CN-122021230-A - Drilling fluid damage simulation method and system based on deep learning

CN122021230ACN 122021230 ACN122021230 ACN 122021230ACN-122021230-A

Abstract

The invention discloses a drilling fluid damage simulation method and system based on deep learning, wherein the method comprises the steps of obtaining an initial static component model of a rock sample according to a three-dimensional pore network and three-dimensional component spatial distribution of the rock sample; and processing the initial static component model of the rock sample according to the chemical reaction module, the morphological conversion module, the fluid action module and the particle migration module to obtain a static component model after the drilling fluid is displaced, and comparing the initial static component model of the rock sample with the static component model after the drilling fluid is displaced to obtain the influence of the drilling fluid displacement on the rock sample. The method effectively overcomes the limitation of indoor experiments, can comprehensively consider various influencing factors, acquires more comprehensive and detailed rock information, effectively predicts reservoir change, optimizes displacement efficiency, evaluates potential risks in advance, and provides a firmer foundation for reservoir sensitivity evaluation and quantitative characterization of reservoir damage degree.

Inventors

YAN KANG
HAN DENGLIN
ZHU CHAOBIN
DU HAO
WANG CHENCHEN

Assignees

长江大学
微纳数智能源科技(武汉)有限公司

Dates

Publication Date: 20260512
Application Date: 20251113

Claims (10)

1. The drilling fluid damage simulation method based on deep learning is characterized by comprising the following steps of: S1, acquiring a three-dimensional gray image of a rock sample obtained by CT scanning of the rock sample, constructing a three-dimensional pore network of the rock sample through the three-dimensional gray image of the rock sample, acquiring a rock two-dimensional component distribution obtained by QEMSCAN mineral scanning of the rock sample, selecting a plurality of components which have obvious influence on a reservoir sensitization effect in the rock two-dimensional component distribution, obtaining a screened rock two-dimensional component distribution, forming a three-dimensional component distribution model based on deep neural network training, and obtaining an initial static component model of the rock sample according to the three-dimensional pore network and the three-dimensional component distribution model of the rock sample; S2, constructing a chemical reaction module, a morphology conversion module, a fluid action module and a particle migration module; S3, processing the initial static component model of the rock sample according to the chemical reaction module, the morphological conversion module, the fluid action module and the particle migration module to obtain a static component model after the drilling fluid is displaced, comparing the initial static component model of the rock sample and the static component model after the drilling fluid is displaced, performing seepage simulation, calculating the damage rate of a reservoir, and obtaining the influence of the drilling fluid displacement on the rock sample.
2. The drilling fluid damage simulation method based on deep learning of claim 1, wherein in step S1, a rock two-dimensional component distribution obtained by QEMSCAN mineral scanning of a rock sample is obtained, and among the rock two-dimensional component distribution, a plurality of components having significant influence on a reservoir sensitization effect are selected to obtain a screened rock two-dimensional component distribution, and an initial static component model of the rock sample is obtained according to a three-dimensional pore network and the two-dimensional component distribution of the rock sample, and specifically comprises the following steps: s11, carrying out QEMSCAN mineral scanning on a rock sample to obtain a plurality of rock two-dimensional component distribution pictures; s12, selecting a plurality of components with obvious influence on the sensitization effect of the reservoir from the rock two-dimensional component distribution picture to obtain a screened rock two-dimensional component distribution picture; s13, CT (computed tomography) slice and back scattering image of the rock sample at the same position are obtained, and rock two-dimensional component distribution pictures, CT slice and back scattering image at the same position are matched; S14, constructing a U-Net neural network model, and training the U-Net neural network model through a plurality of matched groups of rock two-dimensional component distribution pictures, CT slices and back scattering images at the same position to obtain a complete training U-Net neural network model; s15, inputting all CT slices of the rock sample into a well-trained U-Net neural network model to obtain two-dimensional component distribution of each section of the rock sample, and obtaining a three-dimensional component distribution model of the rock sample according to the two-dimensional component distribution of each section; s16, obtaining an initial static component model of the rock sample according to the three-dimensional pore network and the three-dimensional component distribution model of the rock sample.
3. The deep learning-based drilling fluid damage simulation method according to claim 1, wherein the components having a significant influence on the reservoir sensitization effect are pores, tuff, kaolinite, illite and montmorillonite, respectively.
4. The deep learning-based drilling fluid damage simulation method of claim 1, wherein in step S2, the chemical reaction module specifically includes: A database storing physicochemical property data for various formation minerals, drilling fluid compositions, and related fluids, and containing chemical reaction equations and related thermodynamic and kinetic parameters; An input unit for receiving input simulation parameters including initial composition of drilling fluid, type and content of formation minerals, properties of formation fluid, temperature, pressure, and time step of simulation, total duration of simulation; a calculation unit for performing calculation using chemical reaction kinetics and thermodynamic principles based on the input parameters and data in the database; And the output unit is used for outputting the calculated result.
5. The drilling fluid damage simulation method based on deep learning of claim 1, wherein in step S2, the morphology conversion module specifically comprises: the database is used for storing basic data of various water-sensitive minerals, hydration expansion rate data and chemical expansion rate data of various minerals under different conditions, which are determined by physical experiments, and related parameters of reservoir pore structures; The input unit is used for receiving simulation parameters input from the outside and receiving the simulated time step and the simulated total duration; a calculation unit for calculating hydration expansion and chemical expansion processes of minerals according to the input parameters and the stored data; The morphological algorithm comprises expansion operation, migration operation, open operation and close operation, wherein the expansion operation simulates the expansion process of minerals based on hydration expansion rate and chemical expansion rate data in experimental data, and adjusts the volume parameters of the minerals in the model; the migration operation is to simulate the stripping, migration and blocking processes of the particles in the reservoir according to the particle migration experimental result, and adjust the particle distribution parameters in the model; the closed operation is used for filling small holes in the image and connecting the dispersed areas so as to repair small faults in the pore structure and improve the accuracy of permeability simulation; And the output unit is used for outputting the calculated result.
6. The deep learning-based drilling fluid damage simulation method of claim 1, wherein in step S2, the fluid action module specifically comprises: a database for storing a pore network model of the reservoir, fluid property parameters, and boundary condition parameters; An input unit for receiving externally input flow state parameters, fluid property parameters, and boundary condition parameters; the CFD solving unit simulates the flow of drilling fluid in a pore network by adopting a computational fluid dynamics method, discretizes continuous fluid space into a limited grid unit, discretizes a control equation on each grid unit, and solves the distribution of a velocity field, a pressure field and a concentration field of the fluid through iterative calculation; and the output unit is used for outputting the distribution of the speed field, the pressure field and the concentration field obtained by simulation.
7. The deep learning-based drilling fluid damage simulation method of claim 1, wherein in step S2, the particle migration module specifically comprises: the data interaction and input unit is used for carrying out data interaction with the chemical reaction module, receiving the related information about particle generation and dissolution output by the chemical reaction module, and receiving the basic data of the reservoir, the pore network structure parameters and the property parameters of the fluid; The particle migration carrying calculation module is used for carrying out particle migration carrying calculation based on the discrete elements; the particle generation and dissolution calculation unit is used for determining the particle generation and dissolution process according to the data acquired from the chemical reaction module, the input drilling fluid properties and other parameters; A deposition and blockage evaluation unit for simulating a deposition process of particles in the pores according to the generation and dissolution processes of the particles based on a kinetic principle, and determining a position of particle deposition according to the particle size, shape, fluid flow rate and pore structure of the particles; And an output unit for outputting the granular migration result.
8. The deep learning-based drilling fluid damage simulation method of claim 1, wherein in step S3, the initial static component model of the rock sample is processed according to the chemical reaction module, the morphological transformation module, the fluid action module and the particle migration module to obtain a static component model after the drilling fluid displacement, and specifically comprises the following steps: s31, inputting the initial static component model into a fluid action module; s32, the fluid action module receives an input pore network model, fluid property parameters and boundary condition parameters of the reservoir, and a CFD solving unit of the fluid action module obtains the velocity field, the pressure field and the concentration field distribution of the fluid in the pore network of the reservoir, so that the migration path of the fluid is determined; s33, when the fluid action module calculates each step, transmitting the current fluid composition, temperature and pressure information to an input unit of the chemical reaction module, and according to the input information, combining physical and chemical property data and chemical reaction equations of various stratum minerals, drilling fluid components and related fluids stored in a database of the calculation unit of the chemical reaction module, calculating chemical reactions occurring at the moment by using chemical reaction dynamics and thermodynamic principles, determining new substances generated after the reactions and changes of concentration of each substance, and returning the results to the fluid action module, and simultaneously providing data for particle generation in the particle migration module; S34, when the fluid action module calculates each step, the fluid action module transmits the fluid property, temperature and pressure at the moment and initial content information of various minerals in the rock sample to an input unit of a morphology conversion module, a data storage unit of the morphology conversion module reads basic data of various minerals and hydration expansion rate data, chemical expansion rate and particle migration rate data determined through physical experiments, and a calculation unit of the morphology conversion module calculates volume expansion quantity of the minerals under the current condition according to the basic data, hydration expansion rate data, chemical expansion rate and particle migration rate (including denudation and accumulation) data of various minerals stored by the fluid property, temperature and pressure and initial content information of various water-sensitive minerals in the rock sample and data thereof, updates the volume of the minerals, calculates changes of a pore network model and feeds back the changes to the fluid action module; S35, when the fluid action module calculates each step, the fluid action module transmits the current fluid flow rate, fluid composition, pore network model and the like to a data interaction and input unit of the particle migration module, the particle migration module combines particle generation information acquired from the chemical reaction module, the particle generation and dissolution conditions including particle size distribution and chemical composition of particles are determined in a particle generation and dissolution calculation unit, particle migration and transportation calculation is carried out based on discrete elements, a deposition process of the particles in pores is simulated by a deposition and blockage evaluation unit, deposition rate and blockage effect are calculated, change of reservoir permeability is predicted, and the results are fed back to the fluid action module; And S36, the fluid action module advances the simulation process step by step according to the set time step or the calculation step number, and repeatedly calls the operation of other modules at each step, and continuously calculates and interacts until the set ending condition is met, so that the static component model after the drilling fluid is displaced is obtained.
9. Drilling fluid injury simulation system based on deep learning, characterized by comprising: The initial static component model construction module is used for acquiring a three-dimensional gray image of a rock sample obtained by CT scanning of the rock sample, constructing a three-dimensional pore network of the rock sample through the three-dimensional gray image of the rock sample, acquiring a rock two-dimensional component distribution obtained by QEMSCAN mineral scanning of the rock sample, selecting a plurality of components which have obvious influence on a reservoir sensitization effect in the rock two-dimensional component distribution, acquiring a screened rock two-dimensional component distribution, forming a three-dimensional component distribution model based on deep neural network training, and obtaining an initial static component model of the rock sample according to the three-dimensional pore network and the three-dimensional component distribution model of the rock sample; the dynamic reaction function construction module is used for constructing a chemical reaction module, a morphological conversion module, a fluid action module and a particle migration module; The damage simulation module is used for processing the initial static component model of the rock sample according to the chemical reaction module, the morphological conversion module, the fluid action module and the particle migration module to obtain a static component model after the drilling fluid is displaced, comparing the initial static component model of the rock sample with the static component model after the drilling fluid is displaced, performing seepage simulation and calculating the damage rate of the reservoir to obtain the influence of the drilling fluid displacement on the rock sample.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores one or more programs, the one or more programs are executable by the one or more processors to implement the steps in the deep learning based drilling fluid damage simulation method of any of claims 1-8.

