CN113670384B - Multi-variable time sequence diagram convolution multiphase flow virtual metering method and system - Google Patents
Multi-variable time sequence diagram convolution multiphase flow virtual metering method and system Download PDFInfo
- Publication number
- CN113670384B CN113670384B CN202110953143.4A CN202110953143A CN113670384B CN 113670384 B CN113670384 B CN 113670384B CN 202110953143 A CN202110953143 A CN 202110953143A CN 113670384 B CN113670384 B CN 113670384B
- Authority
- CN
- China
- Prior art keywords
- data
- flow
- differential pressure
- pressure
- temperature
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
- G01D21/02—Measuring two or more variables by means not covered by a single other subclass
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention discloses a multivariable time sequence chart convolution multiphase flow virtual metering method, which comprises the steps of firstly obtaining n groups of corresponding pressure, differential pressure and temperature, and calculating the association degree between the input and the output of the n groups of corresponding pressure, differential pressure and temperature, and then carrying out data enhancement on the corresponding pressure, differential pressure and temperature according to the association degree, training and testing a virtual metering model to obtain a flow metering model. The corresponding system comprises a data exploratory analysis module, an algorithm module and a predictive analysis output value module. The method has the remarkable effects that the liquid quantity measurement precision of the virtual flowmeter is improved through the graph neural network iterative algorithm when the multiphase fluid is measured, the measurement precision is higher than that of the impedance water-containing meter under the gas-containing working condition, the debugging is convenient, the purchasing and maintenance cost of the multiphase flowmeter of the oil field can be effectively reduced, the continuous real-time metering of the multiphase flow is realized, and the method has higher data reference value for the production dynamic monitoring, the flow assurance and the oil reservoir management of the oil field.
Description
Technical Field
The invention relates to oil and gas engineering, in particular to a virtual metering technology in petroleum engineering.
Background
With the development and application of artificial intelligence technology and big data algorithm and the influence of global oil price reduction and radioactive source safety control, the oil field market at home and abroad has urgent and vigorous demands for low-cost non-put multiphase flowmeters. Current online multiphase flow meter (MPFM) measurements are mostly based on total flow using radioactive, electromagnetic electronic, ultrasonic, etc. techniques in combination with phase fraction measurements. However, these phase fraction measurement techniques have certain limitations, on the one hand, while radioactive techniques have so far remained the most accurate and reliable multiphase flow measurement method, many countries worldwide have strict limitations on radioactive source technology usage; on the other hand, in the technical field, the measurement error of the flowmeter is drastically deteriorated when the gas content in the medium reaches a certain limit, even if a special design is adopted. Thus, a special low cost design can be used to enable the flow meter measurement range to be met.
Disclosure of Invention
In view of this, the present invention provides a mathematical algorithm metering model based on a software system that can learn a nonlinear relationship between input features and predicted value data to determine a metering result.
The technical scheme is as follows:
a multivariable time sequence diagram convolution multiphase flow virtual metering method is characterized by comprising the following steps:
step one, acquiring historical data related to flow, wherein the historical data comprise seven groups of data including n groups of pressure P, differential pressure DP, temperature T, liquid flow L, water flow W, oil flow O and air flow G which correspond to each other;
step two, independently taking out each row of data X from the three rows of data of pressure P, differential pressure DP and temperature T i The data Y of each column are independently taken out from four columns of data of liquid flow L, water flow W, oil flow O and air flow G i X=p, DP, T, y=l, W, O, G, i=1, 2,3,..n, the correlation degree of which is calculated pairwise using the following formula, resulting in a correlation coefficient r X→Y ；
Wherein:
n is the total data amount of each column of data;
an arithmetic mean value for the corresponding column data;
an arithmetic mean value for the corresponding column data;
step three, according to the correlation coefficient r X→Y Size assessment X of (2) i Each column of data pair Y i The importance degree of each row of data in the system is given to the weight coefficients a, b and c of the pressure P, the differential pressure DP and the temperature T respectively according to the importance degree;
multiplying each data in the pressure P, the differential pressure DP and the temperature T by the corresponding weight coefficient to obtain revised pressure a, revised differential pressure b, DP and revised temperature c;
fifthly, training and testing a virtual metering model by taking the revised pressure a, the revised differential pressure b, the revised differential pressure DP and the revised temperature c, and taking the liquid flow L, the water flow W, the oil flow O and the air flow G as the output to obtain a flow metering model: l, W, O, g=f (P, T, DP).
