Method and system for predicting fracturing effect of heterogeneous reservoir while taking into account fracturing process, and medium

Abstract

A method and system for predicting the fracturing effect of a heterogeneous reservoir while taking into account a fracturing process, and a medium, which belong to the technical field of oil and gas field development. The method comprises: collecting geological parameters of a heterogeneous reservoir, fracturing process parameters, and characteristic parameters related to a fracturing effect, and using the parameters as a data set; using the data set to construct a fracturing-effect prediction model for the heterogeneous reservoir; and inputting the fracturing process parameters and the geological parameters into the fracturing-effect prediction model, such that the prediction model outputs fracturing effect parameters, wherein the fracturing process parameters adapted to the characteristics of the heterogeneous reservoir are optimally selected by means of a multi-objective optimization algorithm and on the basis of the constructed fracturing-effect prediction model, the heterogeneity of the heterogeneous reservoir, and permeability differences among layers. The method can accurately predict the fracturing effect of a heterogeneous reservoir, and effectively reflect the influence of different fracturing process parameters on fracture formation and propagation.

Description

Methods, systems, and media for predicting the fracturing effect of heterogeneous reservoirs considering fracturing technology Technical Field

[0001] This invention belongs to the field of oil and gas field development technology, specifically relating to a method, system, and computer-readable storage medium for predicting the fracturing effect of heterogeneous reservoirs considering fracturing technology. This invention also relates to a real-time prediction method, real-time prediction device, and computer-readable storage medium for fracture propagation. Background Technology

[0002] In the field of oil and gas development, a reservoir is an underground rock layer capable of storing oil and gas and allowing them to flow under certain conditions. Reservoirs can be classified in various ways, the most common being based on rock type, including clastic reservoirs and carbonate reservoirs. Clastic reservoirs are formed by the compaction and cementation of rock fragments (such as quartz and feldspar), and are the most common type of reservoir, typically represented by thick sandstone and conglomerate layers. Carbonate reservoirs are formed by chemical or biochemical sedimentation, accounting for nearly 40% of global oil and gas reserves, and often form large and super-large oil and gas fields. As the exploration and development of conventional oil and gas resources continues to deepen globally, the focus of the oil and gas industry is increasingly shifting to more complex and challenging unconventional resource areas. For example, heterogeneous reservoirs are an important type of unconventional reservoir with enormous resource potential. Heterogeneous reservoirs refer to complex stratigraphic systems formed by the frequent, rhythmic alternation of thin (usually less than 1 meter) reservoir and non-reservoir layers in the vertical direction. In heterogeneous reservoirs, reservoirs (such as fine sandstone and siltstone) possess certain porosity and permeability, serving as the storage space and migration pathway for oil and gas. Non-reservoir layers (mainly mudstone, shale, or tight carbonate rocks) constitute interlayers or partitions, playing a crucial role in blocking vertical fluid flow. Heterogeneous reservoirs are characterized by small single-layer thickness, strong heterogeneity, and large interlayer stress differences. These characteristics present numerous challenges to the application of traditional fracturing techniques in heterogeneous reservoirs. Hydraulic fracturing, as a key technology for improving oil and gas recovery in heterogeneous reservoirs, plays a vital role in optimizing fracturing design and improving development efficiency through its performance prediction.

[0003] Currently, there are two main methods for predicting the fracturing effect of heterogeneous reservoirs: the empirical method and the method based on geological modeling and numerical simulation. The empirical method is based on field construction experience and historical data of similar reservoirs. It predicts the fracturing effect by summarizing empirical rules. Its disadvantages are: (1) strong subjectivity: it relies too much on the experience of engineers and lacks systematicity and scientific basis; (2) poor adaptability: it is difficult to accurately apply to complex heterogeneous reservoirs, especially when the geological conditions are very different; (3) inability to quantify fracture behavior: the prediction of fracture propagation process and conductivity is relatively vague and lacks precision. The geological modeling and numerical simulation method is to establish a geological model (such as finite element or finite difference model), combine reservoir physical parameters (such as porosity, permeability, and stress field distribution) and fracturing design parameters to simulate the process of fracture generation and propagation. Its disadvantages are: (1) High computational cost: the computation of complex models is time-consuming and requires high-performance computing equipment; (2) High sensitivity to input parameters: the accuracy of model parameters is required, and heterogeneous reservoirs usually have large parameter uncertainties; (3) Difficult to predict in real time: it cannot quickly respond to the dynamic changes in the field fracturing process.

[0004] CN118966067A discloses a "method for optimizing hydraulic fracturing technology in thin interbedded reservoirs." This method establishes a formation structure model of thin interbedded reservoirs based on well logging interpretation data, conducts hydraulic fracturing numerical simulations using an orthogonal experimental numerical simulation scheme, and selects the optimal fracturing scheme. Those skilled in the art will understand that thin interbedded reservoirs are a special type of heterogeneous reservoir with extremely strong interlayer heterogeneity. While the method considers the special characteristics of thin interbedded reservoirs, it primarily focuses on optimizing the fracturing process and has limited predictive ability for fracturing effects.

[0005] Therefore, the urgent problem to be solved is to develop a scientific and effective method and device for predicting fracturing effects, as well as a real-time prediction method and device for fracture propagation. This method and device should comprehensively consider fracturing process parameters and geological characteristics to achieve accurate prediction of reservoir fracture formation and propagation, thereby optimizing fracturing design and improving resource extraction efficiency. This method and device are applicable to common clastic and carbonate reservoirs, and are particularly suitable for more complex and challenging heterogeneous reservoirs, such as thin interbedded layers. Summary of the Invention

[0006] The purpose of this invention is to address the problems of strong subjectivity, poor adaptability, inability to quantify fracture behavior in heterogeneous reservoirs, high computational costs, high sensitivity to input parameters, and difficulty in real-time prediction in the existing technologies. This invention provides a method, system, and medium for predicting the fracturing effect of heterogeneous reservoirs, taking into account the fracturing process. Furthermore, this invention provides a real-time prediction method, a real-time prediction device, and a computer-readable storage medium for fracture propagation. By combining real-time monitoring data and intelligent algorithms, this invention achieves accurate prediction of the formation and propagation process of reservoir fractures and efficient prediction of fracturing effects.

[0007] This invention is achieved through the following technical solution:

[0008] According to a first aspect of the present invention, a method for predicting the fracturing effect of heterogeneous reservoirs considering fracturing processes is provided, comprising the following steps:

[0009] Geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs are collected as a dataset;

[0010] A fracturing effect prediction model for heterogeneous reservoirs was constructed using a dataset.

[0011] The fracturing process parameters and geological parameters are input into the fracturing effect prediction model, and the prediction model outputs the fracturing effect parameters. Based on the constructed fracturing effect prediction model, according to the heterogeneity of the heterogeneous reservoir and the permeability difference of each layer, the fracturing process parameters adapted to the characteristics of the heterogeneous reservoir are optimized and selected through a multi-objective optimization algorithm.

[0012] In an embodiment according to a first aspect of the invention, the acquired geological parameters include interlayer thickness, porosity, and permeability of the heterogeneous reservoir; and / or

[0013] The collected fracturing process parameters include fracturing fluid viscosity, sand ratio, and pumping rate; and / or

[0014] The collected characteristic parameters related to fracturing effect include fracture propagation range and fracture conductivity.

[0015] In an embodiment of the first aspect of the present invention, the method for acquiring fracturing process parameters includes:

[0016] A three-dimensional geological model of the heterogeneous reservoir is constructed based on the interlayer thickness, porosity, permeability, fracture density, location of the top and bottom boundaries of the formation, rock mechanical parameters, fracturing fluid viscosity, sand ratio, and pumping rate.

[0017] In an embodiment of the first aspect of the present invention, optimizing the selection of fracturing process parameters using a multi-objective optimization algorithm includes:

[0018] (1) Establish the crack propagation range Lf The objective function is expressed in the following formula:

[0019] Where α is a constant, Q p Here, μ is the pumping rate, μ is the fracturing fluid viscosity, and E is the equivalent Young's modulus of the heterogeneous reservoir. E is calculated using the following formula:

[0020] Where E is the Young's modulus of the heterogeneous reservoir, and ν is the Poisson's ratio of the heterogeneous reservoir;

[0021] (2) The objective function for the fracture conductivity F is established as follows:

[0022] Among them, C p Given the sand ratio, β is a constant, and P... p E is the pumping pressure, and E is the equivalent Young's modulus of the heterogeneous reservoir.

[0023] (3) Generate an initial population P0 containing N individuals, where the fracturing process parameters for each individual are expressed as follows: Where, μ i , Let X represent the fracturing fluid viscosity, sand ratio, pumping rate, and pumping pressure of the i-th individual, respectively. i The objective function values ​​f1 for the crack propagation range and f2 for the crack conductivity are calculated as follows:

[0024] (4) Perform non-dominated ranking on each individual in the population, and select individuals A and B that are in a non-dominated relationship; divide the individuals into different levels according to the non-dominated relationship;

[0025] Calculate the crowding distance d for each non-dominated individual i. i :

[0026] Where M is the number of objective functions;

[0027] For individuals within each non-dominated layer, sort them from largest to smallest by crowding distance;

[0028] (5) Select suitable individuals based on non-dominated ordination and crowding distance, and perform crossover and mutation operations;

[0029] The crossover operation is represented by the following formula:

[0030] Where λ is a uniformly distributed random factor, typically ranging from [-1, 1]; r is a random number in the range [0, 1]; p c The crossover probability has a value range of [0, 1].

[0031] The mutation operation is represented by the formula: X j =X j +N(0,σ)

[0032] Where N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ;

[0033] The current population P t and the newly generated offspring population Q t Merge to form a new population R t The merged population is sorted non-dominated and the top N individuals are selected to form a new population P. t+1 Repeat the above steps until the predetermined number of iterations or convergence conditions are reached, and finally obtain the fracturing process parameters.

[0034] In an embodiment of the first aspect of the present invention, the method for acquiring characteristic parameters related to fracturing effect includes:

[0035] Fracturing operations are carried out using fracturing process parameters adapted to the characteristics of the heterogeneous reservoir, and fracturing monitoring data are collected through a real-time sensor array at the fracturing site.

[0036] The collected fracturing monitoring data were denoised and normalized to obtain preprocessed fracturing monitoring data.

[0037] A joint feature extraction method based on PCA principal component analysis and SVM support vector machine was adopted to extract features related to fracturing effect from the preprocessed fracturing monitoring data, including fracture propagation range and fracture conductivity.

[0038] In an embodiment of the first aspect of the present invention, a fracturing effect prediction model for heterogeneous reservoirs is constructed using a dataset, specifically including:

[0039] Construct a fracturing effect prediction model;

[0040] A fracturing effect prediction model was constructed by training the dataset, and the trained fracturing effect prediction model was obtained.

[0041] The trained fracturing effect prediction model was validated and optimized to obtain the optimized fracturing effect prediction model.

[0042] In an embodiment of the first aspect of the present invention, the constructed fracturing effect prediction model is a multi-input multi-output structure, wherein,

[0043] The input variables of the model include: interlayer thickness of heterogeneous reservoirs, permeability, porosity, fracturing fluid viscosity, pumping rate, and sand ratio;

[0044] The model's output variables include: crack propagation range and crack conductivity.

[0045] In an embodiment of the first aspect of the present invention, the model training employs a gradient boosting decision tree algorithm to obtain a trained prediction model.

[0046] In an embodiment of the first aspect of the present invention, the trained prediction model is validated and optimized to obtain an optimized prediction model, the specific operations of which include:

[0047] The prediction results of the trained prediction model were compared with the actual fracturing effect data, and the training prediction model was verified by error analysis.

[0048] The Bayesian optimization algorithm is used to optimize the trained prediction model in order to minimize the prediction error and improve the prediction accuracy and adaptability of the model, resulting in an optimized prediction model.

