A well pattern design method based on a tight oil hydraulic fracturing test field coring well group

Abstract

This invention discloses a well pattern design method based on cored well groups from a hydraulic fracturing test field in tight oil, relating to the field of oil and gas exploration and development technology. The method includes: acquiring three-dimensional scanning data of the core samples from the cored well group and establishing a standardized three-dimensional quantitative model of reservoir multi-attribute characteristics; using deep learning to extract digital results of sand-mud interbedded layers, fractures, porosity, and oil-bearing properties, constructing an intelligent large-scale model and a deep learning standard model to determine the reservoir's three-dimensional configuration model; constructing a well pattern model and evaluation criteria based on the reservoir's three-dimensional configuration model; determining the well spacing and fracture half-fracture length based on the proppant and tracer distribution of fractured wells and monitoring wells in the hydraulic fracturing test field, and establishing its deep learning standard model and evaluation criteria; finally, optimizing the well pattern type based on the well pattern evaluation criteria and the well spacing-half-fracture length evaluation criteria. This invention improves reservoir characterization accuracy and well pattern adaptability, significantly enhances tight oil development efficiency and economic benefits, and reduces development costs.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration and development technology, specifically to a well network design method based on coring well groups in a tight oil hydraulic fracturing test field. Background Technology

[0002] The design of tight oil well patterns has long been a world-class technical challenge that oil and gas development projects have long needed to address but have not yet fundamentally solved.

[0003] The well pattern design method for tight oil development is mainly a three-dimensional well pattern design guided by integrated geological engineering, aiming to maximize reserve utilization and optimize economic benefits. It features a three-dimensional staggered well pattern, multi-layered integrated development, proactive utilization and optimization of inter-well stress interference, small well spacing, long horizontal sections, dense cutting, and an emphasis on an integrated fracture network. However, tight oil reservoirs exhibit strong macroscopic heterogeneity both between layers and microscopic heterogeneity within layers, resulting in poor well pattern adaptability. Key issues include the inability of existing well spacing to effectively control sand bodies and the inability of fracturing fractures to completely cover the reservoir. For example, in a certain oilfield case, the poor adaptability of a diamond-shaped inverted nine-point well pattern led to poor development results. Current well pattern design methods suffer from several problems: mismatch between well pattern and reservoir; excessively large well spacing; low reservoir control; incompatibility between well pattern and reservoir characteristics; lack of coordination between well pattern and fracturing fractures; inability of fracturing fractures to effectively cover inter-well areas; incoordination between fracture orientation and well pattern deployment, which can easily lead to cross-contamination; gap between geological understanding and engineering implementation; insufficient accuracy of geological models; mismatch between fracturing stimulation and geological conditions; prominent economic and technical contradictions; the contradiction between small well spacing for improved production and high costs; and the complexity of multi-layered, three-dimensional development, resulting in poor tight oil development performance and consequently, low efficiency in tight oil development.

[0004] Therefore, there is an urgent need to provide a well pattern design method for efficient tight oil development, so as to improve the efficiency of tight oil development. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the aforementioned background technology and provide a deep learning design method for optimizing tight oil well patterns based on core sample detection results such as sand-mud interbedded layers and fracture networks. This method addresses the technical problems in existing technologies where inaccurate detection of reservoir heterogeneity leads to low accuracy in obtaining the spatial condition of crude oil in tight oil reservoirs, resulting in low accuracy in tight oil development strategies and low well pattern matching. This invention improves the applicability and accuracy of tight oil reservoir well pattern design, increases the efficiency of tight oil development, and reduces the cost of tight oil development.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention provides a well pattern design method based on cored well groups from tight oil hydraulic fracturing test sites, comprising: Obtain three-dimensional scanning data of the core samples from the core well group, and establish a standardized three-dimensional quantitative model of reservoir multi-attribute characteristics based on the three-dimensional scanning data; Based on the standardized three-dimensional quantitative model, deep learning is used to extract the digital results of reservoir multi-attribute features, and an intelligent large model of reservoir multi-attribute feature parameters is constructed. Based on the aforementioned intelligent large model, a deep learning standard model of reservoir multi-attribute characteristics is established through deep learning, and the reservoir three-dimensional configuration model is determined accordingly. Based on the reservoir three-dimensional configuration model, a well type model and well type evaluation criteria are constructed through deep learning; Based on the test results from the hydraulic fracturing test site, the well spacing and the half-fracture length of the hydraulic fracturing were determined, and a deep learning standard model and evaluation criteria for the well spacing and the half-fracture length of the hydraulic fracturing were established. Based on the well type evaluation criteria and the well spacing and fracture half-fracture length evaluation criteria, the well pattern type is optimally determined.

[0007] As a preferred technical solution of the present invention, the establishment of a standardized three-dimensional quantitative model of reservoir multi-attribute characteristics includes: applying an image coordination algorithm to compare and calibrate the grayscale image of the tight oil layer obtained by focused ion beam scanning electron microscopy with the three-dimensional grayscale image of the tight oil layer obtained by CT scanning, and establishing the standardized three-dimensional quantitative model based on the comparison results; wherein, the reservoir multi-attribute characteristics include sand-mud interlayers, fractures, porosity, and oil content.