Meanwhile, the invention provides a multivariable time sequence diagram convolution multiphase flow virtual metering system, which is characterized in that: the system comprises a data exploratory analysis module, an algorithm module and a predictive analysis output value module which are sequentially arranged;
the data exploratory analysis module is used for acquiring historical data related to the flow, and the algorithm module is used for calculating a correlation coefficient r X→Y The predictive analysis output value module is used for training and testing the virtual metering model.
Drawings
FIG. 1 is a schematic diagram of the present invention;
fig. 2 is a flow chart of the present invention.
Detailed Description
The invention is further described below with reference to examples and figures.
A multivariable time sequence diagram convolution multiphase flow virtual metering method comprises the following steps:
step one, acquiring historical data related to flow, wherein the historical data comprises n groups of seven data of pressure P, differential pressure DP, temperature T, liquid flow L, water flow W, oil flow O and air flow G which correspond to each other, respectively checking the continuity of each column of data, respectively serializing continuous data of discrete data, finally primarily qualifying the data in a histogram form, and analyzing the inter-class distance and quantity difference indexes of the data;
step two, independently taking out each row of data X from the three rows of data of pressure P, differential pressure DP and temperature T i The data Y of each column are independently taken out from four columns of data of liquid flow L, water flow W, oil flow O and air flow G i X=p, DP, T, y=l, W, O, G, i=1, 2,3,..n, the correlation degree of which is calculated pairwise using the following formula, resulting in a correlation coefficient r X→Y ；
Wherein:
n is the total data amount of each column of data;
an arithmetic mean value for the corresponding column data;
an arithmetic mean value for the corresponding column data;
step three, according to the correlation coefficient r X→Y Size assessment X of (2) i Each column of data pair Y i The importance degree of each row of data in the system is given to the weight coefficients a, b and c of the pressure P, the differential pressure DP and the temperature T respectively according to the importance degree;
multiplying each data in the pressure P, the differential pressure DP and the temperature T by the corresponding weight coefficient to obtain revised pressure a, revised differential pressure b, DP and revised temperature c;
fifthly, training and testing a virtual metering model by taking the revised pressure a, the revised differential pressure b, the revised differential pressure DP and the revised temperature c, and taking the liquid flow L, the water flow W, the oil flow O and the air flow G as the output to obtain a flow metering model: l, W, O, g=f (P, T, DP).
The virtual metering model can select various existing models, and in this case, a graph convolution neural network model is taken as an example: firstly, reading an adjacency matrix according to revision pressure a.p, revision differential pressure b.dp, revision temperature c.t, liquid flow L, water flow W, oil flow O and air flow G, then downwards calculating through a graph convolution neural network to realize data transmission between graph networks, and finally outputting a correlation coefficient through a full connection regression mode, thereby training and obtaining a flow metering model: l, W, O, g=f (P, T, DP).
Example 2:
a multivariable timing diagram convolution multiphase flow virtual metering system based on embodiment 1 comprises a data exploratory analysis module, an algorithm module and a predictive analysis output value module which are sequentially arranged;
the data exploratory analysis module is used for acquiring historical data related to the flow, and the algorithm module is used for calculating a correlation coefficient r X→Y The predictive analysis output value module is used for training and testing the virtual metering model.
Example 3:
the graph roll-up neural network model is divided into three modules in total: the method comprises a space association diagram learning module, a time association diagram multi-step transducer module, a loss function and a full-connection module, wherein the core idea is that the optimal prediction point is found by the random walk through nonlinear mapping weighting of an adjacent matrix and a coefficient matrix to construct an effective diagram, and then a prediction value is output in a full-connection mode. The graph learning module is typically configured to learn an adjacency matrix to achieve a time and space varying relationship between adaptively acquired variables in the time series data.
Since the graph rolling network describes the relationship between edges and vertices based on the adjacency matrix, the computational core is the following frequency response matrix.
X＝σ(v·diag[θ N ]v T x)
In the above formula, v represents a point, diag represents a diagonal matrix, X represents an output layer feature map, X represents an input layer feature map, [ θ ] N ]Representing edge feature input, v T Representing the transposed matrix of points.
The mathematical principle of graph learning is that we firstly convert the read data into a matrix of m rows and n columns, and the matrix is calculated as follows by a graph learning module convolution kernel h:
the multi-step Transformer calculates the obtained convolution characteristics through a convolution module
Through Laplace transformation
In the above formulaFor Laplace transformation matrix eigenvalue, y is the predicted value to be output, H output layer normalizesA matrix.
And adopting a neural network design depth regression tree in an artificial intelligent algorithm, and selecting standard residual square sum and minimum residual square sum for the deep learning regression tree model to control the output and growth process of the tree. The weak links exist between random forests, and meanwhile, the method can avoid errors caused by a greedy algorithm in the tree growth process. We first define the loss function as L, where y is the predicted value, using a virtual metrology analysis model while using MSE (mean square error) to evaluate the loss error.
