Shale oil and gas yield dynamic prediction method based on GCN-GRU

Abstract

The invention discloses a shale oil and gas yield dynamic prediction method based on GCN-GRU, and relates to the technical field of oil and gas well yield prediction. According to relevant parameters of all fractured horizontal wells in a shale oil and gas well platform, in combination with the spatial relation of a volume transformation area and artificial fracture distribution characteristics, inter-well interference coefficients are calculated, after a multi-well adjacent matrix is constructed, production dynamic data are obtained according to the yield of all the fractured horizontal wells in the shale oil and gas well platform, and the yield of all the fractured horizontal wells in the shale oil and gas well platform is calculated. And distributing the production dynamic data to the training set and the test set, constructing a GCN-GRU model, predicting the yield of the fractured horizontal well by using the production dynamic data in the training set and the multi-well adjacent matrix training model, and verifying the accuracy of model prediction after training by using the production dynamic data in the test set. According to the method, high-precision prediction of the shale oil and gas yield under three-dimensional development and large-scale hydraulic fracturing is achieved by quantifying inter-well interference.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas well production prediction, and specifically to a GCN-GRU-based shale oil and gas production dynamic prediction method, in particular to a shale oil and gas production dynamic prediction method that integrates graph convolutional neural networks (GCN), gated recurrent neural networks (GRU), and stimulated reservoir volumes (SRVs) spatial relationships and fracture distribution characteristics. Background Art

[0002] With the increasing development of unconventional oil and gas resources, especially shale oil and gas, engineering technologies such as three-dimensional reservoir development, staged horizontal well fracturing, and volumetric fracturing have been widely applied to low-porosity and ultra-low permeability shale reservoirs, significantly increasing oil and gas production. However, due to the complex multi-scale fracture networks commonly developed in the volumetric areas created by three-dimensional development and large-scale hydraulic fracturing, inter-well pressure and production interference are significant, and the dynamic evolution of production is extremely complex, which seriously restricts the optimization of development plans and the improvement of recovery rates.

[0003] At present, shale oil and gas production prediction methods are mainly divided into four categories: the first is the decline analysis method, such as the Arps model and its variants, which have the advantages of few parameters and convenient calculation, but it is difficult to accurately characterize the impact of reservoir parameter changes on production; the second is the analytical method, which can obtain analytical solutions under ideal boundaries and simplified assumptions, and has both computational efficiency and physical significance, but the application of analytical methods in shale reservoirs faces significant challenges, mainly limited by the strong heterogeneity, multi-scale flow mechanism and complex physical and chemical processes of the reservoir; the third is the numerical simulation method, which characterizes the non-steady-state seepage process by numerically solving the seepage differential equations, and is suitable for complex system analysis of multi-physical field coupling problems, but has problems such as cumbersome modeling, parameter sensitivity, and high computational cost; the fourth is the machine learning method, which, with its powerful nonlinear modeling capabilities and good adaptability to unstructured data, is gradually becoming an important means of shale oil and gas production prediction.

[0004] Furthermore, existing production prediction and interference analysis methods generally struggle to systematically account for the distribution characteristics of artificial fractures, the spatial relationships between multiple wells, and the constraints of physical laws, resulting in limited model generalizability and prediction accuracy. Under the three-dimensional development model of shale oil and gas, the mechanisms of interwell pressure and production interference are extremely complex, making it difficult to quantitatively characterize the interference of fracture networks using traditional shale oil and gas production prediction methods.

[0005] Therefore, it is urgent to propose a dynamic prediction method for shale oil and gas production based on GCN-GRU, which can not only reflect the spatial relationship of multiple wells and the distribution characteristics of artificial fractures by integrating physical mechanisms, but also efficiently handle nonlinear and dynamic characteristics in the process of shale oil and gas production prediction. Summary of the Invention

[0006] The present invention aims to solve the above problems and provides a dynamic prediction method for shale oil and gas production based on GCN-GRU. By integrating graph convolutional neural networks, gated recurrent neural networks with the spatial relationship of volume transformation areas and the distribution characteristics of artificial fractures, shale oil and gas production is predicted, and inter-well interference is quantified, achieving high-precision prediction of shale oil and gas production in complex fracture networks.