[0049] According to a second aspect of the present invention, a system for predicting the fracturing effect of heterogeneous reservoirs considering fracturing processes is provided, the system comprising:

[0050] The parameter acquisition unit is used to collect geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs as a dataset.

[0051] The prediction model acquisition unit is used to construct a fracturing effect prediction model for heterogeneous reservoirs using a dataset.

[0052] The prediction unit is used to input fracturing process parameters and geological parameters into the fracturing effect prediction model. The prediction model outputs fracturing effect parameters. Based on the constructed fracturing effect prediction model, according to the heterogeneity of the heterogeneous reservoir and the permeability difference of each layer, the fracturing process parameters adapted to the characteristics of the heterogeneous reservoir are optimized and selected through a multi-objective optimization algorithm.

[0053] In an embodiment of the second aspect of the invention, the acquired geological parameters include interlayer thickness, porosity, and permeability of the heterogeneous reservoir; and / or

[0054] The collected fracturing process parameters include fracturing fluid viscosity, sand ratio, and pumping rate; and / or

[0055] The collected characteristic parameters related to fracturing effect include fracture propagation range and fracture conductivity.

[0056] In an embodiment of the second aspect of the present invention, optimizing the selection of fracturing process parameters adapted to the heterogeneous reservoir characteristics using the multi-objective optimization algorithm includes:

[0057] (1) Establish the crack propagation range L fThe objective function is expressed in the following formula:

[0058] Where α is a constant, Q p Here, μ is the pumping rate, μ is the fracturing fluid viscosity, and E is the equivalent Young's modulus of the heterogeneous reservoir. E is calculated using the following formula:

[0059] Where E is the Young's modulus of the heterogeneous reservoir, and ν is the Poisson's ratio of the heterogeneous reservoir;

[0060] (2) The objective function for the fracture conductivity F is established as follows:

[0061] Among them, C p Given the sand ratio, β is a constant, and P... p E is the pumping pressure, and E is the equivalent Young's modulus of the heterogeneous reservoir.

[0062] (3) Generate an initial population P0 containing N individuals, where the fracturing process parameters for each individual are expressed as follows: Where, μ i , Let X represent the fracturing fluid viscosity, sand ratio, pumping rate, and pumping pressure of the i-th individual, respectively. i The objective function values ​​f1 for the crack propagation range and f2 for the crack conductivity are calculated as follows:

[0063] (4) Perform non-dominated ranking on each individual in the population, and select individuals A and B that are in a non-dominated relationship; divide the individuals into different levels according to the non-dominated relationship;

[0064] Calculate the crowding distance d for each non-dominated individual i. i :

[0065] Where M is the number of objective functions;

[0066] For individuals within each non-dominated layer, sort them from largest to smallest by crowding distance;

[0067] (5) Select suitable individuals based on non-dominated ordination and crowding distance, and perform crossover and mutation operations;

[0068] The crossover operation is represented by the following formula:

[0069] Where λ is a uniformly distributed random factor, typically ranging from [-1, 1]; r is a random number in the range [0, 1]; p c The crossover probability has a value range of [0, 1].

[0070] The mutation operation is represented by the formula: X j =X j +N(0,σ)

[0071] Where N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ;

[0072] The current population P t and the newly generated offspring population Q t Merge to form a new population R t The merged population is sorted non-dominated and the top N individuals are selected to form a new population P. t+1 Repeat the above steps until the predetermined number of iterations or convergence conditions are reached, and finally obtain the fracturing process parameters.

[0073] In an embodiment according to a second aspect of the present invention, the parameter acquisition unit is configured to perform the following operations:

[0074] Fracturing operations are carried out using fracturing process parameters adapted to the characteristics of the heterogeneous reservoir, and fracturing monitoring data are collected through a real-time sensor array at the fracturing site.

[0075] The collected fracturing monitoring data were denoised and normalized to obtain preprocessed fracturing monitoring data.

[0076] A joint feature extraction method based on PCA principal component analysis and SVM support vector machine was adopted to extract features related to fracturing effect from the preprocessed fracturing monitoring data, including fracture propagation range and fracture conductivity.

[0077] In an embodiment according to a second aspect of the invention, the prediction unit is configured to perform the following operations:

[0078] Construct a fracturing effect prediction model;

[0079] A fracturing effect prediction model was constructed by training the dataset, and the trained fracturing effect prediction model was obtained.

[0080] The trained fracturing effect prediction model was validated and optimized to obtain the optimized fracturing effect prediction model.

[0081] In an embodiment of the second aspect of the present invention, the fracturing effect prediction model is a multi-input multi-output structure, wherein,

[0082] The input variables of the model include: interlayer thickness of heterogeneous reservoirs, permeability, porosity, fracturing fluid viscosity, pumping rate, and sand ratio;

[0083] The model's output variables include: crack propagation range and crack conductivity.

[0084] In an embodiment of the second aspect of the present invention, the model training employs a gradient boosting decision tree algorithm to obtain a trained prediction model.

[0085] In an embodiment of the second aspect of the present invention, the trained prediction model is validated and optimized to obtain an optimized prediction model, including:

[0086] The prediction results of the trained prediction model were compared with the actual fracturing effect data, and the training prediction model was verified by error analysis.

[0087] The Bayesian optimization algorithm is used to optimize the trained prediction model in order to minimize the prediction error and improve the prediction accuracy and adaptability of the model, resulting in an optimized prediction model.

[0088] According to a third aspect of the present invention, a real-time prediction method for crack propagation is provided, the method comprising:

[0089] Sample data of geological parameters, pumping image parameters, and corresponding fracture propagation of heterogeneous reservoirs are obtained, wherein the pumping image parameters are images obtained by transforming the pumping parameters;

[0090] The geological parameters are fused with the pumping image parameters to obtain fused feature data;

[0091] A prediction model is constructed based on machine learning algorithms, with the fused feature data as input and the crack propagation amount as output, and the crack propagation prediction model is obtained by training with sample data.

[0092] The fracture propagation prediction model is used to predict fracture propagation in the heterogeneous reservoir in real time.

[0093] In an embodiment of the third aspect of the present invention, the step of acquiring sample data of geological parameters, pumping image parameters, and corresponding fracture propagation includes:

[0094] Geological parameters, pumping parameters, and fracture propagation corresponding to the pumping parameters are obtained. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure.

[0095] The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

[0096] In an embodiment of the third aspect of the present invention, the step of converting the pumping parameters into an image matrix capable of characterizing temporal changes during the pumping process includes:

[0097] The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

[0098] In an embodiment of the third aspect of the present invention, the step of training a crack propagation prediction model using sample data includes:

[0099] The objective function of the initial prediction model is constructed based on the relative error between the predicted crack propagation amount output by the initial prediction model and the sample data of crack propagation amount.

[0100] The initial prediction model is iterated based on the objective function, and the parameters of the initial prediction model are adjusted.

[0101] When the objective function reaches its minimum value, the corresponding parameters are determined as the objective parameters of the initial prediction model.

[0102] The crack propagation prediction model is determined based on the target parameters.

[0103] According to a fourth aspect of the present invention, a real-time prediction device for crack propagation is provided, comprising:

[0104] The acquisition module is used to acquire geological parameters, pumping image parameters, and sample data of fracture propagation corresponding to the heterogeneous reservoir, wherein the pumping image parameters are images obtained based on the pumping parameters.

[0105] The data fusion module is used to fuse the geological parameters and pumping image parameters to obtain fused feature data;

[0106] The model building module is used to build an initial prediction model based on machine learning algorithms. The fused feature data is used as the input of the initial prediction model, the crack expansion amount is used as the output of the initial prediction model, and the initial prediction model is trained using the corresponding sample data to obtain a crack expansion prediction model.

[0107] The prediction module is used to predict the fracture propagation of the heterogeneous reservoir in real time using the fracture propagation prediction model.

[0108] In an embodiment of the fourth aspect of the invention, the acquisition module is configured as follows:

[0109] Geological parameters, pumping parameters, and fracture propagation corresponding to the pumping parameters are obtained. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure.

[0110] The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

[0111] In an embodiment of the fourth aspect of the invention, the acquisition module is configured as follows:

[0112] The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

[0113] According to a fifth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing at least one computer program executable by a computer, the at least one computer program performing steps in the fracturing effect prediction method considering fracturing process according to the first aspect of the present invention when executed by the computer.

[0114] According to a sixth aspect of the present invention, a computer-readable storage medium is provided, characterized in that the computer-readable storage medium stores at least one computer program executable by a computer, which, when executed by the computer, performs a real-time prediction method for crack propagation according to a third aspect of the present invention.

[0115] Compared with the prior art, the beneficial effects of the present invention are:

[0116] This invention can accurately predict the fracturing effect of heterogeneous reservoirs and effectively reflect the influence of different fracturing process parameters on fracture formation and propagation.

[0117] This method, through intelligent prediction and error feedback mechanisms, can optimize fracturing process parameters in real time, ensuring the selection of the optimal construction scheme under different geological conditions, thereby improving fracturing effect and resource recovery rate.

[0118] This invention enables timely adjustment of parameters during construction through real-time monitoring and data processing, thereby reducing construction risks and minimizing economic losses caused by poor fracturing effects, and improving the overall economic benefits of the project.

[0119] This invention employs a real-time prediction method and device for fracture propagation. By combining pumping program data with geological parameters and utilizing a deep learning model for multimodal data fusion, it significantly improves the prediction accuracy of fracture propagation, more accurately reflecting fracture behavior under complex geological conditions. Furthermore, it uses on-site real-time monitoring equipment to collect pumping data and geological parameters, enabling real-time tracking of the fracture propagation process and providing timely decision support for operators in response to geological changes. During pumping operations, it can monitor changes in pumping parameters in real time and readjust fracture propagation prediction results based on new data, demonstrating good adaptability and flexibility, and optimizing the operation process. By converting pumping program data into image data and fusing it with geological parameters, it overcomes the limitations of traditional methods in data processing and analysis, enabling a more comprehensive understanding of the impact of the pumping process on fracture propagation. Attached Figure Description

[0120] Figure 1 is a flowchart of a method for predicting the fracturing effect of heterogeneous reservoirs considering fracturing process according to an embodiment of the present invention.

[0121] Figure 2 is a schematic diagram of heterogeneous reservoir geological modeling according to an embodiment of the present invention.

[0122] Figure 3 is a flowchart of a multi-objective optimization algorithm for selecting fracturing process parameters adapted to the characteristics of heterogeneous reservoirs according to an embodiment of the present invention.

[0123] Figure 4 is a graph of the simulation results of a prediction model according to an embodiment of the present invention, showing the optimal combination of process parameters.

[0124] Figure 5 shows a composite columnar section of a heterogeneous reservoir according to an embodiment of the present invention.

[0125] Figure 6 shows a schematic diagram of the simulation results of a partition in a heterogeneous reservoir according to an embodiment of the present invention.

[0126] Figure 7 shows a flowchart of a real-time prediction method for crack propagation according to an embodiment of the present invention.

[0127] Figure 8 shows a flowchart of a real-time prediction method for crack propagation according to another embodiment of the present invention.

[0128] Figure 9 shows an example of pumping parameters according to an embodiment of the present invention.

[0129] Figure 10 shows a schematic diagram of a real-time prediction device for crack propagation according to an embodiment of the present invention. Detailed Implementation

[0130] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0131] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0132] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0133] The present invention will now be described in further detail with reference to the accompanying drawings.

[0134] This invention provides a method for predicting the fracturing effect of heterogeneous reservoirs considering fracturing technology, as shown in Figure 1. The method includes the following steps:

[0135] The first step is to collect geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs as a dataset;

[0136] The second step is to use the dataset to build a prediction model for the fracturing effect of heterogeneous reservoirs;

[0137] The third step involves inputting the fracturing process parameters and geological parameters into the fracturing effect prediction model, which then outputs the fracturing effect parameters.

