A method and device for intelligently predicting the deliverability of a multi-layer commingled oil well after fracturing

Abstract

The application discloses a kind of multilayer commingled production well pressure production capacity intelligent prediction method and device, comprising: according to the geologic unit of target well, corresponding production capacity prediction integrated model is called;Wherein, production capacity prediction integrated model is obtained by training in the following mode: for different geologic units in target reservoir area, the training data set corresponding to geologic unit is obtained respectively, training data set is constituted by the historical capacity main control factor and historical production data of each reservoir section of each historical well in geologic unit;Based on training data set, train corresponding production capacity prediction integrated model for each geologic unit.The input feature vector of each reservoir section of target well is input to corresponding production capacity prediction integrated model, and the production capacity prediction value corresponding to each reservoir section is output, and input feature vector includes the geologic parameter and fracturing construction parameter of target well.The present application can be aimed at different geologic units, realize the accurate, fast prediction of the production capacity of each reservoir section of multilayer commingled production well.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development technology, and in particular to a method and device for intelligent prediction of post-pressure production capacity of multi-layered syngas wells. Background Technology

[0002] In oil and gas field development, especially for unconventional reservoirs such as low-permeability and tight reservoirs, hydraulic fracturing is a key technology for improving single-well productivity. Accurate prediction of post-fracturing productivity is a crucial prerequisite for optimizing fracturing parameter design, formulating scientific extraction strategies, and evaluating development benefits. It directly impacts the development efficiency and economic benefits of multi-layer synergistic production in low-permeability reservoirs, and plays a vital supporting role in promoting the transformation of oil and gas development towards "intelligent, refined, and efficient" methods.

[0003] However, existing post-fracturing production prediction technologies for multi-layered syngas wells have significant limitations: on the one hand, they do not consider the differences in geological characteristics of different geological units and use a unified model to predict the production of wells in the entire region, resulting in poor model adaptability to specific geological conditions and limited prediction accuracy; on the other hand, existing models mostly use the total production of the well as the prediction target, and cannot accurately and quickly output the independent production prediction values ​​of each reservoir section, thus restricting the scientific nature of stratified fracturing optimization and development decisions.

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

[0005] This specification provides an intelligent prediction method and device for post-pressure production capacity of multi-layer syngas wells to solve the problem of poor predictive adaptability of existing technologies and the inability to accurately and quickly output production capacity prediction values ​​for each reservoir segment.

[0006] Firstly, embodiments of this specification provide a method for intelligent prediction of post-pressure production capacity in multi-layered syndicated oil wells, including: Based on the geological unit to which the target well belongs, the corresponding integrated production capacity prediction model is invoked. The integrated production capacity prediction model is trained as follows: For different geological units within the target reservoir area, training datasets corresponding to each geological unit are obtained. These training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters for each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data for each reservoir segment of the historical well. Based on the training datasets, a corresponding integrated production capacity prediction model is trained for each geological unit. The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model, and the production capacity prediction value of each reservoir segment is output. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0007] In some embodiments, the method further includes: Acquire historical data of each historical well in each geological unit within the target oil reservoir area. The historical data includes historical production data, historical geological parameters of each reservoir section, and historical fracturing operation parameters. The historical production data includes at least historical total production, and the historical geological parameters include at least historical permeability and historical effective thickness. Accordingly, the historical production data is generated based on historical geological parameters and historical production data of each reservoir section of historical wells, including: For each reservoir segment of a historical well, calculate the product of the historical permeability and the historical effective thickness of each reservoir segment, and then calculate the sum of the products of all reservoir segments. The geological weight of each reservoir segment is determined based on the ratio of the product of each reservoir segment to the sum of the products. Based on the geological weights, the total historical production is divided into different reservoir segments to generate independent historical production data for each reservoir segment.

[0008] In some embodiments, the key factors controlling historical production capacity are selected from historical geological parameters and historical fracturing operation parameters of each reservoir section of historical wells, including: Various methods for evaluating the importance of different characteristics were used to evaluate the historical geological parameters and historical fracturing operation parameters of each reservoir section of historical wells; Based on the evaluation results of each feature importance evaluation method for each parameter, the information entropy of each feature importance evaluation method is calculated, and the evaluation results are weighted and summed based on the information entropy. Based on the weighted summation results, the historical geological parameters and historical fracturing operation parameters of each reservoir section of the historical well are sorted, and a preset number of parameters are selected from the sorting results as the main controlling factors of historical production capacity.

[0009] In some embodiments, the various methods for evaluating the importance of different features include at least two of the following: grey relational analysis, maximum mutual information method, SHAP interpretation method, Pearson correlation analysis, and Spearman correlation analysis.

[0010] In some embodiments, the integrated production capacity prediction model includes multiple sub-models of different algorithm types; correspondingly, training a corresponding integrated production capacity prediction model for each geological unit based on the training dataset includes: Based on the training dataset, a multi-round hyperparameter optimization method is used to train and optimize each sub-model in the integrated production capacity prediction model corresponding to each geological unit.

[0011] In some embodiments, the multiple sub-models of different algorithm types include at least two of the following: linear regression model, tree model, and gradient boosting model.

[0012] In some embodiments, the step of inputting the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model and outputting the production capacity prediction value corresponding to each reservoir segment includes: The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model. The outputs of each sub-model in the corresponding integrated production capacity prediction model are weighted and averaged to output the production capacity prediction value corresponding to each reservoir segment. The weighting weight of each sub-model is determined based on the prediction performance index of each sub-model on the validation set.

[0013] In some embodiments, the method further includes: Based on the predicted production capacity of each reservoir segment, calculate the production capacity contribution ratio of each reservoir segment. Based on the aforementioned production capacity contribution ratio, the potential primary and secondary formations of the target well are determined. Based on the potential primary and secondary stratigraphic segments, a differentiated fracturing design scheme is generated for the target well. Specifically, a first fracturing construction parameter is configured for the potential main fracturing layer, and a second fracturing construction parameter is configured for the potential secondary fracturing layer. The design strength of the first fracturing construction parameter is higher than that of the second fracturing construction parameter.

[0014] Secondly, embodiments of this specification also provide an intelligent prediction device for the post-pressure production capacity of multi-layer syndicated oil wells, comprising: The training module is used to acquire training datasets for different geological units within the target reservoir area. These training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data of each reservoir segment of the historical well. Based on these training datasets, a corresponding integrated production capacity prediction model is trained for each geological unit. The calling module is used to call the corresponding integrated production capacity prediction model based on the geological unit to which the target well belongs; The prediction module is used to input the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model and output the production capacity prediction value corresponding to each reservoir segment. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0015] Thirdly, embodiments of this specification also provide a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the steps of the above-described intelligent prediction method for post-pressure production capacity of multi-layered syngas wells.

[0016] This specification provides an embodiment of a method and apparatus for intelligent prediction of post-fracturing production capacity in multi-layer syngas wells. First, based on the geological unit to which the target well belongs, a corresponding integrated production capacity prediction model is invoked. The integrated production capacity prediction model is trained as follows: For different geological units within the target reservoir area, training datasets are obtained for each geological unit. These training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data of each reservoir segment of the historical well. Based on the training dataset, a corresponding integrated production capacity prediction model is trained for each geological unit. Then, the input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model, and the predicted production capacity values ​​for each reservoir segment are output. The input feature vectors include the geological parameters and fracturing operation parameters of the target well. In the embodiments of this specification, a training dataset corresponding to each geological unit is constructed based on the historical production capacity controlling factors of each historical well and each reservoir segment (selected from historical geological parameters and historical fracturing operation parameters of each historical well and reservoir segment) and historical production data (generated based on historical geological parameters and historical production data of each historical well and reservoir segment). This ensures the geological relevance and effectiveness of the training data and provides high-quality data support for differentiated model training. By training a corresponding integrated production capacity prediction model for each geological unit based on the training dataset, the model can deeply fit the reservoir characteristics and production capacity response relationship of a specific geological unit, avoiding cross-geological unit interference and further improving the model's prediction accuracy and reliability. By calling the corresponding integrated production capacity prediction model according to the geological unit to which the target well belongs, the model can be accurately adapted to the specific geological unit, avoiding interference from different geological characteristics on the prediction results. This solves the problem of poor adaptability of the unified model used in the prior art and can effectively improve the model's prediction relevance under specific geological conditions. By inputting the feature vectors of each reservoir segment in the target well into the corresponding integrated production capacity prediction model, accurate and rapid output of independent production capacity prediction values ​​for each reservoir segment can be achieved. This overcomes the limitation of existing technologies that can only predict total production and provides direct data support for stratified fracturing optimization. Through this approach, accurate prediction of production capacity for each reservoir segment in multi-layered synergistic wells is realized, while simultaneously resolving the issues of poor adaptability and inability to output independent production capacity prediction values ​​for single layers found in existing models. This provides crucial support for stratified fracturing optimization and scientific development decision-making. Attached Figure Description

