A virtual metering system for oil and gas two-phase flow

Abstract

A virtual metering system for two-phase oil and gas flow mainly consists of three parts: a flow parameter measurement sensor system, a mobile metering vehicle, and a central processor. The present invention combines a large amount of low-fidelity data from multiphase flow software with a small amount of high-fidelity data measured by the metering vehicle, and establishes a mapping relationship between low-fidelity data and high-fidelity data through a multi-layer neural network, thereby migrating low-fidelity data into high-fidelity data. Real-time and accurate flow can be obtained by installing sensors such as starting pressure, starting temperature, throttling differential pressure, pipeline pressure drop, and pipeline temperature drop on the pipeline. The present invention does not require a physical flow meter, has low metering cost, high precision, and strong generalization, and has broad prospects for promotion and application.

Description

Technical field:

[0001] The invention relates to a multiphase fluid pipeline flow monitoring system, and belongs to the field of oil and gas pipeline flow parameter measurement. Background technology:

[0002] The produced medium at the wellhead is a multiphase flow of gas and liquid. Oilfield production requires the use of multiphase flow metering devices to monitor gas-liquid flow rates. Multiphase flow measurement methods can generally be divided into three categories, depending on whether separation is achieved and the degree of separation: complete separation, non-separation, and split-phase separation.

[0003] The complete separation method uses separation equipment to separate the gas-liquid mixture into single-phase gas and single-phase liquid, which are then measured using a conventional single-phase flowmeter. This converts two-phase flow measurement into single-phase flow measurement. It offers advantages such as reliability, high measurement accuracy, a wide measurement range, and immunity to variations in the gas-liquid two-phase flow pattern. However, the major disadvantage of the complete separation method is the bulky and expensive separation equipment, which requires dedicated metering stations and test pipelines, significantly increasing metering costs.

[0004] The non-separation method places the measuring instrument directly in the two-phase fluid for measurement. This disadvantage is that the sensor operates directly in a two-phase flow environment. Compared to single-phase flow, a notable characteristic of two-phase flow is its strong fluctuation. The distribution of the gas and liquid phases across the pipe cross section, i.e., the flow pattern, changes continuously with the gas-liquid flow rate, exhibiting flow patterns such as stratified flow, wavy flow, annular flow, and slug flow. Therefore, instruments operating directly in two-phase fluid are significantly affected by two-phase flow fluctuations, resulting in low measurement accuracy and a narrow measurement range.

[0005] Chinese Patent 98113068.2 discloses a split-phase two-phase fluid measurement method. Its principle is that the measured two-phase fluid is divided into two parts when it flows through a distributor: one part continues to flow downstream along the original channel, which is called the main fluid, and this loop is the mainstream loop; the other part of the two-phase fluid enters a separator, which is called the branch fluid, and this branch is the branch fluid loop. After the branch fluid is separated by the separator, the gas and liquid are measured using a gas flow meter and a liquid flow meter respectively, and finally reunite with the main fluid. The gas phase flow and liquid phase flow of the measured two-phase fluid are calculated based on their proportional relationship with the gas-liquid phase flow of the branch fluid. How to ensure that the sampled fluid and the measured fluid have a completely consistent phase fraction and a stable proportional relationship is the key to the success of this method. However, the gas-liquid two-phase flow often separates during the sampling process, resulting in an increase in measurement error.

[0006] To overcome the shortcomings of existing technologies, the present invention proposes a multiphase flow virtual metering system. This system combines large amounts of low-fidelity data from multiphase flow software with smaller amounts of high-fidelity data from metering vehicles to establish a multi-fidelity neural network model. This system accurately measures gas-liquid two-phase flow using pressure and temperature sensors placed along the multiphase flow pipeline. Compared to existing multiphase flow metering methods, this system eliminates the need for physical multiphase flow meters and instead utilizes existing flow parameter sensors in the pipeline system to achieve real-time flow monitoring. This system offers advantages such as low metering cost, no resistance loss, a wide measurement range, and high measurement accuracy. Summary of the invention:

[0007] A virtual metering system for two-phase oil and gas flow mainly consists of three parts: a flow parameter measurement sensor system, a mobile metering vehicle, and a central processor. The flow parameter measurement sensors include a starting point temperature sensor, a starting point pressure sensor, a throttling differential pressure sensor, an upstream pressure sensor, a downstream pressure sensor, an upstream temperature sensor, and a downstream temperature sensor. All sensor measurement signals are input to a signal collector via signal lines or remote transmission, and the signal collector is connected to the central processor via wires.