[0007] To achieve the above object, the present invention adopts the following technical solutions: The dynamic prediction method for shale oil and gas production based on GCN-GRU integrates the spatial relationship of the volume transformation area of ​​the shale reservoir and the distribution characteristics of artificial fractures, including the following steps: Step 1: Select a shale oil and gas well platform, obtain the basic geological parameters and engineering parameters of each fractured horizontal well in the shale oil and gas well platform, and calculate the inter-well interference coefficient based on the spatial relationship of the volume transformation area and the fracture distribution characteristics of the shale reservoir; Step 2: Calculate the interference coefficients between the fractured horizontal wells in the shale oil and gas well platform and construct a multi-well adjacency matrix; Step 3: Based on the production of each fractured horizontal well in the shale oil and gas well platform, obtain production dynamic data, distribute the production dynamic data into training and test sets, and build a GCN-GRU model; Step 4: Input the production dynamic data and the multi-well adjacency matrix in the training set into the GCN-GRU model, train the GCN-GRU model, use the GCN model in the GCN-GRU model to extract spatial features, and then input them into the GRU model to extract temporal features. The GCN-GRU model is trained to predict the production of fractured horizontal wells, thereby obtaining a trained GCN-GRU model. Step 5: Input the dynamic production data in the test set into the trained GCN-GRU model for prediction, and use the dynamic production data in the test set to verify the accuracy of the prediction of the trained GCN-GRU model.

[0008] Preferably, the inter-well interference coefficient calculation process is: The first fracture of the No. 1 fractured horizontal well in the shale reservoir is connected to the wellbore as the coordinate origin, and a Cartesian coordinate system is established with the horizontal well direction as the x-axis, the fracture extension direction as the y-axis, and the vertical reservoir direction as the z-axis. The grid is divided in the shale reservoir to obtain the first fracture of the No. 1 fractured horizontal well. Crack No. The coordinates of the segment center are The coordinates of the junction between the first fracture and the wellbore in the No. 2 fractured horizontal well are , No. 2 fractured horizontal well Crack No. The coordinates of the segment center are , then the space straight line passing through the two center positions The point-wise formula is: ; Where, is the horizontal axis, is the vertical axis, is the vertical coordinate; 、 All are crack numbers; 、 All are segment numbers; is the fracture spacing of the No. 1 fractured horizontal well, is the fracture spacing of the No. 2 fractured horizontal well; The first fractured horizontal well The angle between the first fracture and the No. 1 hydraulic fracturing horizontal well, The second fractured horizontal well The angle between the first fracture and the No. 2 fractured horizontal well; is the perforation distance of the No. 1 fractured horizontal well, is the perforation distance of the No. 2 fractured horizontal well; is the well spacing in the x-axis direction; is the well spacing in the y-axis direction; is the well spacing in the z-axis direction.

[0009] Preferably, the volume transformation area of ​​the No. 1 fractured horizontal well and the No. 2 fractured horizontal well is determined, and the half length of the volume transformation area of ​​the No. 1 fractured horizontal well is obtained. Half length of the volume transformation area of ​​No. 2 fractured horizontal well , when the volume transformation areas of the two do not intersect, , when the volume transformation areas of the two intersect, ; The plane where the volume transformation area of ​​the No. 1 fractured horizontal well is located is represented as The plane where the volume transformation area of ​​No. 2 fractured horizontal well is located is represented as ,flat The expression is , plane expression for , space straight line With plane The intersection point is , and plane The intersection point is ; In order to simplify the calculation, the space straight line On point The coordinates are simplified to ,point The coordinates are simplified to ,in, for point The simplified horizontal axis is , for point The simplified vertical axis, , for point The simplified horizontal axis is , for point The simplified vertical axis, ; When the SRV areas of two fractured horizontal wells do not intersect, the intersection point The coordinates are , intersection point The coordinates are ; When the SRV regions of two fractured horizontal wells intersect, the intersection point The coordinates are , intersection point The coordinates are .

