Multi-layer system water injection volume splitting method and system based on heterogeneous graph neural network

Abstract

The present invention discloses a multi-layer system water injection splitting method and system based on heterogeneous graph neural network, which relates to the field of oil and gas reservoir seepage technology. The method comprises: obtaining three-dimensional geometric data, historical production data and formation data of a target well network; constructing a sample database for training a heterogeneous graph neural network based on the historical production data and formation data of the target well network; training and establishing a heterogeneous graph neural network model based on the three-dimensional geometric data of the target well network and the sample database; using the heterogeneous graph neural network model and the water injection volume of the water injection wells at the current stage, predicting the water injection splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells at the current stage, as well as the liquid production of each production well in the target well network. The present invention can clearly characterize the real underground situation and improve the splitting accuracy of the water injection volume of the water injection wells.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas reservoir seepage, and in particular to a multi-layer system water injection volume splitting method and system based on heterogeneous graph neural network. Background Art

[0002] Due to the complexity of underground conditions, the dynamic changes in the oil field development process are even more difficult to grasp. The production data available on site cannot fully and clearly represent the actual underground situation, such as: how the water flows from the injection well to the production well, how the injection volume of the injection well is split in each development layer, and whether there is any impact between the various development layers.

[0003] The existing technical solution is: obtain the geometric information and historical production data of the target well network, and establish a spatiotemporal graph structure data set of the target well network; divide the sliding time window according to the spatiotemporal graph structure data set, and establish a training sample set of the target well network; use the training sample set to train a multi-layer spatiotemporal graph neural network for water drive oil reservoir production prediction; use the multi-layer spatiotemporal graph neural network to obtain the production prediction results of the production wells of the target oil reservoir well network.

[0004] The above technical solutions have the following disadvantages: (1) The above solution only extracts the well network and inter-well characteristics through the graph neural network on a two-dimensional plane, while the actual oil reservoir is a three-dimensional space, and the oil field development is a multi-layer system development. The above solution leads to the failure to clearly characterize the underground internal structure and flow characteristics of the oil field development, so there is a risk of large model calculation, large resource consumption and inaccurate prediction results. (2) The existing technology uses traditional isomorphic graph neural networks for learning. There is only one type of node and relationship type (edge ​​type) in the isomorphic graph. Therefore, when constructing the graph neural network, all nodes share the same model parameters and have the same dimensional feature space, which further reduces the accuracy of the prediction results. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-layer system injection volume splitting method and system based on heterogeneous graph neural network, which can clearly represent the actual underground situation and improve the splitting accuracy of the injection volume of injection wells.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A multi-layer system water injection splitting method based on heterogeneous graph neural network, comprising:

[0008] Acquire three-dimensional geometric data, historical production data and formation data of a target well pattern; the target well pattern includes injection wells and production wells;

[0009] Based on the historical production data and formation data of the target well network, a sample database for training a heterogeneous graph neural network is constructed; the sample database includes the injection volume splitting components of each node of the target well network in three-dimensional space and the liquid production of each production well in different time periods;

[0010] Based on the three-dimensional geometric data of the target well pattern and the sample database, a heterogeneous graph neural network model is trained and established;

[0011] The heterogeneous graph neural network model and the water injection volume of the water injection wells at the current stage are used to predict the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells at the current stage, as well as the liquid production of each production well in the target well network.

[0012] Optionally, the obtaining of three-dimensional geometric data, historical production data and formation data of the target well pattern specifically includes:

[0013] Determine the target well pattern that needs multi-layer injection volume splitting analysis;

[0014] Acquire three-dimensional geometric data of the target well pattern; the three-dimensional geometric data include 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 path relationship between each production well and each water injection well, the path relationship between each production well and each production well, the path relationship between each water injection well and each water injection well, the distance between each perforation in each production well, the distance between each perforation in each water injection well, the distance between each perforation between each production well and each water injection well, the path relationship between each perforation in each production well, the path relationship between each perforation in each water injection well, and the path relationship between each perforation between each production well and each water injection well;

[0015] Obtaining the working sequence of each production well and each water injection well in the target well network, and determining the historical production data of each production well and each water injection well in the target well network based on the working sequence of each production well and each water injection well in the target well network; the working sequence includes the start working time of each production well, the start working time of each water injection well, and the shut-in operation and shut-in duration performed at different times; the historical production data includes the liquid production of each production well in different time periods, and the water injection volume of each water injection well in different time periods;

[0016] Acquire formation data of development layer sections near each perforation of each production well and each water injection well in the target well network; the formation data include permeability, effective thickness, porosity and formation pressure.