[0017] 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: Figure 1 A flowchart illustrating an intelligent prediction method for post-pressure production capacity of multi-layer syndicated oil wells provided in the embodiments of this specification; Figure 2 This is a schematic diagram illustrating one embodiment of the intelligent prediction method for post-pressure production capacity of multi-layer syndicated oil wells provided in this specification, applied in a scenario example. Figure 3 This is a schematic diagram illustrating one embodiment of the intelligent prediction method for post-pressure production capacity of multi-layer syndicated oil wells provided in this specification, applied in a scenario example. Figure 4 This is a schematic diagram illustrating one embodiment of the intelligent prediction method for post-pressure production capacity of multi-layer syndicated oil wells provided in this specification, applied in a scenario example. Figure 5 This is a schematic diagram illustrating one embodiment of the intelligent prediction method for post-pressure production capacity of multi-layer syndicated oil wells provided in this specification, applied in a scenario example. Figure 6 This is a schematic diagram of the structural composition of an intelligent prediction device for post-pressure production capacity of multi-layer syngas wells provided in the embodiments of this specification; Figure 7 This is a schematic diagram of the structural composition of the electronic device provided in the embodiments of this specification. Detailed Implementation

[0018] 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.

[0019] Hydraulic fracturing technology can significantly improve reservoir conductivity by creating artificial fractures, making it a key means for the economical and effective development of low-permeability reservoirs. Accurate prediction of post-fracturing production capacity is crucial for optimizing fracturing parameter design and formulating scientific development strategies. However, current understanding of post-fracturing production dynamics is insufficient, and efficient prediction methods are lacking, which severely restricts the evaluation of fracturing effectiveness and the adjustment of development plans.

[0020] Specifically, existing post-pressure production capacity prediction technologies for multi-layered syndicated oil wells have significant limitations: First, in terms of modeling strategies, existing methods generally adopt a "global uniform model," ignoring the inherent strong heterogeneity of oil and gas reservoirs. Within an oilfield or block, different regions often exhibit significant differences in structural characteristics, sedimentary environments, and reservoir properties (such as porosity and permeability). Using a uniform model to attempt to fit the productivity patterns of all regions essentially forces an average of the "productivity-characteristic" relationship under different geological backgrounds. This leads to unstable model performance in specific geological units (such as high-permeability zones, low-permeability zones, and fractured zones), limited prediction accuracy, and poor generalization ability. Especially for new areas or marginal locations with vastly different geological conditions, the prediction results of the global model are often inaccurate and cannot reliably guide development decisions.

[0021] Second, regarding the prediction target, existing methods only use "total well production" as the prediction target, completely ignoring the hierarchical nature of the production capacity composition of multi-layered commingled wells. Most production wells simultaneously exploit multiple reservoir sections, and the total production measured on the surface is a mixture of contributions from each section. Existing models learn and predict total production as an indivisible label, but their fundamental flaw lies in their inability to learn or output the independent production capacity contribution of each reservoir section. Therefore, the prediction result is only a general total value, unable to identify which section contributes highly or which section is under-fracturing. This makes it difficult for existing intelligent prediction technology to directly serve the refined adjustment and differentiated design of fracturing schemes; its value remains at the macro-level assessment and cannot penetrate into guiding specific construction optimization.

[0022] To address the aforementioned issues, this specification provides an intelligent method and apparatus for predicting post-fracturing production capacity of multi-layer syngas wells. This method is particularly suitable for complex development scenarios in low-permeability, unconventional oil and gas reservoirs where layered fracturing design and effect evaluation of multiple reservoir sections are required. It solves the problem that existing technologies cannot predict the layered production capacity of multi-layer syngas wells for different geological units, achieving a leap from macroscopic prediction of the entire well to microscopic prediction of reservoir sections, and providing direct data support for the refined design and optimization of fracturing schemes.

[0023] 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.

[0024] See Figure 1As shown in the embodiments of this specification, a method for intelligent prediction of post-pressure production capacity of multi-layered syndicated oil wells is provided. In specific implementation, this method may include the following: S101: Based on the geological unit to which the target well belongs, invoke the corresponding integrated production capacity prediction model; wherein, the integrated production capacity prediction model is trained in the following manner: for different geological units within the target reservoir area, obtain the training dataset corresponding to each geological unit. The training dataset consists of the historical production capacity controlling factors and historical production data of each reservoir segment of each historical well in the geological unit. The historical production capacity controlling factors are selected from the historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data of each reservoir segment of the historical well. Based on the training dataset, train the corresponding integrated production capacity prediction model for each geological unit. S102: Input the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model, and output the production capacity prediction value corresponding to each reservoir segment. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0025] Specifically, the target wells mentioned above can be multi-layered synergistic oil wells for which production capacity prediction is to be conducted. These are typically newly designed wells or wells that have been deployed but not yet fractured, with known geological parameters and planned fracturing parameters. The historical wells mentioned above can be multi-layered synergistic oil wells within the target reservoir area that have undergone fracturing and have a production history for some time; their data are used to train the model. Multi-layered synergistic oil wells refer to wells that, through one or more well completions and fracturing operations, simultaneously exploit multiple independent reservoir sections (each reservoir section possessing independent geological characteristics and production potential) within the same well. Their core characteristic is that multiple reservoir sections share the wellbore channel for production, requiring precise differentiation of the production capacity contribution of each section to optimize development strategies. This is a commonly used well type for improving recovery rates in low-permeability, tight reservoirs. The geological units mentioned above can be independent regional units divided based on the consistency of geological conditions such as reservoir lithology, porosity distribution, permeability range, and structural characteristics. Within the same geological unit, the basic reservoir conditions are similar, and the production capacity response patterns are common. However, due to differences in geological characteristics, the factors influencing production capacity and their response relationships differ significantly between different geological units. The target oil reservoir area mentioned above can be a complete oilfield, an exploration block, a structural zone (such as an anticline or fault block), or part of an oil and gas system with a unified stratigraphic framework and reservoir-forming background.

[0026] In practice, based on regional geological research findings, a target oil reservoir (e.g., "Mining Area K") can be divided into several geological units (e.g., "Geological Unit A: High-Permeability Sandstone Zone," "Geological Unit B: Tight Shale Zone"). Each unit has a pre-trained dedicated "Production Prediction Integrated Model" stored in a model library. When a prediction is needed for a target well (new well), the system first determines its geological unit (e.g., "Geological Unit A") based on its well location coordinates and geological design, and then retrieves the corresponding model from the model library. Next, for each designed fracturing reservoir segment of the target well, its geological parameters (obtained from well logging interpretation) and planned construction parameters (obtained from fracturing design) are prepared and combined to form the input feature vector for that reservoir segment. These input feature vectors are then sequentially input into the retrieved model, which outputs the production prediction value for each reservoir segment.