[0008] The starting point temperature sensor and starting point pressure sensor are both arranged close to the wellhead, the two pressure-inducing points of the throttling differential pressure sensor are respectively arranged on the upstream and downstream sides of the throttling device, the upstream pressure sensor and the downstream pressure sensor are respectively arranged upstream and downstream of the pipeline, and the distance between the two is not less than 1 / 2 of the pipe length; the upstream temperature sensor and the downstream temperature sensor are respectively arranged upstream and downstream of the pipeline, and the distance between the two is not less than 1 / 2 of the pipe length.

[0009] The metering vehicle is a temporary metering device located near the pipeline outlet. It primarily consists of a metering separator, a metering inlet pipe, and a metering outlet pipe. One end of the metering inlet pipe is connected to the pipeline, and the other end is connected to the metering separator inlet. The gas-liquid two-phase flow from the pipeline enters the metering separator through the metering inlet pipe, where it undergoes gas-liquid separation. The gas phase flows out through the gas-phase metering line at the top, and the liquid phase flows out through the liquid-phase metering line at the bottom. A gas flowmeter is installed on the gas-phase metering line, and a liquid flowmeter is installed on the liquid-phase metering line. After metering, the gas and liquid phases merge through the metering outlet pipe and return to the pipeline to be measured.

[0010] The central processor has built-in low-fidelity and high-fidelity neural networks. The low-fidelity neural network has a multi-layer structure, with six neurons in its input layer: a throttle differential pressure neuron, a throttle differential pressure fluctuation neuron, a pressure drop neuron, a temperature drop neuron, a starting pressure neuron, and a starting temperature neuron. The low-fidelity neural network has two neurons in its output layer: a low-fidelity liquid mass flow neuron and a low-fidelity gas mass flow neuron. The high-fidelity neural network also has a multi-layer structure, with eight neurons in its input layer: a throttle differential pressure neuron, a throttle differential pressure fluctuation neuron, a pressure drop neuron, a temperature drop neuron, a starting pressure neuron, a starting temperature neuron, as well as a low-fidelity liquid mass flow neuron and a low-fidelity gas mass flow neuron. The high-fidelity neural network has two neurons in its output layer: a high-fidelity liquid mass flow neuron and a high-fidelity gas mass flow neuron.

[0011] The present invention implements the steps as follows:

[0012] (1) Establish a low-fidelity dataset

[0013] Use multiphase flow simulation software to build a pipeline model and obtain parameter data such as gas and liquid flow pipeline pressure and temperature to establish a low-fidelity data set. where {x i} is the input vector, {y i} is the target variable corresponding to the input. i =[M L ,M G ] i , M L Indicates the liquid mass flow rate, M G represents the gas phase mass flow rate, and x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i ,in:

[0014] ΔP V -- represents the differential pressure on both sides of the throttling nozzle, which is predicted by the multiphase flow software;

[0015] ΔP S -- represents the standard deviation of the throttling nozzle differential pressure signal, which is predicted by the multiphase flow software;

[0016] ΔP L -- represents the pipeline pressure drop gradient, which is predicted by multiphase flow software;

[0017] ΔT L -- represents the pipeline temperature drop gradient, which is predicted by multiphase flow software;

[0018] P0--indicates the starting pressure of the pipeline, which is predicted by the multiphase flow software;

[0019] T0--represents the starting temperature of the pipeline, which is predicted by the multiphase flow software;

