Reservoir fracture distribution prediction method and system based on big data

Abstract

The invention discloses a reservoir fracture distribution prediction method and system based on big data, and relates to the technical field of oil and gas field development, and the method comprises the steps of multi-source data collection and fusion, multi-scale feature extraction and optimization, hybrid model construction and three-dimensional fracture probability body generation, and fracture distribution prediction and visual display. According to the method, prediction blind areas are reduced by collecting multi-source heterogeneous data, physical constraints are achieved by introducing a Griffith criterion so as to enhance geological rationality, dynamic data and static geological data of fracturing construction are subjected to combined modeling, dynamic fusion of multi-source data is achieved, timeliness of crack prediction is improved, in addition, the physical constraint hybrid model is adopted, and the prediction efficiency is improved. And spatial characteristics and dynamic response are considered through a CNN-GBDT hybrid architecture, a fracture mechanics criterion is embedded to enhance interpretability, real-time access of logging-while-drilling data in the drilling process is supported, and key technical support can be provided for unconventional oil and gas development.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas field development, and in particular to a reservoir fracture distribution prediction method and system based on big data. Background Art

[0002] Reservoir fractures are natural networks of cracks formed in rocks due to diagenesis or tectonic stress. Like underground "highways," they directly control the storage and flow efficiency of fluids such as oil, gas, and geothermal energy. Especially in the development of unconventional resources such as shale gas and tight oil, the extent of fracture development determines the economic viability of the reservoir. Therefore, accurately predicting the spatial distribution of reservoir fractures has become a core topic in energy exploration.

[0003] Reservoir fractures are key pathways for the migration of oil, gas, geothermal energy, and groundwater. Their spatial distribution and development characteristics directly affect resource development efficiency and engineering safety. In low-permeability reservoirs, the fracture system is both the primary path for fluid seepage and the core source of reservoir heterogeneity. Traditional fracture prediction relies on geological statistics and geophysical inversion, but is limited by data resolution and multi-solutions, making it difficult to accurately characterize complex fracture networks.

[0004] Traditional reservoir fracture distribution prediction methods rely on single geophysical data and ignore the synergistic effect of multi-source data, resulting in low prediction accuracy. Empirical formulas and statistical models are difficult to capture the spatial correlation characteristics of reservoir heterogeneity and fracture networks. There is a lack of effective methods for fusing dynamic data with static geological data, resulting in a large deviation between the prediction results and the actual reservoir fracture distribution. Therefore, the present invention proposes a reservoir fracture distribution prediction method and system based on big data to solve the problems existing in the prior art. Summary of the Invention

[0005] In response to the above problems, the purpose of the present invention is to propose a reservoir fracture distribution prediction method and system based on big data to solve the problems of low prediction accuracy of traditional reservoir fracture distribution prediction methods, difficulty in capturing reservoir heterogeneity and spatial correlation characteristics of fracture networks through empirical formulas and statistical models, and lack of effective methods for fusing dynamic data with static geological data, which leads to large deviations between prediction results and actual reservoir fracture distribution.

[0006] In order to achieve the purpose of the present invention, the present invention is implemented by the following technical solution: a reservoir fracture distribution prediction method based on big data, comprising the following steps:

[0007] Step 1: Collect multi-source heterogeneous data including geological data, geophysical data, engineering data and dynamic monitoring data in the reservoir area to be predicted, and fuse them to generate a fused data volume;

[0008] Step 2: Use wavelet multi-scale decomposition technology to extract multi-scale features in the fusion data volume, and use random forest algorithm and PCA algorithm to optimize the extracted multi-scale features;

[0009] Step 3: Build a CNN-GBDT hybrid model to process the optimized multi-scale features, concatenate the high-dimensional feature vector output by the CNN branch with the features of the GBDT branch, map them into crack probability values through a fully connected layer, and generate a three-dimensional crack probability volume;

[0010] Step 4: Set dynamic thresholds based on the probability volume to delineate fracture development zones and non-development zones. Combined with the in-situ stress direction, generate a 3D fracture network distribution map. The results are integrated into the geological modeling platform to optimize horizontal well trajectory design in real time.

[0011] A further improvement is that in step one, the multi-source heterogeneous data is divided into static data and dynamic data, wherein the static data includes seismic attributes, well logging interpretation results and core brittleness index, and the dynamic data includes fracturing construction pressure curves and the spatiotemporal distribution of microseismic events.

[0012] A further improvement is that in step 1, the fused data volume is gridded by using a Kriging interpolation algorithm to store spatially discrete data, ensuring that the resolution is consistent with the seismic data volume.

[0013] A further improvement is that in step 2, the wavelet multi-scale decomposition technology uses Daubechies wavelet basis function, the decomposition layer number is 5, and the approximate coefficients and detail coefficients at each scale are extracted as initial features.

[0014] A further improvement is that in step 2, a random forest algorithm is used to screen a subset of crack sensitive features, and the PCA algorithm is used to reduce the dimension of the feature subset to generate an optimized feature matrix. The random forest algorithm sets a feature importance threshold ≥ 0.8 and screens out crack density, azimuth deviation, and acoustic time difference constant as sensitive feature subsets.

[0015] A further improvement is that in step 3, the loss function of the CNN-GBDT hybrid model includes a fracture mechanics constraint term, and the formula is:

[0016]

[0017] Among them, L is the total loss function, α is the weight coefficient, and LCE is the cross entropy loss, which measures the difference between the fracture probability distribution predicted by the model and the true label (such as the imaging logging interpretation result) and is used for classification task optimization. is the summation operator, σmodel is the fracture stress value predicted by the model, and σGriffith is the theoretical fracture stress calculated by the Griffith criterion.

[0018] A further improvement is that the formula of σGriffith is:

[0019]

[0020] Where E is the Young's modulus of rock, γ is the surface energy, and α is the half-length of the crack.

[0021] Further improvements lie in: During the entire prediction process, edge computing devices are used to process drilling data in real time at the drilling reservoir site and dynamically correct the reservoir fracture prediction results.

[0022] A reservoir fracture distribution prediction system based on big data, comprising:

[0023] Multi-source data acquisition module, used to collect multi-source heterogeneous data including geological data, geophysical data, engineering data and dynamic monitoring data;

[0024] Data fusion engine, used to generate fused data volumes based on the spatiotemporal alignment algorithm according to the collected multi-source heterogeneous data;

[0025] Feature optimization module, used to perform wavelet multi-scale decomposition, random forest feature selection and PCA dimensionality reduction algorithm for feature optimization;

[0026] The CNN-GBDT hybrid model processor is deployed on the GPU cluster to process the optimized multi-scale features and generate a three-dimensional crack probability volume;

[0027] The 3D visualization module, integrated into the geological modeling software, supports the coordinated design of fracture networks and well trajectories, and displays the predicted results of reservoir fracture distribution.

[0028] The beneficial effects of the present invention are as follows: the present invention reduces prediction blind spots by collecting multi-source heterogeneous data, and implements physical constraints by introducing the Griffith criterion to enhance geological rationality. It also jointly models dynamic data of fracturing construction with static geological data to achieve dynamic fusion of multi-source data and improve the timeliness of fracture prediction. The physical constraint hybrid model is adopted, and the CNN-GBDT hybrid architecture takes into account both spatial characteristics and dynamic responses. The fracture mechanics criterion is embedded to enhance interpretability and support real-time access to logging while drilling data during drilling, so as to optimize in real time and dynamically update the prediction results.

[0029] In addition, the present invention uses big data technology to transform reservoir fracture prediction from traditional experience-driven to data-driven, significantly improving the prediction accuracy under complex geological conditions. The entire prediction process fully reflects the core characteristics of big data technology through multi-source heterogeneous data integration, high-dimensional feature mining, distributed computing architecture and data-driven models, breaking through the limitations of single data and using full-dimensional information to improve prediction accuracy. It solves the problems of traditional prediction methods such as single data utilization and poor model generalization, and provides key technical support for unconventional oil and gas development. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the reservoir fracture distribution prediction method based on big data of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0032] Tight, low-permeability reservoirs possess vast oil and gas resources, but their poor matrix properties, strong heterogeneity, and widespread natural fractures make exploration and development challenging. Natural fractures serve as crucial reservoir spaces and seepage pathways in tight, low-permeability reservoirs, influencing the accumulation, enrichment, single-well productivity, and development outcomes of oil and gas in these reservoirs. Understanding the distribution patterns of natural fractures can provide a geological basis for the exploration and development of these reservoirs. Predicting the distribution patterns of natural fractures is both a difficult and a hot topic in reservoir fracture research.

[0033] Example 1

[0034] See also Figure 1 This embodiment provides a reservoir fracture distribution prediction method based on big data, comprising the following steps:

[0035] Step 1: Multi-source data collection and fusion

[0036] In the reservoir area to be predicted, geological data (such as lithology and tectonic stress field), geophysical data (such as 3D seismic attributes and well logging curves), engineering data (such as drilling trajectory and fracturing operation parameters), and dynamic monitoring data (such as microseismic events and production dynamics) are collected separately to obtain multi-source heterogeneous data. The collected multi-source heterogeneous data are divided into static data and dynamic data. The static data include seismic attributes, well logging interpretation results, and core brittleness index, while the dynamic data include fracturing operation pressure curves and the spatiotemporal distribution of microseismic events. The spatially discrete data are then gridded (grid size ≤ 10m × 10m × 2m) using the Kriging interpolation algorithm to ensure that the resolution is consistent with the seismic data volume, and then fused to generate a fused data volume.

[0037] Step 2: Multi-scale feature extraction and optimization

[0038] Wavelet multiscale decomposition technology is used to extract multiscale features from the fused data volume, and a random forest algorithm is used to screen a subset of crack-sensitive features. The PCA algorithm is then used to reduce the dimensionality of the feature subset to generate an optimized feature matrix. The random forest algorithm sets a feature importance threshold of ≥0.8 and screens out crack density, azimuth deviation, and acoustic time difference constant as sensitive feature subsets.

[0039] The wavelet multi-scale decomposition technology of this embodiment uses the Daubechies wavelet basis function with 5 decomposition layers, extracting the approximate coefficients and detail coefficients at each scale as initial features;

[0040] Step 3: Hybrid model construction and 3D fracture probability volume generation

[0041] Construct a CNN-GBDT hybrid model to process the optimized multi-scale features, concatenate the high-dimensional feature vector output by CNN with the features of the GBDT branch, map them to crack probability values through the fully connected layer, and generate a three-dimensional crack probability body. The CNN-GBDT hybrid model of this embodiment consists of a convolutional neural network (CNN) branch and a gradient boosting tree (GBDT) branch, where

[0042] CNN branch: The input is a three-dimensional feature volume, and a 3D convolution layer is used to extract spatial features (including curvature, coherence volume, and variance volume), and output a high-dimensional feature vector (such as 1024 dimensions);

[0043] GBDT branch: The input is structural features such as well logging curves and ground stress direction, and generates a preliminary prediction of the probability of fracture development;

[0044] Feature fusion: The high-dimensional vector output by CNN is concatenated with the GBDT prediction value and mapped to a crack probability value (0-1) through a fully connected layer;

[0045] The loss function of the CNN-GBDT hybrid model in this embodiment includes a fracture mechanics constraint term, and the formula is:

[0046]

[0047] Among them, L is the total loss function, α is the weight coefficient, and the value range is 0≤α≤1. LCE is the cross entropy loss, which measures the difference between the fracture probability distribution predicted by the model and the true label (such as the imaging logging interpretation result) and is used for classification task optimization. As a summation operator, the physical constraint errors of n samples or spatial locations (such as grid points of seismic data volumes) are accumulated to ensure that the model satisfies the physical laws at all samples or spatial points. σmodel is the fracture stress value predicted by the model. The fracture development probability output by the CNN-GBDT hybrid model is indirectly related to the stress field distribution. σGriffith is the theoretical fracture stress calculated by the Griffith criterion. The formula is:

[0048]

[0049] Where E is the Young's modulus of rock, γ is the surface energy, and α is the half-length of the crack. The model output is constrained by classical fracture mechanics theory to ensure that the prediction results conform to geological laws. In the process of predicting fracture probability, the fracture mechanics criterion (Griffith criterion) is added to the loss function to constrain the model output to conform to geological laws.

[0050] Step 4: Crack distribution prediction and visualization

[0051] Based on the probability volume, a dynamic threshold (ranging from 0.6 to 0.8) is set to delineate fracture development zones from non-development zones. A three-dimensional fracture network distribution map is generated based on the in-situ stress direction (obtained through imaging logging or seismic anisotropy inversion). The results are then integrated into a geological modeling platform to optimize horizontal well trajectory design in real time.

[0052] During the entire prediction process of this embodiment, edge computing devices are used to process the LWD data in real time at the drilling site and dynamically correct the fracture prediction results. The specific operations are as follows: deploying a lightweight model (such as MobileNet-3D), processing the LWD data in real time at the drilling site, generating a local fracture probability volume, receiving edge data, performing global optimization through the CNN-GBDT hybrid model, and dynamically updating the three-dimensional fracture network model.