[0027] Through the core design of "differentiated modeling of geological units + layered feature input + single-layer production output", the core problems of poor prediction adaptability and inability to accurately output single-layer production capacity in existing technologies have been solved. It has realized accurate and layered prediction of post-fracturing production capacity of multi-layer synergistic oil wells, which can provide key data support for subsequent fracturing scheme optimization and development decision-making.

[0028] In some embodiments, prior to S101 above, the following may be included in a specific implementation: Acquire historical data of each historical well in each geological unit within the target oil reservoir area. The historical data includes historical production data, historical geological parameters of each reservoir section, and historical fracturing operation parameters. The historical production data includes at least historical total production, and the historical geological parameters include at least historical permeability and historical effective thickness. Accordingly, the historical production data in S101 above is generated based on the historical geological parameters and historical production data of each reservoir section of historical wells. In specific implementation, it may include: For each reservoir segment of a historical well, calculate the product of the historical permeability and the historical effective thickness of each reservoir segment, and then calculate the sum of the products of all reservoir segments. The geological weight of each reservoir segment is determined based on the ratio of the product of each reservoir segment to the sum of the products. Based on the geological weights, the total historical production is divided into different reservoir segments to generate independent historical production data for each reservoir segment.

[0029] Specifically, the aforementioned historical production data (or historical production parameters) can refer to data related to production output and status recorded during normal production after fracturing operations in historical wells. The core of this data is to reflect the actual production effect after fracturing and is the basis for generating historical production data (single-layer splitting production). This data can include historical total production, daily fluid production, daily oil production, water cut, etc. The aforementioned historical geological parameters can refer to parameters obtained through geological exploration, well logging interpretation, etc., reflecting the inherent geological properties of each reservoir segment. These parameters directly determine the inherent production potential of the reservoir and are the core geological basis for differences in production capacity between different regions and stratigraphic levels. These parameters can include effective thickness (m), porosity (%), permeability (mD), clay content (%), oil saturation (%), etc. The aforementioned historical fracturing operation parameters (or historical engineering parameters) refer to the technical parameters that were actually implemented and could be artificially controlled during the hydraulic fracturing operation of historical wells. These parameters are core indicators reflecting the intensity of artificial reservoir stimulation, directly affecting fracture length, width, and conductivity. These parameters focus on the core controllable aspects of fracturing stimulation and are key engineering means to tap geological potential and improve production efficiency. They may include the strength of the fracturing fluid (m³ / s). 3 / m), total sand volume (t), total liquid volume (m³) 3 Construction discharge (m) 3 / min), sand addition intensity (m 3 / m), sand-carrying liquid volume (m 3 ), pre-liquid volume (m 3 Pump shutdown pressure (MPa), sand ratio during construction (%), etc.

[0030] The collected data can be preprocessed (such as data cleaning, data normalization, etc.) to create a dataset, which will lay the foundation for generating historical output data, screening historical controlling factors, and building a training dataset.

[0031] The following explains the specific process of generating historical production data: First, for each reservoir section of a single historical well, extract the historical permeability (K) of that reservoir section from the well logging interpretation report. i ) and historical effective thickness (h i ), calculate the product of historical permeability and historical effective thickness (K) for each reservoir segment. i ×h i As a production potential indicator, it can directly reflect the reservoir's potential contribution to production. Historical permeability reflects the reservoir's fluid flow capacity, while historical effective thickness reflects the reservoir's effective storage space. The larger the product of the two, the stronger the potential contribution of that layer or reservoir segment to the total production of a single well. After calculating the product of historical permeability and historical effective thickness for all reservoir segments, the sum of these products for the given historical well is then calculated.

[0032] The geological weight of a reservoir segment is calculated by multiplying the historical permeability and historical effective thickness of each segment by the sum of the products of all reservoir segments in that historical well. The percentage obtained is the geological weight of that reservoir segment, ranging from 0 to 100%, and the sum of the weights of all reservoir segments is 100%. The essence of this geological weight is to quantify the "contribution ratio" of a single segment in the entire well reservoir. The higher the weight, the greater its actual contribution to the total historical production.

[0033] Finally, based on the historical total production (Q_total) actually measured by a single well, the geological weight of each reservoir segment is multiplied by the historical total production (i.e., Q_single layer = Q_total × geological weight of that layer) to generate independent historical production data for each reservoir segment. This data can be used as a component of the subsequent training dataset.

[0034] The "permeability-thickness product weighting method" can achieve stratified division of historical total production, solving the technical problem that multi-layer commingled wells cannot directly obtain single-layer historical production data. It can provide high-quality stratified labeled data for the model, ensuring that the model can learn the intrinsic relationship between historical geological parameters, historical fracturing construction parameters and historical production data of a single layer, and can lay a data foundation for subsequent single-layer production capacity prediction.

[0035] In some embodiments, the historical production capacity controlling factors in S101 above are selected from the historical geological parameters and historical fracturing operation parameters of each reservoir section of the historical well. In specific implementation, this may include: Various methods for evaluating the importance of different characteristics were used to evaluate the historical geological parameters and historical fracturing operation parameters of each reservoir section of historical wells; Based on the evaluation results of each feature importance evaluation method for each parameter, the information entropy of each feature importance evaluation method is calculated, and the evaluation results are weighted and summed based on the information entropy. Based on the weighted summation results, the historical geological parameters and historical fracturing operation parameters of each reservoir section of the historical well are sorted, and a preset number of parameters are selected from the sorting results as the main controlling factors of historical production capacity.

[0036] Specifically, the screening process for key historical production capacity control factors can be as follows: We can first screen out n (e.g., 20-30) potential influencing parameters (such as oil saturation and clay content in historical geological parameters, and sand addition strength and pump shutdown pressure in construction parameters) from historical geological parameters and historical fracturing construction parameters as candidate parameters; Then, multiple feature importance evaluation methods are used to independently evaluate the candidate parameters. For example, the evaluation results of each feature importance evaluation method for the n candidate parameters are x. j1 ,x j2 ,……,x jnThen, the evaluation results are normalized to obtain the normalized score p of the i-th parameter under the j-th feature evaluation method. ji ; Then, based on the normalized score p of the i-th parameter under the j-th feature evaluation method... ji Calculate the information entropy H of the j-th feature evaluation method. j :

[0037] Among them, H j The information entropy of the j-th feature evaluation method (within the range of 0 ≤ H) j ≤1); n is the total number of candidate parameters; p ji Let be the normalized score of the i-th parameter under the j-th feature evaluation method; These are the normalization coefficients; Then, based on the information entropy H of the j-th feature evaluation method j Calculate the coefficient of difference d j (d) j =1-H j Then, the difference coefficients are normalized to obtain the weight w of the j-th feature evaluation method. j : ; Then, based on the weight w of the j-th feature evaluation method j Evaluation results x for each method ji Perform a weighted summation to obtain the comprehensive score S of the i-th parameter. i Finally, the historical geological parameters and historical fracturing operation parameters of each reservoir section of the historical wells are ranked according to the comprehensive score. From these, a predetermined number of historical production capacity controlling factors are selected, such as the top 9. These can serve as another component of the subsequent training dataset.

[0038] By employing a multi-method evaluation and entropy weight fusion strategy, the limitations of single feature selection methods are avoided. This ensures that the selected historical production capacity control factors can comprehensively and accurately reflect the core variables affecting production capacity, reduce the interference of irrelevant parameters on model training, and improve model training efficiency and prediction accuracy.

[0039] In some embodiments, the above-mentioned methods for evaluating the importance of different features may include at least two of the following: grey relational analysis, maximum mutual information method, SHAP interpretation method, Pearson correlation analysis, and Spearman correlation analysis.

[0040] Specifically, the grey relational analysis method can be used to analyze the degree of correlation between factors and production capacity, the maximum mutual information method can quantify the information transmission efficiency between factors and production capacity, the SHAP interpretation method can extract the importance of features from the perspective of model decision-making, and the Pearson correlation method and the Spearman correlation method can measure linear and nonlinear correlations, respectively.

