CCUS production well yield metering method based on oil well indicator diagram

Abstract

The invention discloses a CCUS production well yield metering method based on an oil well indicator diagram. The method comprises the steps that S1, CCUS production well data are collected; s2, solving a downhole pump indicator diagram according to the collected data; s3, establishing a downhole pump indicator diagram opening and closing point intelligent pickup model; s4, based on the pump opening and closing point calculation result, the pump effective stroke and the pump filling degree are calculated, the pump leakage amount, the plunger loss and the crude oil volume coefficient of the oil well pump are analyzed, and the CCUS well yield is obtained; s5, the obtained CCUS well yield is compared with the actual oil well yield, the pump indicator diagram conversion condition and the pump opening and closing piece picking condition are analyzed, if the accuracy in the pump indicator diagram conversion process is lower than a first threshold value, the step S2 is executed, and if the pump indicator diagram opening and closing point recognition effect is lower than a second threshold value, the step S3 is executed; and the precision requirement of oil well yield metering is met. According to the invention, the influence of noise, vibration and other external interference on yield metering is reduced, the yield metering precision is improved, and the digitization, intelligence and automation degrees of the metering process are improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial lifting, and in particular, relates to a CCUS production well production measurement method based on an oil well dynamometer diagram. Background Art

[0002] With the growth of energy demand and the increasing pressure of environmental protection, CCUS technology is being used more and more widely in oil production. This technology can not only improve oilfield recovery, but also effectively reduce greenhouse gas emissions, which is of great significance to environmental protection. However, in the management of CCUS production wells, how to accurately measure the production of production wells has become an urgent problem to be solved. At present, most domestic oil fields still use the phase-by-phase metering method of metering stations, and the level of automation is low. One of the prerequisites for oilfield digitization is to realize the automation of single well metering of oil wells. The liquid production metering method based on the dynamometer diagram is a technology that can realize automatic metering of oil well production. This technology determines the effective stroke of the pump by analyzing the position of the opening and closing points in the dynamometer diagram, thereby realizing the analysis of the pump and liquid production, with high accuracy, real-time and reliability. However, on the one hand, the ground dynamometer diagram recorded at the suspension point of the pump is often interfered by factors such as rod and pipe vibration, which affects the accuracy of production evaluation; on the other hand, the current identification of the pump opening and closing points often relies on manual picking by oilfield developers, or the position of the opening and closing points is determined by the five-point curvature method. However, the oil well production determined based on the ground dynamometer diagram and the existing five-point curvature model method is greatly affected by noise, vibration and other external interferences, and the liquid production measurement effect is poor. Summary of the invention

[0003] The embodiment of the present application provides a method for measuring the production of CCUS production wells based on the oil well dynamometer diagram. Combined with the unique scenario of CCUS production wells, the influence of CO2 in the fluid on the dynamometer diagram conversion model is considered to improve the accuracy of the pump dynamometer diagram conversion. The YOLO model in the field of target detection is used to identify the opening and closing point positions of the fixed valve and the floating valve in the pump dynamometer diagram, so as to obtain the fullness and pump efficiency of the oil pump. Compared with the actual oil well production, the pump dynamometer diagram conversion model and the opening and closing point intelligent picking model are optimized, and finally the automatic measurement of the liquid production of CCUS production wells is realized.

[0004] The embodiment of the present application provides a CCUS production well production metering method based on an oil well dynamometer diagram, comprising: S1, collecting CCUS production well data; S2, solving a downhole pump dynamometer diagram based on the collected data; S3, establishing a downhole pump dynamometer diagram opening and closing point intelligent picking model; S4, calculating the effective stroke of the pump and the degree of pump filling based on the pump opening and closing point calculation results, analyzing the pump leakage, plunger loss and crude oil volume coefficient of the oil pump, and obtaining the CCUS well production; S5, comparing the obtained CCUS well production with the actual oil well production, analyzing the pump dynamometer diagram conversion and the pump opening and closing parts picking situation, if the accuracy of the pump dynamometer diagram conversion process is lower than the first threshold, returning to step S2, if the pump dynamometer diagram opening and closing point recognition effect is lower than the second threshold, returning to step S3; until the accuracy requirement of oil well production metering is met.

[0005] Wherein, step S1 comprises:

[0006] S11, collecting load and displacement data at the suspension point of the pumping unit, plotting the load and displacement data as a closed curve, and obtaining a surface indicator diagram of the CCUS production well;

[0007] S12, collecting basic data of CCUS production wells, the basic data including well depth, rod string structure, and tubing string structure data of the oil wells, and collecting production data of the oil wells.

