Method and device for determining fracture parameters of horizontal well fracturing

Abstract

The present specification relates to the technical field of unconventional oil and gas reservoir horizontal well development, and specifically discloses a horizontal well fracturing fracture parameter determination method and device, the method comprising: obtaining static geological data and dynamic fracturing operation data of a target oil and gas reservoir; inputting the static geological data and dynamic fracturing operation data into a target fracture parameter determination model to obtain fracture parameters of the target oil and gas reservoir; the target fracture parameter determination model comprises a combined neural network and a deep neural network; the combined neural network is used for inputting the static geological data and dynamic fracturing operation data; the deep neural network is used for calculating the data input by the combined neural network to obtain the fracture parameters; the loss function of the deep neural network comprises a first loss function based on data driving, a second loss function constructed based on field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on a fracture propagation equation corresponding to the target oil and gas reservoir. The above scheme can accurately and quickly predict fracture parameters in real time.

Description

Technical Field

[0001] This specification relates to the field of horizontal well development technology for unconventional oil and gas reservoirs, and in particular to a method and apparatus for determining fracture parameters in horizontal well fracturing. Background Technology

[0002] Tight / shale oil and gas is an important component of unconventional oil and gas resources. Efficient development and utilization of tight / shale oil has become a crucial guarantee for the development of the oil and gas industry. Horizontal well fracturing technology is an effective means of extracting tight / shale oil. Fracturing effect evaluation provides an important basis for optimizing production enhancement measures, requiring accurate assessment of fracture geometry and conductivity. Compared with conventional horizontal well fracturing, due to the multi-layered development of continental tight / shale oil and gas, multi-layered development is necessary. Horizontal well dimensional development involves more wells, more fracturing stages, and more perforation clusters, resulting in a more complex fracture network morphology and making fracture assessment more difficult. Unlike fracturing effect evaluation in vertical well development or conventional horizontal well development, the fracture propagation path and patterns in dimensional horizontal well fracturing are more complex, requiring more stringent physical constraints and more optimized deep learning models to determine target fracture parameters.

[0003] Conventional performance evaluation methods mainly fall into two categories. The first is direct monitoring methods such as microseismic monitoring and distributed fiber optic monitoring. The drawback is that these methods are inaccurate for monitoring fractures in areas far from the wellbore and in horizontal wells; they can obtain fracture geometry but not fracture conductivity. The second method is indirect inversion, which involves retrieving fracture parameters through production history fitting and well test pressure analysis. Due to the simple model assumptions, this method requires significant time and effort, and the retrieved fracture parameters often differ from actual results. For horizontal well fracturing, interpreting the complex fracture network is inherently difficult, and a real-time, efficient, and accurate fracture diagnosis method is lacking.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This specification provides a method and apparatus for determining fracture parameters in horizontal well fracturing, in order to address the problem of the lack of real-time, efficient and accurate fracture diagnosis methods in the prior art.

[0006] This specification provides an embodiment of a method for determining fracture parameters in horizontal well fracturing, including:

[0007] Acquire static geological data and dynamic fracturing operation data of the target oil and gas reservoir;

[0008] The static geological data and the dynamic fracturing construction data are input into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir;

[0009] The target fracture parameter determination model includes a combined neural network and a deep neural network. The combined neural network is used as input for the static geological data and the dynamic fracturing construction data. The deep neural network is used to calculate the fracture parameters from the data input by the combined neural network. The loss function of the deep neural network includes a data-driven first loss function, a second loss function constructed based on the field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir.

[0010] In one embodiment, the second loss function is determined by: acquiring field monitoring data of the target oil and gas reservoir; performing inversion on the field monitoring data to obtain monitoring data; constructing the second loss function based on the monitoring data; the monitoring data including the range of fracture parameter values ​​obtained from the inversion; and / or

[0011] The third loss function is determined by: determining the fracture type based on the dynamic fracturing construction data, selecting the corresponding fracture propagation equation according to the fracture type, and constructing the third loss function based on the fracture propagation equation.

[0012] In one embodiment, the static geological data and the dynamic fracturing operation data are input into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir, including:

[0013] The dynamic fracturing construction data is preprocessed to obtain preprocessed dynamic fracturing construction data; the preprocessing includes at least one of the following: fracturing segment truncation, data denoising, and feature point extraction.

[0014] The static geological data and the preprocessed dynamic fracturing construction data are input into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

[0015] In one embodiment, the dynamic fracturing operation data includes fracturing operation pressure data, fluid injection volume, operation displacement volume, and proppant concentration at various times during multiple moments in the horizontal well; correspondingly, the dynamic fracturing operation data is preprocessed to obtain preprocessed dynamic fracturing operation data, including:

[0016] Based on the fluid addition volume, the fracturing operation curve generated from the fracturing operation pressure data of the horizontal well at various times is truncated to obtain the fracturing section.

[0017] The fracturing pressure data corresponding to the fracturing section is denoised. The intersection of the fracturing pressure point where the sand concentration change exceeds the first preset threshold and the point where the fracturing pressure data change exceeds the second preset threshold is determined as the characteristic pressure point, thus obtaining the characteristic pressure point corresponding to the fracturing section.

[0018] The characteristic pressure points corresponding to each fracturing segment, the construction displacement and sand concentration corresponding to the characteristic pressure points are determined as the pre-processed dynamic fracturing construction data corresponding to the fracturing segment.

[0019] In one embodiment, the target crack parameter determination model is constructed in the following manner:

[0020] A training sample library is established, which includes an input sample set and an output sample set. The input data in the input sample set includes static geological data and dynamic construction data corresponding to each fracturing segment in multiple fracturing segments. The output data in the output sample set includes fracture evaluation parameters corresponding to each input data in the input sample set.

[0021] The training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a target crack parameter determination model.

[0022] In one embodiment, the output data is determined by one of the following methods:

[0023] The input data in the input sample set is subjected to fracturing construction pressure analysis, and the output data of dynamic fracture parameters during fracture propagation are obtained by inversion.

[0024] By using production dynamics analysis to fit production dynamics, output data of static crack parameters after crack closure are obtained through inversion.

[0025] Based on the on-site fracturing monitoring technology, the fracture parameters are diagnosed and corrected to obtain the output data of comprehensive fracture parameters.

[0026] In one embodiment, the training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a target crack parameter determination model, including:

[0027] The training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a crack parameter determination model;

[0028] The crack parameter determination model was validated using k-fold cross-validation to obtain the target crack parameter determination model.

[0029] In one embodiment, after obtaining the fracture parameters of the target oil and gas reservoir, the method further includes:

[0030] Based on the fracture parameters of the target oil and gas reservoir and the field monitoring data, the fracture stimulation volume and fracture conductivity of the target oil and gas reservoir are determined.

[0031] Based on the fracture modification volume and the fracture conductivity, calculate the fracture productivity and / or fracture economic indicators of the target oil and gas reservoir.

[0032] This specification also provides an embodiment of a device for determining fracture parameters in horizontal well fracturing, comprising:

[0033] The acquisition module is used to acquire static geological data and dynamic fracturing operation data of the target oil and gas reservoir;

[0034] The input module is used to input the static geological data and the dynamic fracturing construction data into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

[0035] The target fracture parameter determination model includes a combined neural network and a deep neural network. The combined neural network is used as input for the static geological data and the dynamic fracturing construction data. The deep neural network is used to calculate the fracture parameters from the data input by the combined neural network. The loss function of the deep neural network includes a data-driven first loss function, a second loss function constructed based on the field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir.

[0036] This specification also provides a computer device, including a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of the method for determining fracture parameters in horizontal well fracturing as described in any of the above embodiments.

[0037] This specification also provides a computer-readable storage medium storing computer instructions that, when executed, implement the steps of the method for determining fracture parameters in horizontal well fracturing as described in any of the above embodiments.

