WO2020215850A1 - 一种测定地层流体组成和性质的方法和系统 - Google Patents

一种测定地层流体组成和性质的方法和系统 Download PDF

Info

Publication number
WO2020215850A1
WO2020215850A1 PCT/CN2020/074353 CN2020074353W WO2020215850A1 WO 2020215850 A1 WO2020215850 A1 WO 2020215850A1 CN 2020074353 W CN2020074353 W CN 2020074353W WO 2020215850 A1 WO2020215850 A1 WO 2020215850A1
Authority
WO
WIPO (PCT)
Prior art keywords
composition
machine learning
properties
data set
model
Prior art date
Application number
PCT/CN2020/074353
Other languages
English (en)
French (fr)
Inventor
左有祥
冯永仁
卢涛
孔笋
沈阳
秦小飞
褚晓冬
陈永超
吴兴方
黄琳
Original Assignee
中海油田服务股份有限公司
中国海洋石油集团有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中海油田服务股份有限公司, 中国海洋石油集团有限公司 filed Critical 中海油田服务股份有限公司
Priority to US17/433,519 priority Critical patent/US20220155275A1/en
Publication of WO2020215850A1 publication Critical patent/WO2020215850A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/26—Oils; Viscous liquids; Paints; Inks
    • G01N33/28—Oils, i.e. hydrocarbon liquids
    • G01N33/2823—Raw oil, drilling fluid or polyphasic mixtures
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B49/00—Testing the nature of borehole walls; Formation testing; Methods or apparatus for obtaining samples of soil or well fluids, specially adapted to earth drilling or wells
    • E21B49/08—Obtaining fluid samples or testing fluids, in boreholes or wells
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B49/00—Testing the nature of borehole walls; Formation testing; Methods or apparatus for obtaining samples of soil or well fluids, specially adapted to earth drilling or wells
    • E21B49/08—Obtaining fluid samples or testing fluids, in boreholes or wells
    • E21B49/087—Well testing, e.g. testing for reservoir productivity or formation parameters
    • E21B49/0875—Well testing, e.g. testing for reservoir productivity or formation parameters determining specific fluid parameters
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
    • G01N30/02—Column chromatography
    • G01N30/86—Signal analysis
    • G01N30/8693—Models, e.g. prediction of retention times, method development and validation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B2200/00—Special features related to earth drilling for obtaining oil, gas or water
    • E21B2200/20—Computer models or simulations, e.g. for reservoirs under production, drill bits
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B2200/00—Special features related to earth drilling for obtaining oil, gas or water
    • E21B2200/22—Fuzzy logic, artificial intelligence, neural networks or the like
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B49/00—Testing the nature of borehole walls; Formation testing; Methods or apparatus for obtaining samples of soil or well fluids, specially adapted to earth drilling or wells
    • E21B49/08—Obtaining fluid samples or testing fluids, in boreholes or wells
    • E21B49/081—Obtaining fluid samples or testing fluids, in boreholes or wells with down-hole means for trapping a fluid sample
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B49/00—Testing the nature of borehole walls; Formation testing; Methods or apparatus for obtaining samples of soil or well fluids, specially adapted to earth drilling or wells
    • E21B49/08—Obtaining fluid samples or testing fluids, in boreholes or wells
    • E21B49/086—Withdrawing samples at the surface
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
    • G01N30/02—Column chromatography
    • G01N2030/022—Column chromatography characterised by the kind of separation mechanism
    • G01N2030/025—Gas chromatography