Let the derivative be 0, minimize the mean square error.
The calculation can be obtained through the full connection layer, so that the prediction y is the minimum and the best of the mean square error.Is the cosine loss value.
The time series prediction tasks may be performed in different ways. Most classical is a method based on statistics and autoregressions. More precisely, based on enhancement and integration algorithms, we have to use the rolling cycle to generate a large number of useful manual features. On the other hand, we can use neural network models that provide more freedom in the development process, providing customizable features that model the order.
The cyclic and convolution structures have had great success in time series prediction. An interesting approach in this field is by employing the transducers and Attention architecture, which are initially local in sequence data.
The use of a structure in which we have a network of different nodes, which are interrelated by some kind of links, appears unusual. What we try to do is to use a graphical representation of the time series to generate future predictions.
We explore the nested structure of data using a graph convolution neural network. A graphical neural network is employed in unusual situations, such as time series prediction. In our deep learning model, graph dependencies are combined with recursive parts in an attempt to provide more accurate predictions. This approach is well suited to our problem because we can emphasize the basic hierarchy in the data and encode it with the correlation matrix.
The graph learning module firstly reads the cleaned data into a memory through GCN and then uses GNN to realize conversion of input features and output predicted values into sparse graph adjacent matrixes, and can obtain whether the existing relation between parameters is linear or nonlinear and visualizes the parameters through coefficient matrix analysis between the adjacent matrixes of the sparse graphs.
In the graph convolution module, a graph learning module is firstly selected to analyze an input characteristic relation, then the input characteristic is distributed to a structure in a network through an encoder to learn characteristic weights and mapped to a low-dimensional space, the learning process is realized through a characteristic embedded matrix, the learning process ensures the learning efficiency of nodes by optimizing parameters of the encoder and realizes the graph convolution module, and the module mainly analyzes the characteristic of a characteristic graph layer through graph convolution to analyze the weights of the free edges.
Example 4:
the invention uses a time sequence diagram convolution neural network algorithm, through which we design a system method for solving the problems as follows. The system can predict liquid amount, oil amount and gas amount for better treatment Pressure, differential Pressure Dp, temperature Temp or other high-energy, low-energy or even multiple input independent variable characteristics and analyze the relation between the characteristics and the nonlinear mapping between the independent variable and the dependent variable.
(1) The first step: the real liquid content, the oil content, the water content, the gas content, the pressure difference and the temperature are read by using a read __ sql_query and taken as training sets to be sliced, quarter, month and day data needing to be analyzed are taken out, discrete data are actually filled by random numbers and then are serialized, and then the discrete data are read into a data EDA analysis module.
(2) And a second step of: before use, the error is scaled by a six-decimal point method, then the pressure, the differential pressure and the temperature are respectively distributed and visual analyzed to obtain the qualitative problem of trend change and the range of removing noise point data, and then the pearson correlation coefficient between independent variables is analyzed.
(3) And a third step of: the type of data relationship and the strength of correlation between the independent and dependent variables were directly and qualitatively analyzed by thermodynamic diagrams through the analysis of pearson's correlation coefficients between independent pressure, differential pressure, temperature and dependent variables using pyhetmap visualization thermodynamic diagrams.
(4) Fourth step: through the first three steps, perfect serialization data are obtained, the data are divided into a training set, a testing set and a verification set, and training weight files are obtained after the data of the training set and the testing set pass through a graph convolution neural network model.
(5) Fifth step: and after the trained weight file is subjected to the test of the optimization model on the verification set data through the model loading part, the model is deployed through an online deployment system, and then the independent variable characteristics read through each different well output predicted values of liquid content, water content, gas content and oil content through the output value module.
Example 5:
a learning training method of a virtual metering model comprises the following steps: the data of a certain well factory for a period of time is measured by adopting a traditional flowmeter system, and then the data is transmitted to a cloud end through data transmission equipment or private cloud or directly transmitted to an edge computing platform.
And a second step of: the data obtained in the first step is subjected to data preprocessing and washed to be converted into sequence data, and the sequence data is divided into discrete type and continuous type.
And a third step of: and (3) carrying out data exploratory analysis on the data obtained in the second step to obtain the optimal prediction factor and the principal component factor of the change influence.
Fourth step: and performing model training and migration learning on the characteristic factors and the predicted values to obtain weight files of the models, and then deploying the weight files obtained by the models into an online algorithm system.
Fifth step: and (3) analyzing, verifying and optimizing the updated model through the data obtained through the fourth-step predictive calculation.