[0053] Example 2

[0054] This embodiment also provides a reservoir fracture distribution prediction system based on big data, which comprises a multi-source data acquisition module for acquiring data, a data fusion engine for fusing data, a feature optimization module for optimizing features, a CNN-GBDT hybrid model processor for data processing, and a three-dimensional visualization module for displaying prediction results;

[0055] The multi-source data acquisition module in this embodiment is used to acquire multi-source heterogeneous data including geological data, geophysical data, engineering data and dynamic monitoring data;

[0056] The data fusion engine in this embodiment is used to generate a fused data volume based on the collected multi-source heterogeneous data and the spatiotemporal alignment algorithm;

[0057] The feature optimization module in this embodiment is used to perform wavelet multi-scale decomposition, random forest feature selection and PCA dimensionality reduction algorithm to perform feature optimization processing;

[0058] The CNN-GBDT hybrid model processor in this embodiment is deployed on a GPU cluster to process the optimized multi-scale features and generate a three-dimensional crack probability volume;

[0059] The 3D visualization module in this embodiment is integrated into the geological modeling software, supports the coordinated design of fracture networks and well trajectories, and displays the predicted results of reservoir fracture distribution.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A reservoir fracture distribution prediction method based on big data, characterized in that: The following steps are involved: Step 1: Collect multi-source heterogeneous data including geological data, geophysical data, engineering data and dynamic monitoring data in the reservoir area to be predicted, and fuse them to generate a fused data volume; Step 2: Use wavelet multi-scale decomposition technology to extract multi-scale features in the fusion data volume, and use random forest algorithm and PCA algorithm to optimize the extracted multi-scale features; Step 3: Build a CNN-GBDT hybrid model to process the optimized multi-scale features, concatenate the high-dimensional feature vector output by the CNN branch with the features of the GBDT branch, map them into crack probability values through a fully connected layer, and generate a three-dimensional crack probability volume; Step 4: Set dynamic thresholds based on the probability volume to delineate fracture development zones and non-development zones. Combined with the in-situ stress direction, generate a 3D fracture network distribution map. The results are integrated into the geological modeling platform to optimize horizontal well trajectory design in real time.

2. The reservoir fracture distribution prediction method based on big data according to claim 1, characterized in that: In step 1, the multi-source heterogeneous data are divided into static data and dynamic data, wherein the static data includes seismic attributes, well logging interpretation results and core brittleness index, and the dynamic data includes fracturing construction pressure curve and spatiotemporal distribution of microseismic events.

3. The reservoir fracture distribution prediction method based on big data according to claim 1, characterized in that: In the step 1, the fused data volume is gridded by using a Kriging interpolation algorithm to obtain spatially discrete data, ensuring that the resolution is consistent with that of the seismic data volume.

4. The method for predicting reservoir fracture distribution based on big data according to claim 1, characterized in that: In the step 2, the wavelet multi-scale decomposition technology uses Daubechies wavelet basis function, the decomposition layer number is 5, and the approximate coefficients and detail coefficients at each scale are extracted as initial features.

5. The method for predicting reservoir fracture distribution based on big data according to claim 1, characterized in that: In step 2, a random forest algorithm is used to screen a subset of crack sensitive features, and the PCA algorithm is used to reduce the dimensionality of the feature subset to generate an optimized feature matrix. The random forest algorithm sets a feature importance threshold of ≥0.8 and screens out crack density, azimuth deviation, and acoustic time difference constant as a subset of sensitive features.

6. The method for predicting reservoir fracture distribution based on big data according to claim 1, characterized in that: In step 3, the loss function of the CNN-GBDT hybrid model includes a fracture mechanics constraint term, and the formula is: Among them, L is the total loss function, α is the weight coefficient, and LCE is the cross entropy loss, which measures the difference between the fracture probability distribution predicted by the model and the true label (such as the imaging logging interpretation result) and is used for classification task optimization. is the summation operator, σmodel is the fracture stress value predicted by the model, and σGriffith is the theoretical fracture stress calculated by the Griffith criterion.

7. The method for predicting reservoir fracture distribution based on big data according to claim 6, characterized in that: The σGriffith's formula is: Where E is the Young's modulus of rock, γ is the surface energy, and α is the half-length of the crack.

8. The method for predicting reservoir fracture distribution based on big data according to claim 1, characterized in that: During the entire prediction process, edge computing devices are used to process downhole data in real time at the drilling reservoir site, and dynamically correct the reservoir fracture prediction results.

9. A reservoir fracture distribution prediction system based on big data as claimed in claim 1, characterized in that: include: Multi-source data acquisition module, used to collect multi-source heterogeneous data including geological data, geophysical data, engineering data and dynamic monitoring data; Data fusion engine, used to generate fused data volumes based on the spatiotemporal alignment algorithm according to the collected multi-source heterogeneous data; Feature optimization module, used to perform wavelet multi-scale decomposition, random forest feature selection and PCA dimensionality reduction algorithm for feature optimization; The CNN-GBDT hybrid model processor is deployed on the GPU cluster to process the optimized multi-scale features and generate a three-dimensional crack probability volume; The 3D visualization module, integrated into the geological modeling software, supports the coordinated design of fracture networks and well trajectories, and displays the predicted results of reservoir fracture distribution.

Images