Definitions

  • the embodiments of the present application relate to, but are not limited to, logging technology, in particular to a method and system for determining the composition and properties of formation fluids.
  • downhole fluid identification technology especially hardware
  • downhole fluid real-time analysis and undisturbed formation fluid sampling especially real-time analysis of downhole spectroscopy to obtain the composition and properties of downhole fluid components (such as oil, gas, water content, gas-oil ratio, mud filtrate pollution, real-time sampling guidance ) Etc.
  • efficiency and accuracy still need to be improved.
  • the embodiments of the present application provide a method and system for determining the composition and properties of formation fluids, which can quickly and accurately determine the composition and properties of formation fluids and improve logging efficiency and accuracy.
  • embodiments of the present application provide a method for determining the composition and properties of formation fluids, and the method may include:
  • the measurement model uses large data about the composition and properties of a variety of reservoir fluids and measurement signals of downhole sensors as a sample data set, and is obtained by training a pre-created machine learning model of.
  • the obtaining a measurement model for the determination of the composition and properties of the formation fluid may include: calling a measurement model created and trained in advance, or creating and training the measurement model in real time.
  • the method before inputting the signal measured in real time by the sensor on the downhole oil and gas formation tester as the input data of the measurement model into the measurement model, the method may further include:
  • the creating and training the measurement model may include:
  • the establishment of a database regarding the composition and properties of a plurality of reservoir fluids and the measured signals of the downhole sensors may include:
  • the single-phase reservoir fluid samples may include: oil-phase reservoir fluid samples, gas-phase reservoir fluid samples And water-phase reservoir fluid samples;
  • composition, the first property, and the related data of the second property of the single-phase reservoir fluid sample are added to the database as part of the formation fluid big data.
  • the first property may include any one or more of the following: gas-oil ratio, American Petroleum Institute API gravity, molecular weight, sulfur content, carbon content, hydrogen content, WatsonK value, SARA content And paraffin content; among them, SARA refers to saturated hydrocarbons, aromatic hydrocarbons, gums and asphaltenes;
  • the second property may include any one or more of the following: bubble point, dew point, constant composition expansion CCE experimental characteristics, differential release DL experimental characteristics, constant volume depletion CVD experimental characteristics, separator experimental characteristics, density, viscosity, conductivity Rate, compressibility, reservoir flow volume coefficient, paraffin production conditions and asphaltene production conditions.
  • the method may further include:
  • the first preset pressure and the first preset temperature perform any one or more of the following measurements on the single-phase reservoir fluid sample: continuous near-infrared spectroscopy, nuclear magnetic resonance NMR, acoustic waves, fluorescence, and The dielectric constant is measured, and the measurement result is added to the database.
  • the method may further include: adding different drilling mud filtrates to different single-phase reservoir fluid samples, and performing analysis on the single-phase reservoir fluid samples added with corresponding drilling mud filtrate.
  • the PVT experiment and, mixing oil and water, measuring oil moisture content, continuous near-infrared spectroscopy, nuclear magnetic resonance NMR, acoustic wave, fluorescence and dielectric constant measurement, and adding the measurement results to the database.
  • the extracting the sample data set from the database, and training a pre-created machine learning model through the sample data set may include:
  • the preprocessing may include any one or more of the following: denoising, removing outliers and smoothing;
  • the first data set is used as the input data vector of the machine learning model
  • the second data set is used as the target data vector of the output data vector of the machine learning model
  • both the input data vector and the target data vector may be one-dimensional or multi-dimensional.
  • the preset machine learning method may include: a supervised machine learning method
  • the training the machine learning model by using a preset machine learning method based on the first data set and the second data set, and obtaining the optimal machine learning model in the training result as the measurement model may include :
  • For each function in the pre-defined function set perform the following operations: input the standardized first data set into the untrained machine learning model; according to the first data set and the machine learning model The currently loaded function calculates the output data vector;
  • the multiple calculated output data vectors are respectively compared with the target data vector, and the target data vector from the multiple comparison results is determined to correspond to the first output data vector with the smallest error between the target data vector
  • the first function of and the current coefficients of the first function
  • the machine learning model loaded with the first function is used as the optimal machine learning model; wherein the loaded first function has the current coefficient.
  • the method may further include: in one or more of the following processes, supplementing the output data vector whose error with the target data vector meets a preset error threshold to the database : During the training process of the machine learning model and the implementation process of determining the composition and properties of the formation fluid.
  • the embodiments of the present application also provide a system for determining the composition and properties of formation fluids.
  • the system may include a processor and a computer-readable storage medium, where instructions are stored,
  • the processor is configured to execute the instructions to perform the method for determining the composition and properties of the formation fluid described in any of the foregoing embodiments.
  • Figure 1 is a flowchart of a method for determining the composition and properties of formation fluids according to an embodiment of the application
  • FIG. 2 is a flowchart of a method for creating and training the measurement model according to an embodiment of the application
  • FIG. 3 is a flowchart of a method for establishing a database regarding the composition and properties of multiple reservoir fluids and the signals measured by the downhole sensors according to an embodiment of the application;
  • FIG. 4 is a schematic diagram of a method for establishing a database about the composition and properties of multiple reservoir fluids and the signals measured by the downhole sensors according to an embodiment of the application;
  • FIG. 5 is a flowchart of a method for extracting the sample data set from the database and training a pre-created machine learning model through the sample data set according to an embodiment of the application;
  • FIG. 6 is a flowchart of a method for training the machine learning model by using a preset machine learning method and obtaining the optimal machine learning model in the training result as the measurement model according to an embodiment of the application;
  • FIG. 7 is a schematic diagram of a supervised machine learning process according to an embodiment of the application.
  • FIG. 8 is a schematic diagram of the application of downhole cable and midway oil and gas formation testing while drilling based on a database and a measurement model according to an embodiment of the application;
  • FIG. 9(a) is a schematic diagram of a spectrogram of a part of fluid (light oil + gas) extracted from the database according to an embodiment of the application;
  • Fig. 9(b) is a schematic diagram of a spectrogram of some fluids (various crude oils) taken from the database according to an embodiment of the application;
  • Figure 10 is a schematic diagram of the comparison between the composition of the reservoir fluid up to the C30+ component and the experimental value predicted by the measurement model of the embodiment of the application;
  • Figure 11(a) is a schematic diagram of the comparison between the composition composition of the reservoir fluid predicted by the measurement model of the embodiment of the application and the experimental value;
  • Figure 11(b) is a schematic diagram of the comparison between the gas-oil ratio of the reservoir fluid predicted by the measurement model of the embodiment of the application and the experimental value;
  • Figure 12(a) is a schematic diagram of the comparison between the API specific gravity of the reservoir fluid predicted by the measurement model of the embodiment of the application and the experimental value;
  • Figure 12(b) is a schematic diagram of the comparison between the reservoir fluid volume coefficient predicted by the measurement model of the embodiment of the application and the experimental value;
  • Figure 13(a) is a schematic diagram of the comparison between the reservoir fluid density predicted by the measurement model of the embodiment of the application and the experimental value;
  • Figure 13(b) is a schematic diagram of the comparison between the reservoir fluid viscosity predicted by the measurement model of the embodiment of the application and the experimental value;
  • FIG. 14 is a schematic diagram of the composition of the system for determining the composition and properties of formation fluids according to an embodiment of the application.
  • the embodiment of the present application provides a method for determining the composition and properties of formation fluid. As shown in Fig. 1, the method may include steps S101-S104:
  • S104 Directly output the processing result as the real-time reservoir fluid composition and property data, or interpret or predict the real-time reservoir fluid composition and property data based on the processing result.
  • the measurement model is obtained by training a pre-created machine learning model based on large data about the composition and properties of a variety of reservoir fluids and measurement signals of downhole sensors as a sample data set. .
  • the obtaining a measurement model for the determination of the composition and properties of the formation fluid may include: calling a measurement model created and trained in advance, or creating and training the measurement model in real time.
  • the creating and training the measurement model may include steps S201-S203:
  • the database may be a large database or a very large database.
  • various surface and downhole measurement methods can be used to measure the composition and property data of a variety of reservoir fluids, and establish a database that includes reservoir fluids (such as gas, condensate gas, oil, and water). , Oil-based mud, water-based mud, etc.) composition (such as CO2, C1, C2, C3,..., C29, C30+), properties and database of sensor measurement signal data.
  • reservoir fluids such as gas, condensate gas, oil, and water.
  • composition such as CO2, C1, C2, C3,..., C29, C30+
  • the establishment of a database regarding the composition and properties of a variety of reservoir fluids and the measurement signals of the downhole sensors may include steps S301-S303:
  • a representative (for example, meeting preset requirements) reservoir fluid sample can be obtained first.
  • surface sampling There are two ways to sample reservoir fluids: surface sampling and downhole sampling. If ground sampling is used, the operating conditions of the ground separator can be kept stable, the oil and gas separated by the ground separator can be collected, and related parameters such as gas-oil ratio can be recorded, so that the obtained gas-oil ratio and other related parameters can be recorded , In the laboratory, sample the collected oil and gas from the ground separator to produce a representative single-phase reservoir fluid sample. If downhole sampling is used, it can be checked in the laboratory whether the sample collected downhole is a representative single-phase reservoir fluid sample.
  • the method may further include:
  • the first preset pressure and the first preset temperature perform any one or more of the following measurements on the single-phase reservoir fluid sample: continuous near-infrared spectroscopy, NMR (nuclear magnetic resonance), acoustic waves, Fluorescence and dielectric constant are measured, and the measurement results are added to the database.
  • the first property may include, but is not limited to, any one or more of the following: gas-oil ratio, API (American Petroleum Institute) specific gravity, molecular weight, sulfur content, carbon content, hydrogen content, Watson K value, SARA content and paraffin content; among them, SARA refers to saturated hydrocarbons, aromatic hydrocarbons, gums and asphaltenes.
  • API American Petroleum Institute
  • the second property may include, but is not limited to, any one or more of the following: bubble point, dew point, CCE (Constant Composition Expansion, constant composition expansion) experimental characteristics, DL (Differential Liberation, Differential release) experimental characteristics, CVD (Constant Volume Depletion) experimental characteristics, separator experimental characteristics, density, viscosity, conductivity, compressibility, reservoir flow volume coefficient, paraffin production conditions and asphaltene production conditions.
  • a representative single-phase reservoir fluid sample is flashed in a single stage at atmospheric pressure (such as standard atmospheric pressure) and room temperature to separate the balanced flash gas and liquid.
  • Analyze the flash gas by gas chromatography, and measure the volume, temperature and pressure of the flash gas.
  • the flash liquid can be analyzed by gas chromatography, and the density and molecular weight of the flash liquid can be measured, and SARA (where SARA refers to saturated hydrocarbons, aromatic hydrocarbons, gums and asphaltenes) content analysis and paraffin content analysis, etc. .
  • the component composition, gas-oil ratio, API gravity, molecular weight, sulfur content, carbon content, and hydrogen content of representative single-phase reservoir fluid samples are obtained , Watson K value, SARA content, paraffin wax content and other properties.
  • a fluid PVT (pressure volume temperature) experiment can be performed on a representative single-phase reservoir fluid sample under high temperature and high pressure (such as the first preset pressure and the first preset temperature), To obtain any one or more of the following fluid properties: bubble point, dew point, CCE test, DL test, CVD test, separator test, density, viscosity, conductivity, compressibility, FVF (Formation Volume Factor, reservoir flow volume) Coefficient), paraffin production conditions, asphaltene production conditions, etc.
  • high temperature and high pressure continuous near-infrared spectroscopy, NMR, acoustic wave, fluorescence, and dielectric constant can be measured on representative single-phase reservoir fluid samples.
  • continuous near-infrared spectroscopy, NMR, acoustic wave, fluorescence, dielectric constant, and PVT data corresponding to the reservoir fluid sample measured during real-time downhole operations as a function of time or pumping volume may include :Component composition, gas-oil ratio, API gravity, molecular weight, sulfur content, carbon content, hydrogen content, Watson K value, SARA content, paraffin content, bubble point, dew point, CCE, DL, CVD, separator, density, viscosity , Electrical conductivity, compressibility, reservoir flow volume coefficient, paraffin production conditions, asphaltene production conditions, etc.
  • gas-water ratio and ion composition analysis can be performed on a water sample.
  • continuous near-infrared spectroscopy, NMR, acoustic wave, fluorescence, dielectric constant, ph value (acid-base value), etc. can be measured under high temperature and high pressure.
  • the parameter data measured in step S302 can all be added to the database.
  • the method may further include: adding different drilling mud filtrates to different single-phase reservoir fluid samples, and performing analysis on the single-phase reservoir fluid samples added with corresponding drilling mud filtrate.
  • the PVT experiment may further include: adding different drilling mud filtrates to different single-phase reservoir fluid samples, and performing analysis on the single-phase reservoir fluid samples added with corresponding drilling mud filtrate. The PVT experiment.
  • the method may further include: mixing oil and water, measuring oil moisture content, continuous near-infrared spectroscopy, nuclear magnetic resonance NMR, acoustic wave, fluorescence, and dielectric constant measurement, and combining the measurement results Join the database.
  • different drilling mud filtrates can be added to different reservoir fluid samples (such as water samples, oil samples, gas samples). ), and perform PVT test on the reservoir fluid sample added with drilling mud filtrate to measure the composition of the reservoir fluid sample, mud filtrate pollution degree, molecular weight, gas-oil ratio, API gravity, molecular weight, sulfur content, carbon content, Hydrogen content, Watson K value, bubble point, dew point, CCE, DL, CVD, separator, density, viscosity, compressibility, reservoir flow volume coefficient, gas-water ratio, etc., and can perform ion composition analysis.
  • continuous near-infrared spectroscopy, NMR, acoustic wave, fluorescence, dielectric constant, ph value, etc. can be measured under high temperature and high pressure. And various measurement results are added to the database to enrich the data in the database.