[0038] This specification provides a method for determining fracture parameters in horizontal well fracturing. This method acquires static geological data and dynamic fracturing operation data of a target oil and gas reservoir. By inputting the static geological data and dynamic fracturing operation data into a target fracture parameter determination model, the fracture parameters of the target oil and gas reservoir can be obtained. The target fracture parameter determination model may include a combined neural network and a deep neural network. The combined neural network receives the input of the static geological data and the dynamic fracturing operation data, while the deep neural network calculates the fracture parameters based on the data input from the combined neural network. The deep neural network uses a loss function that includes a data-driven first loss function, a second loss function constructed based on field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir. This allows the calculation to consider not only the conventional data-driven loss function but also the physical constraints of field monitoring data and the fracture propagation equation on the fracture parameters. This approach, by using dynamic fracturing operation data that controls fracture parameters as part of the input, can make fracture prediction more accurate. The use of combined neural networks enables multi-dimensional data input, solving the problems of varying input data dimensions and the neglect of geological data due to excessive fracturing operation data. By incorporating a second loss function based on field monitoring data and a third loss function based on the fracture propagation equation into the conventional loss function, physical constraints are imposed on the deep learning process. This makes the deep neural network not merely data-driven but physically meaningful, considering actual fracture parameters. Different loss function components further constrain the prediction results from different perspectives, enabling accurate prediction of fracture parameters obtained from three-dimensional fracturing of horizontal wells in tight oil and gas reservoirs. Compared to conventional fracture parameter determination methods, this approach significantly reduces the workload and economic cost of fracture parameter assessment, achieving real-time, accurate, and rapid prediction of fracture parameters. Furthermore, this approach can predict fractures in both vertical and conventional horizontal wells, requiring only adjustments to the model optimization process, thus compensating for the shortcomings of traditional fracture prediction methods for vertical and conventional horizontal wells. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of this specification and form part of it, do not constitute a limitation thereof. In the drawings:

[0040] Figure 1 A flowchart of a method for determining fracture parameters in horizontal well fracturing according to one embodiment of this specification is shown;

[0041] Figure 2 A flowchart of a method for determining fracture parameters in horizontal well fracturing according to one embodiment of this specification is shown;

[0042] Figure 3 A schematic diagram of the crack parameter determination model in one embodiment of this specification is shown;

[0043] Figure 4 This diagram illustrates the effective stimulated volume of the horizontal well fracturing segment calculated using a fracture parameter determination method according to an embodiment of this specification.

[0044] Figure 5 This diagram illustrates a comparison between the predicted and actual values ​​of the fracture half-length in a fracture parameter determination method for horizontal well fracturing according to an embodiment of this specification.

[0045] Figure 6 This diagram illustrates a comparison between the predicted and actual values ​​of the fracture half-height in a fracture parameter determination method for horizontal well fracturing according to an embodiment of this specification.

[0046] Figure 7 This diagram illustrates a comparison between the predicted and actual values ​​of the fracture half-width in a fracture parameter determination method for horizontal well fracturing according to an embodiment of this specification.

[0047] Figure 8 This diagram illustrates a comparison between the predicted and actual values ​​of fracture permeability in a fracture parameter determination method for horizontal well fracturing according to one embodiment of this specification.

[0048] Figure 9 A schematic diagram of a method for determining fracture parameters in horizontal well fracturing according to an embodiment of this specification is shown;

[0049] Figure 10 A schematic diagram of a computer device according to one embodiment of this specification is shown. Detailed Implementation

[0050] The principles and spirit of this specification will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are given merely to enable those skilled in the art to better understand and implement this specification, and are not intended to limit the scope of this specification in any way. Rather, these embodiments are provided to make this disclosure more thorough and complete, and to fully convey the scope of this disclosure to those skilled in the art.

[0051] Those skilled in the art will recognize that the embodiments described in this specification can be implemented as a system, apparatus, method, or computer program product. Therefore, the disclosure of this specification can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0052] This specification provides an embodiment of a method for determining fracture parameters in horizontal well fracturing. Figure 1 A flowchart illustrating a method for determining fracture parameters in horizontal well fracturing according to one embodiment of this specification is provided. While this specification provides method operation steps or apparatus structures as shown in the embodiments or figures below, more or fewer operation steps or module units may be included in the method or apparatus based on conventional or non-inventive effort. In steps or structures where there is no logically necessary causal relationship, the execution order of these steps or the module structure of the apparatus is not limited to the execution order or module structure described in the embodiments and figures of this specification. When the method or module structure is applied in actual devices or end products, it can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed processing environment) according to the method or module structure shown in the embodiments or figures.

[0053] Specifically, such as Figure 1 As shown, one embodiment of this specification provides a method for determining fracture parameters in horizontal well fracturing, which may include the following steps:

[0054] Step S101: Obtain static geological data and dynamic fracturing operation data of the target oil and gas reservoir.

[0055] The method described in this embodiment can be applied to computer equipment or applications installed on computer equipment. It can acquire static geological data and dynamic fracturing operation data of the target oil and gas reservoir. The target oil and gas reservoir can be a shale oil reservoir, shale gas reservoir, tight oil reservoir, or tight gas reservoir with predicted fracture parameters.

[0056] Static geological data can be geological data of the target oil and gas reservoir that does not change over time. In one embodiment, static geological data may include at least one of the following: maximum and minimum horizontal in-situ stress, reservoir pressure, Young's modulus of the rock, Poisson's ratio of the rock, reservoir porosity, and reservoir permeability.

[0057] Dynamic fracturing data can be the data collected during horizontal well fracturing of a target oil and gas reservoir. In one embodiment, dynamic fracturing data may include: fracturing pressure, fluid injection rate, fracturing displacement rate, and proppant concentration at different times for each fracturing stage of each well.

[0058] Step S102: Input the static geological data and the dynamic fracturing construction data into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

[0059] After obtaining static geological data and dynamic fracturing operation data, these data can be input into the target fracture parameter determination model. The resulting output data represents the fracture parameters of the target oil and gas reservoir. These fracture parameters may include, but are not limited to, at least one of the following: fracture length, fracture height, fracture width, and permeability.

[0060] The target fracture parameter determination model may include a combined neural network and a deep neural network. The combined neural network can be used as input for both the static geological data and the dynamic fracturing operation data. The static geological data does not change over time; each fracturing segment corresponds to one set of static geological data, which is a one-dimensional input vector. The static geological data can be calculated using well logging data.

[0061] Dynamic fracturing data can change over time, with each fracturing segment corresponding to a table, representing two-dimensional input vector data. A combining neural network can combine one-dimensional and two-dimensional input vector data as input.

[0062] In one embodiment, the combined neural network can be composed of a first neural network and a second neural network connected in parallel. The first neural network is responsible for inputting geological data, and the second neural network is responsible for inputting dynamic fracturing operation data. The number of layers in the combined neural network, and the number of neurons in each layer, are optimized according to actual needs.

[0063] As an example, and not a limitation, the first neural network can consist of a single-layer neural network containing only 8 neurons. The second neural network can allocate 2000 neurons each as input layers for pressure, proppant concentration, and displacement. The second neural network can also include 6 hidden layers, each containing 800 neurons, and an output layer of 100 neurons.

[0064] The deep neural network is used to calculate the fracture parameters from the data input to the combined neural network. The loss function of the deep neural network may include a first loss function based on data-driven analysis, a second loss function constructed based on field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir.

[0065] In one embodiment, a suitable deep neural network input layer can be established based on the output of the combined neural network. An appropriate number of hidden layers and neurons can be selected to construct the deep neural network, and the output of the deep neural network can be designed as fracture parameters. For the deep neural network, the loss function is redefined. In addition to the conventional data-driven first loss function (e.g., mean squared error), a second loss function based on field monitoring data and a third loss function based on the fracture propagation equation corresponding to the target oil and gas reservoir are added.

