Physical model driven neural network fracturing process parameter optimization method and optimization device

Abstract

The invention relates to the technical field of petroleum and natural gas exploitation, in particular to a neural network fracturing process parameter optimization method and device driven by a physical model, and the method comprises the steps: carrying out the orthogonal experiment design through main fracturing process parameters which affect the fracturing transformation effect, and obtaining a fracturing construction scheme of different fracturing process parameter combinations; and inputting different fracturing process parameter combination construction schemes into the neural network productivity prediction model, and outputting corresponding productivity prediction results. According to the method, multiple fracturing construction parameters can be rapidly optimized at the same time, and time cost brought by simulation is greatly saved; secondly, the neural network is driven by a physical model, limitation caused by oil reservoir complexity is avoided, and the accuracy of a neural network productivity prediction model is improved; and 3, after the neural network productivity prediction model is established, the method can be suitable for different oil reservoirs and has wide applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil and gas exploitation, and is a physical model driven neural network fracturing process parameter optimization method and device. BACKGROUND

[0002] Fracturing is a key technology for the effective development of unconventional reservoirs. With the continuous deepening of unconventional oil and gas exploration and development, the optimization of fracturing process parameters has attracted widespread attention. Researchers at home and abroad have explored the influence of fracturing process parameters on productivity from different angles and have proposed a series of optimization methods and technologies.

[0003] For unconventional reservoirs, the cost of fracturing development is very high, and a reasonable development plan can obtain more benefits with less cost. Fracturing process parameter optimization is particularly important for improving the efficiency of exploitation. Currently, two methods are mainly used, numerical simulation and machine learning. Numerical simulation establishes a mathematical model of hydraulic fracturing to simulate the fracture morphology and productivity under different parameters and optimize the fracturing parameters. Machine learning establishes a prediction model by processing a large amount of experimental data to predict the impact of different fracturing parameters on productivity, thereby achieving parameter optimization. However, both methods have drawbacks. Numerical simulation uses computer power to numerically solve its production formula and is widely used. If a large number of models are simulated, the time cost is high. Machine learning uses data to model from the perspective of data itself and is a data-driven modeling method. It is an effective method when the seepage mechanism is not clear. Machine learning models are very fast when repeatedly called, but they are driven by a large amount of actual reservoir production data and are limited by the complexity and uncertainty of the reservoir itself.

[0004] Fracturing process parameter optimization is still a challenging problem. Therefore, there is an urgent need to develop an efficient fracturing parameter optimization method. SUMMARY

[0005] The present application provides a physical model driven neural network fracturing process parameter optimization method and device, which overcomes the shortcomings of the prior art and effectively solves the problems of high time cost in the optimization process and the need to improve the accuracy of productivity prediction in the existing fracturing process parameter optimization method. It realizes the rapid optimization of fracturing process parameters for different block reservoirs.

[0006] The application considers simplifying the number of fracturing parameter scheme groups through orthogonal experiments, establishing a physical model to predict the productivity of different fracturing process parameter schemes, and using the experimental parameters and prediction results of the physical model to drive machine learning to obtain a neural network productivity prediction model to guide fracturing process parameter optimization.

[0007] One of the technical solutions of the application is realized by the following measures: a neural network fracturing process parameter optimization method driven by a physical model, comprising:

[0008] Orthogonal experiments are designed using the main fracturing process parameters affecting the fracturing reconstruction effect to obtain fracturing construction schemes with different fracturing process parameter combinations.

[0009] The fracturing construction schemes with different fracturing process parameter combinations are input into a neural network productivity prediction model, and corresponding productivity prediction results are output. The optimal fracturing process parameter combination construction scheme is determined by comparing each productivity prediction result. The neural network productivity prediction model is obtained by machine learning using a plurality of training samples, wherein the training samples include fracturing construction schemes with different fracturing process parameter combinations and corresponding productivity calculation results.

[0010] The following is a further optimization or / and improvement of one of the above technical solutions of the application:

[0011] The above orthogonal experiments are designed using the main fracturing process parameters affecting the fracturing reconstruction effect to obtain fracturing construction schemes with different fracturing process parameter combinations, which are as follows:

[0012] The main fracturing process parameters affecting the fracturing reconstruction effect include porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height, and conductivity.