  • different proportions of oil and water can be mixed into different reservoir fluid samples (such as water samples, oil samples, gas samples), and the Continuous near-infrared spectroscopy, NMR, acoustic wave, fluorescence, dielectric constant and other measurements are performed on reservoir fluid samples mixed with corresponding proportions of gas, oil and water under pressure, high temperature and high pressure.
  • CFD computational Fluid Dynamics, computational fluid dynamics
  • reservoir simulator simulations can be used. The large amount of result data obtained is added to the database.
  • -Relational Database such as SQL databases
  • Non-relational databases Non-relational databases
  • the method for establishing a database is shown in FIG. 4 and includes the following steps:
  • SARA saturated hydrocarbons, aromatic hydrocarbons, colloids and Asphaltene
  • a fluid PVT (pressure volume temperature) experiment is performed on the obtained single-phase reservoir fluid sample to obtain any one or more of the following fluid properties: bubble point, dew point, CCE experiment, DL Experiment, CVD experiment, separator experiment, density, viscosity, conductivity, compressibility, FVF, paraffin production conditions, asphaltene production conditions, etc.
  • continuous near-infrared spectroscopy, NMR, acoustic wave, fluorescence, dielectric constant, etc. are measured on representative single-phase reservoir fluid samples. Put the measurement results into the database.
  • continuous near-infrared spectroscopy, NRM, acoustic wave, fluorescence, dielectric constant, and PVT data of corresponding reservoir fluid samples measured during downhole real-time operations including: composition, gas-oil ratio, API gravity, molecular weight, sulfur Content, carbon content, hydrogen content, Watson K value, SARA content, paraffin content, bubble point, dew point, CCE, DL, CVD, separator, density, viscosity, conductivity, compressibility, dielectric constant, paraffin wax generation conditions, Asphaltene production conditions, etc., are also included in the database.
  • the downhole real-time data and laboratory data are put into the database.
  • the extracting the sample data set from the database, and training a pre-created machine learning model through the sample data set may include step S401 -S403:
  • the preprocessing may include any one or more of the following: denoising, removing outliers, and smoothing.
  • the database has been established through the foregoing steps using different methods.
  • the data set (part or all of the data in the database) taken out from the database, that is, the sample data set, can be preprocessed. Turn the original feature vectors (Feature Vectors) into expressions that are more suitable for downstream calculations.
  • One of the methods may include: first denoising, removing outliers, smoothing, etc. on the retrieved data set using filters and mathematical methods, and estimating some default values. Then standardize the data after the above processing, for example, make the data set look like standard normally distributed data: Gaussian distribution with zero mean and unit variance.
  • the standardized data can be divided into a first data set and a second data set, which can be used for supervised machine learning.
  • the first data set may be used as an input data vector of the machine learning model
  • the second data set may be used as a target data vector of an output data vector of the machine learning model
  • S403 Based on the first data set and the second data set, use a preset machine learning method to train the machine learning model, and obtain the trained optimal machine learning model as the measurement model.
  • the preset machine learning method may include: a supervised machine learning method.
  • the machine learning model is trained using a preset machine learning method based on the first data set and the second data set to obtain training
  • the optimal machine learning model in the result as the measurement model may include steps S501-S503:
  • S501 For each function in the predefined function set, perform the following operations: input the standardized first data set into the untrained machine learning model; according to the first data set and the machine learning The function currently loaded in the model calculates the output data vector;
  • the supervised machine learning method may include: (1) inputting a standardized input data vector and a target data vector; (2) defining a function set; (3) according to the standardized input data vector and definition Calculate the value of the output data vector, and compare the calculation result of the function set with the value of the target data vector retrieved from the database to judge the quality of the function set (that is, judge the quality of the currently trained machine learning model Bad); (4) Select the best function as the result of machine learning (that is, the corresponding machine learning model is used as the measurement model).
  • FIG. 7 a schematic diagram of a supervised machine learning process in an exemplary embodiment is provided, including the following operations:
  • Relevant data sets are retrieved from the database for preprocessing and standardization, and the input data vector X and the target vector Y are obtained respectively.
  • Different methods can be used to reduce the dimensions of X and Y as needed.
  • any one or more of the following dimensionality reduction algorithms can be used: PCA (Principal Component Analysis), FA (Factor Analysis), ICA (Independent Component Analysis), NMF (Non -negative Matrix Factorization, non-negative matrix factorization), LDA (Linear Discriminant Analysis, linear discriminant analysis), Laplacian map (Laplacian Eigenmap), LLE (Locally Linear Embedding, local linear embedding algorithm), Isomap (Isometric Feature Mapping) , Isometric feature mapping algorithm) and so on.
  • a function set can be defined, which can be a linear function set or a nonlinear function set.
  • the output data vector value (Y') to be predicted can be expressed as a linear combination of the input data vector (X):
  • W is the value of the coefficient (parameter) to be determined.
  • the output data vector Y'to be predicted can be expressed as a nonlinear combination of the input vector X:
  • V is the value of the coefficient (parameter) to be determined.
  • SVM Small Vector Machines
  • Deep Learning Deep Learning
  • multi-layer neural network K-Nearest Neighbor
  • Gaussian process can be used for the nonlinear model.
  • Regression Gaussian Processes Regression
  • Bayesian regression Bayesian regression
  • Decision Tree Decision Tree
  • Random Forests Random Forests
  • the purpose of these machine learning methods is to find a set of optimal V coefficients (parameters) values to minimize the error between the predicted output data vector Y'and the target vector Y retrieved from the database.
  • the error objective function is: min ⁇ Y-Y' ⁇ .
  • the data retrieved from the database can be divided into two groups: one group is a training data set, and the other is a test data set.
  • the training data set can be used to determine an optimal function (such as the aforementioned first function) and coefficients (parameters) of the optimal function.
  • the test data set can be used for cross-validation to ensure that the best function and the coefficients (parameters) of the best function are used to determine the composition and properties of formation fluids without losing the original accuracy (that is, without over-simulation). Together phenomenon occurs).
  • Gaussian Processes can be used to optimize parameters.
  • the measurement model used can be determined, thereby completing the creation and training of the measurement model.
  • the downhole oil and gas layer tester may be a downhole midway oil and gas layer tester.
  • the method before inputting the signal measured in real time by the sensor on the downhole oil and gas formation tester as the input data of the measurement model into the measurement model, the method may further include:
  • the first method is The offline (off-line) method
  • the other is the online (on-line) method.
  • relevant data can be retrieved from the database, and the relevant model can be trained using the aforementioned machine learning method to obtain the functions and parameters of the relevant model, thereby obtaining the final training Good measurement model. Therefore, the obtained measurement model can be embedded in the equipment and/or software for downhole cable and midway hydrocarbon layer testing while drilling, and can be applied in real time during downhole cable and midway hydrocarbon layer testing while drilling.
  • the application flow chart is shown in Figure 8. Show, including:
  • data can be retrieved from the database, and data can also be stored in the previous database;
  • composition CO2, C1, C2,..., C30+
  • gas-oil ratio API gravity
  • sulfur content sulfur content
  • carbon content hydrogen content
  • Watson K value density viscosity
  • electrical conductivity And gas, oil, moisture rate, and the pollution degree of oil-based and water-based mud filtrate; predict the time and/or volume required for sampling; remove the influence of mud filtrate on the composition and other properties of the reservoir fluid;
  • the downhole cable and the midway oil and gas formation testing instrument while drilling are placed into the required depth downhole, the probe and/or packer are in contact with the reservoir and sealed, and the storage is pumped by a pump.
  • Layer fluid into the flow pipe of the instrument the fluid flows through a variety of sensors on the test instrument (such as one or more of the following: density, viscosity, conductivity, permittivity, continuous near infrared spectroscopy, discontinuous near infrared spectroscopy, Nuclear magnetism, acoustic wave, fluorescence, dielectric constant, etc.).
  • These sensors can measure the properties of flowing fluids in real time (such as one or more of the following: density, viscosity, conductivity, dielectric constant, continuous near infrared spectroscopy, discontinuous near infrared spectroscopy, nuclear magnetism, acoustic waves, fluorescence, etc.).
  • the aforementioned preprocessing and standardization methods can be used to preprocess and standardize these measured fluid properties, which can be used as the input of the aforementioned machine learning method training to obtain the measurement model.
  • the measurement model can be used for interpretation and prediction, and the composition of downhole reservoir fluids can be obtained in real time: including CO2, C1, C2,..., C5, C6+, or CO2, C1, C2,..., C6, C7+, or CO2, C1, C2,..., C7, C8+,..., or CO2, C1, C2,..., C29, C30+; and the properties of downhole reservoir fluids can be obtained in real time: gas-oil ratio, API specific gravity, molecular weight , Density, viscosity, conductivity, gas, oil, moisture content, sulfur content, carbon content, hydrogen content, Watson K value, SARA (saturated hydrocarbons, aromatic hydrocarbons, gums, asphaltenes) content, paraffin content, sound velocity, foam Point, dew point, CCE, DL, CVD, separator experiment, compressibility, reservoir fluid volume factor, paraffin generation conditions, asphaltene generation conditions, and pollution degree of oil-based and water-based mud filtrate.
  • gas-oil ratio including CO2, C1,
  • the on-line mode can be divided into two types.
  • the first method is to retrieve relevant data from the database, train the relevant model using the aforementioned machine learning method, and obtain the parameters of the relevant model (ie, obtain the measurement model).
  • Related models and parameters can be embedded in equipment and/or software for downhole cable and hydrocarbon formation testing while drilling. Adjust and optimize the parameters of the measurement model with the data measured during the test of the downhole cable and the midway oil and gas layer while drilling (especially the real-time measured time series data, and the real-time data can also be stored in the database), and use the related models and parameters obtained (Optimized measurement model) to make predictions.
  • the second method is to directly use the data measured during the downhole cable and midway oil and gas layer test while drilling (especially the real-time measured time series data, and the real-time data can also be stored in the database) as input data, using the aforementioned machine learning Methods to train related models and parameters.
  • the related model and parameters (the measurement model) obtained are then used for prediction.
  • an on-line method may be applied to real-time prediction. For example, predict the time and/or volume required for sampling, the pollution degree of the mud filtrate, the removal of mud filtrate pollution on the composition and other properties of the reservoir fluid (such as one or more of the following: bubble point, dew point, density, viscosity , Conductivity, permittivity, continuous near-infrared spectroscopy, discontinuous near-infrared spectroscopy, nuclear magnetism, acoustic waves, fluorescence, permittivity, gas-oil ratio, API gravity, molecular weight, sulfur content, carbon content, hydrogen content, Watson K value , Oil-water content, SARA content, paraffin content, compressibility, reservoir fluid volume coefficient, paraffin production conditions, asphaltene production conditions).
  • the reservoir fluid such as one or more of the following: bubble point, dew point, density, viscosity , Conductivity, permittivity, continuous near-infrared spectroscopy, discontinuous near-infrared spectroscopy,
  • the method may further include: in one or more of the following processes, supplementing the output data vector whose error with the target data vector meets a preset error threshold to the database : During the training process of the machine learning model and the implementation process of determining the composition and properties of the formation fluid.
  • a large number of reservoir fluid samples dry gas, moisture, condensate gas, volatile oil, black oil, heavy oil
  • a large number of the aforementioned reservoir fluid samples can be measured.
  • an in-house database was established. The continuous or discontinuous near-infrared spectra of a variety of reservoir fluids can be extracted from this database, up to the C30+ component composition, gas-oil ratio, API specific gravity, fluid volume coefficient, density and viscosity data. Part of the fluid spectrum data is shown in Figure 9(a) and Figure 9(b).
  • these data can be preprocessed and standardized.
  • the best model and parameters (that is, the final measurement model) obtained are used to make predictions. That is, using the near-infrared spectroscopy as the input data vector, the obtained machine learning model (that is, the measurement model) is used to predict fluid properties.
  • the prediction results are described as follows:
  • Figure 10 shows the comparison between the composition of the C30+ component predicted by the machine learning model (herein the measurement model) and the experimental value, and the predicted result of the machine learning model is consistent with the experimental value. It is also possible to combine the reorganized components in the reservoir fluid to obtain the composition of C6+, C7+,..., C29+, etc., namely Among them, ZCn+ and ZCi are the components of Cn+ and Ci.
  • Figure 11(a) and Figure 11(b) compare the composition (CO2, C1, C2, C3, C4, C5, C7+) predicted by the machine learning model (here, the measurement model) and the gas-oil ratio with experimental values. Due to the 256-channel near-infrared spectrometer and machine learning method, it has more information and features than other existing downhole spectrometers, so it can get more accurate prediction results of component composition and gas-oil ratio.
  • Figure 12(a) and Figure 12(b) compare the API specific gravity and fluid volume coefficient predicted by the machine learning model (here, the measurement model) with the experimental values.
  • Figure 13(a) and Figure 13(b) show the comparison of the reservoir fluid density and viscosity predicted by the machine learning model with experimental values. It can be seen from these comparison graphs that the machine learning model has better learning and prediction capabilities.
  • the machine learning model (herein referred to as the measurement model) can also be continuously updated and improved.
  • In-house machine learning devices in addition to supervised machine learning methods, can also include semi-supervised machine learning methods and unsupervised machine learning methods. Combining the database and these kinds of machine learning methods can get more intelligent downhole cables and AI (Artificial Intelligence) products for midway hydrocarbon formation testing while drilling.
  • AI Artificial Intelligence
  • the embodiment of the present application also provides a system 1 for determining the composition and properties of formation fluids.
  • the system may include: a processor 11 and a computer-readable storage medium 12, the computer-readable storage medium 12 Instructions are stored in the processor 11, and the processor 11 is configured to execute the instructions and perform the following operations:
  • the processor 11 when the processor 11 executes the instruction, the method for determining the composition and properties of the formation fluid described in any of the foregoing embodiments may be performed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Mining & Mineral Resources (AREA)
  • Geology (AREA)
  • Chemical & Material Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Fluid Mechanics (AREA)
  • Environmental & Geological Engineering (AREA)
  • Geochemistry & Mineralogy (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medicinal Chemistry (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • General Chemical & Material Sciences (AREA)
  • Oil, Petroleum & Natural Gas (AREA)
  • Computing Systems (AREA)
  • Food Science & Technology (AREA)
  • Mathematical Physics (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Sampling And Sample Adjustment (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)
  • Production Of Liquid Hydrocarbon Mixture For Refining Petroleum (AREA)