[0041] Five multi-dimensional feature importance evaluation methods can be used to evaluate the key factors controlling potential historical production capacity. To avoid the limitations of a single evaluation method, the entropy weighting method can be used to objectively integrate the evaluation results of the five methods. The core principle of this method is to calculate the information entropy of the results of each evaluation method (the smaller the entropy value, the greater the dispersion and discrimination of the evaluation results of the method, and the greater the corresponding weight). After converting the information entropy into objective weights for each method, the evaluation results of the five methods are weighted, summed, and then sorted. Finally, nine core parameters can be selected from the sorted results—such as historical geological parameters (reservoir thickness, porosity, permeability, clay content) and historical engineering parameters (construction discharge, total liquid volume, total sand volume, pre-filled liquid volume, and sand-carrying liquid volume)—as the historical input feature vector of the integrated production capacity prediction model, i.e., the key factors controlling historical production capacity.

[0042] In some embodiments, the integrated production capacity prediction model in S101 above may include multiple sub-models of different algorithm types; correspondingly, the step in S101 above, which trains a corresponding integrated production capacity prediction model for each geological unit based on the training dataset, may include, in specific implementation: Based on the training dataset, a multi-round hyperparameter optimization method is used to train and optimize each sub-model in the integrated production capacity prediction model corresponding to each geological unit.

[0043] Specifically, the process of the above-mentioned multi-round hyperparameter optimization can be summarized as follows: First, we can define the hyperparameter search space: for example, we can define the search range of key hyperparameters for different types of sub-models. For example, the search range of regularization strength for a linear regression model is [0.001, 10], the search range of tree depth max_depth for a tree model is [3, 10], and the search range of learning rate for a gradient boosting model is [0.01, 0.3]. Then, perform iterative settings: for example, set the maximum number of iterations to 5 rounds. In the first round, optimize in a fixed order according to the "core hyperparameter priority" (e.g., optimize the regularization strength first, then optimize the number of iterations); in the subsequent 4 rounds, randomly shuffle the hyperparameter optimization order to avoid getting trapped in local optima. In each iteration, a single-factor variable method is used to fix other hyperparameters and perform a grid search on the currently optimized hyperparameter (the step size is set according to the parameter type, such as a learning rate step size of 0.01). The parameter value with the highest prediction accuracy on the validation set is selected as the optimal value of the hyperparameter. Stopping condition judgment: When the improvement in prediction accuracy of the model on the validation set is less than 0.001 after a certain iteration, or when the maximum number of iterations is reached, the hyperparameter optimization is stopped and the current optimal hyperparameter combination is saved.

[0044] By employing a multi-round, randomized hyperparameter optimization strategy, the problem of local optima caused by a single optimization order is avoided, ensuring that each sub-model can achieve optimal training results and guaranteeing the high prediction accuracy of the ensemble model.

[0045] In some embodiments, the sub-models of the above-mentioned multiple different algorithm types may include at least two of the following: linear regression model, tree model, and gradient boosting model.

[0046] Specifically, the aforementioned linear regression models can be either Huber regression or Lasso regression. The Huber regression model, based on standardized master features, optimizes core hyperparameters such as epsilon (outlier tolerance), alpha (regularization strength), and max_iter (number of iterations), employing an improved loss function (squared loss for small errors, linear loss for large errors) to reduce outlier interference, fitting the non-linear relationship between features and single-layer output, and outputting more robust single-layer output predictions. The Lasso regression model, also based on standardized master features, optimizes hyperparameters such as alpha (L1 regularization strength), max_iter (number of iterations), and tol (convergence threshold), using the absolute value of the penalty coefficient to achieve feature sparsity and multicollinearity suppression, fitting the linear mapping relationship between features and single-layer output, and outputting sparsely linearly fitted single-layer output predictions.

[0047] The tree model mentioned above can be a random forest model. Based on the original master features, hyperparameters such as n_estimators (number of decision trees), max_depth (tree depth), and min_samples_split (minimum number of samples for node splitting) can be optimized. Multiple independent decision trees can be constructed by bootstrap sampling. Each tree randomly selects some features to participate in node splitting. Finally, the prediction results of individual trees are fused by averaging to output an integrated single-layer segment yield prediction value that reduces the risk of overfitting.

[0048] The gradient boosting models mentioned above can be either XGBoost or LGBM. The XGBoost model, based on the original master features, optimizes hyperparameters such as `learning_rate`, `n_estimators` (number of weak classifiers), and `max_depth` (tree depth). It iteratively constructs a regularized decision tree using a gradient descent strategy, and fuses the outputs of each weak classifier using a weighted summation method to fit the complex nonlinear mapping between production capacity and features, outputting high-precision single-layer production predictions. The LGBM model, based on the original master features, optimizes hyperparameters such as `learning_rate`, `num_leaves` (number of leaf nodes), and `reg_alpha / reg_lambda` (regularization coefficients). It uses a histogram algorithm to accelerate feature splitting calculation and gradient unilateral sampling to reduce sample size. Through efficient gradient boosting iterative optimization of model performance, it outputs single-layer production predictions that balance speed and accuracy.

[0049] Each sub-model can employ different feature preprocessing methods (standardized or raw features) to suit its algorithmic characteristics. For example, Huber regression and Lasso regression models can use standardized historical capacity control factors or historical input feature vectors (to eliminate the influence of dimensions), while XGBoost, Random Forest, and LGBM can use raw historical input feature vectors.

[0050] By integrating sub-models of different algorithm types, a comprehensive coverage of "linear relationship capture + nonlinear relationship fitting + anti-interference ability + overfitting suppression" is achieved. The advantages of each sub-model complement each other, improving the generalization ability and robustness of the integrated model.

[0051] In some embodiments, the step of inputting the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model in S102 above, and outputting the production capacity prediction value corresponding to each reservoir segment, may include, in specific implementation: The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model. The outputs of each sub-model in the corresponding integrated production capacity prediction model are weighted and averaged to output the production capacity prediction value corresponding to each reservoir segment. The weighting weight of each sub-model is determined based on the prediction performance index of each sub-model on the validation set.

[0052] Specifically, ACC (accuracy based on total absolute error) can be selected as the predictive performance metric, and the weighted weights w of each sub-model can be used. i = ACC i / The sum of the ACC of all sub-models on the validation set (ACC1~ACC5), where w i For the weights of the i-th sub-model, ACC i Let be the prediction performance metric of the i-th sub-model on the validation set.

[0053] For example: Suppose the predictive performance metrics (ACC) of the 5 sub-models on the validation set are 0.86 for the Huber regression model, 0.92 for the XGBoost model, 0.89 for the Random Forest model, 0.93 for the LGBM model, and 0.84 for the Lasso regression model. The sum of the ACC of all sub-models on the validation set is 0.86 + 0.92 + 0.89 + 0.93 + 0.84 = 4.44. Then, the weight of the XGBoost model w_XGB = 0.92 / 4.44 ≈ 0.2072, the weight of the LGBM model w_LGBM = 0.93 / 4.44 ≈ 0.2095, and the weight of the Huber regression model w_Hube... r = 0.86 / 4.44 ≈ 0.1937, the weights of the random forest model w_RF = 0.89 / 4.44 ≈ 0.2005, and the weights of the Lasso regression model w_Lasso = 0.84 / 4.44 ≈ 0.1892. The final result is calculated as follows: "Predicted production capacity of each reservoir segment = w_Huber × predicted value of each reservoir segment + w_XGB × predicted value of each reservoir segment XGBoost + w_RF × predicted value of each reservoir segment Random forest + w_LGBM × predicted value of each reservoir segment LGBM + predicted value of each reservoir segment w_Lasso × predicted value of each reservoir segment Lasso".

[0054] By determining the weights of sub-models based on validation set performance metrics, sub-models with higher prediction accuracy can occupy a larger weight in the final result, further improving the prediction accuracy and reliability of the ensemble model.