[0017] Optionally, constructing a sample database for training a heterogeneous graph neural network based on the historical production data and formation data of the target well pattern specifically includes:

[0018] Determine the preliminary liquid production corresponding to the marked water injection volume based on the marked water injection volume and the trained recurrent neural network; the marked water injection volume is any water injection volume in the historical production data;

[0019] Normalizing the formation data, and multiplying the normalized formation data with the initial liquid production corresponding to the marked water injection volume to obtain a multiplication result, and then using the multiplication result as the water injection volume splitting component of each node in the water injection well of the heterogeneous graph neural network;

[0020] The liquid production in the historical production data is used as the liquid production of each production well in the heterogeneous graph neural network.

[0021] Optionally, the training and establishing of a heterogeneous graph neural network model based on the three-dimensional geometric data of the target well pattern and the sample database specifically includes:

[0022] A heterogeneous graph neural network is established according to the three-dimensional geometric data of the target well pattern; the nodes in the heterogeneous graph neural network represent the perforation positions of different production wells and water injection wells; the heterogeneous graph neural network includes the positions of each production well and each water injection well, the node positions of each development layer section in the water injection well and the production well, and the path relationship between each node and its neighboring nodes;

[0023] Training the heterogeneous graph neural network based on the sample database to obtain a trained heterogeneous graph neural network;

[0024] The heterogeneous graph neural network model includes a trained recurrent neural network, a multiplication module and a trained heterogeneous graph neural network; the input end of the trained recurrent neural network is used to input the water injection volume of the water injection well at the current stage, and the output end of the trained recurrent neural network is used to output the preliminary liquid production corresponding to the water injection volume of the water injection well at the current stage; the input end of the multiplication module is used to input the preliminary liquid production and the normalized formation data corresponding to the preliminary liquid production; the output end of the multiplication module is used to output the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the input end of the trained heterogeneous graph neural network is used to input the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the output end of the trained heterogeneous graph neural network is used to output the liquid production of each production well in the target well network corresponding to the water injection volume of the water injection well at the current stage.

[0025] A multi-layer system water injection volume splitting system based on heterogeneous graph neural network, comprising:

[0026] A target well pattern data acquisition module is used to acquire three-dimensional geometric data, historical production data and formation data of the target well pattern; the target well pattern includes injection wells and production wells;

[0027] A sample database construction module is used to construct a sample database for training a heterogeneous graph neural network based on the historical production data and formation data of the target well network; the sample database includes the injection volume splitting components of each node of the target well network in three-dimensional space and the liquid production of each production well in different time periods;

[0028] A heterogeneous graph neural network model building module is used to train and build a heterogeneous graph neural network model based on the three-dimensional geometric data of the target well pattern and the sample database;

[0029] A prediction module is used to use the heterogeneous graph neural network model and the water injection volume of the water injection wells in the current stage to predict the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells in the current stage, as well as the liquid production of each production well in the target well network.

[0030] Optionally, the target well pattern data acquisition module specifically includes:

[0031] A target well pattern determination unit is used to determine the target well pattern that needs to be subjected to multi-layer injection volume splitting analysis;

[0032] a three-dimensional geometric data acquisition unit, configured to acquire the three-dimensional geometric data of the target well pattern; the three-dimensional 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 path relationship between each production well and each water injection well, the path relationship between each production well and each production well, the path relationship between each water injection well and each water injection well, the distance between each perforation in each production well, the distance between each perforation in each water injection well, the distance between each perforation between each production well and each water injection well, the path relationship between each perforation in each production well, the path relationship between each perforation in each water injection well, and the path relationship between each perforation between each production well and each water injection well;

[0033] A historical production data determination unit is used to obtain the working sequence of each production well and each water injection well in the target well network, and determine the historical production data of each production well and each water injection well in the target well network based on the working sequence of each production well and each water injection well in the target well network; the working sequence includes the start working time of each production well, the start working time of each water injection well, and the shut-in operation and shut-in duration performed at different times; the historical production data includes the liquid production of each production well in different time periods, and the water injection volume of each water injection well in different time periods;

[0034] The formation data acquisition unit is used to acquire the formation data of the development layer section near each perforation of each production well and each water injection well in the target well network; the formation data includes permeability, effective thickness, porosity and formation pressure.