[0138] In one embodiment of the present invention, in the third step, based on the constructed fracturing effect prediction model, and according to the heterogeneity of the heterogeneous reservoir and the permeability difference of each layer, a multi-objective optimization algorithm is used to optimize and select fracturing process parameters that are suitable for the characteristics of the heterogeneous reservoir.

[0139] This invention can accurately predict the fracturing effect, especially in heterogeneous reservoirs, and effectively reflect the influence of different fracturing process parameters on fracture formation and propagation.

[0140] This method, through intelligent prediction and error feedback mechanisms, can optimize fracturing process parameters in real time, ensuring the selection of the optimal construction scheme under different geological conditions, thereby improving fracturing effect and resource recovery rate.

[0141] This invention enables timely adjustment of parameters during construction through real-time monitoring and data processing, thereby reducing construction risks and minimizing economic losses caused by poor fracturing effects, and improving the overall economic benefits of the project.

[0142] In one embodiment of the present invention, in the first step, geological parameters of the heterogeneous reservoir, fracturing process parameters, and characteristic parameters related to fracturing effect are collected as a dataset, wherein...

[0143] The collected geological parameters of heterogeneous reservoirs include the interlayer thickness, porosity, and permeability of the heterogeneous reservoirs. These geological parameters can be obtained through geophysical exploration, well logging, and experiments. These methods are all existing technologies and will not be elaborated here.

[0144] In one embodiment of the present invention, the collected fracturing process parameters include fracturing fluid viscosity, sand ratio, and pumping rate.

[0145] In one embodiment of the present invention, the characteristic parameters collected related to the fracturing effect include: fracture propagation range and fracture conductivity.

[0146] In one embodiment of the present invention, the specific method for collecting fracturing process parameters includes:

[0147] Based on the interlayer thickness, porosity, permeability, fracture density, location of the top and bottom boundaries of the formation, and rock mechanical parameters (including Young's modulus and Burson's ratio) of the heterogeneous reservoir, a three-dimensional geological model of the heterogeneous reservoir is constructed, thereby improving the accuracy of the model in terms of geological features. As shown in Figure 2, this is the three-dimensional geological model of the heterogeneous reservoir constructed in an embodiment of the present invention.

[0148] In one embodiment of the present invention, based on the constructed three-dimensional geological model of the heterogeneous reservoir, and according to the heterogeneity of the heterogeneous reservoir and the permeability differences of each layer, a multi-objective optimization algorithm is used to optimize and select fracturing process parameters that are suitable for the characteristics of the heterogeneous reservoir.

[0149] In one embodiment of the present invention, Figure 3 is a flowchart illustrating the optimization of fracturing process parameters adapted to the characteristics of heterogeneous reservoirs using a multi-objective optimization algorithm according to an embodiment of the present invention. Referring to Figure 3, the specific operation process of the multi-objective optimization algorithm includes:

[0150] (1) Establish the crack propagation range L f The objective function is expressed in the following formula:

[0151] Where α is a constant, Q p Here, μ is the pumping rate, μ is the fracturing fluid viscosity, and E is the equivalent Young's modulus of the heterogeneous reservoir. E can be calculated using the following formula:

[0152] Where E is the Young's modulus of the heterogeneous reservoir, and ν is the Poisson's ratio of the heterogeneous reservoir;

[0153] (2) Establish the objective function for the fracture conductivity F. The specific operations include:

[0154] The conductivity F of a fracture mainly depends on the fracture width w f And sand ratio C p Specifically: F = C p ·w f

[0155] Among them, C p For sand ratio, w f The crack width can be expressed as:

[0156] The objective function for obtaining the fracture conductivity F is:

[0157] Where β is a constant, P p Where E is the pumping pressure, and E is the equivalent Young's modulus of the heterogeneous reservoir, which can be calculated using the following formula:

[0158] Where E is the Young's modulus of the heterogeneous reservoir, and ν is the Poisson's ratio of the heterogeneous reservoir;

[0159] (3) Generate an initial population P0 containing N individuals, where the fracturing process parameters for each individual are expressed as follows: Where, μ i Let X represent the fracturing fluid viscosity, sand ratio, pumping rate, and pumping pressure of the i-th individual, respectively. i The objective function values ​​f1 for the crack propagation range and f2 for the crack conductivity are calculated as follows:

[0160] (4) Perform non-dominated ranking on each individual in the population. For individuals A and B with fracturing process parameters, if A is superior to or equal to B on all objective functions, and is strictly superior to B on at least one objective function, then A is said to dominate B. Otherwise, individuals A and B are considered to have a non-dominated relationship; based on the non-dominated relationship, individuals are divided into different levels.

[0161] Calculate the crowding distance d for each non-dominated individual i. i :

[0162] Where M is the number of objective functions;

[0163] For individuals within each non-dominated layer, sort them from largest to smallest by crowding distance;

[0164] (5) Select suitable individuals based on non-dominated ordination and crowding distance, and perform crossover and mutation operations;

[0165] The crossover operation is represented by the following formula:

[0166] Where λ is a uniformly distributed random factor, typically ranging from [-1, 1], used to control the weight balance between parent solutions and increase diversity; r is a random number in the range [0, 1]; p c This represents the crossover probability, with a value range of [0, 1], typically within the range of [0.6, 0.9].

[0167] The mutation operation is represented by the formula: X j =X j +N(0,σ)

[0168] Where N(0, σ) represents a normal distribution with a mean of 0 and a standard deviation of σ;

[0169] The current population P t and the newly generated offspring population Q t Merge to form a new population R t The merged population is sorted non-dominated and the top N individuals are selected to form a new population P. t+1 Repeat the above steps until the predetermined number of iterations or convergence conditions are reached, and finally obtain the fracturing process parameters adapted to the characteristics of heterogeneous reservoirs.

[0170] In one embodiment of the present invention, the method for collecting characteristic parameters related to fracturing effect includes:

[0171] (1) The fracturing process parameters adapted to the characteristics of heterogeneous reservoirs are used for fracturing construction. Real-time sensor arrays at the fracturing construction site are used to collect fracturing pressure, pumping rate, sand ratio and fracturing monitoring data.

[0172] (2) The collected fracturing monitoring data is denoised and normalized to obtain pre-processed fracturing monitoring data, ensuring the spatiotemporal consistency and accuracy of the data;

[0173] Since noise signals are introduced during data acquisition, this embodiment of the invention first uses a mean filter to remove noise:

[0174] Where Y(t) is the denoised data, X(t) is the original data, N is the window size of the filter, and k is half the size of the window;

[0175] For the two collected time series A(t1) and A(t2), time series alignment and completion are performed using interpolation methods, with the following formula:

[0176] Where A(t) is the interpolated data, and t1 and t2 are the timestamps of the known data points;

[0177] To eliminate the influence of different units of measurement, Z-score normalization is used to normalize the data. The formula is as follows:

[0178] in, X represents the mean of the data; σ represents the standard deviation. z (t) is the Z-score normalized value.

[0179] (3) A joint feature extraction method based on PCA principal component analysis and SVM support vector machine is adopted to extract features related to fracturing effect from the pre-processed fracturing monitoring data, including fracture complexity, fracture propagation range and fracture conductivity, so as to accurately evaluate the influence of different fracturing process parameters on the formation of fractures in heterogeneous reservoirs.

[0180] First, feature extraction is performed using principal component analysis (PCA):

[0181] For the preprocessed fracturing monitoring data in the third step, calculate the covariance matrix using the following formula:

[0182] in, Given the covariance matrix, we obtain its eigenvalues ​​and eigenvectors through eigenvalue decomposition: Cν′=λ′ν′

[0183] Where λ′ is the eigenvalue and ν′ is the eigenvector. The eigenvalues ​​are sorted from largest to smallest, and the eigenvectors corresponding to the first k eigenvalues ​​are selected to form the eigenmatrix W. k W k =[ν′1,ν′2,…,ν′ k ]

[0184] The original data is transformed into a new feature space using the feature matrix: Y = ZW k

[0185] in, It is the data after dimensionality reduction.

[0186] In one embodiment of the invention, a correlation heatmap is used to quantify the degree of information redundancy among independent variables to determine whether the data needs dimensionality reduction. Based on this, principal component analysis (PCA) is introduced to linearly combine the original variables, constructing a set of uncorrelated composite variables. Several representative composite variables are then selected from these to retain the main information in the original data. Using this method, 18 parameters, including geological and engineering parameters, are reduced to 7 principal component characteristic parameters, covering approximately 93% of the original information, thus effectively representing the original variable system with fewer features.

[0187] After feature extraction, SVM is used to further evaluate the extracted features:

[0188] Construct the kernel function:

[0189] Where ω is the bandwidth parameter of the Gaussian kernel, x i x j Sample points, where e is the natural constant;

[0190] Construct the objective function:

[0191] Where w is the weight vector, D is the penalty parameter, and ξ is the weight vector. i These are slack variables; the objective function is transformed into a dual problem using the Lagrange multiplier method.

[0192] Where, α i It is a Lagrange multiplier, y i These are the category labels of the samples;

[0193] Construct a decision function to classify new samples:

[0194] Where b is the bias term, and f(x) is used to determine the category of the new sample x.

[0195] The quality of the extracted features was evaluated by classifying performance indicators and the correlation between features and fracturing effect parameters. Finally, the principal components with the best classification performance and the most relevant correlation with fracturing effect were selected as the key parameters for optimizing the fracturing process and as the key feature data for final extraction.

[0196] In one embodiment of the present invention, in the second step, a model for predicting the fracturing effect of heterogeneous reservoirs is constructed using a dataset, specifically including:

[0197] The first sub-step involves constructing a prediction model for the fracturing effect of heterogeneous reservoirs. Specific operations include:

[0198] A multi-level prediction model combining geological parameters of heterogeneous reservoirs and fracturing process parameters is constructed as a prediction model for the fracturing effect of heterogeneous reservoirs.

[0199] The model employs a multiple-input multiple-output (MIMO) structure. The input variables of the model include: interlayer thickness of heterogeneous reservoir, permeability, porosity, fracturing fluid viscosity, pumping rate, and sand ratio. The output variables of the model include: fracture propagation range, fracture complexity, and fracture conductivity.

[0200] The Multiple Input Multiple Output (MIMO) model is constructed as follows: Y = f(X) + ∈

[0201] Where Y is the output variable matrix and X is the input variable matrix, including the interlayer thickness h, permeability k, porosity φ of the heterogeneous reservoir, and fracturing process parameters such as fracturing fluid viscosity μ and pumping rate Q. p Sand ratio C p Let ∈ be the random error term, and f(X) be the mapping function, represented by multinomial regression:

[0202] Where β0 is the intercept term, X j β is the input feature, p is the total number of input features, and β is the input feature. j and β jk These are the parameters to be learned;

[0203] The input variable matrix X is:

[0204] The second sub-step involves training a heterogeneous reservoir fracturing performance prediction model using the dataset, resulting in the trained model. Specific operations include:

[0205] The model training employs the Gradient Boosting Decision Tree (GBDT) algorithm to obtain a trained prediction model, which captures the nonlinear characteristics of crack network formation and enables accurate prediction of crack propagation range, crack complexity, and crack flow conduction capacity.

[0206] The specific process of the gradient boosting decision tree algorithm includes:

[0207] Establish the initial model:

[0208] Where L is the loss function, y i The actual value;

[0209] In each iteration m, a new tree h is constructed. m (x) is used to fit the residual from the previous round: F m (x)=F m-1 (x)+δh m (x)

[0210] Where δ is the learning rate, typically taking a value between (0,1]; the residual is: r i (m) =y i -F m-1 (x i )

[0211] The mean squared error is calculated as the loss function to update the model parameters using the trained model. The loss function is:

[0212] The third sub-step involves validating and optimizing the trained heterogeneous reservoir fracturing performance prediction model to obtain the optimized model. Specific operations include:

[0213] The prediction results of the trained prediction model were compared with the actual fracturing effect data, and the training prediction model was verified by error analysis.

[0214] The Bayesian optimization algorithm is used to optimize the trained prediction model in order to minimize the prediction error and improve the prediction accuracy and adaptability of the model, resulting in an optimized prediction model.