[0010] Preferably, the interference coefficient of the No. 1 fractured horizontal well on the No. 2 fractured horizontal well is for: ; in, ; ; Where, is the total number of fractures in the No. 1 fractured horizontal well; is the total number of fractures in the No. 2 fractured horizontal well; The first fractured horizontal well The fractures in the No. 2 fractured horizontal well Interference coefficient of cracks; The second fractured horizontal well Half the length of a crack; The first fractured horizontal well The fractures in the No. 2 fractured horizontal well Crack No. Interference coefficient of the segment; The first fractured horizontal well Half the length of a crack; The first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment; When the SRV areas of the two fractured horizontal wells do not intersect, the first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment The calculation formula is: ; in, ; ; ; Where, for point with dot the distance between them; for point Intersection point distance; for point Intersection point distance; The intersection point Intersection point distance; is the permeability of the volume transformation area of ​​No. 1 fractured horizontal well, is the permeability of the volume stimulation area of ​​the No. 2 fractured horizontal well; is the matrix permeability; When the SRV areas of two fractured horizontal wells intersect, the first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment The calculation formula is: ; in, ; ; ; Where, is the permeability of the SRV intersection area.

[0011] Preferably, the GCN-GRU model includes a GCN model and a GRU model connected in sequence; The GCN model is used to update its own nodes based on the features of the surrounding nodes aggregated by the multi-well adjacency matrix and the feature matrix. The GCN model is provided with a graph convolutional network. The propagation mode between the structural layers in the graph convolutional network is: ; Where, is the layer number, The first The feature input of the layer, The first Feature input of the layer; is the sigmoid function; is the degree matrix; is the self-connected adjacency matrix, , is the adjacency matrix; is the identity matrix; The GRU model is built based on the long short-term memory neural network and is used to extract the temporal features of the GCN model output features, including the update gate and reset gate The GRU model integrates the forget gate and input gate of the long short-term memory neural network into an update gate, and uses a gating mechanism to model time series data to capture long-term dependencies and alleviate the gradient disappearance problem.

[0012] Preferably, when the GRU model is used to predict When the production of the shale reservoir is at a certain moment, the GRU model first inputs the current production of the shale reservoir And the hidden state of the GRU model at the previous moment Get Update Gate and reset gate The gating signal of the update gate is calculated and reset gate And the Sigmoid function is used to compress it to the [0,1] interval, where the reset gate Used to control the previous state to the current candidate state If the reset gate If it is close to 0, the historical information is ignored and the candidate state is generated based only on the current input; the update gate Used to dynamically balance the ratio of old state to new candidate state and decide which part to keep and absorption part , and output the hidden state at the current moment ; The calculation formula of the GRU model is: ; Where, To reset the gate, For the update door; for The input of the GRU model at this moment; for The output of the GRU model at this moment, for The output of the GRU model at this moment, is the updated unit status; To reset the gate weight matrix, is the weight matrix of the update gate, is the weight matrix of the current memory content; is the bias matrix of the reset gate, is the bias matrix of the update gate, is the bias matrix of the current memory content; is the Hadamard product operator; is the hyperbolic tangent function.

[0013] Preferably, in step 4, the multi-well adjacency matrix constructed by the production dynamic data in the test set and the interference coefficient between each well is input into the GCN-GRU model, and the GCN-GRU model is trained to predict the production of the fractured horizontal well at the next moment based on the interference coefficient between each fractured horizontal well and the production of the fractured horizontal well at the current moment, and the prediction result of the fractured horizontal well production by the GCN-GRU model is compared with the original production data of the fractured horizontal well in the production dynamic data, and the root mean square error of the GCN-GRU model is calculated each time during training. , mean absolute error , prediction accuracy and mean relative error , when the root mean square error of the GCN-GRU model , mean absolute error , prediction accuracy and mean relative error When the preset accuracy value is reached, the training of the GCN-GRU model is stopped to obtain the trained GCN-GRU model; The trained GCN-GRU model can predict the production at a specified future time based on the production of the fractured horizontal well at the current time, and obtain the predicted value of the production of each fractured horizontal well in the shale oil and gas well platform.

[0014] Preferably, the root mean square error The calculation formula is: ; The mean absolute error The calculation formula is: ; The prediction accuracy The calculation formula is: ; The mean relative error The calculation formula is: ; Where, is the actual production of fractured horizontal wells; The predicted production value of the fractured horizontal well output by the GCN-GRU model; The serial number of the production data of the fractured horizontal well; is the total number of production data for fractured horizontal wells.

[0015] The beneficial technical effects brought about by the present invention are: The present invention proposes a dynamic prediction method for shale oil and gas production based on GCN-GRU. The graph convolutional neural network, gated recurrent neural network, spatial relationship of volume transformation area and distribution characteristics of artificial fractures are integrated to construct a GCN-GRU model with physical constraints. The model not only incorporates the interference relationship between multiple wells, but also can efficiently handle nonlinear and dynamic characteristics. By quantifying the interference between wells, fully considering the spatial relationship of volume transformation area and distribution characteristics of artificial fractures, and considering the interference between each fractured horizontal well, the high-precision prediction of shale oil and gas production is achieved, laying the foundation for the intelligent development of unconventional oil and gas reservoirs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the dynamic prediction method for shale oil and gas production based on GCN-GRU of the present invention.