Through the implementation verification in the oilfield project, the virtual metering error obtained by the method is small and the Mean Square Error (MSE) is smaller than 0.05.
The beneficial effects are that: the technical scheme of the invention improves the liquid quantity measurement precision of the virtual flowmeter when measuring multiphase fluid through a graph neural network iterative algorithm, has higher measurement precision than a resistance water-containing meter under a gas-containing working condition, is convenient to debug, can effectively reduce the purchasing and maintenance cost of the multiphase flowmeter of the oil field, realizes continuous real-time metering of multiphase flow, and has higher data reference value for production dynamic monitoring, flow assurance and oil reservoir management of the oil field.
Finally, it should be noted that the above description is only a preferred embodiment of the present invention, and that many similar changes can be made by those skilled in the art without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (4)
1. A multivariable time sequence diagram convolution multiphase flow virtual metering method is characterized by comprising the following steps:
step one, acquiring historical data related to flow, wherein the historical data comprise seven groups of data including n groups of pressure P, differential pressure DP, temperature T, liquid flow L, water flow W, oil flow O and air flow G which correspond to each other;
step two, respectively pressingThe data X of each column are taken out from the data of three columns of force P, differential pressure DP and temperature T i The data Y of each column are independently taken out from four columns of data of liquid flow L, water flow W, oil flow O and air flow G i X=p, DP, T, y=l, W, O, G, i=1, 2,3,..n, the correlation degree of which is calculated pairwise using the following formula, resulting in a correlation coefficient r X→Y ；
Wherein:
n is the total data amount of each column of data;
an arithmetic mean value for the corresponding column data;
an arithmetic mean value for the corresponding column data;
step three, according to the correlation coefficient r X→Y Size assessment X of (2) i Each column of data pair Y i The importance degree of each row of data in the system is given to the weight coefficients a, b and c of the pressure P, the differential pressure DP and the temperature T respectively according to the importance degree;
multiplying each data in the pressure P, the differential pressure DP and the temperature T by the corresponding weight coefficient to obtain revised pressure a, revised differential pressure b, DP and revised temperature c;
fifthly, training and testing a virtual metering model by taking the revised pressure a, the revised differential pressure b, the revised differential pressure DP and the revised temperature c, and taking the liquid flow L, the water flow W, the oil flow O and the air flow G as the output to obtain a flow metering model: l, W, O, g=f (P, T, DP).
2. The method for virtual metering of a multi-variable timing diagram convolved multiphase flow according to claim 1, wherein: step one, after obtaining historical data related to flow, checking the continuity of each column of data respectively, then carrying out serialization of continuous data of discrete data respectively, finally, primarily qualifying the data in a histogram form, and analyzing the inter-class distance and quantity difference indexes of the data; and then carrying out the second step.
3. The method for virtual metering of a multi-variable timing diagram convolved multiphase flow according to claim 1, wherein: in the fifth step, firstly, reading the adjacency matrix according to the revision pressure a x P, the revision differential pressure b x DP, the revision temperature c x T, the fluid flow L, the water flow W, the oil flow O and the fluid flow G, then calculating downwards through a graph convolution neural network to realize data transmission between the graph networks, and finally outputting a correlation coefficient through a full connection regression mode, thereby training to obtain a fluid flow metering model: l, W, O, g=f (P, T, DP).
