Shale reservoir fracturing pressure channeling early warning method, device, equipment and medium

Abstract

The invention discloses a shale reservoir fracturing pressure channeling early warning method, device, equipment and medium, the method comprises the steps that a training set, a verification set and a test set are obtained, and samples of the training set, the verification set and the test set all comprise data of the following characteristics: oil pressure, casing pressure, Young modulus and Poisson's ratio; constructing an LSTM model and a full three-dimensional crack control equation; determining model parameters of an LSTM model based on the training set and a naive Bayes algorithm to obtain an initial model; verifying the initial model based on the verification set to obtain a target model; and obtaining predicted Young modulus and Poisson's ratio based on the test set, the target model and a naive Bayes algorithm, and obtaining predicted oil pressure and casing pressure based on the predicted Young modulus and Poisson's ratio and a full-three-dimensional fracture control equation to serve as a judgment basis of pressure channeling early warning. According to the invention, real-time artificial intelligence early warning can be carried out on shale reservoir fracturing and pressure channeling, and the prediction accuracy is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of fracturing, and particularly relates to a method, device, equipment and medium for early warning of fracturing and channeling in shale reservoirs. Background Art

[0002] Due to the special geological reservoir structure in China, during the hydraulic fracturing process of shale gas reservoirs, any type of well interference or interaction may lead to the occurrence of downhole "fracture channeling" accidents. Fracture channeling is affected by the interaction of multiple factors such as natural fractures, well section length, pumping rate, fracturing fluid and proppant. When fracture channeling occurs, if effective countermeasures are not taken as soon as possible, it is easy to cause the failure of the fracturing construction, waste fracturing fluid and proppant, increase the cost of the fracturing construction, slow down the speed of the fracturing construction, and reduce the production life of oil wells. Therefore, to ensure the smooth progress of the fracturing construction, it is necessary to effectively identify the risk factors causing fracture channeling, evaluate the risk of downhole fracture channeling accidents, and send out alarm signals in time. However, in the prior art, the curve is usually observed manually and the fracture channeling is judged manually, which has problems such as lag and accident. Summary of the Invention

[0003] The main object of the present invention is to provide a method, device, equipment and medium for early warning of fracturing and channeling in shale reservoirs, so as to solve the problems of lagging judgment and large error caused by manual judgment of fracture channeling in the prior art.

[0004] According to one aspect of the present invention, a method for early warning of fracturing and channeling in shale reservoirs is proposed, including:

[0005] Obtaining a training set, a validation set and a test set, wherein the samples of the training set, the validation set and the test set all include data with the following characteristics: oil pressure, casing pressure, Young's modulus and Poisson's ratio;

[0006] Constructing an LSTM model and a full three-dimensional fracture control equation;

[0007] Determining the model parameters of the LSTM model based on the training set and the Naive Bayes algorithm to obtain an initial model;

[0008] Validating the initial model based on the validation set to obtain a target model;

[0009] Obtaining the predicted Young's modulus and Poisson's ratio based on the test set, the target model and the Naive Bayes algorithm, and obtaining the predicted oil pressure and casing pressure based on the predicted Young's modulus and Poisson's ratio and the full three-dimensional fracture control equation, as the judgment basis for fracture channeling early warning.

[0010] According to an embodiment of the present invention, the sample data of the training set and the validation set are obtained through actual detection or simulated based on the full three-dimensional fracture control equation.

[0011] According to an embodiment of the present invention, the construction of the LSTM model includes:

[0012] The LSTM model constructed at time t is:

[0013] Y t = σ(X t Z xf + h t-1 Z hf + b f )

[0014] The input layer of the LSTM model constructed for the i-th layer is:

[0015] I t = σ(X t Z xi + h t-1 Z hi + b i )

[0016] The output layer of the LSTM model constructed for the i-th layer is:

[0017] Q t = σ(X t Z xo + h t-1 Z ho + b o )

[0018] The computational model of the LSTM layer neuron cell is constructed as:

[0019]

[0020] Where:

[0021]