[0017] Figure 2 Schematic diagram of multi-fractured horizontal wells in a shale oil and gas well platform.

[0018] Figure 3 Schematic diagram of the non-intersecting volume stimulation areas of two fractured horizontal wells.

[0019] Figure 4 Schematic diagram of the intersection of the volume stimulation areas of two fractured horizontal wells.

[0020] Figure 5 Schematic diagram of production dynamic data for a shale oil and gas well pad.

[0021] Figure 6 A comparison of the prediction results of the GCN-GRU model trained using different methods.

[0022] In the figure, W1 is the No. 1 fracturing horizontal well W1, W2 is the No. 2 fracturing horizontal well, W3 is the No. 3 fracturing horizontal well, W4 is the No. 4 fracturing horizontal well, W5 is the No. 5 fracturing horizontal well, and W6 is the No. 6 fracturing horizontal well. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0024] Taking a shale oil and gas well platform in a shale reservoir research area as an example, a shale oil and gas production dynamic prediction method based on GCN-GRU and integrating the spatial relationship of shale reservoir volume transformation area and artificial fracture distribution characteristics proposed in this invention is adopted. Figure 1 As shown, the following steps are included: Step 1: Select a shale oil and gas well platform, where multiple fractured horizontal wells are set up, such as Figure 2 As shown, in this embodiment, the shale oil and gas well platform is provided with 6 fractured horizontal wells, namely, fractured horizontal well No. 1 W1, fractured horizontal well No. 2 W2, fractured horizontal well No. 3 W3, fractured horizontal well No. 4 W4, fractured horizontal well No. 5 W5 and fractured horizontal well No. 6 W6. The basic geological parameters and engineering parameters of each fractured horizontal well in the shale oil and gas well platform are obtained. The basic geological parameters of the fractured horizontal wells include the well location and the permeability of the unreformed area. The engineering parameters of the fractured horizontal wells include the angle between the fracture and the horizontal well, the half-length of the fracture, the first-section perforation distance of the fractured horizontal well, the permeability of the volume reformed area, the fracture spacing, the number of well fracture clusters, the fracture distribution, and the fracture aperture, as shown in Table 1.

[0025] Table 1 Engineering parameters of fractured horizontal wells .

[0026] According to the volume transformation area and the spatial distribution of the fracture network of the shale reservoir, the inter-well interference coefficient is calculated. The calculation process of the inter-well interference coefficient is as follows: The first fracture of the No. 1 horizontal well in the shale reservoir is connected to the wellbore as the coordinate origin, and a Cartesian coordinate system is established with the horizontal well direction as the x-axis, the fracture extension direction as the y-axis, and the vertical reservoir direction as the z-axis. The fracture is discretized into grids with a grid length of one meter, a grid width of the fracture width, and a grid height of the fracture height. Each fracture occupies grids, and obtain the first Crack No. The coordinates of the segment center are The coordinates of the junction between the first fracture and the wellbore in the No. 2 fractured horizontal well are , No. 2 fractured horizontal well Crack No. The coordinates of the segment center are , then the space straight line passing through the two center positions The point-wise formula is: ; Where, is the horizontal axis, is the vertical axis, is the vertical coordinate; 、 All are crack numbers; 、 All are segment numbers; is the fracture spacing of the No. 1 fractured horizontal well, is the fracture spacing of the No. 2 fractured horizontal well; The first fractured horizontal well The angle between the first fracture and the No. 1 hydraulic fracturing horizontal well, The second fractured horizontal well The angle between the first fracture and the No. 2 fractured horizontal well; is the perforation distance of the No. 1 fractured horizontal well, is the perforation distance of the No. 2 fractured horizontal well; is the well spacing in the x-axis direction; is the well spacing in the y-axis direction; is the well spacing in the z-axis direction.