[0035] Optionally, the sample database construction module specifically includes:

[0036] A water injection volume prediction unit, used to determine a preliminary liquid production volume corresponding to the marked water injection volume based on the marked water injection volume and a trained recurrent neural network; the marked water injection volume is any water injection volume in the historical production data;

[0037] A water injection amount split component determination unit is used to normalize the formation data, and multiply the normalized formation data with the preliminary liquid production corresponding to the marked water injection amount to obtain a multiplication result, and then use the multiplication result as the water injection amount split component of each node in the water injection well of the heterogeneous graph neural network;

[0038] The liquid production determination unit is used to use the liquid production in the historical production data as the liquid production of each production well in the heterogeneous graph neural network.

[0039] Optionally, the heterogeneous graph neural network model building module specifically includes:

[0040] A heterogeneous graph neural network construction unit is used to establish a heterogeneous graph neural network according to the three-dimensional geometric data of the target well pattern; the nodes in the heterogeneous graph neural network represent the perforation positions of different production wells and water injection wells; the heterogeneous graph neural network includes the positions of each production well and each water injection well, the node positions of each development layer section in the water injection well and the production well, and the path relationship between each node and its neighboring nodes;

[0041] A heterogeneous graph neural network model building unit, used for training the heterogeneous graph neural network based on the sample database to obtain a trained heterogeneous graph neural network;

[0042] The heterogeneous graph neural network model includes a trained recurrent neural network, a multiplication module and a trained heterogeneous graph neural network; the input end of the trained recurrent neural network is used to input the water injection volume of the water injection well at the current stage, and the output end of the trained recurrent neural network is used to output the preliminary liquid production corresponding to the water injection volume of the water injection well at the current stage; the input end of the multiplication module is used to input the preliminary liquid production and the normalized formation data corresponding to the preliminary liquid production; the output end of the multiplication module is used to output the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the input end of the trained heterogeneous graph neural network is used to input the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the output end of the trained heterogeneous graph neural network is used to output the liquid production of each production well in the target well network corresponding to the water injection volume of the water injection well at the current stage.

[0043] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0044] The present invention provides a multi-layer system water injection volume splitting method and system based on heterogeneous graph neural network, provides a three-dimensional spatial model of multi-layer system development, marks the perforation positions of water injection wells and production wells, considers the water injection volume splitting and oil-water two-phase flow mode in each development layer, is more in line with the actual physical model, reduces the calculation amount of the model, improves the splitting accuracy of the water injection volume of the water injection well, and improves the accuracy of the prediction of the liquid production of the production well. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] 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.

[0046] Figure 1 It is a schematic flow chart of a multi-layer system water injection volume splitting method based on a heterogeneous graph neural network of the present invention;

[0047] Figure 2 This is a schematic diagram of the perforation positions of each development layer section of the production well and the water injection well of the present invention;

[0048] Figure 3 It is a structural schematic diagram of a production well and a water injection well of the present invention;

[0049] Figure 4 A schematic diagram of constructing a heterogeneous graph neural network model of the present invention;

[0050] Figure 5It is a structural schematic diagram of the multi-layer system water injection volume splitting system based on heterogeneous graph neural network of the present invention. DETAILED DESCRIPTION

[0051] 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.

[0052] 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.

[0053] The current method only extracts the well network and inter-well features through graph neural networks in a two-dimensional plane to predict production, while the actual oil reservoir is a three-dimensional space, and the oil field development is mostly multi-layer system development. The above method fails to clearly characterize the underground internal structure of the oil field development, resulting in large calculation amount of the existing model, large resource consumption, and inaccurate prediction results. Therefore, the present invention provides a multi-layer system water injection splitting method and system based on heterogeneous graph neural networks, provides a three-dimensional space model of multi-layer system development, marks the perforation positions of water injection wells and production wells, considers the water injection splitting in each development layer and the oil-water two-phase flow mode, is more in line with the actual physical model, reduces the calculation amount of the model, improves the splitting accuracy of the water injection well water injection, and improves the prediction accuracy of the production well liquid production.

[0054] The internal structure of an actual oil reservoir is complex, and various factors in the process of oilfield development change dynamically and affect each other. Using a homogeneous graph neural network may not be able to learn the relationship between certain parameters. In view of this, the present invention provides a multi-layer system water injection volume splitting method and system based on a heterogeneous graph neural network. Since there can be more than one node type and edge type in a heterogeneous graph, different types of nodes are allowed to have features or attributes of different dimensions, so the generalization ability is stronger, the splitting accuracy of the water injection volume of the water injection well is improved, and the accuracy of the prediction of the liquid production of the production well is improved.

[0055] Embodiment 1

[0056] This embodiment provides a multi-layer system water injection splitting method based on heterogeneous graph neural network, see Figure 1 ,include:

[0057] Step 101: Acquire three-dimensional geometric data, historical production data and formation data of a target well network; the target well network includes injection wells and production wells.

[0058] Wherein, step 101 specifically includes:

[0059] Determine the target well pattern that needs multi-layer injection volume splitting analysis;

[0060] Acquire three-dimensional geometric data of the target well network; the three-dimensional geometric data include the number of injection wells, the number of production wells, the distance between each production well and each injection well, the distance between each production well and each production well, the distance between each injection well and each injection well, the path relationship between each production well and each injection well, the path relationship between each production well and each production well, the path relationship between each injection well and each injection well, the distance between each perforation in each production well, the distance between each perforation in each injection well, the distance between each perforation between each production well and each injection well, the path relationship between each perforation in each production well, the path relationship between each perforation in each injection well, and the path relationship between each perforation between each production well and each injection well.

[0061] The working sequence of each production well and each water injection well in the target well network is obtained, and based on the working sequence of each production well and each water injection well in the target well network, the historical production data of each production well and each water injection well in the target well network is determined; the working sequence includes the start working time of each production well, the start working time of each water injection well, and the shut-in operation and shut-in duration performed at different times. When shutting in the well, each node parameter of each production well and each water injection well is recorded as zero. The historical production data includes the liquid production of each production well in different time periods, and the water injection volume of each water injection well in different time periods.

[0062] Due to personal reasons or machine damage, the data of a certain well in a certain period of time is not on the same order of magnitude as other data, or even violates the flow law, so the data with obvious abnormalities must be eliminated.

[0063] The formation data of the development layer near each perforation of each production well and each water injection well in the target well network are obtained; the formation data include permeability, effective thickness, porosity, and formation pressure. The obviously abnormal data should be eliminated, and then the above four parameters are normalized and processed respectively. Figure 2 and Figure 3 shown.

[0064] Step 102: Based on the historical production data and formation data of the target well network, a sample database for training a heterogeneous graph neural network is constructed; the sample database includes the injection volume splitting components of each node of the target well network in three-dimensional space and the liquid production of each production well in different time periods.

[0065] Production prediction based on recurrent neural network. Since the development of water injection wells is delayed, that is, the real-time water injection volume is not highly correlated with the real-time liquid production. The real-time water injection volume of water injection wells will not be immediately reflected in the liquid production, but the long-term water injection volume has little effect on the liquid production. Therefore, there is no need to worry about the recurrent neural network (RNN) being able to handle only short-term dependencies but not long-term dependencies. Therefore, a recurrent neural network (RNN) is constructed.

[0066] The water injection volume from the current day to 5-10 days is used as training data (x1, x2, ..., xn), and the well spacing and row spacing are selected according to the size (the larger the distance, the longer the selection time), and the daily liquid production is used as the verification data Y. The daily production data of the past five years is used to build a training sample library, 80% as the training set, and 20% as the test set. The Adam activation function in Keras is used in this network model, the dropout rate (Dropout) of each layer of network nodes is set to 0.1, the error calculation method uses the mean square error, the iterative update method of the weight parameter uses the BPTT algorithm, and the number of layers of SimpleRNN is three. Finally, the relationship between the historical water injection volume and the daily liquid production is obtained. Through this model, the liquid production predicted by the historical water injection volume is used as the water injection volume of the water injection well in the subsequent heterogeneous graph neural network.

[0067] Since the splitting of injection volume in each layer from injection well to injection well is carried out underground, with current technology and cost requirements, it is impossible to monitor and measure the splitting of injection volume in each development layer, the flow volume, and the splitting of liquid production in each layer when it converges at the production well. However, it is known that they are related to the formation information.

[0068] The water injection volume predicted by the recurrent neural network (RNN) is multiplied by the four normalized parameters (permeability, effective thickness, porosity, and formation pressure) of each node in the water injection well, and a sample database of water injection volume of each node in the water injection well of the heterogeneous graph neural network is constructed.

[0069] Therefore, step 102 specifically includes:

[0070] Based on the marked water injection volume and the trained recurrent neural network, a preliminary liquid production volume corresponding to the marked water injection volume is determined; the marked water injection volume is any water injection volume in the historical production data; wherein the trained recurrent neural network is obtained by training the recurrent neural network with the water injection volume in the historical production data as input data and the liquid production volume in the historical production data as label data.

[0071] The formation data is normalized, and the normalized formation data is multiplied by the initial liquid production corresponding to the marked water injection volume to obtain a multiplication result, and then the multiplication result is used as the water injection volume splitting component of each node in the water injection well of the heterogeneous graph neural network.

[0072] The liquid production in the historical production data is used as the liquid production of each production well in the heterogeneous graph neural network.

[0073] Step 103: Based on the three-dimensional geometric data of the target well pattern and the sample database, a heterogeneous graph neural network model is trained and established. The heterogeneous graph neural network model is used to predict the relationship between the water injection volume in the target well pattern at the current stage and the liquid production volume in the target well pattern at the next stage.

[0074] A heterogeneous graph neural network is established according to the three-dimensional geometric information of the target well network. G(V, E) describes the topological structure of the graph, and the nodes of the graph neural network represent the perforation positions of different production wells and water injection wells. Each node is a vertex, V represents a set of vertices (node ​​set), V = {v1, v2, ···, vn}, and E is a set of edges in the graph (connection set). Obtain heterogeneous graph neural network topological graph data related to the learning content, the heterogeneous graph neural network includes the positions of each production well and water injection well, the node positions of each development layer segment in the water injection well and production well, and the path relationship between each node and its neighboring nodes (the same water injection well has only a top-down relationship, the same production well has only a bottom-up relationship, the relationship between each node of the water injection well pointing to each node of the production well, and there is a one-to-many or many-to-one path relationship between each node), and the path has at least one type; based on the type of the path, the nodes are grouped so that the path types of the neighboring nodes in the same group are the same.

[0075] The heterogeneous graph neural network uses the injection volume and liquid production sample database of each node in the three-dimensional space, the input is the injection volume splitting component of each node of the water injection well, and the output is the liquid production of the production well. The first 80% of the data is used as the training set data, 10% as the validation set data, and 10% as the test set data. The weights of various parameters (permeability, effective thickness, porosity, formation pressure) at the iterative nodes, the node attention weights and the path attention weights are continuously updated, and the loss function is calculated until the Loss value of the training set, validation set, and test set loss function of the heterogeneous graph neural network approaches zero, and the model training is completed.

[0076] Therefore, step 103 specifically includes:

[0077] A heterogeneous graph neural network is established according to the three-dimensional geometric data of the target well network; the nodes in the heterogeneous graph neural network represent the perforation positions of different production wells and water injection wells; the heterogeneous graph neural network includes the positions of each production well and each water injection well, the node positions of each development layer segment in the water injection well and the production well, and the path relationship between each node and its neighboring nodes.

[0078] The heterogeneous graph neural network is trained based on the sample database to obtain a trained heterogeneous graph neural network.

[0079] The heterogeneous graph neural network model includes a trained recurrent neural network, a multiplication module and a trained heterogeneous graph neural network; the input end of the trained recurrent neural network is used to input the water injection volume of the water injection well at the current stage, and the output end of the trained recurrent neural network is used to output the preliminary liquid production corresponding to the water injection volume of the water injection well at the current stage; the input end of the multiplication module is used to input the preliminary liquid production and the normalized formation data corresponding to the preliminary liquid production; the output end of the multiplication module is used to output the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the input end of the trained heterogeneous graph neural network is used to input the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the output end of the trained heterogeneous graph neural network is used to output the liquid production of each production well in the target well network corresponding to the water injection volume of the water injection well at the current stage.

[0080] The implementation process of step 102 and step 103 is as follows: Figure 4 shown.

[0081] The present invention relates the water injection volume of a water injection well to the liquid production volume in the next stage through a recurrent neural network (RNN), so as to reduce the computational complexity of the heterogeneous graph neural network and improve the generalization ability.

[0082] Step 104: Utilizing the heterogeneous graph neural network model and the water injection volume of the water injection wells at the current stage, predict the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells at the current stage, as well as the liquid production of each production well in the target well network.

[0083] According to the trained heterogeneous graph neural network model, the injection volume splitting components at each node in the water injection well, as well as the relationship between the water injection volume and the liquid production can be predicted. Combined with the actual reservoir conditions, the water injection volume of each well can be adjusted to maximize the economic benefits and guide the actual production on site.

[0084] Embodiment 2

[0085] like Figure 5 As shown, a multi-layer system water injection volume splitting system based on a heterogeneous graph neural network is provided in an embodiment of the present invention, comprising:

[0086] The target well pattern data acquisition module 501 is used to acquire the three-dimensional geometric data, historical production data and formation data of the target well pattern; the target well pattern includes injection wells and production wells.

[0087] The sample database construction module 502 is used to construct a sample database for training heterogeneous graph neural networks based on the historical production data and formation data of the target well network; the sample database includes the injection volume splitting components of each node of the target well network in three-dimensional space and the liquid production of each production well in different time periods.

[0088] The heterogeneous graph neural network model building module 503 is used to train and build a heterogeneous graph neural network model based on the three-dimensional geometric data of the target well network and the sample database.