[0215] Model prediction value The actual observed value is Y. Calculate the following error metrics to verify the model performance:

[0216] Mean square error:

[0217] Mean absolute error:

[0218] Coefficient of determination R 2 :

[0219] in, Let N be the mean of the observed values, and N be the sample size.

[0220] Residuals: By analyzing the residuals r i Identify systematic errors:

[0221] Correlation between calculation parameters and residuals:

[0222] Among them, X i It is a specific input parameter, r i That is the corresponding residual.

[0223] Based on the prediction error, a Gaussian algorithm is used to construct the objective function f(x), and the covariance function is used as the kernel function. The formula is as follows:

[0224] Where, σ 2 It is the signal variance. It is a length scale.

[0225] Desired improvements:

[0226] Among them, f * Given the known optimal target value, the solution is obtained using the predicted mean and variance of the current Gaussian process: EI(x) = (f * -μ(x))Φ(Z)+σ(x)φ(Z)

[0227] Where μ(x) is the predicted mean and σ(x) is the predicted standard deviation. Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0228] Improve the probability:

[0229] The improved model is tested for stability on multiple datasets to verify its adaptability under different conditions. The importance of each input feature is calculated using the model output to allow for further tuning and optimization of the model in the future. Feature importance can be achieved by calculating a measure of the impact of each feature on the model's predictions.

[0230] in, In feature X j The prediction results of the model after the permutation.

[0231] In another embodiment of the present invention, a CNN model is used to extract geological parameters and pumping process parameters, and the geological parameters, fracturing pumping parameters and time series characteristics are integrated to form an initial prediction model, thereby achieving a deep integration of geological conditions and fracturing process.

[0232] In one embodiment of the invention, a correlation heatmap is used to quantify the degree of information redundancy among independent variables to determine whether the data needs dimensionality reduction. Based on this, principal component analysis (PCA) is introduced to linearly combine the original variables, constructing a set of uncorrelated composite variables. Several representative composite variables are then selected from these to retain the main information in the original data. Using this method, 18 parameters, including geological and engineering parameters, are reduced to 7 principal component characteristic parameters, covering approximately 93% of the original information, thus effectively representing the original variable system with fewer features.

[0233] This invention can collect large amounts of data and process sparse data. Large amounts of data can be transmitted using optical fibers or other data interfaces. Furthermore, this invention improves computing power by performing dimensionality reduction or visualization of the data. In addition, this invention exhibits outstanding adaptability and efficiency in data processing and transmission. On one hand, the prediction method and device of this invention can be compatible with and collect large-scale, high-frequency real-time monitoring data. For example, real-time transmission and access of large-scale data can be achieved through high-speed data interfaces such as optical fiber sensing, ensuring the richness and completeness of the data source. On the other hand, addressing the common situation of sparse or incomplete data in practical operations, this invention effectively extracts key features and reduces redundant calculations by performing structured dimensionality reduction or converting the original data into image representations, thereby significantly reducing the demand for computing power while improving data utilization. This design enables this invention to fully utilize massive amounts of data for high-precision modeling and robustly handle sparse datasets, achieving overall optimization of computational efficiency while ensuring prediction accuracy.

[0234] Current commercial fracturing simulation software or fracture propagation simulation software typically relies on extensive numerical simulations and iterative optimizations, with single model training sessions taking hours or even days, making it difficult to meet the timeliness requirements of real-time fracturing decision-making. In contrast, the fracture propagation prediction model constructed in this invention, by fusing geological parameters and pumping image parameters as input, can complete model training in only about 10 minutes under ordinary computing resources, obtaining the calculated fracture geometry results. This significantly improves the efficiency of model updates and deployment, thereby effectively supporting real-time monitoring and dynamic prediction of fracture propagation during fracturing.

[0235] Furthermore, this invention offers significant advantages in model migration and field deployment. When applied to new oil and gas reservoir exploration, existing methods typically require complex remodeling and lengthy training for different geological and engineering conditions, resulting in a prolonged field adaptation period. In contrast, the fracture propagation prediction model constructed in this invention possesses excellent generalization ability and rapid migration characteristics. It only requires input of measured reservoir data (e.g., the heterogeneity of heterogeneous reservoirs and permeability differences between layers), and can complete model adaptation training and updates within 5 to 10 minutes, significantly shortening the cycle from deployment to practical use. This feature significantly improves the model's response speed and engineering efficiency in actual field operations, supporting rapid and dynamic fracturing decisions during exploration and development.

[0236] In one embodiment of the present invention, Figure 4 is a graph showing the simulation results of a prediction model according to an embodiment of the present invention, illustrating the optimal combination of process parameters. As shown in Figure 4, prediction models were established for geological parameters, fracturing parameters, and mechanical parameters, respectively. Points of different colors or gray levels represent different parameter combinations. As can be seen from Figure 4, the circled Y1 represents the optimal combination of process parameters.

[0237] In one embodiment of the present invention, in the third step, the fracturing process parameters and the geological parameters of the heterogeneous reservoir are input into the heterogeneous reservoir fracturing effect prediction model, and the prediction model outputs the heterogeneous reservoir fracturing effect parameters.

[0238] This invention can accurately predict the fracturing effect, especially in heterogeneous reservoirs, and effectively reflect the influence of different fracturing process parameters on fracture formation and propagation.

[0239] In one embodiment of the present invention, the pumping / monitoring time series and geological and technological data are denoised, aligned, and normalized. For example, a joint PCA and SVM method is used to extract key features related to fracture characterization, and the K-Means clustering algorithm is used to cluster the fracturing process and pumping data images to form a feature matrix / encoding and typical pumping templates that can be used for modeling. Then, through image processing, feature extraction, and clustering, data of different types and dimensions are unified into a feature matrix, constructing a unified "template label," thereby improving the stability of training. The following describes the experimental process of a real heterogeneous reservoir.

[0240] Figure 5 shows a composite columnar section of a heterogeneous reservoir according to an embodiment of the present invention. This heterogeneous reservoir exhibits a multi-lithological combination of shale, mudstone, and sandstone, with significant and rapidly changing stress differences between sub-layers. The heterogeneous reservoir comprises three interlayers: interlayer 1, interlayer 2, and interlayer 3. Interlayer 1: A high-stress interlayer approximately 2-4 m thick exists at the top of 1-1, with a stress difference of 3-7 MPa. Interlayer 2: A high-stress interlayer 1-2 m thick exists at the bottom of 2-5, with a stress difference of 5-8 MPa. Interlayer 3: A 3 m thick interlayer develops in the 3-3 sub-layer, with a stress difference of 3-8 MPa. Inter-layer 2: The stress difference between sub-layers 2-3 and 2-4 is 4-5 MPa.

[0241] For the geological model of this heterogeneous reservoir, 120 sets of fracture propagation simulations were conducted under different layered stresses, different layered rock mechanics, and different fracturing scales. A data-enhanced adversarial neural network was built, augmented with 500 sets of data, and a fracture propagation database under different geological-engineering parameters was constructed. Then, the generated fracturing parameters and fracturing effect data were used together with the original data to train the prediction model. The trained prediction model can quickly obtain the fracturing effect in the prediction stage, and the data input process is greatly simplified, significantly reducing training time and improving training efficiency.

[0242] In one embodiment of the present invention, based on the reservoir information of the aforementioned heterogeneous reservoir, the initiation and propagation states of fractures in different sub-layers within the three interlayers are simulated, and the extension length of the fractures in each sub-layer is simulated. For example, for interlayer 2-1, the simulation results are shown in Figure 6. It can be seen that interlayer 2-1 is mainly composed of mudstone and is located in the orogenic section, with fractures extending along the main direction. By using the prediction method according to an embodiment of the present invention, the fracture parameters are obtained as follows: the total length of the fracture network is 305m, the fracture height on the upper fracture surface is 12.4m, and the fracture height on the lower fracture surface is 15.6m, therefore the total height of the fracture network is 28m.

[0243] The actual fracture data of this heterogeneous reservoir are shown in the table below.

[0244] Therefore, the above experiments show that, according to the prediction method of the embodiment of the present invention, the error in the total length of the fracture network is (305-282) / 282 = 8.1%, and the error in the total height of the fracture network is (30-28) / 30 = 6.7%. In other words, the difference between the model prediction and the actual result is less than 10%. That is, in this invention, by inputting the actually collected fracturing effect data (e.g., the heterogeneity of heterogeneous reservoirs and the permeability differences between layers) into the model to update the training model, the reliability of the prediction results is significantly improved. This result shows that the prediction method of the present invention can accurately capture the morphological characteristics of the fracture network, and the prediction results are highly consistent with the actual engineering situation. It can provide reliable data support for optimizing fracturing schemes (e.g., adjusting pumping parameters, adjusting proppant dosage, controlling fracture propagation direction) and evaluating reservoir stimulation effects (e.g., calculating the drainage area), significantly reducing the construction risks and cost waste caused by prediction deviations.

[0245] Embodiments of the present invention also provide a fracturing effect prediction system for heterogeneous reservoirs considering fracturing processes, used to execute the steps in the prediction method of the embodiments of the present invention, the system may include one or more of the following:

[0246] The parameter acquisition unit is used to collect geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs as a dataset.

[0247] The parameter acquisition unit includes:

[0248] The first parameter acquisition subunit is used to acquire geological parameters of heterogeneous reservoirs, including interlayer thickness, porosity, and permeability of heterogeneous reservoirs.

[0249] The second parameter acquisition subunit is used to acquire fracturing process parameters, including fracturing fluid viscosity, sand ratio and pumping rate.

[0250] The third parameter acquisition subunit is used to acquire characteristic parameters related to fracturing effect, including fracture propagation range, fracture complexity, and fracture conductivity.

[0251] The prediction model acquisition unit is used to construct a prediction model for the fracturing effect of heterogeneous reservoirs using a dataset. Specifically, a three-dimensional geological model of the heterogeneous reservoir is constructed based on the geological parameters and fracturing process parameters of the heterogeneous reservoir.

[0252] The prediction model acquisition unit includes:

[0253] The model building sub-unit is used to construct a fracturing effect prediction model for heterogeneous reservoirs. Specific operations include:

[0254] A multi-level prediction model combining geological parameters and fracturing process parameters of heterogeneous reservoirs is constructed as a prediction model for the fracturing effect of heterogeneous reservoirs.

[0255] The model employs a multiple-input multiple-output (MIMO) structure. The input variables of the model include: interlayer thickness of heterogeneous reservoir, permeability, porosity, fracturing fluid viscosity, pumping rate, and sand ratio. The output variables of the model include: fracture propagation range, fracture complexity, and fracture conductivity.

[0256] The model training subunit is used to train a heterogeneous reservoir fracturing performance prediction model using a dataset, resulting in the trained heterogeneous reservoir fracturing performance prediction model. Specific operations include:

[0257] The model training uses the gradient boosting decision tree algorithm (GBDT) to obtain the trained prediction model, which captures the nonlinear characteristics of crack network formation and achieves accurate prediction of crack propagation range, crack complexity and crack conduction capacity.

[0258] The model validation and optimization unit is used to validate and optimize the trained heterogeneous reservoir fracturing effect prediction model, resulting in an optimized heterogeneous reservoir fracturing effect prediction model. Specific operations include:

[0259] The prediction results of the trained prediction model were compared with the actual fracturing effect data, and the training prediction model was verified by error analysis.

[0260] The Bayesian optimization algorithm is used to optimize the trained prediction model in order to minimize the prediction error and improve the prediction accuracy and adaptability of the model, resulting in an optimized prediction model.

[0261] The prediction unit is used to input fracturing process parameters and heterogeneous reservoir geological parameters into the heterogeneous reservoir fracturing effect prediction model. The prediction model outputs heterogeneous reservoir fracturing effect parameters. Based on the constructed three-dimensional geological model of the heterogeneous reservoir, and according to the heterogeneity of the heterogeneous reservoir and the permeability difference of each layer, the fracturing process parameters adapted to the characteristics of the heterogeneous reservoir are optimized and selected through a multi-objective optimization algorithm.