[0013] Through orthogonal experiment design, the main fracturing process parameters are used as experimental factors, each experimental factor is divided into 2 to 3 levels, and an orthogonal experiment table is established to maximize the number of experiments and obtain the fracturing construction scheme after orthogonal experiment design, i.e., the fracturing construction scheme with different fracturing process parameter combinations.

[0014] The productivity (i.e., production) calculation results corresponding to the above fracturing construction schemes with different fracturing process parameter combinations are obtained using a productivity prediction physical model. The productivity prediction physical model is a productivity prediction physical model established based on percolation theory. The productivity prediction physical model is a low-permeability reservoir unsteady percolation equation considering the starting pressure gradient, which is as follows:

[0015] Low-permeability reservoir unsteady percolation equation considering starting pressure gradient:

[0016]

[0017] The initial and boundary conditions of the unstable seepage equation of low-permeability oil reservoirs considering the starting pressure gradient are as follows:

[0018]

[0019] The approximate solution of the unsteady seepage obtained by the integral method is as follows:

[0020]

[0021] The coefficients a0, a1, a2,..., a n+1 in the formula (c) are solved by bringing the formula (c) into the equation group (b), and the production formula is obtained:

[0022]

[0023] In the formula (a) to the formula (d), q is the production, m 3 / s; p is the formation pressure, m; p i 0 is the original formation pressure, Pa; μ is the oil viscosity, Pa·s; K is the formation permeability, mD; r e is the supply boundary radius, m; h is the formation thickness, m; r is the well radius, m; G is the starting pressure gradient, Pa / m; r w is the drainage radius, m; r e (t) is the supply boundary radius at t, m.

[0024] The above starting pressure gradient is calculated according to the following formula:

[0025]

[0026] In the formula, G is the starting pressure gradient, Pa / m; p0 is the core oil production end pressure, Pa; p s is the core closed end pressure, Pa; L is the core length, m; x is the coordinate along the core length direction, m.

[0027] The above neural network production prediction model is constructed according to the following method, comprising:

[0028] The combination of different fracturing process parameters and the corresponding production calculation results are used as a sample set, and the sample set is divided into a training set and a test set;

[0029] The neural network model is trained by using the training set to obtain a neural network initial capacity prediction model, and the neural network initial capacity prediction model is tested by using the test set, if the model accuracy is met, the neural network initial capacity prediction model is output as the optimal neural network capacity prediction model, otherwise the neural network initial capacity prediction model parameters are adjusted, and the training set is used to train the neural network model again.

[0030] The neural network model can adopt a BP neural network model.

[0031] In order to achieve the purpose of accurate prediction, a genetic algorithm can be used to optimize the neural network model to obtain more accurate initial values and threshold values, so as to obtain a more optimized neural network capacity prediction model, and finally output the capacity prediction result corresponding to the optimal fracturing process parameter combination construction scheme.

[0032] The model accuracy is the minimum mean square error (MSE) of the test result and the capacity calculation result of the capacity prediction physical model.

[0033] The second technical solution of the present application is realized by the following measures: one is an optimization device of the physical model driven neural network fracturing process parameter optimization method of the first technical solution, comprising:

[0034] The design unit: using the main fracturing process parameters affecting the fracturing reconstruction effect, orthogonal experiment design is carried out, and the fracturing construction scheme of different fracturing process parameter combinations is obtained;

[0035] The prediction unit: input the different fracturing process parameter combination construction scheme into the neural network capacity prediction model, and output the corresponding capacity prediction result, and determine the optimal fracturing process parameter combination construction scheme by comparing each capacity prediction result, wherein the neural network capacity prediction model is obtained by machine learning using a plurality of training samples, wherein the training samples include different fracturing process parameter combination construction schemes and corresponding capacity calculation results.

[0036] The following is a further optimization or / and improvement of the second technical solution of the above application:

[0037] The design unit comprises:

[0038] The process parameter module: the main fracturing process parameters affecting the fracturing reconstruction effect include porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height and conductivity;

[0039] The orthogonal design module: through orthogonal experiment design, the main fracturing process parameters are used as experimental factors, each experimental factor is divided into 2 to 3 levels, an orthogonal experiment table is established, and the fracturing construction scheme after orthogonal experiment design is obtained, that is, the fracturing construction scheme of different fracturing process parameter combinations is obtained.