Abstract

一种测定地层流体组成和性质的方法和系统，方法包括：获取用于地层流体组成和性质测定的测定模型；将井下油气层测试仪上的传感器实时测量的信号作为输入数据输入测定模型；通过测定模型对输入数据进行处理；直接输出处理结果作为实时地层流体的组成和性质数据，或者，根据处理结果解析出实时地层流体的组成和性质数据，系统包括：处理器和计算机可读存储介质；计算机可读存储介质中存储有指令；处理器执行指令，进行前述测定地层流体组成和性质的方法。该方法和系统可测定地层流体组成和性质，提高测井效率和准确率。

Description

一种测定地层流体组成和性质的方法和系统 技术领域
本申请实施例涉及但不限于测井技术，尤指一种测定地层流体组成和性质的方法和系统。
背景技术
近几年来，井下流体识别技术，尤其是硬件方面，已经取得了长足的进展。但在井下流体实时分析和原状地层流体取样方面，尤其是用井下光谱实时分析来得到井下流体组份的组成和性质(如油、气、水分率、气油比、泥浆滤液污染、实时取样指导)等，还有待提高效率和准确率。
发明概述
以下是对本文详细描述的主题的概述。本概述并非是为了限制权利要求的保护范围。
本申请实施例提供了一种测定地层流体组成和性质的方法和系统，可以快速、准确地测定地层流体组成和性质，提高测井效率和准确率。
一方面，本申请实施例提供了一种测定地层流体组成和性质的方法，所述方法可以包括：
获取用于地层流体组成和性质测定的测定模型；
将井下油气层测试仪上的传感器实时测量的信号作为输入数据输入所述测定模型；
通过所述测定模型对所述输入数据进行处理；
直接输出处理结果作为实时储层流体的组成和性质数据，或者，根据处理结果解析出实时储层流体的组成和性质数据。
在一种示例性实施例中，所述测定模型是以关于多种储层流体的组成和性质，以及井下传感器测量信号的大数据为样本数据集，对预先创建好的机器学习模型进行训练获得的。
在一种示例性实施例中，所述获取用于地层流体组成和性质测定的测定模型可以包括：调取预先创建并训练好的测定模型，或者，实时创建并训练所述测定模型。
在一种示例性实施例中，在将所述井下油气层测试仪上的传感器实时测量的信号作为所述测定模型的输入数据输入所述测定模型之前，所述方法还可以包括：
将预先训练好的所述测定模型加载入井下油气层测试仪中，以在所述井下油气层测试仪进行实时测井过程中，将所述井下油气层测试仪上的传感器实时测量的信号输入所述测定模型。
在一种示例性实施例中，所述创建并训练所述测定模型可以包括：
建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库；
从所述数据库中抽取所述样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练；
从训练后的机器学习模型中获取最优机器学习模型作为所述测定模型。
在一种示例性实施例中，所述建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库可以包括：
通过以下一种或多种方式获取符合预设要求的单相储层流体样品：地面采样、井下采样；所述单相储层流体样品可以包括：油相储层流体样品、气相储层流体样品和水相储层流体样品；
在标准大气压和室温下，对所述单相储层流体样品进行单级闪蒸，以分离出平衡的闪蒸气体和闪蒸液体；对所述闪蒸气体和所述闪蒸液体分别进行气相色谱分析，获取所述单相储层流体样品的组成和第一性质；在第一预设压力和第一预设温度下，对所述单相储层流体样品进行流体压力体积温度PVT实验，获取所述单相储层流体样品的第二性质；其中所述第一预设压力大于所述标准大气压，所述第一预设温度大于所述室温；
将所述单相储层流体样品的组成、所述第一性质和所述第二性质的相关数据作为所述地层流体大数据的一部分加入所述数据库。
在一种示例性实施例中，所述第一性质可以包括以下任意一种或多种：气油比、美国石油学会API比重、分子量、硫含量、碳含量、氢含量、WatsonK值、SARA含量和石蜡含量；其中，SARA是指饱和烃、芳香烃、胶质和沥青质；
所述第二性质可以包括以下任意一种或多种：泡点、露点、恒组成膨胀CCE实验特性、微分释放DL实验特性、恒容衰竭CVD实验特性、分离器实验特性、密度、粘度、电导率、压缩系数、储层流体积系数、石蜡生成条件以及沥青质生成条件。
在一种示例性实施例中，所述方法还可以包括：
进行以下任一种或多种操作：
对所述闪蒸气体进行气相色谱分析时测量所述闪蒸气体的体积、温度和压力加入所述数据库；
对所述闪蒸液体进行气相色谱分析时测量所述闪蒸液体的体积、温度、压力、密度和分子量加入所述数据库；
在所述第一预设压力和所述第一预设温度下，对所述单相储层流体样品进行以下任意一种或多种测量：连续近红外光谱、核磁共振NMR、声波、荧光以及介电常数测量，并将测量结果加入所述数据库。
在一种示例性实施例中，所述方法还可以包括：在不同的单相储层流体样品中加入不同的钻井泥浆滤液，并对加入相应的钻井泥浆滤液的单相储层流体样品进行所述PVT实验；并且，对油和水进行混合，测量油水分率、连续近红外光谱、核磁共振NMR、声波、荧光以及介电常数测量，并将测量结果加入所述数据库。
在一种示例性实施例中，所述从所述数据库中抽取所述样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练可以包括：
对所述样本数据集进行预处理和标准化；所述预处理可以包括以下任意一种或多种：去噪，去离群点和光滑；
将经过所述预处理和标准化的样本数据集划分为第一数据集和第二数据集；
基于所述第一数据集和所述第二数据集，采用预设的机器学习方法对所述机器学习模型进行训练，获取训练出的最优机器学习模型作为所述测定模型；
其中，所述第一数据集作为所述机器学习模型的输入数据向量，所述第二数据集作为所述机器学习模型的输出数据向量的目标数据向量。
其中，所述输入数据向量和目标数据向量均可以为一维的或多维的。
在一种示例性实施例中，所述预设的机器学习方法可以包括：有监督的机器学习方法；
所述基于所述第一数据集和所述第二数据集，采用预设的机器学习方法对所述机器学习模型进行训练，获取训练结果中的最优机器学习模型作为所述测定模型可以包括：
针对预先定义的函数集中的每个函数，分别执行以下操作：将标准化的所述第一数据集输入未经训练的所述机器学习模型；根据所述第一数据集和所述机器学习模型中当前加载的函数计算所述输出数据向量；
将计算出的多个所述输出数据向量分别与所述目标数据向量相比较，从多个比较结果中目标数据向量确定出与所述目标数据向量之间的误差最小的第一输出数据向量对应的第一函数和所述第一函数的当前系数；
将加载有所述第一函数的机器学习模型作为所述最优机器学习模型；其中，所加载的所述第一函数中具有所述当前系数。
在一种示例性实施例中，所述方法还可以包括：在以下一个或多个过程中，将与所述目标数据向量的误差满足预设的误差阈值的输出数据向量补充到所述数据库中：对所述机器学习模型进行训练过程中、进行地层流体组成和性质的实施测定过程中。
另一方面，本申请实施例还提供了一种测定地层流体组成和性质的系统，所述系统可以包括：处理器和计算机可读存储介质，所述计算机可读存储介质中存储有指令，所述处理器设置成执行所述指令，进行上述任意一实施例所述的测定地层流体组成和性质的方法。
在阅读并理解了附图概述和本申请的实施方式后，可以明白其他方面。 附图概述
附图用来提供对本申请实施例技术方案的进一步理解，并且构成说明书的一部分，与本申请实施例一起用于解释技术方案，并不构成对技术方案的限制。
图1为本申请实施例的测定地层流体组成和性质的方法流程图；
图2为本申请实施例的创建并训练所述测定模型的方法流程图；
图3为本申请实施例的建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库的方法流程图；
图4为本申请实施例的建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库的方法示意图；
图5为本申请实施例的从所述数据库中抽取所述样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练的方法流程图；
图6为本申请实施例的采用预设的机器学习方法对所述机器学习模型进行训练，获取训练结果中的最优机器学习模型作为所述测定模型的方法流程图；
图7为本申请实施例的有监督的机器学习流程示意图；
图8为本申请实施例的基于数据库和测定模型的井下电缆和随钻中途油气层测试应用示意图；
图9(a)为本申请实施例的从数据库中取出的部分流体(轻油+气)的光谱图示意图；
图9(b)为本申请实施例的从数据库中取出的部分流体(各种原油)的光谱图示意图；
图10为本申请实施例的测定模型预测的储层流体直到C30+组份的组成与实验值的比较示意图；
图11(a)为本申请实施例的测定模型预测的储层流体组份组成与实验值的比较示意图；
图11(b)为本申请实施例的测定模型预测的储层流体气油比与实验值的比较示意图；
图12(a)为本申请实施例的测定模型预测的储层流体API比重与实验值的比较示意图；
图12(b)为本申请实施例的测定模型预测的储层流体体积系数与实验值的比较示意图；
图13(a)为本申请实施例的测定模型预测的储层流体密度与实验值的比较示意图；
图13(b)为本申请实施例的测定模型预测的储层流体粘度与实验值的比较示意图；
图14为本申请实施例的测定地层流体组成和性质的系统组成示意图。
详述
下文中将结合附图对本申请实施例进行说明。在不冲突的情况下，本申请实施例及实施例中的特征可以相互任意组合。
在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行。并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
本申请实施例提供了一种测定地层流体组成和性质的方法，如图1所示，所述方法可以包括步骤S101-S104：
S101、获取用于地层流体组成和性质测定的测定模型；
S102、将井下油气层测试仪上的传感器实时测量的信号作为输入数据输入所述测定模型；
S103、通过所述测定模型对所述输入数据进行处理；
S104、直接输出处理结果作为实时储层流体的组成和性质数据，或者，根据处理结果解释或预测出实时储层流体的组成和性质数据。
一种示例性实施例中，所述测定模型是以关于多种储层流体的组成和性 质，以及井下传感器测量信号的大数据为样本数据集，对预先创建好的机器学习模型进行训练获得的。
在一种示例性实施例中，所述获取用于地层流体组成和性质测定的测定模型可以包括：调取预先创建并训练好的测定模型，或者，实时创建并训练所述测定模型。
在一种示例性实施例中，如图2所示，所述创建并训练所述测定模型可以包括步骤S201-S203：
S201、建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库；
S202、从所述数据库中抽取所述样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练；
S203、从训练后的机器学习模型中获取最优机器学习模型作为所述测定模型。
在一种示例性实施例中，所述数据库可以是大型数据库或超大型数据库。
在一种示例性实施例中，可以首先采用各种地面和井下的测量手段，测量多种储层流体的组成和性质数据，建立一个包括储层流体(比如气、凝析气、油、水、油基泥浆、水基泥浆等)的组成(比如CO2,C1,C2,C3,…,C29,C30+)、性质和传感器测量信号数据的数据库。
在一种示例性实施例中，如图3所示，所述建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库可以包括步骤S301-S303：
S301、通过以下一种或多种方式获取符合预设要求的单相储层流体样品：地面采样、井下采样；所述单相储层流体样品可以包括：油相储层流体样品、气相储层流体样品和水相储层流体样品。