[0022] In the formula: Y t is the memory threshold at time t; σ is the sigmoid activation function; Z xi , Z xc , Z xo are the threshold weights of the i-th layer of the LSTM model; X t is the collected sample; h t-1 is the output value at time t-1; b i , b o are the bias term parameters; I t is the input threshold of the i-th layer at time t; Q t is the output threshold of the i-th layer at time t; C t is the state of the LSTM model neuron at time t; C t-1 is the state of the LSTM model neuron at time t-1; is the state update at time t; Z hf is the threshold weight value of the LSTM model at time t; Z hi is the threshold weight value of the i-th input layer of the LSTM model at time t; Z ho is the threshold weight value of the i-th output layer of the LSTM model at time t; Z xf is the threshold weight value of the LSTM model at time t; Z hc is the threshold weight value of the i-th layer of the LSTM model neuron; b f is the bias term parameter of the LSTM model; b c is the bias term parameter of the activation function under the updated state.

[0023] According to an embodiment of the present invention, the full three-dimensional fracture control equation includes:

[0024] Width equation:

[0025]

[0026] Two-dimensional flow equation of the fracturing fluid in the fracture:

[0027]

[0028] Total volume conservation equation of the fracturing fluid:

[0029]

[0030] In the formula:

[0031] w = w(x′, y′) is the fracture width at the point (x′, y′) on the fracture surface;

[0032] p(x, y) and σ p (x, y) are the fracturing fluid pressure and in-situ stress at the point (x, y) on the fracture surface;

[0033] r is the distance between the point (x′, y′) and the point (x, y);

[0034] G and v are the shear modulus and Poisson's ratio of the rock respectively;

[0035] Ω is the fracture surface;

[0036] is the fracture front;

[0037] q0 is the injection rate of the fracturing fluid entering a single wing, which is half of the actual injection rate;

[0038] q L is the volume filtration rate of the fracturing fluid per unit area of the fracture surface, and the calculation formula is as follows:

[0039]

[0040] In the formula, C L is the filtration coefficient of the fracturing fluid; τ is the moment when filtration starts at the point (x, y), and t is the construction time;

[0041] s is a calculation coefficient, and the calculation formula is as follows:

[0042]

[0043] In the formula, n is the flow state index of the power-law fluid; k is the consistency index; ω is the fracture width.

[0044] According to an embodiment of the present invention, determining the model parameters of the LSTM model based on the training set and the Naive Bayes algorithm includes:

[0045] Based on the training set and the Bayes formula, calculate the prior probabilities of the model parameters of the LSTM model under the constraint of the training set;

[0046] Assume that the threshold weights and bias term parameters of the LSTM model follow a uniform distribution, and the activation function follows a Gaussian distribution;

[0047] Based on the prior probabilities, the uniform distribution function formula, and the Gaussian distribution function formula, calculate the activation function, threshold weights, and bias term parameters of the LSTM model.

[0048] According to an embodiment of the present invention, obtaining the predicted Young's modulus and Poisson's ratio based on the test set, the target model, and the Naive Bayes algorithm includes:

[0049] Based on the test set, the target model, and the Naive Bayes algorithm, calculate the posterior probability of the occurrence of the feature under the constraint of the test set, and determine the predicted Young's modulus and Poisson's ratio based on the posterior probability.

[0050] According to an embodiment of the present invention, it further includes: comparing the predicted oil pressure and casing pressure with a preset risk database to determine whether a pressure channeling occurs, and sending a warning message when it is determined that a pressure channeling is about to occur.

[0051] According to another aspect of the present invention, a shale reservoir fracturing pressure channeling warning device is proposed, including:

[0052] An acquisition module configured to acquire a training set, a validation set, and a test set, and the samples of the training set, the validation set, and the test set all include data of the following features: oil pressure, casing pressure, Young's modulus, and Poisson's ratio;

[0053] A construction module configured to construct an LSTM model and a full three-dimensional fracture control equation;

[0054] A training module, configured to determine model parameters of the LSTM model based on the training set and the Naive Bayes algorithm to obtain an initial model;

[0055] A verification module, configured to verify the initial model based on the verification set to obtain a target model;

[0056] An early warning module, configured to obtain predicted Young's modulus and Poisson's ratio based on the test set, the target model, and the Naive Bayes algorithm, and obtain predicted oil pressure and casing pressure based on the predicted Young's modulus and Poisson's ratio and the full three-dimensional fracture control equation, as a determination basis for fracturing channeling early warning.

