A rapid prediction method for production performance of tight oil and gas reservoirs with multi-scale fracture development

Abstract

The present invention discloses a method for rapidly predicting the production performance of a tight oil and gas reservoir with multi-scale fracture development, belonging to the technical field of oil exploitation, and comprising the following steps: Step 1, obtaining multi-scale fracture distribution data samples and production simulation data samples to construct a sample library; Step 2, establishing a fracture network graph to represent the data; Step 3, calculating the vertex connectivity strength matrix of the fracture network based on the graph-represented data; Step 4, establishing a depth graph and temporal hybrid learning model for rapidly predicting the production performance of a tight oil and gas reservoir; Step 5, training the depth graph and temporal hybrid learning model, and verifying the performance of the model after training; Step 6, rapidly predicting the production performance of a tight oil and gas reservoir based on the trained model. The present invention can improve the efficiency of predicting the production performance of a tight oil and gas reservoir, and provide a basis for historical matching and production optimization of a tight oil and gas reservoir, etc.

Description

Technical Field

[0001] The present invention belongs to the technical field of oil exploitation, and particularly relates to a method for rapidly predicting the production performance of a tight oil and gas reservoir with multi-scale fracture development. Background Art

[0002] Tight oil and gas reservoirs have low permeability and well-developed natural fractures. Effective development requires hydraulic fracturing to transform the reservoir to form a complex fracture network. Therefore, the fracture distribution greatly affects the production performance of tight oil and gas reservoirs.

[0003] The prediction of reservoir production performance is to predict the production performance data of production wells, such as production rate and bottom-hole pressure, based on known reservoir information, such as geological parameters and development measures. The reservoir production performance prediction technology can analyze the production changes of the reservoir and predict the future production performance of the reservoir, so as to timely optimize the production management of the reservoir. Typical reservoir production performance prediction methods include reservoir engineering methods and numerical simulation methods. Among them, the reservoir engineering method uses empirical formulas or charts to predict the production performance of the reservoir. For example, decline curve analysis or analytical methods are used. The factors considered are relatively simple, the application range is limited, and the prediction accuracy is low. The numerical simulation method is the most commonly used reservoir production performance prediction method at present. It solves the complex partial differential equations of reservoir fluid flow through numerical calculation to simulate and predict the production performance of the reservoir. The calculation accuracy is high and it can fully consider various aspects of information of the reservoir, but it has the disadvantages of low calculation efficiency and long solution time. For tight oil and gas reservoirs with complex multi-scale fracture distributions, typical numerical simulation methods include dual-porosity models, discrete fracture models, and embedded discrete fracture models, etc. Usually, it is necessary to consider the fracture distribution and establish the flow equations of fluids in the matrix and fractures and the fluid exchange equations between fractures and the matrix, and perform production performance calculations by numerically solving complex partial differential equations using finite differences. The calculation efficiency is low and the workload is large.

[0004] Considering that the workload of using numerical simulation methods for predicting the production dynamics of oil and gas reservoirs is large and the computational efficiency is low, to overcome this problem, there have been many attempts to construct alternative models of numerical simulation methods using deep learning methods for predicting the production dynamics of oil and gas reservoirs. However, the current production dynamics prediction models based on deep learning mainly target the continuous geological properties of conventional oil and gas reservoirs, such as parameters like permeability and porosity, and use deep convolutional neural networks for feature learning and extraction. They are difficult to effectively apply to tight oil and gas reservoirs with complex multi-scale discrete fracture network distributions. Fractures are the main flow channels for fluids in tight oil and gas reservoirs, and the fracture distribution plays a dominant role in predicting the production dynamics of tight oil and gas reservoirs. However, fractures are discretely distributed in the reservoir space and have characteristics such as multi-scale and irregularity. Conventional deep convolutional neural networks are difficult to effectively learn data with discrete features in space. Therefore, the present invention provides a method for rapidly predicting the production dynamics of tight oil and gas reservoirs based on a deep learning model, especially establishing a graph representation method and a deep graph learning method suitable for complex discrete fracture networks, which can greatly improve the computational efficiency of predicting the production dynamics of tight oil and gas reservoirs while ensuring accuracy. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a method for rapidly predicting the production dynamics of tight oil and gas reservoirs with multi-scale fracture distributions. Based on quantitatively characterizing the connectivity relationship of the fracture network, an improved Transformer model is combined to establish a deep graph learning module for feature learning and extraction of discrete fractures, and the extracted fracture network features are transmitted to the deep time series learning module to predict the production dynamic indicators of production wells in tight oil and gas reservoirs. The combination of the deep graph learning and the deep time series learning modules constructs an end-to-end deep graph and time series hybrid learning model to achieve rapid prediction of the production dynamics of tight oil and gas reservoirs.