在一种示例性实施例中，首先可以得到具有代表性(如符合预设要求)的储层流体样品。储层流体的取样方式可以包括两种：地面取样和井下取样。如果采用地面取样，则可以使地面分离器的操作条件保持稳定，可以对地面分离器分离出的油和气进行采集，并记录气油比等相关参数，这样可以根据得到的气油比等相关参数，在实验室对采集的地面分离器的油和气进行配样， 产生具有代表性的单相储层流体样品。如果采用井下取样，则可以在实验室中检查井下采集的样品是否是具有代表性的单相储层流体样品。
S302、在标准大气压和室温下，对所述单相储层流体样品进行单级闪蒸，以分离出平衡的闪蒸气体和闪蒸液体；对所述闪蒸气体和所述闪蒸液体分别进行气相色谱分析，获取所述单相储层流体样品的组成和第一性质；在第一预设压力和第一预设温度下，对所述单相储层流体样品进行流体压力体积温度PVT实验，获取所述单相储层流体样品的第二性质；其中所述第一预设压力大于所述标准大气压，所述第一预设温度大于所述室温。
S303、将所述单相储层流体样品的组成、所述第一性质和所述第二性质的相关数据作为所述地层流体大数据的一部分加入所述数据库。
在一种示例性实施例中，所述方法还可以包括：
进行以下任一种或多种操作：
对所述闪蒸气体进行气相色谱分析时测量所述闪蒸气体的体积、温度和压力加入所述数据库；
对所述闪蒸液体进行气相色谱分析时测量所述闪蒸液体的体积、温度、压力、密度和分子量加入所述数据库；
在所述第一预设压力和所述第一预设温度下，对所述单相储层流体样品进行以下任意一种或多种测量：连续近红外光谱、NMR(核磁共振)、声波、荧光以及介电常数测量，并将测量结果加入所述数据库。
在一种示例性实施例中，所述第一性质可以包括但不限于以下任意一种或多种：气油比、API(美国石油学会)比重、分子量、硫含量、碳含量、氢含量、Watson K值、SARA含量和石蜡含量；其中，SARA是指饱和烃、芳香烃、胶质和沥青质。
在一种示例性实施例中，所述第二性质可以包括但不限于以下任意一种或多种：泡点、露点、CCE(Constant Composition Expansion，恒组成膨胀)实验特性、DL(Differential Liberation，微分释放)实验特性、CVD(Constant Volume Depletion，恒容衰竭)实验特性、分离器实验特性、密度、粘度、电导率、压缩系数、储层流体积系数、石蜡生成条件以及沥青质生成条件。
在一种示例性实施例中，在大气压(如标准大气压)和室温下，对具有代表性的单相储层流体样品进行单级闪蒸，分离出平衡的闪蒸气体和液体。对闪蒸气体进行气相色谱分析，并测量闪蒸气体的体积、温度和压力。另外，可以对闪蒸液体进行气相色谱分析，并测量闪蒸液体的密度和分子量，并进行SARA(其中，SARA是指饱和烃、芳香烃、胶质和沥青质)含量分析和石蜡含量分析等。根据所测的相关参数和单级闪蒸时的物质平衡方程，得到具有代表性的单相储层流体样品的组份组成、气油比、API比重、分子量、硫含量、碳含量、氢含量、Watson K值、SARA含量、石蜡含量等性质。
在一种示例性实施例中，在高温高压(如第一预设压力和第一预设温度)下，可以对具有代表性的单相储层流体样品进行流体PVT(压力体积温度)实验，从而得到如下任意一种或多种流体性质：泡点、露点、CCE实验、DL实验、CVD实验、分离器实验、密度、粘度、电导率、压缩系数、FVF(Formation Volume Factor，储层流体积系数)、石蜡生成条件、沥青质生成条件等。另外，在高温高压下，可以对具有代表性的单相储层流体样品进行连续近红外光谱、NMR、声波、荧光、介电常数等测量。
在一种示例性实施例中，井下实时作业时测量的随时间或泵抽体积而变化的连续近红外光谱、NMR、声波、荧光、介电常数以及相对应储层流体样品的PVT数据可以包括：组份组成、气油比、API比重、分子量、硫含量、碳含量、氢含量、Watson K值、SARA含量、石蜡含量、泡点、露点、CCE、DL、CVD、分离器、密度、粘度、电导率、压缩系数、储层流体积系数、石蜡生成条件、沥青质生成条件等。
在一种示例性实施例中，可以对水样品进行气水比、离子组成分析。另外，可以在高温高压下进行连续近红外光谱、NMR、声波、荧光、介电常数、ph值(酸碱值)等测量。
在一种示例性实施例中，通过步骤S302测量的参数数据均可以加入数据库。
在一种示例性实施例中，所述方法还可以包括：在不同的单相储层流体样品中加入不同的钻井泥浆滤液，并对加入相应的钻井泥浆滤液的单相储层流体样品进行所述PVT实验。
在一种示例性实施例中，所述方法还可以包括：对油和水进行混合，测量油水分率、连续近红外光谱、核磁共振NMR、声波、荧光以及介电常数测量，并将测量结果加入所述数据库。
在一种示例性实施例中，为了考虑不同钻井泥浆(水基和油基)滤液对流体的影响，可以加入不同钻井泥浆滤液到不同储层流体样品(如：水样、油样、气样)中，并对加入钻井泥浆滤液的储层流体样品进行PVT试验，测量该储层流体样品组份组成、泥浆滤液污染程度、分子量、气油比、API比重、分子量、硫含量、碳含量、氢含量、Watson K值、泡点、露点、CCE、DL、CVD、分离器、密度、粘度、压缩系数、储层流体积系数、气水比等，并可以进行离子组成分析。此外，在高温高压下可以进行连续近红外光谱、NMR、声波、荧光、介电常数、ph值等测量。并将各种测量结果加入所述数据库中，以丰富数据库的数据。
在一种示例性实施例中，为了考虑井下油、水分率的判别，可以混合不同比例的油和水到不同储层流体样品(如：水样、油样、气样)中，在常温常压和高温高压下，并对混合了相应比例的气、油和水的储层流体样品进行连续近红外光谱、NMR、声波、荧光、介电常数等测量。另外，对混合了相应比例的气、油和水的储层流体样品进行PVT试验，测量储层流体样品组份组成、泥浆滤液污染程度、分子量、气油比、API比重、硫含量、碳含量、氢含量、Watson K值、泡点、露点、CCE、DL、CVD、分离器、密度、粘度、压缩系数、储层流体积系数、气水比等，并可以进行离子组成分析。最后可以将测量结果加入所述数据库中，丰富数据库的数据。
在一种示例性实施例中，根据大量不同油藏特性参数，不同储层流体性质参数和不同测井工具几何参数等，可以用CFD(Computational Fluid Dynamics，计算流体力学)和油藏模拟器仿真得到的大量结果数据加入数据库。
在一种示例性实施例中，包括但不限于上述这些井下实时和实验室测量的数据(和其他数据)，都可以存储在关系数据库(Relational Database如SQL数据库)和/或非关系数据库(Non-Relational Database如NoSQL数据库)中，形成内部(in-house)电缆和随钻中途油气层测试的数据库。
在一种示例性实施例中，数据库的建立方法如图4所示，包括以下步骤：
收集地面分离器分离出的油和气样品，配样成单相储层流体样品；
另外，收集井下单相储层流体样品。
一方面，在大气压和室温下，对得到单相储层流体样品进行单级闪蒸，获得闪蒸平衡的气体和液体；
对气体进行气相色谱分析，并测量闪蒸气体的体积、温度和压力。对闪蒸液体进行气相色谱分析，并测量闪蒸液体的密度、分子量、硫含量、碳含量、氢含量、Watson K值、并进行SARA(其中，SARA是指饱和烃、芳香烃、胶质和沥青质)含量分析和石蜡含量分析等。得到单相储层流体样品的组分组成、气油比、API比重、分子量、硫含量、碳含量、氢含量、Watson K值、SARA含量、石蜡含量等性质，放入数据库。
另一方面，在高温高压下，对所得到的单相储层流体样品进行流体PVT(压力体积温度)实验，从而得到如下任意一种或多种流体性质：泡点、露点、CCE实验、DL实验、CVD实验、分离器实验、密度、粘度、电导率、压缩系数、FVF、石蜡生成条件、沥青质生成条件等。在高温高压下，对具有代表性的单相储层流体样品进行连续近红外光谱、NMR、声波、荧光、介电常数等测量。将测量结果放入所述数据库。
再一方面，井下实时作业时测量的连续近红外光谱、NRM、声波、荧光、介电常数、以及相对应储层流体样品的PVT数据，包括：组成、气油比、API比重、分子量、硫含量、碳含量、氢含量、Watson K值、SARA含量、石蜡含量、泡点、露点、CCE、DL、CVD、分离器、密度、粘度、电导率、压缩系数、介电常数、石蜡生成条件、沥青质生成条件等，也放入所述数据库。
另外，以下数据也放入所述数据库：
根据大量不同油藏特性参数，不同储层流体性质参数和不同测井工具几何参数等，用CFD和油藏模拟器仿真得到的大量的结果数据；
对水样品进行气水比，离子组成分析，在高温高压下进行连续近红外光谱，NMR，声波，荧光，介电常数，ph值等测量的结果；
加入不同钻井泥浆滤液到样品中和混合不同比例的气、油和水，并进行PVT试验，测量组分组成、API比重、硫含量、碳含量、氢含量、Watson K值、粘度、气油比、气水比、离子组成分析，在高温高压下进行连续近红外光谱、NMR、声波、荧光、介电常数，ph值等测量的结果。
其中，井下实时数据和实验室数据都放入所述数据库。
在一种示例性实施例中，如图5所示，所述从所述数据库中抽取所述样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练可以包括步骤S401-S403：
S401、对所述样本数据集进行预处理和标准化；所述预处理可以包括以下任意一种或多种：去噪，去离群点和光滑。
在一种示例性实施例中，通过前述步骤采用不同的方法已经建立了数据库。在机器学习之前，可以对从数据库中取出的数据集(数据库中的部分或全部数据)，即所述样本数据集，进行预处理。将原始特征向量(Feature Vectors)变为更适合于下游计算的表达方式。其中一种方法可以包括：首先对取出的数据集用过滤器和数学方法进行去噪、去离群点、光滑等，对某些缺省值进行估算。然后对经过上述处理的数据进行标准化，例如使数据集看起来像标准的正态分布数据：即零均值和单位方差的高斯分布。
S402、将经过所述预处理和标准化的样本数据集划分为第一数据集和第二数据集。
在一种示例性实施例中，标准化之后的数据可以分为第一数据集和第二数据集，可用于有监督的机器学习。
在一种示例性实施例中，所述第一数据集可以作为所述机器学习模型的输入数据向量，所述第二数据集作为所述机器学习模型的输出数据向量的目标数据向量。
S403、基于所述第一数据集和所述第二数据集，采用预设的机器学习方法对所述机器学习模型进行训练，获取训练出的最优机器学习模型作为所述测定模型。
在一种示例性实施例中，所述预设的机器学习方法可以包括：有监督的 机器学习方法。
在一种示例性实施例中，如图6所示，所述基于所述第一数据集和所述第二数据集，采用预设的机器学习方法对所述机器学习模型进行训练，获取训练结果中的最优机器学习模型作为所述测定模型可以包括步骤S501-S503：
S501、针对预先定义的函数集中的每个函数，分别执行以下操作：将标准化的所述第一数据集输入未经训练的所述机器学习模型；根据所述第一数据集和所述机器学习模型中当前加载的函数计算所述输出数据向量；
S502、将计算出的多个所述输出数据向量分别与所述目标数据向量相比较，从多个比较结果中目标数据向量确定出与所述目标数据向量之间的误差最小的第一输出数据向量对应的第一函数和所述第一函数的当前系数；
S503、将加载有所述第一函数的机器学习模型作为所述最优机器学习模型；其中，所加载的所述第一函数中具有所述当前系数。
在一种示例性实施例中，有监督的机器学习方法可以包括：(1)输入标准化的输入数据向量和目标数据向量；(2)定义函数集；(3)根据标准化的输入数据向量和定义的函数集，计算输出数据向量的值，并将函数集的计算结果与从数据库中取出的目标数据向量的值进行对比，判断函数集的好坏(即判断当前所训练的机器学习模型的好坏)；(4)选择最好的一个函数作为机器学习的结果(即将相应的机器学习模型作为所述测定模型)。