[0066] In one embodiment, the range of fracture parameter values ​​can be determined from field monitoring data. In one embodiment, field monitoring data may include data from microseismic, fiber optic, water hammer, and inclinometer measurements. In one embodiment, a fracture propagation equation can be used to calculate the range of fracture parameter values; this equation may include commonly used two-dimensional, quasi-three-dimensional, and three-dimensional fracture propagation equations. An appropriate fracture propagation equation can be selected based on the actual conditions of the target oil and gas reservoir.

[0067] In the above embodiments, static geological data and dynamic fracturing operation data of the target oil and gas reservoir can be acquired. By inputting the static geological data and the dynamic fracturing operation data into the target fracture parameter determination model, the fracture parameters of the target oil and gas reservoir can be obtained. The target fracture parameter determination model can include a combined neural network and a deep neural network. The combined neural network can input the static geological data and the dynamic fracturing operation data, while the deep neural network can calculate the fracture parameters from the data input by the combined neural network. The loss function of the deep neural network includes a data-driven first loss function, a second loss function constructed based on the field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir. This allows the calculation to consider not only conventional loss functions but also the physical constraints of microseismic data and the fracture propagation equation on the fracture parameters. The above scheme, by using dynamic fracturing operation data that controls fracture parameters as part of the input, can make fracture prediction more accurate. Using a combined neural network enables multi-dimensional data input, solving the problems of different input data dimensions and the neglect of geological data due to excessive fracturing operation data. By incorporating a second loss function based on field monitoring data and a third loss function based on the fracture propagation equation into the conventional loss function, physical constraints can be applied to the deep learning process. This makes the deep neural network not merely data-driven but physically meaningful, considering actual fracture parameters. Different loss function components further constrain the prediction results from different perspectives, enabling accurate prediction of fracture parameters obtained from three-dimensional fracturing of horizontal wells in tight oil and gas reservoirs. Compared to conventional fracture parameter determination methods, this approach significantly reduces the workload and economic cost of fracture parameter assessment, achieving real-time, accurate, and rapid prediction of fracture parameters. Furthermore, this approach can predict fractures in both vertical and conventional horizontal wells, requiring only adjustments to the model optimization process, thus overcoming the shortcomings of conventional fracture prediction methods for vertical and horizontal wells. Vertical wells correspond to a set of input data per well, while horizontal wells are based on fractured sections. Vertical wells may have production data, while horizontal wells rarely have production data in sections. The model can be adjusted based on the differences between vertical and horizontal wells, allowing for the determination of fracturing parameters for vertical wells.

[0068] In some embodiments of this specification, the second loss function may be determined by: acquiring field monitoring data of the target oil and gas reservoir, performing inversion on the field monitoring data to obtain monitoring data; constructing the second loss function based on the monitoring data; the monitoring data includes the range of fracture parameter values ​​obtained by inversion.

[0069] Specifically, field monitoring data of the target oil and gas reservoir can be obtained. Then, monitoring data can be obtained by inverting the field monitoring data. The monitoring data includes the range of fracture parameter values ​​obtained through inversion. After obtaining the monitoring data, a second loss function can be constructed based on the range of fracture parameter values ​​in the monitoring data.

[0070] In some embodiments of this specification, the third loss function may be determined by: determining the fracture type based on the dynamic fracturing construction data, selecting the corresponding fracture propagation equation according to the fracture type, and constructing the third loss function based on the fracture propagation equation.

[0071] Specifically, fracture types can be predicted based on dynamic fracturing construction data, and the corresponding fracture propagation equation can be selected according to the fracture type. After selecting the fracture propagation equation, a third loss function can be constructed based on the fracture propagation equation.

[0072] Crack propagation equations can be used to calculate the range of crack parameters, including commonly used two-dimensional, pseudo-three-dimensional, and three-dimensional crack propagation equations. Each equation makes different model assumptions. Commonly used two-dimensional equations include KGD, PKN, and radial models. Pseudo-three-dimensional models are more realistic than two-dimensional models, while three-dimensional models are the most realistic, but computationally complex. It is necessary to pre-determine the crack type based on the construction conditions and select an appropriate crack propagation equation as a constraint.

[0073] In one embodiment, a KGD geometric model can be used. This model is a two-dimensional model, relatively simple, and basically conforms to the crack propagation state of the block being processed. In another embodiment, a PKN model can also be used.

[0074] The required fracture geometry for fracturing enhancement varies depending on the reservoir type, as described below:

[0075] (1) For single-layer development of low-permeability, tight reservoirs, effective fracturing and production enhancement measures require relatively long fractures and limit the extension of fracture height.

[0076] (2) For multi-layered reservoirs, three-dimensional development is usually adopted, hoping that the fractures can penetrate the layers in height, so as to achieve the purpose of multi-layer development.

[0077] By calculating the net pressure within the fracture through fracturing operation pressure, the propagation characteristics and geometry of the fracture can be determined, and a suitable fracture propagation model can be selected. This is described below:

[0078] (1) For the case where the area of ​​the point source liquid entering the crack increases in a circular manner, a radial model can be used as the crack propagation equation.

[0079] (2) The thickness of the liquid entering the entire reservoir can be approximated by the line source. The fracture area expands in an elliptical shape, and the KGD geometric model can be used as the fracture propagation equation.

[0080] (3) When the stress in the upper and lower layers of the producing layer is greater than that of the producing layer, the growth of the fracture height is restricted, and the fracture expands in a circular pattern. The fracture length is greatly extended, causing the pressure to rise, which is similar to the PKN model. Therefore, the PKN model can be used as the fracture propagation equation.

[0081] Specifically, a double logarithmic curve (Nolte-Smith plot) of net pressure versus time during pumping can be generated based on dynamic fracturing operation data. The double logarithmic relationship between net pressure and time is a straight line, with the slope equal to the exponent of each equation. If the slope is positive and less than 1 / 4, the PKN model is used. For negative slopes, either the KGD geometric model or the radial model can be used.

[0082] In the above embodiments, selecting different fracture propagation equations based on the type of fracture morphology in the target oil and gas reservoir to construct a third loss function can further improve the accuracy of fracture parameter prediction.

[0083] In some embodiments of this specification, inputting the static geological data and the dynamic fracturing operation data into a target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir may include: preprocessing the dynamic fracturing operation data to obtain preprocessed dynamic fracturing operation data; the preprocessing includes at least one of the following: fracturing segment truncation, data denoising, and feature point extraction; inputting the static geological data and the preprocessed dynamic fracturing operation data into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

[0084] Specifically, for static geological data, geological data can be obtained from well logging curves, and the geological data for each fracturing segment can be obtained by averaging according to formation depth. The dynamic fracturing data includes: fracturing pressure, injection rate, and proppant concentration at different times for each fracturing segment of each well. For dynamic fracturing data, preprocessing can be performed, including effective fracturing segment extraction, data denoising, and feature point extraction. In this embodiment, by preprocessing the dynamic fracturing data, effective fracturing data can be extracted, thereby making the fracture parameter prediction results more accurate.

[0085] In some embodiments of this specification, the dynamic fracturing construction data may include fracturing construction pressure data, fluid injection volume, construction displacement, and proppant concentration of the horizontal well at various times within multiple time periods. Correspondingly, preprocessing the dynamic fracturing construction data to obtain preprocessed dynamic fracturing construction data may include: truncating the fracturing construction curve generated based on the fracturing construction pressure data of the horizontal well at various times within multiple time periods according to the fluid injection volume to obtain fracturing segments; performing noise reduction processing on the fracturing construction pressure data corresponding to the fracturing segments; determining the intersection of the fracturing construction pressure point where the sand concentration change exceeds a first preset threshold and the point where the fracturing construction pressure data change exceeds a second preset threshold as a characteristic pressure point, thus obtaining the characteristic pressure point corresponding to the fracturing segment; and determining the characteristic pressure point corresponding to each fracturing segment, the construction displacement corresponding to the characteristic pressure point, and the sand concentration as the preprocessed dynamic fracturing construction data corresponding to the fracturing segment.

