction unit is used to construct an embedded neural network model; the input end of the embedded neural network model is used to input sample geological data of each layer of each production well and sample geological data of each layer of each water injection well; the output end of the embedded neural network model is used to output the predicted liquid production of each layer of each production well at different times and the predicted water injection volume of each layer of each water injection well at different times;

[0131] The second construction unit is used to establish a graph neural network model based on the sample geometric data of the target multi-layer well network, using different nodes to represent each layer of each production well and each layer of each water injection well; wherein, when the node represents a production well and a layer, the feature of the node is the liquid production of the production well and the layer, and when the node represents a water injection well and a layer, the feature of the node is the water injection volume of the water injection well and the layer; the edge of the graph neural network model is used for the distance and relative position between a well point and other well points; a well point is a production well or a water injection well, and the other well points are production wells or water injection wells;

[0132] The input end of the graph neural network model is used to input the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment; the output end of the graph neural network model is used to output the predicted liquid production of each layer of each production well at the second moment; the first moment is the previous moment of the second moment.

[0133] The production well liquid production prediction model determination module specifically includes:

[0134] A first training unit is used to adjust the model parameters of the embedded neural network model based on the predicted liquid production of each layer of each production well at different times, the predicted water injection of each layer of each water injection well at different times, the liquid production of each layer of each production well at different times, and the water injection of each layer of each water injection well at different times, so as to obtain a trained embedded neural network model;

[0135] The second training unit is used to:

[0136] Based on the sample formation data and the trained embedded neural network model, the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment are obtained; the first moment is any moment in the sample data;

[0137] Based on the predicted liquid production of each layer of each production well at the first moment, the predicted water injection volume of each layer of each water injection well at the first moment and the graph neural network model, a production well liquid production prediction model is determined.

[0138] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0139] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting dynamic connectivity between wells based on embedded graph neural network, characterized in that: include: Obtain production data of the target multi-layer well network at the current moment; Based on the production data of the target multi-layer well network at the current moment and the prediction model of the liquid production of the production wells, predict the liquid production of each production well in the target multi-layer well network at the next moment; The determination process of the production well liquid production prediction model is as follows: Determine sample data of the target multi-layer well pattern; the sample data includes sample production data, sample geometry data and sample formation data; the sample production data includes sample water injection volume of water injection wells and sample liquid production volume of production wells; Constructing an embedded graph neural network model; the embedded graph neural network model includes an embedded neural network model and a graph neural network model connected to the output end of the embedded neural network model; wherein the graph neural network model is constructed based on the sample geometry data; the constructing of the embedded graph neural network model specifically includes: According to the sample formation data, the sample production data is divided to obtain the liquid production of each layer of each production well at different times and the water injection volume of each layer of each water injection well at different times; Constructing an embedded neural network model; the input end of the embedded neural network model is used to input sample geological data of each layer of each production well and sample geological data of each layer of each water injection well; the output end of the embedded neural network model is used to output the predicted liquid production of each layer of each production well at different times and the predicted water injection volume of each layer of each water injection well at different times; Based on the sample geometric data of the target multi-layer well network, a graph neural network model is established, which uses different nodes to represent each layer of each production well and each layer of each water injection well; wherein, when the node represents a production well and a layer, the feature of the node is the liquid production of the production well and the layer, and when the node represents a water injection well and a layer, the feature of the node is the water injection volume of the water injection well and the layer; the edges of the graph neural network model are used for the distance and relative position between a well point and other well points; a well point is a production well or a water injection well, and the other well points are production wells or water injection wells; The input end of the graph neural network model is used to input the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment; the output end of the graph neural network model is used to output the predicted liquid production of each layer of each production well at the second moment; the first moment is the previous moment of the second moment; Determining a production well liquid production prediction model based on the sample production data, the sample formation data and the embedded graph neural network model; specifically comprising: Based on the predicted liquid production of each layer of each production well at different times, the predicted water injection of each layer of each water injection well at different times, the liquid production of each layer of each production well at different times, and the water injection of each layer of each water injection well at different times, the model parameters of the embedded neural network model are adjusted to obtain a trained embedded neural network model; Based on the sample formation data and the trained embedded neural network model, the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment are obtained; the first moment is any moment in the sample data; Based on the predicted liquid production of each layer of each production well at the first moment, the predicted water injection volume of each layer of each water injection well at the first moment and the graph neural network model, a production well liquid production prediction model is determined.