[0262] The following describes a real-time prediction method for crack propagation according to an embodiment of the present invention. As shown in FIG7, the real-time prediction method for crack propagation includes:

[0263] Step 110: Obtain geological parameters, pumping image parameters, and sample data of fracture propagation corresponding to the pumping image parameters. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping image parameters are pumping parameters converted into image form. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure.

[0264] Specifically, geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical properties. These parameters can be acquired, transmitted, and stored in real time using sensors, measuring equipment, and core experiments to ensure accuracy and timeliness. Pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure. Data on fracturing fluid type, discharge rate, stage fluid volume, and sand ratio during the pumping operation are acquired, transmitted, and stored in real time using high-precision flow meters, pressure sensors, and sand mixing detection equipment.

[0265] In one example, physical properties include at least one of permeability and rock density, which can be obtained through a combination of laboratory and field tests.

[0266] Step 120: Perform data fusion on the geological parameters and pumping image parameters to obtain fused feature data.

[0267] Specifically, since geological parameters and pumping image parameters come from different time or spatial scales, temporal or spatial matching and alignment are required. By using a suitable data fusion algorithm, the features of geological parameters and pumping image parameters can be fused to generate fused feature data. Based on the fused feature data, the prediction accuracy of fracture propagation can be improved.

[0268] Step 130: Construct an initial prediction model based on a machine learning algorithm, using the fused feature data as the input of the initial prediction model, the crack propagation amount as the output of the initial prediction model, and training the initial prediction model using the corresponding sample data to obtain a crack propagation prediction model.

[0269] Specifically, the collected sample data is divided into a training set, a validation set, and a test set. A suitable machine learning algorithm is selected and trained using the training set data to obtain an initial prediction model. During training, the algorithm attempts to learn the mapping relationship between input features (fused feature data) and output target (crack propagation). The hyperparameters of the model are adjusted using the validation set data to optimize the model's performance. That is, the fused feature data is used as the input to the initial prediction model, the crack propagation is used as the output of the initial prediction model, and the initial prediction model is trained using the corresponding sample data to obtain a crack propagation prediction model. Furthermore, the performance of the trained crack propagation prediction model is evaluated using the test set data.

[0270] Step 140: Real-time prediction of crack propagation is performed using the crack propagation prediction model.

[0271] Specifically, when pumping parameters or geological parameters change during pumping operations, the changed pumping parameters or geological parameters are acquired in real time, and the changed pumping image parameters or geological parameters are fused to obtain adjusted fused feature data. Based on the adjusted fused feature data, a fracture propagation prediction model is used to predict fracture propagation in real time.

[0272] In this embodiment, by combining pumping procedure data with geological parameters and using a deep learning model for multimodal data fusion, the prediction accuracy of fracture propagation is significantly improved. This allows for a more accurate reflection of fracture behavior under complex geological conditions. Furthermore, the use of on-site real-time monitoring equipment to collect pumping data and geological parameters enables real-time tracking of the fracture propagation process, allowing for timely responses to geological changes and providing timely decision support for operators. During pumping operations, changes in pumping parameters can be monitored in real time, and the fracture propagation prediction results can be readjusted based on new data, demonstrating good adaptability and flexibility and optimizing the operation process. By converting pumping procedure data into image data and fusing it with geological parameters, the limitations of traditional methods in data processing and analysis are overcome, enabling a more comprehensive understanding of the impact of the pumping process on fracture propagation.

[0273] In some embodiments, the step of acquiring geological parameters, pumping image parameters, and sample data of fracture propagation corresponding to the pumping image parameters, wherein the geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters, and the pumping image parameters are pumping parameters converted into image form, and the pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure, includes:

[0274] Obtain geological parameters, pumping parameters, and the fracture propagation amount corresponding to the pumping parameters;

[0275] The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

[0276] Specifically, geological parameters are collected and stored through sensors, measuring equipment, and core experiments; and data during the pumping operation are acquired using on-site real-time monitoring equipment, and the pumping parameters are converted into image data to ensure that the input pumping data can be analyzed in image form in subsequent deep learning models (i.e., fracture propagation prediction models), while retaining its original data information for multimodal analysis.

[0277] In one example, strain gauges and fiber optic sensors are used to measure formation stress. Strain gauges directly measure strain changes in rock under external forces, while fiber optic sensors monitor stress changes using fluctuations in optical signals. The strain measurement is described by the following equation:

[0278] Where ∈ represents strain, ΔL represents length change, and L0 represents the original length. By measuring the strain at different locations, the principal stresses are derived using the equilibrium equation: σ1=σ0+E·∈

[0279] Where σ0 is the original stress and E is Young's modulus.

[0280] Using the calculated principal stresses and the Mohr-Coulomb failure criterion, the potential crack propagation direction is analyzed. The formula is: τ=c+σ n tan(φ)

[0281] Where τ is the shear stress, c is the cohesion, and σ n φ is the normal stress, and φ is the internal friction angle.

[0282] In one example, Young's modulus (E) and Poisson's ratio (ν) are measured via core experiments. Young's modulus is calculated using the following formula:

[0283] Where F is the applied force and A is the cross-sectional area. Poisson's ratio is calculated by the following formula:

[0284] Where, ∈ lat For transverse strain, ∈ long For longitudinal strain, the typical range of Poisson's ratio is 0 to 0.5.

[0285] In one example, physical properties include permeability and density. Furthermore, permeability (k) and density (ρ) are measured through a combination of laboratory and field testing. Permeability is calculated based on Darcy's law, with the following formula:

[0286] Where Q is the flow rate, A is the flow cross-sectional area, ΔP is the pressure difference, μ is the fluid viscosity, and L is the flow path length.

[0287] The density of rock is calculated based on its mass and volume:

[0288] Where m is the mass of the rock sample and V is its volume.

[0289] In one example, high-precision flow meters, pressure sensors, and sand mixing monitoring equipment are used to collect data such as fracturing fluid type, discharge rate, stage fluid volume, and sand ratio during the pumping operation in real time, and transmit and store the data in real time.

[0290] Specifically, during the pumping process, a turbine flow meter is used to measure the fracturing fluid discharge rate (Q) in real time. The flow rate formula is Q = A·v.

[0291] Where A is the cross-sectional area of ​​the pipe, and v is the flow velocity, calculated by the relationship between the frequency generated by the turbine rotation and the flow velocity, using the formula: v = k·f

[0292] Where k is the calibration constant of the flow meter, and f is the turbine speed.

[0293] Introducing a correction factor (K) to improve measurement accuracy, the formula is modified to: Q corr =K·A·v

[0294] During pumping operations, pressure sensors are used to monitor the pressure (P) of the fracturing fluid in real time. The relationship between pressure and strain is: P = k·ε

[0295] Where k is the stiffness constant of the material, and ∈ is the corresponding strain value.

[0296] By acquiring pressure data in real time, the volume change of fracturing fluid and the effective injection depth are calculated based on Bernoulli's equation. The formula is:

[0297] Where ρ is the liquid density, g is the gravitational acceleration, h is the liquid height, and C is a constant. The sand ratio (S) in the fracturing fluid is monitored in real time using a sand mixing monitoring device, and the calculation formula is:

[0298] Among them, M s V represents the mass of sand. l This refers to the volume of the liquid. Monitoring the sand ratio requires a high-frequency sensor to provide real-time feedback on the physical properties of the mixture. A closed-loop control system is implemented using a mass flow meter and flow control valve to ensure the stability of the sand ratio under different operating conditions. Through real-time data acquisition, the system can dynamically adjust the sand ratio to maintain it within the optimal range.

[0299] In some embodiments, the step of converting the pumping parameters into an image matrix capable of characterizing temporal changes during the pumping process to obtain pumping image parameters includes:

[0300] The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

[0301] Specifically, the Gram angle field algorithm is used to convert numerical information such as fracturing fluid type, displacement, stage fluid volume and sand ratio in the pumping procedure data into an image matrix. The image matrix represents the temporal changes during the pumping process, while retaining the original numerical data for multimodal fusion input of the model.

[0302] In pumping data, time-series data such as displacement, sand ratio, and pressure typically have different dimensions and magnitudes. Before the data is entered into the image matrix, it needs to be normalized to restrict all data to the [0,1] interval, ensuring consistency between different types of data. For the original data X={x1,x2,…,x…} n The normalization formula is:

[0303] in, It is the normalized data, x i Here are the original data, and min(X) and max(X) are the minimum and maximum values ​​of the original data, respectively. If the data needs to be transformed to the range [-1, 1], the following formula can be used further:

[0304] This formula ensures that the data is between [-1, 1], thus accommodating the needs of subsequent polar coordinate and angle transformations.

[0305] The normalized time series data is mapped to the angle space of the polar coordinate system. The angle calculation formula is as follows:

[0306] Where, θ i It is the angle corresponding to the i-th data point. These are the normalized data values. The arccos function maps the normalized data to angle space, obtaining values ​​between [0, π].

[0307] Based on the angle values ​​at two time points, the relationship between the data is calculated, and the Gram angle field matrix is ​​constructed: GAF(i,j)=cos(θ) i +θ j )

[0308] Where, for each element of the GAF(i,j) Gram angular field matrix, θ i and θ jLet θ be the angles of the i-th and j-th data points in the data sequence, respectively. Using the cosine addition formula: cos(θ) i +θ j )=cos(θ i cos(θ) j )-sin(θ i sin(θ) j )

[0309] At the same time, based on the principle of symmetry, we have: GAF(i,j)=GAF(j,i)

[0310] Difference operations are used to enhance the temporal characteristics of the data, highlight the rate of change in the data, and improve the model's sensitivity to data changes. The formula is: Δx i =x i -x i-1

[0311] Where, Δx i x is the change in data at time i. i and x i-1 These are the original data values ​​at two adjacent time points.

[0312] In some embodiments, the step of fusing the geological parameters and pumping image parameters to obtain fused feature data includes:

[0313] The geological parameters and pumping image parameters are input into a neural network model, and the neural network model is used to perform multimodal data fusion to obtain fused feature data.

[0314] Specifically, appropriate data fusion algorithms, such as weighted average, principal component analysis, and neural networks, should be selected to fuse the features of geological parameters and pumping image parameters. The choice of algorithm should be based on the characteristics of the data and the analysis objectives to maximize the value of the fused data.

[0315] In one example, a neural network model was used to achieve data fusion of geological parameters and pumping image parameters.

[0316] In some embodiments, the step of inputting the geological parameters and pumping image parameters into a neural network model, and using the neural network model to perform multimodal data fusion to obtain fused feature data includes:

[0317] The geological parameters and pumping image parameters are simultaneously input into a long short-term memory neural network model with a coupled convolutional structure. The convolutional neural network extracts features from the pumping image parameters to obtain image features of the pumping image parameters. The long short-term memory layer processes the time-series data of the geological parameters to obtain time-series features of the geological parameters.

[0318] The image features of the pumping image parameters and the temporal features of the geological parameters are spliced ​​together to obtain fused feature data.

[0319] Specifically, the numerical data of pumping image parameters and geological parameters are simultaneously input into a long short-term memory (LSTM) neural network model with a coupled convolutional structure. The LSTM layer is then used to process the time-series data of the pumping procedure image data to achieve multimodal fusion of pumping data and geological parameters.

[0320] Among them, the spatiotemporal features of the image (i.e., pumping image parameters) are extracted from the Gram angle field image matrix based on a convolutional neural network (CNN), and the formula is as follows:

[0321] Among them, G i+p,j+q These are local pixel values ​​in the image matrix; W p,q Y is the weight matrix of the convolution kernel; b is the bias term; i,j It is the output feature after the convolution operation.

[0322] Long Short-Term Memory (LSTM) networks are used to extract time-series data such as formation stress, Young's modulus, and Poisson's ratio. The input gate, forget gate, and output gate of the LSTM are then constructed.

