An interpretable fracturing parameter optimization method and device based on data driving

Abstract

This invention discloses a data-driven, interpretable fracturing parameter optimization method and apparatus, comprising: using a target geological sweet spot clustering model to cluster the effective geological parameters along the well depth of the target well, and outputting the sweet spot probability distribution data of the target well; based on the sweet spot probability distribution data, geologically grouping historical fracturing sections of historical wells, and within each geological group, using the target fracturing parameter clustering model within the group to cluster the effective fracturing parameters of the historical fracturing sections, and outputting an engineering sweet spot template for each geological group; dividing the target well into fracturing sections using the sweet spot probability distribution data, and within each fracturing section, matching initial fracturing parameters from the historical fracturing sections based on the proportion of high-quality sweet spots; querying the engineering sweet spot template of the geological group to which the historical fracturing section belongs, optimizing the initial fracturing parameters, and obtaining the target fracturing parameters of the target well. This invention can provide a scientific, transparent, and practical fracturing parameter optimization solution.

Description

Technical Field

[0001] This invention relates to the field of unconventional oil and gas development technology, and in particular to a data-driven, interpretable fracturing parameter optimization method and apparatus. Background Technology

[0002] Hydraulic fracturing is a key technology for the efficient development of unconventional oil and gas reservoirs, and its effectiveness largely depends on the rationality of fracturing parameter design. However, current methods for optimizing fracturing parameters suffer from the following problems:

[0003] (1) Disconnect between sweet spot evaluation and production capacity: Existing technologies are based on clustering and sweet spot evaluation using reservoir physical parameters, lacking a direct correlation with actual production data. The evaluation criteria for sweet spots are static spatial distribution of geological parameters, rather than dynamic production capacity contribution, which means that the identified "sweet spots" may not be reservoir segments that truly play a dominant role in high production, causing subsequent optimization to lose its precise target orientation; (2) Complex and uninterpretable optimization process: Existing technologies seek optimization between geological sweet spots and engineering sweet spots through multi-objective functions. This optimization process relies on complex mathematical solutions, and the result is a set of Pareto optimal solutions, rather than clear and executable solutions. Field engineers find it difficult to understand why this parameter is chosen instead of that parameter, the optimization logic is opaque, and the interpretability is poor, which is not conducive to rapid decision-making and field application; (3) Neglecting the constraints of geology on engineering: The optimization of fracturing parameters in existing technologies is carried out on a global scale, without fully considering the different effects of different geological conditions on the fracturing parameter effects. Seeking general engineering laws without distinguishing geological backgrounds is prone to leading to vague or misleading conclusions due to the interference of geological heterogeneity, resulting in insufficient geological adaptability of the optimization scheme. In summary, existing technologies cannot provide a scientific, transparent, and practical solution for optimizing fracturing parameters.

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

[0005] This specification provides a data-driven, interpretable fracturing parameter optimization method and apparatus to address the problem that existing technologies cannot provide a scientific, transparent, and practical fracturing parameter optimization solution.

[0006] Firstly, embodiments of this specification provide a data-driven, interpretable fracturing parameter optimization method, including:

[0007] The target geological sweet spot clustering model is used to cluster the effective geological parameters of the target well along the well depth, and output the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between each effective geological parameter of historical wells and the oil production.

[0008] Based on the sweet spot probability distribution data, the historical fracturing sections of historical wells are geologically grouped. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and the fluid production of historical wells.

[0009] The target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data.

[0010] Query the engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs, optimize the initial fracturing parameters, and obtain the target fracturing parameters for the target well fracturing section.

[0011] In some embodiments, the cross-well correlation coefficients between the effective geological parameters of each historical well and the oil production are obtained in the following manner:

[0012] For each historical well, calculate the first correlation coefficient between each geological parameter and the oil production of each historical well;

[0013] A significance test is performed on the first correlation coefficient between each geological parameter and oil production. Based on the significance test results and the first correlation coefficient, effective geological parameters of each historical well are selected from each geological parameter of each historical well. The significance test results of the effective geological parameters reach the preset significance level and the absolute value of the first correlation coefficient is not less than the preset correlation coefficient threshold.

[0014] Fisher Z-transform is performed on the second correlation coefficient between each effective geological parameter and oil production of each historical well, and then the target value obtained by the transformation is inversely transformed to obtain the cross-well correlation coefficient between each effective geological parameter and oil production of the historical well.

[0015] In some embodiments, the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficients between effective geological parameters and oil production of historical wells, including:

[0016] The cross-well weight of each effective geological parameter is calculated based on the absolute value of the cross-well correlation coefficient between each effective geological parameter and oil production in historical wells.

[0017] Based on the cross-well weight of each effective geological parameter, a first weighted Euclidean distance function is constructed for the target geological sweet spot clustering model; the first weighted Euclidean distance function is used to calculate the distance from each depth point to the center of each cluster.

[0018] Accordingly, the method of using the target geological sweet spot clustering model to cluster the effective geological parameters along the well depth of the target well and output the sweet spot probability distribution data of the target well includes:

[0019] Initialize the normalized effective quality parameters of each depth point to the membership matrix of each cluster, wherein the membership matrix satisfies that the sum of the membership degrees of each depth point to each cluster is 1;

[0020] Based on the initialized membership matrix and the normalized effective quality parameters of each depth point, the cluster center of each cluster is calculated;

[0021] Based on the cluster centers of each cluster and the first weighted Euclidean distance function, the membership degree of the normalized effective quality parameters of each depth point to each cluster is recalculated to form an updated membership matrix;

[0022] Determine whether the difference between the updated membership matrix and the membership matrix is ​​less than the first preset convergence threshold;

[0023] If so, the clustering is determined to have converged. The updated membership matrix is ​​used as the final target membership matrix. Based on the final target cluster center and the cross-well weights of each effective genomic parameter, the comprehensive score of the target cluster center of each cluster is determined. The target cluster center is calculated based on the target membership matrix and the normalized effective genomic parameters of each depth point.

[0024] By comparing the comprehensive scores of each target cluster center, clusters with comprehensive scores greater than a preset score threshold are identified as high-quality sweet spot clusters. The membership degree of each depth point in the target membership matrix to the high-quality sweet spot clusters is taken as the sweet spot probability of each depth point, thus forming the sweet spot probability distribution data.

[0025] In some embodiments, determining the comprehensive score of the target cluster center for each cluster based on the final target cluster center and the cross-well weights of each effective geology parameter includes:

[0026] The overall score of the target cluster center is determined according to the following formula:

[0027]

[0028] Among them, S k S is the comprehensive score of the target cluster center of the k-th cluster; j S represents the relevant direction of the j-th geological parameter. j Take -1 or +1; w j c is the cross-well weight of the j-th effective quality parameter; kj Let be the j-th geological parameter of the k-th cluster; D is the total number of effective geological parameters.

[0029] In some embodiments, the geological grouping of historical fracturing sections of historical wells based on the sweet spot probability distribution data includes:

[0030] From the sweet spot probability distribution data, the sweet spot probability of all depth points of each historical fracturing segment is obtained, and the proportion of high-quality sweet spots in each historical fracturing segment is calculated based on the sweet spot probability.

[0031] The quartiles of the proportion of high-quality sweet spots in historical fracturing sections were determined, and based on these quartiles, the historical fracturing sections of all historical wells were divided into four geological groups: high, medium-high, medium-low, and low.

[0032] In some embodiments, the target fracturing parameter clustering model is constructed based on the intra-group correlation coefficients between each effective fracturing parameter and production rate of historical wells, including:

[0033] The intra-group weight of each effective fracturing parameter is calculated based on the absolute value of the intra-group correlation coefficient between each effective fracturing parameter and the production of fluid in historical wells.

[0034] Based on the intra-group weight of each effective fracturing parameter, a second weighted Euclidean distance function is constructed for the target fracturing parameter clustering model; the second weighted Euclidean distance function is used to calculate the distance from the historical fracturing segment to the cluster center of the fracturing parameter.

[0035] Accordingly, the effective fracturing parameters of historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output, including:

[0036] Based on the cluster center of the initial fracturing parameter cluster and the second weighted Euclidean distance function, each historical fracturing segment is assigned to the nearest fracturing parameter cluster;

[0037] Based on the normalized effective fracturing parameters of all historical fracturing segments within each fracturing parameter cluster, the cluster center of the fracturing parameter cluster is recalculated.

[0038] The steps of historical fracturing segment allocation and cluster center recalculation are performed iteratively until the cluster center change between two adjacent iterations is less than the second preset convergence threshold.

[0039] After clustering convergence, the average fluid production of all historical fracturing segments within each fracturing parameter cluster is calculated. The cluster center of the fracturing parameter cluster with the highest average fluid production is determined as the engineering sweet spot template corresponding to the geological grouping.

[0040] In some embodiments, the step of dividing the target well into fracturing sections using the sweet spot probability distribution data includes:

[0041] The sweet spot probability distribution data of the target well is processed by moving average to obtain the smoothed sweet spot probability;

[0042] Calculate the fuzziness index for each depth point of the target well; the fuzziness index is used to characterize the uncertainty of the geological category at each depth point;

[0043] Using the smoothed dessert probability and the fuzzy index together as input signals, the Planck energy minimization change point detection algorithm is used for preliminary segment boundary detection;

[0044] For the segment boundaries obtained from the preliminary segment boundary detection, engineering constraints of minimum and maximum segment lengths are applied. Excessively short segments are merged and excessively long segments are divided to obtain a fracturing segment division scheme. Based on the fracturing segment division scheme, fracturing segments are divided for the target well.

[0045] In some embodiments, matching initial fracturing parameters from historical fracturing sections based on a high-quality sweet spot ratio within each fracturing section includes:

[0046] Based on the proportion of high-quality sweet spots and the length of each fractured section in the target well, a feature vector for each fractured section is constructed.

[0047] Calculate the Euclidean distance between the feature vector and the corresponding feature vector of each historical fracturing segment;

[0048] The fracturing parameters used in historical fracturing sections where the Euclidean distance from the target well fracturing section is less than a preset distance threshold are used as the initial fracturing parameters for the target well fracturing section.

[0049] In some embodiments, the engineering sweet spot template corresponding to the geological group to which the fractured section of the queried historical well belongs optimizes the initial fracturing parameters and outputs the target fracturing parameters for the target well fractured section, including:

[0050] Query the engineering sweet spot templates corresponding to the geological groupings, use the engineering sweet spot template library of historical wells as the training set, train the CART decision tree model, and generate decision rules for determining whether the fracturing parameter combination is an engineering sweet spot;

[0051] The initial fracturing parameters are input into the trained CART decision tree model for discrimination.

[0052] If the determination result is a non-engineering sweet spot, the initial fracturing parameters are adjusted according to the decision rules until the determination result is an engineering sweet spot, and the adjusted initial fracturing parameters are used as the final optimized target fracturing parameters.

[0053] Secondly, embodiments of this specification also provide a data-driven, interpretable fracturing parameter optimization device, comprising:

[0054] The first clustering module is used to cluster the effective geological parameters of the target well along the well depth using the target geological sweet spot clustering model, and output the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between the effective geological parameters of each historical well and the oil production.