[0089] The prediction module 504 is used to use the heterogeneous graph neural network model and the water injection volume of the water injection wells in the current stage to predict the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells in the current stage, as well as the liquid production of each production well in the target well network.

[0090] The target well pattern data acquisition module 501 specifically includes:

[0091] The target well pattern determination unit is used to determine the target well pattern that needs to be subjected to multi-layer injection volume splitting analysis.

[0092] A three-dimensional geometric data acquisition unit is used to acquire the three-dimensional geometric data of the target well network; the three-dimensional geometric data include 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 path relationship between each production well and each water injection well, the path relationship between each production well and each production well, the path relationship between each water injection well and each water injection well, the distance between each perforation in each production well, the distance between each perforation in each water injection well, the distance between each perforation between each production well and each water injection well, the path relationship between each perforation in each production well, the path relationship between each perforation in each water injection well, and the path relationship between each perforation between each production well and each water injection well.

[0093] A historical production data determination unit is used to obtain the working timing of each production well and each water injection well in the target well network, and determine the historical production data of each production well and each water injection well in the target well network based on the working timing of each production well and each water injection well in the target well network; the working timing includes the start working time of each production well, the start working time of each water injection well, and the shut-in operations and shut-in durations performed at different times; the historical production data includes the liquid production of each production well in different time periods, and the water injection volume of each water injection well in different time periods.

[0094] The formation data acquisition unit is used to acquire the formation data of the development layer section near each perforation of each production well and each water injection well in the target well network; the formation data includes permeability, effective thickness, porosity and formation pressure.

[0095] The sample database construction module 502 specifically includes:

[0096] The water injection volume prediction unit is used to determine the preliminary liquid production volume corresponding to the marked water injection volume based on the marked water injection volume and the trained recurrent neural network; the marked water injection volume is any water injection volume in the historical production data.

[0097] The water injection volume splitting component determination unit is used to normalize the formation data, and multiply the normalized formation data with the preliminary liquid production corresponding to the marked water injection volume to obtain a multiplication result, and then use the multiplication result as the water injection volume splitting component of each node in the water injection well of the heterogeneous graph neural network.

[0098] The liquid production determination unit is used to use the liquid production in the historical production data as the liquid production of each production well in the heterogeneous graph neural network.

[0099] The heterogeneous graph neural network model building module 503 specifically includes:

[0100] A heterogeneous graph neural network construction unit is used to establish a heterogeneous graph neural network based on the three-dimensional geometric data of the target well network; the nodes in the heterogeneous graph neural network represent the perforation positions of different production wells and water injection wells; the heterogeneous graph neural network includes the positions of each production well and each water injection well, the node positions of each development layer segment in the water injection well and the production well, and the path relationship between each node and its neighboring nodes.

[0101] The heterogeneous graph neural network model building unit is used to train the heterogeneous graph neural network based on the sample database to obtain a trained heterogeneous graph neural network.

[0102] The heterogeneous graph neural network model includes a trained recurrent neural network, a multiplication module and a trained heterogeneous graph neural network; the input end of the trained recurrent neural network is used to input the water injection volume of the water injection well at the current stage, and the output end of the trained recurrent neural network is used to output the preliminary liquid production corresponding to the water injection volume of the water injection well at the current stage; the input end of the multiplication module is used to input the preliminary liquid production and the normalized formation data corresponding to the preliminary liquid production; the output end of the multiplication module is used to output the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the input end of the trained heterogeneous graph neural network is used to input the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the output end of the trained heterogeneous graph neural network is used to output the liquid production of each production well in the target well network corresponding to the water injection volume of the water injection well at the current stage.

[0103] Compared with the prior art, the present invention has the following innovative features:

[0104] 1. The heterogeneous graph neural network is constructed using three-dimensional geometric information. The positions of each node and the path relationship between them play a crucial role in the final result. The existing technology only has the relationship between wells on a two-dimensional plane. This model is more in line with the actual reservoir physical model, with higher generalization ability and accuracy.

[0105] 2. Considering the delay of water injection on the same day, a recurrent neural network is used to improve the generalization ability and accuracy of the model.

[0106] 3. Use heterogeneous graph neural networks instead of traditional homogeneous graph neural networks.

[0107] Compared with the prior art, the present invention has the following advantages:

[0108] 1. The model is established based on the three-dimensional reservoir multi-layer development, which is more in line with the actual situation, optimizes the structure, reduces the calculation amount of the model, and improves the prediction accuracy.