[0086] Specifically, a feature point data extraction method can be established (which can be based on MATLAB or Python code or use computational software). The fracturing operation curve is truncated according to the amount of fluid added. The pressure data is denoised using a noise reduction method (e.g., wavelet denoising). The intersection of the pressure point where the sand concentration change exceeds a certain threshold and the point where the pressure itself changes by a certain threshold is taken as the feature pressure point. The displacement and sand concentration corresponding to the feature pressure point are used together as the input fracturing operation data.

[0087] In some embodiments of this specification, the target fracture parameter determination model may be constructed in the following manner: establishing a training sample library; the training sample library includes an input sample set and an output sample set; the input data in the input sample set includes static geological data and corresponding dynamic construction data corresponding to each fracturing segment in multiple fracturing segments; the output data in the output sample set includes fracture evaluation parameters corresponding to each input data in the input sample set; the training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain the target fracture parameter determination model.

[0088] Specifically, a sample library can be established first. The training sample library includes an input sample set and an output sample set. The input data in the input sample set includes static geological data and dynamic fracturing data. The static geological data includes: maximum and minimum horizontal in-situ stress, reservoir pressure, Young's modulus of the rock, Poisson's ratio of the rock, reservoir porosity, and reservoir permeability; the dynamic fracturing data includes: fracturing pressure, fluid injection volume, fracturing displacement, and proppant concentration at different times for each fracturing stage of each well.

[0089] Geological data can be obtained from well logging curves, and averaged according to formation depth to obtain the geological data for each fracturing segment. Dynamic fracturing operation data requires preprocessing, including effective fracturing segment truncation, data denoising, and feature point extraction. A feature point data extraction method should be established (which can be based on MATLAB or Python code or use computational software). The fracturing operation curve should be truncated according to the fluid injection volume. Denoising methods should be used to reduce the noise in the pressure data (e.g., wavelet denoising). The intersection of the pressure point where the sand concentration change exceeds a certain threshold and the point where the pressure itself changes by a certain threshold is taken as the feature pressure point. The displacement and sand concentration corresponding to the feature pressure point are used together as the input fracturing operation data. Static geological data and dynamic fracturing operation data of the same fracturing segment can be combined to obtain the input data for the target fracture parameter determination model for all fracturing segments.

[0090] The output data consists of crack assessment parameters, including crack geometry and conductivity, including but not limited to crack length, height, width, and permeability.

[0091] Combinatorial neural networks and deep neural networks considering physical constraints can be constructed. Combinatorial neural networks are constructed based on different characteristics of dynamic and static input data. Different numbers of neurons are allocated according to the characteristics of geological and fracturing data, resulting in two neural networks. The number of network layers and the number of neurons in each layer are optimized based on data characteristics. The two neural networks are then connected in parallel to form a combined neural network. Deep neural networks can also be built. Appropriate input layers for the deep neural network are established based on the output of the combined neural network. An appropriate number of hidden layers and neurons in each hidden layer are selected to construct the deep neural network. The model output is designed to include the fracture half-length, half-height, half-width, and permeability. The loss function of the deep neural network can be redefined. Specifically, based on the conventional first loss function (mean squared error), the range of fracture parameters obtained from field monitoring data can be added as a second loss function, and the corresponding fracture propagation equation can be added as a third loss function.

[0092] In one embodiment, the loss function of a deep neural network can be expressed by the following formula.

[0093]

[0094] Among them, Loss all This represents the total loss function, and α1, α2, and α3 represent the weight coefficients, the specific values ​​of which need to be determined through model training.

[0095] MSE stands for Mean Squared Error, and its expression is as follows:

[0096]

[0097] Where N represents the number of samples, yi predict It is the predicted value, y i data This is the actual value.

[0098] Loss f The second loss function is expressed as follows:

[0099]

[0100] Where the subscript f represents the crack parameter, y f The predicted crack parameters are represented by y. fmin and y fmax denoted as the minimum and maximum values ​​of the crack geometric parameters obtained from the field monitoring data, j = 1, 2, and 3 respectively represent the crack length, crack height, and crack width, and i represents the number of samples.

[0101] Here, we take the two-dimensional crack propagation equation as an example, Loss 2D The third loss function corresponding to the two-dimensional crack propagation equation is expressed as follows:

[0102]

[0103] in, The crack parameters determine the crack length predicted by the model. The crack width is predicted by the crack parameter determination model, and w and L represent the crack width and length calculated by the two-dimensional crack propagation equation.

[0104] When the two-dimensional crack propagation equation is based on the KGD geometric model, the formulas for calculating w and L are as follows:

[0105]

[0106]

[0107] When the two-dimensional crack propagation equation is based on the PKN model, the formulas for calculating w and L are as follows:

[0108]

[0109]

[0110] When the two-dimensional crack propagation equation is a radial model, the formulas for calculating w and L are as follows:

[0111]

[0112]

[0113] Where w is the seam width, L is the seam length, and h fν is the fracture height, q is the displacement, μ is the fracturing fluid viscosity, G is the shear modulus, t is the time, and E' is the plane modulus.

[0114] After constructing the combined neural network and the deep neural network that considers physical constraints, the training sample library can be input into the pre-constructed combined neural network and deep neural network for training to obtain the target crack parameter determination model.

[0115] In some embodiments of this specification, the output data may be determined by one of the following methods: performing fracturing construction pressure analysis on the input data in the input sample set to obtain output data of dynamic fracture parameters during fracture propagation; fitting production dynamics using production dynamic analysis cases to obtain output data of static fracture parameters after fracture closure; or diagnosing and correcting fracture parameters based on on-site fracturing monitoring technology to obtain output data of comprehensive fracture parameters.

[0116] Specifically, based on preprocessed dynamic fracturing data and static geological data, the fracturing module of fracturing simulation software can be used to analyze fracturing construction pressure and invert the output data of dynamic fracture parameters during fracture propagation. Alternatively, production dynamic analysis can be used to fit production dynamics and invert the output data of static fracture parameters after fracture closure. Alternatively, fracture parameters can be diagnosed based on on-site fracturing monitoring technology to obtain comprehensive fracture parameter output data. The reliability of the data can be verified using microseismic or other on-site monitoring data. The output data includes fracture geometry and conductivity, including but not limited to: fracture length, height, width, and permeability. Considering the high cost of on-site monitoring and the difficulty in obtaining or having limited samples of fracture effect evaluation parameters output by the model on-site, numerical simulation methods can be used to establish a large number of output model samples.

[0117] In some embodiments of this specification, the training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a target crack parameter determination model. This may include: inputting the training sample library into a pre-constructed combined neural network and a deep neural network for training to obtain a crack parameter determination model; after model optimization, k-fold cross-validation is used to validate the crack parameter determination model to obtain a target crack parameter determination model.

[0118] The hidden layers and number of neurons in a deep neural network can be optimized based on the prediction results during training. Model validation can utilize k-fold cross-validation to redivide the dataset, dividing all samples into k (k can be an integer from 2 to 10) equal-sized subsets. These k subsets are iterated sequentially, with each subset used as the validation set and all other samples as the training set for model training and evaluation. Finally, the average of the k evaluation metrics is used as the final evaluation metric to further optimize the model. The final target crack parameters are obtained to determine the model. Compared to simple cross-validation, k-fold cross-validation can improve model performance, allowing the neural network to apply samples more effectively and reducing overfitting.