[0027] Furthermore, the volume transformation area of ​​the No. 1 fractured horizontal well and the No. 2 fractured horizontal well is determined, and the half length of the volume transformation area of ​​the No. 1 fractured horizontal well is obtained. Half length of the volume transformation area of ​​No. 2 fractured horizontal well , where the half-length of the volume transformation region is half the length of the volume transformation region; when the volume transformation regions of the two do not intersect, such as Figure 3 As shown, at this time , when the volume transformation areas of the two intersect, such as Figure 4 As shown, at this time .

[0028] The plane where the volume transformation area of ​​the No. 1 fractured horizontal well is located is represented as The plane where the volume transformation area of ​​No. 2 fractured horizontal well is located is represented as ,flat The expression is , plane expression for , space straight line With plane The intersection point is , and plane The intersection point is .

[0029] In order to simplify the calculation, the space straight line On point The coordinates are simplified to ,point The coordinates are simplified to ,in, for point The simplified horizontal axis is , for point The simplified vertical axis, , for point The simplified horizontal axis is , for point The simplified vertical axis, ; When the SRV areas of two fractured horizontal wells do not intersect, the intersection point The coordinates are , intersection point The coordinates are ; When the SRV regions of two fractured horizontal wells intersect, the intersection point The coordinates are , intersection point The coordinates are .

[0030] Furthermore, the interference coefficient of the No. 1 fractured horizontal well on the No. 2 fractured horizontal well is for: ; in, ; ; Where, is the total number of fractures in the No. 1 fractured horizontal well; is the total number of fractures in the No. 2 fractured horizontal well; The first fractured horizontal well The fractures in the No. 2 fractured horizontal well Interference coefficient of cracks; The second fractured horizontal well Half the length of a crack; The first fractured horizontal well The fractures in the No. 2 fractured horizontal well Crack No. Interference coefficient of the segment; The first fractured horizontal well Half the length of a crack; The first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. The interference coefficient of the segment.

[0031] When the SRV areas of the two fractured horizontal wells do not intersect, the first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment The calculation formula is: ; in, ; ; ; Where, for point with dot the distance between them; for point Intersection point distance; for point Intersection point distance; The intersection point Intersection point distance; is the permeability of the volume transformation area of ​​No. 1 fractured horizontal well, is the permeability of the volume stimulation area of ​​the No. 2 fractured horizontal well; is the matrix permeability.

[0032] When the SRV areas of two fractured horizontal wells intersect, the first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment The calculation formula is: ; in, ; ; ; Where, is the permeability of the SRV intersection area.

[0033] Step 2: Calculate the interference coefficients between the horizontal wells in the shale oil and gas well platform and construct a multi-well adjacency matrix. The multi-well adjacency matrix is ​​expressed as: ; Where, The first shale oil and gas well in Taichung The fractured horizontal well Fractured horizontal well Interference coefficient of fractured horizontal wells.

[0034] The multi-well adjacency matrix constructed in this embodiment is: .

[0035] In order to highlight the superiority of the method of the present invention, a common method is used to construct a 0-1 matrix, and the following is obtained: .

[0036] Step 3: According to the production of each fractured horizontal well in the shale oil and gas well platform, obtain production dynamic data, such as Figure 5 As shown, in this embodiment, the ratio of the number of dynamic data produced in the training set and the test set is 8:2, and a GCN-GRU model is constructed.

[0037] In this embodiment, the GCN-GRU model includes a GCN model and a GRU model connected in sequence.

[0038] The GCN model consists of an input layer, a hidden layer, and an output layer. The hidden layer is the core of the GCN model. The hidden layer includes multiple graph convolution layers. Each graph convolution layer is constructed by performing an adjacency matrix With the characteristic matrix Convolution operation is performed to update the node feature representation. The propagation mode between the structural layers in the graph convolution network is: ; Where, is the layer number, The first The feature input of the layer, The first Feature input of the layer; is the sigmoid function; is the degree matrix; is the self-connected adjacency matrix, , is the adjacency matrix; is the identity matrix.

[0039] The GRU model is built based on the long short-term memory neural network (LSTM), which is a variant of the long short-term memory neural network and is used to extract the temporal features of the GCN model output features, including the update gate and reset gate , which not only solves the vanishing gradient problem but also preserves time series features. The GRU model improves training speed by integrating the forget gate and input gate of the long short-term memory neural network into an update gate. It also uses a gating mechanism to model time series data, effectively capturing long-term dependencies and alleviating the vanishing gradient problem.