[0006] The technical solution of the present invention is as follows:

[0007] A method for rapidly predicting the production dynamics of tight oil and gas reservoirs with multi-scale fracture development, comprising the following steps:

[0008] Step 1, obtain multi-scale fracture distribution data samples and production simulation data samples, and construct a sample library;

[0009] Step 2, establish graph representation data of the fracture network;

[0010] Step 3, calculate the vertex connectivity strength matrix of the fracture network based on the graph representation data;

[0011] Step 4, establish a deep graph and time series hybrid learning model for rapidly predicting the production dynamics of tight oil and gas reservoirs;

[0012] Step 5, train the deep graph and time series hybrid learning model, and verify the performance of the model after training is completed;

[0013] Step 6: Based on the trained model, perform rapid prediction on the production dynamics of tight oil and gas reservoirs.

[0014] Further, the specific process of Step 1 is as follows:

[0015] Step 1.1: Conduct stochastic modeling on the multi-scale fracture distribution of tight oil and gas reservoirs to establish a multi-scale fracture distribution data sample containing several groups of different fracture distributions;

[0016] Step 1.2: Use the embedded discrete fracture numerical simulation method to perform numerical simulation calculations on different fracture distributions to obtain corresponding production simulation data samples. The production simulation data samples are the production dynamic data of production wells, including daily oil production, daily gas production, daily water production, and bottom hole pressure;

[0017] Step 1.3: Combine the fracture distribution data samples and the corresponding production simulation data samples to form a sample library, and divide the sample library into a training set and a test set according to a certain proportion.

[0018] Further, in Step 2, a group of fracture distributions is a fracture network, which contains multiple discrete fractures. The parameters of each fracture include length, position, strike, aperture, porosity, and permeability; based on the discrete fracture parameters, establish the graph representation data of the fracture network, that is , represents the graph representation data of the fracture network diagram, represents the vertex set of the graph, represents the edge set of the graph; among them, the two endpoints of each single fracture are used as the vertices of the graph, and the position coordinates of the fracture endpoints are used as vertex features; the fracture is used as the edge of the graph, and the length, strike, aperture, porosity, and permeability of the fracture are used as edge features; intersecting fractures are re-divided from the intersection point.

[0019] Further, in Step 3, calculate the connectivity strength of the fracture vertices based on the graph representation data of the fracture network, and construct the vertex connectivity strength matrix of the fracture network; the calculation formula for the connectivity strength between any two vertices in the matrix is:

[0020] (1);

[0021] In the formula, represents the th vertex and the th vertex row and the th column element in the corresponding vertex connectivity strength matrix; represents the th vertex and the th vertex Represents the total number of vertices.

[0022] Furthermore, the specific process of step 4 is as follows:

[0023] Step 4.1: Establish a deep graph learning module for feature learning and extraction of the graph representation data of the fracture network;

[0024] Step 4.2: Establish a deep time series learning module for predicting the production dynamic data of production wells;

[0025] Step 4.3: Combine the deep graph learning module and the deep time series learning module to obtain an end-to-end deep graph and time series hybrid learning model; the input data of the deep graph and time series hybrid learning model is the graph representation data of the fracture network and the vertex connectivity strength matrix of the fracture network, and the output is the production dynamic prediction data.

[0026] Furthermore, in step 4.1, the deep graph learning module includes an embedding layer, an information propagation aggregation model, an improved Transformer model, and a global pooling layer; the specific working process is as follows:

[0027] Step 4.1.1: Input the graph representation data of the fracture network into the embedding layer, and the embedding layer uses two independent fully connected layers to perform a feature mapping on the vertex features and edge features of the graph data respectively;

[0028] Step 4.1.2: Use the information propagation aggregation model to fuse the edge features mapped by the embedding layer into the vertex features; specifically: the information propagation aggregation model propagates the features of the edge between the th vertex and the th vertex to the th vertex and the th vertex , uses a vector concatenation and merging operation to merge the edge features into the vertex feature vector, and uses a single layer of fully connected neural network to perform a mapping calculation on the merged vector at the vertex to obtain the updated vertex features;

[0029] Step 4.1.3: Based on the improved Transformer model, perform multi-level updates and feature transformations on the vertex features after fusing the edge features; the improved Transformer model embeds the connectivity strength matrix into the learning and prediction process of the original Transformer model, and uses the vertex connectivity strength matrix as the fracture network structure encoding information to participate in the calculation process of a single attention operation, specifically as follows:

[0030] (2);

[0031] In the formula, represents the updated vertex feature output by the -th layer encoder; represents the dimension of the updated vertex feature of the -th layer encoder; is the softmax function; , and represent the query matrix, key matrix, and value matrix in the attention operation; is the dimension of the updated vertex feature of the -th layer encoder; is a learnable weight matrix; is the transpose symbol;

[0032] Step 4.1.4: Use the global pooling layer to pool all vertex features into a feature vector as the output of the depth map learning module. The feature vector is the high-level feature of the extracted fracture network. .