[0020] (2) Establishing a high-fidelity dataset

[0021] High-fidelity data is obtained by regular measurements from the metering vehicle 14. The metering inlet pipe 16 and metering outlet pipe 17 are connected to the pipeline 1, and then the control valve on the pipeline 1 is opened to allow all the multiphase fluid in the pipeline 1 to flow into the metering separator 15. After the flow stabilizes, the metering operation begins. The high-reliability gas phase flow rate M of the pipeline is measured by the gas flow meter 20 and liquid flow meter 21 on the metering separator 15. G-H , and liquid mass flow rate M L-H At the same time, the starting point temperature sensor 5 on the pipeline is used to measure the starting point temperature T0 of the pipeline, the starting point pressure sensor 6 is used to measure the starting point pressure P0 of the pipeline, and the throttling differential pressure sensor 7 is used to measure the throttling nozzle differential pressure ΔP V , the upstream pressure sensor 8 measures the upstream pressure P1 of the pipeline, the downstream pressure sensor 9 measures the downstream pressure P2 of the pipeline, the upstream temperature sensor 10 measures the upstream temperature T1 of the pipeline, and the downstream temperature sensor 11 measures the downstream pressure T2 of the pipeline, thereby obtaining a high-fidelity data set where y i =[M L ,M G ] i , represents the liquid phase mass flow rate and gas phase mass flow rate measured by the metering separator 15, x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i , which means that when the metering separator 15 is working stably, the sensor measures the throttling nozzle differential pressure ΔP at this time. V , the standard deviation of the differential pressure signal ΔP S , pressure drop per unit pipe length ΔP L , temperature drop per unit pipe length ΔT L , pipeline starting point pressure P0, pipeline starting point temperature T0, where:

[0022] ΔP V --Indicates the differential pressure on both sides of the throttle nozzle, measured by the throttle differential pressure sensor 7;

[0023] ΔP S --Indicates the standard deviation of the throttling nozzle differential pressure signal;

[0024] ΔP L--Indicates the pipeline pressure drop gradient, which is the ratio of the pressure difference measured by the upstream and downstream pressure sensors to the distance between the two pressure sensors;

[0025] ΔT L --Indicates the pipeline temperature drop gradient, which is the ratio of the temperature difference measured by the upstream and downstream temperature sensors to the distance between the two temperature sensors;

[0026] P0--indicates the starting pressure of the pipeline, measured by the starting pressure sensor 6;

[0027] T0--indicates the starting temperature of the pipeline, measured by the starting temperature sensor 5.

[0028] (3) Building and training low-fidelity neural networks

[0029] The low-fidelity neural network 22 uses a multi-layer neural network structure, and the training set of the network is a low-confidence training set. The network parameters are iteratively updated using standard neural network training methods until the network converges.

[0030] (4) High-fidelity neural network construction and training

[0031] High-confidence data Input the low-fidelity neural network 22 trained in step (3) and output the predicted value The predicted value is concatenated with the high-confidence input data to form a new input feature vector, The number of neurons in the input layer of the high-confidence neural network is 8, corresponding to 8 eigenvectors respectively, and the number of neurons in the output layer is 2, corresponding to high-fidelity liquid flow rate 32 and gas flow rate 33 respectively. As input vectors, the liquid and gas flow rates of the high-confidence data set collected by the measurement calibration module Use it as a label to train until the network converges.

[0032] (5) Deployment of multiphase flow virtual metering system

[0033] The sensor module is used to collect the differential pressure ΔP of the throttling nozzle on the pipeline in real time V , pipeline pressure drop ΔP, pipeline temperature drop ΔT, pipeline starting pressure P0, pipeline starting temperature T0, input the collected signal into the signal collector, and establish the input vector x=[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0], input the vector into the low-confidence neural network 22 trained in step (3) to obtain the low-confidence prediction value [M L-L ,M G-L], and then concatenate it with x to form a new input vector x=[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0,M L-L ,M G-L ], the vector is input into the high-confidence neural network 23 trained in step (4), and finally the high-confidence liquid phase flow rate 32 and the high-confidence gas phase flow rate 33 are predicted [M L-H ,M G-H ]. Description of the drawings:

[0034] Figure 1 Schematic diagram of the multiphase flow monitoring system;

[0035] Figure 2 Schematic diagram of the traffic prediction module;

[0036] Figure 3 On-site deployment diagram of the present invention.