[0057] According to another aspect of the present invention, a computer device is provided, including:

[0058] At least one processor; and

[0059] A memory, the memory stores a computer program that can run on the processor, and when the processor executes the program, the method described in any of the above embodiments is implemented.

[0060] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method described in any of the above embodiments is implemented.

[0061] In the technical solution of the present invention, based on the LSTM model, the full three-dimensional fracture control equation, and the Naive Bayes algorithm, the predicted oil pressure and casing pressure are calculated as the determination basis for fracturing channeling early warning, so as to realize early prediction of fracturing channeling through an artificial intelligence method and ensure the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0063] Figure 1 A flowchart showing a method for early warning of fracturing channeling in a shale reservoir according to an embodiment of the present invention;

[0064] Figure 2 A graph showing the comparison of the prediction result accuracies of the training set according to an embodiment of the present invention;

[0065] Figure 3 A graph showing the comparison of the prediction result accuracies of the test set according to an embodiment of the present invention;

[0066] Figure 4 Shows the classification prediction result graph of the mixing matrix of the training set according to an embodiment of the present invention;

[0067] Figure 5 Shows the real-time prediction classification result graph of the field data according to an embodiment of the present invention;

[0068] Figure 6 Shows the longitudinal profile and the corresponding oil pressure curve measured by reservoir perforation according to an embodiment of the present invention;

[0069] Figure 7 Shows the schematic diagram of the shale reservoir fracturing and perforating warning device according to an embodiment of the present invention. Detailed implementation manners

[0070] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the following further describes the embodiments of the present invention in detail with reference to specific embodiments and the accompanying drawings.

[0071] It should be noted that all the expressions using "first" and "second" in the embodiments of the present invention are used to distinguish two entities or parameters with the same name but different, so it can be seen that "first" and "second" are only for the convenience of expression and should not be understood as a limitation on the embodiments of the present invention. This will not be elaborated one by one in the subsequent embodiments.

[0072] Refer to Figure 1 , the present invention proposes a method for warning shale reservoir fracturing and perforating, including the following steps:

[0073] S1. Obtain a training set, a validation set and a test set, wherein the samples of the training set, the validation set and the test set all include data with the following characteristics: oil pressure, casing pressure, Young's modulus and Poisson's ratio;

[0074] S2. Construct an LSTM model and a full three-dimensional fracture control equation;

[0075] S3. Determine the model parameters of the LSTM model based on the training set and the Naive Bayes algorithm to obtain an initial model;

[0076] S4. Validate the initial model based on the validation set to obtain a target model;

[0077] S5. Obtain the predicted Young's modulus and Poisson's ratio based on the test set, the target model and the Naive Bayes algorithm, and obtain the predicted oil pressure and casing pressure based on the predicted Young's modulus and Poisson's ratio and the full three-dimensional fracture control equation, as the judgment basis for perforating warning.

[0078] In an embodiment of the present invention, based on the LSTM model, the full three-dimensional fracture control equation, and the Naive Bayes algorithm, the predicted oil pressure and casing pressure are calculated as the judgment basis for fracturing channeling early warning, so as to realize predicting fracturing channeling in advance through artificial intelligence methods and ensure the accuracy of prediction. The present invention combines the LSTM model and the Bayes algorithm for prediction, which has the advantages of a wide discrimination range and strong pertinence, and can greatly shorten the time for preventing accidents.

[0079] In some embodiments, in step S1, the sample data of the training set and the validation set are obtained through actual detection, for example, obtained based on well logging data, core data, etc. In other embodiments, the sample data of the training set and the validation set can also be obtained by simulating based on the full three-dimensional fracture control equation. The data of the training set can be obtained by collecting construction data at the operation site. The data of each feature can be further normalized. After normalization, the data is scaled to the range of 0 to 1 to ensure that the data is trained on the same scale. For each collected sample, preliminary training feature data can be extracted according to the practical experience of fracturing operations.