[0033] Furthermore, in Step 4.2, the depth-time series learning module uses the original Transformer model for production dynamic time series prediction. The specific working process is as follows:

[0034] Step 4.2.1: Copy the high-level feature of the fracture network to time steps as the input vector, and add it to the time series encoding information as the input of the encoder. The time series encoding information is obtained by encoding the time order of production dynamic data.

[0035] Step 4.2.2: The multi-head self-attention mechanism inside the encoder performs feature transformation on the time series.

[0036] Step 4.2.3: Use the fully connected layer as the output layer to output the production dynamic prediction value corresponding to each time step , , where is the number of production prediction indicators; represents the number of production wells; represents the number of production dynamic indicator types; represents the number of prediction time steps of production dynamics.

[0037] Furthermore, in Step 5, the depth map and time series hybrid learning model is trained by combining the training set divided in Step 1, and the model parameters are optimized to minimize the loss function. The mean absolute error is used as the loss function of the model, and the calculation formula is as follows:

[0038] (3);

[0039] Wherein, is the mean absolute error; represents the number of samples; and respectively represent the true value and the model prediction result of the th sample;

[0040] After the training is completed, the test set divided in Step 1 is used to verify the performance of the model. The verification index is the coefficient of determination, and the formula is as follows:

[0041] (4);

[0042] Wherein, is the coefficient of determination; represents the average value of the true values of the samples.

[0043] Furthermore, in the said Step 6, the depth map and the time series hybrid learning model that are trained and have good test performance are output. The fracture distribution parameters of the tight oil and gas reservoir and the vertex connectivity strength matrix corresponding to the calculated fracture distribution are used as the model inputs to quickly predict the corresponding production dynamic data.

[0044] The beneficial technical effects brought by the present invention: The present invention establishes a depth map and time series hybrid learning model based on the discrete connectivity structure of the fracture network and the improved Transformer model for the production dynamic prediction task of tight oil and gas reservoirs with complex fracture network distributions. This model can quickly predict the production dynamics of production wells in tight oil and gas reservoirs according to the fracture distribution. The fast prediction model constructed by the present invention can be used to predict the production dynamics of tight oil and gas reservoirs, and can replace the reservoir numerical simulation method in reservoir engineering tasks involving production dynamic prediction such as production optimization and automatic history matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flowchart of the method for quickly predicting the production dynamics of a tight oil and gas reservoir with multi-scale fracture distributions according to the present invention.

[0046] Figure 2 is a schematic diagram of a tight oil reservoir block with multi-scale fracture distributions in an embodiment of the present invention.

[0047] Figure 3 is a schematic diagram of three randomly selected different fracture distributions in an embodiment of the present invention; wherein, (a), (b), and (c) respectively represent the schematic diagrams of the first group of fracture distributions, the second group of fracture distributions, and the third group of fracture distributions.

[0048] Figure 4It is the daily water production curve graph of production well 1 corresponding to three groups of different fracture distributions in the embodiments of the present invention.

[0049] Figure 5 It is the daily oil production curve graph of production well 1 corresponding to three groups of different fracture distributions in the embodiments of the present invention.

[0050] Figure 6 It is the graph of the change of the loss function with the iteration during the training process of the depth map and time series hybrid learning model in the embodiments of the present invention.

[0051] Figure 7 It is the fracture network distribution graph of a test sample in the embodiments of the present invention.

[0052] Figure 8 It is the comparison graph between the predicted result of the daily water production model of production well 1 and the numerical simulation result in the embodiments of the present invention.

[0053] Figure 9 It is the comparison graph between the predicted result of the daily oil production model of production well 1 and the numerical simulation result in the embodiments of the present invention. Detailed implementation manners

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners:

[0055] The present invention models the multi-scale complex discrete fracture network widely distributed in tight oil and gas reservoirs as graph structure data; establishes a quantification method to calculate the connectivity relationship of the fracture network; combines the fracture network connectivity relationship with the Transfomer neural network model to establish a depth graph learning module and a depth time series learning module to respectively perform multi-scale discrete fracture network feature learning and prediction of production dynamic data of tight oil and gas reservoirs; integrates the depth graph learning module and the depth time series learning module to form a production dynamic rapid prediction method for tight oil and gas reservoirs.