如图7所示，给出了一种示例性实施例中有监督的机器学习流程示意图，包括如下操作：
从数据库中取出有关的数据集进行预处理和标准化，分别得到输入数据向量X和目标向量Y。根据需要可用不同方法对X和Y进行降维。例如，可以采用如下的任意一种或多种降维算法：PCA(Principal Component Analysis，主成分分析)、FA(Factor Analysis，因子分析)、ICA(Independent Component Analysis，独立成分分析)、NMF(Non-negative Matrix Factorization，非负矩阵分解)、LDA(Linear Discriminant Analysis，线性判别式分析)、拉普拉斯映射(Laplacian Eigenmap)、LLE(Locally Linear Embedding，局部线性嵌入算法)、Isomap(Isometric Feature Mapping，等距特征映射算法)等。可以定义函数集，该函数集可以是线性函数集或非线性函数集。
对于线性模型，要预测的输出数据向量值(Y’)可以表达成输入数据向量(X)的线性组合：
Y′＝F1(X，W)＝XW＝w 0+w 1x 1+…+w px p  (1)
其中W是要确定的系数(参数)值。
对于非线性模型，要预测的输出数据向量Y’可表达成输入向量X的非线性组合：
Y′＝F2(X，V)  (2)
其中V是要确定的系数(参数)值。
对于线性模型，可以采用普通最小二乘法(Ordinary Least Squares)、岭回归(Ridge Regression)、贝叶斯回归、贝叶斯岭回归、Logistic回归、随机梯度下降法、感知器法、多项式回归、PLS(Partial Least Squares，偏最小二乘法)等算法回归得到线性模型的参数(W)。
在一种示例性实施例中，对于非线性模型，可以采用SVM(Support VectorMachines，支持向量机)、深度学习(Deep Learning)、多层神经网络、K近邻算法(K-Nearest Neighbor)、高斯过程回归(Gaussian Processes Regression)、贝叶斯回归、决策树(Decision Tree)、随机森林(Random Forests)等算法。每次得到输出数据向量Y’后，判断与目标向量Y的误差是否最小，如果是则输出F(X)，如果不是则更新F(X)使误差目标函数最小，再返回到降维的步骤。这些机器学习方法的目的是找到一套最佳V系数(参数)值，使预测的输出数据向量Y’与从数据库中取出的目标向量Y之间的误差最小。通常误差目标函数为：min‖Y-Y′‖。
在一种示例性实施例中，在机器学习过程中，可以将从数据库中取出的数据分为两组：一组是训练数据集，另一组是测试数据集。训练数据集可以用来确定一个最佳函数(如前述的第一函数)和该最佳函数的系数(参数)。测试数据集可以用来做交叉验证，以确保得出的最佳函数和该最佳函数的系数(参数)用于进行测定地层流体组成和性质预测时不致失去原有的精度(即没有过度拟合现象发生)。通常，过于简单的函数难以保证预测精度，而过于复杂的函数又会造成过度拟合现象。因此，高斯过程(Gaussian Processes) 可用来优化参数。
在一种示例性实施例中，从上述步骤获取最优机器学习模型后，便可以确定出所用到的测定模型，从而完成测定模型的创建和训练。
一种示例性实施例中，所述井下油气层测试仪可以为井下中途油气层测试仪。
在一种示例性实施例中，在将所述井下油气层测试仪上的传感器实时测量的信号作为所述测定模型的输入数据输入所述测定模型之前，所述方法还可以包括：
将预先训练好的所述测定模型加载入井下油气层测试仪中，以在所述井下油气层测试仪进行实时测井过程中，将所述井下油气层测试仪上的传感器实时测量的信号输入所述测定模型。
在一种示例性实施例中，结合上述的数据库和测定模型，在井下电缆和随钻中途油气层测试时，可以包含两种关于储层流体的组成和性质的测定方式：第一种方式是线下(off-line)方式，另一种是线上(on-line)方式。
在一种示例性实施例中，在线下(off-line)方式中，可以从数据库中取出相关数据，用前述的机器学习方法训练好相关模型，得到相关模型的函数和参数，从而获得最终训练好的测定模型。因此，得到的测定模型可嵌入到井下电缆和随钻中途油气层测试的设备和/或软件中，可在井下电缆和随钻中途油气层测试时实时应用，应用时的流程图如图8所示，包括：
开始作业后，抽取储层流体到仪器的流动管道中；
测量流体的连续或非连续近红外光谱、NMR、声波、荧光、介电常数、密度、粘度、电导率；
进行数据预处理和标准化；
进行线下机器预学习和/或线上学习训练得到最佳模型，在此操作中可以从数据库取出数据，还可以往数据库存储数据；
将得到的最佳模型用于预测；
实时得到井下储层流体的组成和性质，包括组成(CO2,C1,C2,…,C30+)、气油比、API比重、硫含量、碳含量、氢含量、Watson K值密度、粘度、电 导率、以及气、油、水分率，以及油基、水基泥浆滤液的污染程度等；预测取样所需的时间和/或体积；去除泥浆滤液对储层流体组分组成和其它性质的影响等；
判断是否结束作业；如果是则结束作业，如果不是则在下一个时间从测量流体的连续或非连续近红外光谱、NMR、声波、荧光、介电常数、密度、粘度、电导率的步骤开始继续执行。
在一种示例性实施例中，将井下电缆和随钻中途油气层测试仪器放入井下所需的深度，将探针和/或封隔器与储层接触并密封好，用泵抽抽取储层流体到仪器的流动管道中，流体流动通过测试仪器上的多种传感器(如以下一种或多种：密度、粘度、电导率、介电常数、连续近红外光谱、非连续近红外光谱、核磁、声波、荧光、介电常数等)。这些传感器可实时测量流动流体的性质(如以下一种或多种：密度、粘度、电导率、介电常数、连续近红外光谱、非连续近红外光谱、核磁、声波、荧光等)。可以用前面所述的预处理和标准化的方法对这些测量的流体性质进行预处理和标准化，作为前述机器学习方法训练得到测定模型的输入。可以用测定模型进行解释和预测，可以实时得到井下储层流体的组成：包括CO2、C1、C2、...、C5、C6+，或者CO2、C1、C2、...、C6、C7+，或者CO2、C1、C2、...、C7、C8+、…，或者CO2、C1、C2、...、C29、C30+；并且可以实时得到井下储层流体的性质：气油比、API比重、分子量、密度、粘度、电导率、气、油、水分率、硫含量、碳含量、氢含量、Watson K值、SARA(饱和烃、芳香烃、胶质、沥青质)含量、石蜡含量、声速、泡点、露点、CCE、DL、CVD、分离器实验、压缩系数、储层流体体积系数、石蜡生成条件、沥青质生成条件以及油基、水基泥浆滤液的污染程度等。由于在同一深度不仅会进行井下实时测量而且会取样并送到实验室进行PVT测量。这些测量数据通过严格的质量检测后，可以加到数据库中不断充实数据库。另外，可以定时或不定时地更新机器学习的模型和参数，从而得到更好的井下测量结果。
在一种示例性实施例中，在线(on-line)方式可以分为两种。第一种方式是从数据库中取出相关数据，用前述的机器学习方法训练好相关模型，得到相关模型的参数(即获得所述测定模型)。相关模型和参数(所述测定模 型)可嵌入到井下电缆和随钻中途油气层测试的设备和/或软件中。用井下电缆和随钻中途油气层测试时测量的数据(尤其是实时测量的时间系列数据，并且实时数据也可储存到数据库中)对测定模型的参数进行调整优化，用得到的相关模型和参数(优化后的测定模型)进行预测。第二种方式是直接用井下井下电缆和随钻中途油气层测试时测量的数据(尤其是实时测量的时间系列数据，并且实时数据也可储存到数据库中)作为输入数据，用前述的机器学习方法训练相关模型和参数。再用得到的相关模型和参数(所述测定模型)进行预测。
在一种示例性实施例中，在线(on-line)方式可应用于实时预测。例如，预测取样所需的时间和/或体积、泥浆滤液的污染程度、去除泥浆滤液污染对储层流体组份组成和其他性质(如以下一种或多种：泡点、露点、密度、粘度、电导率、介电常数、连续近红外光谱、非连续近红外光谱、核磁、声波、荧光、介电常数、气油比、API比重、分子量、硫含量、碳含量、氢含量、Watson K值、油水分率、SARA含量、石蜡含量、压缩系数、储层流体体积系数、石蜡生成条件、沥青质生成条件)的影响等。
在一种示例性实施例中，所述方法还可以包括：在以下一个或多个过程中，将与所述目标数据向量的误差满足预设的误差阈值的输出数据向量补充到所述数据库中：对所述机器学习模型进行训练过程中、进行地层流体组成和性质的实施测定过程中。
在一种示例性实施例中，可以预先收集大量的(干气、湿气、凝析气、挥发油、黑油、稠油)储层流体样品，并对这些储层流体样品测量了大量的前述实验数据，建立了内部(in-house)的数据库。可以从此数据库中取出多种储层流体连续或非连续的近红外光谱，直到C30+的组份组成、气油比、API比重、流体体积系数、密度和粘度数据。部分流体光谱的数据如图9(a)和图9(b)所示。
在本实施例中，可以将这些数据进行预处理和标准化。用近红外光谱作为输入数据向量，而其他性质作为目标数据向量。将处理好的数据向量放到内部(in-house)编好的机器学习器中，对这些数据进行有监督的机器学习训练，并进行交叉验证。得到的最佳模型和参数(即最终的测定模型)则用来 进行预测。即用近红外光谱作为输入数据向量，用得到的机器学习模型(即所述测定模型)进行流体性质的预测，预测结果说明如下：
图10显示了机器学习模型(这里指测定模型)预测的直到C30+组份的组成与实验值的比较，机器学习模型预测的结果与实验值一致。也可以将储层流体中重组份合并在一起，得到C6+、C7+、…、C29+等组份的组成，即
Figure PCTCN2020074353-appb-000001
其中，ZCn+和ZCi是组份Cn+和Ci的组成。
图11(a)和图11(b)比较了机器学习模型(这里指测定模型)预测的组份(CO2、C1、C2、C3、C4、C5、C7+)组成和气油比与实验值。由于采用了256个通道的近红外光谱仪和机器学习方法，比其他现有井下光谱仪具有更多的信息和特征，因此能够得到更加准确的组份组成和气油比的预测结果。
图12(a)和图12(b)比较了机器学习模型(这里指测定模型)预测的API比重和流体体积系数与实验值。图13(a)和图13(b)显示了机器学习模型预测的储层流体密度和粘度与实验值的比较。由这些比较图可以看出，机器学习模型具有较好的学习和预测能力。
在一种示例性实施例中，随着数据库不断地完善和丰富，机器学习模型(这里指测定模型)也能不断地更新和完善。在内部(in-house)机器学习器中，除了有监督的机器学习方法之外，也可以包括半监督的机器学习方法和无监督的机器学习方法。结合数据库和这几种机器学习的方法，可以得到更加智能化的井下电缆和随钻中途油气层测试的AI(Artificial Intelligence，人工智能)化产品。
本申请实施例还提供了一种测定地层流体组成和性质的系统1，如图14所示，所述系统可以包括：处理器11和计算机可读存储介质12，所述计算机可读存储介质12中存储有指令，所述处理器11设置成执行所述指令，进行如下操作：
获取用于地层流体组成和性质测定的测定模型；
将井下油气层测试仪上的传感器实时测量的信号作为输入数据输入所述测定模型；
通过所述测定模型对所述输入数据进行处理；
直接输出处理结果作为实时储层流体的组成和性质数据，或者，根据处理结果解析出实时储层流体的组成和性质数据。
在一种示例性实施例中，所述处理器11执行所述指令时，可以进行上述任一实施例所述的测定地层流体组成和性质的方法。
虽然本申请实施例所揭露的实施方式如上，但所述的内容仅为便于理解本申请实施例而采用的实施方式，并非用以限定本申请实施例。任何本申请所属领域内的技术人员，在不脱离本申请实施例所揭露的精神和范围的前提下，可以在实施的形式及细节上进行任何的修改与变化，但本申请的专利保护范围，仍须以所附的权利要求书所界定的范围为准。