[0055] The second clustering module is used to geologically group the historical fracturing sections of historical wells based on the sweet spot probability distribution data. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and the production volume of historical wells.

[0056] The initial fracturing parameter matching module is used to divide the target well into fracturing segments using the sweet spot probability distribution data. Within each fracturing segment, initial fracturing parameters are matched from historical fracturing segments based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data.

[0057] The initial fracturing parameter optimization module is used to query the engineering sweet spot template corresponding to the geological group to which the fracturing section of the historical well belongs, optimize the initial fracturing parameters, and output the target fracturing parameters of the target well fracturing section.

[0058] This specification provides a data-driven, interpretable fracturing parameter optimization method and apparatus. First, a target geological sweet spot clustering model is used to cluster the effective geological parameters along the well depth of the target well, outputting sweet spot probability distribution data. This model is constructed based on the cross-well correlation coefficients between effective geological parameters and production from historical wells. Second, based on the sweet spot probability distribution data, historical fracturing sections of historical wells are geologically grouped. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, outputting an engineering sweet spot template corresponding to each geological group. This model is constructed based on the intra-group correlation coefficients between effective fracturing parameters and production from historical wells. Then, the target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. Finally, the engineering sweet spot templates corresponding to the geological groups to which the historical fracturing sections belong are queried, and the initial fracturing parameters are optimized to obtain the target fracturing parameters for the target well's fracturing sections. In the embodiments of this specification, a target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between effective geological parameters and oil production of historical wells, and a target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between effective fracturing parameters and fluid production of historical wells. This avoids the disconnect between sweet spot evaluation and production capacity, thereby improving the accuracy of sweet spot evaluation and its engineering relevance. The target geological sweet spot clustering model can accurately output the sweet spot probability distribution data of the target well, and the intra-group target fracturing parameter clustering model can accurately output the engineering sweet spot template corresponding to each geological group. By geologically grouping the historical fracturing sections of historical wells based on the sweet spot probability distribution data, the problem of existing technologies neglecting the constraints of geology on engineering can be avoided, ensuring relatively uniform geological conditions within the group. Independent clustering of effective fracturing parameters within each geological group can eliminate interference from geological heterogeneity. By using the sweet spot probability distribution data to divide the target well into fracturing sections, scientific segmentation can be achieved. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections, enabling a "one-section-one-policy" approach and ensuring the scientific validity and reliability of initial fracturing parameter matching. By querying the engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs, the initial fracturing parameters can be optimized, avoiding the problems of complex and uninterpretable optimization process. The target fracturing parameters with transparent and interpretable results can be obtained, which effectively improves the operability and field acceptance of the well completion optimization design scheme. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0060] Figure 1 This specification provides a data-driven, interpretable fracturing parameter optimization method for embodiments of the invention.

[0061] Figure 2 A schematic diagram of the structural composition of a data-driven interpretable fracturing parameter optimization device provided in the embodiments of this specification;

[0062] Figure 3 This is a schematic diagram of the structural composition of the electronic device provided in the embodiments of this specification. Detailed Implementation

[0063] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0064] Hydraulic fracturing is crucial for the efficient development of unconventional reservoirs, and its effectiveness largely depends on the rationality of the well completion design. However, current well completion design optimization methods mainly suffer from three types of problems:

[0065] The first category is based on statistical and empirical methods. These methods use historical data from multiple wells to establish empirical relationships between parameters and production. However, they typically assume linear or monotonic relationships between variables, making it difficult to capture the complex nonlinear coupling mechanisms between geology, engineering, and production capacity. This results in poor adaptability of optimization schemes in highly heterogeneous reservoirs.

[0066] The second category is based on numerical simulation methods. These methods utilize fracture propagation and reservoir simulation models to assess the sensitivity of different fracturing parameters. While they can reflect the physical processes, they are computationally complex, time-consuming, and costly. Furthermore, their results heavily depend on input parameters that are difficult to obtain accurately, resulting in high uncertainty and making them unsuitable for rapid on-site decision-making and large-scale application.

[0067] The third category is data-driven intelligent methods. These methods construct "black box" models such as neural networks and random forests to predict yields and inversely determine optimal parameters. While they can fit complex nonlinear relationships, their inherent structural complexity and insufficient interpretability make it difficult to extract clear on-site decision-making rules, thus limiting the model's transparency and engineers' trust.

[0068] Furthermore, it is worth noting that most production prediction models are based on the average parameters of the entire well or the fractured section, which ignores the heterogeneity at the depth point scale within the wellbore and cannot provide guidance for fine optimization of cluster parameters within the section.

[0069] In the existing technology, geological sweet spots are evaluated by clustering and segmented and optimized by combining engineering constraints. However, it still has the following shortcomings: (1) its sweet spot evaluation lacks direct correlation with actual production data, making it difficult to truly reflect the coupling relationship between reservoir-engineering-production capacity; (2) the optimization process relies on complex multi-objective function solutions, the results have poor interpretability and are not conducive to field application; (3) insufficient consideration is given to the geological heterogeneity within the fracturing section.

[0070] In summary, existing technologies cannot provide a scientific, transparent, and practical solution for optimizing fracturing parameters.

[0071] To address the aforementioned issues, this specification provides a data-driven, interpretable fracturing parameter optimization method and apparatus. First, a target geological sweet spot clustering model is used to cluster the effective geological parameters along the well depth of the target well, outputting sweet spot probability distribution data for the target well. This model is constructed based on the cross-well correlation coefficients between effective geological parameters and production from historical wells. Second, based on the sweet spot probability distribution data, historical fracturing sections of historical wells are geologically grouped. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, outputting an engineering sweet spot template corresponding to each geological group. This model is constructed based on the intra-group correlation coefficients between effective fracturing parameters and production from historical wells. Then, the target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. Finally, the engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs is queried, and the initial fracturing parameters are optimized to obtain the target fracturing parameters for the target well fracturing section.

[0072] This invention combines data-driven approaches with mechanistic analysis, overcoming the subjectivity of traditional empirical methods, the high cost of numerical simulation, and the lack of interpretability of black-box models. It provides a scientific, transparent, and practical solution for optimizing fracturing parameters in unconventional oil and gas reservoirs.

[0073] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate for the embodiments of this application described herein.

[0074] It is understood that the methods described in the embodiments of this specification can be applied to electronic devices, which can refer to electronic devices with data computing, processing, and storage capabilities. These electronic devices can be terminals such as PCs (Personal Computers), tablets, smartphones, wearable devices, and intelligent robots; they can also be servers. A server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0075] See Figure 1 As shown in the embodiments of this specification, a data-driven, interpretable fracturing parameter optimization method is provided. In specific implementation, this method may include the following:

[0076] S101: Using the target geological sweet spot clustering model, the effective geological parameters along the well depth of the target well are clustered to output the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between the effective geological parameters of each historical well and the oil production.

[0077] In some embodiments, a unified and standardized multi-source dataset can be constructed before modeling. Subsequent modeling can then directly obtain the required data from this dataset, thereby improving modeling speed. Specifically, readily available multi-source heterogeneous data from the oilfield can be collected, including geological parameters, fracturing parameters, and tracer monitoring data (production capacity data), as shown in Table 1.

[0078] Table 1

[0079]

[0080] The geological parameters are at the decimeter level for the entire well section, while the fracturing parameters and tracer monitoring parameters are at the section scale. To ensure spatial consistency and engineering relevance of the parameters, a series of processing rules were formulated. Finally, the multi-source dataset used subsequently was obtained. The data processing rules are as follows:

[0081] Considering the limited longitudinal extension of hydraulic fractures along the wellbore, all geological parameters at the cluster scale are represented by the average value within ±1m of the corresponding perforation cluster, serving as a characterization of the local geological properties of that cluster. At the fractured section scale, the overall reservoir characteristics of a fractured section are obtained by averaging the geological properties of all clusters within that section. The fractured section length is defined as a continuous wellbore segment extending 5m above the first cluster to 5m below the last cluster, to fully cover the potential fracture propagation range during fracturing and to serve as a spatial reference for assigning operational parameters and related indices.

[0082] By processing multi-source heterogeneous data through precise scale alignment, a unified and standardized multi-source dataset can be constructed, which can lay a solid data foundation for establishing a reliable relationship between geology, engineering, and production capacity, and avoid erroneous conclusions caused by data scale mismatch.

[0083] In some embodiments, the cross-well correlation coefficients between the effective geological parameters of each historical well and the oil production in S101 above can be obtained in the following manner:

[0084] For each historical well, calculate the first correlation coefficient between each geological parameter and the oil production of each historical well;

[0085] A significance test is performed on the first correlation coefficient between each geological parameter and oil production. Based on the significance test results and the first correlation coefficient, effective geological parameters of each historical well are selected from each geological parameter of each historical well. The significance test results of the effective geological parameters reach the preset significance level and the absolute value of the first correlation coefficient is not less than the preset correlation coefficient threshold.

[0086] Fisher's Z-transform is performed on the second correlation coefficient between each effective quality parameter and oil production of each historical well. Then, the target value obtained by the transformation is inversely transformed to obtain the cross-well correlation coefficient between each effective quality parameter and oil production of the historical well.

[0087] Specifically, to explain the control mechanism of fracturing stage production capacity and overcome the limitations of single-parameter analysis, it is necessary to systematically identify the key factors, i.e., effective geological parameters, in multi-source datasets that have a significant impact on oil production. Based on this, a weighting system with both physical significance and modeling value can be established. The calculation process for the cross-well correlation coefficients between each effective geological parameter of historical wells and oil production can be as follows:

[0088] First, for each historical well, calculate the first correlation coefficient between each geological parameter and oil production using the following formula:

[0089] (1)

[0090] Where, ρ l,jd represents the first correlation coefficient between the j-th geological parameter and oil production in the l-th historical well. l,i,j Let n be the difference between the rank of the j-th geological parameter and the rank of oil production in the i-th historical fracturing segment of the l-th well. l This represents the number of historical fracturing sections (total number) in the l-th well.

[0091] After calculating the primary correlation coefficient between each geological parameter and oil production for each historical well, the significance of the primary correlation coefficient between each geological parameter and oil production can be tested using the following formula:

[0092] (2)

[0093] (3)

[0094] Among them, t l,j p is the statistical measure of the j-th geological parameter in the l-th historical well; l,j The significance value of the j-th geological parameter in the l-th historical well; CDF t df is the cumulative distribution function of the t-distribution; df is the degree of freedom, which determines the specific shape of the t-distribution.

[0095] Then, p can be calculated based on the significance test results (i.e., formula (3)). l,j Using the first correlation coefficient mentioned above, effective geological parameters for each historical well are selected from each geological parameter of each historical well, thus allowing for the selection of p. l,j <0.1 and ρ l,j Geological parameters with an absolute value ≥ 0.2 are considered as effective geological parameters, meaning that the significance verification results of effective geological parameters reach the preset significance level and the absolute value of the first correlation coefficient is not less than the preset correlation coefficient threshold.