[0008] As a preferred technical solution of the present invention, the digital results of extracting reservoir multi-attribute features using deep learning include: using the U-Net convolutional neural network algorithm to extract digital result models of sand-mud interlayers, fractures, porosity, and oil content of horizontal well oil layers; the digital results of reservoir multi-attribute features include the percentage content of sand-mud interlayers, the density of sand-mud interlayers, fracture aperture, porosity, and oil saturation of the oil layer.

[0009] As a preferred technical solution of the present invention, the construction of a large intelligent model of reservoir multi-attribute characteristic parameters includes: constructing a large intelligent model of sand-mud interbedded layers, fractures, porosity, and oil-bearing characteristic parameters based on datasets of sand-mud interbedded layers, fractures, porosity, and oil-bearing properties under different geological conditions and different genetic types.

[0010] As a preferred technical solution of the present invention, the step of establishing a deep learning standard model of reservoir multi-attribute features through deep learning includes: using a convolutional neural network as the deep learning model architecture, using the reservoir multi-attribute feature data as training data, adjusting the model parameters through a backpropagation algorithm, and creating the deep learning standard model.

[0011] As a preferred technical solution of the present invention, the determination of the reservoir three-dimensional configuration model includes: obtaining the digital results of reservoir multi-attribute characteristics and their judgment criteria based on the deep learning standard model, and determining the evaluation criteria for different types of tight oil layers; based on the evaluation criteria, establishing a reservoir three-dimensional configuration model including sand-mud interbedded layer type, fracture type, and oil-bearing type.

[0012] As a preferred technical solution of the present invention, the step of constructing well type models and well type evaluation standards through deep learning includes: using a convolutional neural network architecture, using the reservoir three-dimensional configuration model as a training model, and constructing well type models and well type evaluation standards for different types of tight oil layers.

[0013] As a preferred technical solution of the present invention, the determination of well spacing and fracture half-fracture length based on the test results of the hydraulic fracturing test field includes: determining the distribution of fracturing proppant and tracer based on the sealed core and cuttings test results of the fracturing wells and test wells in the hydraulic fracturing test field, thereby determining the well spacing and fracture half-fracture length.

[0014] As a preferred technical solution of the present invention, the establishment of a deep learning standard model of well spacing and fracture half-length includes: applying the U-Net convolutional neural network algorithm to establish a standardized quantitative model of well spacing and fracture half-length for different types of tight oil layers, and creating the deep learning standard model of well spacing and fracture half-length through deep learning.

[0015] As a preferred technical solution of the present invention, the establishment of the evaluation criteria for well spacing and fracture half-length includes: applying a convolutional neural network algorithm to obtain datasets of well spacing and fracture half-length of tight oil reservoirs under different geological conditions and of different genesis, constructing an intelligent large model of characteristic parameters of well spacing and fracture half-length of different types of tight oil reservoirs, and determining the evaluation criteria for well spacing and fracture half-length based on the intelligent large model.

[0016] Compared with the prior art, the beneficial effects of the present invention include: This invention integrates three-dimensional grayscale images from focused ion beam scanning electron microscopy and CT scans, and applies image coordination algorithms to establish a standardized three-dimensional quantitative model. This achieves accurate, multi-scale quantitative description of multiple attributes such as sand-mud interlayers, fractures, porosity, and oil content, overcoming the inaccuracy of traditional reservoir heterogeneity detection. Utilizing deep learning algorithms such as U-Net and convolutional neural networks, this invention automatically extracts digital results of multiple reservoir attributes, constructs intelligent large-scale models and deep learning standard models, and then establishes three-dimensional reservoir configuration models, well type models, and well spacing-fracture half-fracture length models, achieving intelligent, standardized, and transferable well pattern design. Based on the three-dimensional reservoir configuration model and deep learning standard model, this invention establishes well type evaluation standards and well spacing-fracture half-fracture length evaluation standards. It can optimize appropriate well types, well spacing, and row spacing for different types of tight oil reservoirs, significantly improving the matching degree between well patterns and reservoirs and avoiding the problem of poor adaptability of traditional well patterns. Based on the proppant and tracer distribution measured in the hydraulic fracturing test field, the half-fracture length is accurately determined, and a deep learning standard model of well spacing and half-fracture length is established to ensure the coordinated optimization of well network deployment and fracturing network, improve fracture network coverage and reserve utilization, and reduce inter-well interference and fracturing crossover risks. This invention reduces the uncertainty of well network design through high-precision reservoir modeling and multi-objective optimization (maximizing reserve utilization, maximizing production capacity, minimizing costs, etc.), avoids inefficient investment due to unreasonable well networks, shortens the investment payback period, and improves the internal rate of return. After applying this method in the Chang 8 tight oil test field in the Ordos Basin, the initial production capacity of a single well increased by 16.2%, the EUR of a single well increased by 15.7%, the number of wells decreased by 21.3%, and the overall development cost decreased by 12.5%. In multiple major basins, the field application compliance rate reached over 90%, driving a reduction in overall development cost of over 10% and an average increase in single-well production capacity of over 15%. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0018] Figure 1 This is a grayscale image of the dense oil-sand-mud interlayers and fractures obtained from core scanning.