[0119] In some embodiments of this specification, model evaluation metrics may also be set, including but not limited to root mean square error (RMSE) and correlation coefficient (R²). 2 The mean relative error (MRE) can be used to evaluate the fracture parameters of fracturing performance. This can be done by comparing the fracture parameters of models using deep neural networks and combined neural networks that consider physical constraints (referring to considering the second and third loss functions) with those of models using convolutional neural networks as input networks, purely data-driven deep neural networks, and models employing simple validation. The impact of combined neural networks, physical constraints, and k-fold cross-validation on the accuracy of model predictions can be compared and analyzed.

[0120] In some embodiments of this specification, after obtaining the fracture parameters of the target oil and gas reservoir, the method may further include: determining the fracture stimulation volume and fracture conductivity of the target oil and gas reservoir based on the fracture parameters of the target oil and gas reservoir and the field monitoring data; and calculating the fracture productivity and / or fracture economic indicators of the target oil and gas reservoir based on the fracture stimulation volume and the fracture conductivity.

[0121] Specifically, fracture parameters obtained from the model can be determined based on the target fracture parameters, the fracturing stimulation volume can be calculated, and combined with parameters such as fluid efficiency, fracturing effect evaluation indicators can be obtained. These evaluation indicators may include, but are not limited to, parameters such as fracture productivity, economic indicators, effective fracture stimulation volume, and fracture conductivity, thereby enabling the assessment of the fracturing effect on horizontal wells in the reservoir being evaluated.

[0122] The fracture network volume and fracture conductivity can be obtained from the fracture length, width, height, and permeability predicted by the model. Combining field monitoring data such as microseismic data, the affected area is divided into rectangles of equal size, with the outermost fracture point as the outer boundary of the affected volume. The sum of the areas of these rectangles, multiplied by their formation depth (fracture height), yields the affected volume. Further calculations of the effective affected volume and fracture conductivity are performed using methods including, but not limited to, analytical methods and discrete grid methods. Based on steady-state flow and continuous medium theory, simplified fracture treatments can be established by setting assumptions to create a fractured horizontal well productivity model, thus obtaining the fracture productivity. Economic indicators, including but not limited to net present value (NPV), can be calculated based on oil prices, fracturing costs, and fracture productivity.

[0123] A threshold for fracture stimulation volume is set based on fluid efficiency. Typical fluid efficiency ranges from 30% to 60%. If the assessed SRV (Surge Reduction Value) is less than the fracture volume achieved with 30% fluid efficiency, the stimulation effect is considered poor. The effectiveness of fracturing can also be judged based on the effective stimulation volume, fracture conductivity, production capacity, and the number of economic indicators. A larger effective stimulation volume indicates a better fracturing effect; greater fracture conductivity also indicates a better effect; higher production capacity also indicates a better effect; a net present value (NPV) greater than zero is considered a valid stimulation, and a higher NPV indicates a better effect.

[0124] In this embodiment, the fracture modification volume threshold can be calculated, multiple indicators can be used to evaluate the fracturing effect, and the horizontal well fracturing effect can be further evaluated based on the fracture parameters.

[0125] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. For details, please refer to the foregoing descriptions of the relevant processing embodiments; they will not be repeated here.

[0126] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0127] The above method will be described below with reference to a specific embodiment. However, it is worth noting that this specific embodiment is only for better illustration of this specification and does not constitute an improper limitation of this specification.

[0128] This specific embodiment provides a method for determining fracture parameters in horizontal well fracturing considering physical constraints. Please refer to... Figure 2 This diagram illustrates a flowchart of a method for evaluating the fracturing effect of horizontal wells using a deep neural network that considers physical constraints, according to an embodiment of the present invention. The method in this embodiment may include the following:

[0129] Step 1: Construct a deep learning-based model for determining target fracture parameters in horizontal wells of tight oil reservoirs (or a model for evaluating the fracturing effect of horizontal wells).

[0130] like Figure 2 As shown, the construction of the fracturing effect evaluation model may include: establishing a sample library, constructing a fracture parameter determination model, model optimization and verification, model effect evaluation, and fracturing effect evaluation. Figure 2 As shown, this combines neural networks and physically constrained deep neural networks.

[0131] The establishment of the sample library includes input data and output data, specifically including: obtaining input data for the selected reservoir, including static geological data and dynamic fracturing operation data, and may also include production data and monitoring data such as microseismic data.

[0132] The static geological data include: maximum and minimum horizontal geostress, reservoir pressure, Young's modulus of the rock, Poisson's ratio of the rock, reservoir porosity, and reservoir permeability.

[0133] The dynamic fracturing operation data includes: fracturing operation pressure, fluid injection volume, operation displacement and proppant concentration at different times for each fracturing stage of each well.

[0134] For static geological data, geological data is obtained from well logging curves, and the geological data for each fracturing segment is obtained by averaging according to formation depth. For dynamic fracturing operation data, preprocessing is required, including effective fracturing segment truncation, data denoising, and feature point extraction. A feature point data extraction method is established, the fracturing operation curve is truncated according to the fluid injection volume, the pressure data is denoised using a denoising method, and the intersection of the pressure point corresponding to the sand concentration change exceeding a certain threshold and the point where the pressure itself changes by an amplitude exceeding a certain threshold is taken as the feature pressure point. The displacement and sand concentration corresponding to the feature pressure point are used together as the input fracturing operation data.

[0135] The static geological data and dynamic fracturing construction data of the same fracturing section are combined to obtain the input data for the target fracture parameter determination model of different fracturing sections.

[0136] The method for acquiring the output data of the sample library specifically includes: based on pre-processed fracturing construction data and reservoir geological data, performing fracturing pressure analysis through the fracturing module of fracturing simulation software, and inverting to obtain the output data of dynamic fracture parameters during fracture propagation; or fitting production dynamics using production dynamic analysis cases, and inverting to obtain the output data of static fracture parameters after fracture closure; or diagnosing fracture parameters based on on-site fracturing monitoring technology, and obtaining comprehensive fracture parameter output data, including fracture geometry and conductivity, including but not limited to: fracture length, fracture height, fracture width, and permeability. A sample library is established using geological data and pre-processed fracturing construction data as model inputs and fracture evaluation parameters as model outputs.

[0137] Please refer to Figure 3 The diagram shows a structural schematic of the crack parameter determination model in an embodiment of this specification. Figure 3 As shown, the combined neural network is used for data input and consists of two neural networks connected in parallel. Different numbers of neurons are allocated according to the characteristics of static geological data and dynamic fracturing construction data. The deep neural network considering physical constraints consists of an input layer, hidden layers, an output layer, and a loss function considering physical constraints. The number of input, hidden, and output layers, as well as the number of neurons in each hidden layer, needs to be obtained by optimizing the model. The deep neural network calculates the results of the combined neural network to obtain the predicted results of the fracture evaluation parameters.

[0138] A combined neural network was constructed based on the characteristics of dynamic and static data: geological data served as one-dimensional input vector data, while dynamic fracturing operation data served as two-dimensional input vector data. The combined neural network consisted of two parallel neural networks. The first neural network, responsible for the geological data input, comprised a single layer containing only 8 neurons. The second neural network handled the hydraulic fracturing operation data input. Pressure, proppant concentration, and displacement were each allocated 2000 neurons as input layers. The second neural network also included 6 hidden layers, each containing 800 neurons, and an output layer with 100 neurons. The number of neurons in each layer of the combined neural network was also optimized.

[0139] like Figure 3 As shown, a deep neural network is established. Based on the output of the combined neural network, a suitable input layer is created. An appropriate number of hidden layers and neurons are selected. The model output is designed to represent the crack half-length, half-height, half-width, and permeability. The loss function is redefined. Based on the conventional loss function (mean squared error), the parameter range obtained from field monitoring data is added as field experience, and the crack propagation equation is added as part of the physical constraints.