[0008] Wherein, step S2 comprises:

[0009] S21, the pump work diagram is transformed into a wave equation:

[0010] Combining the structural parameters of the oil well pipe string and rod string, and using the Gibbs viscous damping wave equation, the displacement and load at the pumping unit suspension point are converted into the displacement and load of the oil well pump.

[0011] The one-dimensional damped wave equation based on Gibbs is:

[0012]

[0013] Where U(x,t) is the displacement of the rod string section x at time t; a is the propagation speed of the stress wave in the sucker rod string; C is the damping coefficient;

[0014] ① Initial conditions

[0015] The bottom dead point position of the plunger during the upward stroke of the rod pump is taken as the starting point of a stroke. The initial displacement and velocity of the plunger are both 0, so the initial condition of the wave equation can be expressed as:

[0016]

[0017] ② Boundary conditions

[0018] The dynamic load function D(t) and displacement function U(t) at a certain position of the column are the boundary conditions for solving the model, which can be expressed as:

[0019]

[0020] By expanding the load and displacement at the suspension point of the polished rod according to different Fourier series, we can get the equation:

[0021]

[0022] Where ω is the angular velocity; T is the period; σ n , τ n ,υ n ,δ n is the Fourier coefficient;

[0023] ③Connection conditions between poles

[0024] The rod pumping system is often composed of a combination of rods of different levels. For a multi-level rod structure, assuming that the displacement and load of the connecting surface of two adjacent rods are continuous and unchanged, the connection condition between the rods can be expressed as:

[0025]

[0026] S22, pump power diagram wave equation solution:

[0027] In this stage, based on the Fourier transform method, combined with the initial conditions, boundary conditions and connection conditions analyzed in the previous stage, the variables are separated and solved to obtain the following formula:

[0028]

[0029] From Hooke's law we can get:

[0030]

[0031] Then the relationship between the dynamic load and time at the depth x of the column can be expressed as:

[0032]

[0033] Where E is the elastic modulus of the sucker rod string, kg / m 3 ;

[0034] The analytical expressions On(x), Pn(x) are expressed as:

[0035] O n (x)=(K n chβ n x+δ n shβ n x)sinα n x+(μn shβ n x+v n chβ n x)cosα n xP n (x)

[0036] P n (x)=(K n shβ n x+δ n chβ n x)cosα n x-(μ n chβ n x+v n shβ n x)sinα n x

[0037] Calculate the viscosity of the crude oil at different locations in the wellbore:

[0038]

[0039] In the formula, the correlation coefficients a1-a5 are determined by experiments;

[0040] The corresponding damping coefficient of the rod column is as follows:

[0041]

[0042] Where m, B1, and B2 are as follows:

[0043]

[0044] For the Fourier coefficients of the multi-stage column, according to the principle of force continuity, the end point of the previous stage column of the multi-stage column is used as the starting point and connection condition of the next stage column, and the corresponding relationship of the Fourier coefficients of the i-th stage column is as follows:

[0045] σ n (i) = EA r O n '(x) (i-1)

[0046] τ n (i) = EA r P n '(x) (i-1)

[0047] v n (i) = O n (x) (i-1)

[0048] δ n (i) = P n (x)(i-1)

[0049] Finally, after step-by-step calculations, the load and displacement values ​​of the last stage of the sucker rod string are obtained, and the downhole pump work diagram of the corresponding CCUS production well is drawn.

[0050] Wherein, step S3 comprises:

[0051] S31, the pump performance diagram data is a grayscale image of a preset size, the ball valve opening and closing point positions are marked for the converted pump performance diagram, the LabelImg tool is used to make label data, a file in a format suitable for YOLO network training is generated, and a pump performance diagram data set is made;

[0052] S32, randomly dividing the pump performance diagram data set into a training set, a validation set, and a test set according to a preset ratio;

[0053] S33, performing data enhancement on the pump performance diagram data by rotating the pump performance diagram data at a preset angle;

[0054] S34, building a pump opening and closing point picking model based on the YOLO network;

[0055] S35, after the model training is completed, the trained model is evaluated using the validation set, and the precision, recall and F1 score are calculated to ensure that the model can accurately detect the opening and closing points;

[0056] The accuracy calculation formula is:

[0057]

[0058] The recall calculation formula is:

[0059]

[0060] The F1 score calculation formula is:

[0061]

[0062] Wherein, in step S31, the pump performance diagram data is a grayscale image of 416×416 pixels.

[0063] In step S32, the pump performance diagram data set is randomly divided into a training set, a validation set and a test set in a ratio of 6:2:2.