Claims (13)

  1. 一种测定地层流体组成和性质的方法，包括：
    获取用于地层流体组成和性质测定的测定模型；
    将井下油气层测试仪上的传感器实时测量的信号作为输入数据输入所述测定模型；
    通过所述测定模型对所述输入数据进行处理；
    直接输出处理结果作为实时储层流体的组成和性质数据，或者，根据处理结果解析出实时储层流体的组成和性质数据。
  2. 根据权利要求1所述的测定地层流体组成和性质的方法，其中，所述获取用于地层流体组成和性质测定的测定模型包括：调取预先创建并训练好的测定模型，或者，实时创建并训练所述测定模型。
  3. 根据权利要求1或2所述的测定地层流体组成和性质的方法，其中，在将所述井下油气层测试仪上的传感器实时测量的信号作为所述测定模型的输入数据输入所述测定模型之前，还包括：
    将预先训练好的所述测定模型加载入井下油气层测试仪中，以在所述井下油气层测试仪进行实时测井过程中，将所述井下油气层测试仪上的传感器实时测量的信号输入所述测定模型。
  4. 根据权利要求3所述的测定地层流体组成和性质的方法，其中，所述创建并训练所述测定模型包括：
    建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库；
    从所述数据库中抽取样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练；
    从训练后的机器学习模型中获取最优机器学习模型作为所述测定模型。
  5. 根据权利要求4所述的测定地层流体组成和性质的方法，其中，所述建立关于多种储层流体的组成和性质，以及所述井下传感器测量信号的数据库包括：
    通过以下一种或多种方式获取符合预设要求的单相储层流体样品：地面采样、井下采样；所述单相储层流体样品包括：油相储层流体样品、气相储层流体样品和水相储层流体样品；
    在标准大气压和室温下，对所述单相储层流体样品进行单级闪蒸，以分离出平衡的闪蒸气体和闪蒸液体；对所述闪蒸气体和所述闪蒸液体分别进行气相色谱分析，获取所述单相储层流体样品的组成和第一性质；在第一预设压力和第一预设温度下，对所述单相储层流体样品进行流体压力体积温度PVT实验，获取所述单相储层流体样品的第二性质；其中所述第一预设压力大于所述标准大气压，所述第一预设温度大于所述室温；
    将所述单相储层流体样品的组成、所述第一性质和所述第二性质的相关数据作为所述地层流体大数据的一部分加入所述数据库。
  6. 根据权利要求5所述的测定地层流体组成和性质的方法，其中：
    所述第一性质包括以下任意一种或多种：气油比、美国石油学会API比重、分子量、硫含量、碳含量、氢含量、Watson K值、SARA含量和石蜡含量；其中，SARA是指饱和烃、芳香烃、胶质和沥青质；
    所述第二性质包括以下任意一种或多种：泡点、露点、恒组成膨胀CCE实验特性、微分释放DL实验特性、恒容衰竭CVD实验特性、分离器实验特性、密度、粘度、电导率、压缩系数、储层流体积系数、石蜡生成条件以及沥青质生成条件。
  7. 根据权利要求5所述的测定地层流体组成和性质的方法，还包括：
    进行以下任一种或多种操作：
    对所述闪蒸气体进行气相色谱分析时测量所述闪蒸气体的体积、温度和压力加入所述数据库；
    对所述闪蒸液体进行气相色谱分析时测量所述闪蒸液体的体积、温度、压力、密度和分子量加入所述数据库；
    在所述第一预设压力和所述第一预设温度下，对所述单相储层流体样品进行以下任意一种或多种测量：连续近红外光谱、核磁共振NMR、声波、荧光以及介电常数测量，并将测量结果加入所述数据库。
  8. 根据权利要求5所述的测定地层流体组成和性质的方法，还包括：在不同的单相储层流体样品中加入不同的钻井泥浆滤液，并对加入相应的钻井泥浆滤液的单相储层流体样品进行所述PVT实验。
  9. 根据权利要求4所述的测定地层流体组成和性质的方法，其中，所述从所述数据库中抽取所述样本数据集，并通过所述样本数据集对预先创建好的机器学习模型进行训练包括：
    对所述样本数据集进行预处理和标准化；所述预处理包括以下任意一种或多种：去噪，去离群点和光滑；
    将经过所述预处理和标准化的样本数据集划分为第一数据集和第二数据集；
    基于所述第一数据集和所述第二数据集，采用预设的机器学习方法对所述机器学习模型进行训练，获取训练出的最优机器学习模型作为所述测定模型；
    其中，所述第一数据集作为所述机器学习模型的输入数据向量，所述第二数据集作为所述机器学习模型的输出数据向量的目标数据向量。
  10. 根据权利要求9所述的测定地层流体组成和性质的方法，其中，所述预设的机器学习方法包括：有监督的机器学习方法；
    所述基于所述第一数据集和所述第二数据集，采用预设的机器学习方法对所述机器学习模型进行训练，获取训练结果中的最优机器学习模型作为所述测定模型包括：
    针对预先定义的函数集中的每个函数，分别执行以下操作：将标准化的所述第一数据集输入未经训练的所述机器学习模型；根据所述第一数据集和所述机器学习模型中当前加载的函数计算所述输出数据向量；
    将计算出的多个所述输出数据向量分别与所述目标数据向量相比较，从多个比较结果中目标数据向量确定出与所述目标数据向量之间的误差最小的第一输出数据向量对应的第一函数和所述第一函数的当前系数；
    将加载有所述第一函数的机器学习模型作为所述最优机器学习模型；其中，所加载的所述第一函数中具有所述当前系数。
  11. 根据权利要求10所述的测定地层流体组成和性质的方法，还包括：在以下一个或多个过程中，将与所述目标数据向量的误差满足预设的误差阈值的输出数据向量补充到所述数据库中：
    对所述机器学习模型进行训练过程中、进行地层流体组成和性质的实施测定过程中。
  12. 根据权利要求1至11中任一所述的测定地层流体组成和性质的方法，其中，所述测定模型是以关于多种储层流体的组成和性质，以及井下传感器测量信号的大数据为样本数据集，对预先创建好的机器学习模型进行训练获得的。
  13. 一种测定地层流体组成和性质的系统，包括：处理器和计算机可读存储介质，所述计算机可读存储介质中存储有指令，所述处理器执行所述指令，进行如权利要求1至12任意一项所述的测定地层流体组成和性质的方法。
PCT/CN2020/074353 2019-04-25 2020-02-05 一种测定地层流体组成和性质的方法和系统 WO2020215850A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/433,519 US20220155275A1 (en) 2019-04-25 2020-02-05 Method and system for measuring composition and property of formation fluid

