Shale comprehensive fracturing dessert evaluation method based on movable-compressible feature fusion

Abstract

The invention belongs to the technical field of shale reservoir comprehensive fracturing dessert evaluation, and particularly discloses a shale comprehensive fracturing dessert evaluation method based on movable-compressible characteristic fusion, which is used for solving the problem that shale reservoir dessert evaluation focuses on reservoir compressibility characterization. Comprising the following steps: constructing a comprehensive fracturing dessert evaluation index reflecting potential mobility and compressibility of a reservoir and carrying out calculation; based on a self-adaptive weight determination method, mutual information correction and kernel density estimation are introduced, and calculation of evaluation index weights is achieved; based on the comprehensive fracturing dessert values in the wellbore direction, three-dimensional reservoir comprehensive fracturing dessert distribution construction is achieved by combining a sequential Gaussian simulation method; carrying out dimensionality reduction projection based on a t-SNE algorithm; and performing adaptive classification on the one-dimensional comprehensive fracturing dessert distribution based on a deep clustering algorithm, and constructing a quantitative shale reservoir classification and grading evaluation standard. According to the method, accurate quantitative evaluation of the fracturing transformation potential of the shale reservoir is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of comprehensive fracturing sweet spot evaluation for shale reservoirs, and particularly relates to a method for evaluating shale comprehensive fracturing sweet spots based on the fusion of movable and compressible characteristics. Background Art

[0002] In unconventional oil and gas reservoirs, the "sweet spot reservoir" is a local high-yield block affected by the combined action of sedimentation and tectonics, and is the key area of concern during the fracturing transformation process. Its evaluation and prediction are the core issues in the exploration and development of unconventional oil and gas. Restricted by problems such as extremely tight reservoirs, strong heterogeneity, and rapid lateral changes, traditional evaluation of the quality of heterogeneous shale reservoirs faces huge challenges.

[0003] In the initial stage of research, scholars mainly used rock mechanics parameters represented by the brittleness index to characterize the fracturability of reservoirs. Many scholars proposed various methods for characterizing the brittleness index based on reservoir geological conditions such as sedimentary background and tectonic changes, including those based on elastic mechanics parameters and rock mineral composition. Considering the strong heterogeneity and large physical property differences of unconventional reservoirs, single-factor evaluation methods are no longer applicable. Some existing patents propose using multi-factor evaluation methods to evaluate the sweet spots of shale reservoirs. The Chinese invention patent with the publication number CN111188612A proposes a method for identifying sweet spots of shale oil based on the fusion of multiple logging parameters. According to the standardized sensitive logging curves, a comprehensive evaluation index curve of shale oil is obtained. However, this method is only based on logging curve data and does not perform secondary interpretation, and the calculated parameters lack physical meaning. The Chinese invention patent with the publication number CN113090258A proposes to calculate the "geology-engineering" comprehensive evaluation index of the horizontal section of shale gas wells based on logging data to divide the fracturability grades, but does not consider the mobility index of shale oil. The Chinese invention patent with the publication number CN117474368A selects sweet spot evaluation parameters through entropy value analysis, realizes the fusion of multiple evaluation parameters by introducing an adaptive weighted fusion method, and constructs a method for characterizing double sweet spots of shale oil. However, this method still does not clearly introduce the evaluation dimension of the mobility of shale reservoirs.

[0004] In summary, the current methods for evaluating sweet spots in shale reservoirs mainly have the following problems: lack of accurate description of the heterogeneity of the reservoirs drilled by horizontal wells, and the relevant research mainly focuses on the characterization of reservoir compressibility, with less consideration of the shale occurrence mobility that characterizes properties such as shale oil content, shale reservoir property, and movable hydrocarbon content. Shale mobility refers to all geological elements and actions that are conducive to the exploitation of free fluids in shale formations, which is a combination of dense shale formations and the fluid properties within the formations, and can characterize the development degree of micro-nano pores in shale reservoirs, the enrichment of organic matter, and the occurrence state of oil and gas, etc., which also need to be emphasized in the evaluation of sweet spots in shale reservoirs. Summary of the Invention

[0005] The object of the present invention is to provide a comprehensive fracturing sweet spot evaluation method for shale based on the fusion of movable-compressible characteristics, effectively solving the problem that the evaluation of sweet spots in shale reservoirs mainly focuses on the characterization of reservoir compressibility, while less consideration is given to the movability of shale reservoirs.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a comprehensive fracturing sweet spot evaluation method for shale based on the fusion of movable-compressible characteristics, including the following steps: S1. Based on the logging interpretation results of a single well reservoir of shale oil, construct a comprehensive fracturing sweet spot evaluation index reflecting the potential movability and compressibility of the reservoir and carry out calculations; S2. Based on the constructed method for determining adaptive weights based on information entropy, introduce mutual information correction and kernel density estimation to calculate the weights of the evaluation indexes, thereby determining the distribution of comprehensive fracturing sweet spots along the wellbore; S3. Based on the numerical values of comprehensive fracturing sweet spots along the wellbore direction of a shale oil horizontal well, combine with the sequential Gaussian simulation method to construct the three-dimensional reservoir comprehensive fracturing sweet spot distribution; S4. Based on the t-SNE algorithm, obtain the dimensionality reduction projection of the three-dimensional comprehensive fracturing sweet spot distribution, calculate the distribution of the longitudinal dominant area of the sweet spot, and realize the three-dimensional characterization of the sand body geometry and the spatial distribution of the "sweet spot reservoir" under complex geological conditions of the shale reservoir; S5. Based on the deep clustering algorithm, perform adaptive classification on the one-dimensional comprehensive fracturing sweet spot distribution after projection, and construct a quantitative classification and grading evaluation standard for shale reservoirs.

[0007] Further, in step S1, the positive factors in the evaluation index of reservoir potential movability are organic carbon content, oil saturation, organic matter maturity, porosity, permeability, free hydrocarbons, and saturated adsorption coefficient; the positive factors in the evaluation index of reservoir potential compressibility are brittleness index and Young's modulus; the reverse factors in the evaluation index of reservoir potential compressibility are Poisson's ratio, fracture toughness, horizontal maximum / minimum in-situ stress, and stress difference; no treatment is performed on the positive factors, and the reciprocal of the reverse factors is taken.

[0008] Further, in step S2, the continuous entropy of the evaluation index based on kernel density estimation:

[0009] H(x) = -∫p(x)logp(x)dx;

[0010] where H(x) is the continuous entropy of index x; p(x) is the probability density function of index x.

[0011]

[0012] where n' is the number of samples for comprehensive fracturing sweet spot evaluation; K is the kernel function, and the Gaussian kernel function is used; x I is the I-th sample point; h is the bandwidth parameter, and the bandwidth parameter uses the Silverman criterion to determine the optimal bandwidth of kernel density estimation.

[0013] h = 0.9AN-1 / 5 ;

[0014] Among them, A = min(σ, IQR / 1.34), where σ is the sample standard deviation, IQR is the interquartile range, and N is the sample size.

[0015]

[0016] Among them, MI(x, y) is the mutual information between indicators x and y; p(x, y) is the joint probability density function between indicators x and y; p(y) is the marginal probability density function of indicator y.