[0140] Model optimization specifically includes optimizing the hidden layers and the number of neurons in the deep neural network based on the prediction results. Model validation involves re-dividing the dataset using k-fold cross-validation, dividing all samples into 10 equal-sized subsets. Each time, these 10 subsets are iterated, with the current subset used as the validation set and all other samples used as the training set for model training and evaluation. Finally, the average of the 10 evaluation metrics is used as the final evaluation metric to further optimize the model, resulting in the final fracturing effect evaluation model.

[0141] Model evaluation specifically includes setting model evaluation metrics. These metrics include, but are not limited to, root mean square error (RMSE) and correlation coefficient (R²). 2 The mean relative error (MRE) was used to evaluate the fracture parameters of fracturing performance. Based on the model evaluation metrics, the fracture parameter prediction accuracy of different models was compared using a controlled variable method. This involved comparing fracturing performance evaluation models composed of deep neural networks considering physical constraints and combined neural networks with models using convolutional neural networks as input, pure data-driven deep neural networks, and models employing simple validation. The influence of combined neural networks, physical constraints, and k-fold cross-validation on the accuracy of the model prediction results was analyzed.

[0142] Step 2: Obtain input data for the reservoir to be predicted. The input data includes static geological data and preprocessed dynamic fracturing operation data.

[0143] Step 3: Input the input data into the fracture parameter determination model to obtain fracture parameters: fracture parameters include the geometric dimensions and conductivity of the fracture, including but not limited to: fracture length, fracture height, fracture width and permeability.

[0144] Step 4: Use the obtained fracture parameters to evaluate the fracturing effect.

[0145] Based on the fracture length, width, height, and permeability predicted by the model, the fracture network volume and fracture conductivity can be obtained. Combining microseismic data, the affected area is divided into rectangles of equal size, with the outermost fracture point as the outer boundary of the affected volume. The areas of these rectangles are summed and multiplied by their formation depth (fracture height) to obtain the affected volume. Further calculations are made to determine the effective affected volume and fracture conductivity, using methods including, but not limited to, analytical methods and discrete grid methods. Based on steady-state flow and continuous medium theory, the fractures are simplified by setting assumptions to establish a fracturing horizontal well productivity model, thereby obtaining the fracture productivity. Economic indicators, including but not limited to net present value (NPV), are calculated based on oil prices, fracturing costs, and fracture productivity.

[0146] A threshold for fracture stimulation volume is set based on fluid efficiency. Typical fluid efficiency ranges from 30% to 60%. If the assessed SRV (Surge Reduction Value) is less than the fracture volume achieved with 30% fluid efficiency, the stimulation effect is considered poor. The effectiveness of fracturing can also be judged based on the effective stimulation volume, fracture conductivity, production capacity, and the number of economic indicators. A larger effective stimulation volume indicates a better fracturing effect; greater fracture conductivity also indicates a better effect; higher production capacity also indicates a better effect; a net present value (NPV) greater than zero is considered a valid stimulation, and a higher NPV indicates a better effect.

[0147] The method described in the above embodiments is applied to the evaluation of the fracturing effect of a horizontal well on a three-dimensional development platform for a tight oil reservoir in western China. The specific steps include the following steps.

[0148] Step 1: Establishing the sample library (acquiring input and output data).

[0149] Input data includes static geological data and dynamic fracturing operation data.

[0150] Geological data was obtained from well logging curves, and the average geological data for each fracturing segment was calculated based on formation depth. For fracturing operation data, a feature point data extraction method was established using MATLAB. The fracturing operation curve was truncated based on whether the fluid addition was zero. The pressure data was denoised using a wavelet denoising toolbox. A wavelet method was employed to denoise the fracturing pressure data, selecting bioorthogonal spline wavelets as the wavelet family to distinguish weak signals. Stein's Unbiased Risk Estimate was used for denoising, employing soft threshold constraints to decompose the 10-layer signal. The intersection of pressure points where sand concentration changes exceeded a certain threshold and points where pressure itself changes exceeded a certain threshold was taken as feature pressure points. The displacement and sand concentration corresponding to the feature pressure points were used as input data. The geological data and fracturing operation data of the same fracturing segment were combined to obtain the input data for the fracturing effect evaluation model of different fracturing segments.

[0151] The output data consists of crack parameters, including crack length, crack width, crack height, and permeability.

[0152] Fracturing pressure analysis was performed using the Kinetix module of Petrel software. A geomechanical model and fracturing operation plan were used to fit the fracturing pressure. After correction for in-situ stress and frictional resistance of the fracturing fluid and proppant, the fracture length, height, width, and permeability were obtained through inversion. To ensure the accuracy of the inverted data, the obtained fracture geometry parameters were compared with field microseismic data to ensure the effectiveness of the neural network samples.

[0153] A sample library was established using geological data and pre-processed fracturing construction data as model inputs and fracture assessment parameters as model outputs.

[0154] Step 2: Construct a model to determine crack parameters.

[0155] The crack parameter determination model consists of a combined neural network and a deep neural network that takes into account physical constraints.

[0156] The combined neural network consists of two parallel neural networks. The first neural network, containing only 8 neurons, is responsible for the input of geological data. The second neural network is responsible for the input of preprocessed dynamic fracturing data. Pressure, proppant concentration, and displacement are each allocated 2000 neurons as input layers. The second neural network also includes 6 hidden layers, each containing 800 neurons, and an output layer of 100 neurons. The model's input includes geological data and hydraulic fracturing operation data. The geological data includes reservoir pressure, maximum and minimum horizontal principal stresses, Poisson's ratio, Young's modulus, porosity, permeability, and oil saturation, allocated 8 neurons. The hydraulic fracturing operation data includes the changes in operation pressure, operation displacement, and sand concentration over time, allocated 2000 neurons according to the operation time. The neural network for hydraulic fracturing operation data is further divided into 6 hidden layers, each containing 800 neurons, with the final layer containing 100 neurons.

[0157] A deep neural network consists of one input layer, three hidden layers, and one output layer. Each hidden layer has 300 neurons, and the output layer has 4 neurons.

[0158] The physical constraints comprise a loss function composed of a conventional data-driven loss function, microseismic data, and a physical model. The loss function for the conventional data-driven model is the mean squared error (MSE). The microseismic data consists of fracture geometry parameters obtained through inversion from in-situ microseismic monitoring, providing a range of values ​​for these parameters; values ​​exceeding this range are considered unreasonable. The difference between the model's predictions and the KGD model's calculations is used as part of the loss function. The model is trained based on this loss function, and the KGD model's calculations constrain the model, preventing unreasonable predictions from being made by the deep neural network model.

[0159] Step 3: Model optimization and verification to obtain the optimal fracturing effect evaluation model.

[0160] The main optimization focused on the deep neural network. Based on the calculation results, the number of hidden layers and the number of neurons in each hidden layer were first optimized, determining the optimal number of hidden layers to be 5 and the number of neurons in each hidden layer to be 60. Next, k-fold cross-validation was used to re-divide the dataset, dividing all samples into 10 equal-sized subsets. These 10 subsets were iterated sequentially, with the current subset used as the validation set and all other samples used as the training set for model training and evaluation. Finally, the average of the 10 evaluation metrics was used as the final evaluation metric to further optimize the model, resulting in the final fracturing effect evaluation model.

[0161] Step 4: Model performance evaluation.

[0162] Define model evaluation metrics, including root mean square error (RMSE) and correlation coefficient (R²). 2 The mean relative error (MRE) was calculated. Based on the model evaluation index, the accuracy of crack parameters of different models was evaluated using the control variable method. The influence of combined neural networks, physical constraints and k-fold cross-validation on the accuracy of model prediction results was compared and analyzed.

[0163] Step 5: Evaluation of fracturing effect.

