Intelligent reservoir transformation method

Abstract

The invention discloses an intelligent reservoir transformation method, and belongs to the technical field of reservoir transformation. The method comprises the following steps: firstly, extracting data from a historical database, constructing fusion features containing engineering parameters and expert knowledge, and training and storing a construction parameter prediction model and a transformation effect prediction model according to the fusion features; and for a well to be transformed, obtaining a category label of the well to be transformed through an improved k-means algorithm, processing fusion features of the well to be transformed through a perturbation function family, and inputting the processed fusion features into the trained XGBoost model to obtain transformation parameters. And then, splicing the reconstruction parameters, the fusion features of the well to be reconstructed and the category labels, and inputting the spliced reconstruction parameters, the fusion features and the category labels into the trained deep neural network model to obtain a reconstruction effect prediction vector. And finally, automatic optimization is carried out through a fitness calculation function, and optimal transformation parameters are output. According to the method, the intelligent level and accuracy of reservoir transformation parameter selection are remarkably improved, and therefore the transformation effect is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of reservoir reconstruction, and particularly relates to an intelligent reservoir reconstruction method. BACKGROUND

[0002] In the process of oil and gas exploration and development, the physical characteristics (such as porosity, permeability, brittleness index, oil and gas saturation, etc.) of the reservoir not only determine the occurrence state and recoverable potential of the oil and gas, but also directly determine the engineering design and construction strategy of the exploration well. However, these physical characteristic parameters cannot be independently obtained only by relying on seismic or logging data, and must be identified and quantified through the well connecting geology and engineering. The well bears a dual mission in the whole life cycle: it is a detection window for recognizing the characteristics of the reservoir, and reveals the underground physical properties through logging, coring and testing, etc.; it is also an implementation carrier for reconstructing the structure of the reservoir, and actively regulates the reservoir by means of fracturing, acidizing and other processes.

[0003] In the reconstruction process, a large number of engineering parameters (discharge, pump pressure, proppant concentration, etc.) generated by well construction reflect the dynamic response characteristics of the reservoir around the wellbore in real time, not only embodying the sensitivity of the reservoir to the reconstruction measures, but also providing a key basis for evaluating the reconstruction effect and optimizing the subsequent scheme. The complexity of the reservoir determines the adaptability and challenge of the well technology, and the data obtained in the implementation process of the well and the reconstruction effect continuously promote the revision and cognitive iteration of the reservoir model, forming a closed-loop system of "recognizing the reservoir by the well, reconstructing the reservoir by the well, and optimizing the model by the well".

[0004] In the prior art, although there are analysis methods for single well or single data, there is a lack of an intelligent modeling technology system that can systematically integrate the physical characteristics of the reservoir, well engineering parameters and expert experience. This deficiency makes it difficult to realize the optimization migration of the reservoir reconstruction effect across wells and across work areas, and also restricts the reconstruction efficiency and capacity improvement under complex reservoir conditions. SUMMARY

[0005] The purpose of the present application is to solve the problems of the prior art that the reservoir reconstruction effect is difficult to realize the optimization migration across wells and across work areas, and the reconstruction efficiency and capacity under complex reservoir conditions.

[0006] An intelligent reservoir reconstruction method, comprising the following steps:

[0007] S1, based on a historical reservoir reconstruction database, constructing the fusion features, construction parameter labels, reconstruction effects and category labels of the historical reconstruction wells.

[0008] S2, construct a construction parameter prediction model and a reconstruction effect prediction model, train the construction parameter prediction model to obtain a trained construction parameter prediction model using the fused features and the construction parameter label, and train the reconstruction effect prediction model to obtain a trained reconstruction effect prediction model using the fused features, the construction parameter label and the category label.

[0009] S3, obtain a fused feature set of the well to be reconstructed, and calculate the category label of the well to be reconstructed using the improved k-means algorithm.

[0010] S4, apply disturbance to the fused feature set of the well to be reconstructed using a family of disturbance functions to obtain disturbed fused features, and input the disturbed fused features into the trained construction parameter prediction model to obtain reconstruction parameters.

[0011] S5, concatenate the reconstruction parameters, the fused features of the well to be reconstructed and the category label of the well to be reconstructed, and input them into the trained reconstruction effect prediction model to obtain a reconstruction effect prediction vector;

[0012] S6, calculate the fitness of the reconstruction effect prediction vector using a fitness calculation function, and select the optimal reconstruction parameters based on the fitness.