[0017] Based on continuous entropy and mutual information, construct a weight calculation model for evaluation indicators:

[0018]

[0019] Among them, W i is the weight of the i-th comprehensive fracturing sweet spot evaluation indicator; x j is the weight of the j-th comprehensive fracturing sweet spot evaluation indicator; H i is the continuous entropy of the i-th comprehensive fracturing sweet spot evaluation indicator; H j is the continuous entropy of the j-th comprehensive fracturing sweet spot evaluation indicator; N' is the total number of evaluation indicators; MI(x i , x j ) is the mutual information between evaluation indicators x i and x j ; α is the information redundancy penalty coefficient.

[0020]

[0021] Among them, α opt is the optimal information redundancy penalty coefficient obtained through the optimization process; argmin α is to find the α that minimizes the objective function value within the value range of parameter α; M is the number of cross-validation folds; L(D m , α) is the validation loss on the m-th fold of data; D m represents the m-th fold of the validation data set.

[0022] Furthermore, in step S2, to ensure that the numerical range of each evaluation indicator is between 0 and 1, normalize each factor and perform a linear transformation on the original data. The transformation function is as follows:

[0023]

[0024] Among them, Y i′j′ is the normalized value of the j'-th indicator of the i'-th sample; X i′j′ is the value of the j'-th indicator of the i'-th sample in the original data; Xj′ Denote the set of the \(j'\)-th index among all samples; \(min(X j′ )\) and \(max(X j′ )\) are the minimum and maximum values of the \(j'\)-th index respectively.

[0025] The comprehensive fracturing sweet spot expression is:

[0026]

[0027] where \(CFS\) is the comprehensive fracturing sweet spot; \(r\) is the positive factor index; \(t\) is the negative factor index; \(P\) is the positive factor; \(N\) is the negative factor; \(r'\) is the number of positive factors; \(t'\) is the number of negative factors; \(W r is the positive factor weight, \(W t is the negative factor weight; \(P r is the original value of the \(r\)-th positive evaluation index, \(N t is the original value of the \(t\)-th negative evaluation index; \(P min is the minimum value of the \(r\)-th positive evaluation index, \(P max is the maximum value of the \(r\)-th positive evaluation index, \(N mon is the minimum value of the \(t\)-th negative evaluation index, \(N max is the maximum value of the \(t\)-th negative evaluation index.

[0028] Furthermore, in step S3, with the known sweet spot value at the wellbore position as the hard data constraint, through spatial correlation analysis and conditional probability estimation, predict the sweet spot value at the un-drilled position within the study area, so as to obtain the complete three-dimensional sweet spot distribution.

[0029] First, perform spatial structuring on the sweet spot values along the wellbore of each well, expressed as a discrete point set \(CFS(x k ,y k ,z k )\) in the spatial coordinate system, where \(CFS(x k ,y k ,z k )\) is the reservoir comprehensive fracturing sweet spot value at the wellbore position \((x k ,y k ,z k ).

[0030] Due to the heterogeneity characteristics of the sweet spot distribution in shale reservoirs, establish a spatial variogram considering anisotropy:

[0031]

[0032] where \(\gamma\) is the variogram; \(h vec is the spatial position vector; \(A\) is the anisotropy matrix, used to describe the correlation difference between horizontal and vertical directions; \(C oNugget effect value; C is the sill value; a is the range.

[0033] Construct a three-dimensional sweet spot distribution based on multi-well constraints. First, define the multi-well constraint set D = {(x k , y k , z k , CFS k ) | k = 1,..., nn}, where nn is the total number of wellbore sampling points, and CFS k is the comprehensive reservoir fracturing sweet spot value at the wellbore position (x k , y k , z k ).

[0034] For any point x 00 to be estimated in the grid system: x 00 = (x 0 , y 0 , z 0 ), the conditional probability distribution of the sweet spot value is:

[0035]

[0036] where P represents the conditional probability, CFS 0 represents the comprehensive reservoir fracturing sweet spot value at the point x 00 to be estimated, μ 0 represents the conditional expectation at the point x 00 to be estimated, represents the variance at the point x 00 to be estimated; Normal represents the normal distribution.

[0037]

[0038] where γ is the variogram value; λ kk is the Kriging weight considering multi-well constraints, obtained by solving the Kriging equations; kk is the wellbore sampling point index; mm is the total number of wells; CFS k represents the comprehensive reservoir fracturing sweet spot value at the point x kk to be estimated.

[0039] Kriging equations:

[0040]

[0041] where λ jj and λ kk are both Kriging weights considering multi-well constraints, jj = 1,..., mm, x kk and x jj are the three-dimensional coordinates of the known wellbore positions, and μ is the Lagrange multiplier.

[0042] Further, in step S3, the process of constructing the comprehensive fracturing sweet spot distribution of the three-dimensional reservoir by using the sequential Gaussian simulation method includes: First, establish a three-dimensional grid system covering the research area, and reasonably determine the grid resolution according to the research requirements and computing resources; input the known sweet spot values at the wellbore positions as hard data into the system, and generate a random access path considering the geological trend; for each grid point to be simulated in the path, first collect conditional data within the specified search range of the grid point to be simulated, and the conditional data includes the hard data points of the wellbore and the data of the grid points that have completed simulation; based on the collected conditional data, obtain the conditional probability distribution by solving the Kriging equations, and perform random sampling from the conditional probability distribution to obtain the simulated value; immediately after completing the simulation of each grid point, add the simulated value of the grid point as new conditional data to the dataset for the simulation of subsequent grid points.

[0043] Further, in step S4, the t-SNE algorithm realizes dimensionality reduction by minimizing the difference between data points in the high-dimensional space and the low-dimensional space, and maintains the local relationship between data points during the dimensionality reduction process; for each grid point in the three-dimensional space, define its high-dimensional feature vector as a i =(X ai ,Y ai ,Z ai ,CFS ai ), where (X ai ,Y ai ,Z ai ) are the three-dimensional position coordinates, and CFS ai is the comprehensive fracturing sweet spot value at the position (X ai ,Y ai ,Z ai ).

[0044] Therefore, for the similarity between any two high-dimensional feature vectors a i and a j containing three-dimensional position and comprehensive fracturing sweet spot information, the Gaussian kernel function A aiaj is used for calculation:

[0045] A aiaj =(A(aj|ai)+A(ai|aj)) / (2nn′);

[0046] The conditional probabilities A(ai|aj) and A(aj|ai) are respectively:

[0047]

[0048] where, σ t is the bandwidth parameter of the Gaussian kernel of the t-SNE algorithm, nn′ is the total number of grid points in the research area, and a k is the position (Xak , Y ak , Z ak The high-dimensional feature vectors corresponding to the grid points of (), ||a i -a j || and ||a j -a i || are both the Euclidean distances between two grid points in the high-dimensional space.

[0049] In the one-dimensional projection space, the similarity between two three-dimensional reservoir comprehensive fracturing sweet spots is defined by the t-distribution:

[0050] q aiaj = (1 + ‖‖y ai -y aj || 2 ) -1 / ∑ ak≠al (1 + ||y ak -y al || 2 ) -1 ;

[0051] where y ai and y aj are the projection coordinates of a i and a j in the one-dimensional space respectively, y ak and y al are any two projection coordinates different from y ai and y aj in the study area, ||y ak -y al || is the Euclidean distance between two projection coordinates in the one-dimensional space, and the denominator summation term traverses all grid point pairs where ak ≠ al in the study area.