[0055] In some embodiments, after S102 above, in specific implementation, it may further include: Based on the predicted production capacity of each reservoir segment, calculate the production capacity contribution ratio of each reservoir segment. Based on the aforementioned production capacity contribution ratio, the potential primary and secondary formations of the target well are determined. Based on the potential primary and secondary stratigraphic segments, a differentiated fracturing design scheme is generated for the target well. Specifically, a first fracturing construction parameter is configured for the potential main fracturing layer, and a second fracturing construction parameter is configured for the potential secondary fracturing layer. The design strength of the first fracturing construction parameter is higher than that of the second fracturing construction parameter.

[0056] Specifically, for the target well, the sum of the predicted production capacity of all reservoir segments can be calculated first. Then, the ratio of the predicted production capacity of each reservoir segment to the sum of the predicted production capacity of all reservoir segments can be determined to obtain the production capacity contribution ratio of each reservoir segment. Next, the production capacity contribution ratio of each reservoir segment is compared with a preset contribution ratio threshold (which may include a first preset contribution ratio threshold and a second preset contribution ratio threshold, the second preset contribution ratio threshold being less than the first preset contribution ratio threshold) to classify the reservoir segments of the target well into potential main segments and potential secondary segments. For example, if the production capacity contribution ratio is greater than or equal to the first preset contribution ratio threshold, such as 30%, then the reservoir segment is classified or defined as a potential main segment; if the production capacity contribution ratio is less than the first preset contribution ratio threshold, such as 30%, but greater than the second preset contribution ratio threshold, such as 10%, then the reservoir segment is classified or defined as a potential secondary segment; if the production capacity contribution ratio is less than the second preset contribution ratio threshold, such as 10%, then the reservoir segment is classified or defined as a low-potential segment (the threshold can be adjusted according to the production capacity distribution characteristics of the geological unit).

[0057] Specifically, first fracturing construction parameters can be configured for potential main strata, and second fracturing construction parameters can be configured for potential secondary strata. The design strength of the first fracturing construction parameters can be higher than that of the second fracturing construction parameters. Low-potential strata can have reduced construction intensity or be temporarily not fracturing to avoid ineffective investment.

[0058] By directly linking the capacity forecast results with the fracturing scheme design, a closed loop of "prediction-evaluation-optimization" is achieved. Through differentiated parameter configuration, resources can be ensured to be tilted towards high-potential sections, reducing the risk of ineffective construction and improving the economic benefits and targeting of fracturing construction.

[0059] In some embodiments, during the model training phase, an independent model or group of models can be trained for each geological unit (such as a block-stratum combination). During prediction, the corresponding trained model can be selected based on the block and stratum where the target well is located, and the geological and engineering parameters of the well can be input to obtain the predicted production value for each reservoir segment.

[0060] 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.

[0061] 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.

[0062] The following section, using specific scenario examples, explains the entire process of the intelligent prediction method for post-pressure production capacity of multi-layered syngas wells described above. (See reference...) Figure 2 As shown, the process of model training and new well prediction can be summarized as follows (S1-S4): S1. Geological Unit Division and Historical Data Preparation: Suppose a target area (e.g., mining area K) is divided into two main geological units: Geological Unit A (a high-yield sandstone unit, primarily composed of high-porosity, high-permeability sandstone at shallow depths) and Geological Unit B (a tight oil unit, primarily composed of low-porosity, low-permeability tight sandstone and shale, requiring large-scale fracturing). Each of Geological Unit A and Geological Unit B contains several historical wells that have undergone fracturing and production. Historical data can be collected from all historical wells in Geological Unit A and all historical wells in Geological Unit B.

[0063] S2. Construct a training dataset for each geological unit: Taking geological unit A as an example: Assume there are 3 historical wells in geological unit A: A_H1, A_H2, and A_H3. Taking historical well A_H1 (a multi-layered syndicated oil well) as an example: S21. Collect historical data from historical well A_H1: Historical production data: The cumulative oil production of historical well A_H1 in the 30 days after compression was 1200 tons (historical total production Q_total).

[0064] Historical geological parameters (from well logging interpretation): Reservoir section A_H1-L1: historical effective thickness h=8m, historical permeability K=50mD, historical porosity=18%.

[0065] Reservoir section A_H1-L2: historical effective thickness h=12m, historical permeability K=80mD, historical porosity=22%.

[0066] Historical fracturing operation parameters (from the construction design report): For reservoir section A_H1-L1: Historical construction discharge = 10m³ 3 / min, historical total liquid volume = 800m 3 The total historical sand volume is 60 tons.

[0067] For reservoir section A_H1-L2: historical construction discharge = 12m³ 3 / min, historical total liquid volume = 1200m 3 The total historical sand volume is 100 tons.

[0068] S22. Perform production allocation (distribute the total historical production of each well to each reservoir section according to the weight of K×h): Calculate the capacity potential indicator (the product of historical penetration rate and historical effective thickness): A_H1-L1: K×h =50×8=400.

[0069] A_H1-L2: K×h =80×12=960.

[0070] The sum of the products of all reservoir segments = 400 + 960 = 1360.

[0071] Calculate geological weights: A_H1-L1 weight = 400 / 1360 ≈ 0.294.

[0072] A_H1-L2 weight = 960 / 1360 ≈ 0.706.

[0073] Generate independent historical production data (tags) for each reservoir segment: Historical production data for A_H1-L1 = 1200 tons × 0.294 ≈ 353 tons.

[0074] Historical production data for A_H1-L2 = 1200 tons × 0.706 ≈ 847 tons.

[0075] S23. Forming training samples (training dataset): Sample A1 (corresponding to A_H1-L1): The historical production capacity controlling factors selected from the historical geological parameters and historical fracturing construction parameters of reservoir section A_H1-L1 constitute the historical input feature vector. For example, the historical input feature vector of reservoir section A_H1-L1 = [thickness = 8m, porosity = 18%, permeability = 50mD, clay content = X%, displacement = 10, liquid volume = 800, sand volume = 60...]; tag = 353 tons.

[0076] Sample A2 (corresponding to A_H1-L2): The historical input feature vector of this reservoir section A_H1-L2 = [thickness = 12m, porosity = 22%, permeability = 80mD, clay content = Y%, discharge rate = 12, liquid volume = 1200, sand volume = 100...]; tag = 847 tons.

[0077] Similarly, all historical wells (A_H2, A_H3) within geological unit A and all historical wells within geological unit B are processed. This results in two independent training datasets: Dataset_A (Training Sample Library_A): Contains samples from all reservoir segments of geological unit A.

[0078] Dataset_B (Training Sample Library_B): Contains samples from all reservoir segments of geological unit B.

[0079] For example, assuming historical well A_H2 has one reservoir segment A_H2-L1, one sample is generated; historical well A_H3 has three reservoir segments A_H3-L1, A_H3-L2, and A_H3-L3, three samples are generated. Therefore, there are a total of 6 samples for all reservoir segments in geological unit A.

[0080] S3, Model Training: The training dataset consisting of 6 samples from geological unit A can be divided into a training set and a validation set to train the ensemble model for productivity prediction corresponding to geological unit A (which may include multiple sub-models). For each sub-model, a multi-round hyperparameter optimization method can be used for training. For example, for the XGBoost model, the search space for learning_rate, n_estimators, and max_depth can be defined. A maximum of 5 rounds of iteration are performed: the first round optimizes sequentially, and subsequent rounds randomly shuffle the order, with single-factor stepwise optimization used in each round to find the optimal value.

[0081] Similarly, train the integrated model for production capacity prediction corresponding to geological unit B. The trained model can be stored in the model library.

[0082] After training, each sub-model is evaluated on the validation set, and the prediction performance metric (ACC) of each sub-model on the validation set is determined as the weighting weight of each sub-model.