[0013] The above technical scheme forms a closed-loop intelligent system covering data input, model training and parameter output through the construction of a historical data modeling, a well to be reconstructed classification, a parameter disturbance optimization, an effect prediction and an optimal decision-making whole-process intelligent technical framework, the construction of features by fusing engineering parameters and expert knowledge, the use of improved k-means clustering, XGBoost parameter prediction, deep neural network (DNN) effect evaluation and fitness screening model. The technical scheme solves the technical problems of the existing reservoir reconstruction method, such as the dependence on artificial experience, the lack of intelligent parameter optimization logic of multi-model cooperation, the low accuracy of reconstruction parameter selection for complex reservoirs (such as sand-shale interaction and heterogeneous formation), the poor adaptability of cross-work area schemes, and the inability to balance productivity, cost and risk.

[0014] Further, the S1 is specifically based on a historical reservoir reconstruction database, and engineering parameters G and expert knowledge rule data K of historical reconstruction wells are constructed, and the fused features F are calculated based on the engineering parameters G and the expert knowledge rule vector K of the historical reconstruction wells.

[0015]

[0016] wherein Concat() is a vector concatenation function; is a weight parameter for data fusion;

[0017] The construction parameter label t = (Q, P, S) is constructed, wherein Q is the construction displacement, P is the pump injection pressure, and S is the proppant concentration.

[0018] The construction modification effect W=(E, V, B) is constructed, wherein E is the capacity improvement efficiency; V is the fracture conductivity efficiency; and B is the economic index;

[0019] The category label of the historical modification well is constructed, wherein the category label includes: a high permeability well, a medium permeability well and a low permeability well.

[0020] The above technical scheme clearly defines the standardized construction logic of the fusion features of the historical modification well (the weighted splicing of the engineering parameter vector and the expert knowledge rule vector), defines the specific dimensions of the construction parameter label (displacement, pump pressure and proppant concentration) and the modification effect label (capacity improvement, fracture conductivity and economy / safety), and provides a standardized and high-quality data set for subsequent model training. The technical problems of the prior art, such as no unified weight distribution in feature fusion, ambiguous label dimensions, low model training data quality, slow convergence speed, large effect prediction error, and low utilization rate of historical data (such as unreasonable combination of expert knowledge and engineering parameters), are solved.

[0021] Further, the S3 specifically includes the following steps:

[0022] S3-1, the engineering parameter vector and the expert knowledge rule vector of the m wells to be modified are constructed into the fusion feature set X of the wells to be modified {T j};

[0023] S3-2, the initial cluster center U is extracted from the fusion feature set X of the wells to be modified by using the improved k-means algorithm {C v};

[0024] S3-3, the improved k-means algorithm is initialized by using the initial cluster center U {C v}, and the category label of the well to be modified is calculated by using the improved k-means algorithm .

[0025] In the above implementation process, the engineering parameters and the expert knowledge rule vector of the well to be modified are converted into standardized fusion features, the clustering classification of the well to be modified is realized by improving the k-means algorithm, and the category label is output to provide a category for subsequent XGBoost parameter prediction, so as to ensure that the parameter prediction is consistent with the reservoir characteristics of the well to be modified.

[0026] Further, the S3-2 specifically includes the following steps:

[0027] S3-2-1, an initial cluster center C1 is randomly selected from the fusion feature set X of the well to be modified:

[0028]

[0029] wherein random (1,...,m) represents a number randomly selected from 1,...,m;

[0030] S3-2-2, selecting the vth cluster center.

[0031] 1) Calculate the minimum bi-space weighted distance of each data point T j in the fusion feature set X={T j of the well to be transformed to all the selected cluster centers {C r}=(C1,...,C v-1 ):

[0032]

[0033] wherein C r represents all the selected cluster centers; G j represents the geological sub-vector of the fusion feature T j of the jth well to be transformed, G j =Sg*T j , Sg represents the selection matrix; represents the center vector of the geological subspace; represents the geological attribute weight matrix; and a and β represent the center weight values;

[0034] 2) Calculate the probability P(T j ) that all the data in the fusion feature X of the well to be transformed are selected as the next cluster center, which is calculated as follows:

[0035]

[0036] wherein represents the sum of the minimum bi-space weighted distances of all the data in X={T j} to all the selected initial cluster center points; and y=1,...,m.

[0037] 3) Randomly select the next cluster center according to the probability P(T j ), which is specifically as follows:

[0038] Calculate the cumulative distribution Y(T j ) of the probability P(T j ):

[0039]

[0040] generate a random number in the interval [0,1);

[0041] select the first one that satisfies the cumulative distribution Y(T j )≥​​ T j As the v-th cluster center, and set v = v + 1;

[0042] S3-2-3, Complete the extraction of the initial cluster centers, specifically as follows:

[0043] Repeat S3-2-2 to extract K initial cluster centers U={C}. v Save it for later use.

[0044] In the above implementation process, the initial cluster centers of the fusion features of the well to be modified are extracted through random point selection, dual-space weighted distance calculation, and probabilistic point selection iteration process of the improved k-means algorithm. The dual-space weighted distance takes into account both overall features and key geological parameters, and the probabilistic point selection avoids the initial centers from being concentrated in local areas.