Description

Drilling fluid damage simulation method and system based on deep learning Technical Field The invention relates to the technical field of drilling fluid displacement numerical simulation, in particular to a drilling fluid damage simulation method and system based on deep learning. Background In the field of oil and gas exploration and development, it is important to accurately evaluate the influence of drilling fluid on a reservoir, wherein the research of the sensitization effect of the drilling fluid is a key link. At present, an indoor core physical simulation experiment takes the dominant role in a method system for researching the sensitization effect of drilling fluid. Scientific researchers can carry out omnibearing and refined observation on a rock core sample by means of a multi-scale analysis system (MAPS) scanning technology to acquire detailed information of a microstructure in the rock, the type and the content of rock minerals can be accurately measured by X-ray diffraction (XRD), reliable data are provided for analyzing the composition of the minerals, quantitative mineral scanning (QEMSCAN) can carry out quantitative analysis on the minerals, the characteristics of the minerals are further clarified, and in-situ displacement CT scanning can monitor the change condition of the inside of the rock core in the process of displacement of drilling fluid in real time. Through the advanced experimental techniques, researchers can obtain microscopic characteristic data such as rock mineral composition, pore structure, physical and chemical reaction and the like in detail, and a solid foundation is laid for reservoir sensitivity evaluation and quantitative characterization of reservoir damage degree. However, there are significant limitations to the physical simulation of indoor cores. Limited to the number of samples, it is difficult to cover all possible combinations of particle properties, fluid properties and formation properties. In the actual drilling process, the reservoir is subjected to the combined action of multiple complex factors such as physics, chemistry, biology, hydrodynamic force and the like, and the influence on the reservoir under the combined action of the factors cannot be comprehensively reflected only by means of a physical simulation experiment. For example, migration of particles of different size distributions in complex formation pore structures under the action of fluids of different flow rates is difficult to fully demonstrate through limited core sample experiments. In view of this, numerical simulation methods have been developed. The numerical simulation can comprehensively consider various influencing factors of reservoir damage, such as multi-physical field coupling effects of particle migration, chemical reaction, fluid flow and the like by means of strong computing power. By establishing an accurate mathematical model, reservoir change in the drilling fluid displacement process can be effectively predicted, the displacement efficiency is optimized, and potential risks are estimated in advance, so that uncertainty in the drilling process is reduced, and smooth progress of oil and gas exploration and development work is ensured. Disclosure of Invention The invention aims to overcome the technical defects, and provides a drilling fluid damage simulation method and system based on deep learning, which solve the technical problems that the particle characteristics, the fluid properties and the stratum properties cannot be comprehensively reflected due to the limited number of physical simulation experiment samples in the prior art and the comprehensive influence on a reservoir under multiple complex actions is difficult to accurately and efficiently simulate and predict the reservoir damage degree and the displacement efficiency in the process of drilling fluid displacement by the traditional method, thereby providing a more comprehensive and accurate decision basis for drilling engineering and effectively reducing drilling risks. In order to achieve the technical purpose, the invention adopts the following technical scheme: The invention provides a drilling fluid damage simulation method based on deep learning, which comprises the following steps: S1, acquiring a three-dimensional gray image of a rock sample obtained by CT scanning of the rock sample, constructing a three-dimensional pore network of the rock sample through the three-dimensional gray image of the rock sample, acquiring a rock two-dimensional component distribution obtained by QEMSCAN mineral scanning of the rock sample, selecting a plurality of components which have obvious influence on a reservoir sensitization effect in the rock two-dimensional component distribution, obtaining a screened rock two-dimensional component distribution, forming a three-dimensional component distribution model based on deep neural network training, and obtaining an initial static component model of the rock sample accordi