[0040] The prediction unit comprises:

[0041] The data acquisition module acquires the productivity calculation results corresponding to different fracturing process parameter combination construction schemes, and the productivity calculation results comprise:

[0042] The productivity calculation results corresponding to different fracturing process parameter combination construction schemes are obtained by using a productivity prediction physical model, and the productivity prediction physical model is a productivity prediction physical model established based on percolation theory.

[0043] The low-permeability reservoir unsteady percolation equation considering the starting pressure gradient is as follows:

[0044]

[0045] The initial and boundary conditions of the low-permeability reservoir unsteady percolation equation considering the starting pressure gradient are as follows:

[0046]

[0047] The approximate solution of the unsteady percolation obtained by the integral method is as follows:

[0048]

[0049] The equation group (b) is substituted into the formula (c), and the coefficients a0, a1, a2,..., a n+1 in the formula (c) are solved to obtain a production formula:

[0050]

[0051] In the formula (a) to the formula (d), q is the production, m 3 / s; p is the formation pressure, m; p i is the original formation pressure, Pa; μ is the oil viscosity, Pa·s; K is the formation permeability, mD; r e is the supply boundary radius, m; h is the formation thickness, m; r is the well radius, m; G is the starting pressure gradient, Pa / m; r w is the drainage radius, m; r e (t) is the supply boundary radius at time t, m.

[0052] The model construction module constructs a neural network productivity prediction model, and the neural network productivity prediction model comprises:

[0053] The different fracturing process parameter combination construction schemes and the corresponding productivity calculation results are taken as a sample set, and the sample set is divided into a training set and a test set;

[0054] The neural network model is trained using the training set to obtain a neural network initial capacity prediction model, the neural network initial capacity prediction model is tested using the test set, if the model accuracy is met, the neural network initial capacity prediction model is output and used as the optimal neural network capacity prediction model, otherwise the neural network initial capacity prediction model parameters are adjusted, and the training set is used to train the neural network model again;

[0055] The capacity prediction module inputs different fracturing process parameter combination construction schemes into the neural network capacity prediction model, and outputs corresponding capacity prediction results, and determines the optimal fracturing process parameter combination construction scheme by comparing each capacity prediction result.

[0056] The physical model driven neural network fracturing process parameter optimization method provided by the application is a fracturing process parameter optimization method suitable for different block reservoirs, and has the following effects: 1. It can realize the rapid optimization of multiple fracturing construction parameters at the same time, and save a lot of time cost caused by simulation; 2. The neural network is driven by a physical model (i.e. the capacity prediction physical model), which avoids the limitation caused by the complexity of the reservoir and improves the accuracy of the neural network capacity prediction model; 3. After the neural network capacity prediction model is established, it can be applied to different reservoirs and has wide applicability; 4. After the fracturing process parameter optimization is completed, it can guide the design of the fracturing construction scheme and improve the single well capacity. The application selects a single well in the M2 well area of the Baikouquan group reservoir in the Mahu oilfield in the Junggar basin for fracturing process parameter optimization, can quickly obtain the optimization results of multiple fracturing process parameters, and the optimized capacity is 10% higher than that of the original fracturing construction scheme, which has good economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0057] ATTACHED Figure 1 The flowchart of the physical model driven neural network fracturing process parameter optimization method is shown in the application. DETAILED DESCRIPTION

[0058] The application is not limited by the following examples, and the specific implementation can be determined according to the technical scheme of the application and the actual situation.

[0059] Based on the data mining of big data, the main fracturing process parameters affecting the fracturing reconstruction effect of the horizontal well are determined by combining the oilfield geology, drilling, logging, microseismic, rock mechanics and fracturing construction and post-fracturing production data (engineering parameters, geological parameters and production data), the number of parameter combinations is reduced through orthogonal experiment design, the fracturing construction schemes of different fracturing process parameter combinations are obtained, the capacity of all fracturing construction schemes is calculated using the physical model (i.e. the capacity prediction physical model), and the capacity calculation results of different parameter combination fracturing construction schemes are obtained.