[0164] The fracture network volume was calculated based on the fracture length, height, width, and permeability parameters predicted by the model. By comparing and verifying this with microseismic data, the effective stimulation volume of one horizontal well could be determined. Please refer to [reference needed]. Figure 4 This diagram illustrates the effective stimulation volume of the horizontal well fracturing section calculated in this embodiment. Figure 4 As shown, the effective modification volume range is 43.98 × 10⁻⁶. 4 -148.17×10 4 m 3 The renovation was quite effective.

[0165] Crack parameters were predicted using the methods described in the embodiments of this specification, and the results were compared with actual results. Please refer to... Figures 5 to 8 The diagram illustrates a comparison between the predicted and actual (true) values ​​of the crack parameters in this embodiment. Figures 5 to 8 As shown, the relative errors, RMSE, MRE, and R of the fracture length, fracture height, fracture width, and fracture permeability were calculated respectively. 2 The accuracy of the model's predictions was evaluated using various metrics. Results showed that the absolute relative errors of the crack geometry and crack permeability were generally less than 10%. The combined neural network reduced the root mean square error (RMSE) of the model by 71.9%, and the addition of physical constraints reduced the irrationality of the predictions, resulting in a 56% reduction in the RMSE. k-fold cross-validation, by improving the irrationality of the training data, reduced the RMSE of crack parameters by 42%–80% compared to simple cross-validation.

[0166] Based on the same inventive concept, this specification also provides a fracture parameter determination device for horizontal well fracturing, as described in the following embodiments. Since the principle of the fracture parameter determination device for horizontal well fracturing is similar to that of the fracture parameter determination method for horizontal well fracturing, the implementation of the fracture parameter determination device for horizontal well fracturing can refer to the implementation of the fracture parameter determination method for horizontal well fracturing, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated. Figure 9 This is a structural block diagram of a fracture parameter determination device for horizontal well fracturing according to an embodiment of this specification, such as... Figure 9 As shown, it includes: an acquisition module 901 and an input module 902. The structure is described below.

[0167] The acquisition module 901 is used to acquire static geological data and dynamic fracturing operation data of the target oil and gas reservoir.

[0168] The input module 902 is used to input the static geological data and the dynamic fracturing construction data into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

[0169] The target fracture parameter determination model includes a combined neural network and a deep neural network. The combined neural network is used as input for the static geological data and the dynamic fracturing construction data. The deep neural network is used to calculate the fracture parameters from the data input by the combined neural network. The loss function of the deep neural network includes a data-driven first loss function, a second loss function constructed based on the field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir.

[0170] In some embodiments of this specification, the second loss function may be determined by: acquiring field monitoring data of the target oil and gas reservoir, performing inversion on the field monitoring data to obtain monitoring data; constructing the second loss function based on the monitoring data; the monitoring data includes the range of fracture parameter values ​​obtained by inversion.

[0171] In some embodiments of this specification, the third loss function may be determined by: determining the fracture type based on the dynamic fracturing construction data, selecting the corresponding fracture propagation equation according to the fracture type, and constructing the third loss function based on the fracture propagation equation.

[0172] In some embodiments of this specification, the input module may be specifically used to: preprocess the dynamic fracturing construction data to obtain preprocessed dynamic fracturing construction data; the preprocessing includes at least one of the following: fracturing segment truncation, data denoising, and feature point extraction; input the static geological data and the preprocessed dynamic fracturing construction data into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

[0173] In some embodiments of this specification, the dynamic fracturing operation data may include fracturing operation pressure data, fluid injection volume, operation displacement, and proppant concentration of the horizontal well at various times. Correspondingly, the input module may be specifically used to: extract the fracturing operation curve generated based on the fracturing operation pressure data of the horizontal well at various times according to the fluid injection volume to obtain a fracturing segment; perform noise reduction processing on the fracturing operation pressure data corresponding to the fracturing segment, and determine the intersection of the fracturing operation pressure point where the sand concentration change exceeds a first preset threshold and the point where the fracturing operation pressure data change exceeds a second preset threshold as the characteristic pressure point, thereby obtaining the characteristic pressure point corresponding to the fracturing segment; and determine the characteristic pressure point corresponding to each fracturing segment, the operation displacement corresponding to the characteristic pressure point, and the sand concentration as the preprocessed dynamic fracturing operation data corresponding to the fracturing segment.

[0174] In some embodiments of this specification, the target fracture parameter determination model may be constructed in the following manner: establishing a training sample library; the training sample library includes an input sample set and an output sample set; the input data in the input sample set includes static geological data and corresponding dynamic construction data corresponding to each fracturing segment in multiple fracturing segments; the output data in the output sample set includes fracture evaluation parameters corresponding to each input data in the input sample set; the training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain the target fracture parameter determination model.

[0175] In some embodiments of this specification, the output data may be determined by one of the following methods: performing fracturing construction pressure analysis on the input data in the input sample set to obtain output data of dynamic fracture parameters during fracture propagation; fitting production dynamics using production dynamic analysis cases to obtain output data of static fracture parameters after fracture closure; or diagnosing and correcting fracture parameters based on on-site fracturing monitoring technology to obtain output data of comprehensive fracture parameters.

[0176] In some embodiments of this specification, the training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a target crack parameter determination model. This may include: inputting the training sample library into a pre-constructed combined neural network and a deep neural network for training to obtain a crack parameter determination model; and validating the crack parameter determination model using k-fold cross-validation to obtain a target crack parameter determination model.

[0177] In some embodiments of this specification, the apparatus may further include a calculation module, which is specifically used to: determine the fracture stimulation volume and fracture conductivity of the target oil and gas reservoir based on the fracture parameters of the target oil and gas reservoir and the field monitoring data; and calculate the fracture productivity and / or fracture economic indicators of the target oil and gas reservoir based on the fracture stimulation volume and the fracture conductivity.

[0178] From the above description, it can be seen that the embodiments of this specification achieve the following technical effects: Using dynamic fracturing construction data that controls fracture parameters as part of the input makes fracture prediction more accurate. Using a combined neural network enables multi-dimensional data input, solving the problem of geological data being ignored due to different input data dimensions and excessive fracturing construction data. By adding a second loss function based on field monitoring data and a third loss function based on the fracture propagation equation to the conventional loss function, physical constraints can be applied to the deep learning process. This makes the deep neural network not simply data-driven but physically meaningful, considering actual fracture parameters. Different loss function parts further constrain the prediction results from different angles, thus enabling accurate prediction of fracture parameters obtained from three-dimensional fracturing of horizontal wells in tight oil and gas reservoirs. Compared with conventional fracture parameter determination methods, this scheme can significantly reduce the workload and economic cost of fracture parameter evaluation, achieving real-time, accurate, and rapid prediction of fracture parameters. Furthermore, this scheme can predict fractures in both vertical and conventional horizontal wells, requiring only adjustments to the model optimization process, thus compensating for the shortcomings of traditional fracture prediction methods for vertical and conventional horizontal wells.

[0179] This specification also provides a computer device, which can be found in the following description. Figure 10 The diagram shown illustrates the computer device structure for a horizontal well fracturing fracture parameter determination method provided in the embodiments of this specification. Specifically, the computer device may include an input device 11, a processor 12, and a memory 13. The memory 13 stores processor-executable instructions. When the processor 12 executes these instructions, it implements the steps of the horizontal well fracturing fracture parameter determination method described in any of the above embodiments.

[0180] In this embodiment, the input device can specifically be one of the main devices for information exchange between the user and the computer system. The input device may include a keyboard, mouse, camera, scanner, light pen, handwriting input tablet, voice input device, etc.; the input device is used to input raw data and programs for processing these data into the computer. The input device can also receive data transmitted from other modules, units, and devices. The processor can be implemented in any suitable manner. For example, the processor can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc. The memory can specifically be a memory device used to store information in modern information technology. The memory can include multiple layers; in digital systems, anything that can store binary data can be considered memory; in integrated circuits, a circuit without physical form but with storage function is also called memory, such as RAM, FIFO, etc.; in a system, a storage device with physical form is also called memory, such as a memory stick, TF card, etc.