[0037] In the figure: 1- pipeline; 2- wellhead; 3- joint station; 4- throttling nozzle; 5- starting point temperature sensor; 6- starting point pressure sensor; 7- throttling differential pressure sensor; 8- upstream pressure sensor; 9- downstream pressure sensor; 10- upstream temperature sensor; 11- downstream temperature sensor; 12- signal collector; 13- central processor; 14- metering vehicle; 15- metering separator; 16- metering inlet pipe; 17- metering outlet pipe; 18- gas phase metering pipeline; 19- liquid phase metering pipe Line; 20-gas flowmeter; 21-liquid flowmeter; 22-low-fidelity neural network; 23-high-fidelity neural network; 24-throttle differential pressure neuron; 25-throttle differential pressure fluctuation neuron; 26-pressure gradient neuron; 27-temperature gradient neuron; 28-starting pressure neuron; 29-starting temperature neuron; 30-low-fidelity liquid volume neuron; 31-low-fidelity gas volume neuron; 32-high-fidelity liquid phase flow neuron; 33-high-fidelity gas phase flow neuron. Specific implementation:

[0038] In oil and gas production, the starting point of the oil and gas mixed pipeline 1 that transports gas-liquid two-phase flow is usually the wellhead 2, and the end point is usually the joint station 3. A throttling device 4 is often provided on the pipeline 1 near the wellhead to control the flow and adjust the pressure of the gathering pipeline.

[0039] like Figure 1As shown, the present invention consists of three parts: a flow parameter measurement sensor system, a mobile metering vehicle 14 and a central processor 13. The flow parameter measurement sensor includes a starting temperature sensor 5, a starting pressure sensor 6, a throttling differential pressure sensor 7, an upstream pressure sensor 8, a downstream pressure sensor 9, an upstream temperature sensor 10, and a downstream temperature sensor 11. All sensor measurement signals are input to a signal collector 12 by a signal line or through remote transmission. The signal collector 12 is connected to the central processor 13 through a wire; the starting temperature sensor 5 and the starting pressure sensor 6 are both arranged close to the wellhead 2, the upstream pressure-inducing point and the downstream pressure-inducing point of the throttling differential pressure sensor 7 are respectively on the upstream and downstream sides of the throttling device, the upstream pressure sensor 8 and the downstream pressure sensor 9 are respectively arranged upstream and downstream of the pipeline 1, and the distance between the two is not less than 1 / 2 of the pipe length, the upstream temperature sensor 10 and the downstream temperature sensor 11 are respectively arranged upstream and downstream of the pipeline 1, and the distance between the two is not less than 1 / 2 of the pipe length.

[0040] Mobile metering vehicle 14 is positioned near the outlet of pipeline 1 and primarily comprises a metering separator 15, a metering inlet pipe 16, and a metering outlet pipe 17. One end of metering inlet pipe 16 is connected to pipeline 1, and the other end is connected to the inlet of metering separator 15. The gas-liquid two-phase flow from pipeline 1 enters metering separator 15 through metering inlet pipe 16, where it undergoes gas-liquid separation. The gas phase flows out through gas-phase metering line 18 at the top, and the liquid phase flows out through liquid-phase metering line 19 at the bottom. A gas flowmeter 20 is installed on gas-phase metering line 18, and a liquid flowmeter 21 is installed on liquid-phase metering line 19. After metering, the gas and liquid phases return to pipeline 1 through metering outlet pipe 17.

[0041] The central processor 13 has a built-in low-fidelity neural network 22 and a high-fidelity neural network 23. The low-fidelity neural network 22 is a multi-layer network structure, and the input layer has 6 neurons, namely, throttling differential pressure neuron 24, throttling differential pressure fluctuation neuron 25, pressure drop neuron 26, temperature drop neuron 27, starting pressure neuron 28, and starting temperature neuron 29. The output layer of the low-fidelity neural network 22 has 2 neurons, namely, low-fidelity liquid mass flow neuron 30 and low-fidelity gas mass flow neuron 31; the high-fidelity neural network 23 is also a multi-layer network structure, and the input layer has 8 neurons, namely, throttling differential pressure neuron 24, throttling differential pressure fluctuation neuron 25, pressure drop neuron 26, temperature drop neuron 27, starting pressure neuron 28, starting temperature neuron 29, as well as low-fidelity liquid mass flow neuron 30 and low-fidelity gas mass flow neuron 31. The output layer of the high-fidelity neural network 23 has two neurons, namely the high-fidelity liquid mass flow neuron 32 and the high-fidelity gas mass flow neuron 33.