[0056] As Figure 1 shown, a production dynamic rapid prediction method for tight oil and gas reservoirs with multi-scale fracture distributions specifically includes the following steps:

[0057] Step 1, obtain multi-scale fracture distribution data samples and production simulation data samples, and construct a sample library; the specific process is as follows:

[0058] Step 1.1. Randomly model the multi-scale fracture distribution of the tight oil and gas reservoir to establish a multi-scale fracture distribution data sample containing several groups of different fracture distributions; the description parameters of the fracture distribution include statistical information such as the position, strike, length, aperture, porosity, and permeability of the fractures. Among them, the position of the fractures can be obtained by using one of the methods of uniform distribution, Poisson process, Cluster process, Cox process, or Markov process; the strike of the fractures can be obtained by using one of the methods of uniform distribution, Fisher distribution, von-Mises distribution, or normal distribution; the fracture length distribution can be obtained by using one of the methods of uniform distribution, fractal simulation, or power-law distribution; the aperture and porosity of the fractures need to be obtained from core analysis data and well logging analysis data; the permeability of the fractures can be calculated according to the fracture length in combination with the cubic law. The present invention uses a discrete fracture model to explicitly characterize the fracture distribution, and the fractures are randomly statistically modeled according to fracture groups to construct a multi-scale fracture distribution data sample containing 1000 to 2000 groups of different fracture distributions.

[0059] Step 1.2. Use the embedded discrete fracture numerical simulation method to numerically simulate different fracture distributions to obtain corresponding production simulation data samples. The production simulation data samples are the production dynamic data of production wells, including the oil production rate per day (OPR), gas production rate per day (GPR), water production rate per day (WPR), and bottom hole pressure (BHP).

[0060] Step 1.3. Combine the fracture distribution data samples and the corresponding production simulation data samples to form a sample library, and divide the sample library into a training set and a test set according to a certain proportion.

[0061] Step 2. Establish fracture network graph representation data.

[0062] A group of fracture distributions is a fracture network, which contains multiple discrete fractures. Each fracture is described by parameters such as length, position, strike, aperture, porosity, and permeability. Based on the discrete fracture parameters, establish the graph representation data of the fracture network, that is , represents the fracture network graph representation data, represents the vertex set of the graph, represents the edge set of the graph, which is composed of each fracture. Among them, the two endpoints of each single fracture are used as the vertices of the graph, and the position coordinates of the fracture endpoints are used as vertex features, also known as vertex attributes; the fracture is used as the edge of the graph, and the length, strike, aperture, porosity, and permeability of the fracture are used as edge features, also known as edge attributes; intersecting fractures are re-divided from the intersection point.

[0063] Step 3. Calculate the vertex connectivity strength matrix of the fracture network based on the graph representation data.

[0064] Calculate the connectivity strength of fracture vertices based on the graph representation data of the fracture network obtained in Step 2, obtain the vertex connectivity strength matrix of the fracture network, quantitatively calculate the connection distance between fracture vertices through the shortest connection path between fracture vertices and the fracture edge length, calculate the connectivity strength of fracture vertices according to the connection distance, and quantitatively model the discrete structural relationship of the fracture network. Specifically, the connectivity strength between any two vertices is quantitatively calculated through the connection distance between the vertices. In the fracture network, the vertices are the endpoints of the fractures and are only connected to each other through the fracture edges. Therefore, the distance between the vertices can be accurately calculated through the length of the fracture edges connected between the vertices; there may be one or more connection paths between two vertices, and the connection distance is calculated by the shortest path. The greater the distance between fracture vertices indicates that the connectivity relationship between the two vertices is weaker, that is, the connectivity strength index is smaller; on the contrary, the smaller the distance between the vertices indicates that the connectivity relationship between the vertices is stronger, that is, the connectivity strength index is larger. After calculating the connection distances corresponding to all the vertices of the fracture network, perform normalization calculation to obtain the vertex connectivity strength matrix of the fracture network ; The calculation formula for the connectivity strength between any two vertices is:

[0065] (1);

[0066] In the formula, represents the connectivity strength between the th vertex and the th vertex, corresponding to the element in the th row and the th column of the vertex connectivity strength matrix; represents the connection distance between the th vertex and the th vertex; represents the total number of vertices.

[0067] Through the connectivity strength calculated by this method, it can be realized that the connectivity strength between the vertices with the smallest connection distance is the largest, and the connectivity strength between the vertices with the largest connection distance is the smallest. For two fracture vertices without a connection relationship, their connectivity strength is set to 0.