[0052] Furthermore, in step S5, a network structure combining a deep autoencoder and a clustering layer is adopted to achieve end-to-end learning of the shale reservoir feature extraction and clustering method; the network structure includes a deep autoencoder and a clustering layer, where the deep autoencoder network consists of two parts: an encoder f(B) and a decoder g(z), and its structure is as follows:

[0053]

[0054] where B is the input reservoir comprehensive fracturing sweet spot data, z is the feature representation output by the encoder, σ e and σ d are the activation functions of the encoder and the decoder respectively, W e is the encoder weight matrix, W d is the decoder weight matrix, b e and b dBias vectors for the encoder and decoder respectively.

[0055] The clustering layer maps the encoded feature representation Z to the cluster centers, and constructs a joint optimization objective LL by combining the reconstruction loss and the clustering loss:

[0056] LL = Lr + βLc;

[0057] Where β is the balance parameter, and the reconstruction loss Lr is used to ensure the effectiveness of feature extraction:

[0058] Lr = ||B - g(f(B))|| 2 ;

[0059] The clustering loss Lc adopts an auxiliary target distribution based on the student t-distribution:

[0060] Lc = JD KL (P||Q);

[0061] Where J is the preset number of clustering categories, P is the auxiliary target distribution, Q is the predicted soft assignment probability, and D KL (P||Q) represents the KL divergence, which is used to measure the difference between distribution P and distribution Q.

[0062] The auxiliary target distribution P is calculated by the following formula:

[0063]

[0064] Where the soft assignment probability q mo represents the probability that the m-th sample belongs to the o-th cluster, z m is the encoded feature representation of the m-th sample, μ o is the o-th cluster center, v is the degree of freedom of the t-distribution, f o is the normalization factor of the o-th cluster, and p mo is the target probability that the m-th sample belongs to the o-th cluster.

[0065] Furthermore, in step S5, based on the deep clustering calculation results, further combined with reservoir engineering practice experience, a classification and grading evaluation standard for shale reservoirs is established; among them, the sweet spot value range of the first-class reservoir is located in [V 1 , 1], V 1 represents the lower limit of the first-class reservoir. The first-class reservoir has the best fracturing conditions and needs to be preferentially fractured; the sweet spot value range of the second-class reservoir is located in [V 2 , V 1 ), V 2 represents the lower limit of the developable reservoir. The second-class reservoir has good fracturing conditions and is suitable for conventional fracturing; the sweet spot value range of the third-class reservoir is [0, V 2), the fracturing and reconstruction conditions of three types of reservoirs are poor and the reconstruction is not considered for the time being. Among them, the demarcation values V 1 and V 2 are obtained based on the depth clustering results first, and then fine-tuned and optimized in combination with engineering practice experience.

[0066] Compared with the prior art, the beneficial technical effects of the present invention are as follows: (1) The present invention establishes a comprehensive fracturing sweet spot evaluation method for shale that comprehensively considers the shale reservoir mobility evaluation index and the fracturability evaluation index. Combining with the self-adaptive weight determination method based on information entropy, it realizes the accurate quantitative evaluation of the fracturing and reconstruction potential of shale reservoirs. Through sequential Gaussian simulation and t-SNE dimensionality reduction algorithm, an effective mapping from the wellbore sweet spot distribution to the one-dimensional characterization of the large-scale longitudinal reservoir sweet spot is realized, providing a more comprehensive theoretical basis for the fracturing and reconstruction design.

[0067] (2) The reservoir classification and grading evaluation method based on depth clustering established by the present invention extracts reservoir features through a deep autoencoder and realizes the self-adaptive classification of shale reservoirs in combination with a clustering algorithm. This method breaks through the subjective limitations of the traditional manual experience in dividing reservoir categories, can automatically identify the intrinsic characteristics of different types of reservoirs, and realizes the accurate grading of reservoir quality, providing reliable decision-making support for the optimization design and implementation of fracturing and reconstruction plans. Description of the Drawings

[0068] Figure 1 shows the distribution of the comprehensive fracturing sweet spots along the wellbore of a certain shale oil horizontal well in Example 1.

[0069] Figure 2 is a calculation example of the three-dimensional comprehensive fracturing sweet spot of a reservoir in Example 1.

[0070] Figure 3 is a calculation example of data dimensionality reduction for the three-dimensional comprehensive fracturing sweet spot in Example 1. Detailed Embodiments

[0071] Example 1: Taking the shale oil in the X research area of Oilfield A as an example, the shale oil in the Chang 7 section is located in the Mesozoic Triassic of Basin B and is mainly controlled by the semi-deep lake - deep lake gravity flow sedimentary system. The sweet spot of the shale oil in the Chang 7 section is the thin sandstone interlayer formed by gravity flow in the thick mud shale series. The reservoir has strong lateral heterogeneity, poor continuity, and variable oil-bearing properties. Coupled with the influence of the low-pressure characteristics of the basin, it shows problems such as low recovery rate, rapid decline of single wells, low effective recoverable degree, and high development cost. In production, it shows the characteristics of large differences in single-well productivity and low production and low efficiency of some wells. The single-well productivity of this example is improved and the development cost is reduced through the shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable - fracturable characteristics provided by the present invention.

[0072] Shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible characteristics, comprising the following steps: S1. Based on the logging interpretation results of the single-well reservoir of shale oil, construct a comprehensive fracturing sweet spot evaluation index reflecting the potential movability and compressibility of the reservoir and carry out calculations.

[0073] Taking movability as an example, seven indexes including organic carbon content, oil saturation, organic matter maturity, porosity, permeability, free hydrocarbons and saturated adsorption coefficient are selected from three aspects of shale oiliness, shale reservoir property and movable hydrocarbon content for evaluation; the evaluation indexes for compressibility are selected as Young's modulus, Poisson's ratio, brittleness index, fracture toughness, maximum / minimum horizontal in-situ stress and stress difference for evaluation. For the positive factors in the evaluation indexes, no treatment is done, and for the negative factors, their reciprocals are taken. The detailed evaluation indexes are shown in Table 1.

[0074] Table 1 Comprehensive fracturing sweet spot evaluation indexes for horizontal well reservoir

[0075]

[0076] (1) Calculation of reservoir compressibility evaluation indexes.

[0077] ① Brittleness index: The calculation of the brittleness index requires calculating dynamic parameters based on logging data and converting the dynamic Young's modulus and Poisson's ratio into static Young's modulus and Poisson's ratio. The calculation formulas for dynamic Poisson's ratio and dynamic Young's modulus are as follows:

[0078]

[0079] Among them, σ d is the dynamic Poisson's ratio; E d is the dynamic Young's modulus; Δt s is the shear wave travel time; Δt p is the compressional wave travel time; ρ b is the rock density.

[0080] The conversion formulas for dynamic and static parameters are as follows:

[0081] σ s = 0.16σ d + 0.23;

[0082] E s = 0.75E d + 2.12;

[0083] Among them, σ s is the static Poisson's ratio; E s is the static Young's modulus.

[0084] After calculating the required static parameters, the brittleness index is calculated using the following formula:

[0085]

[0086] Among them, BI is the brittleness index; σ smax is the maximum static Poisson's ratio, and σ smin is the minimum static Poisson's ratio; E smax is the maximum static Young's modulus, and E smin is the minimum static Young's modulus.