[0042] The present invention implements the steps as follows:

[0043] (1) Establish a low-fidelity dataset

[0044] Low-fidelity flow data samples come from conventional multiphase flow simulation software such as OLGA and LedaFlow. Multiphase flow software is used to build a pipeline model to obtain parameter data such as gas and liquid flow, pipeline pressure, and temperature under various working conditions, thereby establishing a low-fidelity data set. N l Represents the amount of low-fidelity data. The output feature vector y i =[M L ,M G ] i , M L Indicates the liquid mass flow rate, M G Indicates the gas phase mass flow rate.

[0045] The input eigenvector has 6 eigenvalues, namely x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i ,in:

[0046] ΔP V --Indicates the differential pressure on both sides of the throttle nozzle, measured by the throttle differential pressure sensor 7;

[0047] ΔP S -- represents the standard deviation of the throttling nozzle differential pressure signal, calculated by the following formula:

[0048]

[0049] In the above formula, n represents the number of throttling nozzle differential pressure test points, i represents the order of the test points, Indicates the average value of the throttling nozzle differential pressure.

[0050] ΔP L --Indicates the pressure drop per unit length of the pipeline, calculated using the following formula:

[0051]

[0052] In the above formula, P1 is the measurement value of the upstream pressure sensor 8, P2 is the measurement value of the downstream pressure sensor 9, and L P is the distance between the two sensors.

[0053] ΔT L --Indicates the pressure drop per unit length of the pipeline, calculated using the following formula:

[0054]

[0055] In the above formula, T1 is the measurement value of the upstream temperature sensor 10, T2 is the measurement value of the downstream temperature sensor 11, and L T is the distance between the two sensors.

[0056] P0--is the starting pressure of the pipeline, measured by the starting pressure sensor 6;

[0057] T0--is the starting temperature of the pipeline, measured by the starting temperature sensor 5.

[0058] (2) Establishing a high-fidelity dataset

[0059] The high-fidelity flow data set comes from the metering vehicle 14, which regularly separates, measures, and calibrates the multiphase fluid flow in the pipeline 1. When the metering vehicle 14 is used to measure the fluid in the pipeline, the metering inlet pipe 16 and the metering outlet pipe 17 are first connected to the pipeline 1, and then the control valve on the pipeline 1 is closed, so that all the multiphase fluid in the pipeline 1 flows into the metering separator 15, and the metering operation is started after the flow stabilizes. After the gas and liquid flows are measured, they return to the main pipeline. The single collection time of the metering vehicle 14 is 1-2 hours. When the metering vehicle 14 is working, the high-reliability gas phase flow M of the pipeline is measured by the gas flow meter 20 and the liquid flow meter 21 on the metering separator 15. G-H , and liquid mass flow rate M L-H At the same time, the starting point temperature sensor 5 on the pipeline is used to measure the starting point temperature T0 of the pipeline, the starting point pressure sensor 6 is used to measure the starting point pressure P0 of the pipeline, and the throttling differential pressure sensor 7 is used to measure the throttling nozzle differential pressure ΔP V The upstream pressure sensor 8 measures the upstream pressure P1 of the pipeline, the downstream pressure sensor 9 measures the downstream pressure P2 of the pipeline, the upstream temperature sensor 10 measures the upstream temperature T1 of the pipeline, and the downstream temperature sensor 11 measures the downstream pressure T2 of the pipeline.

[0060] Following the same data set preprocessing method as step (1), the collected signals are processed to obtain a high-fidelity data set. where N h is the number of high-fidelity data. i =[M L ,M G ] i , represents the liquid phase mass flow rate or gas phase mass flow rate. x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i It indicates that when the metering separator 15 is working stably, the sensor measures the throttling nozzle differential pressure ΔP at this time. V , the standard deviation of the differential pressure signal ΔP S, pressure drop per unit pipe length ΔP L , temperature drop per unit pipe length ΔT L , pipeline starting point pressure P0, pipeline starting point temperature T0.

[0061] Since the metering vehicle 14 is expensive to use and can only be measured periodically, the high-fidelity data input N h Much smaller than the number of low-fidelity data N l .

[0062] (3) Building and training low-fidelity neural networks

[0063] Construct a low-fidelity neural network 22 to build low-confidence data points x i and the corresponding low confidence label y i The corresponding relationship between them. Figure 2 As shown, the low-fidelity neural network 22 adopts a multi-layer neural network structure, and the training set of the network is a low-confidence training set. Among them, x i is the network input, which has 6 characteristic parameters, namely: y i represents low confidence sample labels, This is the low-confidence liquid and gas flow rate. Standard neural network training methods are used to iteratively update network parameters until the network converges, resulting in a converged low-confidence neural network 22. Once training is complete, the module can generate corresponding low-fidelity gas and liquid flow rate predictions for any low-fidelity data vector, effectively capturing the distribution characteristics of the low-fidelity training set and laying the foundation for subsequent model training.