[0045] Furthermore, S3-3 specifically includes the following steps:

[0046] S3-3-1, Set the initial cluster centers U={C v} as the initial cluster center of the improved k-means algorithm (0) Specifically:

[0047] The initial cluster centers U={C v} is recorded as

[0048] S3-3-2 Clustering the fusion feature set X of the well to be modified includes the following steps:

[0049] ① Calculate the fusion characteristic T of each well to be modified j The weighted distance to all cluster centers generated in the (t-1)th iteration is used, and this distance is assigned to the cluster represented by the nearest cluster center. ;

[0050] Sample T j Cluster number assigned The calculation method is as follows:

[0051]

[0052] In the formula, ; ; It is an integer. Represents sample features T j The cluster number assigned in the t-th iteration, and ;

[0053] By ① combining the fusion characteristics T of each well to be modified jis divided into K clusters, denoted as wherein:

[0054]

[0055] wherein, represents the cluster numbered v, represents the T j of the cluster numbered v that satisfies

[0056] ②Calculate the mean value of all data points contained in each cluster , and record this mean value as the new center of the cluster :

[0057]

[0058]

[0059] wherein, represents the number of data points in each cluster ; represents the sum of the features of the data points in the cluster ;

[0060] ③Check the convergence of the improved k-means algorithm;

[0061] When the algorithm satisfies any of the following convergence conditions, it is considered to have converged, and the iteration process is stopped, otherwise let t=t+1, and return to ① to continue the iteration process:

[0062] 1、 and , wherein, is the geological space convergence threshold; is the geological subspace convergence threshold;

[0063] 2、 ;

[0064] 3、 , wherein, is the maximum iteration step;

[0065] S3-3-3, output the clustering result;

[0066] When the iteration process is finished, each fusion feature T j of the well to be transformed is assigned a final cluster label, which is the class label of the class to which the fusion feature T j of the well to be transformed belongs .

[0067] In the implementation process, the clustering iteration process of the improved k-means algorithm is used, and three convergence judgment conditions (center change threshold, objective function change threshold, cluster assignment consistency, and maximum iteration number) are set to ensure efficient and stable termination of the clustering process, thereby obtaining the class label.

[0068] Further, the S4 specifically comprises the following steps:

[0069] After perturbing the fusion feature set X of the jth well to be transformed, the perturbed fusion feature is obtained, specifically:

[0070]

[0071]

[0072] Among them: are the standard basis vectors corresponding to the displacement, pump pressure, and proppant concentration, respectively; are the perturbation amplitudes of the displacement, pump pressure, and proppant concentration, respectively;

[0073] The perturbed fusion feature is input into the trained construction parameter prediction model to obtain the corresponding transformation parameters , and the calculation method is as follows:

[0074]

[0075] Among them, , L = 0,..., 4; represents the construction displacement of the Lth group of perturbation schemes of the jth well to be transformed, represents the pump injection pressure of the Lth group of perturbation schemes of the jth well to be transformed, represents the proppant concentration of the Lth group of perturbation schemes of the jth well to be transformed.

[0076] In the implementation process, the perturbation function family is used to perturb the fusion feature of the well to be transformed, generate perturbed features, and input them into the trained XGBoost model to output multiple groups of transformation parameters, forming a parameter candidate set covering different scenarios.

[0077] Further, the reconstruction effect prediction vector in S5 is:

[0078]

[0079] Among them, L = 0,..., 4; represents the productivity improvement efficiency of the Lth group of perturbation schemes of the jth well to be transformed, ​a fracture conductivity of the jth well to be transformed in the Lth group of disturbance schemes, an economic / safety performance of the jth well to be transformed in the Lth group of disturbance schemes.

[0080] In the implementation process, the transformation parameters, the fusion features of the well to be transformed, and the category labels are spliced into an extended input vector, which is input into a DNN model to realize transformation effect prediction, and a three-dimensional effect vector including a productivity improvement performance, a fracture conductivity performance, and an economic / safety index is output, thereby providing a quantitative basis for fitness calculation and solving the problem of large prediction error.

[0081] Further, the S6 specifically includes the following steps:

[0082] ① Calculate the fitness of each transformation effect prediction value using a fitness calculation function .

[0083]

[0084] wherein, represents construction resource consumption of the candidate parameter combination, and the specific calculation method is as follows:

[0085]

[0086] In the formula, represents the construction displacement; represents the construction time; represents the construction liquid volume; represents the proppant usage; represents the number of stages; represents a unit cost coefficient, and , represents an adjustment coefficient, .