[0083] S4. Model Prediction Stage: For a newly designed well or target well (Target_Well) with plans to fracture three reservoir sections: T-L1, T-L2, and T-L3, if the geological unit to which the well belongs is determined to be A, then the integrated production prediction model corresponding to geological unit A (integrated model A1) is automatically invoked; if the geological unit to which the well belongs is determined to be B, then the integrated production prediction model corresponding to geological unit B (integrated model B1) is automatically invoked. The input feature vector of T-L1 of the new well or target well is as follows: [thickness = 9m, porosity = 19%, permeability = 55mD, clay content = 9%, design flow rate = 11m³ / s]. 3 / min, Design total liquid volume = 950 m³ 3 Total sand volume designed = 75 tons, Design pre-fluid volume designed = 250 m³3 Design sand-carrying capacity = 700m³ 3 The input is fed into the integrated production capacity prediction model corresponding to the geological unit. The outputs of each sub-model are then weighted and averaged according to their respective weights, for example: Predicted production of T-L1 reservoir section = 0.194×310+0.207×330+0.200×320+0.210×325+0.189×305≈320.5 tons.

[0084] Similarly, the input feature vectors of T-L2 and T-L3 are weighted and averaged to obtain predicted production values ​​of 715.2 tons and 185.3 tons, respectively. This outputs independent production prediction values ​​for each reservoir segment, which are then aggregated to obtain the total predicted production value for the entire well.

[0085] S5. Results Analysis and Optimization Decision-Making: S51. Calculate the contribution ratio: The predicted total production of the whole well = 320.5 + 715.2 + 185.3 = 1221.0 tons.

[0086] The proportion of the T-L1 reservoir segment is approximately 26.2% (320.5 / 1221.0).

[0087] The proportion of the T-L2 reservoir segment is approximately 58.6% (715.2 / 1221.0).

[0088] The proportion of the T-L3 reservoir segment is approximately 15.2% (185.3 / 1221.0).

[0089] S52. Determine the potential primary and secondary reservoir segments: Set thresholds, with a contribution ratio >40% for primary reservoir segments and <20% for secondary reservoir segments. Therefore, the T-L2 reservoir segment is the potential primary reservoir segment, and the T-L1 and T-L3 reservoir segments are potential secondary reservoir segments (the T-L3 reservoir segment is close to the economic lower limit).

[0090] S53. Generate differentiated fracturing design schemes: For the main reservoir section T-L2: adopt the first fracturing operation parameters (high strength). It is recommended to increase the fracturing displacement to 14-15m³ based on the initial design. 3 / min, the total sand volume is increased by 10%-15% to maximize production capacity.

[0091] For the secondary reservoir section T-L1: Use the second fracturing parameters (medium intensity). The original design can be maintained, or minor adjustments and optimizations can be made.

[0092] For the minor reservoir section T-L3: Use the second fracturing parameters (economic intensity). Given its low predictive contribution, it is recommended to assess its economic viability. Consider postponing fracturing of this section or reducing the fracturing scale (e.g., reducing the flow rate to 8 m³). 3(Sand volume reduced by 20% per minute) to save costs. Simultaneously, it can be adjusted to a monitoring section to provide data for subsequent adjacent well design.

[0093] See Figure 3 As shown, the principle of the integrated capacity forecasting model can be summarized as follows: a. Collect and organize the geological, engineering, and production parameters of the target well and adjacent wells, preprocess the data, and establish a dataset. The dataset can be found in [reference needed]. Figure 4 As shown, some geological parameters may include oil layer thickness, gamma curve, minimum clay content, maximum clay content, average clay content, minimum porosity (%), maximum porosity (%), average porosity (%), minimum permeability (mD), maximum permeability (mD), average permeability (mD), minimum oil saturation (%), maximum oil saturation (%), and average oil saturation (%); some engineering parameters may include total sand content (m³ / s). 3 Total liquid volume during construction (m³) 3 ), construction fluid strength (m) 3 / m), construction sand strength (m 3 / m), sand-carrying liquid volume (m 3 ), substitution (m) 3 Pre-treatment liquid ratio (%), construction sand ratio (%), and maximum construction sand concentration (kg / m³) 3 ), Rupture pressure (MPa), Pump stop pressure (MPa), Construction pressure (MPa), 30-day cumulative stroke, 30-day cumulative production time; Some production parameters may include: 30-day liquid increase, 30-day liquid production, 60-day liquid increase, 60-day liquid production, 90-day liquid increase, 90-day liquid production, 180-day liquid increase, 180-day liquid production, 360-day liquid increase, 360-day liquid production, 720-day liquid increase, 720-day liquid production, hourly liquid increase, hourly liquid production.

[0094] b. Addressing the characteristic that the historical total production of multi-layered syndicated oil wells is contributed by multiple reservoir segments, making it impossible to directly obtain the independent production of a single segment through testing, a targeted data processing scheme was designed to adapt to the production capacity prediction scenario of multi-layered syndicated oil wells. Specifically, by reasonably splitting the historical total production of a single well, independent historical production data for a single segment is obtained, which can provide training samples for subsequent single-layer production capacity prediction models. The specific splitting method can be found above and will not be elaborated upon here.

[0095] c. Regarding the identification of key historical production capacity control factors, based on the full data collected in step a, it is first clarified that candidate key control factors cover two major categories: geological parameters and engineering parameters. Five multi-dimensional feature importance evaluation methods are comprehensively used to evaluate potential key historical production capacity control factors. The entropy weighting method can be used to objectively integrate the evaluation results of the five methods. The core principle of this method is to calculate the information entropy of the results of each evaluation method, convert the information entropy into objective weights for each method, and then perform a weighted summation and ranking of the evaluation results of the five methods. Finally, nine core parameters (key control features) can be selected from the ranking results, such as thickness, porosity, permeability, total sand content, pump shutdown pressure, sand-carrying fluid volume, total fluid volume, sand addition intensity, and sand-carrying fluid ratio.

[0096] d. Using the optimized geological and engineering parameters as historical input feature vectors (historical main control factors of production capacity), and historical production data (such as 30-day cumulative liquid production) as output labels, train an integrated production capacity prediction model (which may include multiple different sub-models). Specifically, 12 models (such as Lasso, Ridge, Random Forest, PLS, XGBoost, Catboost, SVR, and Huber) can be trained using the selected main control features as input. Five optimal models (such as Huber, Lasso, Random Forest, LGBM, and XGBoost) are then selected through a single-factor stepwise optimization method. A weighted layer is introduced to construct an integrated learning model (Huber, Lasso, Random Forest, LGBM, and XGBoost correspond to w1=w_Huber, w2=w_Lasso, w3=w_Random Forest, w4=w_LGBM, and w5=w_XGBoost, respectively), achieving automatic training and hyperparameter optimization. Evaluation is completed using ACC (accuracy) as the core indicator.

[0097] Specifically, a unified "multi-round randomized sequential optimization method" can be used for model training and hyperparameter optimization. First, a search space for key hyperparameters is defined for each model (such as the regularization strength of linear models, the tree depth and node splitting threshold of tree-based models, etc.). Then, optimization is carried out through a maximum of 5 rounds of iteration. In the first round, the core hyperparameters are optimized in a fixed order. In subsequent rounds, the optimization order is randomly shuffled to avoid local optima. Within each round, a single-factor stepwise optimization method is used (other parameters are fixed, and the optimal value of each core hyperparameter is searched one by one in detail) until the model's ACC improvement on the validation set is less than 0.001 or the maximum number of iterations is reached.

[0098] The training results of each sub-model in the integrated capacity forecasting model can be found in [reference]. Figure 5As shown, Huber's accuracy is 0.8358, XGBoost's accuracy is 0.8446, the Random Forest model's accuracy is 0.8456, LGBM's prediction accuracy is 0.8403, Lasso's prediction accuracy is 0.9297, and the overall prediction accuracy of the ensemble model is 0.8461.

[0099] The final production capacity forecast is derived by weighted averaging of the predictions from the five models. This ensemble approach combines the advantages of different models—linear models capture linear correlations, nonlinear models fit complex relationships, and tree-based and gradient-enhanced models mine feature interactions—significantly improving the model's stability and generalization ability. It achieves an organic unity of automatic model training, intelligent hyperparameter optimization, and accurate production capacity forecasting. The "single-layer production forecast" outputs from the five models can be fused using a weighted average method to obtain the final production forecast result for a single layer.