[0087] ② Fracture toughness: Calculating the fracture toughness requires calculating the compressive and tensile strengths based on the shale content interpreted from well logging. The calculation method is as follows:

[0088] K IC = 0.271 + 0.107S t ;

[0089] σ c = (0.0045 + 0.0035V cl )E d ;

[0090]

[0091] I γ = (γ - γ min ) / (γ max - γ min );

[0092] S t = σ c / K;

[0093] Among them, K IC is the fracture toughness; S t is the tensile strength; V cl is the shale content; G cur is the Hilchie index; I γ is the shale content index; γ is the natural gamma log value at the current measurement point; γ min is the gamma value of the pure sandstone standard layer; γ max is the gamma value of the pure shale standard layer; σ c is the uniaxial compressive strength of the rock; K is 12.26.

[0094] ③ In-situ stress distribution: The characteristics of in-situ stress distribution have an important impact on the formation of fracture networks in shale formations. The influence degree of the horizontal principal stress difference on the shale fracture morphology can be characterized by the horizontal stress difference coefficient K h :

[0095]

[0096] Among them, σ H is the maximum horizontal principal stress, and σ h is the minimum horizontal principal stress. σH and σ h can be obtained from differential strain experiments. For well logging prediction, Huang's model is adopted:

[0097]

[0098] where p p is the formation pore pressure; ξ h and ξ H are both formation tectonic stress coefficients; σ v is the overburden pressure; ρ b is the density logging value; α is the predicted value of the formation pore pressure gradient; H S is the burial depth of the target interval, that is, the vertical distance from the surface to the bottom boundary of the target interval; h S is the current calculation depth, representing any depth position from the surface to the target interval during the integration process.

[0099] (2) Calculation of reservoir mobility evaluation index.

[0100] ① Conventional reservoir physical property parameters: The distribution data of shale reservoir permeability, porosity and oil saturation results are obtained through secondary well logging interpretation data.

[0101] ② Organic carbon content: The ΔTOC model is used to evaluate the organic carbon content. The calculation formula is:

[0102] Δlog R = log R + log(R max / R min ) / (Δt max -Δt min ) × (Δt - Δt max ) - logR min ;

[0103] TOC = ΔlgR × 10 (2.297-0.1688LOM) +ΔTOC;

[0104] where Δt is the result of acoustic travel time; R is the true resistivity of the formation; R min and R max are the minimum and maximum values of the resistivity curve scale when the acoustic travel time and resistivity curves are superimposed respectively; Δt min and Δt max are the minimum and maximum values of the acoustic travel time curve scale when the acoustic travel time and resistivity curves are superimposed respectively; LOM is the organic matter maturity; ΔTOC is the background value of the organic carbon content.

[0105] ③ Organic matter maturity: Vitrinite reflectance is an important index reflecting the organic matter maturity. The deeper the organic matter thermal metamorphism, the greater the vitrinite reflectance. The ΔRRS method is used to evaluate the organic matter maturity. The calculation formula is as follows:

[0106] ΔRRS = DT Cum -RR Cum ;

[0107] Wherein, ΔRRS is the separation amount of the cumulative acoustic wave value and the cumulative resistivity ratio. The smaller ΔRRS is, the lower the maturity of the source rock is characterized; DT Cum is the cumulative frequency of the acoustic time difference with depth; RR Cum is the cumulative frequency of the resistivity ratio with depth, wherein, DT Cum and RR Cum take opposite directions.

[0108] The organic matter maturity R o can be calculated from ΔRRS as shown in the following formula:

[0109] R o = 0.5615e (0.7143GG-1.1593)×ΔRRS ;

[0110] Wherein, GG is the geothermal gradient.

[0111] ④ Free hydrocarbons: The free hydrocarbons are calculated by the statistical regression method. Using the good linear regression relationship between the free hydrocarbons and TOC, the quantitative evaluation formula of the free hydrocarbons S 1 can be obtained:

[0112] S 1 = 0.6614 × w(TOC) + 1.5269.

[0113] ⑤ Saturated adsorption coefficient S 1 / TOC: This index can be used to characterize the adsorption capacity of the shale. Adsorptivity is also the basic standard for evaluating whether the shale reservoir has mobility. Only when S 1 / TOC is greater than the adsorption critical value, the oil and gas in the shale formation are mobile, and numerically it is equal to the ratio of the calculated S 1 to the TOC content.

[0114] S2. Based on the constructed adaptive weight determination method based on information entropy, mutual information correction and kernel density estimation are introduced to achieve the accurate calculation of the evaluation index weights, so as to determine the distribution of the comprehensive fracturing sweet spots along the wellbore.

[0115] Since the influence degrees of the selected evaluation indicators on the comprehensive fracturing sweet spots are different, establishing a reasonable mathematical evaluation model to quantify the influence degrees of each evaluation indicator is the key to realizing the quantitative evaluation of comprehensive fracturing sweet spots. The comprehensive fracturing sweet spot index analysis method is carried out based on each interpreted fracturing sweet spot evaluation indicator. To ensure that the values of each fracturing sweet spot index are in the range of 0 to 1, it is necessary to normalize each physical property parameter respectively, so as to eliminate the adverse effects caused by singular sample data. The min-max standardization method is used for normalization, and the original data is linearly transformed. The transformation function is as follows:

[0116]

[0117] where Y i′j′ is the normalized value of the j'-th index of the i'-th sample; X i′j′ is the value of the j'-th index of the i'-th sample in the original data; X j′ represents the set of the j'-th index in all samples; min(X j′ ) and max(X j′ ) are the minimum and maximum values of the j'-th index respectively.

[0118] The finally calculated expression of the comprehensive fracturing sweet spot is:

[0119]

[0120] where CFS is the comprehensive fracturing sweet spot; r is the positive factor index; t is the negative factor index; P is the positive factor; N is the negative factor; r' is the number of positive factors; t' is the number of negative factors; W r is the positive factor weight, W t is the negative factor weight; P r is the original value of the r-th positive evaluation index, N t is the original value of the t-th negative evaluation index; P min is the minimum value of the r-th positive evaluation index, P max is the maximum value of the r-th positive evaluation index, N min is the minimum value of the t-th negative evaluation index, N max is the maximum value of the t-th negative evaluation index.

[0121] To determine the weight coefficients of each evaluation indicator, an adaptive weight determination method based on information entropy is introduced. This method is based on the information entropy theory and realizes the accurate calculation of the index weights by introducing mutual information correction and kernel density estimation.

[0122] (1) Continuous entropy of the evaluation indicator based on kernel density estimation:

[0123] H(x) = -∫p(x) log p(x) dx;

[0124] Among them, H(x) is the continuous entropy of the index x; p(x) is the probability density function of the index x. The larger the continuous entropy value, the higher the uncertainty of the evaluation index and the less effective information it provides.

[0125]

[0126] Among them, n′ is the number of samples for comprehensive fracturing sweet spot evaluation; K is the kernel function, and the Gaussian kernel function is used; x I is the I-th sample point; h is the bandwidth parameter, and the Silverman criterion is used to determine the optimal bandwidth for kernel density estimation.

[0127] h = 0.9AN -1 / 5 ;

[0128] Among them, A = min(σ, IQR / 1.34), σ is the sample standard deviation, IQR is the interquartile range, and N is the number of samples.

[0129] (2) Mutual information correction: Considering that there may be information redundancy among shale reservoir evaluation indicators, mutual information is introduced to correct the weights. The mutual information between two indicators is defined as:

[0130]

[0131] Among them, MI(x, y) is the mutual information between indicators x and y; p(x, y) is the joint probability density function between indicators x and y; p(y) is the marginal probability density function of indicator y. The larger the mutual information, the higher the degree of information overlap between the two indicators.