[0087] represents an operation risk, and the specific calculation method is as follows:

[0088]

[0089] In the formula, is a fracture deviation / overextension risk function; is a casing integrity risk function; is a sand plugging and sand return risk function; is a liquid excess induced formation damage risk function; r1, r2, r3, and r4 represent risk weight coefficients;

[0090] ② Based on the fitness calculated in ①, determine the optimal transformation parameter;

[0091] ​For the jth reformed well, the reformation parameter corresponding to the highest fitness is the optimal reformation parameter of the jth reformed well .

[0092] In the above implementation process, the comprehensive performance of different perturbation schemes is quantified through the fitness calculation function, the reformation parameter corresponding to the highest fitness is selected as the optimal scheme by comparing the fitness values of the five perturbation schemes.

[0093] Compared with the prior art, the beneficial effects of the present application are:

[0094] 1. Form a data benchmark, use improved k-means clustering to reformed wells and assign class labels, combine model adjustment parameters to achieve efficient adaptation of the scheme, improve migration accuracy and avoid reformation failure.

[0095] 2. The application generates candidate features with a perturbation function, inputs multiple sets of parameters to the model, breaks through the limitations of artificial selection, splices parameters and feature effect vectors input to the DNN, and predicts the effect to avoid risks; the optimal parameter is selected by using the fitness function, multiple targets are balanced, efficiency and capacity are improved, and the reformation effect is improved.

[0096] 3. The application constructs a closed-loop intelligent framework to realize seamless connection of data at each link and feedback data to the model; multiple models are integrated to form a decision chain, replacing artificial decision-making, shortening the scheme cycle, improving efficiency, reducing personnel dependence, and promoting the transformation to intelligent decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0097] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0098] Figure 1 The flowchart of the overall method involved in an embodiment of a reservoir reformation intelligent method. DETAILED DESCRIPTION

[0099] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0100] It should be noted that similar reference numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance, or implying any such actual relationship or order between these entities or operations. In addition, the terms "connected", "connected", and the like can be direct connection between elements, or indirect connection via other elements.

[0101] Referring to Figure 1 The present embodiment provides an intelligent reservoir reconstruction method, specifically comprising S1 to S6:

[0102] S1, based on a historical reservoir reconstruction database, constructing fusion features, construction parameter labels, reconstruction effects and category labels of historical reconstruction wells.

[0103] There are several historical reconstruction wells in the historical reservoir reconstruction database. The engineering parameters G of each historical reconstruction well are 42 dimensions, specifically including 12 geological parameters and 30 construction parameters.

[0104] The 12-dimensional geological parameters include: porosity, permeability, layer thickness, shale content, brittleness index, Poisson's ratio, Young's modulus, minimum horizontal principal stress, closure pressure, formation temperature, formation pressure, gas / oil saturation.

[0105] The 30-dimensional construction parameters include: design displacement, measured displacement, design pump pressure, measured pump pressure, segment length, stage injection time, preflush volume, sand-carrying fluid volume, total fluid volume, total amount of proppant, stage number, cluster number, cluster spacing, sand ratio average, sand ratio average peak value, sand ratio average rising slope, viscosity, acid volume, liquid density, friction estimation, proppant particle size, fixed barrel, quartz sand proppant, coated sand proppant, ceramic proppant, slickwater / crosslinked glue / acidification; construction strategy signs include: staged synchronous sand carrying or stepwise sand adding.

[0106] All parameters in engineering parameter G are represented by One-Hot encoding.

[0107] The expert knowledge rule data K of each historical reconstruction well includes reservoir type, brittle and plastic discrimination, oil and gas bearing property, stress difference characteristics, fracture complexity experience, preferred fracturing fluid system, preferred proppant type, construction strategy selection, common risk discrimination, and economic and safety experience coefficient.

[0108] The engineering parameter G of the historical reconstruction well is spliced with the expert knowledge rule data K to obtain the fusion feature F, which is specifically represented as:

[0109]

[0110] Wherein, Concat() is a vector splicing function; is a weight parameter of data fusion, satisfying , and .

[0111] The construction parameter label t=(Q, P, S) is constructed, wherein Q represents the construction displacement (unit: m³ / min), P represents the pump injection pressure (unit: MPa), and S represents the proppant concentration (unit: kg / m³).

[0112] The reconstruction effect W=(E, V, B) of the historical reconstruction well is constructed, wherein E represents the capacity improvement efficiency, V represents the fracture conductivity efficiency, and B represents the economic index.

[0113] The category label of the historical reconstruction well is constructed, and the category label of the historical reconstruction well is divided into three categories, namely, high permeability well, medium permeability well and low permeability well, and each historical reconstruction well has only one category label.

[0114] S2, the fusion feature and the construction parameter label are used to train the construction parameter prediction model to obtain the trained construction parameter prediction model, and the fusion feature, the construction parameter label and the category label are spliced to train the reconstruction effect prediction model to obtain the trained reconstruction effect prediction model.