[0181] In this embodiment, the specific functions and effects implemented by the computer device can be explained in comparison with other embodiments, and will not be repeated here.

[0182] This specification also provides a computer storage medium for a method of determining fracture parameters based on horizontal well fracturing, wherein the computer storage medium stores computer program instructions that, when executed, implement the steps of the method for determining fracture parameters of horizontal well fracturing described in any of the above embodiments.

[0183] In this embodiment, the storage medium includes, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), cache, hard disk drive (HDD), or memory card. The memory can be used to store computer program instructions. The network communication unit can be an interface configured according to standards specified in the communication protocol for network connection communication.

[0184] In this embodiment, the specific functions and effects implemented by the program instructions stored in the computer storage medium can be explained by comparison with other embodiments, and will not be repeated here.

[0185] Obviously, those skilled in the art will understand that the modules or steps of the embodiments described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of this specification are not limited to any particular combination of hardware and software.

[0186] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and applications beyond the provided examples will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this specification should not be determined by reference to the above description, but rather by reference to the foregoing claims and the full scope of their equivalents.

[0187] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to the embodiments described herein by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A method for determining fracture parameters in horizontal well fracturing, characterized in that, include: Acquire static geological data and dynamic fracturing operation data of the target oil and gas reservoir; The static geological data and the dynamic fracturing construction data are input into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir; The target fracture parameter determination model includes a combined neural network and a deep neural network. The combined neural network is used as input for the static geological data and the dynamic fracturing operation data. The deep neural network is used to calculate the fracture parameters from the data input by the combined neural network. The loss function of the deep neural network includes a data-driven first loss function, a second loss function constructed based on the field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir. The loss function of the deep neural network is expressed by the following formula: in, Loss all This represents the loss function of the deep neural network. α 1. α 2 and α 3 represents the weighting coefficient; N Indicates the number of samples; MSE stands for Mean Squared Error, and its expression is as follows: in, It is a predicted value. This is the actual value; Loss f The second loss function is expressed as follows: Among them, subscript f Indicates crack parameters, y f This represents the predicted results of the crack parameters. y fmin and y fmax These represent the minimum and maximum values ​​of the crack geometric parameters obtained from field monitoring data, respectively, with subscripts indicating the values. j =1, 2, and 3 represent the crack length, crack height, and crack width, respectively. i Indicates the number of samples; Loss 2D This represents the third loss function.

2. The method for determining fracture parameters in horizontal well fracturing according to claim 1, characterized in that, The second loss function is determined by: acquiring field monitoring data of the target oil and gas reservoir; performing inversion on the field monitoring data to obtain monitoring data; constructing the second loss function based on the monitoring data; the monitoring data includes the range of fracture parameter values ​​obtained from the inversion; and / or The third loss function is determined by: determining the fracture type based on the dynamic fracturing construction data, selecting the corresponding fracture propagation equation according to the fracture type, and constructing the third loss function based on the fracture propagation equation.

3. The method for determining fracture parameters in horizontal well fracturing according to claim 1, characterized in that, The static geological data and the dynamic fracturing operation data are input into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir, including: The dynamic fracturing construction data is preprocessed to obtain preprocessed dynamic fracturing construction data; the preprocessing includes at least one of the following: fracturing segment truncation, data denoising, and feature point extraction. The static geological data and the preprocessed dynamic fracturing construction data are input into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir.

4. The method for determining fracture parameters in horizontal well fracturing according to claim 3, characterized in that, The dynamic fracturing operation data includes fracturing operation pressure data, fluid injection volume, operation displacement, and proppant concentration at various times during multiple moments in the horizontal well; correspondingly, the dynamic fracturing operation data is preprocessed to obtain preprocessed dynamic fracturing operation data, including: Based on the fluid addition volume, the fracturing operation curve generated from the fracturing operation pressure data of the horizontal well at various times is truncated to obtain the fracturing section. The fracturing pressure data corresponding to the fracturing section is subjected to noise reduction processing. The intersection of the fracturing construction pressure point corresponding to the sand concentration change exceeding the first preset threshold and the point where the fracturing construction pressure data change exceeds the second preset threshold is determined as the characteristic pressure point, thus obtaining the characteristic pressure point corresponding to the fracturing section. The characteristic pressure points corresponding to each fracturing segment, the construction displacement and sand concentration corresponding to the characteristic pressure points are determined as the pre-processed dynamic fracturing construction data corresponding to the fracturing segment.

5. The method for determining fracture parameters in horizontal well fracturing according to claim 1, characterized in that, The target crack parameter determination model is constructed in the following way: A training sample library is established, which includes an input sample set and an output sample set. The input data in the input sample set includes static geological data and dynamic construction data corresponding to each fracturing segment in multiple fracturing segments. The output data in the output sample set includes fracture evaluation parameters corresponding to each input data in the input sample set. The training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a target crack parameter determination model.

6. The method for determining fracture parameters in horizontal well fracturing according to claim 5, characterized in that, The output data is determined by one of the following methods: The input data in the input sample set is subjected to fracturing construction pressure analysis, and the output data of dynamic fracture parameters during fracture propagation are obtained by inversion. By using production dynamics analysis to fit production dynamics, output data of static crack parameters after crack closure are obtained through inversion. Based on the on-site fracturing monitoring technology, the fracture parameters are diagnosed and corrected to obtain the output data of comprehensive fracture parameters.

7. The method for determining fracture parameters in horizontal well fracturing according to claim 5, characterized in that, The training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a target crack parameter determination model, including: The training sample library is input into a pre-constructed combined neural network and a deep neural network for training to obtain a crack parameter determination model; The crack parameter determination model was validated using k-fold cross-validation to obtain the target crack parameter determination model.

8. The method for determining fracture parameters in horizontal well fracturing according to claim 1, characterized in that, After obtaining the fracture parameters of the target oil and gas reservoir, the process also includes: Based on the fracture parameters of the target oil and gas reservoir and the field monitoring data, the fracture stimulation volume and fracture conductivity of the target oil and gas reservoir are determined. Based on the fracture modification volume and the fracture conductivity, calculate the fracture productivity and / or fracture economic indicators of the target oil and gas reservoir.

9. A device for determining fracture parameters in horizontal well fracturing, characterized in that, include: The acquisition module is used to acquire static geological data and dynamic fracturing operation data of the target oil and gas reservoir; The input module is used to input the static geological data and the dynamic fracturing construction data into the target fracture parameter determination model to obtain the fracture parameters of the target oil and gas reservoir. The target fracture parameter determination model includes a combined neural network and a deep neural network. The combined neural network is used as input for the static geological data and the dynamic fracturing operation data. The deep neural network is used to calculate the fracture parameters from the data input by the combined neural network. The loss function of the deep neural network includes a data-driven first loss function, a second loss function constructed based on the field monitoring data of the target oil and gas reservoir, and a third loss function constructed based on the fracture propagation equation corresponding to the target oil and gas reservoir. The loss function of the deep neural network is expressed by the following formula: in, Loss all This represents the loss function of the deep neural network. α 1. α 2 and α 3 represents the weighting coefficient; N Indicates the number of samples; MSE stands for Mean Squared Error, and its expression is as follows: in, It is a predicted value. This is the actual value; Loss f The second loss function is expressed as follows: Among them, subscript f Indicates crack parameters, y f This represents the predicted results of the crack parameters. y fmin and y fmax These represent the minimum and maximum values ​​of the crack geometric parameters obtained from field monitoring data, respectively, with subscripts indicating the values. j =1, 2, and 3 represent the crack length, crack height, and crack width, respectively. i Indicates the number of samples; Loss 2D This represents the third loss function.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 8.

Images