[0068] Step 4: Establish a depth graph and temporal hybrid learning model for rapid prediction of the production dynamics of tight oil and gas reservoirs; the specific process is as follows:

[0069] Step 4.1: Establish a depth graph learning module suitable for feature learning of the graph representation data of the fracture network. The main function of the depth graph learning module is to perform feature learning and extraction on the graph representation data of the fracture network;

[0070] The deep graph learning module uses an information propagation aggregation model to fuse the edge features of the graph representation data of the fracture network into the vertex features. It further performs multi-level updates and feature transformations on the vertex features of the graph based on the improved Transformer model. The connectivity strength matrix of the fracture vertices provides additional structural encoding information for the improved Transformer model, models the discrete structural information of the fracture network, and extracts high-level features of the discrete fracture network.

[0071] The input data of the deep graph learning module is the fracture network graph representation data established in step 2 And the vertex connectivity strength matrix of the fracture network obtained in step 3 The deep graph learning module includes an embedding layer, an information propagation aggregation model, an improved Transformer model, and a global pooling layer; the specific working process is as follows:

[0072] Step 4.1.1: Represent the data in the fracture network diagram Input embedding layer, which uses two independent fully connected layers to perform feature mapping on vertex features and edge features of graph data respectively;

[0073] Step 4.1.2: Use the information propagation aggregation model to fuse the edge features of the graph representation data after the embedding layer mapping into the vertex features; specifically: the information propagation aggregation model fuses the first The vertices and The edges between vertices The feature is propagated to the edge corresponding to Vertices and Vertices , using vector concatenation and merging operations, the edge features are merged into the vertex feature vector, and a fully connected neural network is used to perform a mapping calculation on the merged vectors at the vertex to obtain the updated vertex features; this operation can achieve the fusion of the edge features of the crack into the vertex features, and only the vertex features are updated in subsequent learning, avoiding the need to update and learn the edge and vertex features independently;

[0074] Step 4.1.3, based on the improved Transformer model, the vertex features after the edge features are integrated are updated and transformed at multiple levels. The vertex connectivity strength matrix of the crack network provides the improved Transformer model with the spatial structure encoding information between the vertices of the crack network, and models the discrete structure information of the crack network. The improved Transformer model uses the encoder as the basic unit. An improved Transformer model consists of Composed of multiple encoders in series, one encoder consists of multi-head attention operation, normalization operation, linear operation, and addition operation. To embed the vertex connectivity strength matrix into the learning and prediction process of the original Transformer model, the vertex connectivity strength matrix is used as the encoded information of the fracture network structure to participate in the calculation process of one attention operation, as shown below:

[0075] (2);

[0076] In the formula, represents the updated vertex features output by the -th layer encoder; represents the dimension of the updated vertex features of the -th layer encoder; is the softmax function; , and represent the query matrix (vector), key matrix (vector), and value matrix (vector) in the attention operation, which can be calculated through the vertex features input by the -th layer encoder (i.e., the updated vertex features output by the -th layer encoder) , is the dimension of the updated vertex features of the -th layer encoder, , , , , and are learnable weight matrices; is a learnable weight matrix used to adaptively adjust the importance of the vertex connectivity strength matrix in the attention calculation process; is the transpose symbol.

[0077] Formula (2) represents one attention calculation, and multi-head attention is to repeat the calculation times, and the output of the encoder is obtained through the concatenation operation. Formula (2) integrates the vertex connectivity strength matrix of the fracture network on the basis of the basic attention calculation operation, which is an improvement to the attention calculation method of the Transformer encoder, is the basic attention calculation, which distributes and calculates the attention for all fracture vertex features and has global characteristics; while is the vertex connectivity strength matrix of the fracture network. The vertex connectivity strength of non-connected vertices is 0. It is a characterization of the discrete spatial structure of the fracture network and has local characteristics. Therefore, formula (2) integrates the global characteristics and local characteristics, that is, it retains the relationship between the fracture vertices learned independently by the original Transformer model and embeds the inherent structural relationship of the fracture vertices, enabling the improved Transformer model to actively learn and update the vertex features of the fracture network.