[0064] (4) High-fidelity neural network construction and training

[0065] like Figure 2 As shown, the parameters of the low-confidence neural network 22 are kept fixed, and the high-confidence data are used As input, the low-fidelity neural network 22 outputs a predicted value The predicted value is concatenated with the high-confidence input data to form a new input feature vector, It can be seen that the new input vector has 8 features.

[0066] The high-confidence neural network 23 is also composed of an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is 8, corresponding to 8 feature vectors respectively, and the number of neurons in the output layer is 2, corresponding to high-fidelity liquid flow 32 and gas flow 33 respectively. Liquid and gas flow rates of high-confidence data sets collected by the measurement and calibration module It is used as a label for training until the network converges and is deployed to the central processor to measure gas and liquid flow.

[0067] (5) Deployment of multiphase flow virtual metering system

[0068] like Figure 3 As shown, the present invention does not require a metering vehicle 14 when implementing multiphase flow measurement. The sensor system installed on the pipeline and the central processor 13 can achieve real-time flow measurement. The sensor module is used to collect the differential pressure ΔP of the throttling device on the pipeline 1 in real time. V , the standard deviation of the differential pressure signal ΔP S , pipeline pressure drop ΔP, pipeline temperature drop ΔT, pipeline starting point pressure P0, pipeline starting point temperature T0, the collected signal is input into the signal collector 12, and the signal collector 12 inputs the signal into the central processor 13 for flow prediction. The central processor 13 first pre-processes the collected flow signal, calculates the differential pressure fluctuation signal using formula (1), calculates the unit length pressure drop using formula (2), and calculates the unit length temperature drop using formula (3), thereby forming the input vector x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i The vector is input into the low-confidence neural network 22 trained in step (3) to obtain the low-confidence prediction value [M L-L ,M G-L ], and then combine it with x i Splicing to form a new input vector x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0,M L-L ,M G-L ], the vector is input into the high-confidence neural network 23 trained in step (4), and finally the high-confidence liquid phase flow rate 32 and the high-confidence gas phase flow rate 33 are predicted [M L-H ,M G-H Since the input vector is the real-time data collected by the pipeline sensor, the real-time and accurate measurement of the liquid phase can be achieved.

[0069] The metering vehicle adopts a separation metering mode to separate the gas-liquid two-phase flow in the pipe into single phase and gas and liquid, and uses a gas flow meter and a liquid flow meter for measurement respectively, which has the advantage of high metering accuracy. However, in actual applications, due to the high application cost of the metering vehicle, it is usually used as a mobile metering device to measure the pipeline flow on a regular basis. Since the data set obtained by the metering vehicle is small, it is impossible to directly apply a small amount of high-fidelity data for flow prediction. On the other hand, the correspondence between parameters such as pipeline pressure drop and pipeline temperature drop and flow rate can be established through multiphase flow software such as OLGA and LedaFlow. The advantage of using software is that the application cost is low and it is easy to accumulate a large amount of data samples. Affected by the accuracy of the software's two-phase flow model, the prediction error is very large. The present invention combines a large amount of low-fidelity data from the multiphase flow software with a small amount of high-fidelity data measured by the metering vehicle, establishes a mapping relationship between low-fidelity data and high-fidelity data through a neural network, and migrates the low-fidelity data to high-fidelity data, thereby expanding the data set and improving the model prediction accuracy.

[0070] In summary, in engineering applications, the present invention only requires sensors such as starting pressure, starting temperature, throttling differential pressure, pipeline pressure drop, and pipeline temperature drop installed on the pipeline, and inputting these into a trained neural network to directly and accurately predict flow rate, thereby achieving real-time flow monitoring. While the present invention only uses sensors such as starting pressure, starting temperature, throttling differential pressure, pipeline pressure drop, and pipeline temperature drop as examples, other flow measurement sensors can actually be used depending on the actual conditions of existing pipelines. This invention eliminates the need for a physical flow meter, significantly reduces metering costs, and has broad prospects for widespread application.