[0096] It should be noted that ρ calculated by the above formula (1) l,j The values ​​are 0.8~1.0 for very strong correlation, 0.6~0.8 for strong correlation, 0.4~0.6 for moderate correlation, 0.2~0.4 for weak correlation, and 0.0~0.2 for very weak or no correlation. The p calculated using formula (3) above... l,j When ρ is less than 0.05, the differences or correlations between variables are considered statistically significant, and the results are unlikely to be caused by random factors; when ρ is less than 0.1 but greater than or equal to 0.05, it is considered marginally significant, and there may be a weak correlation between the variables; when ρ is greater than or equal to 0.1, the correlation between the variables is considered insufficient, and the observed results may originate from random fluctuations or errors. By limiting ρ... l,j An absolute value ≥ 0.2 ensures that the correlation is at least "weak" and has a real impact, while limiting p. l,jThe value <0.1 ensures that the correlation is significant at a 90% confidence level. These two constraints can effectively filter out some parameters with negligible or insignificant relationships.

[0097] Subsequently, inter-well robustness can be achieved using the Fisher Z-transform. Given that the number of samples from each historical well is typically limited (approximately 15-30), the aforementioned first-phase coefficients may differ significantly between different historical wells. To improve robustness, a weighted method based on the Fisher Z-transform is used to calculate the inter-well correlation coefficients between each effective geological parameter and oil production of the historical wells. The calculation process is as follows:

[0098] First, perform Fisher's Z-transform on the second correlation coefficient between each effective geological parameter and oil production for each historical well according to the following formula to obtain the first target value for each effective geological parameter of each historical well:

[0099] (4)

[0100] Then, using the following formula, a weighted average is applied to the first target value to obtain the target value after averaging each effective quality parameter across wells:

[0101] (5)

[0102] Then, using the following formula, the target value is inversely transformed to obtain the cross-well correlation coefficients between the effective geological parameters of each historical well and the oil production:

[0103] (6)

[0104] Among them, z l,j Let z be the first target value (first z value) of the j-th effective quality parameter in the l-th historical well. The target value (z-value) of the effective quality parameter across wells is denoted as j; L is the number of historical wells (total number). The cross-well correlation coefficient between the effective quality parameter of the j-th historical well and the oil production.

[0105] Through a dual screening mechanism of "significance test" and "correlation coefficient threshold," noisy parameters that are irrelevant to production capacity or have a weak correlation are effectively eliminated, ensuring that only truly statistically and engineeringally significant geological parameters are included in subsequent models. By employing Fisher's Z-transform to fuse inter-well correlation coefficients, a robust cross-well correlation coefficient that can represent the general laws of the entire work area is obtained, effectively overcoming the shortcomings of small data volume and high randomness of single wells. The final "cross-well correlation coefficient" is a rigorously statistically validated and quantified weight calculation basis, providing a clear and scientific basis for the "weighting" operation in the subsequent clustering model.

[0106] In some embodiments, the target geological sweet spot clustering model in S101 above is constructed based on the cross-well correlation coefficient between the effective geological parameters of each historical well and the oil production. In specific implementation, it may include:

[0107] The cross-well weight of each effective geological parameter is calculated based on the absolute value of the cross-well correlation coefficient between each effective geological parameter and oil production in historical wells.

[0108] Based on the cross-well weight of each effective geological parameter, a first weighted Euclidean distance function is constructed for the target geological sweet spot clustering model; the first weighted Euclidean distance function is used to calculate the distance from each depth point to the center of each cluster.

[0109] Accordingly, the above-mentioned S101, which uses a target geological sweet spot clustering model to cluster the effective geological parameters of the target well along the well depth and outputs the sweet spot probability distribution data of the target well, may include the following in specific implementations:

[0110] Initialize the normalized effective quality parameters of each depth point to the membership matrix of each cluster, wherein the membership matrix satisfies that the sum of the membership degrees of each depth point to each cluster is 1;

[0111] Based on the initialized membership matrix and the normalized effective quality parameters of each depth point, the cluster center of each cluster is calculated;

[0112] Based on the cluster centers of each cluster and the first weighted Euclidean distance function, the membership degree of the normalized effective quality parameters of each depth point to each cluster is recalculated to form an updated membership matrix;

[0113] Determine whether the difference between the updated membership matrix and the membership matrix is ​​less than the first preset convergence threshold;

[0114] If so, the clustering is determined to have converged. The updated membership matrix is ​​used as the final target membership matrix. Based on the final target cluster center and the cross-well weights of each effective genomic parameter, the comprehensive score of the target cluster center of each cluster is determined. The target cluster center is calculated based on the target membership matrix and the normalized effective genomic parameters of each depth point.

[0115] By comparing the comprehensive scores of each target cluster center, clusters with comprehensive scores greater than a preset score threshold are identified as high-quality sweet spot clusters. The membership degree of each depth point in the target membership matrix to the high-quality sweet spot clusters is taken as the sweet spot probability of each depth point, thus forming the sweet spot probability distribution data.

[0116] Specifically, after obtaining the cross-well correlation coefficients between the effective geological parameters and oil production of each historical well, the cross-well weight of each effective geological parameter can be calculated using the following formula:

[0117] (7)

[0118] Where D represents the total number of valid geological parameters selected; The absolute value of the cross-well correlation coefficient between the j-th effective quality parameter of a historical well and its oil production; The sum of the absolute values ​​of the cross-well correlation coefficients of all effective quality parameters; w j Let be the cross-well weight of the j-th effective quality parameter, and the sum of the cross-well weights of all effective quality parameters is 1.

[0119] Subsequently, to identify reservoir conditions with good oil production potential and quantitatively characterize the reservoir performance of fracturing sections, a production-oriented geological sweet spot clustering model (i.e., a target geological sweet spot clustering model) can be established based on the determination of effective geological parameters and their cross-well weights. Specifically, a first weighted Euclidean distance function for the target geological sweet spot clustering model can be constructed based on the cross-well weight of each effective geological parameter. This first weighted Euclidean distance function can be used to calculate the distance from each depth point to the cluster center (the vector of each cluster center).

[0120] This target geological sweet spot clustering model uses effective geological parameters along the well depth of the target well (i.e., the well to be designed) as input. It filters features based on their correlation with oil production data and assigns corresponding weight coefficients, thus highlighting the influence of effective geological parameters in the clustering process. Unlike traditional Euclidean distance clustering, which assumes equal weights for all parameters, the target geological sweet spot clustering model constructed in this invention can automatically identify sweet spot segments with good oil production potential and similar rock physical characteristics. Compared with classification methods that rely on empirical thresholds, this target geological sweet spot clustering model combines effective geological parameters with production feedback, exhibiting higher objectivity, adaptability, and engineering guidance value.

[0121] The following explains the construction of the first weighted Euclidean distance function and the clustering process of the target geological sweet spot clustering model:

[0122] First, obtain the effective geological parameters (total number D, D-dimensional parameters) of the target well along the well depth direction. This will form the following feature matrix of the effective geological parameters of the target well along the well depth direction:

[0123] (8)

[0124] It is important to note that a single-well independent modeling strategy should be adopted, rather than merging all wells into a single model. This is because hydraulic fracturing operations are conducted at the single-well scale, while the goal of "sweet spot" evaluation is to identify relatively favorable intervals within a single well. Furthermore, there are systematic differences in logging baselines and amplitudes between wells; directly merging data may mask intra-well correlations, thus biasing the clustering results. Single-well modeling better highlights the relative distribution of "sweet spots" within each well, more closely reflecting actual fracturing design requirements.

[0125] Based on this, the effective quality parameter feature matrix can be further normalized to obtain the normalized effective quality parameters for each depth point. Normalization eliminates the influence of different dimensions; specific normalization formulas can be found in existing technologies, and will not be elaborated upon here.

[0126] Then, based on the cross-well weight of each effective geological parameter and the normalized effective geological parameter at each depth point mentioned above, the first weighted Euclidean distance function of the target geological sweet spot clustering model is constructed according to the following formula:

[0127] (9)

[0128] in, Let the depth point i be the distance from the cluster center c. k The weighted Euclidean distance indicates that the depth point is more similar to the geological features of that category. For the normalized effective quality parameters of the i-th depth point; w is the j-th normalized effective quality parameter at the i-th depth point; j c is the cross-well weight of the j-th effective quality parameter; kj Let be the j-th effective quality parameter of the k-th cluster.

[0129] After constructing the first weighted Euclidean distance function of the target geological sweet spot clustering model, the model can be used to cluster the effective geological parameters along the well depth of the target well, and output the sweet spot probability distribution data of the target well. The specific process is as follows:

[0130] 1) Initialization: Randomly initialize the normalized effective quality parameters of each depth point to each cluster's membership matrix. The membership matrix satisfies that the sum of the membership degrees of each depth point to each cluster is 1. The expression for the initialized membership matrix can be as follows:

[0131] (10)

[0132] Among them, u ikis the membership degree of the i-th depth point to the k clusters; N is the total number of depth points; C is the total number of clusters (or the best cluster).

[0133] 2) Cluster center update: Based on the membership degrees in the initialized membership matrix and the normalized effective quality parameters of each depth point, the cluster centers of each cluster can be calculated according to the following formula:

[0134] (11)

[0135] Among them, c k The cluster center of cluster k; This is the m-th power of the membership degree of the i-th depth point to k clusters. It is a weight, where m is a fuzziness factor used to control the degree of fuzziness in clustering. When m=1, it degenerates into hard clustering (either / or). The larger m is, the more even the membership degree. m is usually taken as 1.5-2.5. is the normalized effective quality parameter of the i-th depth point (or sample point); N is the total number of depth points or sample points.

[0136] 3) Membership update: Based on the cluster centers and the first weighted Euclidean distance function of each cluster, the membership degree of the normalized effective prime parameters of each depth point to each cluster is recalculated according to the following formula, forming an updated membership matrix:

[0137] (12)

[0138] Among them, u 更新_ik The updated membership degree of the i-th depth point to the k clusters; Let the depth point i be the distance from the cluster center c. k The weighted Euclidean distance can be calculated using formula (9); C is the total number of clusters.

[0139] 4) Iterative optimization: Determine and update the membership matrix u 更新_ik With the membership matrix u ik If the difference between the membership matrix and the new membership matrix is ​​less than the first preset convergence threshold, or if the maximum number of iterations is reached, then the clustering is considered converged. If not, based on the updated membership matrix, return to step 2) to continue iterating. At this point, step 2 becomes calculating the cluster center of each cluster based on the membership degree in the updated membership matrix and the normalized effective quality parameter of each depth point. Repeat steps 2) to 4) until the difference between the updated membership matrix and the new membership matrix is ​​less than the first preset convergence threshold, or the maximum number of iterations is reached. At this point, the membership matrix converges.

[0140] 5) Output results: When the clustering converges, the updated membership matrix can be used as the final target membership matrix and the target membership matrix can be output, which is the membership distribution of each sample point to each category. Its essence can be understood as the "sweet spot probability distribution". This soft partitioning result describes the continuous transition characteristics of geological parameters between high-quality and low-quality sweet spots, which is more in line with the heterogeneity of geological bodies.