[0109] 2. Due to the delay in water injection development, the real-time water injection volume of the injection well will not be immediately reflected in the liquid production, that is, it is not highly correlated with the liquid production. Therefore, constructing a recurrent neural network is more in line with the actual situation, can better learn the internal structure, has stronger generalization ability, and improves the prediction accuracy.

[0110] 3. Using graph neural networks based on graph heterogeneous networks, there can be more than one type of nodes and edges in a heterogeneous graph, thus allowing different types of nodes to have features or attributes of different dimensions, better learning the internal structure, stronger generalization ability, and improved prediction accuracy.

[0111] 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.

[0112] 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 multi-layer system water injection splitting method based on heterogeneous graph neural network, characterized in that: include: Acquire three-dimensional geometric data, historical production data and formation data of a target well pattern; the target well pattern includes injection wells and production wells; Based on the historical production data and formation data of the target well network, a sample database for training a heterogeneous graph neural network is constructed; the sample database includes the injection volume splitting components of each node of the target well network in three-dimensional space and the liquid production of each production well in different time periods; Based on the three-dimensional geometric data of the target well pattern and the sample database, a heterogeneous graph neural network model is trained and established; the nodes in the heterogeneous graph neural network represent the perforation positions of different production wells and water injection wells; the heterogeneous graph neural network includes the positions of each production well and each water injection well, the node positions of each development layer section in the water injection well and the production well, and the path relationship between each node and its neighboring nodes; The heterogeneous graph neural network model and the water injection volume of the water injection wells at the current stage are used to predict the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells at the current stage, as well as the liquid production of each production well in the target well network.

2. A multi-layer system water injection volume splitting method based on heterogeneous graph neural network according to claim 1, characterized in that: The acquisition of the three-dimensional geometric data, historical production data and formation data of the target well pattern specifically includes: Determine the target well pattern that needs multi-layer injection volume splitting analysis; Acquire three-dimensional geometric data of the target well pattern; the three-dimensional geometric data include 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 path relationship between each production well and each water injection well, the path relationship between each production well and each production well, the path relationship between each water injection well and each water injection well, the distance between each perforation in each production well, the distance between each perforation in each water injection well, the distance between each perforation between each production well and each water injection well, the path relationship between each perforation in each production well, the path relationship between each perforation in each water injection well, and the path relationship between each perforation between each production well and each water injection well; Obtaining the working sequence of each production well and each water injection well in the target well network, and determining the historical production data of each production well and each water injection well in the target well network based on the working sequence of each production well and each water injection well in the target well network; the working sequence includes the start working time of each production well, the start working time of each water injection well, and the shut-in operation and shut-in duration performed at different times; the historical production data includes the liquid production of each production well in different time periods, and the water injection volume of each water injection well in different time periods; Acquire formation data of development layer sections near each perforation of each production well and each water injection well in the target well network; the formation data include permeability, effective thickness, porosity and formation pressure.

3. The multi-layer system water injection volume splitting method based on heterogeneous graph neural network according to claim 1 is characterized in that: The constructing of a sample database for training a heterogeneous graph neural network based on the historical production data and formation data of the target well pattern specifically includes: Determine the preliminary liquid production corresponding to the marked water injection volume based on the marked water injection volume and the trained recurrent neural network; the marked water injection volume is any water injection volume in the historical production data; Normalizing the formation data, and multiplying the normalized formation data with the initial liquid production corresponding to the marked water injection volume to obtain a multiplication result, and then using the multiplication result as the water injection volume splitting component of each node in the water injection well of the heterogeneous graph neural network; The liquid production in the historical production data is used as the liquid production of each production well in the heterogeneous graph neural network.

4. A multi-layer system water injection volume splitting method based on heterogeneous graph neural network according to claim 3, characterized in that: The training and establishing of a heterogeneous graph neural network model based on the three-dimensional geometric data of the target well pattern and the sample database specifically includes: Establishing a heterogeneous graph neural network according to the three-dimensional geometric data of the target well pattern; Training the heterogeneous graph neural network based on the sample database to obtain a trained heterogeneous graph neural network; The heterogeneous graph neural network model includes a trained recurrent neural network, a multiplication module and a trained heterogeneous graph neural network; the input end of the trained recurrent neural network is used to input the water injection volume of the water injection well at the current stage, and the output end of the trained recurrent neural network is used to output the preliminary liquid production corresponding to the water injection volume of the water injection well at the current stage; the input end of the multiplication module is used to input the preliminary liquid production and the normalized formation data corresponding to the preliminary liquid production; the output end of the multiplication module is used to output the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the input end of the trained heterogeneous graph neural network is used to input the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the output end of the trained heterogeneous graph neural network is used to output the liquid production of each production well in the target well network corresponding to the water injection volume of the water injection well at the current stage.