[0115] The construction parameter prediction model is constructed based on the XGBoost network, the fusion feature F and the construction parameter label t are used to supervise the training of the construction parameter prediction model, and the trained construction parameter prediction model is obtained.

[0116] The reconstruction effect prediction model is constructed based on the deep neural network (DNN), the fusion feature F, the construction parameter label t and the category label are spliced to train the reconstruction effect prediction model, and the trained reconstruction effect prediction model is obtained.

[0117] S3, obtain a fusion feature set of the to-be-reformed well, and calculate a category label of the to-be-reformed well by using an improved k-means algorithm .

[0118] The S3 specifically includes the following steps:

[0119] S3-1, engineering parameters Q of m to-be-reformed wells and expert knowledge rule data R are preprocessed and constructed into a fusion feature set X = {T j} of the to-be-reformed wells, wherein j = 1,..., m.

[0120] It should be noted that the method of constructing the fusion feature set of the to-be-reformed well is consistent with the method of constructing the historical reformed well fusion feature in S1, and therefore will not be repeated here.

[0121] S3-2, an initial cluster center {C v} is extracted from the fusion feature set X of the to-be-reformed well by using an improved k-means algorithm, wherein v = 1,..., K.

[0122] The S3-2 specifically includes the following steps:

[0123] S3-2-1, a first initial cluster center C1: j is randomly selected from the fusion feature set X = {T

[0124]

[0125] Wherein, random(1,..., m) represents a number randomly selected from 1,..., m.

[0126] S3-2-2, the vth cluster center is selected.

[0127] 1) the minimum double-space weighted distance of all data in the fusion feature set X of the to-be-reformed well to all selected cluster centers C r = (C1,..., C v-1 ) is calculated :

[0128]

[0129] Wherein, C r represents all selected cluster centers; G j represents a geological sub-vector of the fusion feature T j of the jth to-be-reformed well, G j = Sg*T j , Sg is a selection matrix; represents a center vector of the geological subspace, ; The geological attribute weight matrix , , represents a diagonal matrix with diagonal elements on the main diagonal line being , and the diagonal elements of the matrix are used to characterize the importance of the pth geological parameter in well classification and transformation effect, where p = 1,..., 12; the central weight value is greater than 0, .

[0130] The selection matrix is constructed in the following way: the selection matrix Sg takes the value of 1 for the geological parameter dimension of the fusion features of the well to be transformed, and takes the value of 0 for the remaining positions;

[0131] It is worth mentioning that the value of the internal weight of the geological attribute weight matrix is determined according to rock mechanics and seepage theory, wherein the sensitivity of porosity and permeability to productivity and flow conductivity is the highest, so the weight value range of porosity and permeability is set to 0.2-0.3, while the sensitivity of brittleness index and Young's modulus to productivity and flow conductivity is the second, so the weight value range of brittleness index and Young's modulus is set to 0.1-0.15, while the sensitivity of horizon thickness, shale content, Poisson's ratio, minimum horizontal principal stress, closure pressure, formation temperature, formation pressure, and gas / oil saturation to productivity and flow conductivity is weaker, so the corresponding weight value range is set to 0.05-0.1.

[0132] 2) Calculate the probability P(T j ) that all data of the fusion feature set X of the well to be transformed are selected as the vth cluster center.

[0133]

[0134] wherein, represents the sum of the minimum bispacial weighted distances from all data in the fusion feature set X = {T j} of the well to be transformed to all selected initial cluster centers; y = 1,..., m.

[0135] 3) Select the vth cluster center according to the probability P(T j ), and the specific method is as follows:

[0136] Calculate the cumulative distribution Y(T j ) of the probability P(T j ) as follows:

[0137]

[0138] Generate a random number in the interval [0, 1) .

[0139] select the first T j that satisfies the cumulative distribution Y(T as the vth cluster center, and let v=v+1. j

[0140] S3-2-3, complete the extraction of initial cluster centers, specifically:

[0141] repeat S3-2-2 until K initial cluster centers U={C v} are extracted, where v=1,...,K.

[0142] S3-3, based on the initial cluster centers U={C v} extracted in S3-2, initialize the improved k-means algorithm, and use the improved k-means algorithm to cluster the fusion features {T j} of the wells to be transformed.

[0143] (1) take the initial cluster centers U={C v} as the initial cluster centers Center (0) of the improved k-means algorithm, specifically:

[0144] record the initial cluster centers U={C v} as

[0145] It is easy to understand that the K initial cluster centers {C1,C2,...,C K} are recorded as:

[0146]

[0147] and .

[0148] (2) cluster the fusion feature set X of the wells to be transformed, specifically including the following steps:

[0149] ① calculate the weighted distance of each fusion feature T j of the well to be transformed to all cluster centers produced in the first t-1 iterations, and assign it to the cluster represented by the nearest cluster center .