[0019] Figure 2 To obtain a three-dimensional digital percentage curve of dense oil-sand-mud interlayer using deep learning.

[0020] Figure 3 To obtain a three-dimensional digital quantitative model of the bedding fractures and fracture porosity curves of the interlayered oil-sand-mud layer using deep learning.

[0021] Figure 4 To obtain a three-dimensional digital quantitative model of the oil content of tight oil and a digital quantitative result curve of oil saturation through deep learning.

[0022] Figure 5 This is a diagram showing the test results of hydraulic fracturing in tight oil.

[0023] Figure 6 This document presents a well pattern design diagram for a well pattern design method based on cored well groups in a tight oil hydraulic fracturing test field. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] In a typical embodiment of this invention, a well network design method based on core well groups from a hydraulic fracturing test site for tight oil is provided. The specific embodiment of this invention is illustrated in detail using a core well group from a hydraulic fracturing test site in a basin as an example. This test site is located in the southwest of the basin, with the main development stratum being the Chang 8 section of the Triassic Yanchang Formation. It is a typical continental tight oil reservoir with extremely high reservoir heterogeneity, featuring millimeter-centimeter-scale development of interbedded sand and mud layers, complex and significantly heterogeneous natural fracture distribution, an average reservoir porosity of 8.5%, an average permeability of only 0.23 mD, and an average oil saturation of 45%. Previously, when using conventional diamond-shaped inverse nine-point well networks for development, problems such as mismatch between the well network and the reservoir, insufficient fracture coverage, and severe inter-well interference existed, resulting in low single-well productivity and poor development performance. This invention, by integrating core 3D scanning technology, deep learning algorithms, and measured data from the hydraulic fracturing test site, achieves refined and intelligent design of the well network for tight oil reservoirs, effectively solving the aforementioned technical problems.

[0026] Step 1: Establishment of a standardized three-dimensional quantitative model of reservoir multi-attribute characteristics The core of this step is to construct a high-precision, standardized three-dimensional quantitative model of reservoir multi-attribute using multi-scale core scanning technology and image coordination algorithms, providing basic data for subsequent deep learning analysis.

[0027] 1. Three-dimensional core scanning data acquisition: Twelve cored wells from Well X in the tight oil hydraulic fracturing test field were selected, including four fracturing wells and eight monitoring wells. Full-section closed-loop core sampling was performed on eight sections of the main oil-bearing layer (depth range 2550m-2590m). Industrial CT scanning equipment was used to perform three-dimensional scanning of all cores at a resolution of 50μm, obtaining three-dimensional grayscale images of the tight oil layer. The images clearly identified millimeter-level sand-mud interlayers and micron-level pore structures. Simultaneously, representative core samples were selected and nanoscale scanning was performed using focused ion beam scanning electron microscopy (FIB-SEM) at a resolution of 10nm, obtaining microscopic grayscale images of the tight oil layer to characterize nanoscale pores and microfractures.

[0028] 2. Image Preprocessing and Quantitative Model Extraction: Digital core analysis software was used to preprocess the grayscale images acquired from CT and FIB-SEM scans. This included image denoising, contrast enhancement, artifact removal, and signal-to-noise ratio improvement to eliminate noise and interference generated during the scanning process. Using image segmentation algorithms, three-dimensional quantitative models of sand-mud interlayers, fractures, porosity, and oil content were extracted from the preprocessed grayscale images, resulting in sand body distribution models, fracture network models, porosity distribution models, and oil content distribution models, respectively.

[0029] 3. Multi-scale Image Coordination and Calibration: An image coordination algorithm is applied to spatially register and compare nanoscale grayscale images acquired by FIB-SEM with micrometer-scale three-dimensional grayscale images acquired by CT scans. By comparing the sand-mud interlayer boundaries, fracture morphology, pore structure, and oil-bearing characteristics at the same locations in the two images, the correspondence between grayscale values ​​of CT scan images and reservoir properties is calibrated and verified, using the high-precision measurement results of FIB-SEM as a benchmark. For example, FIB-SEM is used to determine the porosity and oil saturation corresponding to different grayscale values, establishing a quantitative conversion relationship between grayscale values ​​and reservoir properties, and correcting the quantitative interpretation errors of CT scan images.

[0030] 4. Construction of a Standardized 3D Quantitative Model: Based on the comparison and calibration results of multi-scale images, scanning data from all cored wells were integrated to establish a standardized 3D quantitative model of reservoir multi-attribute characteristics covering the entire hydraulic fracturing test site. This model has a planar range of 400m × 400m, vertically covering eight main oil-bearing segments with a thickness of approximately 20m, and a resolution of 50μm. It can quantitatively characterize the spatial distribution of sand-mud interbedded layers, fracture geometry, three-dimensional porosity variations, and the heterogeneity of oil-bearing properties.

[0031] Step 2: Digital extraction of reservoir multi-attribute features and construction of intelligent large model This step utilizes the U-Net convolutional neural network algorithm to automatically extract digital results of reservoir multi-attribute features from a standardized three-dimensional quantitative model, and constructs an intelligent large model of reservoir multi-attribute feature parameters covering different geological conditions.

[0032] 1. U-Net Convolutional Neural Network Model Construction: The U-Net convolutional neural network is used as the feature extraction model. This model consists of an encoder and a decoder. The encoder extracts deep features of the image through convolutional and pooling layers, while the decoder restores the spatial resolution of the image through upsampling layers and skip connections. It is suitable for processing geological image segmentation and feature extraction tasks, and can effectively identify small targets and edge features such as interbedded sand and mud layers and fractures. The model input is a grayscale image slice of a standardized 3D quantitative model, and the output is the corresponding reservoir multi-attribute feature segmentation image.

[0033] 2. Digital Extraction of Reservoir Multi-Attribute Features: The standardized three-dimensional quantitative model of the test site is input into the U-Net model. Through automatic segmentation and identification of the model, digital results models of sand-mud interbedded layers, fractures, porosity, and oil content of horizontal well oil reservoirs are extracted. Specific extracted quantitative parameters include: sand-mud interbedded layer percentage (%), sand-mud interbedded layer density (layers / meter), fracture aperture (μm), fracture density (fractures / meter), porosity (%), and oil saturation (%). The extraction results are output in the form of three-dimensional data volumes and two-dimensional curves, for example... Figure 2 The three-dimensional digital percentage curve of the sand-mud interlayer shown is as follows: Figure 3 The crack porosity curves shown are Figure 4 The curve showing the digital quantitative result of oil saturation is shown.

[0034] 3. Model Training and Validation: Manually labeled data from cored wells in the test field were used as the training and test sets, with the training set accounting for 80% and the test set for 20%. The U-Net model underwent multiple rounds of iterative training, continuously adjusting the model's weights and bias parameters using the backpropagation algorithm to optimize the model's feature extraction accuracy. After training, the model was validated using the test set data. The results showed that the model's accuracy in identifying sand-mud interbedded layers was 92.3%, the consistency rate in characterizing fractures was 98.7%, and the prediction errors for porosity and oil saturation were less than 5% and 6%, respectively, meeting the requirements for engineering applications.

[0035] 4. Construction of a Large-Scale Intelligent Model for Multi-Attribute Reservoir Features: Based on the test field data, core scanning data of tight oil reservoirs with different geological conditions and genetic types from multiple basins were collected. This included datasets on sand-mud interbedded layers, fractures, porosity, and oil-bearing properties of reservoirs with different genesis, such as delta front sand bodies, underwater distributary channel sand bodies, and mouth bar sand bodies, totaling over 10TB of data. Based on these multi-source heterogeneous datasets, the pre-trained U-Net model was fine-tuned using transfer learning to construct a large-scale intelligent model for multi-attribute feature parameters of reservoirs covering major tight oil reservoir types nationwide. This model can be directly applied to the extraction of reservoir features from tight oil reservoirs in different basins without requiring extensive data annotation and model retraining, significantly improving work efficiency.

[0036] Step 3: Establishment of a standard deep learning model for reservoir multi-attribute features This step is based on a large intelligent model of reservoir multi-attribute features. It adopts a convolutional neural network architecture to establish a general deep learning standard model for reservoir multi-attribute features, providing a unified judgment standard for the construction of reservoir three-dimensional configuration models.

[0037] 1. Deep Learning Model Architecture Design: An improved convolutional neural network architecture is adopted, including an input layer, 12 convolutional layers, 6 pooling layers, 6 upsampling layers, and an output layer. The convolutional layers use 3×3 convolutional kernels with a stride of 1 and the ReLU activation function to extract nonlinear features of multiple reservoir attributes. The pooling layers use 2×2 max pooling with a stride of 2 to reduce feature dimensionality and computational cost. The output layer uses the Softmax activation function (for classification tasks) and the linear activation function (for regression tasks) to output the classification results and quantitative values ​​of reservoir attributes, respectively.

[0038] 2. Model Training and Parameter Optimization: All data from the intelligent large-scale model of reservoir multi-attribute features were used as training data to train the aforementioned convolutional neural network model. Batch gradient descent was employed during training, with a batch size of 32, a learning rate of 0.001, and 100 training epochs. The model parameters were continuously adjusted using backpropagation to minimize the loss function; cross-entropy loss was used for classification tasks, and mean squared error loss was used for regression tasks. To prevent overfitting, Dropout was employed, randomly discarding 20% ​​of the neurons during training.

[0039] 3. Model Cross-Validation: A 5-fold cross-validation method was used to validate the model's generalization ability. All datasets were randomly divided into 5 parts, with 4 parts selected as the training set and 1 part as the validation set in each iteration. This training and validation process was repeated 5 times, and the average of the 5 validation results was taken as the model's final performance metric. Validation results show that the model has good generalization ability on tight oil reservoir data from different basins and of different genetic types, with an average classification accuracy of 90.2% and an average regression prediction error of less than 7%.

[0040] 4. Establishment of a Deep Learning Standard Model: Based on the training and validation results, a deep learning standard model for reservoir multi-attribute characteristics was established. This model can automatically input grayscale images from core scans, output standardized digital results of reservoir multi-attribute characteristics, and provide unified judgment criteria, such as the sand-mud interbedded type, fracture type, porosity grade, and oil-bearing grade corresponding to different grayscale values, laying the foundation for the subsequent construction of reservoir three-dimensional configuration models.

[0041] Step 4: Construction of the three-dimensional reservoir configuration model This step is based on the deep learning standard model of reservoir multi-attribute features, formulates evaluation criteria for different types of tight oil layers, and establishes a high-precision three-dimensional reservoir configuration model to achieve a fine characterization of the internal structure of the reservoir.

[0042] 1. Development of Evaluation Standards for Tight Oil Reservoirs: Based on the digital results of multi-attribute reservoir characteristics output by the deep learning standard model, and combined with the actual needs of tight oil reservoir development, comprehensive evaluation standards for different types of tight oil reservoirs are developed. Sand-mud interlayer classification: According to the percentage content of sand-mud interlayer, it is divided into high content type (>30%), medium content type (10%-30%), and low content type (<10%); according to the density of sand-mud interlayer, it is divided into dense type (>5 layers / meter), medium type (2-5 layers / meter), and sparse type (<2 layers / meter).

[0043] Crack classification: According to crack aperture, cracks are classified into microcracks (<10μm), small cracks (10-50μm), medium cracks (50-100μm), and large cracks (>100μm); according to crack origin, cracks are classified into bedding cracks, tectonic cracks, and diagenetic cracks.

[0044] Oil content classification: According to oil saturation, it is divided into high oil content type (>60%), medium oil content type (30%-60%), and low oil content type (<30%).

[0045] Comprehensive reservoir classification: Based on the combined characteristics of sand-mud interbedded layers, fractures, and oil content, tight oil reservoirs are divided into Class I high-quality reservoirs (low sand-mud interbedded layer content, well-developed fractures, high oil content), Class II medium-quality reservoirs (medium sand-mud interbedded layer content, relatively well-developed fractures, medium oil content), and Class III poor-quality reservoirs (high sand-mud interbedded layer content, poorly developed fractures, low oil content).

[0046] 2. Establishment of a 3D Reservoir Configuration Model: Based on the aforementioned evaluation criteria, the digitized results of the multi-attribute characteristics of the reservoir in the test field were classified and 3D modeled. Using geostatistical methods, combined with measured data from core wells and seismic data, interpolation and prediction of reservoir attributes between wells were performed to establish a 3D reservoir configuration model that includes sand-mud interbedded layers, fracture types, and oil-bearing types. This model has a resolution of 1m × 1m × 0.1m, clearly demonstrating the sand body stacking relationships, fracture network distribution, and spatial variations of oil-bearing properties within the reservoir, accurately characterizing the macroscopic and microscopic heterogeneity of the reservoir.

[0047] 3. Model Validation and Correction: The 3D reservoir configuration model was validated using measured data from three core wells not involved in the modeling process at the test site. Comparing the model predictions with the measured results, the sand body boundary agreement rate was 91.5%, the fracture distribution agreement rate was 89.3%, and the oil saturation prediction error was 6.2%. The model accuracy meets the well network design requirements. For areas with larger errors, the model was iteratively corrected by supplementing core scanning data and adjusting interpolation parameters to further improve model accuracy.

[0048] Step 5: Construction of well type model and well type evaluation criteria This step is based on the three-dimensional reservoir configuration model. It uses convolutional neural networks and multimodal large model technology to construct well type models and well type evaluation criteria for different types of tight oil reservoirs, so as to achieve the optimal matching between well type and reservoir.

[0049] 1. Multimodal Well Type Model Construction: A well type prediction model is constructed using an architecture that fuses convolutional neural networks with a multimodal large model. The model inputs include 3D reservoir configuration data (sand-mud interbedded layer type, fracture type, oil-bearing type), geomechanical data (stress direction, rock mechanical parameters), and historical production data (production capacity, water cut, and recovery rate for different well types). The model output is the optimal well type corresponding to different reservoir types. The multimodal large model is used to fuse different types of input data, explore potential correlations between data, and improve the model's prediction accuracy.

[0050] 2. Model Training and Optimization: Reservoir and production data from over 1000 tight oil wells in the Ordos, Songliao, and Junggar basins were collected to construct a training dataset. The well types included vertical wells, directional wells, horizontal wells, fractured horizontal wells, and cluster wells. The model underwent multiple training runs, and model parameters were adjusted using the backpropagation algorithm to optimize its predictive performance. After training, the model was validated using a test set. The results showed that the model's prediction accuracy for the optimal well type was 93.5%.

[0051] 3. Well Type Evaluation Standards: Based on the technical and economic requirements of tight oil reservoir development, comprehensive well type evaluation standards are formulated. The evaluation indicators include three categories: Production capacity indicators: initial production capacity of a single well, cumulative production of a single well, recoverable reserves of a single well (EUR), and reserve utilization level.

[0052] Development indicators: rate of increase in water cut, waterless oil recovery period, ultimate recovery rate, degree of inter-well interference.

[0053] Economic indicators: drilling costs, fracturing costs, payback period, internal rate of return (IRR), and net present value (NPV).

[0054] The weights of each evaluation indicator were determined using the analytic hierarchy process (AHP), with production capacity having a weight of 0.4, development indicators having a weight of 0.3, and economic indicators having a weight of 0.3. Based on the comprehensive evaluation score, well types were categorized into four levels: excellent, good, average, and poor. A score ≥80 was considered excellent, 60-80 was good, 40-60 was average, and <40 was poor.

[0055] 4. Determination of Optimal Well Types for Different Reservoir Types: Based on well type models and evaluation criteria, the optimal well types corresponding to different types of tight oil reservoirs are determined. Class I high-quality reservoirs: Prefer horizontal wells with long horizontal sections and segmented fracturing, with horizontal sections of 1500-2000m in length and 15-20 segments, which can make full use of natural fractures and improve the reservoir utilization rate.

[0056] Class II medium-sized reservoirs: Prefer horizontal wells with medium-length horizontal sections that are fractured in stages, with a horizontal section length of 1000-1500m and 10-15 stages, balancing production capacity and cost.

[0057] Class III poor reservoirs: cluster directional wells or short horizontal wells are preferred, with a horizontal section length of 500-1000m and 5-10 sections, to reduce development costs and improve economic benefits.

[0058] Step 6: Determining the well spacing and fracture half-length, and establishing the deep learning standard model. This step, based on measured data from the hydraulic fracturing test site, determines the quantitative relationship between well spacing and fracture half-length, and establishes a deep learning standard model for well spacing and fracture half-length to achieve collaborative optimization of well spacing and fracturing fracture.

[0059] 1. Hydraulic fracturing test site implementation: A large-scale hydraulic fracturing test was conducted at the tight oil hydraulic fracturing test site, with 4 fracturing wells (X-2-3X9, X-2-3X10, X-20-38X, X-20-39X) and 8 monitoring wells. The fracturing operation used a fracturing fluid system of slickwater + ceramic proppant, with a single-well fracturing fluid volume of 1500-2000 m³ / h. 3 The proppant dosage is 100-150 mg / m³. 3 The construction discharge volume is 8-10m³. 3The fracturing rate is 50-70 MPa per minute. During the fracturing process, different types of tracers are injected into different fracturing sections to monitor the direction and length of the fractures.

[0060] 2. Actual Measurement of Fracture Half-Length: After the fracturing operation, sealed core sampling and cuttings testing were conducted on all monitoring wells. Fluorescence microscopy and scanning electron microscopy (SEM) energy dispersive spectroscopy were used to analyze the distribution of proppant (ceramsite) and tracers in the cores and cuttings. Based on the location of proppant and tracers in the cores at different depths of the monitoring wells, and considering the distance between the fracturing wells and the monitoring wells, the half-length and extension direction of the fracturing fractures were calculated. The results show that the half-length of the fracturing fractures in the Chang 8 tight oil reservoir in the Ordos Basin is mainly distributed between 180-220 m, with an average of 200 m. The fracture extension direction is mainly northeast 67°-75°, consistent with the direction of the maximum principal stress in the region.

[0061] 3. Establishment of a Standardized Quantitative Model for Well Spacing and Fracture Half-Length: A standardized quantitative model for well spacing and fracture half-length was established using the U-Net convolutional neural network algorithm. The model's input parameters include reservoir configuration parameters (sand-mud interbedded content, fracture density, porosity), geomechanical parameters (stress difference, rock brittleness index, Young's modulus), and fracturing operation parameters (fracturing fluid volume, proppant volume, and displacement). The model's output parameters are the fracture half-length and the optimal well spacing. The model was trained and validated using measured data from the test field. The results show that the model's prediction error for fracture half-length is less than 8%, and its prediction error for optimal well spacing is less than 10%.

[0062] 4. Construction of a Standard Deep Learning Model for Well Spacing and Fracture Half-Length: Based on test field data, experimental data from over 500 tight oil fracturing wells in four major basins across China were collected to construct a dataset covering different geological conditions and construction parameters. The U-Net model was fine-tuned using transfer learning to establish a general standard deep learning model for well spacing and fracture half-length. This model can quickly predict fracture half-length and optimal well spacing based on the geological characteristics of different reservoirs and fracturing conditions, providing a basis for well pattern design.

[0063] Step 7: Establishment of evaluation criteria for well spacing and fracture half-length This step is based on a deep learning standard model of well spacing and fracture half-length. It constructs an intelligent large model of the characteristic parameters of well spacing and fracture half-length and formulates a comprehensive evaluation standard to ensure the synergy between well spacing and fracture.

[0064] 1. Construction of a large-scale intelligent model for well spacing and fracture half-fracture length: Utilizing a convolutional neural network algorithm, this model integrates datasets of well spacing and fracture half-fracture length in tight oil reservoirs under different geological conditions and genetic types, including both successful and failed cases, to construct a large-scale intelligent model of the characteristic parameters of well spacing and fracture half-fracture length. This model contains over 1000 sets of data corresponding to the relationship between well spacing and fracture half-fracture length, covering different reservoir types, fracturing techniques, and development methods.

[0065] 2. Development of Evaluation Criteria for Well Spacing and Fracture Half-Length: Based on the analysis results of the intelligent large-scale model and combined with the technical and economic requirements for tight oil reservoir development, a comprehensive evaluation standard for well spacing and fracture half-length was developed. The evaluation indicators include: Fracture network coverage: The fracture network is required to cover ≥95% of the reservoir to ensure that the reserves are fully utilized.

[0066] Inter-well interference level: Control inter-well interference within a reasonable range to avoid communication between fracturing fractures leading to cross-linking. The inter-well interference coefficient should be ≤0.3.

[0067] Reserve utilization rate: It is required that the utilization rate of the controlled reserves of a single well is ≥85% to improve resource utilization.

[0068] Economic indicators: The unit capacity investment is required to be ≤3000 yuan / ton, and the investment payback period is ≤5 years.

[0069] Based on the comprehensive evaluation score, the matching degree between well spacing and fracture half-length is divided into four levels: excellent, good, medium, and poor. A score ≥85 is excellent, 70-85 is good, 50-70 is medium, and <50 is poor.

[0070] 3. Determination of Optimal Well Spacing Range for Different Reservoir Types: Based on evaluation criteria, the optimal well spacing range for different types of tight oil reservoirs is determined. Type I high-quality reservoir: fractures are well-developed, with large extension lengths of hydraulic fractures. The optimal well spacing is 200-250m, and the row spacing is 400-500m.

[0071] Type II medium-sized reservoirs: fractures are relatively well-developed, with moderate extension length of hydraulic fractures. The optimal well spacing is 150-200m, and the row spacing is 300-400m.

[0072] Class III poor reservoirs: fractures are not well developed, the extension length of hydraulic fractures is small, the optimal well spacing is 120-150m, and the row spacing is 240-300m.

[0073] Step 8: Optimization and Determination of Well Pattern Type This step is based on well type evaluation criteria and well spacing and fracture half-fracture length evaluation criteria. A multi-objective optimization algorithm is used to select the optimal well pattern suitable for different tight oil layers, and the results are verified by production data from the test field.

[0074] 1. Multi-objective optimization model construction: A well network optimization model is constructed using a non-dominated sorting genetic algorithm (NSGA-II). The optimization objectives include: maximizing reserve utilization, maximizing single-well productivity, minimizing development costs, and minimizing the investment payback period. Constraints include: fracture network coverage ≥ 95%, inter-well interference coefficient ≤ 0.3, fracturing operation safety, and horizontal well orientation perpendicular to the direction of maximum principal stress.

[0075] 2. Comparative Analysis of Different Well Patterns: Based on the geological characteristics of the tight oil test site, various well pattern types were designed for comparative analysis, including rhomboid inverted nine-point well patterns, square well patterns, rectangular well patterns, staggered horizontal well patterns, and three-dimensional well patterns. Evaluation indicators for each well pattern type were calculated, and a comprehensive score was obtained using the analytic hierarchy process (AHP).

[0076] 3. Determination of Optimal Well Pattern: Comparative results show that the rectangular staggered horizontal well pattern has the highest comprehensive score of 92 points, making it the optimal well pattern for this test site. Specific well pattern parameters are: horizontal section length 1500m, well spacing 200m, row spacing 400m, horizontal well azimuth 160° NW, perpendicular to the direction of the maximum principal stress in the region (70° NE). This well pattern allows for uniform fracture network coverage of the entire reservoir, achieving a fracture network coverage of 96.8%, a reserve utilization rate of 88.5%, and an inter-well interference coefficient of 0.25, meeting development requirements. The well pattern design diagram is shown below. Figure 6 As shown.

[0077] 4. Model Validation and Iterative Correction: The optimized well pattern was applied to the development of the test site, and the well pattern model was validated using dynamic production data. Test results showed that after adopting this well pattern design method, the initial production capacity of a single well in the test site reached 18.5 tons / day, an increase of 16.2% compared to conventional well patterns; the EUR (Earnings per Hour) of a single well reached 12,000 tons, an increase of 15.7% compared to conventional well patterns; the number of wells was reduced by 21.3%, and the overall development cost decreased by 12.5%, achieving significant development results. Based on the dynamic production data, the parameters of the well pattern model were iteratively corrected to further optimize the model accuracy, forming a modular well pattern design model that can be widely applied to tight oil reservoirs in different basins.