[0132] (3) Weight calculation: Based on continuous entropy and mutual information, a weight calculation model for evaluation indicators is constructed:

[0133]

[0134] Among them, W i is the weight of the i-th comprehensive fracturing sweet spot evaluation indicator; x j is the weight of the j-th comprehensive fracturing sweet spot evaluation indicator; H i is the continuous entropy of the i-th comprehensive fracturing sweet spot evaluation indicator; H j is the continuous entropy of the j-th comprehensive fracturing sweet spot evaluation indicator; N′ is the total number of evaluation indicators; MI(x i , x j ) is the evaluation indicator x i and x jThe mutual information between them; α is the information redundancy penalty coefficient, which is used to adjust the intensity of mutual information correction. Its value is determined by cross - validation to enable the model to have good generalization ability. The calculation method is as follows:

[0135]

[0136] Among them, α opt is the optimal information redundancy penalty coefficient obtained through the optimization process; argmin α is to find the α that minimizes the objective function value within the value range of the parameter α; M is the number of cross - validation folds; L(D m ,α) is the validation loss on the m - th fold of data; B m represents the m - th validation data set.

[0137] Among them, the numerical calculations of the continuous entropy H(x) and the mutual information MI(x,y) are carried out using the Monte Carlo method:

[0138]

[0139] Among them, M c is the number of Monte Carlo sampling points, is the m - th c sampling point pair, and p is the probability density function.

[0140] After the above steps, the weights of each evaluation index can be obtained. Substituting them into the comprehensive fracturing sweet spot expression can realize the quantitative evaluation of the comprehensive fracturing sweet spot of the shale reservoir.

[0141] The weights of each evaluation index calculated in this embodiment are shown in Table 2. Among them, the normalized weights of porosity, permeability (mobility index), and brittleness index (compressibility index) are relatively large, and they have a greater impact on the comprehensive sweet spot evaluation.

[0142] Table 2 Weight distribution of comprehensive fracturing sweet spot evaluation indexes

[0143]

[0144] Through the determined weights of each evaluation index, the distribution of the comprehensive fracturing sweet spot along the wellbore of a horizontal shale oil well in this embodiment can be calculated as Figure 1 shown.

[0145] S3. Based on the numerical values of the comprehensive fracturing sweet spot along the wellbore direction of the horizontal shale oil well, combined with the sequential Gaussian simulation method, the construction of the three - dimensional reservoir comprehensive fracturing sweet spot distribution is realized.

[0146] After calculating the weights of each evaluation index, the comprehensive fracturing sweet spot values of horizontal wells in shale reservoirs in the study area along the wellbore direction can be obtained. However, the sweet spot distribution of a single well can only reflect the reservoir characteristics near the wellbore and is difficult to comprehensively characterize the spatial distribution law of sweet spot reservoirs in the study area. To realize the conversion from discrete wellbore sweet spots to continuous regional sweet spot distributions, the sequential Gaussian simulation method is used to construct the three-dimensional reservoir comprehensive fracturing sweet spot distribution. This method takes the comprehensive fracturing sweet spot values of reservoirs along the wellbores of multiple wells as constraint conditions and estimates the sweet spot distribution of the entire study area by considering spatial correlation and randomness. The core idea is: using the known sweet spot values at the wellbore positions as hard data constraints, predicting the sweet spot values at the positions of undrilled wells in the study area through spatial correlation analysis and conditional probability estimation, so as to obtain a complete three-dimensional sweet spot distribution.

[0147] First, spatially structure the sweet spot values along the wellbores of each well, expressed as a discrete point set CFS(x k ,y k ,z k ) in the spatial coordinate system, where CFS(x k ,y k ,z k ) is the comprehensive fracturing sweet spot value of the reservoir at the wellbore position (x k ,y k ,z k ).

[0148] Due to the heterogeneity characteristics of the sweet spot distribution in shale reservoirs, establish a spatial variogram considering anisotropy:

[0149]

[0150] where γ is the variogram; h vec is the spatial position vector; A is the anisotropy matrix used to describe the correlation difference between the horizontal and vertical directions; C o is the nugget effect value, generally taking 0.1 - 0.2 of the total variance of the data in actual application; C is the sill value, usually equal to the total variance of the data set; a is the range, generally taking 1 / 3 - 1 / 2 of the well spacing in the horizontal direction and 1 / 4 - 1 / 3 of the target layer thickness in the vertical direction.

[0151] After constructing the spatial variogram considering anisotropy, start constructing the three-dimensional sweet spot distribution based on multi-well constraints. First, define the multi-well constraint set D = {(x k ,y k ,z k ,CFS k )|k = 1,...,nn}, where nn is the total number of wellbore sampling points, and CFS k is the comprehensive fracturing sweet spot value of the reservoir at the wellbore position (x k ,yk , z k The integrated reservoir fracturing sweet spot value at

[0152] For any point x to be estimated in the grid system 00 : x 00 =(x 0 , y 0 , z 0 ), the conditional probability distribution of the sweet spot value is as follows:

[0153]

[0154] where P represents the conditional probability, CFS 0 represents the integrated reservoir fracturing sweet spot value at the point x to be estimated 00 , μ 0 represents the conditional expectation at the point x to be estimated 00 , represents the variance at the point x to be estimated 00 ; Normal represents the normal distribution

[0155]

[0156] where γ is the variogram value, λ kk is the Kriging weight considering multi-well constraints, obtained by solving the Kriging equations; kk is the wellbore sampling point index, mm is the total number of wellbores, CFS k represents the integrated reservoir fracturing sweet spot value at the point x to be estimated kk

[0157] The Kriging equations:

[0158]

[0159] where λ jj and λ kk are both Kriging weights considering multi-well constraints, jj = 1,..., mm, x kk and x jj are the three-dimensional coordinates of the known wellbore positions, and μ is the Lagrange multiplier

[0160] ​Based on the above estimation principle, the process of constructing a three-dimensional reservoir comprehensive fracturing sweet spot distribution using the sequential Gaussian simulation method includes: first, establishing a three-dimensional grid system covering the study area, and the grid resolution is reasonably determined according to research needs and computing resources; the known sweet spot value at the wellbore position is input into the system as hard data, and a random access path considering the geological trend is generated; for each grid point to be simulated in the path, first, conditional data is collected within the specified search range of the grid point to be simulated, and the conditional data includes the hard data points of the wellbore and the grid point data of the completed simulation; based on the collected conditional data, the conditional probability distribution is obtained by solving the Kriging equation group, and random sampling is performed from the conditional probability distribution to obtain simulation values; after each simulation of a grid point is completed, the simulated value of the grid point is immediately added to the data set as new conditional data for the simulation of subsequent grid points.