[0040] When the GRU model is used to predict When the production of the shale reservoir is at a certain moment, the GRU model first inputs the current production of the shale reservoir And the hidden state of the GRU model at the previous moment Get Update Gate and reset gate The gating signal of the update gate is calculated and reset gate And the Sigmoid function is used to compress it to the [0,1] interval, where the reset gate Used to control the previous state to the current candidate state If the reset gate If it is close to 0, the historical information is ignored and the candidate state is generated based only on the current input; the update gate Used to dynamically balance the ratio of old state to new candidate state and decide which part to keep and absorption part , and output the hidden state at the current moment .

[0041] Furthermore, the calculation formula of the GRU model is: ; Where, To reset the gate, For the update door; for The input of the GRU model at this moment; for The output of the GRU model at this moment, for The output of the GRU model at this moment, is the updated unit status; To reset the gate weight matrix, is the weight matrix of the update gate, is the weight matrix of the current memory content; is the bias matrix of the reset gate, is the bias matrix of the update gate, is the bias matrix of the current memory content; is the Hadamard product operator; is the hyperbolic tangent function.

[0042] Step 4: Input the production dynamic data and multi-well adjacency matrix in the training set into the GCN-GRU model, train the GCN-GRU model, use the GCN model in the GCN-GRU model to extract spatial features and then input them into the GRU model to extract temporal features, train the GCN-GRU model to predict the production of fractured horizontal wells, and obtain the trained GCN-GRU model.

[0043] In this embodiment, the multi-well adjacency matrix constructed by the production dynamic data in the test set and the interference coefficient between each well is input into the GCN-GRU model. The GCN-GRU model is trained to predict the production of the fractured horizontal well at the next moment based on the adjacency matrix and the production of the fractured horizontal well at the current moment. The prediction result of the fractured horizontal well production by the GCN-GRU model is compared with the original production data of the fractured horizontal well in the production dynamic data. The root mean square error of the GCN-GRU model is calculated each time during training. , mean absolute error , prediction accuracy and mean relative error , the root mean square error The calculation formula is: ; The mean absolute error The calculation formula is: ; The prediction accuracy The calculation formula is: ; The mean relative error The calculation formula is: ; Where, is the actual production of fractured horizontal wells; The predicted production value of the fractured horizontal well output by the GCN-GRU model; The serial number of the production data of the fractured horizontal well; is the total number of production data for fractured horizontal wells.

[0044] Calculate the root mean square error of the GCN-GRU model , mean absolute error , prediction accuracy and mean relative error , the root mean square error , mean absolute error and mean relative error The smaller the value, the better the prediction accuracy The closer the value is to 1, the better the prediction effect of the GCN-GRU model.

[0045] In this embodiment, when the root mean square error of the GCN-GRU model , mean absolute error , prediction accuracy and mean relative error When the preset accuracy value is reached, the training of the GCN-GRU model is stopped to obtain the trained GCN-GRU model.

[0046] The trained GCN-GRU model can predict the production at a specified future time based on the production of the fractured horizontal well at the current time, and obtain the predicted value of the production of each fractured horizontal well in the shale oil and gas well platform.

[0047] Step 5: Input the dynamic production data in the test set into the trained GCN-GRU model for prediction, and use the dynamic production data in the test set to verify the accuracy of the prediction of the trained GCN-GRU model.

[0048] Furthermore, the GCN-GRU models trained by the conventional method and the method of the present invention are compared. The conventional method uses a 0-1 matrix to train the GCN-GRU model, while the method of the present invention uses a multi-well adjacency matrix of the fused shale reservoir volume transformation area to train the GCN-GRU model. The root mean square error of the GCN-GRU models obtained by the two methods is compared. , mean absolute error , prediction accuracy and mean relative error , as shown in Table 2.

[0049] Table 2 Comparison of evaluation indicators of the two methods .

[0050] Comparing the prediction results of the GCN-GRU model trained by the conventional method and the method of the present invention, Figure 6 As shown in the figure, after comparison, it is found that the prediction error of the GCN-GRU model trained by the method of the present invention is smaller than that of the GCN-GRU model trained by the commonly used method. That is, the method of the present invention quantitatively characterizes the interference between fractured horizontal wells by integrating the spatial relationship of the shale reservoir volume transformation area and the distribution characteristics of artificial fractures, thereby achieving accurate prediction of shale oil and gas production, which is beneficial to the optimization of shale reservoir development plans and the improvement of recovery rates.