[0141] It should be noted that the total number of clusters or the optimal number of clusters C can be determined using a combination of the Fuzzy Partition Coefficient (FPC) and the Chern-Bennie Index (XB). A higher FPC value indicates a clearer partition, while a lower XB value indicates stronger intra-cluster compactness and greater inter-cluster separation. The optimal C is the value that simultaneously produces a relatively high FPC and a minimum XB index.

[0142] In some embodiments, the comprehensive score for determining the target cluster center of each cluster based on the final target cluster center and the cross-well weights of each effective quality parameter may, in specific implementations, include:

[0143] The overall score of the target cluster center is determined according to the following formula:

[0144] (13)

[0145] Among them, S k S is the comprehensive score of the target cluster center of the k-th cluster; j S represents the relevant direction of the j-th geological parameter. j Take -1 or +1; w j c is the cross-well weight of the j-th effective quality parameter; kj S represents the j-th geological parameter of the k-th cluster; D represents the total number of effective geological parameters. Where S... j A value of +1 indicates a positive correlation with productivity, while a value of -1 indicates a negative correlation.

[0146] Specifically, the target cluster centers can be calculated using the target membership matrix and the normalized effective quality parameters of each depth point according to the formula (11) above. When the clustering is determined to be converged, the comprehensive score of the target cluster centers can be calculated using the formula (13) above. Then, the comprehensive score of each target cluster center is compared, and the clusters with comprehensive scores greater than the preset score threshold are regarded as high-quality sweet spot clusters. The membership degree of each depth point in the final target membership matrix to the high-quality sweet spot clusters is then used as the sweet spot probability of each depth point, forming the sweet spot probability distribution data.

[0147] For example: Assume the target well is Well-X, to be evaluated. Effective quality parameters (D=3): POR (porosity, positively correlated with productivity), OIL (oil saturation, positively correlated with productivity), SH (shale content, negatively correlated with productivity). Cross-well weight (w)j ) is calculated based on formula (7), assuming it is [w_ POR =0.50, w _OIL =0.30, w _SH =0.20]. Related directions (S) j ) is [s _POR =+1, s _OIL =+1, s _SH =-1]. The optimal number of clusters C is 2, the fuzzy factor m is 2, and the preset score threshold is 0.5 (for example purposes, the actual threshold may be determined based on the score distribution).

[0148] When clustering is determined to be converged, the final target membership matrix is ​​as follows: Depth point A has a membership of 0.9 to cluster 1 and 0.1 to cluster 2; Depth point B has a membership of 0.75 to cluster 1 and 0.25 to cluster 2; Depth point C has a membership of 0.20 to cluster 1 and 0.80 to cluster 2; Depth point D has a membership of 0.95 to cluster 1 and 0.05 to cluster 2. The sum of the memberships of each depth point to the two clusters is 1. The final target cluster center is calculated according to the above formula (11): Cluster center c1 is: [POR=0.85, OIL=0.88, SH=0.12], c 11 =0.85, c 12 =0.88, c 13 =0.12. Cluster center c2 is: [POR=0.45, OIL=0.55, SH=0.60], c 21 =0.45, c 22 =0.55, c 23 =0.60.

[0149] The comprehensive score S of cluster center c1 is calculated according to the above formula (13). _1 :

[0150] POR term: S1×w1×c 11 =(+1)×0.50×0.85=0.425 S1×w1×c 11 =(+1)×0.50×0.85=0.425

[0151] OIL item: S2×w2×c 12 =(+1) ×0.30×0.88=0.264 S2×w2×c 12 =(+1)×0.30×0.88=0.264

[0152] SH item: S3×w3×c 13=(-1)×0.20×0.12=-0.024 S3×w3×c 13 =(-1)×0.20×0.12=-0.024

[0153] S _1 =0.425 + 0.264 - 0.024 = 0.665

[0154] Similarly, calculate the overall score S of cluster center c2. _2 :S _2 =0.225 + 0.165 - 0.120 = 0.270

[0155] Due to S _1 >S _2 And S _1 Since the score is greater than the preset score threshold (0.5), the cluster represented by cluster core 1 is identified as a high-quality dessert cluster. The membership degree of each depth point in the target membership matrix to the high-quality dessert cluster (cluster 1) is used as the dessert probability of that depth point. That is, depth point A: p(z) = 0.90, depth point B: p(z) = 0.75, depth point C: p(z) = 0.20, and depth point D: p(z) = 0.95 are used as the dessert probability distribution data.

[0156] By assigning cross-well weights to each effective geological parameter, a first weighted Euclidean distance function is constructed for the target geological sweet spot clustering model. Combined with the correlation weights between geological parameters and production capacity (specifically, oil production), a probabilistic and fuzzy representation of reservoir sweet spots is achieved. This avoids the disconnect between sweet spot evaluation and production capacity, thereby improving the accuracy and engineering relevance of sweet spot evaluation. Utilizing the target geological sweet spot clustering model, the probability distribution data of sweet spots in target wells can be accurately output, providing a more accurate data foundation for subsequent geological grouping.

[0157] S102: Based on the sweet spot probability distribution data, the historical fracturing sections of historical wells are geologically grouped. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and the production volume of historical wells.

[0158] In some embodiments, the geological grouping of the historical fracturing sections of the historical wells based on the sweet spot probability distribution data in S102 above may, in specific implementation, include:

[0159] From the sweet spot probability distribution data, the sweet spot probability of all depth points of each historical fracturing segment is obtained, and the proportion of high-quality sweet spots in each historical fracturing segment is calculated based on the sweet spot probability.

[0160] Determine the quartiles of the high-quality sweet spot ratio of historical fracturing intervals. Based on the quartiles, divide the historical fracturing intervals of all historical wells into four geological groups: high, medium-high, medium-low, and low.

[0161] Specifically, first obtain the sweet spot probabilities of all depth points of each historical fracturing interval from the sweet spot probability distribution data (continuous sweet spot probability curve). Then count the number of depth points with sweet spot probabilities greater than the preset sweet spot probability threshold within each historical fracturing interval. Then, according to the number of depth points with sweet spot probabilities greater than the preset sweet spot probability threshold and the total number of depth points within the fracturing interval, calculate the high-quality sweet spot ratio of each historical fracturing interval according to the following formula:

[0162] (14)

[0163] where R sweet is the high-quality sweet spot ratio of a single fracturing interval; N sweet is the number of depth points with sweet spot probabilities greater than the preset sweet spot probability threshold in a single fracturing interval; N total is the total number of depth points in a single fracturing interval.

[0164] After determining the high-quality sweet spot ratio of each historical fracturing interval, the quartiles of the high-quality sweet spot ratio of historical fracturing intervals can be determined according to the following geological grouping rules:

[0165] (15)

[0166] where R sweet ≥Q3 is the high group; Q2≤R sweet <Q3 is the medium-low group; Q1≤R sweet <Q2 is the medium-low group; R sweet <Q1 is the low group; Q1, Q2, Q3 are the 25th, 50th, and 75th percentiles of all segment R sweet values respectively.

[0167] By performing geological grouping on the historical fracturing intervals of historical wells based on the sweet spot probability distribution data, the problem of ignoring the constraint of geology on engineering in the prior art can be avoided, ensuring relatively uniform geological conditions within the group, and independent effective fracturing parameter clustering can be carried out within each geological group to eliminate the interference of geological heterogeneity.

[0168] In some embodiments, the within-group correlation coefficient between each effective fracturing parameter of the historical well and the liquid production volume in S102 above can be obtained in the following manner:

[0169] For each geological group, calculate the third correlation coefficient between each fracturing parameter and the liquid production volume within each geological group;

[0170] A significance test is performed on the third correlation coefficient between each fracturing parameter and the production volume. Based on the significance test results and the third correlation coefficient, effective fracturing parameters for each geological group are selected. The significance test results of the effective fracturing parameters reach the preset significance level and the absolute value of the third correlation coefficient is not less than the preset correlation coefficient threshold.

[0171] Fisher's Z-transform is performed on the fourth correlation coefficient between the effective fracturing parameters and the production of each geological group, and then the target value obtained by the transformation is inversely transformed to obtain the intra-group correlation coefficient between the effective fracturing parameters and the production of each historical well.

[0172] In some embodiments, the target fracturing parameter clustering model in S102 above is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and production rate of historical wells. In specific implementation, it may include:

[0173] The intra-group weight of each effective fracturing parameter is calculated based on the absolute value of the intra-group correlation coefficient between each effective fracturing parameter and the production of fluid in historical wells.

[0174] Based on the intra-group weight of each effective fracturing parameter, a second weighted Euclidean distance function is constructed for the target fracturing parameter clustering model; the second weighted Euclidean distance function is used to calculate the distance from the historical fracturing segment to the cluster center of the fracturing parameter.

[0175] Accordingly, the clustering model of target fracturing parameters within the group mentioned above in S102 clusters the effective fracturing parameters of historical fracturing sections and outputs the engineering sweet spot template corresponding to each geological group. In specific implementation, this can include:

[0176] Based on the cluster center of the initial fracturing parameter cluster and the second weighted Euclidean distance function, each historical fracturing segment is assigned to the nearest fracturing parameter cluster;

[0177] Based on the normalized effective fracturing parameters of all historical fracturing segments within each fracturing parameter cluster, the cluster center of the fracturing parameter cluster is recalculated.

[0178] The steps of historical fracturing segment allocation and cluster center recalculation are performed iteratively until the cluster center change between two adjacent iterations is less than the second preset convergence threshold.

[0179] After clustering convergence, the average fluid production of all historical fracturing segments within each fracturing parameter cluster is calculated. The cluster center of the fracturing parameter cluster with the highest average fluid production is determined as the engineering sweet spot template corresponding to the geological grouping.

[0180] Specifically, the calculation process for the intra-group correlation coefficient between each effective fracturing parameter and production of historical wells is the same as the calculation process for the inter-well correlation coefficient between each effective geological parameter and oil production of historical wells. Both use formulas (1)-(6) to screen p. l,j<0.1 and ρ l,j Fracturing parameters with an absolute value ≥ 0.2 are considered effective fracturing parameters, meaning the significance verification result of the effective fracturing parameters reaches the preset significance level and the absolute value of the third correlation coefficient is not less than the preset correlation coefficient threshold. In calculating the intra-group correlation coefficient, the oil production in the above formulas (1)-(6) can be adjusted to the fluid production, the first correlation coefficient can be adjusted to the third correlation coefficient, the second correlation coefficient can be adjusted to the fourth correlation coefficient, and the cross-well correlation coefficient can be adjusted to the intra-group correlation coefficient, etc. The specific calculation formulas are not listed here in this manual.