Claims

1. A virtual metering system for oil and gas two-phase flow, characterized by: The system mainly consists of three parts: a flow parameter measurement sensor system, a mobile metering vehicle, and a central processor. The flow parameter measurement sensor includes a starting point temperature sensor (5), a starting point pressure sensor (6), a throttling differential pressure sensor (7), an upstream pressure sensor (8), a downstream pressure sensor (9), an upstream temperature sensor (10), and a downstream temperature sensor (11). All sensor measurement signals are input to a signal collector (12) via a signal line or remote transmission. The signal collector (12) is connected to a central processor (13) via a wire. The starting point temperature sensor (5) and the starting point pressure sensor (6) are both arranged near the wellhead; the two pressure-inducing points of the throttling differential pressure sensor (7) are respectively located on the upstream and downstream sides of the throttling device; the upstream pressure sensor (8) and the downstream pressure sensor (9) are respectively arranged on the upstream and downstream sides of the pipeline (1), and the distance between them is not less than 1 / 2 of the length of the pipeline; the upstream temperature sensor (10) and the downstream temperature sensor (11) are respectively arranged on the upstream and downstream sides of the pipeline (1), and the distance between them is not less than 1 / 2 of the length of the pipeline; The metering vehicle (14) is a temporary metering device, which is arranged near the outlet of the pipeline and mainly consists of a metering separator (15), a metering inlet pipe (16), and a metering outlet pipe (17). One end of the metering inlet pipe (16) is connected to the pipeline (1), and the other end is connected to the inlet of the metering separator (15). After the gas-liquid two-phase flow from the pipeline (1) enters the metering separator (15) through the metering inlet pipe (16), gas-liquid separation is completed in the metering separator (15). The gas phase flows out from the gas phase metering pipeline (18) at the top, and the liquid phase flows out from the liquid phase metering pipeline (19) at the bottom. A gas flowmeter (20) is installed on the gas phase metering pipeline (18), and a liquid phase flowmeter (21) is installed on the liquid phase metering pipeline (19). After the gas and liquid phases are metered, they are merged through the metering outlet pipe (17) and returned to the pipeline to be measured. The central processor (13) is equipped with a low-fidelity neural network (22) and a high-fidelity neural network (23). The low-fidelity neural network (22) is a multi-layer network structure. The input layer has six neurons, namely, a throttle differential pressure neuron (24), a throttle differential pressure fluctuation neuron (25), a pressure drop neuron (26), a temperature drop neuron (27), a starting point pressure neuron (28), and a starting point temperature neuron (29). The output layer of the low-fidelity neural network (22) has two neurons, namely, a low-fidelity liquid mass flow neuron (30) and a low-fidelity gas mass flow neuron (31). ; The high-fidelity neural network (23) is also a multi-layer network structure. The input layer has 8 neurons, namely, throttling differential pressure neuron (24), throttling differential pressure fluctuation neuron (25), pressure drop neuron (26), temperature drop neuron (27), starting pressure neuron (28), starting temperature neuron (29), as well as low-fidelity liquid mass flow neuron (30) and low-fidelity gas mass flow neuron (31); the output layer of the high-fidelity neural network (23) has 2 neurons, namely, high-fidelity liquid mass flow neuron (32) and high-fidelity gas mass flow neuron (33).