[0051] Of course, 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 technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. A dynamic prediction method for shale oil and gas production based on GCN-GRU, characterized by: The integration of the spatial relationship of the volume transformation area of ​​the shale reservoir and the distribution characteristics of artificial fractures includes the following steps: Step 1: Select a shale oil and gas well platform, obtain the basic geological parameters and engineering parameters of each fractured horizontal well in the shale oil and gas well platform, and calculate the inter-well interference coefficient based on the spatial relationship of the volume transformation area and the fracture distribution characteristics of the shale reservoir; Step 2: Calculate the interference coefficients between the fractured horizontal wells in the shale oil and gas well platform and construct a multi-well adjacency matrix; Step 3: Based on the production of each fractured horizontal well in the shale oil and gas well platform, obtain production dynamic data, distribute the production dynamic data into training and test sets, and build a GCN-GRU model; Step 4: Input the production dynamic data and the multi-well adjacency matrix in the training set into the GCN-GRU model, train the GCN-GRU model, use the GCN model in the GCN-GRU model to extract spatial features, and then input them into the GRU model to extract temporal features. The GCN-GRU model is trained to predict the production of fractured horizontal wells, thereby obtaining a trained GCN-GRU model. Step 5: Input the dynamic production data in the test set into the trained GCN-GRU model for prediction, and use the dynamic production data in the test set to verify the accuracy of the prediction of the trained GCN-GRU model.

2. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 1, characterized in that: The calculation process of the inter-well interference coefficient is: The first fracture of the No. 1 fractured horizontal well in the shale reservoir is connected to the wellbore as the coordinate origin, and a Cartesian coordinate system is established with the horizontal well direction as the x-axis, the fracture extension direction as the y-axis, and the vertical reservoir direction as the z-axis. The grid is divided in the shale reservoir to obtain the first fracture of the No. 1 fractured horizontal well. Crack No. The coordinates of the segment center are The coordinates of the junction between the first fracture and the wellbore in the No. 2 fractured horizontal well are , No. 2 fractured horizontal well Crack No. The coordinates of the segment center are , then the space straight line passing through the two center positions The point-wise formula is: ； Where, is the horizontal axis, is the vertical axis, is the vertical coordinate; 、 All are crack numbers; 、 All are segment numbers; is the fracture spacing of the No. 1 fractured horizontal well, is the fracture spacing of the No. 2 fractured horizontal well; The first fractured horizontal well The angle between the first fracture and the No. 1 hydraulic fracturing horizontal well, The second fractured horizontal well The angle between the first fracture and the No. 2 fractured horizontal well; is the perforation distance of the No. 1 fractured horizontal well, is the perforation distance of the No. 2 fractured horizontal well; is the well spacing in the x-axis direction; is the well spacing in the y-axis direction; is the well spacing in the z-axis direction.

3. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 2, characterized in that: Determine the volume transformation area of ​​the No. 1 and No. 2 fractured horizontal wells, and obtain the half-length of the volume transformation area of ​​the No. 1 fractured horizontal well Half length of the volume transformation area of ​​No. 2 fractured horizontal well , when the volume transformation areas of the two do not intersect, , when the volume transformation areas of the two intersect, ; The plane where the volume transformation area of ​​the No. 1 fractured horizontal well is located is represented as The plane where the volume transformation area of ​​No. 2 fractured horizontal well is located is represented as ,flat The expression is , plane expression for , space straight line With plane The intersection point is , and plane The intersection point is ; In order to simplify the calculation, the space straight line On point The coordinates are simplified to ,point The coordinates are simplified to ,in, for point The simplified horizontal axis is , for point The simplified vertical axis, , for point The simplified horizontal axis is , for point The simplified vertical axis, ; When the SRV areas of two fractured horizontal wells do not intersect, the intersection point The coordinates are , intersection point The coordinates are ; When the SRV regions of two fractured horizontal wells intersect, the intersection point The coordinates are , intersection point The coordinates are .

4. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 3, characterized in that: The interference coefficient of the No. 1 fractured horizontal well on the No. 2 fractured horizontal well for: ； in, ； ； Where, is the total number of fractures in the No. 1 fractured horizontal well; is the total number of fractures in the No. 2 fractured horizontal well; The first fractured horizontal well The fractures in the No. 2 fractured horizontal well Interference coefficient of cracks; The second fractured horizontal well Half the length of a crack; The first fractured horizontal well The fractures in the No. 2 fractured horizontal well Crack No. Interference coefficient of the segment; The first fractured horizontal well Half the length of a crack; The first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment; When the SRV areas of the two fractured horizontal wells do not intersect, the SRV area of ​​the first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment The calculation formula is: ； in, ； ； ； Where, for point with dot the distance between them; for point Intersection point distance; for point Intersection point distance; The intersection point Intersection point distance; is the permeability of the volume transformation area of ​​No. 1 fractured horizontal well, is the permeability of the volume stimulation area of ​​the No. 2 fractured horizontal well; is the matrix permeability; When the SRV areas of two fractured horizontal wells intersect, the first fractured horizontal well Crack No. Section 2 and No. 2 fractured horizontal well Crack No. Interference coefficient of the segment The calculation formula is: ； in, ； ； ； Where, is the permeability of the SRV intersection area.

5. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 1, characterized in that: The GCN-GRU model includes a GCN model and a GRU model connected in sequence; The GCN model is used to update its own nodes based on the features of the surrounding nodes aggregated by the multi-well adjacency matrix and the feature matrix. The GCN model is provided with a graph convolutional network. The propagation mode between the structural layers in the graph convolutional network is: ； Where, is the layer number, The first The feature input of the layer, The first Feature input of the layer; is the sigmoid function; is the degree matrix; is the self-connected adjacency matrix, , is the adjacency matrix; is the identity matrix; The GRU model is built based on the long short-term memory neural network and is used to extract the temporal features of the GCN model output features, including the update gate and reset gate The GRU model integrates the forget gate and input gate of the long short-term memory neural network into an update gate, and uses a gating mechanism to model time series data to capture long-term dependencies and alleviate the gradient disappearance problem.

6. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 5, characterized in that: When the GRU model is used to predict When the production of the shale reservoir is at a certain moment, the GRU model first inputs the current production of the shale reservoir And the hidden state of the GRU model at the previous moment Get Update Gate and reset gate The gating signal of the update gate is calculated and reset gate And the Sigmoid function is used to compress it to the [0,1] interval, where the reset gate Used to control the previous state to the current candidate state If the reset gate If it is close to 0, the historical information is ignored and the candidate state is generated based only on the current input; the update gate Used to dynamically balance the ratio of old state to new candidate state and decide which part to keep and absorption part , and output the hidden state at the current moment ; The calculation formula of the GRU model is: ； Where, To reset the gate, For the update door; for The input of the GRU model at this moment; for The output of the GRU model at this moment, for The output of the GRU model at this moment, is the updated unit status; To reset the gate weight matrix, is the weight matrix of the update gate, is the weight matrix of the current memory content; is the bias matrix of the reset gate, is the bias matrix of the update gate, is the bias matrix of the current memory content; is the Hadamard product operator; is the hyperbolic tangent function.

7. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 6, characterized in that: In step 4, the multi-well adjacency matrix constructed by the production dynamic data in the test set and the interference coefficient between each well is input into the GCN-GRU model, and the GCN-GRU model is trained to predict the production of the fractured horizontal well at the next moment based on the interference coefficient between each fractured horizontal well and the production of the fractured horizontal well at the current moment, and the prediction result of the fractured horizontal well production by the GCN-GRU model is compared with the original production data of the fractured horizontal well in the production dynamic data, and the root mean square error of the GCN-GRU model is calculated each time during training. , mean absolute error , prediction accuracy and mean relative error , when the root mean square error of the GCN-GRU model , mean absolute error , prediction accuracy and mean relative error When the preset accuracy value is reached, the training of the GCN-GRU model is stopped to obtain the trained GCN-GRU model; The trained GCN-GRU model can predict the production at a specified future time based on the production of the fractured horizontal well at the current time, and obtain the predicted value of the production of each fractured horizontal well in the shale oil and gas well platform.

8. The method for dynamic prediction of shale oil and gas production based on GCN-GRU according to claim 7, characterized in that: The root mean square error The calculation formula is: ； The mean absolute error The calculation formula is: ； The prediction accuracy The calculation formula is: ； The mean relative error The calculation formula is: ； Where, is the actual production of fractured horizontal wells; The predicted production value of the fractured horizontal well output by the GCN-GRU model; The serial number of the production data of the fractured horizontal well; is the total number of production data for fractured horizontal wells.

Images