[0060] Based on different parameter combinations of fracturing construction scheme and corresponding productivity calculation results, machine learning is performed by using a neural network to find the relationship between fracturing process parameters and productivity calculation results, a neural network productivity prediction model based on a physical model is trained to obtain an efficient and accurate fracturing process parameter optimization method applicable to different blocks of reservoirs.

[0061] The application will be further described below in combination with examples:

[0062] Example 1: A physical model driven neural network fracturing process parameter optimization method, comprising:

[0063] The main fracturing process parameters affecting the fracturing reconstruction effect are used to design an orthogonal experiment, and fracturing construction schemes with different fracturing process parameter combinations are obtained.

[0064] The different fracturing process parameter combination construction schemes are input into the neural network productivity prediction model, and the corresponding productivity prediction results are output. The optimal fracturing process parameter combination construction scheme is determined by comparing each productivity prediction result. The neural network productivity prediction model is obtained by machine learning using a plurality of training samples, wherein the training samples include different fracturing process parameter combination construction schemes and corresponding productivity calculation results.

[0065] Example 2: As an optimization of the above-mentioned example, the main fracturing process parameters affecting the fracturing reconstruction effect are used to design an orthogonal experiment, and fracturing construction schemes with different fracturing process parameter combinations are obtained, as follows:

[0066] The main fracturing process parameters affecting the fracturing reconstruction effect include porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height and conductivity.

[0067] Through orthogonal experimental design, the main fracturing process parameters are used as experimental factors, each experimental factor is divided into 2 to 3 levels, and an orthogonal experiment table is established to maximize the number of experiments, thereby obtaining the fracturing construction scheme after orthogonal experimental design, i.e., the fracturing construction scheme with different fracturing process parameter combinations.

[0068] Example 3: As an optimization of the above-mentioned example, the productivity calculation results corresponding to the different fracturing process parameter combination construction schemes are obtained by using a productivity prediction physical model. The productivity prediction physical model is a productivity prediction physical model established based on percolation theory. The productivity prediction physical model is a low-permeability reservoir unsteady percolation equation considering the starting pressure gradient, as follows:

[0069] Low-permeability reservoir unsteady percolation equation considering starting pressure gradient:

[0070]

[0071] The initial and boundary conditions of the unstable seepage equation of the low-permeability reservoir considering the start-up pressure gradient are as follows:

[0072]

[0073] The approximate solution of the unstable seepage obtained by the integral method is as follows:

[0074]

[0075] The coefficients a0, a1, a2,..., a n+1 in the formula (c) are solved by bringing the formula (c) into the equation group (b), and the production formula is obtained:

[0076]

[0077] In the formula (a) to the formula (d), q is the production, m 3 / s; p is the formation pressure, m; p i 0 is the original formation pressure, Pa; μ is the oil viscosity, Pa·s; K is the formation permeability, mD; r e is the supply boundary radius, m; h is the formation thickness, m; r is the well radius, m; G is the start-up pressure gradient, Pa / m; r w is the drainage radius, m; r e (t) is the supply boundary radius at t, m.

[0078] The start-up pressure gradient is calculated according to the following formula:

[0079]

[0080] In the formula, G is the start-up pressure gradient, Pa / m; p0 is the core oil production end pressure, Pa; p s is the core closed end pressure, Pa; L is the core length, m; x is the coordinate along the core length, m.

[0081] In the above examples, the neural network productivity prediction model is constructed according to the following method, which comprises the following steps:

[0082] Combining the construction schemes of different fracturing process parameters and the corresponding productivity calculation results as a sample set, the sample set is divided into a training set and a test set;

[0083] The neural network model is trained using the training set to obtain a neural network productivity initial prediction model, and the test set is used to test the neural network productivity initial prediction model. If the model accuracy is met, the neural network productivity initial prediction model is output and used as the optimal neural network productivity prediction model. Otherwise, the neural network productivity initial prediction model parameters are adjusted, and the training set is used to train the neural network model again.

[0084] Embodiment 5: As an optimization of the above-mentioned embodiment 4, the model accuracy is the minimum mean square error (MSE) of the test results and the production capacity calculation results of the production capacity prediction physical model.