2. The oil-gas two-phase flow virtual metering system according to claim 1 is characterized in that The measurement implementation steps are as follows: ① Establish a low-fidelity data set: Use multiphase flow software to establish a pipeline model, obtain pipeline pressure and temperature parameter data for different gas and liquid flow rates, and thus establish a low-fidelity data set where {y i } is the label corresponding to the input, y i =[M L ,M G ] i , M L Indicates the liquid mass flow rate, M G represents the gas phase mass flow rate, {x i } is the input vector, x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i ,in: ΔP V -- represents the differential pressure on both sides of the throttling nozzle, which is predicted by the multiphase flow software; ΔP S -- represents the standard deviation of the throttling nozzle differential pressure signal, which is predicted by the multiphase flow software; ΔP L -- represents the pipeline pressure drop gradient, which is predicted by multiphase flow software; ΔT L -- represents the pipeline temperature drop gradient, which is predicted by multiphase flow software; P0—indicates the starting pressure of the pipeline, which is predicted by the multiphase flow software; T0--represents the starting temperature of the pipeline, which is predicted by the multiphase flow software; ② Establish a high-fidelity data set: High-fidelity data is obtained by regular measurements by the metering vehicle (14); the metering inlet pipe (16) and the metering outlet pipe (17) are connected to the pipeline (1), and then the control valve on the pipeline (1) is closed, so that all the multiphase fluid in the pipeline (1) flows into the metering separator (15). After the flow stabilizes, the metering operation is started, and the high-reliability gas phase flow rate M of the pipeline is measured by the gas flow meter (20) and the liquid flow meter (21) on the metering separator (15). G-H , and liquid mass flow rate M L-H； At the same time, the starting point temperature sensor (5) on the pipeline is used to measure the starting point temperature T0 of the pipeline, the starting point pressure sensor (6) is used to measure the starting point temperature P0 of the pipeline, and the throttling differential pressure sensor (7) is used to measure the differential pressure ΔP of the throttling nozzle. V , the upstream pressure sensor (8) measures the upstream pressure P1 of the pipeline, the downstream pressure sensor (9) measures the downstream pressure P2 of the pipeline, the upstream temperature sensor (10) measures the upstream temperature T1 of the pipeline, and the downstream temperature sensor (11) measures the downstream pressure T2 of the pipeline, thereby obtaining a high-fidelity data set where y i =[M L ,M G ] i , represents the liquid phase mass flow rate and gas phase mass flow rate measured by the metering separator (15), x i =[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0] i , which indicates the throttling nozzle differential pressure ΔP measured by the sensor when the metering separator (15) is working stably V , the standard deviation of the differential pressure signal ΔP S , pressure drop per unit pipe length ΔP L , temperature drop per unit pipe length ΔT L , pipeline starting point pressure P0, pipeline starting point temperature T0, where: ΔP V -- indicates the differential pressure on both sides of the throttle nozzle, measured by the throttle differential pressure sensor (7); ΔP S --Indicates the standard deviation of the throttling nozzle differential pressure signal; ΔP L --Indicates the pipeline pressure drop gradient, which is the ratio of the pressure difference measured by the upstream and downstream pressure sensors to the distance between the two pressure sensors; ΔT L --Indicates the pipeline temperature drop gradient, which is the ratio of the temperature difference measured by the upstream and downstream temperature sensors to the distance between the two temperature sensors; P0—indicates the starting pressure of the pipeline, measured by the starting pressure sensor (6); T0 - represents the starting temperature of the pipeline, measured by the starting temperature sensor (5); ③ Construct and train a low-fidelity neural network (22): The low-fidelity neural network (22) adopts a multi-layer neural network structure, and the training set of the network is a low-confidence training set. Iteratively update network parameters through standard neural network training methods until the network converges; ④ Construction and training of high-fidelity neural networks (23): Input the low-fidelity neural network (22) trained in step ③ and output the predicted value The predicted value is concatenated with the high-confidence input data to form a new input feature vector, The number of neurons in the input layer of the high-confidence neural network (23) is 8, corresponding to 8 eigenvectors respectively, and the number of neurons in the output layer is 2, corresponding to high-fidelity liquid phase flow (32) and gas phase flow (33) respectively. As input vectors, the liquid and gas flow rates of the high-confidence data set collected by the measurement calibration module Use it as a label to train until the network converges; ⑤ Deployment of a multiphase flow virtual metering system: Applying a sensor module to collect the differential pressure ΔP of the throttling nozzle on the pipeline in real time V , pipeline pressure drop ΔP, pipeline temperature drop ΔT, pipeline starting pressure P0, pipeline starting temperature T0, input the collected signal into the signal collector (12), establish the input vector x=[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0], input the vector into the low-confidence neural network (22) trained in step ③, and obtain the low-confidence prediction value [M L-L ,M G-L ], and then concatenate it with x to form a new input vector x=[ΔP V ,ΔP S ,ΔP L ,ΔT L ,P0,T0,M L-L ,M G-L ], input the vector into the high-confidence neural network (23) trained in step ④, and finally predict the high-confidence liquid phase flow rate (32) and high-confidence gas phase flow rate (33) [M L-H ,M G-H ].

Images