[0100] e. A standardized prediction process can be established to address the production capacity forecasting needs of new well fracturing: using the nine core parameters selected in step c as input, and the weighted integration model optimized in step d as the core tool, the final output is the 30-day cumulative fluid production after single-well fracturing, completing the entire process of new well production capacity forecasting. This forecast result can provide support from multiple dimensions, including development decision-making, scheme optimization, and benefit assessment. Specific application values ​​include: 1. Guiding development deployment: By predicting the 30-day cumulative fluid production of different blocks and layers, high-potential areas can be accurately identified, providing data for new well deployment and well location selection, improving the scientific nature of development layout; 2. Optimizing fracturing schemes: By comparing the difference between the predicted results and the target production capacity, construction parameters can be adjusted in reverse, forming a closed loop of "prediction-optimization-implementation," reducing the risk of ineffective construction; 3. Evaluating development benefits: Based on the 30-day cumulative fluid production, the initial production capacity of a single well can be estimated. Combined with data such as oil production costs and oil prices, the economic benefits of a single well can be predicted in advance, providing a reference for development investment decisions.

[0101] In summary, the dataset can be manually compiled from documents such as daily oil well data, measured data tables of fracturing operations, and well logging curves. Outliers are then removed, and missing values ​​are filled using algorithms. During the analysis of controlling factors, five methods—grey relational analysis, maximum mutual information, SHAP post-interpretation, Pearson correlation, and Spearman correlation—were used to evaluate the controlling factors of fracturing production capacity. A comprehensive evaluation was then performed by integrating these five methods using the entropy weighting method, calculating the Pearson correlation coefficient for each parameter, and obtaining the final ranking of the controlling factors.

[0102] Using the selected key features as input, 12 models were trained. A single-factor stepwise optimization method was employed to optimize hyperparameters, and the five best-performing models were selected for ensemble learning. Weight layers were introduced to stack multiple intelligent models to improve accuracy, and automatic training and hyperparameter optimization were implemented. Model evaluation was conducted using ACC as the core metric for comprehensive verification.

[0103] Comparing the production capacity prediction results using single-algorithm models, the results show that the intelligent production capacity allocation model integrating multiple algorithms significantly outperforms other methods in multi-layer synergistic mining scenarios. Single models have poor robustness; some algorithms excel at capturing global trends, while others excel at capturing local trends. In fracturing production prediction, a single model cannot cater to all these aspects. In contrast, the integrated production capacity prediction model combining multiple algorithms has strong generalization ability, not only improving prediction accuracy but also more intuitively demonstrating the contribution of each production layer in multi-layer synergistic mining scenarios. Furthermore, it can transfer and apply the production response patterns from old wells to each segment of new wells, guiding the fracturing operations of new wells.

[0104] Although this specification provides the following examples or appendices Figure 6 The method or apparatus structure shown may include more or fewer operational steps or modular units, depending on conventional or non-inventive methods. In steps or structures where there is no logically necessary causal relationship, the execution order of these steps or the modular structure of the apparatus is not limited to the execution order or modular structure shown in the embodiments or drawings of this specification. When the method or modular structure is applied in actual devices, servers, or terminal products, it can be executed sequentially or in parallel according to the method or modular 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 intelligent prediction method for post-pressure production capacity of multi-layer syndicated oil wells, this specification also proposes an embodiment of an intelligent prediction device for post-pressure production capacity of multi-layer syndicated oil wells. Figure 6 As shown, the device may specifically include the following modules: Training module 601 can be used to acquire training datasets corresponding to different geological units within a target reservoir area. The training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well in the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data are generated based on historical geological parameters and historical production data of each reservoir segment of the historical well. Based on the training datasets, a corresponding integrated production capacity prediction model is trained for each geological unit. Module 602 can be used to call the corresponding integrated production capacity prediction model based on the geological unit to which the target well belongs; The prediction module 603 can be used to input the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model and output the production capacity prediction value corresponding to each reservoir segment. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0105] In some embodiments, the training module 601 may further be used to acquire historical data of each historical well in each geological unit within the target reservoir area. The historical data includes historical production data, historical geological parameters of each reservoir segment, and historical fracturing operation parameters. The historical production data includes at least historical total production, and the historical geological parameters include at least historical permeability and historical effective thickness. Correspondingly, the training module 601 may be used to calculate the product of historical permeability and historical effective thickness for each reservoir segment of the historical well, and to calculate the sum of the products of all reservoir segments. Based on the ratio of the product of each reservoir segment to the sum of the products, the geological weight of each reservoir segment is determined. Based on the geological weight, the historical total production is divided among the reservoir segments to generate independent historical production data for each reservoir segment.

[0106] In some embodiments, the training module 601 can be specifically used to evaluate the historical geological parameters and historical fracturing operation parameters of each reservoir segment of a historical well using a variety of different feature importance evaluation methods; calculate the information entropy of each feature importance evaluation method based on the evaluation results of each parameter using each feature importance evaluation method, and perform a weighted summation of each evaluation result based on the information entropy; sort the historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well based on the weighted summation result, and select a preset number of parameters from the sorting results as the main controlling factors of historical production capacity.

[0107] In some embodiments, the various feature importance evaluation methods in the training module 601 described above may include at least two of the following: grey relational analysis, maximum mutual information method, SHAP interpretation method, Pearson correlation analysis, and Spearman correlation analysis.

[0108] In some embodiments, the integrated model for production capacity prediction in the training module 601 may include multiple sub-models with different algorithm types; correspondingly, the training module 601 may be specifically used to train and optimize each sub-model in the integrated model for production capacity prediction corresponding to each geological unit based on the training dataset and using a multi-round hyperparameter optimization method.

[0109] In some embodiments, the multiple sub-models of different algorithm types in the training module 601 described above may include at least two of the following: linear regression model, tree model, and gradient boosting model.

[0110] In some embodiments, the prediction module 603 can be specifically used to input the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model, perform a weighted average of the outputs of each sub-model in the corresponding integrated production capacity prediction model, and output the production capacity prediction value corresponding to each reservoir segment. The weighting weight of each sub-model is determined based on the prediction performance index of each sub-model on the validation set.

[0111] In some embodiments, after the prediction module 603, it can also be used to calculate the production contribution ratio of each reservoir segment based on the predicted production value of each reservoir segment; determine the potential main segment and potential secondary segment of the target well based on the production contribution ratio; generate a differentiated fracturing design scheme for the target well based on the potential main segment and potential secondary segment; wherein, a first fracturing construction parameter is configured for the potential main segment, and a second fracturing construction parameter is configured for the potential secondary segment, and the design strength of the first fracturing construction parameter is higher than that of the second fracturing construction parameter.

[0112] As can be seen from the above, the intelligent prediction device for post-fracturing production capacity of multi-layer synergistic oil wells provided in the embodiments of this specification not only has stronger generalization ability and higher prediction accuracy, but also can intuitively present the contribution of each production layer under multi-layer synergistic production, and can transfer the production response law of old wells to each section of new wells, providing guidance for fracturing construction of new wells.

[0113] This specification also provides an electronic device based on the above-described intelligent prediction method for post-fracturing production capacity of multi-layer syndicated oil wells, including a processor and a memory for storing executable programs / instructions. In specific implementation, the processor can execute the following steps according to the program / instructions: Based on the geological unit to which the target well belongs, the corresponding integrated production capacity prediction model is invoked; wherein, the integrated production capacity prediction model is trained in the following manner: For different geological units within the target reservoir area, training datasets corresponding to each geological unit are obtained. The training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well in the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data of each reservoir segment of the historical well; Based on the training dataset, a corresponding integrated production capacity prediction model is trained for each geological unit; The input feature vectors of each reservoir segment of the target well are input to the corresponding integrated production capacity prediction model, and the predicted production capacity value corresponding to each reservoir segment is output. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0114] To execute the above instructions more accurately, please refer to... Figure 7As shown in the embodiments of this specification, another specific electronic device is also provided, wherein the electronic device includes a network communication port 701, a processor 702, and a memory 703. The above structures are connected by internal cables so that the various structures can perform specific data interaction.