[0085] Embodiment 6: An optimization device, comprising:

[0086] A design unit: using the main fracturing process parameters affecting the fracturing reconstruction effect, orthogonal experimental design is carried out to obtain the fracturing construction scheme of different fracturing process parameter combinations;

[0087] A prediction unit: inputting the fracturing construction scheme of different fracturing process parameter combinations into a neural network production capacity prediction model, outputting the corresponding production capacity prediction results, and determining the optimal fracturing process parameter combination construction scheme by comparing each production capacity prediction result, wherein the neural network production capacity prediction model is obtained by machine learning using a plurality of training samples, wherein the training samples include different fracturing process parameter combination construction schemes and their corresponding production capacity calculation results.

[0088] Embodiment 7: As an optimization of the above-mentioned embodiment 6, the design unit comprises:

[0089] A process parameter module: the main fracturing process parameters affecting the fracturing reconstruction effect include porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height and conductivity;

[0090] An orthogonal design module: through orthogonal experimental design, using the main fracturing process parameters as experimental factors, dividing 2 to 3 levels for each experimental factor, establishing an orthogonal experiment table, and obtaining the fracturing construction scheme after orthogonal experimental design, that is, the fracturing construction scheme of different fracturing process parameter combinations.

[0091] Embodiment 8: As an optimization of the above-mentioned embodiment 6, the prediction unit comprises:

[0092] A data acquisition module acquires the production capacity calculation results corresponding to the fracturing construction scheme of different fracturing process parameter combinations, including:

[0093] The production capacity calculation results corresponding to the fracturing construction scheme of different fracturing process parameter combinations are obtained by using a production capacity prediction physical model, and the production capacity prediction physical model is a production capacity prediction physical model established based on percolation theory. The production capacity prediction physical model is a low-permeability reservoir unsteady percolation equation considering the starting pressure gradient, and is specifically as follows:

[0094] Low-permeability reservoir unsteady percolation equation considering starting pressure gradient:

[0095]

[0096] The initial and boundary conditions of the low-permeability reservoir unsteady percolation equation considering the starting pressure gradient are:

[0097]

[0098] The approximate solution for the unsteady seepage obtained by the integral method is as follows:

[0099]

[0100] Substituting equation (c) into the system of equations (b), we can solve for the coefficients a0, a1, a2...a0 in equation (c). n+1 The production formula is derived as follows:

[0101]

[0102] In equations (a) to (d), q represents output or production capacity, and m 3 / s; p is the formation pressure, m; p i The original formation pressure is given by: μ, crude oil viscosity is given by: K, formation permeability is given by: r e The supply boundary radius is m; h is the formation thickness, m; r is the well radius, m; G is the starting pressure gradient, Pa / m; r w Let r be the oil drain radius, in meters (m); e (t) is the supply boundary radius at time t, in meters.

[0103] The model building module constructs a neural network-based capacity prediction model, including:

[0104] The construction schemes with different combinations of fracturing process parameters and their corresponding production capacity calculation results are used as the sample set, which is then divided into a training set and a test set.

[0105] The neural network model is trained using the training set to obtain the initial neural network capacity prediction model. The initial neural network capacity prediction model is then tested using the test set. If the model accuracy is met, the initial neural network capacity prediction model is output and used as the optimal neural network capacity prediction model. Otherwise, the parameters of the initial neural network capacity prediction model are adjusted, and the neural network model is retrained using the training set.

[0106] Production capacity prediction module: Input different fracturing process parameter combination construction schemes into the neural network production capacity prediction model, output the corresponding production capacity prediction results, and determine the optimal fracturing process parameter combination construction scheme by comparing the various production capacity prediction results.

[0107] Example 9: As Figure 1 As shown, the neural network fracturing process parameter optimization method driven by the above physical model was applied to the Baikouquan Formation reservoir in the Ma131 well area of ​​the Mahu Oilfield in the Junggar Basin.

[0108] Step (1), collect data to determine the main fracturing process parameters affecting the effect of fracturing reconstruction. Collect and analyze the reservoir parameters, engineering parameters and production data of the fractured horizontal wells in the block, including Young's modulus, formation pressure, porosity, permeability, fracture half-length, fracture height, conductivity, production rate, cluster spacing, total sand volume and horizontal section length, etc. The main fracturing process parameters affecting the effect of fracturing reconstruction are porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height and conductivity.