4. A system based on the multivariate timing diagram convolution multiphase flow virtual metering method of claim 1, wherein: the system comprises a data exploratory analysis module, an algorithm module and a predictive analysis output value module which are sequentially arranged;
the data exploratory analysis module is used for acquiring historical data related to the flow, and the algorithm module is used for calculating a correlation coefficient r X→Y The predictive analysis output value module is used for training and testing the virtual metering model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110953143.4A CN113670384B (en) | 2021-08-19 | 2021-08-19 | Multi-variable time sequence diagram convolution multiphase flow virtual metering method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110953143.4A CN113670384B (en) | 2021-08-19 | 2021-08-19 | Multi-variable time sequence diagram convolution multiphase flow virtual metering method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113670384A CN113670384A (en) | 2021-11-19 |
| CN113670384B true CN113670384B (en) | 2023-09-08 |
Family
ID=78543952
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110953143.4A Active CN113670384B (en) | 2021-08-19 | 2021-08-19 | Multi-variable time sequence diagram convolution multiphase flow virtual metering method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113670384B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| NO990111D0 (en) * | 1999-01-11 | 1999-01-11 | Audun Aspelund | Multiphase flow measurement in tubes |
| CN101666770A (en) * | 2009-09-18 | 2010-03-10 | 中国石油大学(华东) | Device and method for measuring crude oil with low air void |
| WO2017112839A1 (en) * | 2015-12-24 | 2017-06-29 | General Electric Company | Calibration device and sensitivity determining device for virtual flow meter, and associated methods |
| CN110321646A (en) * | 2019-07-10 | 2019-10-11 | 海默潘多拉数据科技(深圳)有限公司 | A kind of multiphase flow rates virtual metrology method promoting regression tree model based on gradient |
| CN113188613A (en) * | 2021-03-05 | 2021-07-30 | 深圳市联恒星科技有限公司 | Multiphase flow measurement method and system based on uncertainty analysis |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11280141B2 (en) * | 2019-07-23 | 2022-03-22 | Cameron International Corporation | Virtual multiphase flowmeter system |
-
2021
- 2021-08-19 CN CN202110953143.4A patent/CN113670384B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| NO990111D0 (en) * | 1999-01-11 | 1999-01-11 | Audun Aspelund | Multiphase flow measurement in tubes |
| CN101666770A (en) * | 2009-09-18 | 2010-03-10 | 中国石油大学(华东) | Device and method for measuring crude oil with low air void |
| WO2017112839A1 (en) * | 2015-12-24 | 2017-06-29 | General Electric Company | Calibration device and sensitivity determining device for virtual flow meter, and associated methods |
| CN110321646A (en) * | 2019-07-10 | 2019-10-11 | 海默潘多拉数据科技(深圳)有限公司 | A kind of multiphase flow rates virtual metrology method promoting regression tree model based on gradient |
| CN113188613A (en) * | 2021-03-05 | 2021-07-30 | 深圳市联恒星科技有限公司 | Multiphase flow measurement method and system based on uncertainty analysis |
Non-Patent Citations (1)
| Title |
|---|
| 基于数据融合技术的水下油气井计量方法研究;马文礼 等;《石油化工设备》;第48卷(第4期);9-14 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113670384A (en) | 2021-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Roshni et al. | Development and evaluation of hybrid artificial neural network architectures for modeling spatio-temporal groundwater fluctuations in a complex aquifer system | |
| Shiri et al. | Estimation of daily suspended sediment load by using wavelet conjunction models | |
| CN108647839A (en) | Voltage-stablizer water level prediction method based on cost-sensitive LSTM Recognition with Recurrent Neural Network | |
| Fu et al. | Physics-data combined machine learning for parametric reduced-order modelling of nonlinear dynamical systems in small-data regimes | |
| CN111784061B (en) | Training method, device and equipment for power grid engineering cost prediction model | |
| CN114662414A (en) | Oil reservoir production prediction method based on graph wavelet neural network model | |
| CN109376926A (en) | Error prediction method, device, equipment and storage medium for electric energy metering appliance | |
| CN116796799A (en) | Method for creating small-river basin flood rainfall threshold model in area without hydrologic data | |
| CN110852415B (en) | Vegetation index prediction method, system and equipment based on neural network algorithm | |
| Kapangaziwiri | Regional application of the Pitman monthly rainfall-runoff model in southern Africa incorporating uncertainty | |
| CN117709488B (en) | Dam seepage prediction method based on RUN-XGBoost | |
| Nasseri et al. | Uncertainty assessment of monthly water balance models based on incremental modified fuzzy extension principle method | |
| CN113670384B (en) | Multi-variable time sequence diagram convolution multiphase flow virtual metering method and system | |
| KR102196820B1 (en) | Apparatus and method for release modeling of toxic gases | |
| CN116502539B (en) | VOCs gas concentration prediction method and system | |
| Tangrand | Some new contributions to neural networks and wavelets with applications | |
| CN111859799A (en) | Method and device for evaluating data accuracy based on complex electromechanical system coupling relation model | |
| CN111582634A (en) | Multi-factor safety grading method and system for underground large-space construction | |
| Ferreira et al. | Efficient selection of reservoir model outputs within an emulation-based bayesian history-matching uncertainty analysis | |
| Keshavarzi et al. | Developing pedotransfer functions for estimating field capacity and permanent wilting point using fuzzy table look-up scheme | |
| CN111582632B (en) | Multi-factor safety stage prediction method for whole process of underground large space construction | |
| Santhirasekaran et al. | Is Artificial Intelligent the Future for AHM? Decision Making between an Automated Optimisation of ANN Proxy versus Full Numerical Simulation Optimisation Technique | |
| Hatchett et al. | How accurate is a hydraulic model? | |
| 賀斌 et al. | Application of the artificial neural network method to estimate the missing hydrologic data | |
| CN117909927B (en) | Precipitation quantitative estimation method and device based on multisource data fusion model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |