CN111707579A - Oil-gas two-phase flow pattern thermodynamic diagram visualization method and parameter measurement method - Google Patents
Oil-gas two-phase flow pattern thermodynamic diagram visualization method and parameter measurement method Download PDFInfo
- Publication number
- CN111707579A CN111707579A CN202010494978.3A CN202010494978A CN111707579A CN 111707579 A CN111707579 A CN 111707579A CN 202010494978 A CN202010494978 A CN 202010494978A CN 111707579 A CN111707579 A CN 111707579A
- Authority
- CN
- China
- Prior art keywords
- flow
- gas
- oil
- thermodynamic diagram
- capacitance
- 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.)
- Pending
Links
- 238000010586 diagram Methods 0.000 title claims abstract description 78
- 230000005514 two-phase flow Effects 0.000 title claims abstract description 73
- 238000007794 visualization technique Methods 0.000 title claims abstract description 17
- 238000000691 measurement method Methods 0.000 title abstract description 8
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 24
- 238000011144 upstream manufacturing Methods 0.000 claims abstract description 22
- 238000003325 tomography Methods 0.000 claims abstract description 16
- 238000013507 mapping Methods 0.000 claims abstract description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 42
- 238000000034 method Methods 0.000 claims description 35
- 230000004913 activation Effects 0.000 claims description 9
- 238000003062 neural network model Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 abstract description 23
- 238000002474 experimental method Methods 0.000 abstract description 6
- 238000013528 artificial neural network Methods 0.000 description 15
- 230000006870 function Effects 0.000 description 11
- 238000005259 measurement Methods 0.000 description 11
- 210000002569 neuron Anatomy 0.000 description 5
- 238000003909 pattern recognition Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000003068 static effect Effects 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 2
- 239000003208 petroleum Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 239000011541 reaction mixture Substances 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N11/00—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01F—MEASURING VOLUME, VOLUME FLOW, MASS FLOW OR LIQUID LEVEL; METERING BY VOLUME
- G01F1/00—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow
- G01F1/05—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow by using mechanical effects
- G01F1/34—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow by using mechanical effects by measuring pressure or differential pressure
- G01F1/36—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow by using mechanical effects by measuring pressure or differential pressure the pressure or differential pressure being created by the use of flow constriction
- G01F1/40—Details of construction of the flow constriction devices
- G01F1/44—Venturi tubes
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/02—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance
- G01N27/22—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance by investigating capacitance
Landscapes
- Chemical & Material Sciences (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Fluid Mechanics (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- Measuring Volume Flow (AREA)
Abstract
The application provides a visualization method and a parameter measurement method for a flow pattern thermodynamic diagram of oil-gas two-phase flow. Capacitance data under 52 working conditions are obtained through two 8-Electrode Capacitance Tomography (ECT) sensors at the upstream and the downstream of the Venturi tube, and flow charts under 52 working conditions are obtained through an image reconstruction algorithm. Obtaining a flow chart feature vector through a CNN model, and obtaining a flow chart thermodynamic diagram based on the CNN model by using a Grad-CAM model; and obtaining a key identification flow pattern area of CNN model flow pattern identification through thermodynamic diagram hot spot position distribution. An ECT attention inverse mapping algorithm (EARM) is designed, and effective capacitance data based on a hot spot region of the flow thermodynamic diagram are obtained according to the relation between the hot spot distribution rule of the flow thermodynamic diagram and the ECT image reconstruction principle. And predicting the oil-gas two-phase flow parameters according to the effective capacitance data. The control group experiment proves that the effective capacitance value extracted by the EARM algorithm is correct and effective.
Description
Technical Field
The application belongs to the technical field of flow measurement, and particularly relates to a visualization method of an oil-gas two-phase flow thermodynamic diagram and a parameter measurement method.
Background
The oil-gas two-phase flow widely exists in the petroleum industry, in recent years, various parameters of the oil-gas two-phase flow are widely researched, and the correct measurement of the parameters of the oil-gas two-phase flow has important significance for reasonable and safe exploitation of petroleum. The flow pattern is an important parameter of the oil-gas two-phase flow, which not only affects the flow form, heat transfer and mass transfer performance of the two-phase flow, but also has profound influence on the accurate measurement of other parameters of the oil-gas two-phase flow. The current oil-gas two-phase flow measurement method only uses a neural network model for flow pattern identification and flow prediction, does not explore the relation between the neural network and the physical meanings related to an Electric Capacitance Tomography (ECT) image reconstruction algorithm, and lacks the interpretability of the flow pattern identification of the neural network.
Convolutional neural networks and other deep neural networks have highlighted their superior performance in various fields such as image classification, target detection, semantic segmentation, and the like. Improving the interpretability of neural networks has been a widespread concern and is currently a focus of research. Visualization methods are the most widely used methods in data-based interpretability analysis. The visualization method mainly labels important parts in the data through a visualization tool and combines the neural network learning process with the original measurement data, so that the learning process of deep learning is more clearly understood. Thermodynamic diagrams are a common visualization method that allows one to have a more direct understanding of the intrinsic mechanisms of neural networks.
Currently, in the research of oil-gas two-phase flow, the research of interpretability of a neural network in oil-gas two-phase flow application is lacked, and meanwhile, the accuracy of predicting the parameters of the oil-gas two-phase flow through the neural network is low.
Disclosure of Invention
1. Technical problem to be solved
Based on the problem that the neural network lacks of interpretability exploration in oil-gas two-phase flow application in the current oil-gas two-phase flow research, and meanwhile, the accuracy of predicting the oil-gas two-phase flow parameters through the neural network is low. In order to solve the problems, the application provides a visualization method and a parameter measurement method for a flow pattern thermodynamic diagram of oil-gas two-phase flow.
2. Technical scheme
In order to achieve the above object, the present application provides a method for visualizing an oil-gas two-phase flow pattern thermodynamic diagram, the method comprising the steps of:
1) designing an experimental scheme, setting different experimental working conditions, acquiring first capacitance data and second capacitance data of fixed duration under each working condition by using a Venturi tube upstream capacitance tomography sensor and a Venturi tube downstream capacitance tomography sensor, and measuring each phase flow of oil-gas two-phase flow by using an oil phase flow meter and a gas flow meter;
2) normalizing the first capacitance data and the second capacitance data, and then reconstructing an image;
3) establishing a convolution neural network model, and respectively extracting the characteristics of the oil-gas two-phase flow pattern diagrams at the upstream and the downstream of the venturi tube;
4) and (3) obtaining a thermodynamic diagram of the flow pattern recognition of the convolutional neural network through the weighted gradient type activation mapping model, and determining a main recognition area of the flow pattern recognition of the convolutional neural network.
Another embodiment provided by the present application is: and 2) reconstructing an image by adopting a linear projection algorithm.
Another embodiment provided by the present application is: the convolutional neural network model in the step 3) is an initiation-v 3 model, the input of the model is a manifold graph, and a feature vector is extracted from the middle layer of the model and is used as the input of a weighted gradient type activation mapping model.
Another embodiment provided by the present application is: and 3) adopting MSE regression as a loss function in the convolutional neural network model in the step 3).
Another embodiment provided by the present application is: the step 4) scales the thermodynamic diagram to a manifold size.
The application also provides a method for measuring the parameters of the flow pattern thermodynamic diagram of the oil-gas two-phase flow, which comprises the following steps:
a. determining corresponding third capacitance data according to a convolution neural network-flow thermodynamic diagram hot spot region by adopting an EARM algorithm in combination with the reconstruction of a capacitance tomography image;
b. predicting the oil content and the gas content in the oil-gas two-phase flow according to the third capacitance data and the original capacitance data of the capacitance tomography sensor;
c. and comparing the predicted results.
Another embodiment provided by the present application is: and b, using a Lenet-5 model in the convolutional neural network to predict the flow.
Another embodiment provided by the present application is: the fourth capacitance data which do not completely cover the flow pattern hot spot area and the fifth capacitance data which exceed the flow pattern hot spot area are used as a comparison group to predict the oil content and the gas content in the oil-gas two-phase flow; and comparing the prediction result of the control group with the prediction result of the capacitance data of the flow thermodynamic diagram hot spot region of the Venturi tube upstream capacitance tomography sensor.
3. Advantageous effects
Compared with the prior art, the oil-gas two-phase flow pattern thermodynamic diagram visualization method and the parameter measurement method have the advantages that:
the visualization method of the oil-gas two-phase flow thermodynamic diagram comprises the steps of collecting capacitance data of ECT sensors at the upstream and the downstream of a Venturi tube, obtaining flow diagrams at the upstream and the downstream of the Venturi tube through an image reconstruction algorithm, and obtaining the flow thermodynamic diagram through a CNN network visualization method.
The oil-gas two-phase flow pattern thermodynamic diagram visualization method and the parameter measurement method provided by the application explore the relationship between the CNN and the ECT image reconstruction algorithm-related physical meanings, and improve the interpretability of CNN flow pattern identification.
The visualization method of the flow pattern thermodynamic diagram of the oil-gas two-phase flow can obtain the thermodynamic diagrams of the flow patterns of the oil-gas two-phase flow at the upstream and the downstream of the Venturi tube, and can determine the main identification area of CNN network flow pattern identification according to the distribution of the hot spot positions of the flow pattern thermodynamic diagram.
According to the visualization method of the oil-gas two-phase flow thermodynamic diagram, an EARM algorithm is provided for the first time according to the distribution of the hot spot positions of the oil-gas two-phase flow thermodynamic diagram, and the effective capacitance value corresponding to the key flow pattern identification position is obtained by combining the ECT image reconstruction principle.
According to the oil-gas two-phase flow pattern thermodynamic diagram parameter measuring method, the oil content and the gas content of the oil-gas two-phase flow are predicted through the effective capacitance value extracted based on the flow pattern thermodynamic diagram.
Compared with the method for measuring the flow thermodynamic diagram parameters of the oil-gas two-phase flow, which uses the original 28 capacitance values of the ECT sensor to predict the flow, the method for measuring the flow thermodynamic diagram parameters of the oil-gas two-phase flow predicts the flow through the effective capacitance values, removes invalid capacitance data which may influence the flow parameter measurement, and improves the feature effectiveness of the input CNN network.
According to the oil-gas two-phase flow pattern thermodynamic diagram parameter measuring method, a main identification area of CNN network flow pattern identification is obtained according to a flow pattern thermodynamic diagram hot spot, an EARM algorithm is provided, key identification area capacitance data are obtained by combining an ECT image reconstruction principle and the flow pattern thermodynamic diagram hot spot, and oil-gas two-phase flow parameters are predicted based on the key identification area capacitance data.
According to the method for measuring the parameters of the flow thermodynamic diagram of the oil-gas two-phase flow, two groups of comparison experiments are designed, and compared with the capacitance prediction result corresponding to the hot spot area of the flow thermodynamic diagram, the method ensures that the effective capacitance extracted by the EARM algorithm is correct and effective.
Drawings
FIG. 1 is a flow chart of an upstream oil-gas two-phase flow of a venturi tube under different flow conditions according to the present application;
FIG. 2 is a flow chart of a two-phase flow of oil and gas at the downstream of a venturi tube under different flow conditions according to the application;
FIG. 3 is a graph showing the sensitivity profiles of ECT sensors of the present application at different electrode spacings;
FIG. 4 is a graph illustrating the effective capacitance distribution of the ECT sensor upstream of the venturi for low air content conditions in accordance with the present application;
FIG. 5 is a graph illustrating the effective capacitance distribution of the ECT sensor upstream of the venturi for mid-air conditions according to the present application;
FIG. 6 is a graph illustrating the effective capacitance distribution of the ECT sensor upstream of the venturi for large gas content conditions according to the present application
FIG. 7 is a graph illustrating the effective capacitance distribution of the ECT sensor downstream of the venturi for low air content conditions in accordance with the present application;
FIG. 8 is a graph illustrating the effective capacitance distribution of the ECT sensor downstream of the venturi for high air content conditions in accordance with the present application;
FIG. 9 is a graph illustrating the predicted relative error distribution of flow based on capacitance data for an ECT sensor upstream of a venturi for different air content conditions according to the present application;
FIG. 10 is a graph illustrating the relative error distribution of flow prediction based on capacitance data for a downstream ECT sensor of a venturi tube for different air content conditions according to the present application;
FIG. 11 is a schematic diagram of the flow prediction relative error distribution of the control group 1 under different gas contents of the present application;
FIG. 12 is a graph showing the relative error distribution of the flow prediction of the control group 2 under different gas contents according to the present application;
fig. 13 is a schematic view of the principle of the method of the present application.
Detailed Description
Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, and it will be apparent to those skilled in the art from this detailed description that the present application can be practiced. Features from different embodiments may be combined to yield new embodiments, or certain features may be substituted for certain embodiments to yield yet further preferred embodiments, without departing from the principles of the present application.
The flow pattern is one of the most basic characteristics of oil-gas two-phase flow, and has great influence on accurate measurement of other parameters of the two-phase flow. Convolutional Neural Networks (CNN) have been widely used for the study of flow patterns in the past decade, but current flow pattern studies based on CNN networks are more model-based optimizations, and there is no way to explore the relationship between CNN networks and the physical meaning of flow patterns. The internal principle of the neural network flow pattern identification is analyzed by researching the thermodynamic diagram of the oil-gas two-phase flow pattern.
Grad-CAM is a commonly used visualization model for finding the thermodynamic diagram of an image. Firstly, a feature map obtained after the last convolution of the image after feature extraction is obtained. The weights of all the feature maps are different in the full connection layer, and the weight of each feature map is obtained by utilizing back propagation. Finally, features useful for each category are retained using the Relu activation function, where positive numbers represent features useful for that category and negative numbers represent features useful for other categories (i.e., garbage features). And finally, scaling the thermodynamic diagram to the size of the image size so as to be convenient for weighting with the image.
Referring to fig. 1 to 13, the application provides a visualization method of an oil-gas two-phase flow pattern thermodynamic diagram, which comprises the following steps:
1) designing an experimental scheme, setting different experimental working conditions, acquiring first capacitance data and second capacitance data of fixed duration under each working condition by using capacitance tomography sensors of two 8 electrodes at the upstream and the downstream of a Venturi tube, and measuring each phase flow of the oil-gas two-phase flow by using an oil phase flow meter and a gas flow meter;
2) normalizing the first capacitance data and the second capacitance data, and then reconstructing an image;
3) establishing a convolution neural network model, and respectively extracting the characteristics of the oil-gas two-phase flow pattern diagrams at the upstream and the downstream of the venturi tube;
4) and obtaining a thermodynamic diagram of the flow pattern identification of the convolutional neural network through the weighted gradient type activation mapping model, and determining a main identification area of the flow pattern identification of the convolutional neural network.
Different conditions here refer to conditions at different total flows and gas-oil ratios. The working pressure of the experiment is 0.6MPa, and the working temperature is 33 ℃. The experimental data acquisition time under each working condition is 10 min. And acquiring capacitance data under different working conditions, wherein the different working conditions refer to different total flow and oil-gas ratio. The experimental data acquisition time under each working condition is 10 min. The oil content is measured in the range of 1-10m3H, the measurement range of the gas content is 20-150m3/h。
In step 3), normalization is performed according to the following formula:
in the above formula, CnIs a normalized capacitance value, CmFor the actual measured capacitance value, CgIs the static capacitance value, C, when the pipeline is full of gas phaseoIs the static capacitance value when the pipeline is filled with oil phase.
Further, a linear projection algorithm (LBP) is used for image reconstruction in the step 2).
Further, the convolutional neural network model in the step 3) is an initiation-v 3 model, the input of the model is a manifold graph, and a feature vector is extracted from the middle layer of the model and is used as the input of the Grad-CAM model.
Further, in the convolutional neural network model in step 5), an MSE regression is adopted as a loss function, and an objective function expression of the MSE regression is as follows:
in the above equation, Loss represents the error between the true value and the predicted value,
representing predicted oil and gas contents, yiRepresenting the actual oil and gas contents, and n representing the number of data sets.
Further, in the step 4), a flow pattern thermodynamic diagram is obtained by using the Grad-CAM model, and the thermodynamic diagram is scaled to the size of the flow pattern so as to be weighted with the flow pattern.
The application also provides a method for measuring the parameters of the flow pattern thermodynamic diagram of the oil-gas two-phase flow, which comprises the following steps:
a. determining corresponding third capacitance data according to a flow thermodynamic diagram hot spot region of the convolutional neural network by adopting an EARM algorithm in combination with the reconstruction of a capacitance tomography image;
b. predicting the oil content and the gas content in the oil-gas two-phase flow according to the third capacitance data and the original 28 capacitance data of the capacitance tomography sensor;
c. and comparing the predicted results.
Further, in the step b, a Lenet-5 model in the convolutional neural network is used for flow prediction, the input of the model is capacitance data, and the output is the oil content and the gas content of the oil-gas two-phase flow obtained through prediction.
And further, the method comprises the steps of using the capacitance data (comparison 1 group) which does not completely cover the flow pattern hot spot area and the capacitance data (comparison 2 group) which exceeds the flow pattern hot spot area as comparison groups to predict the oil content and the gas content in the oil-gas two-phase flow, and comparing the prediction results with the capacitance data prediction results of the flow pattern thermodynamic diagram hot spot area.
In the experiment, besides the capacitance values measured by the ECT sensor at different moments, the respective flow rates of oil and gas measured by the gas flowmeter and the oil phase flowmeter are recorded, and the specific detection method is as follows.
The method comprises the following steps: the experimental data under different working conditions are measured by designing an experimental scheme, wherein the different working conditions refer to different total flow and oil-gas ratios. In the experimental process, the oil phase flow is fixed, the gas flow is changed, and the oil content and the gas content in the oil-gas two-phase flow in a certain period of specific time under each working condition are respectively recorded.
The measurement range of the oil content in this application is 1-10m3H, the measurement range of the gas content is 20-150m3/h。
Step two: and (4) measuring the flow of the oil-gas two-phase flow by adopting an ECT sensor. And measuring capacitance values of the oil-gas two-phase flow before and after passing through the Venturi tube by using ECT sensors at the upstream and downstream of the Venturi tube respectively. The schematic structure of the ECT sensor is shown in fig. 3, and is composed of 8 electrode plates, and the number of capacitance values per frame is 28. For a sensor consisting of M electrodes, when only one electrode is energized and all other electrodes are held at zero potential, the number of independent capacitances is M (M-1)/2.
The working pressure in the experiment is 0.6MPa, and the working temperature is 33 ℃. The experimental data acquisition time under each working condition is 10 min. In order to ensure the flow stability under each working condition, the data collection of the ECT sensor is not started until the single-phase flow is stable and the oil-gas two-phase flow is mixed in the experimental process. Capacitance data of an Electric Capacitance Tomography (ECT) sensor is acquired through experiments, and before image reconstruction is carried out on the capacitance data, the instability of gas and oil-gas two-phase flow patterns is considered, and the ECT capacitance data of each 100 frames are subjected to average processing, so that stable capacitance data are obtained. Image reconstruction is performed using the averaged processed capacitance data.
Step three: and carrying out normalization processing on the capacitance value under each working condition.
In the above formula, CnIs a normalized capacitance value, CmFor the actual measured capacitance value, CgIs the static capacitance value, C, when the pipeline is full of gas phaseoIs the static capacitance value when the pipeline is filled with oil phase. All the flow charts in fig. 4 and 5 are obtained by the LBP image reconstruction algorithm based on the normalized capacitance value.
Step four: the application uses a linear projection algorithm (LBP) to carry out image reconstruction on a manifold before and after a Venturi tube. ECT image reconstruction is the inverse problem of ECT, i.e. determining the distribution of permittivity within a pipe from the capacitance measurements between pairs of electrodes. The non-linear relationship between the measured capacitance and the dielectric constant can be simplified as follows:
λ＝Sg (2)
where λ is the normalized capacitance vector, S is the normalized sensitive field matrix, and g is the internal distribution matrix of the medium. The normalization method of the capacitance values is shown in step three.
Step five: the method applies an inclusion-V3 model in a CNN algorithm to perform feature extraction on the oil-gas two-phase flow pattern. The input of the CNN model is a flow pattern diagram before and after a Venturi tube, and a feature vector is extracted in the middle layer of the model and used as the input of the Grad-CAM method.
The forward propagation algorithm in a convolutional neural network can be expressed as:
in the above formula, wherein
Is the h ththMth in layer neural networkthOutput value of individual neuron, kn h-1Is the first (h-1)thN-th in layer neural networkthOutput of individual neuron, wmn hIs the first (h-1)thN th of layerthFrom neuron to hthM th of layerthWeight of individual neuron, qm hIs the h ththMth in layer neural networkthDeviation term for individual neurons. The Relu function is used herein as the activation function.
f(x)＝max(x,0) (4)
Step six: the loss function of the convolutional neural network in the present application employs MSE regression. The objective function is shown in formula (5).
In the above equation, Loss represents the error between the true value and the predicted value,
representing predicted oil and gas contents, yiRepresenting the actual oil and gas contents, and n representing the number of data sets.
The method comprises the steps of obtaining a thermodynamic diagram of CNN network flow pattern identification by using a Grad-CAM method, and determining a main identification area of the CNN network flow pattern identification. Firstly, carrying out feature extraction on the oil-gas two-phase flow pattern diagram through a CNN model, and selecting a feature diagram obtained after the last convolution. The weights of all the feature maps are different in the full connection layer, and the weight of each feature map is obtained by utilizing back propagation. The weight solving formula of Grad-CAM is shown in formula (6).
In the formula (6), the reaction mixture is,
representing the weight of the qth th feature map corresponding to the category a, K being the expected number of feature maps, xaIn order to represent the gradient of the class a score,
represents the expected value of the (m, n) position in the qth th feature map. And multiplying each feature map by the weight to obtain a weighted feature map, summing all the feature maps and activating by using a Relu function, wherein the calculation formula is shown as a formula (7).
In equation (7), the Relu activation function leaves only the features useful for the class, where positive numbers represent features useful for the class and negative numbers represent features useful for other classes (i.e., useless features). Without the Relu activation function, the thermodynamic diagram represents a multi-class feature. And finally, scaling the thermodynamic diagram to the size of the image size so as to be convenient for weighting with the image. Fig. 1 and 2 are schematic diagrams of distribution of flow charts of oil-gas two-phase flow upstream and downstream of a venturi tube respectively.
According to the method, an EARM algorithm is provided by analyzing the distribution rule of the flow thermodynamic diagram hot spot region of the oil-gas two-phase flow, and the algorithm is combined with an ECT image reconstruction principle to determine the corresponding key identification region capacitance data according to the main identification region (namely the flow thermodynamic diagram hot spot region) identified by the CNN network flow pattern. FIG. 3 is a graph showing the sensitivity distribution of ECT sensor flow patterns at different plate electrode spacings. For the hot spot of the flow chart thermodynamic diagram upstream of the Venturi tube, under the condition of small air content, the hot spot region of the thermodynamic diagram is mainly distributed on the right side of the flow chart. Under the condition of medium air content, the hot spot area of the thermodynamic diagram is mainly distributed on the upper side of the flow chart. Under the condition of large gas content, the hot spot area of the thermodynamic diagram is mainly distributed on the upper left side of the rheological diagram. With the increase of the gas content, the whole distribution of the hot spot area of the flow thermodynamic diagram is in a counterclockwise rotation trend. FIGS. 4-6 are graphs showing the distributions of effective capacitance values of ECT sensors upstream of the venturi for different air content conditions, respectively.
By analyzing the distribution rule of the hot spot area of the flow thermodynamic diagram of the downstream oil-gas two-phase flow of the venturi tube, the hot spots of the flow thermodynamic diagram are mainly distributed on the left side of the flow pattern under the condition of small gas content, and the hot spots of the flow thermodynamic diagram are stably distributed on the upper left side of the flow pattern under the condition of large gas content. Fig. 6-7 are schematic diagrams illustrating the effective capacitance distribution of the ECT sensor downstream of the venturi tube under low and high air content conditions, respectively.
The method uses the selected effective capacitance values under different gas content conditions to predict the parameters of the oil-gas two-phase flow, and compares the parameter prediction results with the original 28 capacitance values of the ECT sensor. And (3) using a Lenet-5 model in the CNN network to measure parameters, wherein the model inputs capacitance data and outputs predicted oil content and gas content of the oil-gas two-phase flow. Fig. 8 and 9 are schematic diagrams illustrating predicted relative error distributions of flow based on capacitance data for ECT sensors upstream and downstream of the venturi, respectively. Compared with the original 28 capacitance values of the ECT sensor, the relative error of capacitance data prediction obtained based on the EARM algorithm is obviously reduced, and the EARM algorithm has a positive effect on improving the prediction accuracy of the oil-gas two-phase flow.
The present application proposes an EARM algorithm for predicting oil and gas content at the current state. Compared with the prediction result of using the ECT sensor to obtain the original 28 capacitance values, the method for predicting the oil-gas two-phase flow parameters by using the capacitance data of the hot spot area of the flow thermodynamic diagram improves the effectiveness of the CNN network model characteristics, removes invalid capacitance data which may influence the flow parameter measurement, and enables the input CNN network characteristics to have higher effectiveness.
The application provides a CNN-based oil-gas two-phase flow pattern thermodynamic diagram visualization method, and by researching the thermodynamic diagram of the oil-gas two-phase flow pattern, the relation between a convolutional neural network and an ECT image reconstruction principle is explored, the internal principle of neural network flow pattern recognition is analyzed, and a theoretical basis is provided for subsequently improving the accuracy of CNN network flow pattern recognition.
Although the present application has been described above with reference to specific embodiments, those skilled in the art will recognize that many changes may be made in the configuration and details of the present application within the principles and scope of the present application. The scope of protection of the application is determined by the appended claims, and all changes that come within the meaning and range of equivalency of the technical features are intended to be embraced therein.
Claims (8)
1. A visualization method of a flow pattern thermodynamic diagram of oil-gas two-phase flow is characterized by comprising the following steps: the method comprises the following steps:
1) designing an experimental scheme, setting different experimental working conditions, acquiring first capacitance data and second capacitance data of fixed duration under each working condition by using a Venturi tube upstream capacitance tomography sensor and a Venturi tube downstream capacitance tomography sensor, and measuring each phase flow of oil-gas two-phase flow by using an oil phase flow meter and a gas flow meter;
2) normalizing the first capacitance data and the second capacitance data, and then reconstructing an image;
3) establishing a convolution neural network model, and respectively extracting the characteristics of the oil-gas two-phase flow pattern diagrams at the upstream and the downstream of the venturi tube;
4) and obtaining a thermodynamic diagram of the flow pattern identification of the convolutional neural network through the weighted gradient type activation mapping model, and determining a main identification area of the flow pattern identification of the convolutional neural network.
2. The method of claim 1, wherein: and 2) reconstructing an image by adopting a linear projection algorithm.
3. The method of claim 1, wherein: the convolutional neural network model in the step 3) is an initiation-v 3 model, the input of the model is a manifold graph, and a feature vector is extracted from the middle layer of the model and is used as the input of a weighted gradient type activation mapping model.
4. The method of claim 3, wherein: and 3) adopting MSE regression as a loss function in the convolutional neural network model in the step 3).
5. The method of claim 1, wherein: the step 4) scales the thermodynamic diagram to a manifold size.
6. A method for measuring parameters of a flow pattern thermodynamic diagram of oil-gas two-phase flow is characterized by comprising the following steps: the method comprises the following steps:
a. determining corresponding third capacitance data according to a flow thermodynamic diagram hot spot region of the convolutional neural network by adopting an EARM algorithm in combination with the reconstruction of a capacitance tomography image;
b. predicting the oil content and the gas content in the oil-gas two-phase flow according to the third capacitance data and the original capacitance data of the capacitance tomography sensor;
c. and comparing the predicted results.
7. The method of claim 6, wherein: and b, using a Lenet-5 model in the convolutional neural network to predict the flow.
8. The method of claim 6 or 7, wherein: the fourth capacitance data which do not completely cover the flow pattern hot spot area and the fifth capacitance data which exceed the flow pattern hot spot area are used as a comparison group to predict the oil content and the gas content in the oil-gas two-phase flow; and comparing the prediction result of the control group with the prediction result of the capacitance data of the flow thermodynamic diagram hot spot region of the Venturi tube upstream capacitance tomography sensor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010494978.3A CN111707579A (en) | 2020-06-03 | 2020-06-03 | Oil-gas two-phase flow pattern thermodynamic diagram visualization method and parameter measurement method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010494978.3A CN111707579A (en) | 2020-06-03 | 2020-06-03 | Oil-gas two-phase flow pattern thermodynamic diagram visualization method and parameter measurement method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111707579A true CN111707579A (en) | 2020-09-25 |
Family
ID=72539263
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010494978.3A Pending CN111707579A (en) | 2020-06-03 | 2020-06-03 | Oil-gas two-phase flow pattern thermodynamic diagram visualization method and parameter measurement method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111707579A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112651407A (en) * | 2020-12-31 | 2021-04-13 | 中国人民解放军战略支援部队信息工程大学 | CNN visualization method based on discriminative deconvolution |
| CN112712007A (en) * | 2020-12-25 | 2021-04-27 | 天津城建大学 | Algorithm for identifying automobile exhaust remote sensing data based on DCNN-inclusion network |
| CN113124949A (en) * | 2021-04-06 | 2021-07-16 | 深圳市联恒星科技有限公司 | Multiphase flow detection method and system |
| CN113887135A (en) * | 2021-09-30 | 2022-01-04 | 江苏如清智造科技有限公司 | Background prediction method for conductivity and flow pattern, image reconstruction method, device and medium |
| CN114964396A (en) * | 2022-05-18 | 2022-08-30 | 清华大学深圳国际研究生院 | Method and device for measuring liquid phase flow of oil-gas-water three-phase flow based on ERT and EMF |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104089985A (en) * | 2014-07-10 | 2014-10-08 | 天津大学 | Visual multiphase flow test method based on electric and ultrasonic sensing principle |
| US20170212024A1 (en) * | 2016-01-21 | 2017-07-27 | Atout Process Limited | Method and Apparatus for Determining Properties of a Contained Fluid |
| CN109002629A (en) * | 2018-08-01 | 2018-12-14 | 苏州慧德仿真技术有限公司 | A kind of convolutional neural networks and quick visualization method of Simulation of Multiphase Flow emulation |
| US20190025105A1 (en) * | 2017-07-19 | 2019-01-24 | Government Of The United States, As Represented By The Secretary Of The Air Force | Method for estimating physical characteristics of two materials |
| CN110084867A (en) * | 2019-04-24 | 2019-08-02 | 河北科技大学 | Arteriovenous image rebuilding method based on CNN and multi-electrode electromagnetic measurement |
-
2020
- 2020-06-03 CN CN202010494978.3A patent/CN111707579A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104089985A (en) * | 2014-07-10 | 2014-10-08 | 天津大学 | Visual multiphase flow test method based on electric and ultrasonic sensing principle |
| US20170212024A1 (en) * | 2016-01-21 | 2017-07-27 | Atout Process Limited | Method and Apparatus for Determining Properties of a Contained Fluid |
| US20190025105A1 (en) * | 2017-07-19 | 2019-01-24 | Government Of The United States, As Represented By The Secretary Of The Air Force | Method for estimating physical characteristics of two materials |
| CN109002629A (en) * | 2018-08-01 | 2018-12-14 | 苏州慧德仿真技术有限公司 | A kind of convolutional neural networks and quick visualization method of Simulation of Multiphase Flow emulation |
| CN110084867A (en) * | 2019-04-24 | 2019-08-02 | 河北科技大学 | Arteriovenous image rebuilding method based on CNN and multi-electrode electromagnetic measurement |
Non-Patent Citations (8)
| Title |
|---|
| MJIANSUN: "《https://blog.csdn.net/u013066730/article/details/102599871》", 17 October 2019 * |
| REALLOCING: "《https://www.jianshu.com/p/3c4aa73fd598》", 14 February 2019 * |
| ZHUOQUN XU等: "ECT Attention Reverse Mapping algorithm visualization of flow pattern heatmap based on convolutional neural network and its impact on ECT image reconstruction", 《MEASUREMENT SCIENCE AND TECHNOLOGY》 * |