[0109] Step (2), determine the fracturing construction scheme of different fracturing process parameter combinations. The above main fracturing process parameters are used as experimental factors, each experimental factor is divided into 2-3 levels, a part of representative experimental combinations are selected to establish an orthogonal experiment table, so as to maximize the simplification of the number of experiments, and 200 groups of fracturing construction schemes of different fracturing process parameter combinations are obtained, part of which are shown in Table 1.

[0110] Step (3), carry out productivity calculation under the construction scheme of different fracturing process parameter combinations. According to the formula described in embodiment 3, the productivity (i.e. production rate q) of the horizontal well is predicted, and the productivity calculation results under the construction scheme of different fracturing process parameter combinations are obtained. The fracturing construction scheme and its corresponding productivity calculation results are used as a sample set;

[0111] Step (4), establish a neural network productivity prediction model. The 200 groups of fracturing construction schemes and their corresponding productivity calculation results are divided into training set and test set according to the quantity ratio of 85:15. After segmentation, the training set is 170 groups and the test set is 30 groups. The neural network model is trained using the training set data, and the initial neural network productivity prediction model is obtained.

[0112] Step (5), verify the accuracy of the initial neural network productivity prediction model. The 30 groups of test set schemes are brought into the initial neural network productivity prediction model for testing, and the test results are compared with the productivity calculation results obtained according to the formula described in embodiment 3. When the mean square error of the test results and the productivity calculation results of the physical productivity prediction model is the smallest, it is proved that the model has excellent performance, good generalization ability and certain applicability. The initial neural network productivity prediction model is used as the optimal neural network productivity prediction model.

[0113] Step (6), the single well of different blocks of the target reservoir is applied to quickly optimize the fracturing process parameters. A single well in a certain well area of Well Ma2 in Baikouquan Reservoir of Mahu Oilfield in Junggar Basin is selected to optimize the fracturing process parameters. The main fracturing parameters of the single well are determined as porosity, permeability, cluster spacing, fracture half-length, formation pressure, fracture height and conductivity, and the range values are 8% to 12%, 0.1 mD to 2 mD, 15 m to 40 m, 80 m to 150 m, 40 MPa to 55 MPa, 20 m to 40 m, 5D·cm to 10D·cm, respectively. Step (2) is repeated, the fracturing construction scheme of different fracturing process parameter combinations is obtained, the obtained fracturing construction scheme of different parameter combinations is input into the optimal neural network productivity prediction model obtained in step (5), a large number of productivity prediction results corresponding to the construction scheme of different fracturing process parameter combinations are quickly obtained, the productivity prediction results are compared, and the economic adaptability evaluation is combined, and finally the optimal fracturing process parameter combination is determined as fracture half-length 120 m, cluster spacing 25 m, conductivity 8D·cm, and fracture height 35 m. The corresponding 15-year productivity is 34,000 tons, which is 10% higher than that of the original fracturing construction scheme. It has good economic benefits.

[0114] The above technical features constitute an embodiment of the present application, which has strong adaptability and implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the needs of different situations.

[0115] Table 1 part of the fracturing construction scheme

[0116]

Claims

1. A physical model-driven neural network-based method for optimizing fracturing process parameters, characterized in that... include: By utilizing the main fracturing process parameters that affect the fracturing effect, orthogonal experimental design was conducted to obtain fracturing construction schemes with different combinations of fracturing process parameters; Different fracturing process parameter combinations are input into the neural network capacity prediction model, which outputs the corresponding capacity prediction results. By comparing the various capacity prediction results, the optimal fracturing process parameter combination is determined. The neural network capacity prediction model is obtained by machine learning using several training samples, which include different fracturing process parameter combinations and their corresponding capacity calculation results.

2. The physical model-driven neural network fracturing process parameter optimization method according to claim 1, characterized in that... Using the main fracturing process parameters that affect the fracturing effect, an orthogonal experimental design was conducted to obtain fracturing operation schemes with different combinations of fracturing process parameters, as follows: The main fracturing process parameters that affect the fracturing effect include porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height, and conductivity. By using orthogonal experimental design, with the main fracturing process parameters as experimental factors, each experimental factor is divided into 2 to 3 levels. An orthogonal experimental table is established to obtain the fracturing construction scheme after the orthogonal experimental design, that is, the fracturing construction scheme with different combinations of fracturing process parameters.