[0150] The number of clusters to which the sample T j is assigned is calculated as follows:

[0151]

[0152] It is easy to understand that​ ; ; is an integer, denotes the sample feature T j the number of the cluster to which the sample is assigned in the t-th iteration, and .

[0153] By ① merging the features T j of each well to be transformed into K clusters, denoted as , where:

[0154]

[0155] where, denotes the cluster numbered v, denotes the set of T j satisfying the cluster numbered v.

[0156] ② Calculate the mean of all data points contained in each cluster , and record this mean as the new center of the cluster:

[0157]

[0158]

[0159] where, denotes the number of data points in each cluster ; denotes the sum of the features of the data points in the cluster .

[0160] ③ Check the convergence of the improved k-means algorithm.

[0161] When the algorithm meets any of the following convergence conditions, it means that the algorithm has converged, and the iteration process is stopped, otherwise let t = t + 1, and return to ① to continue the iteration process:

[0162] 1、 and , where, is the convergence threshold of the geological space = 0.001 and is the convergence threshold of the geological subspace = 0.001;

[0163] 2、 , which means that the cluster assignment results of the adjacent two iterations are exactly the same;

[0164] 3、 , where, is the maximum iteration step, = 10000.

[0165] (3) Output the clustering result.

[0166] When the iteration process ends, each fusion feature T j of the well to be transformed is assigned a final cluster label, which is a class label representing the class of the fusion feature T j of the well to be transformed.

[0167] S4, input the fusion feature set X of the well to be transformed into the family of perturbation functions , to obtain the perturbed fusion feature , and input the perturbed fusion feature into the trained construction parameter prediction model to obtain the transformation parameter.

[0168] All data in the fusion feature set X = {T j} of the well to be transformed have the following calculation process:

[0169]

[0170]

[0171] Wherein: are the standard basis vectors corresponding to the displacement, pump pressure, and proppant concentration, respectively; are the perturbation amplitudes of the displacement, pump pressure, and proppant concentration, respectively, and the value range is set according to engineering experience.

[0172] It is easy to understand that after applying perturbation to each fusion feature T j of the well to be transformed, the perturbed fusion feature can be obtained, and each perturbed fusion feature contains five different perturbed to-be-measured sample sub-features , L = 0,..., 4, as follows:

[0173]

[0174] Input the perturbed fusion feature into the trained construction parameter prediction model to obtain the corresponding transformation parameter , and the calculation method is as follows:

[0175]

[0176] Wherein, ; , L = 0,..., 4; ​​QjL represents the construction rate (in m3 / min) of the jth well to be transformed in the Lth group of disturbance schemes, PjL represents the injection pressure (in MPa) of the jth well to be transformed in the Lth group of disturbance schemes, CjL represents the proppant concentration (in kg / m3) of the jth well to be transformed in the Lth group of disturbance schemes.

[0177] S5, the transformation parameters of the jth well are determined according to the transformation effect prediction vector , the fusion features T of the well to be transformed j and the category label After splicing, the transformation effect prediction parameters are obtained by inputting into the trained transformation effect prediction model :

[0178]

[0179] wherein, L = 0,..., 4; QjL represents the capacity improvement efficiency of the jth well to be transformed in the Lth group of disturbance schemes, QjL represents the fracture conductivity efficiency of the jth well to be transformed in the Lth group of disturbance schemes, QjL represents the economic index of the jth well to be transformed in the Lth group of disturbance schemes.

[0180] It is easy to understand that since L = 0,..., 4, for each determined j, five transformation effect prediction vectors can be obtained, which are Since j = 1,..., m, a total of 5*m transformation effect prediction values will be obtained.

[0181] S6, the fitness of each transformation effect prediction value is calculated by using the fitness calculation function , and the optimal transformation parameters are selected based on the fitness .

[0182] ①The fitness of each transformation effect prediction value is calculated by using the fitness calculation function , and the specific calculation method is as follows:

[0183]

[0184] wherein, w1 represents the first fitness weight, w2 represents the second fitness weight, w3 represents the third fitness weight, w1, w2 and w3 are all greater than 0 and less than 1, and w1+w2+w3=1.

[0185] represents the construction resource consumption of the candidate parameter combination, and the specific calculation method is as follows:

[0186]

[0187] wherein, represents the construction displacement; represents the construction time; represents the construction liquid volume; represents the proppant dosage; represents the number of stages; represents the unit cost coefficient, and

[0188] , represents the adjustment coefficient, .

[0189] represents the operation risk, and the specific calculation method is as follows:

[0190]

[0191] wherein, represents the fracture deviation / over-extended risk function; represents the casing integrity risk function; represents the sand plug and sand return risk function; represents the liquid excess induced formation damage risk function; r1, r2, r3 and r4 represent risk weight coefficients, and r1+r2+r3+r4=1, represents the adjustment coefficient, .

[0192] Further, the fracture deviation / over-extended risk function is calculated as follows:

[0193]

[0194] wherein, is the critical pressure of fracture expansion, which is obtained by field detection.

[0195] Further, the casing integrity risk function is calculated as follows:

[0196]

[0197] wherein, is the maximum pressure-bearing capacity of the casing, which is determined by the type of casing deployed in the field.

[0198] Further, the sand plug and sand return risk function is calculated as follows:

[0199]

[0200] wherein, The critical sand concentration is determined by the construction site conditions, The critical sand-carrying discharge is determined by the construction site conditions, The adjustment coefficient.

[0201] Further, the liquid excess initiates the formation damage risk function The calculation method is as follows:

[0202]

[0203] Wherein, The upper limit of the safety injection is determined by the construction site safety regulations.

[0204] ②Based on the fitness calculated in ①, the optimal reconstruction parameter is determined.

[0205] For the jth reconstruction well, five fitnesses are calculated in ① The reconstruction parameter corresponding to the highest fitness is selected as the optimal reconstruction parameter of the jth reconstruction well Similarly, for all m reconstruction wells, the above decision-making process is repeated, and the optimal reconstruction scheme set of the m reconstruction wells can be output j=1, 2,..., m.

[0206] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent method of reservoir modification, characterized in that, The method comprises the following steps: S1, based on the historical reservoir reconstruction database, the fusion features, construction parameter tags, reconstruction effects and category tags of the historical reconstruction wells are constructed; S2, construction parameter prediction model and reconstruction effect prediction model are constructed, the fusion features and construction parameter tags are used to train the construction parameter prediction model to obtain the trained construction parameter prediction model, the fusion features, construction parameter tags and category tags are spliced to train the reconstruction effect prediction model to obtain the trained reconstruction effect prediction model; S3, the fusion feature set of the well to be reconstructed is obtained, and the category tag of the well to be reconstructed is calculated by using the improved k-means algorithm; S4, the fusion features of the well to be reconstructed are disturbed by using the family of perturbation functions to obtain the disturbed fusion features, and the disturbed fusion features are input into the trained construction parameter prediction model to obtain the reconstruction parameters; S5, the reconstruction parameters, the fusion features of the well to be reconstructed and the category tag of the well to be reconstructed are spliced and input into the trained reconstruction effect prediction model to obtain the reconstruction effect prediction vector; S6, the fitness of the reconstruction effect prediction vector is calculated by using the fitness calculation function, and the optimal reconstruction parameter is selected based on the fitness.

2. The intelligent reservoir stimulation method of claim 1, wherein, The S1 is specifically: Based on the historical reservoir reconstruction database, the engineering parameters G and expert knowledge rule data K of the historical reconstruction well are constructed, and the fusion features F are calculated based on the engineering parameters G and the expert knowledge rule vector K of the historical reconstruction well: wherein Concat() is a vector concatenation function; is a weight parameter for data fusion; The construction parameter tag t=(Q, P, S) is constructed, wherein Q is the construction displacement, P is the pump injection pressure, and S is the proppant concentration; The reconstruction effect W=(E, V, B) is constructed, wherein E is the capacity improvement efficiency, V is the fracture conductivity efficiency, and B is the economic index; The category tag of the historical reconstruction well is constructed, wherein the category tag includes high-permeability well, medium-permeability well and low-permeability well.

3. The intelligent reservoir stimulation method of claim 2, wherein, The S3 specifically comprises the following steps: S3-1, the engineering parameter vector and the expert knowledge rule vector of the mth well to be transformed are constructed into a fusion feature set X={T j} of the well to be transformed. S3-2, extracting initial cluster centers U={C from the fusion feature set X of the well to be transformed by using the improved k-means algorithm v}; S3-3, initialize the improved k-means algorithm with the initial clustering centers U, and calculate the category label of the well to be transformed by using the improved k-means algorithm .