2. The method for predicting dynamic well connectivity based on embedded graph neural network according to claim 1, characterized in that: The sample data for determining the target multi-layer well pattern specifically includes: Acquire sample geometric data of the target multi-layer well network; the sample geometric data includes the number of water injection wells, the number of production wells, the distance between each production well and each water injection well, the distance between each production well and each production well, the distance between each water injection well and each water injection well, the relative position between each production well and each water injection well, the relative position between each production well and each production well, and the relative position between each water injection well and each water injection well; Acquire sample production data of the target multi-layer well pattern; the sample production data includes the liquid production of each production well at different times and the water injection volume of each water injection well at different times; Obtain sample formation data of the target multi-layer well network; the sample formation data include sample geological data of each layer of each production well and sample geological data of each layer of each injection well; the geological data include perforation parameters, layer depth, effective thickness, permeability and porosity.

3. A well dynamic connectivity prediction system based on embedded graph neural network, characterized in that: include: A data acquisition module is used to obtain the production data of the target multi-layer well network at the current moment; A liquid production prediction module, used to predict the liquid production of each production well in the target multi-layer well network at the next moment based on the production data of the target multi-layer well network at the current moment and the liquid production prediction model of the production wells; The determination process of the production well liquid production prediction model is as follows: A sample data determination module is used to determine the sample data of the target multi-layer well pattern; the sample data includes sample production data, sample geometry data and sample formation data; the sample production data includes sample water injection volume of water injection wells and sample liquid production volume of production wells; A neural network model construction module is used to construct an embedded graph neural network model; the embedded graph neural network model includes an embedded neural network model and a graph neural network model connected to the output end of the embedded neural network model; wherein the graph neural network model is constructed based on the sample geometry data; the neural network model construction module specifically includes: A division unit, used to divide the sample production data according to the sample formation data, to obtain the liquid production of each layer of each production well at different times and the water injection of each layer of each water injection well at different times; The first construction unit is used to construct an embedded neural network model; the input end of the embedded neural network model is used to input sample geological data of each layer of each production well and sample geological data of each layer of each water injection well; the output end of the embedded neural network model is used to output the predicted liquid production of each layer of each production well at different times and the predicted water injection volume of each layer of each water injection well at different times; The second construction unit is used to establish a graph neural network model based on the sample geometric data of the target multi-layer well network, using different nodes to represent each layer of each production well and each layer of each water injection well; wherein, when the node represents a production well and a layer, the feature of the node is the liquid production of the production well and the layer, and when the node represents a water injection well and a layer, the feature of the node is the water injection volume of the water injection well and the layer; the edge of the graph neural network model is used for the distance and relative position between a well point and other well points; a well point is a production well or a water injection well, and the other well points are production wells or water injection wells; The input end of the graph neural network model is used to input the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment; the output end of the graph neural network model is used to output the predicted liquid production of each layer of each production well at the second moment; the first moment is the previous moment of the second moment; A production well liquid production prediction model determination module is used to determine the production well liquid production prediction model based on the sample production data, the sample formation data and the embedded graph neural network model; the production well liquid production prediction model determination module specifically includes: A first training unit is used to adjust the model parameters of the embedded neural network model based on the predicted liquid production of each layer of each production well at different times, the predicted water injection of each layer of each water injection well at different times, the liquid production of each layer of each production well at different times, and the water injection of each layer of each water injection well at different times, so as to obtain a trained embedded neural network model; The second training unit is used to: Based on the sample formation data and the trained embedded neural network model, the predicted liquid production of each layer of each production well at the first moment and the predicted water injection volume of each layer of each water injection well at the first moment are obtained; the first moment is any moment in the sample data; Based on the predicted liquid production of each layer of each production well at the first moment, the predicted water injection volume of each layer of each water injection well at the first moment and the graph neural network model, a production well liquid production prediction model is determined.

4. The well dynamic connectivity prediction system based on embedded graph neural network according to claim 3, characterized in that: The sample data determination module specifically includes: A first acquisition unit is used to acquire sample geometric data of the target multi-layer well network; the sample geometric data includes the number of water injection wells, the number of production wells, the distance between each production well and each water injection well, the distance between each production well and each production well, the distance between each water injection well and each water injection well, the relative position between each production well and each water injection well, the relative position between each production well and each production well, and the relative position between each water injection well and each water injection well; A second acquisition unit is used to acquire sample production data of the target multi-layer well pattern; the sample production data includes the liquid production of each production well at different times and the water injection volume of each water injection well at different times; The third acquisition unit is used to acquire sample formation data of the target multi-layer well network; the sample formation data includes sample geological data of each layer of each production well and sample geological data of each layer of each injection well; the geological data includes perforation parameters, layer depth, effective thickness, permeability and porosity.

Images