[0078] Step 4.1.4. Finally, use the global pooling layer to pool all vertex features into a feature vector as the output of the depth map learning module. The feature vector is the high-level feature of the fracture network extracted. 。

[0079] Step 4.2. Establish a depth time series learning module for predicting the production dynamic data of production wells;

[0080] The input of the depth time series learning module is the high-level feature of the fracture network extracted in Step 4.1. The basic unit of this module is the encoder, which combines time series coding to maintain the time order information of the production dynamic data and predicts the production dynamic data of the production well corresponding to each time step. Specifically, the depth time series learning module uses the original Transformer model for production dynamic time series prediction. The encoder is used as the basic unit, and an encoder consists of multi-head attention operation, normalization operation, linear operation, addition operation, and time series coding. The encoder predicts the production dynamic data through the time series change of the fracture features extracted by the multi-head self-attention mechanism, and the time series coding information is obtained from the time order coding of the production dynamic data.

[0081] The Transformer model of the depth time series learning module consists of encoders connected in series. The specific working process is as follows:

[0082] Step 4.2.1. Copy the high-level feature of the fracture network to time steps as the input vector, and add it to the time series coding information as the input of the encoder;

[0083] Step 4.2.2. The multi-head self-attention mechanism inside the encoder performs feature transformation in the time series;

[0084] Step 4.2.3. Finally, use the fully connected layer as the output layer to output the production dynamic prediction value corresponding to each time step , ,where is the number of production prediction indicators; represents the number of production wells; represents the number of production dynamic indicator types; Indicates the number of prediction time steps for production dynamics.

[0085] Step 4.3: Combine the deep graph learning module and the deep time series learning module to obtain an end-to-end hybrid deep graph and time series learning model; the input data of this model is the graph representation data of the fracture network and the vertex connectivity strength matrix of the fracture network, and the output is the production dynamics prediction data.

[0086] Step 5: Use the divided training set to train the hybrid deep graph and time series learning model, and use the test set to verify the performance of the model after training;

[0087] Train the hybrid deep graph and time series learning model in combination with the training set divided in Step 1, optimize the model parameters to minimize the loss function, and use the mean absolute error (MAE) as the loss function of the model. The calculation formula of MAE is as follows:

[0088] (3);

[0089] In the formula, is the mean absolute error; represents the number of samples; and respectively represent the true value and the model prediction result of the th sample.

[0090] After training is completed, use the test set divided in Step 1 to verify the performance of the model. The verification index is the coefficient of determination, and the formula is as follows:

[0091] (4);

[0092] In the formula, is the coefficient of determination; represents the average value of the true values of

[0093] Step 6: Based on the trained model, perform fast prediction of the production dynamics of tight oil and gas reservoirs;

[0094] Output the hybrid deep graph and time series learning model that has been trained and has good test performance. Use the fracture distribution parameters of the tight oil and gas reservoir and the vertex connectivity strength matrix corresponding to the calculated fracture distribution as the model input, and the corresponding production dynamics data can be quickly predicted.

[0095] To prove the feasibility and superiority of the present invention, the following embodiments are given.

[0096] Figure 2It is a tight oil reservoir block with a multi-scale fracture distribution, which includes four production wells (numbered Production Well 1 to Production Well 4) and one injection well (numbered Injection Well 1). For Figure 2 random simulation, a multi-scale fracture distribution data sample containing 1,200 groups of different fracture distributions was constructed; three different fracture distributions were randomly selected as Figure 3 shown. Figure 3 Among them, (a), (b), and (c) represent the first group of fracture distribution, the second group of fracture distribution, and the third group of fracture distribution respectively. The embedded discrete fracture numerical simulation method was used to numerically simulate different fracture distributions to obtain a production performance data sample corresponding to 1,200 different fracture distributions. In this embodiment, the production performance data sample includes the daily oil production and daily water production of the four production wells. Figure 4 And Figure 5 are the comparison results of the daily water production and daily oil production of Production Well 1 corresponding to three different fracture distributions respectively. From Figure 4 and Figure 5 , it can be seen that different fracture distributions will lead to different production performances, that is, the daily water production and daily oil production of the production wells are different.

[0097] All fracture distributions and corresponding production data are combined into a sample library, and the sample library is divided according to a ratio of 5:1. Among them, 1,000 samples are used as the training set, and the other 200 samples are used as the test set. According to the process of Step 3, the graph representation data of the discrete fracture network is established, and the vertex connectivity strength matrix of the fracture network is calculated based on the graph representation data of the fractures;