[0181] After obtaining the intragroup correlation coefficients between each effective fracturing parameter and the production rate of historical wells, the intragroup weight of each effective fracturing parameter can be calculated. The weight calculation formula is similar to formula (7), such as:

[0182] (16)

[0183] Where P is the total number of effective fracturing parameters selected; is the absolute value of the intragroup correlation coefficient between the p-th effective fracturing parameter and the fluid production of a historical well; w p Let be the intra-group weight of the p-th effective fracturing parameter, and the sum of the intra-group weights of all effective fracturing parameters is 1.

[0184] Then, based on the intra-group weights of each effective fracturing parameter mentioned above, the second weighted Euclidean distance function of the target fracturing parameter clustering model can be constructed according to the following formula:

[0185] (17)

[0186] in, From historical fracturing segment a to the center of the fracturing parameter cluster The weighted Euclidean distance is such that the smaller the distance, the more similar the parameter combination of the fracturing section is to the typical mode of the fracturing parameter cluster; Here are the normalized effective fracturing parameters for the a-th historical fracturing segment; This is the p-th normalized effective fracturing parameter in the a-th fracturing segment; Let p be the fracturing parameter in the b-th cluster.

[0187] After constructing the second weighted Euclidean distance function of the target fracturing parameter clustering model, the effective fracturing parameters of historical fracturing segments can be clustered using the in-group target fracturing parameter clustering model, and the engineering sweet spot template corresponding to each geological group can be output. The specific process is as follows:

[0188] 1) Initialization: From all historical fracturing segments in the current geological group, randomly select K samples with a cluster size of K as the initial cluster centers, denoted as μ1. (0)μ2 (0) ,...,μ K (0) μ1 (0) μ2 (0) ,...,μ K (0) The superscript (0) indicates the 0th iteration. The number of clusters K can be determined by combining the sum of squared errors (SSE) method with the elbow method.

[0189] 2) Weighted Euclidean distance allocation: The second weighted Euclidean distance can be calculated according to the cluster center of the fracturing parameter cluster after initialization and the second weighted Euclidean distance function, according to the above formula (17), and then each historical fracturing segment can be allocated to the nearest fracturing parameter cluster according to the second weighted Euclidean distance.

[0190] 3) Cluster center update: Based on the normalized effective fracturing parameters of all historical fracturing segments within each fracturing parameter cluster, the cluster center can be recalculated according to the following formula:

[0191] (18)

[0192] in, The cluster center of the recalculated fracturing parameter cluster; S b The set of fracturing segments assigned to the b-th cluster; Let set S b The number of fracturing sections in the middle; represents the normalized effective fracturing parameters for the a-th historical fracturing segment.

[0193] 4) Iteration: Iterate through the historical fracturing segment allocation and cluster center recalculation steps, i.e., steps 2)-3) above, until the cluster center change between two adjacent iterations is less than the second preset convergence threshold, or the maximum number of iterations is reached. At this point, the algorithm converges and the iteration stops.

[0194] 5) Results Analysis: After clustering convergence, the average production rate of all historical fracturing segments within each fracturing parameter cluster was calculated using the following formula. The cluster center of the fracturing parameter cluster with the highest average production rate was determined as the engineering sweet spot template corresponding to the geological grouping. Clusters with significantly higher production rates were defined as high-production templates. To ensure robustness, the parameter range of this template was limited to the interquartile range (25%–75%) of the target category distribution, thereby minimizing the impact of outliers and avoiding unrealistic parameter deviations. This high-production template (engineering sweet spot template) will subsequently serve as the basis for parameter recommendations and well completion optimization.

[0195] For example: Geological grouping focuses on the "High" geological group. Historical fracturing segments: This group contains M=5 historical fracturing segments (S1-S5). Effective fracturing parameters (P=2) are: CN (cluster number): normalized value, FV (fluid volume): normalized value. Intra-group weights (w) p ):w_ CN =0.7 (cluster number has a significant impact on liquid production), w_ FV =0.3 (Liquid volume has an impact on liquid production, but the impact is relatively small). The preset cluster number K=2, and the second preset convergence threshold is 0.01.

[0196] Assume the initial dataset is as follows: historical fracturing segment S1 has a CN (cluster number) of 0.9, an FV (fluid volume) of 0.2, and a production volume of 80; historical fracturing segment S2 has a CN (cluster number) of 0.8, an FV (fluid volume) of 0.9, and a production volume of 75; historical fracturing segment S3 has a CN (cluster number) of 0.7, an FV (fluid volume) of 0.7, and a production volume of 70; historical fracturing segment S4 has a CN (cluster number) of 0.2, an FV (fluid volume) of 0.2, and a production volume of 80; historical fracturing segment S5 has a CN (cluster number) of 0.1, an FV (fluid volume) of 0.9, and a production volume of 25.

[0197] Two fracturing segments are randomly selected as initial cluster centers, such as cluster center μ1 = parameters of historical fracturing segment S1 = [0.9, 0.2], and cluster center μ2 = parameters of historical fracturing segment S4 = [0.2, 0.8]. The distance from each segment to the two cluster centers is calculated and assigned according to the formula (17) above. If the distance from S1 to μ1 is less than the distance from S1 to μ2, then S1 is assigned to cluster 1. Finally, cluster 1 includes S1, S2, and S3, and cluster 2 includes S4 and S5. The new cluster center μ is recalculated according to the formula (18) above. new_1 : CN=(0.9+0.8+0.7) / 3=0.800, FV=(0.2+0.9+0.7) / 3=0.600, μ new_1 =[0.800, 0.600], similarly, μ is calculated. new_2 =[0.150, 0.850]. Since the cluster center μ1 changes from [0.9, 0.2] to [0.8, 0.6], which is a large change, while μ2 changes from [0.2, 0.8] to [0.15, 0.85], which is a small change, the iteration has not converged and continues.

[0198] Repeat the above process. Recalculate the distances and assign samples using the new cluster centers [0.8, 0.6] and [0.15, 0.85]. Assume that after the second iteration, the cluster affiliation of the samples no longer changes, and the cluster centers stabilize at: final cluster center 1 (μ1) = [0.8, 0.6]; final cluster center 2 (μ2) = [0.15, 0.85], final assignment: cluster 1 (C1): S1, S2, S3; cluster 2 (C2): S4, S5. At this point, the algorithm converges.

[0199] Then, the average liquid production of cluster 1 can be calculated as (80+75+70) / 3=75.0m³ / d; the average liquid production of cluster 2 is (30+25) / 2=27.5m³ / d. The average liquid production of cluster 1 is greater than that of cluster 2. Therefore, the final cluster core of cluster 1 [CN=0.8, FV=0.6] is determined as the engineering sweet spot template corresponding to the "High" geological group.

[0200] By incorporating geological grouping into fracturing parameter clustering, the interference of geological heterogeneity is effectively eliminated, the optimal fracturing parameter configuration under different geological conditions is determined, and a high-yield engineering sweet spot template that can be applied to the design of fracturing sections before fracturing is established, which can provide important engineering input for subsequent well completion optimization and decision support.

[0201] S103: The target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data.

[0202] In some embodiments, the process of dividing the target well into fracturing sections using the sweet spot probability distribution data in S103 above may, in specific implementation, include:

[0203] The sweet spot probability distribution data of the target well is processed by moving average to obtain the smoothed sweet spot probability;

[0204] Calculate the fuzziness index for each depth point of the target well; the fuzziness index is used to characterize the uncertainty of the geological category at each depth point;

[0205] Using the smoothed dessert probability and the fuzzy index together as input signals, the Planck energy minimization change point detection algorithm is used for preliminary segment boundary detection;

[0206] For the segment boundaries obtained from the preliminary segment boundary detection, engineering constraints of minimum and maximum segment lengths are applied. Excessively short segments are merged and excessively long segments are divided to obtain a fracturing segment division scheme. Based on the fracturing segment division scheme, fracturing segments are divided for the target well.

[0207] Specifically, to improve the adaptability of fracturing segments to reservoir property continuity and geological boundary changes, a dynamic segmentation method integrating geological sweet spot probability and category ambiguity is proposed. Unlike traditional geometric design or design based on hard clustering results, this method fully utilizes the high-quality sweet spot probability and information entropy signal to jointly characterize the sweet spot enrichment characteristics and boundary uncertainties along the wellbore, thereby enabling the simultaneous identification of geological transition zones and continuous sweet spot areas.

[0208] Specifically, since the sweet spot probability distribution data (continuous sweet spot probability curve) output by the above steps may contain noise and small fluctuations, directly using it for segmentation would result in too many and unreasonable boundaries. Therefore, the sweet spot probability distribution data of the target well can be processed by a moving average according to the following formula:

[0209] (19)

[0210] in, Let be the smoothed sweet spot probability at depth point z; W is the width of the sliding window (usually set to the number of sampling points corresponding to the minimum fracturing section length); p(t) is the original sweet spot probability at the t-th depth point within the window (i.e., the membership degree to the high-quality sweet spot cluster). This involves summing the dessert probabilities for all depths within the window from z to z+W-1.

[0211] Then, using the following formula (introducing the fuzziness index based on Shannon entropy), calculate the fuzziness index for each depth point of the target well (the fuzziness index is used to characterize the uncertainty of the geological category at each depth point):

[0212] (20)

[0213] in, is the fuzziness index at depth point z, with a value range of [0, 1]; C is the total number of clusters (geological categories) of effective geological parameters. Let be the window-averaged smoothed sweet spot probability of the k-th class calculated at depth point z; log is the natural logarithm. The sweet spot probability curve of the horizontal well section reflects the enrichment trend of reservoir sweet spots, while the entropy curve highlights the fuzzy region of the class boundary. The combination of the two curves can more comprehensively characterize the spatial variability of reservoir properties.

[0214] Then, the smoothed dessert probability (smoothed dessert probability curve) and the fuzziness index (fuzziness index curve) are used together as input signals, and the Planck energy minimization change point detection algorithm is used for preliminary segment boundary detection:

[0215] (twenty one)

[0216] (twenty two)

[0217] in, To select the optimal set of change points { To minimize the objective function; } represents the set of locations of the points of change (segment boundaries). and These are the top and bottom of the wellbore. K is the number of points of change; Cost(.) is the cost function within a segment, measuring the degree of heterogeneity within a fracturing segment; This is a penalty coefficient used to control the coarseness of the segments; the larger it is, the more the model tends to divide the data into fewer segments. Let be the starting and ending depths of the k-th fracturing segment; To balance the weights (0≤ ≤ 1), used to adjust the relative importance of dessert probability and ambiguity index in the cost; for Inside The average value; for Inside The average value; This represents the local residual of the probability of a sweet spot; a larger value indicates a greater difference between that point and the average geological quality of the segment. This represents the local residual of the fuzziness index; a larger value indicates a greater difference between the point and the average uncertainty of the segment.

[0218] Next, engineering constraints for minimum and maximum segment lengths are applied to the segment boundaries obtained from the initial segment boundary detection. The specific constraints are as follows:

[0219] (twenty three)

[0220] Among them, L min L is the minimum segment length required by the project. max This is the maximum segment length required by the project.

[0221] Subsequently, under engineering constraints, excessively short segments can be merged and excessively long segments can be divided to obtain a fracturing segment division scheme, and fracturing segments can be divided for target wells based on this fracturing segment division scheme.