[0323] Forgotten Gate: f t =σ(W f ·[h t-1 ,x t ]+b f )

[0324] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i )

[0325] Memory update formula:

[0326] Output gate: o t =σ(W o ·[h t-1 ,x t ]+b o h t =o t ·tanh(C t )

[0327] A series of temporal features are obtained through multi-layer recursive operations of LSTM. Where d represents the dimension of the temporal feature vector. The image features of the pumping procedure... Temporal characteristics of geological parameters Multimodal fusion is performed to ensure that data from these different sources can be uniformly analyzed in the deep learning model. Feature concatenation is used to combine the image features F output by the convolutional neural network. CNN Geological parameter characteristics F output by LSTM LSTM Scattered together: F fusion =[F CNN ;F LSTM ]

[0328] in This fused feature vector combines the temporal characteristics of pumping procedure data and geological parameters.

[0329] In some embodiments, the crack propagation amount includes the crack length, width, and height, and the step of real-time prediction of crack propagation using the crack propagation prediction model further includes:

[0330] A crack propagation image is generated based on the predicted crack propagation amount output by the crack propagation prediction model.

[0331] Specifically, based on the crack propagation amount, i.e. the length, width and height of the crack, crack propagation images are generated and displayed in real time through the user interface at different time periods, i.e. different geological parameters or different pumping image parameters. The dynamic process of crack propagation is recorded in its entirety, and reports are generated for subsequent analysis and decision support.

[0332] In this embodiment of the invention, the crack propagation data generated by the model is converted into a visual image, and the three-dimensional propagation of the cracks is updated and displayed in real time. The crack length, width, and height for each time period are displayed on the user interface in real time, helping operators monitor crack changes. This function ensures that on-site personnel can intuitively see the real-time morphology of the cracks, avoid operational errors, and provide decision-making basis at critical moments. Furthermore, the entire crack propagation process is dynamically recorded, including not only real-time crack propagation data but also background data such as pumping parameters and geological conditions for each time period, for subsequent analysis and optimization decision-making.

[0333] Another embodiment of the present invention relates to a real-time dynamic analysis method for fracture propagation based on pumping program data and geological parameters. The specific process is shown in Figure 8, and includes the following steps:

[0334] Step S1: Geological parameter acquisition. Geological parameters are acquired and stored through sensors, measuring equipment, and core experiments.

[0335] Step S2, Pumping data acquisition: Data during the pumping operation is acquired using on-site real-time monitoring equipment.

[0336] Step S3, data conversion, converts the pumping procedure data into image data, ensuring that the input pumping data can be analyzed in the form of images in the deep learning model, while retaining its original numerical information for multimodal analysis.

[0337] Step S4, multimodal data fusion: input geological parameters and pumping procedure image data into the neural network model, and use the model to perform multimodal data fusion.

[0338] Step S5, crack propagation prediction: The crack propagation process is predicted using a deep learning model, generating crack propagation data for each time period or stage, and outputting it as a visual image or numerical result.

[0339] Step S6, Dynamic Adjustment and Re-prediction: During the pumping operation, the changes in pumping parameters are monitored and recorded in real time. If the pumping parameters are adjusted, the adjusted data is re-inputted into the model to update the crack propagation prediction results in real time.

[0340] Step S7: Real-time display and recording to generate an adjusted crack propagation image, while recording the dynamic changes of the entire crack propagation process to provide real-time monitoring support for on-site operations.

[0341] In one embodiment, step S1, geological parameter acquisition, involves real-time acquisition, transmission, and storage of formation stress, Young's modulus, Poisson's ratio, and physical property parameters through sensors, measuring equipment, and core experiments to ensure data accuracy and timeliness. Strain gauges and fiber optic sensors are used to measure formation stress. Strain gauges can directly measure the strain changes of rocks under external forces, while fiber optic sensors utilize fluctuations in optical signals to monitor stress changes.

[0342] The stress measurement is described by the following formula:

[0343] Where ∈ represents strain, ΔL represents length change, and L0 represents the original length.

[0344] By measuring the strain at different locations, the principal stresses are derived using the equilibrium equations. The formula is: σ1=σ0+E·∈

[0345] Where σ0 is the original stress and E is Young's modulus.

[0346] Using the calculated principal stresses and the Mohr-Coulomb failure criterion, the potential crack propagation direction is analyzed. The formula is: τ=c+σ n tan(φ)

[0347] Where τ is the shear stress, c is the cohesion, and σ n φ is the normal stress, and φ is the internal friction angle.

[0348] Young's modulus (E) and Poisson's ratio (ν) were measured using core experiments. Young's modulus was calculated using the following formula:

[0349] Where F is the applied force and A is the cross-sectional area. Poisson's ratio is calculated by the following formula:

[0350] Where, ∈ lat For transverse strain, ∈ long For longitudinal strain, the typical range of Poisson's ratio is 0 to 0.5.

[0351] Permeability (k) and density (ρ) are measured through a combination of laboratory and field tests. Permeability is calculated based on Darcy's law, using the following formula:

[0352] Where Q is the flow rate, A is the flow cross-sectional area, ΔP is the pressure difference, μ is the fluid viscosity, and L is the flow path length.

[0353] The density of rock is calculated based on its mass and volume:

[0354] Where m is the mass of the rock sample and V is its volume.

[0355] In some examples, step S2, pumping data acquisition, involves real-time collection of data such as fracturing fluid type, flow rate, stage fluid volume, and sand ratio during the pumping operation using high-precision flow meters, pressure sensors, and sand mixing monitoring equipment, and then transmitting and storing this data in real time. During the pumping process, a turbine flow meter is used to measure the fracturing fluid flow rate (Q) in real time. The flow rate formula is Q = A·v.

[0356] Where A is the cross-sectional area of ​​the pipe, and v is the flow velocity, calculated by the relationship between the frequency generated by the turbine rotation and the flow velocity, using the formula: v = k·f

[0357] Where k is the calibration constant of the flow meter, and f is the turbine speed.

[0358] Introducing a correction factor (K) to improve measurement accuracy, the formula is modified to: Q corr =K·A·v

[0359] During pumping operations, pressure sensors are used to monitor the pressure (P) of the fracturing fluid in real time. The relationship between pressure and strain is: P = k·ε

[0360] Where k is the stiffness constant of the material, and ∈ is the corresponding strain value.

[0361] By acquiring pressure data in real time, the volume change of fracturing fluid and the effective injection depth are calculated based on Bernoulli's equation. The formula is:

[0362] Where ρ is the liquid density, g is the gravitational acceleration, h is the liquid height, and C is a constant. The sand ratio (S) in the fracturing fluid is monitored in real time using a sand mixing monitoring device, and the calculation formula is:

[0363] Among them, M s V represents the mass of sand. l This refers to the volume of the liquid. Monitoring the sand ratio requires a high-frequency sensor to provide real-time feedback on the physical properties of the mixture. A closed-loop control system is implemented using a mass flow meter and flow control valve to ensure the stability of the sand ratio under different operating conditions. Through real-time data acquisition, the system can dynamically adjust the sand ratio to maintain it within the optimal range.

[0364] As shown in Figure 9, which is an example of pumping parameters.

[0365] In one example, step S3, data conversion, uses the Gram angle field algorithm to convert numerical information such as fracturing fluid type, displacement, stage fluid volume and sand ratio in the pumping procedure data into an image matrix. The image matrix represents the temporal changes during the pumping process, while retaining the original numerical data for multimodal fusion input of the model.

[0366] In pumping data, time-series data such as displacement, sand ratio, and pressure typically have different dimensions and magnitudes. Before the data is entered into the image matrix, it needs to be normalized to restrict all data to the [0,1] interval, ensuring consistency between different types of data. For the original data X={x1,x2,…,x…} n The normalization formula is:

[0367] in, It is the normalized data, x i Here are the original data, and min(X) and max(X) are the minimum and maximum values ​​of the original data, respectively. If the data needs to be transformed to the range [-1, 1], the following formula can be used further:

[0368] This formula ensures that the data is between [-1, 1], thus accommodating the needs of subsequent polar coordinate and angle transformations.

[0369] The normalized time series data is mapped to the angle space of the polar coordinate system. The angle calculation formula is as follows:

[0370] Where, θ iIt is the angle corresponding to the i-th data point. These are the normalized data values. The arccos function maps the normalized data to angle space, obtaining values ​​between [0, π].

[0371] Based on the angle values ​​at two time points, the relationship between the data is calculated, and the Gram angle field matrix is ​​constructed: GAF(i,j)=cos(θ) i +θ j )

[0372] Where, for each element of the GAF(i,j) Gram angular field matrix, θ i and θ j Let θ be the angles of the i-th and j-th data points in the data sequence, respectively. Using the cosine addition formula: cos(θ) i +θ j )=cos(θ i cos(θ) j )-sin(θ i sin(θ) j )

[0373] At the same time, based on the principle of symmetry, we have: GAF(i,j)=GAF(j,i)

[0374] Difference operations are used to enhance the temporal characteristics of the data, highlight the rate of change in the data, and improve the model's sensitivity to data changes. The formula is: Δx i =x i -x i-1

[0375] Where, Δx i x is the change in data at time i. i and x i-1 These are the original data values ​​at two adjacent time points.

[0376] In one example, step S4, multimodal data fusion, involves simultaneously inputting pumping procedure data converted into image form and numerical data of geological parameters into a long short-term memory (LSTM) neural network model with a coupled convolutional structure. The pumping procedure image data is then processed by a convolutional neural network (CNN) to extract features, and the time-series data of geological parameters is then processed by an LSTM layer to achieve multimodal fusion of pumping data and geological parameters.

[0377] Based on a convolutional neural network (CNN), the spatiotemporal features of the Gram angle field image matrix obtained in step S3 are extracted using the following formula:

[0378] Among them, G i+p,j+q These are local pixel values ​​in the image matrix; W p,qY is the weight matrix of the convolution kernel; b is the bias term; i,j It is the output feature after the convolution operation.

[0379] Long Short-Term Memory (LSTM) networks are used to extract time-series data such as formation stress, Young's modulus, and Poisson's ratio. The input gate, forget gate, and output gate of the LSTM are then constructed.

[0380] Forgotten Gate: f t =σ(W f ·[h t-1 ,x t ]+b f )

[0381] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i )

[0382] Memory update formula:

[0383] Output gate: o t =σ(W o ·[h t-1 ,x t ]+b o h t =o t ·tanh(C t )

[0384] A series of temporal features are obtained through multi-layer recursive operations of LSTM. Where d represents the dimension of the temporal feature vector. The image features of the pumping procedure... Temporal characteristics of geological parameters Multimodal fusion is performed to ensure that data from these different sources can be uniformly analyzed in the deep learning model. Feature concatenation is used to combine the image features F output by the convolutional neural network. CNN Geological parameter characteristics F output by LSTM LSTM Scattered together: F fusion =[F CNN ;F LSTM ]

[0385] in This fused feature vector combines the temporal characteristics of pumping procedure data and geological parameters.

[0386] In one example, the data collected and processed in steps S1 to S3 are divided into training, validation, and test sets using a 70% training, 15% validation, and 15% test ratio. The mean squared error is used as the loss function to calculate the difference between the model's predicted values ​​and the actual values, as shown in the following formula:

[0387] The model parameters are updated using optimization algorithms such as Adam, using the following formula:

[0388] Where, θ t Here, η is the current parameter, and η is the learning rate. It is a moving average of the gradient. It is the moving average of the square of the gradient, and ∈ is a small constant to avoid division by zero.

[0389] In one example, step S5, crack propagation prediction, generates real-time crack propagation data through a long short-term memory (LSTM) neural network model with coupled convolutional structures. The crack propagation data includes the length, width, and height of the crack and is output in image form for visual analysis of crack propagation.

[0390] In one example, using the LSTM network with coupled convolutional structure obtained in step S4, the features are input into the model to predict the length, width, and height of the crack. The output formula is:

[0391] in, f represents the predicted crack length, width, and height, respectively. L f W f H It is a learned function generated from the input feature X.

[0392] Simultaneously, the outputs of the CNN and LSTM in step S4 are made independently, and then the prediction results of the two are combined through weighted fusion: y final =α·y CNN +β·y LSTM

[0393] Where y CNN and y LSTM These are the independent outputs of CNN and LSTM, and α and β are the corresponding fusion weights.

[0394] In this invention, step S6, dynamic adjustment and re-prediction, involves inputting the adjusted pumping data into a deep learning model in real time when parameters such as displacement, sand ratio, or liquid volume are adjusted during the pumping operation. This recalculates the fracture propagation process and outputs new fracture propagation data and images. During the pumping process, if on-site parameters are adjusted, such as displacement Q, sand ratio S, and pumping pressure P, the relevant parameters for fracture propagation need to be recalculated. The initial parameters are Q0, S0, and P0, and the adjusted parameters are Q0, S0, and P0. new S new and P new The updated state of the pumping parameters can be represented as: D new ={Q new ,S new ,P new}

[0395] At this point, the geological parameters remain unchanged or are slightly adjusted based on the time series data, denoted as G={σ,E,ν} (which represent formation stress, Young's modulus, and Poisson's ratio, respectively).

[0396] The feature vector F input to the neural network model new Includes updated pumping parameters D new And geological parameters G: F new ={D new ,G}

[0397] After adjusting the pumping parameters, based on the feature extraction method described in step S5, a convolutional neural network (CNN) is used to extract the spatiotemporal features of the pumping data, and a long short-term memory network (LSTM) is used to process the time series of geological parameters. After real-time input adjustment, the model recalculates the length, width, and height of the cracks.

[0398] Crack length L t The variation over time is influenced by both pumping parameters (pressure, displacement, sand ratio) and geological parameters. Based on convolution and temporal characteristics, the prediction formula for length can be expressed as:

[0399] Among them, F fusion W represents the feature vector obtained after multimodal fusion. L and b L These are the weight matrix and bias term for length prediction, respectively; ReLU is the activation function, which is:

[0400] Crack width W t The prediction method is similar, calculating the fracture opening width by combining pumping data and geological parameters:

[0401] Among them, W W and bW These are the model parameters for width prediction, and ReLU is the activation function.

[0402] Crack height H t The vertical propagation of the crack is mainly affected by the pumping pressure, and the prediction formula is as follows:

[0403] Among them, W h and b h These are the model parameters for vertical prediction, and ReLU is the activation function.

[0404] In one example, step S7 involves real-time display and recording of crack propagation images across different time periods, generating and displaying these images via a user interface. This process records the entire dynamic process of crack propagation and generates reports for subsequent analysis and decision support. By converting the crack propagation data generated by the model into visualized images, the three-dimensional propagation of the cracks is updated and displayed in real time. The crack length, width, and height for each time period are displayed on the user interface in real time, helping operators monitor crack changes. This function ensures that on-site personnel can intuitively see the real-time morphology of the cracks, avoiding operational errors and providing decision-making support at critical moments. The entire crack propagation process is dynamically recorded. The recorded content includes not only real-time crack propagation data but also background data such as pumping parameters and geological conditions for each time period, for subsequent analysis and optimization decision-making.

[0405] In this embodiment, by combining pumping procedure data with geological parameters and using a deep learning model for multimodal data fusion, the prediction accuracy of fracture propagation is significantly improved. This allows for a more accurate reflection of fracture behavior under complex geological conditions. Furthermore, the use of on-site real-time monitoring equipment to collect pumping data and geological parameters enables real-time tracking of the fracture propagation process, allowing for timely responses to geological changes and providing timely decision support for operators. During pumping operations, changes in pumping parameters can be monitored in real time, and the fracture propagation prediction results can be readjusted based on new data, demonstrating good adaptability and flexibility and optimizing the operation process. By converting pumping procedure data into image data and fusing it with geological parameters, the limitations of traditional methods in data processing and analysis are overcome, enabling a more comprehensive understanding of the impact of the pumping process on fracture propagation.

[0406] Therefore, the present invention also relates to a real-time prediction method for crack propagation, the real-time prediction method comprising:

[0407] Sample data of geological parameters, pumping image parameters, and corresponding fracture propagation amounts are obtained, wherein the pumping image parameters are images obtained by transforming the pumping parameters.

[0408] The geological parameters are fused with the pumping image parameters to obtain fused feature data;

[0409] A prediction model is constructed based on machine learning algorithms, with the fused feature data as input and the crack propagation amount as output, and the crack propagation prediction model is obtained by training with sample data.

[0410] The crack propagation prediction model is used to predict crack propagation in real time.

[0411] In the real-time prediction method for fracture propagation according to an embodiment of the present invention, the step of acquiring sample data of geological parameters, pumping image parameters, and corresponding fracture propagation amounts includes:

[0412] Geological parameters, pumping parameters, and fracture propagation corresponding to the pumping parameters are obtained. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure.

[0413] The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

[0414] In the real-time prediction method for crack propagation according to an embodiment of the present invention, the step of converting the pumping parameters into an image matrix capable of characterizing temporal changes during the pumping process includes:

[0415] The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

[0416] In the real-time prediction method for crack propagation according to an embodiment of the present invention, the step of training a crack propagation prediction model using sample data includes:

[0417] The objective function of the initial prediction model is constructed based on the relative error between the predicted crack propagation amount output by the initial prediction model and the sample data of crack propagation amount.

[0418] The initial prediction model is iterated based on the objective function, and the parameters of the initial prediction model are adjusted.

[0419] When the objective function reaches its minimum value, the corresponding parameters are determined as the objective parameters of the initial prediction model.

[0420] The crack propagation prediction model is determined based on the target parameters.

[0421] Another embodiment of this application relates to a real-time prediction device for crack propagation. The implementation details of the real-time prediction device for crack propagation in this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution. The schematic diagram of the real-time prediction of crack propagation in this embodiment can be shown in Figure 10, including an acquisition module 801, a data fusion module 802, a model building module 803, and a prediction module 804.

[0422] The acquisition module 801 is used to acquire geological parameters, pumping image parameters, and sample data of fracture propagation corresponding to the pumping image parameters. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping image parameters are pumping parameters converted into image form. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure.

[0423] The data fusion module 802 is used to fuse the geological parameters and pumping image parameters to obtain fused feature data.

[0424] The model building module 803 is used to build an initial prediction model based on a machine learning algorithm. The fused feature data is used as the input of the initial prediction model, the crack expansion amount is used as the output of the initial prediction model, and the initial prediction model is trained using the corresponding sample data to obtain a crack expansion prediction model.

[0425] The prediction module 804 is used to make real-time predictions of crack propagation using the crack propagation prediction model.

[0426] In some optional embodiments, the data fusion module is further configured to input the geological parameters and pumping image parameters into a neural network model, and use the neural network model to perform multimodal data fusion to obtain fused feature data.

[0427] In some optional embodiments, the data fusion module is further configured to simultaneously input the geological parameters and pumping image parameters into a long short-term memory neural network model with a coupled convolutional structure, extract features from the pumping image parameters through the convolutional neural network to obtain image features of the pumping image parameters, and process the time-series data of the geological parameters through the long short-term memory layer to obtain time-series features of the geological parameters.

[0428] The image features of the pumping image parameters and the temporal features of the geological parameters are spliced ​​together to obtain fused feature data.

[0429] In some optional embodiments, the prediction module is further configured to generate a crack propagation image based on the predicted crack propagation amount output by the crack propagation prediction model.

[0430] In some optional embodiments, the acquisition module is further configured to acquire geological parameters, pumping parameters, and the amount of fracture propagation corresponding to the pumping parameters;

[0431] The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

[0432] In some optional embodiments, the acquisition module is further configured to convert the pumping parameters into an image matrix using the Gram corner field algorithm.

[0433] In the embodiments of this invention, by combining pumping procedure data with geological parameters and using a deep learning model for multimodal data fusion, the prediction accuracy of fracture propagation is significantly improved. This allows for a more accurate reflection of fracture behavior under complex geological conditions. Furthermore, by employing on-site real-time monitoring equipment to collect pumping data and geological parameters, real-time tracking of the fracture propagation process is achieved, enabling timely responses to geological changes and providing timely decision support for operators. During pumping operations, changes in pumping parameters can be monitored in real time, and the fracture propagation prediction results can be readjusted based on new data, demonstrating good adaptability and flexibility, and optimizing the operation process. By converting pumping procedure data into image data and fusing it with geological parameters, the limitations of traditional methods in data processing and analysis are overcome, enabling a more comprehensive understanding of the impact of the pumping process on fracture propagation.

[0434] Therefore, the present invention also relates to a real-time prediction device for crack propagation, the real-time prediction device comprising:

[0435] The acquisition module is used to acquire geological parameters, pumping image parameters, and sample data of fracture propagation corresponding to the pumping image parameters, wherein the pumping image parameters are images obtained based on the pumping parameters.

[0436] The data fusion module is used to fuse the geological parameters and pumping image parameters to obtain fused feature data;

[0437] The model building module is used to build an initial prediction model based on machine learning algorithms. The fused feature data is used as the input of the initial prediction model, the crack expansion amount is used as the output of the initial prediction model, and the initial prediction model is trained using the corresponding sample data to obtain a crack expansion prediction model.

[0438] The prediction module is used to make real-time predictions of crack propagation using the crack propagation prediction model.

[0439] In a real-time crack propagation prediction device according to an embodiment of the present invention, the acquisition module is configured as follows:

[0440] Geological parameters, pumping parameters, and fracture propagation corresponding to the pumping parameters are obtained. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure.

[0441] The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

[0442] In a real-time crack propagation prediction device according to an embodiment of the present invention, the acquisition module is configured as follows:

[0443] The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

[0444] This invention also provides a computer-readable storage medium storing at least one computer-executable program. When executed by the computer, the at least one program causes the computer to perform steps in the method for predicting the fracturing effect of heterogeneous reservoirs considering fracturing technology. The method includes:

[0445] The first step is to collect geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs as a dataset;

[0446] The second step is to use the dataset to build a fracturing effect prediction model for heterogeneous reservoirs;

[0447] The third step involves inputting the fracturing process parameters and the geological parameters of the heterogeneous reservoir into the fracturing effect prediction model of the heterogeneous reservoir, and the prediction model outputs the fracturing effect parameters of the heterogeneous reservoir.

[0448] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0449] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the fracturing effect of heterogeneous reservoirs considering fracturing technology, characterized in that, Includes the following steps: Geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs are collected as a dataset; A fracturing effect prediction model for heterogeneous reservoirs was constructed using a dataset. The fracturing process parameters and geological parameters are input into the fracturing effect prediction model, and the prediction model outputs the fracturing effect parameters. Based on the constructed fracturing effect prediction model, according to the heterogeneity of the heterogeneous reservoir and the permeability difference of each layer, the fracturing process parameters adapted to the characteristics of the heterogeneous reservoir are optimized and selected through a multi-objective optimization algorithm.

2. The method according to claim 1, characterized in that, The collected geological parameters include interlayer thickness, porosity, and permeability of heterogeneous reservoirs; and / or The collected fracturing process parameters include fracturing fluid viscosity, sand ratio, and pumping rate; and / or The collected characteristic parameters related to fracturing effect include fracture propagation range and fracture conductivity.

3. The method according to claim 1 or 2, characterized in that, Methods for acquiring fracturing process parameters include: A three-dimensional geological model of the heterogeneous reservoir is constructed based on the interlayer thickness, porosity, permeability, fracture density, location of the top and bottom boundaries of the formation, rock mechanical parameters, fracturing fluid viscosity, sand ratio, and pumping rate.