[0098] Then, a deep graph learning module is established to extract the features of the current crack distribution through the deep graph learning module. First, through the embedding layer, an initial feature transformation is performed. The vertex attribute features and edge attribute features are mapped into vertex embedding features and edge embedding features through two different one-layer fully connected neural networks. The number of neurons in the fully connected layer is set to [128×1], the activation function is set to the Sigmoid function, and the embedding feature dimensions of vertices and edges are uniformly set to [128×1]. Second, for the fusion of edge features and vertex features, the information propagation aggregation model is used to fuse the edge features mapped by the embedding layer into the vertex features. The features of each edge are propagated to the two vertices corresponding to each edge, and the vector concatenation merge operation is used to merge the edge features into the vertex feature vector. A one-layer fully connected neural network is used to perform a mapping calculation on the merged vector at the vertex. The number of neurons in the fully connected layer is set to [128×1], the activation function is set to the ReLU function, and the updated vertex features with a dimension of [128×1] are obtained. Then, the vertex features after fusing the edge features and the obtained vertex connectivity strength matrix are input into the encoder calculation process of the Transformer. In this step, 2 layers of encoders are used to update the vertex features, and the vertex feature dimension is maintained as [128×1] during the update process. Finally, a vertex-based global pooling layer is used to perform a max pooling operation on all vertex features to obtain the high-level features of the crack network.

[0099] Construct a deep time series learning module to predict production dynamic data; the deep time series module consists of 2 layers of encoders and 1 layer of fully connected layers distributed over time steps. The input is the high-level features of the extracted crack network. Through the vector replication operation, it is replicated over time steps as the input parameters for each time step. Through 2 layers of Transformer encoders, feature transformation in the time series is performed. Finally, a fully connected layer wrapped by a time distribution layer is used as the output layer to output the production dynamic prediction values corresponding to each time step.

[0100] The deep graph learning module and the deep time series learning module are combined to construct a deep graph and time series hybrid learning model. The input of the model is the graph representation data of the crack network and the vertex connectivity strength matrix, and the output is the production dynamic data. The mean absolute error (MAE) is used as the loss function of the model. The model is trained by combining the divided training set to optimize the weight parameters of the model to minimize the loss function. Figure 6 It shows the change of the mean absolute error loss function with the iteration during the model training process. It can be seen that the loss function has tended to converge after 80 iterations.

[0101] After the training is completed, the divided test set is used to verify the performance of the model, and the coefficient of determination of the model on the test set is calculated. Figures 7 to 9The prediction effect of the trained model on a test sample is shown. The test sample is selected according to the median value of the determination coefficients of the prediction results of 200 test samples, so it is representative. Figure 7 It is the fracture network distribution of the test sample. For the test sample, production performance data prediction is carried out, and the production performance data (including daily water production and daily oil production) of one of the production wells are selected for display. Figure 8 and Figure 9 They are respectively the comparisons between the model prediction results and the numerical simulation results of the daily water production and daily oil production corresponding to Production Well 1. It can be seen that the model prediction results are consistent with the numerical simulation results. In particular, accurate predictions can be achieved at the water breakthrough time and the turning point of the decline in oil production.

[0102] Output the depth map and the time series hybrid learning model with good training completion and test performance. Using the fracture distribution parameters of the tight oil and gas reservoir and calculating the vertex connection strength corresponding to the fracture distribution as the model input, the corresponding production performance data can be quickly predicted.

[0103] Certainly, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development, characterized in that: The steps include: Step 1: Obtain multi-scale fracture distribution data samples and production simulation data samples, and build a sample library; Step 2, establishing a crack network diagram to represent the data; Step 3, calculating the vertex connectivity strength matrix of the fracture network based on the graph representation data; Step 4: Establish a depth map and time series hybrid learning model for rapid prediction of tight oil and gas reservoir production dynamics; the specific process is as follows: Step 4.1, establish a deep graph learning module for feature learning and extraction of graph representation data of the fracture network; In step 4.1, the deep graph learning module includes an embedding layer, an information propagation aggregation model, an improved Transformer model, and a global pooling layer; the specific working process is as follows: Step 4.1.1: Represent the data in the fracture network diagram Input embedding layer, which uses two independent fully connected layers to perform feature mapping on vertex features and edge features of graph data respectively; Step 4.1.2: Use the information propagation aggregation model to fuse the edge features mapped by the embedding layer into the vertex features; Specifically: the information propagation aggregation model will The vertices and The edges between vertices The feature is propagated to the edge corresponding to Vertices and Vertices , use vector concatenation and merging operations to merge edge features into vertex feature vectors, and use a layer of fully connected neural network to perform a mapping calculation on the merged vectors at the vertex to obtain the updated vertex features; Step 4.1.3: Based on the improved Transformer model, perform multi-level update and feature transformation on the vertex features after integrating the edge features; The improved Transformer model embeds the connectivity strength matrix into the learning and prediction process of the original Transformer model, using the vertex connectivity strength matrix The calculation process of the crack network structure encoding information participating in an attention operation is as follows: (2)； In the formula, Indicates Updated vertex features output by the layer encoder; Indicates The dimension of the vertex features after the layer encoder updates; is the softmax function; , and Represents the query matrix, key matrix, and value matrix in the attention operation; Indicates the total number of vertices; For the The dimension of the vertex features after the layer encoder updates; is a learnable weight matrix; is the transpose symbol; Step 4.1.4: Use the global pooling layer to pool all vertex features into a feature vector as the output of the deep graph learning module. The feature vector is the high-level feature of the extracted crack network. ; Step 4.2: Establish a deep time series learning module for production dynamic data prediction of production wells; Step 4.3, combining the deep graph learning module and the deep time series learning module to obtain an end-to-end deep graph and time series hybrid learning model; the input data of the deep graph and time series hybrid learning model is the graph representation data of the fracture network and the vertex connectivity strength matrix of the fracture network, and the output is the production dynamic prediction data; Step 5: Train the depth map and time series hybrid learning model, and verify the performance of the model after the training is completed; Step 6: Rapidly predict the production dynamics of tight oil and gas reservoirs based on the trained model.