[0222] By introducing a dynamic fracturing segmentation method based on "sweet spot probability" and category fuzziness, the segmentation can reflect reservoir continuity and geological boundary changes.

[0223] In some embodiments, the process of matching initial fracturing parameters from historical fracturing sections based on a high-quality sweet spot ratio in each fracturing section in S103 above may, in specific implementation, include:

[0224] Based on the proportion of high-quality sweet spots and the length of each fractured section in the target well, a feature vector for each fractured section is constructed.

[0225] Calculate the Euclidean distance between the feature vector and the corresponding feature vector of each historical fracturing segment;

[0226] The fracturing parameters used in historical fracturing sections where the Euclidean distance from the target well fracturing section is less than a preset distance threshold are used as the initial fracturing parameters for the target well fracturing section.

[0227] Specifically, the proportion of high-quality sweet spots can be calculated according to the above formula (14), and then the feature vector of each fractured section can be constructed based on the proportion of high-quality sweet spots and the section length of each fractured section of the target well:

[0228] f=[R sweet ,L stage ], where f is the feature vector of the fracturing section; R sweet For a high-quality dessert ratio; L stage This represents the length of the fracturing section.

[0229] Then, the Euclidean distance between the eigenvectors and the corresponding eigenvectors of each historical fracturing segment is calculated using the following formula:

[0230] (twenty four)

[0231] in, The distance between the target segment and the feature vector of the h-th historical fracturing segment is: The proportion of high-quality desserts in the target segment; The length of the fracturing section in the target segment; The proportion of high-quality desserts in the h-th historical segment; is the length of the fracturing segment in the h-th historical segment.

[0232] Then, the Euclidean distance between the target well fractured segment and the h-th historical fractured segment (i.e., the Euclidean distance between the target segment and the eigenvector of the h-th historical fractured segment) can be calculated. The fracturing parameters used in historical fracturing sections with distances less than a preset threshold are used as the initial fracturing parameters for the target well fracturing section.

[0233] For example, the target well has completed dynamic segmentation, and three fracturing sections require fracturing parameter design. The historical well database contains fracturing section data for four historically operational wells, including their characteristics and the fracturing parameters actually used. Assuming the target well is segmented and the calculated R... sweet as follows:

[0234] Target segment T1: Segment length L stage =30m, there are 24 points within the segment with a probability of sweetness greater than 0.5, and a total of 30 points, R sweetT1=24 / 30=0.8; Target segment T2: Segment length L stage =25m, there are 10 points within the segment with a probability greater than 0.5 for the sweet spot, and a total of 25 points. R sweetT2 =10 / 25=0.4; Target segment T3: Segment length L stage =35m, there are 7 points within the segment with a probability of sweetness greater than 0.5, and a total of 35 points, R sweetT3 =7 / 35=0.2. The eigenvectors constituting the target well are: f T1 =[R sweetT1 =0.8,L stage =30], f T2 =[R sweetT2 =0.4,L stage =25], f T3 =[R sweetT3 =0.2,L stage =35].

[0235] Assume the historical fracturing database consists of historical fracturing segments H1, H2, and H3, with corresponding feature vectors [R]. sweet ,L stage =[0.85,28], [0.75,32], [0.45,26], the corresponding actual fracturing parameters [CN,FV] are [6 clusters, 1200m³], [5 clusters, 1100m³], [4 clusters, 900m³].

[0236] Taking target segment T1 as an example, calculate its Euclidean distance from the historical fracturing segments H1, H2, and H3. Find the smallest Euclidean distance. If the Euclidean distance between T1 and H1 is the smallest, then the fracturing parameters of the historical segment H1, which has the smallest distance from target segment T1, are transferred to T1. That is, the initial fracturing parameters of the target segment of the target well (i.e., the target well fracturing segment) are [6 clusters, 1200m³]. Perform the same operation on all target segments to finally form the initial fracturing parameter scheme of the target well, such as: segment T1 (high-quality segment): [CN=6 clusters, FV=1200m³]; segment T2 (medium-quality segment): [CN=4 clusters, FV=850m³]; segment T3 (low-quality segment): [CN=3 clusters, FV=750m³].

[0237] By constructing an intelligent initialization process of "feature vector-similarity matching-parameter migration", a high-quality and highly reliable initial fracturing parameter optimization scheme is provided for the entire fracturing parameter optimization process.

[0238] S104: Query the engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs, optimize the initial fracturing parameters, and obtain the target fracturing parameters for the target well fracturing section.

[0239] In some embodiments, the engineering sweet spot template corresponding to the geological group to which the fracturing segment of the historical well belongs in S104 above optimizes the initial fracturing parameters and outputs the target fracturing parameters for the target well fracturing segment. In specific implementation, this may include:

[0240] Query the engineering sweet spot templates corresponding to the geological groupings, use the engineering sweet spot template library of historical wells as the training set, train the CART decision tree model, and generate decision rules for determining whether the fracturing parameter combination is an engineering sweet spot;

[0241] The initial fracturing parameters are input into the trained CART decision tree model for discrimination.

[0242] If the determination result is a non-engineering sweet spot, the initial fracturing parameters are adjusted according to the decision rules until the determination result is an engineering sweet spot, and the adjusted initial fracturing parameters are used as the final optimized target fracturing parameters.

[0243] Specifically, you can query the engineering sweet spot template corresponding to a geological group and retrieve the typical fracturing parameter combination that has been verified as the most productive under that group. For example, the template parameter vector ["High"] = [CN=7, FV=1100]. Then, using the engineering sweet spot templates of historical wells as positive samples and other historical parameter combinations of "non-engineering sweet spots" as negative samples, a training set can be constructed to train a CART decision tree classification model. This model can automatically learn and extract clear rules to distinguish between "engineering sweet spots" and "non-engineering sweet spots." The initial fracturing parameters are then input into the trained CART decision tree model, and automatic discrimination is performed according to the rule logic. If the discrimination result is a non-engineering sweet spot, the relevant parameters are adjusted according to the specific rule clauses triggered, and then the discrimination is repeated until the engineering sweet spot criteria are met.

[0244] Among them, the decision tree described above is a typical white-box supervised learning method that can decompose the complex classification process into a transparent "if-then" logical structure defined by features and thresholds, making it highly interpretable and practical for engineering applications. Its basic principle is to recursively partition the feature space until each sample falls into the leaf node with the highest purity. Compared to black-box ensemble methods (such as random forests and GBDT), a single CART tree has a simple structure, and its nodes and rules are clear and easy to understand.

[0245] For example, to optimize target segment T1 (belonging to the "High" geological group), its initial fracturing parameters are [CN=6 cluster, FV=1200m³]. First, the engineering sweet spot template for the "High" geological group is found to be [CN=7 cluster, FV=1100m³]. A ​​CART decision tree is trained using the engineering sweet spot template library to generate decision rules. Assume the generated core rules are as follows:

[0246] Rule 1: IF CN < 7 THEN Non-engineered dessert

[0247] Rule 2: IF CN>=7 AND FV<1000 THEN Non-engineered dessert

[0248] Rule 3: IF CN>=7 AND FV>=1000 THEN (Engineering Dessert)

[0249] Inputting the initial fracturing parameters [CN=6, FV=1200] of T1 into the decision tree: First, rule 1 is encountered: CN=6 < 7, the condition is true. It is immediately judged as a "non-engineering sweet spot". According to the triggered rule 1, the optimization direction is: the cluster number (CN) must be increased to 7 or more. Inputting the optimized parameters [CN=7, FV=1200] into the decision tree again: Rule 1: CN=7>=7, the condition is false, proceed to the next rule. Rule 2: CN>=7 is true, but FV=1200>=1000, the condition is false, so rule 2 is not triggered. Rule 3: CN>=7 is true, and FV>=1000 is true, therefore, it is judged as an "engineering sweet spot". Finally, the final optimized fracturing parameters of the target segment T1 can be output as: [CN=7 clusters, FV=1200m³].

[0250] In some embodiments, training a CART decision tree model using a library of engineering sweet spot templates from historical wells as a training set may include:

[0251] Using the Gini coefficient as the node splitting criterion, a decision tree is constructed through a recursive partitioning method, where:

[0252] Calculate the weighted Gini coefficient for each potential split point;

[0253] Select the feature and threshold that minimize the weighted Gini coefficient as the split point;

[0254] The above process is repeated recursively at each split point until the stopping condition is met.

[0255] In some embodiments, the specific optimization steps for fracturing parameters under different geological conditions are as follows:

[0256] 1) Decision Tree Construction: CART decision tree classification models were developed for fracturing parameters and their categories under different geological conditions. These models established a mapping between parameter features and production levels, laying the foundation for subsequent rule extraction.

[0257] 2) Threshold extraction: The trained decision tree is used to extract the "if-then" rule path leading to high yield level, and the key parameter thresholds are determined to form interpretable design rules.

[0258] 3) Parameter Optimization: The high-yield rule range is used as the primary constraint; the initialization scheme first incorporates these ranges. Based on this, targeted fine-tuning is performed using decision tree thresholds to ensure parameters align with the high-yield path. When a conflict arises between the rule range and the threshold, the rule range is prioritized based on engineering feasibility (e.g., 2-8 clusters per segment; cluster spacing greater than or equal to 5 meters), and the parameters are moved towards the nearest threshold according to the principle of minimum change.

[0259] 4) Result Validation: The optimized parameter scheme was re-evaluated using a decision tree model to validate its category assignment. The number of fracturing segments that successfully transitioned from a non-high-yield category to a high-yield category was recorded as a measure of optimization performance and the effectiveness of the proposed method.

[0260] Through the above solution, the present invention can achieve the following beneficial effects:

[0261] A more scientific approach to sweet spot evaluation: The proposed mechanism-guided clustering model for target geological sweet spots is based on the cross-well correlation coefficients between effective geological parameters and oil production of historical wells. It calculates the cross-well weight of each effective geological parameter and constructs a clustering model based on these cross-well weights. This approach breaks through the limitations of traditional empirical threshold methods and equal-weight clustering methods, enabling sweet spot evaluation to not only reflect nonlinear coupling relationships but also quantitatively reveal the fuzzy transition characteristics between high-quality and low-quality reservoirs. This significantly improves the accuracy and geological rationality of sweet spot identification.

[0262] The engineering model is more representative: Under the constraint of geological sweet spot grouping, weighted clustering is used to perform pattern recognition on fracturing parameters, eliminating the interference of geological heterogeneity and obtaining typical high-yield engineering templates under different geological conditions. Compared with the method of directly performing statistical analysis in the global parameter space, this model can better reveal the completion rules of zonal and stratified wells, providing more reliable engineering guidance for personalized fracturing.

[0263] More intelligent segmentation design: In the division of fracturing sections, a dynamic segmentation method based on sweet spot probability and ambiguity is introduced. Combined with PELT change point detection and engineering constraints, the segment boundary can adaptively identify the "continuous sweet spot area" and accurately capture the "transition zone". This avoids the problem of intra-segment imbalance caused by traditional equal-length or experience-based segmentation, and improves the rationality and automation of segmentation design.