3. The physical model-driven neural network fracturing process parameter optimization method according to claim 1 or 2, characterized in that... The production capacity calculation results corresponding to different fracturing process parameter combinations are obtained using a production capacity prediction physical model. The production capacity prediction physical model is an unsteady flow equation for low-permeability reservoirs that considers the initiation pressure gradient, as detailed below: Unsteady flow equation for low-permeability reservoirs considering the initiation pressure gradient: The initial and boundary conditions for the unsteady flow equation of a low-permeability reservoir considering the initiation pressure gradient are as follows: The approximate solution for the unsteady seepage obtained by the integral method is as follows: Substituting equation (c) into the system of equations (b), we can solve for the coefficients a0, a1, a2...a0 in equation (c). n+1 The production formula is derived as follows: In equations (a) to (d), q represents output or production capacity, and m 3 / s; p is the formation pressure, m; p i The original formation pressure is given by: μ, crude oil viscosity is given by: K, formation permeability is given by: r e The supply boundary radius is m; h is the formation thickness, m; r is the well radius, m; G is the starting pressure gradient, Pa / m; r w Let r be the oil drain radius, in meters (m); e (t) is the supply boundary radius at time t, in meters.

4. The physical model-driven neural network fracturing process parameter optimization method according to claim 1 or 2, characterized in that... The neural network capacity prediction model is constructed using the following method, including: The construction schemes with different combinations of fracturing process parameters and their corresponding production capacity calculation results are used as the sample set, which is then divided into a training set and a test set. The neural network model is trained using the training set to obtain the initial neural network capacity prediction model. The initial neural network capacity prediction model is then tested using the test set. If the model accuracy is met, the initial neural network capacity prediction model is output and used as the optimal neural network capacity prediction model. Otherwise, the parameters of the initial neural network capacity prediction model are adjusted, and the neural network model is retrained using the training set.

5. The physical model-driven neural network fracturing process parameter optimization method according to claim 3, characterized in that... The neural network capacity prediction model is constructed using the following method, including: The construction schemes with different combinations of fracturing process parameters and their corresponding production capacity calculation results are used as the sample set, which is then divided into a training set and a test set. The neural network model is trained using the training set to obtain the initial neural network capacity prediction model. The initial neural network capacity prediction model is then tested using the test set. If the model accuracy is met, the initial neural network capacity prediction model is output and used as the optimal neural network capacity prediction model. Otherwise, the parameters of the initial neural network capacity prediction model are adjusted, and the neural network model is retrained using the training set.

6. The physical model-driven neural network fracturing process parameter optimization method according to claim 5, characterized in that... The model accuracy is defined as having the smallest mean square error between the test results and the capacity calculation results of the physical model for capacity prediction.

7. An optimization apparatus for implementing the physical model-driven neural network fracturing process parameter optimization method according to any one of claims 1 to 6, characterized in that... include: Design Unit: Using the main fracturing process parameters that affect the fracturing effect, orthogonal experimental design is carried out to obtain fracturing construction schemes with different combinations of fracturing process parameters; Prediction Unit: Input different fracturing process parameter combination construction schemes into the neural network capacity prediction model, output the corresponding capacity prediction results, and determine the optimal fracturing process parameter combination construction scheme by comparing the various capacity prediction results. The neural network capacity prediction model is obtained by machine learning using several training samples, which include different fracturing process parameter combination construction schemes and their corresponding capacity calculation results.

8. The optimization device according to claim 7, characterized in that... Design units include: Process parameter module: The main fracturing process parameters that affect the fracturing effect include porosity, formation permeability, cluster spacing, fracture half-length, formation pressure, fracture height, and conductivity. Orthogonal design module: Through orthogonal experimental design, the main fracturing process parameters are used as experimental factors. Each experimental factor is divided into 2 to 3 levels. An orthogonal experimental table is established to obtain the fracturing construction scheme after orthogonal experimental design, that is, the fracturing construction scheme with different combinations of fracturing process parameters.