2. The method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development according to claim 1 is characterized in that: The specific process of step 1 is as follows: Step 1.1, stochastically model the multi-scale fracture distribution of tight oil and gas reservoirs, and establish a multi-scale fracture distribution data sample containing several groups of different fracture distributions; Step 1.2, using an embedded discrete fracture numerical simulation method to perform numerical simulation calculations on different fracture distributions, and obtain corresponding production simulation data samples, where the production simulation data samples are production dynamic data of production wells, including daily oil production, daily gas production, daily water production, and bottom hole pressure; Step 1.3: The fracture distribution data samples and the corresponding production simulation data samples are combined into a sample library, and the sample library is divided into a training set and a test set in proportion.

3. The method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development according to claim 2, characterized in that: In step 2, a group of fractures are distributed as a fracture network, including multiple discrete fractures, and the parameters of each fracture include length, position, direction, aperture, porosity and permeability; a graphical representation of the fracture network is established based on the discrete fracture parameters, that is, , Represents the crack network diagram to represent the data, represents the set of vertices of the graph, Represents the edge set of the graph; the two endpoints of each single fracture are used as the vertices of the graph, and the position coordinates of the fracture endpoints are used as vertex features; the fractures are used as the edges of the graph, and the length, direction, aperture, porosity and permeability of the fractures are used as edge features; intersecting fractures are re-divided from the intersection.

4. The method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development according to claim 3 is characterized in that: In step 3, the connectivity strength of the fracture vertices is calculated based on the graph representation data of the fracture network, and a vertex connectivity strength matrix of the fracture network is constructed; the calculation formula for the connectivity strength between any two vertices in the matrix is: (1)； In the formula, Indicates The vertices and The connectivity strength between vertices corresponds to the first Line Elements of a column; Indicates The vertices and The connection distance between vertices.

5. The method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development according to claim 4, characterized in that: In step 4.2, the deep time series learning module uses the original Transformer model to perform production dynamic time series prediction; the specific working process is: Step 4.2.1: High-level features of the fracture network Copy to time steps as input vectors, and added with the time sequence coding information as the input of the encoder; the time sequence coding information is obtained by encoding the time sequence of the production dynamic data; Step 4.2.2: The multi-head self-attention mechanism inside the encoder performs feature transformation on the time series; Step 4.2.3: Use the fully connected layer as the output layer to output the production dynamic prediction value corresponding to each time step , ,in, To forecast the number of indicators for production; Indicates the number of producing wells; Indicates the number of production dynamic indicator types; The number of forecast time steps representing the production dynamics.

6. The method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development according to claim 5, characterized in that: In step 5, the depth map and time series hybrid learning model is trained in combination with the training set divided in step 1, the model parameters are optimized to minimize the loss function, and the mean absolute error is used as the loss function of the model. The calculation formula is as follows: （3）； In the formula, is the mean absolute error; Indicates the number of samples; and Respectively represent The true value of the sample and the model prediction result; After training is completed, the performance of the model is verified using the test set divided in step 1. The verification indicator is the determination coefficient, and the formula is as follows: （4）； In the formula, is the coefficient of determination; express The average of the true values ​​of the samples.

7. The method for rapid prediction of production performance of tight oil and gas reservoirs with multi-scale fracture development according to claim 6, characterized in that: In step 6, the depth map and time series hybrid learning model that has been trained and tested with good performance is output, and the fracture distribution parameters of the tight oil and gas reservoir and the vertex connectivity strength matrix corresponding to the calculated fracture distribution are used as model input to quickly predict the corresponding production dynamic data.

Images