[0161] Based on the distribution of comprehensive fracturing sweet spots along the wellbore calculated in step S2, this embodiment uses the sequential Gaussian simulation method to generate the three-dimensional spatial distribution of comprehensive fracturing sweet spots, thereby constructing a three-dimensional comprehensive sweet spot evaluation model applicable to the reservoir in the entire study area. Figure 2 As shown in the figure, the evaluation results of the comprehensive fracturing sweet spot along the wellbore of a horizontal well Well X and three surrounding horizontal wells (Well 1, Well 2, Well 3) are used to discretize the comprehensive fracturing sweet spot along the wellbore into a three-dimensional distribution through the sequential Gaussian simulation method. Similar to the two-dimensional reservoir comprehensive fracturing sweet spot along the wellbore, the three-dimensional comprehensive fracturing sweet spot of the reservoir is also a dimensionless evaluation index ranging from 0 to 1. The closer the value is to 1, the larger the three-dimensional comprehensive fracturing sweet spot is and the higher the reservoir quality is.

[0162] S4. Based on the t-SNE algorithm, the dimensionality reduction projection of the three-dimensional comprehensive fracturing sweet spot distribution is obtained, and the vertical distribution of the sweet spot is calculated to achieve a three-dimensional depiction of the geometric shape of the sand body and the spatial distribution of the "sweet spot reservoir" under complex geological conditions of the shale reservoir.

[0163] Considering the difficulties in visualization and interpretation of the three-dimensional comprehensive fracturing sweet spot distribution in practical applications, a dimensionality reduction projection method based on the t-SNE algorithm is proposed to transform the three-dimensional sweet spot distribution into a one-dimensional representation that is easier to use in practical applications. This method achieves an effective mapping from three dimensions to one dimension while maintaining the data structure, so that the one-dimensional sweet spot distribution after projection not only contains information along the wellbore, but also integrates the regional longitudinal spatial distribution characteristics.

[0164] The t-SNE algorithm achieves dimensionality reduction by minimizing the differences between data points in high-dimensional space and low-dimensional space. The core idea is to maintain the local relationship between data points as much as possible during the dimensionality reduction process.

[0165] For each grid point in three-dimensional space, its high-dimensional feature vector is defined as a i =(X ai , Y ai , Z ai , CFS ai ), where (X ai , Y ai , Z ai ) are three-dimensional position coordinates, and CFS ai is the comprehensive fracturing sweet spot value at the position (X ai , Y ai , Z ai ).

[0166] Therefore, for the similarity between any two high-dimensional feature vectors a i and a j that contain three-dimensional position and comprehensive fracturing sweet spot information, the Gaussian kernel function A aiaj is used for calculation:

[0167] A aiaj =(A(aj|ai)+A(ai|aj)) / (2nn');

[0168] The conditional probabilities A(ai|aj) and A(aj|ai) are respectively:

[0169]

[0170] where σ t is the bandwidth parameter of the Gaussian kernel of the t-SNE algorithm, nn' is the total number of grid points in the study area, a k is the high-dimensional feature vector corresponding to the grid point at the position (X ak , Y ak , Z ak ), and ||a i - a j || and ||a j - a i || are both the Euclidean distances between two grid points in the high-dimensional space.

[0171] In the one-dimensional projection space, the similarity between the three-dimensional reservoir comprehensive fracturing sweet spots of two points is defined by the t-distribution:

[0172] q aiaj =(1 + ||y ai - y aj || 2 ) -1 / ∑ ak≠al (1 + ||y ak - y al || 2 ) -1 ;

[0173] where y ai and y aj are the projection coordinates of a i and a j in one-dimensional space respectively, y ak and y al are any two projection coordinates different from y ai and y aj in the study area, ‖|y ak - y al |‖ is the Euclidean distance between the two projection coordinates in one-dimensional space, and the denominator summation term traverses all grid point pairs of ak≠al in the study area.

[0174] The one-dimensional sweet spot distribution obtained by projection not only contains the sweet spot variation characteristics along the wellbore, but also integrates the regional spatial distribution information, which is convenient for engineering applications and decision-making analysis. The projection result maintains the main structural characteristics of the original data and provides effective technical support for the engineering application of sweet spot evaluation in shale reservoirs.

[0175] In this embodiment, it is considered that there are difficulties in directly using the three-dimensional comprehensive fracturing sweet spot to carry out fracturing design in the actual application process. Therefore, in the actual application process, it is necessary to reduce the dimension of the data of the three-dimensional comprehensive fracturing sweet spot to obtain the projection distribution of the longitudinal advantageous area, and the calculation example is Figure 3 as shown. The projection of the three-dimensional comprehensive fracturing sweet spot obtained by using the t-SNE algorithm in this embodiment can fully reflect the longitudinal arrangement law of the comprehensive fracturing sweet spot, and the one-dimensional curve form can more intuitively present the specific values of the reservoir comprehensive fracturing sweet spot.

[0176] S5. Based on the deep clustering algorithm, adaptively classify the projected one-dimensional comprehensive fracturing sweet spot distribution, and construct a quantitative classification and grading evaluation standard for shale reservoirs.

[0177] After completing the evaluation of the reservoir comprehensive fracturing sweet spot, in order to further clarify the quality characteristics of different types of shale reservoirs and realize the accurate classification and grading evaluation of the reservoir, a deep clustering method is used to adaptively classify the projected one-dimensional sweet spot distribution, learn the implicit feature representation of the data through a deep neural network, and combine the clustering algorithm to realize the automatic grading of the reservoir, and finally establish a quantitative reservoir classification and grading evaluation standard.

[0178] This embodiment adopts a network structure combining a deep autoencoder and a clustering layer to realize the end-to-end learning of shale reservoir feature extraction and clustering methods. The network structure includes a deep autoencoder and a clustering layer. Among them, the deep autoencoder network consists of two parts: an encoder f(B) and a decoder g(z), and its structure is as follows:

[0179]

[0180] Among them, B is the input comprehensive fracturing sweet spot data of the reservoir, z is the feature representation output by the encoder, σ e and σ d are the activation functions of the encoder and decoder respectively, W e is the encoder weight matrix, W d is the decoder weight matrix, b e and b d are the bias vectors of the encoder and decoder respectively. The dimensions of each layer of the network decrease layer by layer, which is convenient for extracting the intrinsic features of the data.

[0181] The clustering layer maps the encoded feature representation z to the cluster centers, and constructs a joint optimization objective LL by combining the reconstruction loss and the clustering loss:

[0182] LL = Lr + βLc;

[0183] Among them, β is the balance parameter, and the reconstruction loss Lr is used to ensure the effectiveness of feature extraction:

[0184] Lr = ||B - g(f(B))|| 2 ;

[0185] The clustering loss Lc adopts an auxiliary target distribution based on the student t-distribution:

[0186] Lc = JD KL (P||Q);

[0187] Among them, J is the preset number of clustering categories, P is the auxiliary target distribution, Q is the predicted soft assignment probability, and D KL (P||Q) represents the KL divergence, which is used to measure the difference between distribution P and distribution Q.

[0188] The auxiliary target distribution P is calculated by the following formula:

[0189]

[0190] Among them, the soft assignment probability q mo represents the probability that the m-th sample belongs to the o-th cluster, and is calculated by the student t-distribution; z m is the encoded feature representation of the m-th sample, μ o is the o-th cluster center, v is the degree of freedom of the t-distribution; f o is the normalization factor of the o-th cluster, and p mo is the target probability that the m-th sample belongs to the o-th cluster.