[0264] Cluster parameter optimization is more interpretable: A two-stage fracturing strategy combining spatial similarity transfer and decision tree optimization is proposed. Based on historical experience, interpretable high-yield rules are extracted to optimize key parameters such as cluster number, spacing, and liquid sand volume. Unlike black-box algorithms, this method outputs clear "if-then" logical rules, ensuring that the optimized scheme conforms to both geological conditions and engineering feasibility, thereby improving the transparency and practicality of cluster design.

[0265] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. For details, please refer to the foregoing descriptions of the relevant processing embodiments; they will not be repeated here.

[0266] The foregoing description of this method is for illustrative purposes only and describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0267] Although this specification provides the following examples or appendices Figure 2 The method or apparatus structure shown may include more or fewer combined operational steps or module units based on conventional or non-creative labor. In steps or structures where there is no logically necessary causal relationship, the execution order of these steps or the module structure of the apparatus is not limited to the execution order or module structure shown in the embodiments or drawings of this specification. When the method or module structure is applied in actual devices, servers, or terminal products, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or drawings (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed processing or server cluster implementation environment). Based on the above-described data-driven interpretable fracturing parameter optimization method, this specification also proposes an embodiment of a data-driven interpretable fracturing parameter optimization apparatus. Figure 2 As shown, the device may specifically include the following modules:

[0268] The first clustering processing module 201 can be used to cluster the effective geological parameters of the target well along the well depth using the target geological sweet spot clustering model, and output the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between the effective geological parameters of each historical well and the oil production.

[0269] The second clustering processing module 202 can be used to geologically group the historical fracturing sections of historical wells based on the sweet spot probability distribution data. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and the fluid production of the historical well.

[0270] The initial fracturing parameter matching module 203 can be used to divide the target well into fracturing segments using the sweet spot probability distribution data. Within each fracturing segment, initial fracturing parameters are matched from historical fracturing segments based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data.

[0271] The initial fracturing parameter optimization module 204 can be used to query the engineering sweet spot template corresponding to the geological group to which the fracturing section of the historical well belongs, optimize the initial fracturing parameters, and output the target fracturing parameters of the target well fracturing section.

[0272] In some embodiments, the cross-well correlation coefficients between effective geological parameters and oil production of each historical well in the first clustering processing module 201 are obtained as follows: For each historical well, a first correlation coefficient between each geological parameter and oil production of each historical well is calculated; a significance test is performed on the first correlation coefficient between each geological parameter and oil production; based on the significance test result and the first correlation coefficient, effective geological parameters of each historical well are selected from each geological parameter of each historical well; the significance test result of the effective geological parameters reaches a preset significance level and the absolute value of the first correlation coefficient is not less than a preset correlation coefficient threshold; a Fisher Z-transform is performed on the second correlation coefficient between each effective geological parameter and oil production of each historical well, and then an inverse transform is performed on the target value obtained by the transformation to obtain the cross-well correlation coefficients between each effective geological parameter and oil production of the historical well.

[0273] In some embodiments, the first clustering processing module 201 described above can be specifically used to calculate the cross-well weight of each effective geological parameter based on the absolute value of the cross-well correlation coefficient between each effective geological parameter and oil production of historical wells; construct a first weighted Euclidean distance function for the target geological sweet spot clustering model based on the cross-well weight of each effective geological parameter; the first weighted Euclidean distance function is used to calculate the distance from each depth point to the center of each cluster; initialize the membership matrix of the normalized effective geological parameters of each depth point to each cluster, wherein the membership matrix satisfies that the sum of the membership degrees of each depth point to each cluster is 1; calculate the center of each cluster based on the initialized membership matrix and the normalized effective geological parameters of each depth point; and recalculate the center of each cluster based on the center of each cluster and the first weighted Euclidean distance function. The normalized effective primality parameters of each depth point are used to determine the membership degree of each cluster, forming an updated membership matrix. The difference between the updated membership matrix and the original membership matrix is ​​determined to be less than a first preset convergence threshold. If so, the clustering is considered converged, and the updated membership matrix is ​​used as the final target membership matrix. Based on the final target cluster centers and the cross-well weights of each effective primality parameter, a comprehensive score for the target cluster center of each cluster is determined. The target cluster centers are calculated based on the target membership matrix and the normalized effective primality parameters of each depth point. The comprehensive scores of each target cluster center are compared, and clusters with comprehensive scores greater than a preset score threshold are designated as high-quality sweet spot clusters. The membership degree of each depth point in the target membership matrix to these high-quality sweet spot clusters is used as the sweet spot probability for each depth point, forming the sweet spot probability distribution data.

[0274] In some embodiments, the first clustering processing module 201 described above can also be used to determine the comprehensive score of the target cluster centers according to the following formula:

[0275]

[0276] Among them, S k S is the comprehensive score of the target cluster center of the k-th cluster; j S represents the relevant direction of the j-th geological parameter. j Take -1 or +1; w j c is the cross-well weight of the j-th effective quality parameter; kj Let be the j-th geological parameter of the k-th cluster; D is the total number of effective geological parameters.

[0277] In some embodiments, the second clustering processing module 202 described above can be specifically used to obtain the sweet spot probability of all depth points of each historical fracturing segment from the sweet spot probability distribution data, calculate the proportion of high-quality sweet spots for each historical fracturing segment based on the sweet spot probability, determine the quartiles of the proportion of high-quality sweet spots for the historical fracturing segment, and divide the historical fracturing segments of all historical wells into four geological groups: high, medium-high, medium-low, and low, based on the quartiles.

[0278] In some embodiments, the second clustering processing module 202 described above can also be used to calculate the intra-group weight of each effective fracturing parameter based on the absolute value of the intra-group correlation coefficient between each effective fracturing parameter and the production of fluid in historical wells.

[0279] Based on the intra-group weights of each effective fracturing parameter, a second weighted Euclidean distance function is constructed for the target fracturing parameter clustering model. The second weighted Euclidean distance function is used to calculate the distance from the historical fracturing segment to the cluster center of the fracturing parameter. Based on the cluster center of the initialized fracturing parameter cluster and the second weighted Euclidean distance function, each historical fracturing segment is assigned to the nearest fracturing parameter cluster. Based on the normalized effective fracturing parameters of all historical fracturing segments within each fracturing parameter cluster, the cluster center of the fracturing parameter cluster is recalculated. The steps of historical fracturing segment assignment and cluster center recalculation are iteratively executed until the change in cluster center between two adjacent iterations is less than a second preset convergence threshold. After clustering convergence, the average fluid production of all historical fracturing segments within each fracturing parameter cluster is calculated, and the cluster center of the fracturing parameter cluster with the highest average fluid production is determined as the engineering sweet spot template corresponding to the geological grouping.

[0280] In some embodiments, the initial fracturing parameter matching module 203 described above can be used to perform moving average processing on the sweet spot probability distribution data of the target well to obtain a smoothed sweet spot probability; calculate the fuzziness index of each depth point of the target well; the fuzziness index is used to characterize the uncertainty of the geological category at each depth point; use the smoothed sweet spot probability and the fuzziness index together as input signals, and use the Planck energy minimization change point detection algorithm to perform preliminary segment boundary detection; apply minimum segment length and maximum segment length engineering constraints to the segment boundaries obtained by the preliminary segment boundary detection, merge excessively short segments and divide excessively long segments to obtain a fracturing segment division scheme, and divide the target well into fracturing segments based on the fracturing segment division scheme.

[0281] In some embodiments, the initial fracturing parameter matching module 203 can also be used to construct a feature vector for each fracturing segment based on the proportion of high-quality sweet spots and the segment length of each fracturing segment in the target well; calculate the Euclidean distance between the feature vector and the corresponding feature vector of each historical fracturing segment; and use the fracturing parameters adopted by the historical fracturing segments whose Euclidean distance from the target well fracturing segment is less than a preset distance threshold as the initial fracturing parameters of the target well fracturing segment.

[0282] In some embodiments, the initial fracturing parameter optimization module 204 can be specifically used to query the engineering sweet spot template corresponding to the geological group, use the engineering sweet spot template library of historical wells as the training set to train a CART decision tree model, generate decision rules for determining whether the combination of fracturing parameters is an engineering sweet spot, input the initial fracturing parameters into the trained CART decision tree model for determination, and if the determination result is not an engineering sweet spot, adjust the initial fracturing parameters according to the decision rules until the determination result is an engineering sweet spot, and use the adjusted initial fracturing parameters as the final target fracturing parameters for optimization.

[0283] As can be seen from the above, the data-driven interpretable fracturing parameter optimization device provided in the embodiments of this specification can provide a scientific, transparent, and practical fracturing parameter optimization solution.

[0284] This specification also provides an electronic device based on the aforementioned data-driven interpretable fracturing parameter optimization method, including a processor and a memory for storing processor-executable programs / instructions. Specifically, the processor can execute the following steps according to the program / instructions: using a target geological sweet spot clustering model, clustering the effective geological parameters along the well depth of the target well, and outputting the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficients between the effective geological parameters and oil production of each historical well; based on the sweet spot probability distribution data, geologically grouping the historical fracturing sections of the historical wells, and within each geological group, using... The effective fracturing parameters of historical fracturing sections are clustered using an intra-group target fracturing parameter clustering model, outputting an engineering sweet spot template for each geological group. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and production volume of historical wells. The target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. The engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs is queried, and the initial fracturing parameters are optimized to obtain the target fracturing parameters for the target well fracturing section.

[0285] To execute the above instructions more accurately, please refer to... Figure 3 As shown in the embodiments of this specification, another specific electronic device is also provided, wherein the electronic device includes a network communication port 301, a processor 302, and a memory 303. The above structures are connected by internal cables so that the various structures can perform specific data interaction.

[0286] Specifically, the processor 302 can be used to cluster the effective geological parameters of the target well along the well depth using a target geological sweet spot clustering model, and output the sweet spot probability distribution data of the target well. The target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between each effective geological parameter of historical wells and the oil production. Based on the sweet spot probability distribution data, the historical fracturing sections of historical wells are geologically grouped. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter of historical wells and the fluid production. The target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from the historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. The engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs is queried, and the initial fracturing parameters are optimized to obtain the target fracturing parameters of the target well fracturing section.

[0287] The memory 303 can be used to store the corresponding instruction program.

[0288] In this embodiment, the network communication port 301 can be a virtual port bound to different communication protocols, thereby enabling the sending or receiving of different data. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. Furthermore, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM or CDMA; it can also be a Wi-Fi chip; or it can be a Bluetooth chip.

[0289] In this embodiment, the processor 302 can be implemented in any suitable manner. For example, the processor can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc. This specification is not limiting.