[0115] Specifically, the processor 702 can be used to call the corresponding integrated production capacity prediction model according to the geological unit to which the target well belongs. The integrated production capacity prediction model is trained as follows: for different geological units within the target reservoir area, training datasets corresponding to each geological unit are obtained. These training datasets consist of historical production capacity control factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity control factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data of each reservoir segment of the historical well. Based on the training datasets, a corresponding integrated production capacity prediction model is trained for each geological unit. The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model, and the predicted production capacity value for each reservoir segment is output. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0116] The memory 703 can be used to store the corresponding instruction program.

[0117] In this embodiment, the network communication port 701 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.

[0118] In this embodiment, the processor 702 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.

[0119] In this embodiment, the memory 703 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.

[0120] This specification also provides a computer storage medium based on the above-described intelligent prediction method for post-fracturing production capacity of multi-layer syngas wells. The computer storage medium stores a computer program / instruction, which, when executed, performs the following: Based on the geological unit to which the target well belongs, the corresponding integrated production capacity prediction model is invoked. The integrated production capacity prediction model is trained as follows: For different geological units within the target reservoir area, training datasets corresponding to each geological unit are obtained. These training datasets consist of historical production capacity control factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity control factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on the historical geological parameters and historical production data of each reservoir segment of the historical well. Based on the training dataset, a corresponding integrated production capacity prediction model is trained for each geological unit. The input feature vectors of each reservoir segment of the target well are input to the corresponding integrated production capacity prediction model, and the predicted production capacity value for each reservoir segment is output. The input feature vectors include the geological parameters and fracturing operation parameters of the target well.

[0121] 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.

[0122] 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.

[0123] 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.

[0124] 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.

[0125] 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.

[0126] 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.

[0127] 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.

[0128] 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 method for intelligent prediction of post-pressure production capacity of multi-layered syndicated oil wells, characterized in that, include: Based on the geological unit to which the target well belongs, the corresponding integrated production capacity prediction model is invoked. The integrated production capacity prediction model is trained as follows: For different geological units within the target reservoir area, training datasets corresponding to each geological unit are obtained. These training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters for each reservoir segment of the historical well. The historical production data is generated based on historical geological parameters and historical production data for each reservoir segment of the historical well. Based on the training datasets, a corresponding integrated production capacity prediction model is trained for each geological unit. The geological unit is an independent regional unit divided according to the consistency of geological conditions, including reservoir lithology, porosity distribution, permeability range, and structural features. The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model, and the production capacity prediction value of each reservoir segment is output. The input feature vectors include the geological parameters and fracturing operation parameters of the target well. The step of inputting the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model and outputting the production capacity prediction value corresponding to each reservoir segment includes: The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model. The outputs of each sub-model in the integrated production capacity prediction model are weighted and averaged to output the production capacity prediction value corresponding to each reservoir segment. The weight of each sub-model is determined based on the prediction performance index of each sub-model on the validation set. Among them, the accuracy based on the total absolute error is selected as the prediction performance index, and the weight of each sub-model is the ratio of the prediction performance index of the sub-model on the validation set to the prediction performance index of all sub-models on the validation set.

2. The method according to claim 1, characterized in that, The method further includes: Acquire historical data of each historical well in each geological unit within the target oil reservoir area. The historical data includes historical production data, historical geological parameters of each reservoir section, and historical fracturing operation parameters. The historical production data includes at least historical total production, and the historical geological parameters include at least historical permeability and historical effective thickness. Accordingly, the historical production data is generated based on historical geological parameters and historical production data of each reservoir section of historical wells, including: For each reservoir segment of a historical well, calculate the product of the historical permeability and the historical effective thickness of each reservoir segment, and then calculate the sum of the products of all reservoir segments. The geological weight of each reservoir segment is determined based on the ratio of the product of each reservoir segment to the sum of the products. Based on the geological weights, the total historical production is divided into different reservoir segments to generate independent historical production data for each reservoir segment.

3. The method according to claim 1, characterized in that, The key factors controlling historical production capacity were selected from historical geological parameters and historical fracturing operation parameters of each reservoir section of historical wells, including: Various methods for evaluating the importance of different characteristics were used to evaluate the historical geological parameters and historical fracturing operation parameters of each reservoir section of historical wells; Based on the evaluation results of each feature importance evaluation method for each parameter, the information entropy of each feature importance evaluation method is calculated, and the evaluation results are weighted and summed based on the information entropy. Based on the weighted summation results, the historical geological parameters and historical fracturing operation parameters of each reservoir section of the historical well are sorted, and a preset number of parameters are selected from the sorting results as the main controlling factors of historical production capacity.

4. The method according to claim 3, characterized in that, The various methods for evaluating the importance of different features include at least two of the following: grey relational analysis, maximum mutual information method, SHAP interpretation method, Pearson correlation analysis, and Spearman correlation analysis.

5. The method according to claim 1, characterized in that, The integrated production capacity prediction model includes multiple sub-models with different algorithm types; correspondingly, training a corresponding integrated production capacity prediction model for each geological unit based on the training dataset includes: Based on the training dataset, a multi-round hyperparameter optimization method is used to train and optimize each sub-model in the integrated production capacity prediction model corresponding to each geological unit.

6. The method according to claim 5, characterized in that, The sub-models of the various different algorithm types include at least two of the following: linear regression model, tree model, and gradient boosting model.

7. The method according to claim 1, characterized in that, The method further includes: Based on the predicted production capacity of each reservoir segment, calculate the production capacity contribution ratio of each reservoir segment. Based on the aforementioned production capacity contribution ratio, the potential primary and secondary formations of the target well are determined; Based on the potential primary and secondary stratigraphic segments, a differentiated fracturing design scheme is generated for the target well. Specifically, a first fracturing construction parameter is configured for the potential main fracturing layer, and a second fracturing construction parameter is configured for the potential secondary fracturing layer. The design strength of the first fracturing construction parameter is higher than that of the second fracturing construction parameter.

8. A smart prediction device for post-pressure production capacity of multi-layered syndicated oil wells, characterized in that, include: The training module is used to acquire training datasets for different geological units within the target reservoir area. These training datasets consist of historical production capacity controlling factors and historical production data for each reservoir segment of each historical well within the geological unit. The historical production capacity controlling factors are selected from historical geological parameters and historical fracturing operation parameters of each reservoir segment of the historical well. The historical production data is generated based on historical geological parameters and historical production data of each reservoir segment of the historical well. Based on the training datasets, a corresponding integrated production capacity prediction model is trained for each geological unit. Each geological unit is an independent regional unit defined based on the consistency of geological conditions, including reservoir lithology, porosity distribution, permeability range, and structural features. The calling module is used to call the corresponding integrated production capacity prediction model based on the geological unit to which the target well belongs; The prediction module is used to input the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model and output the production capacity prediction value corresponding to each reservoir segment. The input feature vectors include the geological parameters and fracturing operation parameters of the target well. The step of inputting the input feature vectors of each reservoir segment of the target well into the corresponding integrated production capacity prediction model and outputting the production capacity prediction value corresponding to each reservoir segment includes: The input feature vectors of each reservoir segment of the target well are input into the corresponding integrated production capacity prediction model. The outputs of each sub-model in the integrated production capacity prediction model are weighted and averaged to output the production capacity prediction value corresponding to each reservoir segment. The weight of each sub-model is determined based on the prediction performance index of each sub-model on the validation set. Among them, the accuracy based on the total absolute error is selected as the prediction performance index, and the weight of each sub-model is the ratio of the prediction performance index of the sub-model on the validation set to the prediction performance index of all sub-models on the validation set.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images