[0191] Based on the deep clustering calculation results, further combined with the reservoir engineering practice experience, a classification and grading evaluation standard for shale reservoirs is established. Among them, the sweet spot value range of the first-class reservoir (Class I) is located in [V 1 , 1], V 1Characterize the lower limit of a certain type of reservoir. This type of reservoir has the optimal conditions for fracturing transformation and requires priority implementation of fracturing transformation; the sweet spot value range of the second type of reservoir (Type II) is located in [V 2 , V 1 ), V 2 Characterize the lower limit of the developable reservoir. The second type of reservoir has good conditions for fracturing transformation and is suitable for conventional fracturing transformation; the sweet spot value range of the third type of reservoir (Type III) is [0, V 2 ), and the fracturing transformation conditions of the third type of reservoir are poor, and special fracturing techniques need to be used or the transformation is not considered for the time being. Among them, the demarcation values V 1 and V 2 are first obtained based on the depth clustering results, and then fine-tuned and optimized in combination with engineering practice experience. Realize the scientific classification of shale reservoirs and provide an important decision-making basis for the design and implementation of fracturing transformation.

[0192] After establishing the comprehensive fracturing sweet spot evaluation model for shale reservoirs in this embodiment, the constructed machine learning depth clustering algorithm is used to carry out the classification and grading evaluation of shale reservoirs, obtain the classification boundaries of the normalized comprehensive fracturing sweet spot evaluation indexes for different types of reservoirs, and obtain the classification boundaries of the mobility and fracturability evaluation indexes for Class I, II, and III reservoirs according to the corresponding relationship between the normalized evaluation indexes and the original data, as shown in Tables 3 and 4.

[0193] Table 3 Classification boundaries of mobility evaluation indexes for different types of reservoirs

[0194]

[0195] Table 4 Classification boundaries of fracturability evaluation indexes for different types of reservoirs

[0196]

[0197] Secondly, the classification and grading evaluation criteria for the comprehensive fracturing sweet spot of shale reservoirs calculated can also be obtained by using the depth clustering analysis method, as shown in Table 5, forming a method for identifying the comprehensive fracturing sweet spot of horizontal wells.

[0198] Table 5 Comprehensive fracturing sweet spot distribution of different types of reservoirs

[0199] Type Class I reservoir (excellent) Class II reservoir (medium) Class III reservoir (poor) Comprehensive fracturing sweet spot value ≧0.57 0.37-0.57 ≦0.37

[0200] The comprehensive fracturing sweet spot identification model for shale reservoirs considers the oil-bearing property, reservoir property, and movable hydrocarbon content of shale. The identification result has higher evaluation accuracy than the traditional fracturability index, and is of great significance for the efficient development of shale oil and cost reduction and efficiency increase.

[0201] Certainly, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. A shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible characteristics, characterized in that: The following steps are involved: S1. Based on the logging interpretation results of shale oil single well reservoir, a comprehensive fracturing sweet spot evaluation index reflecting the potential mobility and compressibility of the reservoir is constructed and calculated; S2. Based on the constructed information entropy-based adaptive weight determination method, mutual information correction and kernel density estimation are introduced to realize the calculation of evaluation index weights, thereby determining the distribution of comprehensive fracturing sweet spots along the wellbore; S3, based on the comprehensive fracturing sweet spot values ​​of shale oil horizontal wells along the wellbore direction, combined with the sequential Gaussian simulation method, the three-dimensional reservoir comprehensive fracturing sweet spot distribution is constructed; S4. Based on the t-SNE algorithm, the dimensionality reduction projection of the three-dimensional comprehensive fracturing sweet spot distribution is obtained, and the vertical distribution of the sweet spot is calculated to achieve a three-dimensional depiction of the geometric shape of the sand body and the spatial distribution of the "sweet spot reservoir" under complex geological conditions of the shale reservoir; S5. Based on the deep clustering algorithm, the projected one-dimensional comprehensive fracturing sweet spot distribution is adaptively classified to construct a quantitative shale reservoir classification and grading evaluation standard.

2. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible characteristics according to claim 1 is characterized in that: In step S1, the positive factors in the reservoir potential mobility evaluation index are organic carbon content, oil saturation, organic matter maturity, porosity, permeability, free hydrocarbon and saturated adsorption coefficient, the positive factors in the reservoir potential compressibility evaluation index are brittleness index and Young's modulus, and the negative factors in the reservoir potential compressibility evaluation index are Poisson's ratio, fracture toughness, horizontal maximum / minimum ground stress and stress difference; No treatment is given to positive factors, and the reciprocal of negative factors is taken.

3. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 2 is characterized in that: In step S2, the continuous entropy of the evaluation index based on kernel density estimation is: H(x)=-∫p(x)logp(x)dx; Among them, H(x) is the continuous entropy of indicator x; p(x) is the probability density function of indicator x; Among them, n′ is the number of samples for comprehensive fracturing sweet spot evaluation; K is the kernel function, which uses the Gaussian kernel function; x I is the Ith sample point; h is the bandwidth parameter, and the bandwidth parameter uses the Silverman criterion to determine the optimal bandwidth of the kernel density estimation; <h2 style=";text-align:left;direction:ltr">h=0.9AN<h2 style=";text-align:left;direction:ltr"> -1 / 5 <h2 style=";text-align:left;direction:ltr"> ; Where A = min(σ, IQR / 1.34), σ is the sample standard deviation, IQR is the interquartile range, and N is the number of samples; Among them, MI(x,y) is the mutual information between indicators x and y; p(x,y) is the joint probability density function between indicators x and y; p(y) is the marginal probability density function of indicator y; Based on continuous entropy and mutual information, a weight calculation model for evaluation indicators is constructed: Among them, W i is the weight of the evaluation index of the i-th comprehensive fracturing sweet spot; x j is the weight of the j-th comprehensive fracturing sweet spot evaluation index; H i is the continuous entropy of the evaluation index of the i-th comprehensive fracturing sweet spot; H j is the continuous entropy of the jth comprehensive fracturing sweet spot evaluation index; N′ is the total number of evaluation indicators; MI(x i ,x j ) is the evaluation index x i and x j The mutual information between them; α is the information redundancy penalty coefficient; Among them, α opt is the optimal information redundancy penalty coefficient obtained through the optimization process; argmin α To find the α that minimizes the objective function value within the range of the parameter α; M is the number of cross-validation folds; L(D m ,α) is the validation loss on the m-th fold data; D m Represents the m-th fold validation dataset.

4. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 3 is characterized in that: In step S2, in order to ensure that the value range of each evaluation index is between 0 and 1, each factor is normalized and the original data is linearly transformed. The conversion function is as follows: Among them, Y i′j′ is the normalized value of the j′th indicator of the i′th sample; X i′j′ is the value of the j′th index of the i′th sample in the original data; X j′ represents the set of the j′th index in all samples; min(X j′ ) and max(X j′ ) are the minimum and maximum values ​​of the j′th indicator respectively; The comprehensive fracturing sweet spot expression is: Wherein, CFS is the comprehensive fracturing sweet spot; r is the positive factor index; t is the negative factor index; P is the positive factor; N is the negative factor; r′ is the number of positive factors; t′ is the number of negative factors; W r is the weight of positive factors, W t is the weight of the reverse factor; P r is the original value of the rth positive evaluation index, N t is the original value of the tth reverse evaluation index; P min is the minimum value of the rth positive evaluation index, P max is the maximum value of the rth positive evaluation index, N min is the minimum value of the tth reverse evaluation index, N max is the maximum value of the tth reverse evaluation index.

5. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 4 is characterized in that: In step S3, the known sweet spot value at the wellbore position is used as a hard data constraint, and the sweet spot value at the undrilled position in the study area is predicted through spatial correlation analysis and conditional probability estimation, so as to obtain a complete three-dimensional sweet spot distribution; First, the sweet spots along the wellbore of each well are spatially structured and represented as a discrete point set CFS(x k ,y k ,z k ), where CFS(x k ,y k ,z k ) is the wellbore position (x k ,y k ,z k ) of the reservoir comprehensive fracturing sweet spot; Due to the heterogeneous characteristics of the sweet spot distribution of shale reservoirs, a spatial variation function considering anisotropy is established: Among them, γ is the variogram; h vec is the spatial position vector; A is the anisotropy matrix, which is used to describe the horizontal and vertical correlation differences; C o is the nugget value; C is the base value; a is the range; Constructing a three-dimensional sweet spot distribution based on multi-well constraints, first define the multi-well constraint set D = {(x k ,y k ,z k ,CFS k )|k=1,...,nn}, where nn is the total number of wellbore sampling points, CFS k is the wellbore position (x k ,y k ,z k ) of the reservoir comprehensive fracturing sweet spot; For any point x to be estimated in the grid system 00 :x 00 =(x0,y0,z0) is the conditional probability distribution of the sweet spot value: Among them, P represents the conditional probability, CFS0 represents the point x to be estimated 00 The comprehensive fracturing sweet spot value of the reservoir at, μ0 represents the estimated point x 00 The conditional expectation at Represents the point x to be estimated 00 The variance at ; Normal means normal distribution; Among them, γ is the value of the variogram; λ kk is the kriging weight considering multiple well constraints, obtained by solving the kriging equations; kk is the wellbore sampling point index; mm is the total number of wellbores; CFS k Represents the point x to be estimated kk The comprehensive fracturing sweet spot value of the reservoir at; Kriging equations: Among them, λ jj and λ kk are all Kriging weights considering multi-well constraints, jj=1,...,mm,x kk and x jj is the three-dimensional coordinate of the known wellbore position, and μ is the Lagrange multiplier.

6. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 5 is characterized in that: In step S3, the process of constructing the three-dimensional reservoir comprehensive fracturing sweet spot distribution using the sequential Gaussian simulation method includes: first, establishing a three-dimensional grid system covering the study area, and the grid resolution is reasonably determined according to the research requirements and computing resources; The known sweet spot values ​​at the wellbore locations are input into the system as hard data and a random access path is generated that takes into account the geological trends; For each grid point to be simulated in the path, firstly, condition data is collected within a specified search range of the grid point to be simulated, wherein the condition data includes hard data points of the wellbore and grid point data of completed simulation; Based on the collected conditional data, a conditional probability distribution is obtained by solving a Kriging equation group, and a random sample is taken from the conditional probability distribution to obtain a simulation value; After each simulation of a grid point is completed, the simulated value of the grid point is immediately added to the data set as new conditional data for use in the simulation of subsequent grid points.

7. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 6 is characterized in that: In step S4, the t-SNE algorithm achieves dimensionality reduction by minimizing the difference between data points in high-dimensional space and low-dimensional space, and maintains the local relationship between data points during the dimensionality reduction process; For each grid point in three-dimensional space, its high-dimensional feature vector is defined as a i =(X ai ,Y ai ,X ai ,CFS ai ), where (X ai ,Y ai ,Z ai ) is the three-dimensional position coordinate, CFS ai is the position (X ai ,Y ai ,Z ai ) at the comprehensive fracturing sweet spot; Therefore, for any two high-dimensional feature vectors a containing the three-dimensional position and comprehensive fracturing sweet spot information i and a j The similarity between them is measured using the Gaussian kernel function A aiaj Perform the calculation: THE aiaj =(A(aj|ai)+A(ai|aj)) / (2nn′); The conditional probabilities A(ai|aj) and A(aj|ai) are: Among them, σ t is the bandwidth parameter of the Gaussian kernel of the t-SNE algorithm, nn′ is the total number of grid points in the study area, and a k For the position (X ak ,Y ak ,Z ak ) corresponds to the high-dimensional feature vector of the grid point, ||a i -a j || and ||a j -a i || are the Euclidean distances between two grid points in high-dimensional space; In the one-dimensional projection space, the similarity between two points of the three-dimensional reservoir comprehensive fracturing sweet spot is defined by t distribution: q aiaj =(1+||and ai -and aj || 2 ) -1 / ∑ ak≠al (1+||and ak -and al || 2 ) -1 ; Among them, y ai and aj They are respectively i and a j Projected coordinates in one-dimensional space, y ak and al is different from y in the study area ai and aj Any two projection coordinates of ||y ak -y al || is the Euclidean distance between two projection coordinates in one-dimensional space, and the denominator summation term traverses all grid point pairs with ak≠al in the study area.

8. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 7 is characterized in that: In step S5, a network structure combining a deep autoencoder and a clustering layer is used to achieve end-to-end learning of shale reservoir feature extraction and clustering methods; The network structure includes a deep autoencoder and a clustering layer, wherein the deep autoencoder network consists of two parts: an encoder f(B) and a decoder g(z), and its structure is as follows: Among them, B is the input reservoir comprehensive fracturing sweet spot data, z is the feature representation output by the encoder, σ e and σ d are the activation functions of the encoder and decoder respectively, W e is the encoder weight matrix, W d is the decoder weight matrix, b e and b d are the bias vectors of the encoder and decoder respectively; The clustering layer maps the encoded feature representation z to the cluster center and combines the reconstruction loss and clustering loss to construct the joint optimization objective LL: LL=Lr+βLc; Among them, β is a balance parameter, and the reconstruction loss Lr is used to ensure the effectiveness of feature extraction: Lr=||B-g(f(B))|| 2 ; The clustering loss Lc adopts an auxiliary target distribution based on Student's t distribution: Lc=JD KL (P||Q); Among them, J is the preset number of cluster categories, P is the auxiliary target distribution, Q is the predicted soft assignment probability, D KL (P||Q) represents the KL divergence, which is used to measure the difference between distribution P and distribution Q; The auxiliary target distribution P is calculated by the following formula: Among them, the soft assignment probability q mo represents the probability that the mth sample belongs to the oth cluster, z m is the encoding feature representation of the mth sample, μ o is the oth cluster center, v is the degree of freedom of t distribution, f o is the normalization factor of the o-th cluster, p mo is the target probability that the mth sample belongs to the oth cluster.

9. The shale comprehensive fracturing sweet spot evaluation method based on the fusion of movable and compressible features according to claim 8, characterized in that: In step S5, based on the deep clustering calculation results and further combined with reservoir engineering practice experience, a shale reservoir classification and grading evaluation standard is established; Among them, the sweet spot value interval of a type of reservoir is located at [V1,1], V1 represents the lower limit of a type of reservoir, and a type of reservoir has the best fracturing transformation conditions, and fracturing transformation needs to be implemented first; The sweet spot value interval of the second type reservoir is located at [V2, V1), V2 represents the lower limit of the exploitable reservoir, and the second type reservoir has good fracturing conditions and is suitable for conventional fracturing. The sweet spot value interval of the three types of reservoirs is [0, V2). The fracturing conditions of the three types of reservoirs are poor and are not considered for the time being; Among them, the cutoff values ​​V1 and V2 are first obtained based on the deep clustering results to obtain the initial cutoff values, and then fine-tuned and optimized in combination with engineering practice experience.

Images