4. The intelligent reservoir stimulation method of claim 3, wherein, The S3-2 specifically comprises the following steps: S3-2-1, an initial cluster center C1 is randomly selected from the fusion feature set X of the well to be reconstructed: Wherein, random(1,...,m) represents a number randomly selected from 1,...,m; S3-2-2, the vth cluster center is selected; 1) Calculate the minimum bi-space weighted distance of each data point T j in the fusion feature set X = {T j} of the well to be transformed to all the currently selected cluster centers {C r} = (C1,...,C v-1 ) : where C r represents all the selected cluster centers; G j represents the fusion feature T of the jth well to be transformed j geological sub-vector, G j =Sg*T j , Sg represents a selection matrix; represents a center vector of the geological sub-space; represents a geological attribute weight matrix; and α and β represent center weight values; 2) Calculate the probability P(T) that all data in the fusion feature X of the well to be modified is selected as the next cluster center. j The calculation method is as follows: wherein, represents X = {T j the sum of the minimum bi-space weighted distances of all data in to all selected initial cluster centers; y = 1,..., m; 3) According to the probability P(T j ), the next cluster center is randomly selected, specifically: The cumulative distribution Y(T j ) of the probability P(T j ) is calculated. generating a random number in the interval [0, 1) ; Choose the first one that satisfies the cumulative distribution Y(T) j )≥ T j As the v-th cluster center, and set v = v + 1; S3-2-3, the extraction of the initial cluster center is completed, specifically: Repeat S3-2-2 to extract K initial cluster centers U = {C v} in turn and save as standby.

5. The intelligent reservoir stimulation method of claim 4, wherein, The S3-3 specifically comprises the following steps: S3-3-1、Set the initial clustering center U={C v} as the initial clustering center Center of the improved k-means algorithm (0) , specifically: Let the initial cluster centers U = {C v} be denoted as 6. S3-3-2, the fusion feature set X of the well to be reconstructed is clustered, specifically comprising the following steps: j the weighted distance of all cluster centers produced up to the t-1th iteration and assigned to the cluster represented by the nearest cluster center ;​ Sample T j Number of cluster to which it is assigned The calculation is as follows: wherein ; ; is an integer, denotes the sample feature T j the number of the cluster to which the sample is assigned in the t-th iteration, and ; By ① merging the features T of each well to be reformed into a set of K clusters, denoted as j wherein:​ wherein, represents a cluster numbered v, represents T j of the cluster numbered v that satisfies the set; ii. Calculate the mean of all data points contained within each cluster and record this mean as the new center of the cluster : wherein, represents the number of data points in each cluster ; represents summing the features of the data points in the cluster ; ③ Check the convergence of the k-means algorithm; When the algorithm meets any of the following convergence conditions, the algorithm converges, the iteration process is stopped, otherwise t=t+1, and return to ① to continue the iteration process: Sq1、 and wherein, is a geologic spatial convergence threshold; is a geologic subspace convergence threshold; Sq2, ; Sq3、 wherein, is the maximum iteration step. S3-3-3, output the clustering result; When the iteration process ends, the fusion feature T of each well to be transformed is given a final cluster label, i.e. a cluster label representing the category of the fusion feature T of the well to be transformed j j the category label of the category to which the well to be transformed belongs .​ 7. The intelligent reservoir reformation method of claim 5, wherein, The S4 specifically comprises the following steps: After the fusion feature set X of the jth well to be transformed is disturbed, a disturbed fusion feature set X' is obtained Specifically, wherein: are standard basis vectors corresponding to displacement, pump pressure, and proppant concentration, respectively; are perturbation amplitudes for displacement, pump pressure, and proppant concentration, respectively. the perturbed fused features are input into the trained construction parameter prediction model to obtain corresponding retrofit parameters , and the calculation method is as follows: wherein, , L = 0,..., 4; QjLrepresents the operating flow rate of the jth well to be transformed in the Lth group of perturbation scenarios, PjLrepresents the injection pressure of the jth well to be transformed in the Lth group of perturbation scenarios, CjLrepresents the proppant concentration of the jth well to be transformed in the Lth group of perturbation scenarios.

8. The intelligent reservoir reformation method of claim 6, wherein, The S5 modified effect prediction vector : where L = 0,..., 4; represents the productivity improvement performance of the Lth set of perturbation scenarios for the jth well to be stimulated, represents the fracture conductivity performance of the Lth set of perturbation scenarios for the jth well to be stimulated, represents the economic / safety performance of the Lth set of perturbation scenarios for the jth well to be stimulated.

9. The intelligent reservoir reformation method of claim 7, wherein, The S6 specifically comprises the following steps: the fitness of each modification effect prediction value is calculated using a fitness calculation function :​ wherein, The construction resource consumption representing the candidate parameter combination is calculated in the following manner: In the formula, represents the construction displacement; represents the construction time; represents the construction liquid volume; represents the proppant amount; represents the number of stages; represents the unit cost coefficient, and , represents the adjustment coefficient, ; Representative operational risk, the specific calculation method is: wherein is a fracture deviation / overreach risk function; is a casing integrity risk function; is a sand plug and sand flowback risk function; is a fluid excess induced formation damage risk function; r1, r2, r3, and r4 represent risk weighting factors; ② Based on the fitness calculated in ①, the optimal reconstruction parameter is determined; For the jth reformed well, the reformed parameter corresponding to the highest fitness is selected as the optimal reformed parameter of the jth reformed well

Images