| ZHUOQUN XU等: "Imaging of flow pattern of gas-oil flows with convolutional neural network", 《2019 IEEE INTERNATIONAL CONFERENCE ON IMAGING SYSTEMS AND TECHNIQUES (IST)》 * |
| 宋蕾: "基于神经网络的电容层析成像系统流型识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 彭珍瑞等: "电容层析成像技术的研究现状综述", 《传感器与微系统》 * |
| 颜华等: "基于ECT投影数据的两相流流型辨识", 《仪器仪表学报》 * |
| 黄志尧等: "电容层析成像两相流流型可视化系统", 《仪器仪表学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112712007A (en) * | 2020-12-25 | 2021-04-27 | 天津城建大学 | Algorithm for identifying automobile exhaust remote sensing data based on DCNN-inclusion network |
| CN112651407A (en) * | 2020-12-31 | 2021-04-13 | 中国人民解放军战略支援部队信息工程大学 | CNN visualization method based on discriminative deconvolution |
| CN112651407B (en) * | 2020-12-31 | 2023-10-20 | 中国人民解放军战略支援部队信息工程大学 | CNN visualization method based on discriminative deconvolution |
| CN113124949A (en) * | 2021-04-06 | 2021-07-16 | 深圳市联恒星科技有限公司 | Multiphase flow detection method and system |
| CN113887135A (en) * | 2021-09-30 | 2022-01-04 | 江苏如清智造科技有限公司 | Background prediction method for conductivity and flow pattern, image reconstruction method, device and medium |
| CN114964396A (en) * | 2022-05-18 | 2022-08-30 | 清华大学深圳国际研究生院 | Method and device for measuring liquid phase flow of oil-gas-water three-phase flow based on ERT and EMF |
| CN114964396B (en) * | 2022-05-18 | 2024-09-17 | 清华大学深圳国际研究生院 | Method and device for measuring oil-gas-water three-phase flow liquid-phase flow based on ERT and EMF |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111707579A (en) | Oil-gas two-phase flow pattern thermodynamic diagram visualization method and parameter measurement method | |
| Gamboa et al. | Validation of the rapid detection approach for enhancing the electronic nose systems performance, using different deep learning models and support vector machines | |
| Jin et al. | Prediction model of velocity field around circular cylinder over various Reynolds numbers by fusion convolutional neural networks based on pressure on the cylinder | |
| Liu et al. | Supervised learning method for the physical field reconstruction in a nanofluid heat transfer problem | |
| CN111537032A (en) | Oil-gas two-phase flow parameter measuring device and method | |
| Pan et al. | Oil well production prediction based on CNN-LSTM model with self-attention mechanism | |
| Dang et al. | A novel deep learning framework for industrial multiphase flow characterization | |
| CN101984340B (en) | Method for identifying flow type of soft grain two-phase turbulence based on artificial intelligence | |
| CN107958267B (en) | Oil product property prediction method based on spectral linear representation | |
| Xiong et al. | An odor recognition algorithm of electronic noses based on convolutional spiking neural network for spoiled food identification | |
| CN105447857A (en) | Feature extraction method of pulsed eddy-current infrared thermal image | |
| Huang et al. | Prediction of minimum miscibility pressure (MMP) of the crude oil-CO2 systems within a unified and consistent machine learning framework | |
| Xu et al. | Numerical investigating the effect of Al2O3-water nanofluids on the thermal efficiency of flat plate solar collectors | |
| CN112926767A (en) | Annular fog flow gas phase apparent flow velocity prediction method based on particle swarm BP neural network | |
| CN114626304A (en) | Soft measurement modeling method for online prediction of copper grade in ore pulp | |
| CN103186696B (en) | Towards the auxiliary variable reduction method of high dimensional nonlinear soft-sensing model | |
| OuYang et al. | Soft measurement of oil–water two-phase flow using a multi-task sequence-based CapsNet | |
| Impagnatiello et al. | Systematic assessment of data-driven approaches for wall heat transfer modelling for LES in IC engines using DNS data | |
| CN211824568U (en) | Oil-gas two-phase flow parameter measuring device | |
| Liu et al. | Bioinspired integrated triboelectric electronic tongue | |
| Hu et al. | A self-adapting Landweber algorithm for the inverse problem of electrical capacitance tomography (ECT) | |
| Fu et al. | Progressive prediction algorithm by multi-interval data sampling in multi-task learning for real-time gas identification | |
| Lu et al. | Identification of Gas-Solid Two-Phase Flow Regimes Based on Electrostatic Sensor and CB-ResNext Network | |
| Chunyu et al. | Research on fault diagnosis method of turbocharger rotor based on Hu-SVM-RFE | |
| Wang et al. | E-Nose-Based Cross-Domain Recognition of Combustion-Supporting Agents via M-BFFNet |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200925 |
|
| RJ01 | Rejection of invention patent application after publication |