[0080] In some embodiments, in step S2, the full three-dimensional fracture control equation includes:

[0081] Width equation:

[0082]

[0083] Two-dimensional flow equation of fracturing fluid in the fracture:

[0084]

[0085] Total volume conservation equation of fracturing fluid:

[0086]

[0087] In the formula:

[0088] w = w(x′, y′) is the fracture width at the point (x′, y′) on the fracture surface;

[0089] p(x, y) and σ p (x, y) are the fracturing fluid pressure and in-situ stress at the point (x, y) on the fracture surface;

[0090] r is the distance between the point (x′, y′) and the point (x, y);

[0091] G and v are the shear modulus and Poisson's ratio of the rock respectively;

[0092] Ω is the fracture surface;

[0093] is the fracture front;

[0094] q0 is the injection rate of the fracturing fluid entering the single wing and is half of the actual injection rate;

[0095] q L is the volumetric filtration rate of the fracturing fluid per unit area of the fracture surface (filtrating simultaneously to both sides of the fracture), and the calculation formula is as follows:

[0096]

[0097] In the formula, C L is the fracturing fluid filtration coefficient; τ is the starting time of filtration at the point (x, y), and t is the construction time;

[0098] s is the calculation coefficient, and the calculation formula is as follows:

[0099]

[0100] In the formula, n is the flow behavior index of the power-law fluid; k is the consistency index; ω is the fracture width.

[0101] Most of the existing fracture propagation models assume that the fracture is straight and does not change direction during the extension process, ignoring the inertial terms in the fracture opening and fluid flow processes, and not considering the curved fractures in the laminated heterogeneous formation. The accuracy of establishing the empirical database of fracture extension by the original model is relatively low, and the targeting of each layer section is poor. To solve this problem, the present invention designs the above-mentioned full three-dimensional fracture control equation, which considers the inertial terms in the fracture opening and fluid flow processes to characterize the propagation of curved fractures in heterogeneous formations, and has strong applicability and good accuracy.

[0102] It should be noted that there is a parameter representing the pressure of the fracturing fluid in the fracture width equation. During the fracturing operation construction, the pressure of the fracturing fluid is directly or indirectly affected by the construction pump pressure and the oil pressure or casing pressure, that is, the pressure of the fracturing fluid is associated with the oil pressure and the casing pressure. In addition, the shear modulus and Young's modulus can be converted into each other, and the conversion formula is as follows:

[0103]

[0104] In the formula, G is the shear modulus; E is the Young's modulus; v is the Poisson's ratio.

[0105] Therefore, the full three-dimensional fracture control equation of the present application involves parameters associated with the four characteristics of Young's modulus, Poisson's ratio, oil casing, and casing pressure. Based on Young's modulus, Poisson's ratio, and the full three-dimensional fracture control equation, the corresponding oil pressure and casing pressure can be obtained.

[0106] In some embodiments, in step S2, constructing the LSTM model includes: constructing a machine learning LSTM network including an input layer of the breakdown warning system, a flattening layer, an LSTM layer, a fully connected layer, and a warning output layer. This network is used to learn the features of the data and issue warnings. Specifically:

[0107] Construct the LSTM model at time t as:

[0108] Y t = σ(X t Z xf + h t-1 Z hf + b f )

[0109] Construct the input layer of the LSTM model for the i-th layer as:

[0110] I t = σ(X t Z xi + h t-1 Z hi + b i )

[0111] Construct the output layer of the LSTM model for the i-th layer as:

[0112] Q t = σ(X t Z xo + h t-1 Z ho + b o )

[0113] Construct the neuron cell calculation model (i.e., the update function) of the LSTM layer as:

[0114]

[0115] Where:

[0116]

[0117] In the formula: Y t is the memory threshold at time t; σ is the sigmoid activation function; Z xi , Z xc , Z xo are the threshold weights of the i-th layer of the LSTM model; X t is the collected sample; h t-1 is the output value at time t-1; b i , b o are the bias term parameters; I t is the input threshold at time t of the i-th layer; Q t is the output threshold at time t of the i-th layer; Ct is the state of the neuron at time t in the LSTM model; C t-1 is the state of the neuron at time t-1 in the LSTM model; is the state update at time t; Z hf is the threshold weight value at time t in the LSTM model; Z hi is the threshold weight value at time t in the i-th input layer of the LSTM model; Z ho is the threshold weight value at time t in the i-th output layer of the LSTM model; Z xf is the threshold weight value at time t in the LSTM model; Z hc is the threshold weight value of the i-th layer of the neuron in the LSTM model; b f is the bias term parameter of the LSTM model; b c is the bias term parameter of the activation function in the updated state.

[0118] In some embodiments, in step S3, determining the model parameters of the LSTM model based on the training set and the Naive Bayes algorithm includes:

[0119] Calculating the prior probabilities of the respective model parameters of the LSTM model under the constraint of the training set based on the training set and Bayes' formula;

[0120] Assuming that the threshold weight values and bias term parameters of the LSTM model follow a uniform distribution, and the activation function follows a Gaussian distribution;

[0121] Calculating the activation function, threshold weight values, and bias term parameters of the LSTM model based on the prior probabilities, the uniform distribution function formula, and the Gaussian distribution function formula.

[0122] For a given training sample set D = (x1, x2,..., x n ), according to Bayes' formula, calculate the prior probabilities C i (i = 1, 2, 3, 4) of the respective parameters of the LSTM model under this sample condition. The prior probability calculation formula is as follows:

[0123] P(C i |D) = P(C i )P(D|C i ) / P(D)

[0124] Among them, P(C i |D) represents the probability of the occurrence of the model parameters under the constraint of the training set; P(C i ) represents the probability of the occurrence of the model parameters (i.e., the prior probability); P(D|C i ) represents the probability of the occurrence of the training set under the constraint of the model parameters; P(D) represents the probability of the occurrence of the training set.

[0125] The uniform distribution function formula is as follows:

[0126]

[0127] The Gaussian distribution function is as follows:

[0128]

[0129] In some embodiments, in step S4, validating the initial model based on the validation set includes: calculating the posterior probability of the occurrence of the feature under the constraint of the validation set and known model parameters based on the validation set, the initial model, and the Naive Bayes algorithm, determining Young's modulus and Poisson's ratio based on the posterior probability, and substituting Young's modulus and Poisson's ratio into the full three-dimensional fracture control equation to obtain the oil pressure and casing pressure, so as to perform validation based on the values of each feature (oil pressure, casing pressure, Young's modulus, and Poisson's ratio). According to Bayes' theorem, the MCMC algorithm is usually adopted, and the total probability formula is used to calculate the posterior probability under the constraint of the validation set R, and then the generalization ability of the LSTM model is obtained, that is, the correctness and robustness of the validation model are verified. The calculation formula for the posterior probability under the constraint of the validation set R is:

[0130] P(Y|R) = P(R|Y)P(Y) / P(R)

[0131] Wherein, P(Y|R) represents the probability of the occurrence of Young's modulus and Poisson's ratio under the constraint of the validation set (i.e., the posterior probability); P(R|Y) represents the probability of the validation set occurring under the constraint of Young's modulus and Poisson's ratio; P(Y) represents the probability of the occurrence of Young's modulus and Poisson's ratio; P(R) represents the probability of the validation set occurring.

[0132] In some embodiments, in step S5, obtaining the predicted Young's modulus and Poisson's ratio based on the test set, the target model, and the Naive Bayes algorithm includes:

[0133] Calculating the posterior probability of the occurrence of the feature under the constraint of the test set based on the test set, the target model, and the Naive Bayes algorithm, and determining the predicted Young's modulus and Poisson's ratio based on the posterior probability (selecting the feature value with the highest probability as the warning value, that is, as the basis for determining the pressure channeling warning). The method for calculating the posterior probability here is similar to the method for calculating the posterior probability in step S4. The calculation formula for the posterior probability under the constraint of the test set T is:

[0134] P(Y|T) = P(T|Y)P(Y) / P(T)

[0135] Wherein, P(Y|T) represents the probability of the occurrence of Young's modulus and Poisson's ratio under the constraint of the test set (i.e., the posterior probability); P(T|Y) represents the probability of the test set occurring under the constraint of Young's modulus and Poisson's ratio; P(Y) represents the probability of the occurrence of Young's modulus and Poisson's ratio; P(T) represents the probability of the test set occurring.

[0136] In some embodiments, in step S5, after obtaining the predicted oil pressure and casing pressure data, the data can be further visualized as, for example, a curve graph, and judgment can be made based on the visualized graph. In some embodiments, the method further includes: comparing the predicted oil pressure and casing pressure with a preset risk database to determine whether a pressure channeling occurs, and sending a warning message when it is determined that pressure channeling is about to occur.

[0137] According to the calculation results, in one embodiment, by Figure 2 it can be seen that the accuracy rate of the prediction result of the training set compared with the true value reaches 96.667%, and by Figure 3 it can be seen that the accuracy rate of the prediction result of the test set compared with the true value reaches 93.1624%. The present invention improves the traditional LSTM model by integrating the Naive Bayes algorithm to obtain the BO-LSTM model. By performing Bayesian inference on the model parameters, the robustness and generalization ability of the model are improved. In the BO-LSTM model, the weights and biases of each LSTM unit are modeled as random variables, and the posterior probability is calculated through the Bayesian formula. In this way, the BO-LSTM model can better process unbalanced data sets and has better classification and warning performance. In this way, the BO-LSTM model can obtain more accurate model parameters and can better adapt to different data sets and warning tasks.

[0138] The present invention has been verified by examples on shale gas wells, and the prediction results are as shown in Figure 4 and Figure 5 , and the figure respectively shows four classification prediction results. Figure 4 The classification prediction result of the confusion matrix from the training set gives the prior probability of the model; Figure 5 The real-time warning classification result from on-site data. Figure 5 The measured results are consistent with the predicted results, indicating the correctness of the Naive Bayes algorithm integrated based on the BO-LSTM model. Moreover, the model also has simplicity and high efficiency. The integrated Bayesian algorithm has a low computational complexity, has a fast computational speed and warning visualization ability for large-scale data sets existing in real time on site, is applicable to real-time pressure breakthrough warning identification at the fracturing well site, especially for the large-scale reservoir volume transformation fracturing mode at the fracturing site, and can improve the success rate of large-scale construction of volume fracturing. In addition, the Naive Bayes algorithm has relatively good robustness in dealing with missing data. Figure 6 The upper figure of Figure 6 is the measured monitoring result graph of reservoir pressure breakthrough. In the figure, the fracturing crack extends downward in the later stage of the operation, and it is determined that there is downward pressure channeling; correspondingly,

[0139] Reference Figure 7 , the present invention further provides a shale reservoir fracturing and channeling warning device 100, including:

[0140] An acquisition module 10, configured to acquire a training set, a validation set, and a test set, and the samples of the training set, the validation set, and the test set all include data with the following characteristics: oil pressure, casing pressure, Young's modulus, and Poisson's ratio;

[0141] A construction module 20, configured to construct an LSTM model and a full three-dimensional fracture control equation;

[0142] A training module 30, configured to determine the model parameters of the LSTM model based on the training set and the Naive Bayes algorithm to obtain an initial model;

[0143] A verification module 40, configured to verify the initial model based on the validation set to obtain a target model;

[0144] An early warning module 50, configured to obtain the predicted Young's modulus and Poisson's ratio based on the test set, the target model, and the Naive Bayes algorithm, and obtain the predicted oil pressure and casing pressure based on the predicted Young's modulus and Poisson's ratio and the full three-dimensional fracture control equation, as the determination basis for fracturing and channeling early warning.

[0145] The present invention further provides a computer device, including: at least one processor; and a memory, the memory stores a computer program that can run on the processor, and when the processor executes the program, it implements the method described in any of the above embodiments.

[0146] The present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the method described in any of the above embodiments.

[0147] In summary, the present invention provides a method for real-time artificial intelligence warning of shale reservoir fracturing and channeling based on a machine learning BO-LSTM model fused with the Naive Bayes algorithm, and the method includes the following steps:

[0148] (1) Based on the theoretical model of full three-dimensional hydraulic fracturing crack extension under heterogeneous conditions, collect sample data including oil pressure, casing pressure, Young's modulus, and Poisson's ratio as the training set, validation set, and test set;

[0149] (2) Construct a BO-LSTM model fused with the Naive Bayes algorithm based on deep learning;

[0150] (3) Calculate the model parameters using the prior probability under the training set;

[0151] (4) Calculate the Young's modulus, Poisson's ratio, and the oil pressure and casing pressure parameters of the real-time monitoring data of fracture extension for the shale gas fracturing well under the validation set by using the LSTM model.

[0152] (5) Establish a time-axis probability window under the test set, adopt the posterior probability method to calculate the maximum probability of the parameter occurrence, and conduct real-time analysis of the risk factors to achieve breakdown warning.

[0153] The present invention also provides a full three-dimensional fracture control equation for the bending fracture propagation in heterogeneous formations considering the crack opening and the inertial term during fluid flow. The constructed full three-dimensional fracture control equation applies the long short-term memory neural network and integrates the naive Bayes algorithm to construct a breakdown risk judgment factor warning LSTM model; using the Bayesian probability method, the prior probability and posterior probability algorithms are adopted to calculate the breakdown warning features, and the one with the maximum probability of the feature occurrence is used as the judgment basis for breakdown warning.

[0154] The present invention analyzes and calculates the oil pressure and casing pressure data that are most sensitive to breakdown, and uses the deep learning method for time series prediction to respectively predict the values of Young's modulus, Poisson's ratio, oil pressure, and casing pressure, so as to give an early warning of breakdown. The present invention is based on the machine learning BO-LSTM model integrating the naive Bayes algorithm, which greatly improves the warning effect for accidents such as breakdown, and to a certain extent solves the problems of lag and accident existing in the existing method of manually observing curves on site and manually judging breakdown.

[0155] It should be noted that those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments.

[0156] In addition, it should be understood that the computer-readable storage medium herein (for example, a memory) can be a volatile memory or a non-volatile memory, or can include both a volatile memory and a non-volatile memory.

[0157] Those of ordinary skill in the art should understand that: the discussion of any above embodiment is only exemplary, and is not intended to imply that the scope (including the claims) disclosed by the embodiments of the present invention is limited to these examples; under the idea of the embodiments of the present invention, the technical features between the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the embodiments of the present invention as described above, which are not provided in detail for the sake of brevity. Therefore, any omission, modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present invention shall be included in the protection scope of the embodiments of the present invention.

Claims

1. A method for predicting hydraulic fracturing and crossflow in shale reservoirs, characterized in that Including: Obtain a training set, a validation set, and a test set, where the samples of the training set, the validation set, and the test set all include data with the following characteristics: oil pressure, casing pressure, Young's modulus, and Poisson's ratio; Construct an LSTM model and a full three-dimensional fracture control equation; Based on the training set and the Naive Bayes algorithm, determine the model parameters of the LSTM model to obtain an initial model; Based on the validation set, validate the initial model to obtain a target model; Based on the test set, the target model, and the Naive Bayes algorithm, obtain the predicted Young's modulus and Poisson's ratio, and based on the predicted Young's modulus and Poisson's ratio and the full three-dimensional fracture control equation, obtain the predicted oil pressure and casing pressure as the determination basis for fracturing channeling warning.

2. The method according to claim 1, wherein The sample data of the training set and the validation set are obtained through actual detection or simulated based on the full three-dimensional fracture control equation.

3. The method according to claim 1, wherein The constructing of the LSTM model includes: Construct the LSTM model at time t as: Y t = σ(X t Z xf + h t-1 Z hf + b f ) Construct the input layer of the LSTM model of the i-th layer as: I t = σ(X t Z xi + h t-1 Z hi + b i ) Construct the output layer of the LSTM model of the i-th layer as: Q t = σ(X t Z xo + h t-1 Z ho + b o ) Construct the neuron cell calculation model of the LSTM layer as: Where: Where: Y t is the memory threshold at time t; σ is the sigmoid activation function; Z xi 、Z xc 、Z xo are the threshold weights of the i-th layer of the LSTM model; X t is the collected sample; h t-1 is the output value at time t-1; b i 、b o are the bias term parameters; I t is the input threshold of the i-th layer at time t; Q t is the output threshold of the i-th layer at time t; C t is the state of the LSTM model neuron at time t; C t-1 is the state of the LSTM model neuron at time t-1; is the state update at time t; Z hf is the threshold weight of the LSTM model at time t; Z hi is the threshold weight of the i-th input layer of the LSTM model at time t; Z ho is the threshold weight of the i-th output layer of the LSTM model at time t; Z xf is the threshold weight of the LSTM model at time t; Z hc is the threshold weight of the i-th layer of the LSTM model neuron; b f is the bias term parameter of the LSTM model; b c is the bias term parameter of the activation function in the updated state.

4. The method according to claim 1, wherein The full three-dimensional fracture control equation includes: Width equation: Two-dimensional flow equation of fracturing fluid in the fracture: Total volume conservation equation of fracturing fluid: In the formula: w = w(x′, y′) is the fracture width at the point (x′, y′) on the fracture surface; p(x,y) and σ p (x, y) is the fracturing fluid pressure and in-situ stress at the point (x, y) on the fracture surface; r is the distance between the point (x′, y′) and the point (x, y); G and v are the shear modulus and Poisson's ratio of the rock respectively; Ω is the fracture surface; is the crack front; q0 is the injection rate of fracturing fluid entering a single wing, which is half of the actual injection rate; q L is the volume filtration rate of the fracturing fluid per unit area of the fracture surface, and the calculation formula is as follows: where C L is the filtration coefficient of the fracturing fluid; τ is the moment when filtration starts at the point (x, y), and t is the construction time; s is a calculation coefficient, and the calculation formula is as follows: In the formula, n is the flow state index of the power-law fluid; k is the consistency index; ω is the fracture width.

5. The method according to claim 3, wherein The determining of the model parameters of the LSTM model based on the training set and the Naive Bayes algorithm includes: Based on the training set and the Bayes formula, calculate the prior probabilities of the model parameters of the LSTM model under the constraint of the training set; Assume that the threshold weights and bias term parameters of the LSTM model follow a uniform distribution, and the activation function follows a Gaussian distribution; Based on the prior probabilities, the uniform distribution function formula, and the Gaussian distribution function formula, calculate the activation function, threshold weights, and bias term parameters of the LSTM model.

6. The method according to claim 1, characterized in that, The obtaining of the predicted Young's modulus and Poisson's ratio based on the test set, the target model, and the Naive Bayes algorithm includes: Based on the test set, the target model, and the Naive Bayes algorithm, calculate the posterior probabilities of the appearance of the characteristics under the constraint of the test set, and determine the predicted Young's modulus and Poisson's ratio based on the posterior probabilities.

7. The method according to claim 1, characterized in that It also includes: Compare the predicted oil pressure and casing pressure with a preset risk database to determine whether fracturing channeling occurs, and send a warning message when it is judged that fracturing channeling is about to occur.

8. A fracturing and fracturing breakthrough warning device for shale reservoirs, characterized in that, Including: An acquisition module configured to acquire a training set, a validation set, and a test set, where the samples of the training set, the validation set, and the test set all include data with the following characteristics: oil pressure, casing pressure, Young's modulus, and Poisson's ratio; A construction module configured to construct an LSTM model and a full three-dimensional fracture control equation; A training module, configured to determine model parameters of the LSTM model based on the training set and the Naive Bayes algorithm, and obtain an initial model; A verification module, configured to verify the initial model based on the verification set to obtain a target model; An early warning module, configured to obtain predicted Young's modulus and Poisson's ratio based on the test set, the target model and the Naive Bayes algorithm, and obtain predicted oil pressure and casing pressure based on the predicted Young's modulus and Poisson's ratio and the full three-dimensional fracture control equation, as a determination basis for fracturing warning.

9. A computer device, comprising: At least one processor; And A memory, the memory storing a computer program that can run on the processor, wherein when the processor executes the program, the method according to any one of claims 1-7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the method according to any one of claims 1-7 is implemented.

Images