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910341093.7A CN110056348B (zh) 2019-04-25 2019-04-25 一种测定地层流体组成和性质的方法和系统
CN201910341093.7 2019-04-25

Publications (1)

Publication Number Publication Date
WO2020215850A1 true WO2020215850A1 (zh) 2020-10-29

Family

ID=67320860

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/074353 WO2020215850A1 (zh) 2019-04-25 2020-02-05 一种测定地层流体组成和性质的方法和系统

Country Status (3)

Country Link
US (1) US20220155275A1 (zh)
CN (1) CN110056348B (zh)
WO (1) WO2020215850A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113361771A (zh) * 2021-06-04 2021-09-07 合肥工业大学 确定储层压力的方法及装置

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110056348B (zh) * 2019-04-25 2021-05-11 中国海洋石油集团有限公司 一种测定地层流体组成和性质的方法和系统
CN110598340B (zh) * 2019-09-18 2020-09-01 中国石油大学(北京) 一种注气驱油实验流体的确定方法及装置
CN110905493B (zh) * 2019-11-21 2023-05-12 中国海洋石油集团有限公司 一种测量井下地层流体污染率的方法
US11933159B2 (en) * 2020-03-26 2024-03-19 Aspentech Corporation System and methods for developing and deploying oil well models to predict wax/hydrate buildups for oil well optimization
CN111811988B (zh) * 2020-07-14 2023-05-12 中海石油(中国)有限公司 一种基于流体分析预测圈闭中气油界面的方法及其用途
CN111794744B (zh) * 2020-07-22 2023-08-15 中国海洋石油集团有限公司 一种井下实时监测地层水污染程度的方法
CN112462438B (zh) * 2020-11-16 2021-08-24 中国石油大学(北京) 基于核磁共振测井定量评价地层流体的方法、装置及设备
CN113239621B (zh) * 2021-05-11 2022-07-12 西南石油大学 一种基于弹性网络回归算法的pvt测量方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2245568A1 (en) * 2008-02-20 2010-11-03 McMaster University Expert system for determining patient treatment response
US9460408B2 (en) * 2013-12-23 2016-10-04 International Business Machines Corporation Method and apparatus for generating test scenarios for a set of business rules
CN108533254A (zh) * 2018-03-20 2018-09-14 盘锦中录油气技术服务有限公司 储层油气水层录井参数主控因素权重分析方法
CN108573320A (zh) * 2018-03-08 2018-09-25 中国石油大学(北京) 页岩气藏最终可采储量的计算方法和系统
CN108875122A (zh) * 2018-04-25 2018-11-23 杭州迅美科技有限公司 利用随钻测井数据计算地质参数的人工智能方法和系统
CN109446735A (zh) * 2018-12-18 2019-03-08 中国石油大学(北京) 一种模拟测井数据的生成方法、设备以及系统
CN110056348A (zh) * 2019-04-25 2019-07-26 中国海洋石油集团有限公司 一种测定地层流体组成和性质的方法和系统

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7379819B2 (en) * 2003-12-04 2008-05-27 Schlumberger Technology Corporation Reservoir sample chain-of-custody
US7091719B2 (en) * 2004-04-30 2006-08-15 Schlumberger Technology Corporation Method for determining properties of formation fluids
US7966273B2 (en) * 2007-07-27 2011-06-21 Schlumberger Technology Corporation Predicting formation fluid property through downhole fluid analysis using artificial neural network
CN101498215B (zh) * 2008-02-01 2014-12-10 普拉德研究及开发股份有限公司 增强的井下流体分析
US9249659B2 (en) * 2009-04-15 2016-02-02 Halliburton Energy Services, Inc. Formation fluid property determination
CN102339326B (zh) * 2010-07-16 2014-01-15 中国石油化工股份有限公司 一种分析模拟缝洞型油藏流体流动的方法
CN106570524A (zh) * 2016-10-25 2017-04-19 中国石油天然气股份有限公司 一种储层流体类型的识别方法及装置
CN108596251A (zh) * 2018-04-25 2018-09-28 中国地质大学（北京） 一种基于委员会机器利用测井数据进行储层流体识别方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2245568A1 (en) * 2008-02-20 2010-11-03 McMaster University Expert system for determining patient treatment response
US9460408B2 (en) * 2013-12-23 2016-10-04 International Business Machines Corporation Method and apparatus for generating test scenarios for a set of business rules
CN108573320A (zh) * 2018-03-08 2018-09-25 中国石油大学(北京) 页岩气藏最终可采储量的计算方法和系统
CN108533254A (zh) * 2018-03-20 2018-09-14 盘锦中录油气技术服务有限公司 储层油气水层录井参数主控因素权重分析方法
CN108875122A (zh) * 2018-04-25 2018-11-23 杭州迅美科技有限公司 利用随钻测井数据计算地质参数的人工智能方法和系统
CN109446735A (zh) * 2018-12-18 2019-03-08 中国石油大学(北京) 一种模拟测井数据的生成方法、设备以及系统
CN110056348A (zh) * 2019-04-25 2019-07-26 中国海洋石油集团有限公司 一种测定地层流体组成和性质的方法和系统

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113361771A (zh) * 2021-06-04 2021-09-07 合肥工业大学 确定储层压力的方法及装置

Also Published As

Publication number Publication date
CN110056348A (zh) 2019-07-26
CN110056348B (zh) 2021-05-11
US20220155275A1 (en) 2022-05-19

Similar Documents

Publication Publication Date Title
WO2020215850A1 (zh) 一种测定地层流体组成和性质的方法和系统
Arabloo et al. Toward a predictive model for estimating dew point pressure in gas condensate systems
CN111046341B (zh) 一种基于主成分分析的非常规天然气压裂效果评价及产能预测方法
Ahmadi et al. Robust intelligent tool for estimating dew point pressure in retrograded condensate gas reservoirs: application of particle swarm optimization
Ozdogan et al. Optimization of well placement under time-dependent uncertainty
Al-Bulushi et al. Development of artificial neural network models for predicting water saturation and fluid distribution
Barbosa et al. Time-domain proton nuclear magnetic resonance and chemometrics for identification and classification of Brazilian petroleum
US20170261640A1 (en) System And Methods For Cross-Tool Optical Fluid Model Validation And Real-Time Application
Karimnezhad et al. A new empirical correlation for estimating bubble point oil formation volume factor
Mahdiani et al. Modeling viscosity of crude oil using k-nearest neighbor algorithm
US8099241B2 (en) Method and apparatus for real time oil based mud contamination monitoring
Liu et al. Predictive model for water absorption in sublayers using a machine learning method
Kang et al. Initial model selection for efficient history matching of channel reservoirs using Ensemble Smoother
Gaskari et al. An integrated technique for production data analysis (PDA) with application to mature fields
US20220074303A1 (en) Determining reservoir fluid properties from downhole fluid analysis data using machine learning
Tadjer et al. Bayesian Deep Decline Curve Analysis: A New Approach for Well Oil Production Modeling and Forecasting
Hashemi et al. An inclusive consistency check procedure for quality control methods of the black oil laboratory data
Wang et al. Hybrid application of unsupervised and supervised learning in forecasting absolute open flow potential for shale gas reservoirs
Yang et al. Predicting Reservoir Fluid Properties from Advanced Mud Gas Data
Lashkenari et al. A further study in the prediction of viscosity for Iranian crude oil reservoirs by utilizing a robust radial basis function (RBF) neural network model
Ibrahim Prediction of shale wettability using different machine learning techniques for the application of CO2 sequestration
Kharazi Esfahani et al. Enhanced machine learning—ensemble method for estimation of oil formation volume factor at reservoir conditions
Zuo et al. EOS-based downhole fluid characterization
El Aily et al. Experimental studies on constant mass–volume depletion of gas-condensate systems
Sureshjani et al. A generalized approach for fast estimation of reservoir permeability distribution from analysis of pressure transient data

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20795783

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20795783

Country of ref document: EP

Kind code of ref document: A1