5. A multi-layer system water injection splitting system based on heterogeneous graph neural network, characterized in that: include: The target well pattern data acquisition module is used to acquire the three-dimensional geometric data, historical production data and formation data of the target well pattern; The target well pattern includes water injection wells and production wells; A sample database construction module is used to construct a sample database for training a heterogeneous graph neural network based on the historical production data and formation data of the target well network; the sample database includes the injection volume splitting components of each node of the target well network in three-dimensional space and the liquid production of each production well in different time periods; A heterogeneous graph neural network model building module is used to train and build a heterogeneous graph neural network model based on the three-dimensional geometric data of the target well network and the sample database; the nodes in the heterogeneous graph neural network represent the perforation positions of different production wells and water injection wells; the heterogeneous graph neural network includes the positions of each production well and each water injection well, the node positions of each development layer section in the water injection well and the production well, and the path relationship between each node and its neighboring nodes; A prediction module is used to use the heterogeneous graph neural network model and the water injection volume of the water injection wells in the current stage to predict the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection wells in the current stage, as well as the liquid production of each production well in the target well network.

6. A multi-layer water injection volume splitting system based on heterogeneous graph neural network according to claim 5, characterized in that: The target well pattern data acquisition module specifically includes: A target well pattern determination unit is used to determine the target well pattern that needs to be subjected to multi-layer injection volume splitting analysis; a three-dimensional geometric data acquisition unit, configured to acquire the three-dimensional geometric data of the target well pattern; the three-dimensional 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 path relationship between each production well and each water injection well, the path relationship between each production well and each production well, the path relationship between each water injection well and each water injection well, the distance between each perforation in each production well, the distance between each perforation in each water injection well, the distance between each perforation between each production well and each water injection well, the path relationship between each perforation in each production well, the path relationship between each perforation in each water injection well, and the path relationship between each perforation between each production well and each water injection well; A historical production data determination unit is used to obtain the working sequence of each production well and each water injection well in the target well network, and determine the historical production data of each production well and each water injection well in the target well network based on the working sequence of each production well and each water injection well in the target well network; the working sequence includes the start working time of each production well, the start working time of each water injection well, and the shut-in operation and shut-in duration performed at different times; the historical production data includes the liquid production of each production well in different time periods, and the water injection volume of each water injection well in different time periods; The formation data acquisition unit is used to acquire the formation data of the development layer section near each perforation of each production well and each water injection well in the target well network; the formation data includes permeability, effective thickness, porosity and formation pressure.

7. The multi-layer water injection volume splitting system based on heterogeneous graph neural network according to claim 5 is characterized in that: The sample database construction module specifically includes: A water injection volume prediction unit, used to determine a preliminary liquid production volume corresponding to the marked water injection volume based on the marked water injection volume and a trained recurrent neural network; the marked water injection volume is any water injection volume in the historical production data; A water injection amount split component determination unit is used to normalize the formation data, and multiply the normalized formation data with the preliminary liquid production corresponding to the marked water injection amount to obtain a multiplication result, and then use the multiplication result as the water injection amount split component of each node in the water injection well of the heterogeneous graph neural network; The liquid production determination unit is used to use the liquid production in the historical production data as the liquid production of each production well in the heterogeneous graph neural network.

8. The multi-layer water injection volume splitting system based on heterogeneous graph neural network according to claim 5 is characterized in that: The heterogeneous graph neural network model building module specifically includes: A heterogeneous graph neural network construction unit, used for establishing a heterogeneous graph neural network according to the three-dimensional geometric data of the target well pattern; A heterogeneous graph neural network model building unit, used for the heterogeneous graph neural network model building unit, used to train the heterogeneous graph neural network based on the sample database to obtain a trained heterogeneous graph neural network; The heterogeneous graph neural network model includes a trained recurrent neural network, a multiplication module and a trained heterogeneous graph neural network; the input end of the trained recurrent neural network is used to input the water injection volume of the water injection well at the current stage, and the output end of the trained recurrent neural network is used to output the preliminary liquid production corresponding to the water injection volume of the water injection well at the current stage; the input end of the multiplication module is used to input the preliminary liquid production and the normalized formation data corresponding to the preliminary liquid production; the output end of the multiplication module is used to output the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the input end of the trained heterogeneous graph neural network is used to input the water injection volume splitting components at each perforation of each water injection well in the target well network corresponding to the water injection volume of the water injection well at the current stage; the output end of the trained heterogeneous graph neural network is used to output the liquid production of each production well in the target well network corresponding to the water injection volume of the water injection well at the current stage.

Images