9. The optimization device according to claim 7, characterized in that... Prediction unit, including: The data acquisition module obtains the production capacity calculation results corresponding to different fracturing process parameter combinations and construction schemes, including: The production capacity calculation results corresponding to different fracturing process parameter combinations are obtained using a production capacity prediction physical model. The production capacity prediction physical model is an unsteady flow equation for low-permeability reservoirs that considers the initiation pressure gradient, as detailed below: Unsteady flow equation for low-permeability reservoirs considering the initiation pressure gradient: The initial and boundary conditions for the unsteady flow equation of a low-permeability reservoir considering the initiation pressure gradient are as follows: The approximate solution for the unsteady seepage obtained by the integral method is as follows: Substituting equation (c) into the system of equations (b), we can solve for the coefficients a0, a1, a2...a0 in equation (c). n+1 The production formula is derived as follows: In equations (a) to (d), q represents output or production capacity, and m 3 / s; p is the formation pressure, m; p i The original formation pressure is given by: μ, crude oil viscosity is given by: K, formation permeability is given by: r e The supply boundary radius is m; h is the formation thickness, m; r is the well radius, m; G is the starting pressure gradient, Pa / m; r w Let r be the oil drain radius, in meters (m); e (t) represents the supply boundary radius at time t, in meters. The model building module constructs a neural network-based capacity prediction model, including: The construction schemes with different combinations of fracturing process parameters and their corresponding production capacity calculation results are used as the sample set, which is then divided into a training set and a test set. The neural network model is trained using the training set to obtain the initial neural network capacity prediction model. The initial neural network capacity prediction model is then tested using the test set. If the model accuracy is met, the initial neural network capacity prediction model is output and used as the optimal neural network capacity prediction model. Otherwise, the parameters of the initial neural network capacity prediction model are adjusted, and the neural network model is retrained using the training set. Production capacity prediction module: Input different fracturing process parameter combination construction schemes into the neural network production capacity prediction model, output the corresponding production capacity prediction results, and determine the optimal fracturing process parameter combination construction scheme by comparing the various production capacity prediction results.

10. The optimization device according to claim 8, characterized in that... Prediction unit, including: The data acquisition module obtains the production capacity calculation results corresponding to different fracturing process parameter combinations and construction schemes, including: The production capacity calculation results corresponding to different fracturing process parameter combinations are obtained using a production capacity prediction physical model. This model is based on seepage theory and is an unsteady seepage equation for low-permeability reservoirs that considers the initiation pressure gradient, as detailed below: Unsteady flow equation for low-permeability reservoirs considering the initiation pressure gradient: The initial and boundary conditions for the unsteady flow equation of a low-permeability reservoir considering the initiation pressure gradient are as follows: The approximate solution for the unsteady seepage obtained by the integral method is as follows: Substituting equation (c) into the system of equations (b), we can solve for the coefficients a0, a1, a2...a0 in equation (c). n+1 The production formula is derived as follows: In equations (a) to (d), q represents output, and m... 3 / s; p is the formation pressure, m; p i The original formation pressure is given by: μ, crude oil viscosity is given by: K, formation permeability is given by: r e The supply boundary radius is m; h is the formation thickness, m; r is the well radius, m; G is the starting pressure gradient, Pa / m; r w Let r be the oil drain radius, in meters (m); e (t) represents the supply boundary radius at time t, in meters. The model building module constructs a neural network-based capacity prediction model, including: The construction schemes with different combinations of fracturing process parameters and their corresponding production capacity calculation results are used as the sample set, which is then divided into a training set and a test set. The neural network model is trained using the training set to obtain the initial neural network capacity prediction model. The initial neural network capacity prediction model is then tested using the test set. If the model accuracy is met, the initial neural network capacity prediction model is output and used as the optimal neural network capacity prediction model. Otherwise, the parameters of the initial neural network capacity prediction model are adjusted, and the neural network model is retrained using the training set. Production capacity prediction module: Input different fracturing process parameter combination construction schemes into the neural network production capacity prediction model, output the corresponding production capacity prediction results, and determine the optimal fracturing process parameter combination construction scheme by comparing the various production capacity prediction results.

Images