[0290] In this embodiment, the memory 303 may include multiple layers. In a digital system, anything that can store binary data can be a memory. In an integrated circuit, a circuit with storage function but no physical form is also called a memory, such as RAM, FIFO, etc. In a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0291] This specification also provides a computer storage medium based on the aforementioned data-driven interpretable fracturing parameter optimization method. The computer storage medium stores a computer program / instruction that, when executed, performs the following: using a target geological sweet spot clustering model, clustering the effective geological parameters along the well depth of the target well, and outputting the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficients between the effective geological parameters and oil production of each historical well; based on the sweet spot probability distribution data, geologically grouping the historical fracturing sections of historical wells, and within each geological group, using the target... A target fracturing parameter clustering model is used to cluster the effective fracturing parameters of historical fracturing sections, outputting an engineering sweet spot template for each geological group. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and production volume of historical wells. The target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. The engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs is queried to optimize the initial fracturing parameters, thereby obtaining the target fracturing parameters for the target well fracturing section.

[0292] In this embodiment, the storage medium includes, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Cache, Hard Disk Drive (HDD), or Memory Card. The memory can be used to store computer program instructions. The network communication unit can be an interface configured according to standards specified in the communication protocol for network connection communication.

[0293] In this embodiment, the specific functions and effects implemented by the program instructions stored in the computer storage medium can be explained in comparison with other implementation methods, and will not be repeated here.

[0294] While this specification provides the steps of operation for the methods described in the embodiments or flowcharts, more or fewer steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible order of execution among many steps and does not represent the only possible order. In actual device or client product execution, the methods shown in the embodiments or drawings may be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded. The terms "first," "second," etc., are used to denote names and do not indicate any particular order.

[0295] Those skilled in the art will also know that, besides implementing the controller using purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller function as logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices within it used to implement various functions can also be considered structures within that hardware component. Alternatively, the devices used to implement various functions can be considered as both software modules implementing the method and structures within a hardware component.

[0296] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0297] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this specification can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions of this specification can essentially be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments of this specification.

[0298] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. This specification can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0299] Although this specification has been described by way of examples, those skilled in the art will recognize that many variations of this specification are possible without departing from its spirit, and it is intended that the appended claims cover such variations without departing from the spirit of this specification.

Claims

1. A data-driven, interpretable fracturing parameter optimization method, characterized in that, include: Using the target geological sweet spot clustering model, the effective geological parameters along the well depth of the target well are clustered, and the sweet spot probability distribution data of the target well is output. The target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between the effective geological parameters of each historical well and the oil production. Based on the sweet spot probability distribution data, the historical fracturing sections of historical wells are geologically grouped. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and the fluid production of historical wells. The target well is divided into fracturing sections using the sweet spot probability distribution data. Within each fracturing section, initial fracturing parameters are matched from historical fracturing sections based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. Query the engineering sweet spot template corresponding to the geological group to which the historical fracturing section belongs, optimize the initial fracturing parameters, and obtain the target fracturing parameters for the target well fracturing section.

2. The method according to claim 1, characterized in that, The cross-well correlation coefficients between the effective geological parameters of each historical well and the oil production were obtained in the following manner: For each historical well, calculate the first correlation coefficient between each geological parameter and the oil production of each historical well; A significance test is performed on the first correlation coefficient between each geological parameter and oil production. Based on the significance test results and the first correlation coefficient, effective geological parameters of each historical well are selected from each geological parameter of each historical well. The significance test results of the effective geological parameters reach the preset significance level and the absolute value of the first correlation coefficient is not less than the preset correlation coefficient threshold. Fisher Z-transform is performed on the second correlation coefficient between each effective geological parameter and oil production of each historical well, and then the target value obtained by the transformation is inversely transformed to obtain the cross-well correlation coefficient between each effective geological parameter and oil production of the historical well.

3. The method according to claim 1, characterized in that, The target geological sweet spot clustering model is constructed based on the cross-well correlation coefficients between effective geological parameters and oil production of historical wells, including: The cross-well weight of each effective geological parameter is calculated based on the absolute value of the cross-well correlation coefficient between each effective geological parameter and oil production in historical wells. Based on the cross-well weight of each effective geological parameter, a first weighted Euclidean distance function is constructed for the target geological sweet spot clustering model; the first weighted Euclidean distance function is used to calculate the distance from each depth point to the center of each cluster. Accordingly, the method of using the target geological sweet spot clustering model to cluster the effective geological parameters along the well depth of the target well and output the sweet spot probability distribution data of the target well includes: Initialize the normalized effective quality parameters of each depth point to the membership matrix of each cluster, wherein the membership matrix satisfies that the sum of the membership degrees of each depth point to each cluster is 1; Based on the initialized membership matrix and the normalized effective quality parameters of each depth point, the cluster center of each cluster is calculated; Based on the cluster centers of each cluster and the first weighted Euclidean distance function, the membership degree of the normalized effective quality parameters of each depth point to each cluster is recalculated to form an updated membership matrix; Determine whether the difference between the updated membership matrix and the membership matrix is ​​less than the first preset convergence threshold; If so, the clustering is determined to have converged. The updated membership matrix is ​​used as the final target membership matrix. Based on the final target cluster center and the cross-well weights of each effective genomic parameter, the comprehensive score of the target cluster center of each cluster is determined. The target cluster center is calculated based on the target membership matrix and the normalized effective genomic parameters of each depth point. By comparing the comprehensive scores of each target cluster center, clusters with comprehensive scores greater than a preset score threshold are identified as high-quality sweet spot clusters. The membership degree of each depth point in the target membership matrix to the high-quality sweet spot clusters is taken as the sweet spot probability of each depth point, thus forming the sweet spot probability distribution data.

4. The method according to claim 3, characterized in that, The comprehensive score for the target cluster center of each cluster is determined based on the final target cluster center and the cross-well weights of each effective quality parameter, including: The overall score of the target cluster center is determined according to the following formula: where S k is the overall score of the target cluster center of the kth cluster; S j is the relevant direction of the jth geological parameter, S j takes -1 or +1; w j is the cross-well weight of the jth effective geological parameter; c kj is the jth geological parameter of the kth cluster; D is the total number of effective geological parameters.

5. The method according to claim 1, characterized in that, The geological grouping of the historical fracturing sections of the historical wells based on the sweet spot probability distribution data includes: From the sweet spot probability distribution data, the sweet spot probability of all depth points of each historical fracturing segment is obtained, and the proportion of high-quality sweet spots in each historical fracturing segment is calculated based on the sweet spot probability. The quartiles of the proportion of high-quality sweet spots in historical fracturing sections were determined, and based on these quartiles, the historical fracturing sections of all historical wells were divided into four geological groups: high, medium-high, medium-low, and low.

6. The method according to claim 1, characterized in that, The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficients between each effective fracturing parameter and production rate in historical wells, including: The intra-group weight of each effective fracturing parameter is calculated based on the absolute value of the intra-group correlation coefficient between each effective fracturing parameter and the production of fluid in historical wells. Based on the intra-group weight of each effective fracturing parameter, a second weighted Euclidean distance function is constructed for the target fracturing parameter clustering model; the second weighted Euclidean distance function is used to calculate the distance from the historical fracturing segment to the cluster center of the fracturing parameter. Accordingly, the effective fracturing parameters of historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output, including: Based on the cluster center of the initial fracturing parameter cluster and the second weighted Euclidean distance function, each historical fracturing segment is assigned to the nearest fracturing parameter cluster; Based on the normalized effective fracturing parameters of all historical fracturing segments within each fracturing parameter cluster, the cluster center of the fracturing parameter cluster is recalculated. The steps of historical fracturing segment allocation and cluster center recalculation are performed iteratively until the cluster center change between two adjacent iterations is less than the second preset convergence threshold. After clustering convergence, the average fluid production of all historical fracturing segments within each fracturing parameter cluster is calculated. The cluster center of the fracturing parameter cluster with the highest average fluid production is determined as the engineering sweet spot template corresponding to the geological grouping.

7. The method according to claim 1, characterized in that, The process of dividing the target well into fracturing sections using the sweet spot probability distribution data includes: The sweet spot probability distribution data of the target well is processed by moving average to obtain the smoothed sweet spot probability; Calculate the fuzziness index for each depth point of the target well; the fuzziness index is used to characterize the uncertainty of the geological category at each depth point; Using the smoothed dessert probability and the fuzzy index together as input signals, the Planck energy minimization change point detection algorithm is used for preliminary segment boundary detection; For the segment boundaries obtained from the preliminary segment boundary detection, engineering constraints of minimum and maximum segment lengths are applied. Excessively short segments are merged and excessively long segments are divided to obtain a fracturing segment division scheme. Based on the fracturing segment division scheme, fracturing segments are divided for the target well.

8. The method according to claim 1 or 7, characterized in that, Within each fracturing segment, initial fracturing parameters are matched from historical fracturing segments based on a high-quality sweet spot ratio, including: Based on the proportion of high-quality sweet spots and the length of each fractured section in the target well, a feature vector is constructed for each fractured section. Calculate the Euclidean distance between the feature vector and the corresponding feature vector of each historical fracturing segment; The fracturing parameters used in historical fracturing sections where the Euclidean distance from the target well fracturing section is less than a preset distance threshold are used as the initial fracturing parameters for the target well fracturing section.

9. The method according to claim 1, characterized in that, The engineering sweet spot template corresponding to the geological group to which the fractured section of the historical well belongs is used to optimize the initial fracturing parameters and output the target fracturing parameters for the target well's fracturing section, including: Query the engineering sweet spot templates corresponding to the geological groupings, use the engineering sweet spot template library of historical wells as the training set, train the CART decision tree model, and generate decision rules for determining whether the fracturing parameter combination is an engineering sweet spot; The initial fracturing parameters are input into the trained CART decision tree model for discrimination. If the determination result is a non-engineering sweet spot, the initial fracturing parameters are adjusted according to the decision rules until the determination result is an engineering sweet spot, and the adjusted initial fracturing parameters are used as the final optimized target fracturing parameters.

10. A data-driven, interpretable fracturing parameter optimization device, characterized in that, include: The first clustering processing module is used to cluster the effective geological parameters of the target well along the well depth using the target geological sweet spot clustering model, and output the sweet spot probability distribution data of the target well; the target geological sweet spot clustering model is constructed based on the cross-well correlation coefficient between the effective geological parameters of each historical well and the oil production. The second clustering module is used to geologically group the historical fracturing sections of historical wells based on the sweet spot probability distribution data. Within each geological group, the effective fracturing parameters of the historical fracturing sections are clustered using the target fracturing parameter clustering model within the group, and the engineering sweet spot template corresponding to each geological group is output. The target fracturing parameter clustering model is constructed based on the intra-group correlation coefficient between each effective fracturing parameter and the production volume of historical wells. The initial fracturing parameter matching module is used to divide the target well into fracturing segments using the sweet spot probability distribution data. Within each fracturing segment, initial fracturing parameters are matched from historical fracturing segments based on the proportion of high-quality sweet spots. The proportion of high-quality sweet spots is determined based on the sweet spot probability distribution data. The initial fracturing parameter optimization module is used to query the engineering sweet spot template corresponding to the geological group to which the fracturing section of the historical well belongs, optimize the initial fracturing parameters, and output the target fracturing parameters of the target well fracturing section.

Images