[0078] This invention establishes a complete intelligent design method for tight oil well networks by integrating 3D core scanning technology, deep learning algorithms, and measured data from hydraulic fracturing test sites. This method achieves integrated refinement of reservoir characterization, well type optimization, and coordinated design of well spacing and fracturing fractures, significantly improving the accuracy and applicability of tight oil reservoir well network design. Currently, this method has been deployed and validated in four major basins: Ordos, Songliao, Junggar, and Bohai Bay, with a field application compliance rate exceeding 90%. It has driven a reduction of over 10% in the overall cost of tight oil development and an average increase of over 15% in single-well productivity, providing strong technical support for the large-scale and efficient development of tight oil in my country.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A well pattern design method based on cored well groups from tight oil hydraulic fracturing test sites, characterized in that, include: Obtain three-dimensional scanning data of the core samples from the core well group, and establish a standardized three-dimensional quantitative model of reservoir multi-attribute characteristics based on the three-dimensional scanning data; Based on the standardized three-dimensional quantitative model, deep learning is used to extract the digital results of reservoir multi-attribute features, and an intelligent large model of reservoir multi-attribute feature parameters is constructed. Based on the aforementioned intelligent large model, a deep learning standard model of reservoir multi-attribute characteristics is established through deep learning, and the reservoir three-dimensional configuration model is determined accordingly. Based on the reservoir three-dimensional configuration model, a well type model and well type evaluation criteria are constructed through deep learning; Based on the test results from the hydraulic fracturing test site, the well spacing and the half-fracture length of the hydraulic fracturing were determined, and a deep learning standard model and evaluation criteria for the well spacing and the half-fracture length of the hydraulic fracturing were established. Based on the well type evaluation criteria and the well spacing and fracture half-fracture length evaluation criteria, the well pattern type is optimally determined.

2. The well pattern design method according to claim 1, characterized in that, The establishment of a standardized three-dimensional quantitative model for reservoir multi-attribute characteristics includes: applying an image coordination algorithm to compare and calibrate the grayscale image of the tight oil layer obtained by focused ion beam scanning electron microscopy with the three-dimensional grayscale image of the tight oil layer obtained by CT scanning, and establishing the standardized three-dimensional quantitative model based on the comparison results; wherein, the reservoir multi-attribute characteristics include sand-mud interlayers, fractures, porosity, and oil content.

3. The well pattern design method according to claim 1, characterized in that, The digital results of extracting reservoir multi-attribute features using deep learning include: using the U-Net convolutional neural network algorithm to extract digital result models of sand-mud interlayers, fractures, porosity, and oil content of horizontal well oil layers; the digital results of reservoir multi-attribute features include the percentage content of sand-mud interlayers, sand-mud interlayer density, fracture aperture, porosity, and oil saturation of the oil layer.

4. The well pattern design method according to claim 1, characterized in that, The construction of the intelligent large model of reservoir multi-attribute characteristic parameters includes: constructing an intelligent large model of sand-mud interbedded layers, fractures, porosity, and oil-bearing characteristic parameters based on datasets of sand-mud interbedded layers, fractures, porosity, and oil-bearing properties under different geological conditions and genetic types.

5. The well pattern design method according to claim 1, characterized in that, The establishment of a deep learning standard model for reservoir multi-attribute features through deep learning includes: using a convolutional neural network as the deep learning model architecture, using the reservoir multi-attribute feature data as training data, adjusting the model parameters through a backpropagation algorithm, and creating the deep learning standard model.

6. The well pattern design method according to claim 1, characterized in that, The determination of the reservoir three-dimensional configuration model includes: obtaining digital results of reservoir multi-attribute characteristics and their judgment criteria based on the deep learning standard model, and determining the evaluation criteria for different types of tight oil layers; based on the evaluation criteria, establishing a reservoir three-dimensional configuration model including sand-mud interbedded layer type, fracture type, and oil-bearing type.

7. The well pattern design method according to claim 1, characterized in that, The method of constructing well type models and well type evaluation criteria through deep learning includes: using a convolutional neural network architecture, using the reservoir three-dimensional configuration model as a training model, and constructing well type models and well type evaluation criteria for different types of tight oil layers.

8. The well pattern design method according to claim 1, characterized in that, The determination of well spacing and fracture half-fracture length based on the test results of the hydraulic fracturing test field includes: determining the distribution of fracturing proppant and tracer based on the sealed core and cuttings test results of the fracturing wells and test wells in the hydraulic fracturing test field, thereby determining the well spacing and fracture half-fracture length.

9. The well pattern design method according to claim 1, characterized in that, The establishment of a deep learning standard model for well spacing and fracture half-length includes: applying the U-Net convolutional neural network algorithm to establish a standardized quantitative model of well spacing and fracture half-length for different types of tight oil layers, and creating a deep learning standard model of well spacing and fracture half-length through deep learning.

10. The well pattern design method according to claim 1, characterized in that, The establishment of evaluation criteria for well spacing and fracture half-length includes: applying a convolutional neural network algorithm to obtain datasets of well spacing and fracture half-length in tight oil reservoirs under different geological conditions and with different genesis, constructing intelligent large models of characteristic parameters of well spacing and fracture half-length in different types of tight oil reservoirs, and determining the evaluation criteria for well spacing and fracture half-length based on the intelligent large model.

Images