4. The method according to claim 1 or 2, characterized in that, The selection of fracturing process parameters through multi-objective optimization algorithms includes: (1) Establish the crack propagation range L f The objective function is expressed in the following formula: Where α is a constant, Q p Here, μ is the pumping rate, μ is the fracturing fluid viscosity, and E is the equivalent Young's modulus of the heterogeneous reservoir. E is calculated using the following formula: Where E is the Young's modulus of the heterogeneous reservoir, and ν is the Poisson's ratio of the heterogeneous reservoir; (2) The objective function for the fracture conductivity F is established as follows: Among them, C p Given the sand ratio, β is a constant, and P... p E is the pumping pressure, and E is the equivalent Young's modulus of the heterogeneous reservoir. (3) Generate an initial population P0 containing N individuals, where the fracturing process parameters for each individual are expressed as follows: Where, μ i Let X represent the fracturing fluid viscosity, sand ratio, pumping rate, and pumping pressure of the i-th individual, respectively. i The objective function values ​​f1 for the crack propagation range and f2 for the crack conductivity are calculated as follows: (4) Perform non-dominated ranking on each individual in the population, and select individuals A and B that are in a non-dominated relationship; divide the individuals into different levels according to the non-dominated relationship; Calculate the crowding distance d for each non-dominated individual i. i : Where M is the number of objective functions; For individuals within each non-dominated layer, sort them from largest to smallest by crowding distance; (5) Select suitable individuals based on non-dominated ordination and crowding distance, and perform crossover and mutation operations; The crossover operation is represented by the following formula: Where λ is a uniformly distributed random factor, typically ranging from [-1, 1]; r is a random number in the range [0, 1]; p c The crossover probability has a value range of [0, 1]. The mutation operation, the formula is: X j =X j +N(0,σ) Where N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ; The current population P t and the newly generated offspring population Q t Merge to form a new population R t The merged population is sorted non-dominated and the top N individuals are selected to form a new population P. t+1 Repeat the above steps until the predetermined number of iterations or convergence conditions are reached, and finally obtain the fracturing process parameters.

5. The method according to claim 3, characterized in that, Methods for acquiring characteristic parameters related to fracturing effectiveness include: Fracturing operations are carried out using fracturing process parameters adapted to the characteristics of the heterogeneous reservoir, and fracturing monitoring data are collected through a real-time sensor array at the fracturing site. The collected fracturing monitoring data were denoised and normalized to obtain preprocessed fracturing monitoring data. A joint feature extraction method based on PCA principal component analysis and SVM support vector machine was adopted to extract features related to fracturing effect from the preprocessed fracturing monitoring data, including fracture propagation range and fracture conductivity.

6. The method according to claim 1, characterized in that, The specific steps involve constructing a fracturing performance prediction model for heterogeneous reservoirs using a dataset, including: Construct a fracturing effect prediction model; A fracturing effect prediction model was constructed by training the dataset, and the trained fracturing effect prediction model was obtained. The trained fracturing effect prediction model was validated and optimized to obtain the optimized fracturing effect prediction model.

7. The method according to claim 6, characterized in that, The constructed fracturing effect prediction model is a multi-input multi-output structure, in which, The input variables of the model include: interlayer thickness of heterogeneous reservoirs, permeability, porosity, fracturing fluid viscosity, pumping rate, and sand ratio; The model's output variables include: crack propagation range and crack conductivity.

8. The method according to claim 6, characterized in that, The model training uses the gradient boosting decision tree algorithm to obtain the trained prediction model.

9. The method according to claim 6, characterized in that, The trained prediction model is validated and optimized to obtain the optimized prediction model. Specific operations include: The prediction results of the trained prediction model were compared with the actual fracturing effect data, and the training prediction model was verified by error analysis. The Bayesian optimization algorithm is used to optimize the trained prediction model in order to minimize the prediction error and improve the prediction accuracy and adaptability of the model, resulting in an optimized prediction model.

10. A system for predicting the fracturing effect of heterogeneous reservoirs considering fracturing technology, characterized in that, The system includes: The parameter acquisition unit is used to collect geological parameters, fracturing process parameters, and characteristic parameters related to fracturing effect of heterogeneous reservoirs as a dataset. The prediction model acquisition unit is used to construct a fracturing effect prediction model for heterogeneous reservoirs using a dataset. The prediction unit is used to input fracturing process parameters and geological parameters into the fracturing effect prediction model. The prediction model outputs fracturing effect parameters. Based on the constructed fracturing effect prediction model, according to the heterogeneity of the heterogeneous reservoir and the permeability difference of each layer, the fracturing process parameters adapted to the characteristics of the heterogeneous reservoir are optimized and selected through a multi-objective optimization algorithm.

11. The system according to claim 10, characterized in that, The collected geological parameters include interlayer thickness, porosity, and permeability of heterogeneous reservoirs; and / or The collected fracturing process parameters include fracturing fluid viscosity, sand ratio, and pumping rate; and / or The collected characteristic parameters related to fracturing effect include fracture propagation range and fracture conductivity.

12. The system according to claim 10, characterized in that, The multi-objective optimization algorithm is used to optimize and select fracturing process parameters that are suitable for the heterogeneous reservoir characteristics, including: (1) Establish the crack propagation range L f The objective function is expressed in the following formula: Where α is a constant, Q p Here, μ is the pumping rate, μ is the fracturing fluid viscosity, and E is the equivalent Young's modulus of the heterogeneous reservoir. E is calculated using the following formula: Where E is the Young's modulus of the heterogeneous reservoir, and ν is the Poisson's ratio of the heterogeneous reservoir; (2) The objective function for the fracture conductivity F is established as follows: Among them, C p Given the sand ratio, β is a constant, and P... p E is the pumping pressure, and E is the equivalent Young's modulus of the heterogeneous reservoir. (3) Generate an initial population P0 containing N individuals, where the fracturing process parameters for each individual are expressed as follows: Where, μ i , Let X represent the fracturing fluid viscosity, sand ratio, pumping rate, and pumping pressure of the i-th individual, respectively. i The objective function values ​​f1 for the crack propagation range and f2 for the crack conductivity are calculated as follows: (4) Perform non-dominated ranking on each individual in the population, and select individuals A and B that are in a non-dominated relationship; divide the individuals into different levels according to the non-dominated relationship; Calculate the crowding distance d for each non-dominated individual i. i : Where M is the number of objective functions; For individuals within each non-dominated layer, sort them from largest to smallest by crowding distance; (5) Select suitable individuals based on non-dominated ordination and crowding distance, and perform crossover and mutation operations; The crossover operation is represented by the following formula: Where λ is a uniformly distributed random factor, typically ranging from [-1, 1]; r is a random number in the range [0, 1]; p c The crossover probability has a value range of [0, 1]. The mutation operation, the formula is: X j =X j +N(0,σ) Where N(0,σ) represents a normal distribution with a mean of 0 and a standard deviation of σ; The current population P t and the newly generated offspring population Q t Merge to form a new population R t The merged population is sorted non-dominated and the top N individuals are selected to form a new population P. t+1 Repeat the above steps until the predetermined number of iterations or convergence conditions are reached, and finally obtain the fracturing process parameters.

13. The system according to claim 10, characterized in that, The parameter acquisition unit is configured to perform the following operations: Fracturing operations are carried out using fracturing process parameters adapted to the characteristics of the heterogeneous reservoir, and fracturing monitoring data are collected through a real-time sensor array at the fracturing site. The collected fracturing monitoring data were denoised and normalized to obtain preprocessed fracturing monitoring data. A joint feature extraction method based on PCA principal component analysis and SVM support vector machine was adopted to extract features related to fracturing effect from the preprocessed fracturing monitoring data, including fracture propagation range and fracture conductivity.

14. The system according to claim 10, characterized in that, The prediction unit is configured to perform the following operations: Construct a fracturing effect prediction model; A fracturing effect prediction model was constructed by training the dataset, and the trained fracturing effect prediction model was obtained. The trained fracturing effect prediction model was validated and optimized to obtain the optimized fracturing effect prediction model.

15. The system according to claim 14, characterized in that, The fracturing effect prediction model is a multi-input multi-output structure, wherein, The input variables of the model include: interlayer thickness of heterogeneous reservoirs, permeability, porosity, fracturing fluid viscosity, pumping rate, and sand ratio; The model's output variables include: crack propagation range and crack conductivity.

16. The system according to claim 14, characterized in that, The model training uses the gradient boosting decision tree algorithm to obtain the trained prediction model.

17. The system according to claim 14, characterized in that, The trained prediction model is validated and optimized to obtain the optimized prediction model, including: The prediction results of the trained prediction model were compared with the actual fracturing effect data, and the training prediction model was verified by error analysis. The Bayesian optimization algorithm is used to optimize the trained prediction model in order to minimize the prediction error and improve the prediction accuracy and adaptability of the model, resulting in an optimized prediction model.

18. A real-time prediction method for crack propagation, characterized in that, include: Sample data of geological parameters, pumping image parameters, and corresponding fracture propagation of heterogeneous reservoirs are obtained, wherein the pumping image parameters are images obtained by transforming the pumping parameters; The geological parameters are fused with the pumping image parameters to obtain fused feature data; A prediction model is constructed based on machine learning algorithms, with the fused feature data as input and the crack propagation amount as output, and the crack propagation prediction model is obtained by training with sample data. The fracture propagation prediction model is used to predict fracture propagation in the heterogeneous reservoir in real time.

19. The real-time prediction method for crack propagation according to claim 18, characterized in that, The steps for obtaining sample data of geological parameters, pumping image parameters, and corresponding fracture propagation include: Geological parameters, pumping parameters, and fracture propagation corresponding to the pumping parameters are obtained. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure. The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

20. The real-time prediction method for crack propagation according to claim 19, characterized in that, The step of converting the pumping parameters into an image matrix capable of characterizing temporal changes during the pumping process includes: The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

21. The real-time prediction method for crack propagation according to claim 18, characterized in that, The steps for training a crack propagation prediction model using sample data include: The objective function of the initial prediction model is constructed based on the relative error between the predicted crack propagation amount output by the initial prediction model and the sample data of crack propagation amount. The initial prediction model is iterated based on the objective function, and the parameters of the initial prediction model are adjusted. When the objective function reaches its minimum value, the corresponding parameters are determined as the objective parameters of the initial prediction model. The crack propagation prediction model is determined based on the target parameters.

22. A real-time prediction device for crack propagation, characterized in that, include: The acquisition module is used to acquire geological parameters, pumping image parameters, and sample data of fracture propagation corresponding to the heterogeneous reservoir, wherein the pumping image parameters are images obtained based on the pumping parameters. The data fusion module is used to fuse the geological parameters and pumping image parameters to obtain fused feature data; The model building module is used to build an initial prediction model based on machine learning algorithms. The fused feature data is used as the input of the initial prediction model, the crack expansion amount is used as the output of the initial prediction model, and the initial prediction model is trained using the corresponding sample data to obtain a crack expansion prediction model. The prediction module is used to predict the fracture propagation of the heterogeneous reservoir in real time using the fracture propagation prediction model.

23. The real-time prediction device for crack propagation according to claim 22, characterized in that, The acquisition module is configured as follows: Geological parameters, pumping parameters, and fracture propagation corresponding to the pumping parameters are obtained. The geological parameters include at least one of formation stress, Young's modulus, Poisson's ratio, and physical property parameters. The pumping parameters include at least one of fracturing fluid discharge rate, stage fluid volume, sand ratio, and pumping pressure. The pumping parameters are converted into an image matrix that can characterize the temporal changes during the pumping process to obtain pumping image parameters.

24. The real-time prediction device for crack propagation according to claim 23, characterized in that, The acquisition module is configured as follows: The pumping parameters are converted into an image matrix using the Gram angle field algorithm.

25. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program that is executable by a computer, and when executed by the computer, the at least one computer program performs the steps in the fracturing effect prediction method considering fracturing process as described in any one of claims 1-9.

26. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program that is executable by a computer, which, when executed by the computer, performs the real-time prediction method for crack propagation as described in any one of claims 18 to 21.

Images