[0064] Wherein, step S4 comprises:

[0065] S41, from step S3, the four ball valve opening and closing point positions A, B, C, and D in the pump power diagram can be obtained, point A corresponds to the closing point of the floating valve, point B corresponds to the opening point of the fixed valve, point C corresponds to the closing point of the fixed valve, and point D corresponds to the opening point of the floating valve;

[0066] S42, calculation of plunger effective stroke:

[0067] The effective stroke of the plunger represents the displacement that actually produces the pumping effect within one stroke, which can be expressed as

[0068] S pd =X d -X a

[0069] Where, X a , X d is the horizontal coordinate of points A and D;

[0070] S43, rod pump output calculation:

[0071] The factors that affect the production of rod pumps include the effective stroke of the pump, the elastic coefficient of the pipe rod, the volume coefficient, and the leakage coefficient.

[0072] Elastic coefficient of pipe rod:

[0073]

[0074] S pu =X c -X b

[0075] The influence of filling coefficient:

[0076]

[0077] S pd =X d -X a

[0078] The influence of volume coefficient:

[0079]

[0080] The influence of leakage coefficient:

[0081]

[0082] The pump efficiency of the rod pump can be expressed as:

[0083] η=η λ β η B ·η L

[0084] The final rod pump output can be expressed as:

[0085] Q t ′=1440SNA p η

[0086] In the formula, Qt ′ is the theoretical production of CCUS production wells; S is the stroke of the rod pump, N is the number of rod pump strokes, A p is the pump area, B1 is the crude oil volume coefficient, S pu S is the distance between the opening and closing points of the fixed valve; pd is the distance between the opening and closing points of the floating valve; X a , X b , X c , X d The horizontal coordinates of the four points are A, B, C, and D respectively;

[0087] First assume that the pump efficiency η is 100%, and then iteratively calculate the output of the rod pump.

[0088] Wherein, step S5 comprises:

[0089] S51, compared with the actual liquid production of the oil well Q t The theoretical liquid production Q calculated in step S4 t ', calculate the output difference, if the output difference is greater than the preset value, go to step S52;

[0090] S52, changing the Fourier series K used in the Fourier transform method in the power diagram conversion model, and then retraining the pump opening and closing point intelligent picking model until the model accuracy meets the accuracy requirement of oil well production measurement.

[0091] Wherein, after step S5, the method further includes: deploying a CCUS production well production intelligent metering model.

[0092] The CCUS production well production measurement method based on the oil well dynamometer diagram in the embodiment of the present application has the following beneficial effects:

[0093] (1) The automated pump opening and closing point identification method is adopted to reduce the time of manual processing and analysis of the performance diagram. Currently, the acquisition of the surface performance diagram of oil wells has been realized in real time. The method of the present invention can be used to quickly analyze the production of oil wells, saving a lot of manpower and equipment costs for the oil field.

[0094] (2) Compared with conventional oil wells, the fluid of CCUS production wells has the characteristic of high CO2 content. The present invention uses the LBC method in the work diagram construction work diagram conversion model, corrects the viscosity calculation model of CO2-containing crude oil, improves the conversion accuracy of the pump work diagram, and can more accurately analyze the effective filling degree of the rod pump.

[0095] (3) Compared with the conventional pump opening and closing point identification method, the present invention adopts the YOLO model in the field of target detection to identify the opening and closing point positions of the fixed valve and the floating valve in the pump performance diagram, which reduces the influence of noise, vibration and other external interference on the production measurement, improves the accuracy of production measurement, and improves the digitization, intelligence and automation of the measurement process.

[0096] (4) By comparing the theoretical liquid production with the actual liquid production, the Fourier series K in the pump power diagram conversion model and the pump opening and closing point picking model can be further optimized, thereby further improving the accuracy and availability of the power diagram oil measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 This is a schematic diagram of the flow chart of a CCUS production well production measurement method based on an oil well indicator diagram according to an embodiment of the present application;

[0098] Figure 2 Another schematic diagram of the flow chart of the CCUS production well production measurement method based on the oil well indicator diagram according to the embodiment of the present application;

[0099] Figure 3 The effect of the power diagram conversion model used in this application;

[0100] Figure 4 YOLO network structure adopted by the present invention;

[0101] Figure 5 This is a schematic diagram of the feature pyramid effect;

[0102] Figure 6 It is a schematic diagram of the pump filling degree;

[0103] Figure 7 This is the transformation effect of the pump work diagram under different Fourier series. DETAILED DESCRIPTION

[0104] Embodiment 1

[0105] like Figure 1 As shown, the CCUS production well production metering method based on the oil well dynamometer diagram of the present application includes: S1, collecting CCUS production well data; S2, solving the downhole pump dynamometer diagram according to the collected data; S3, establishing an intelligent picking model for the opening and closing points of the downhole pump dynamometer diagram; S4, calculating the effective stroke of the pump and the degree of pump filling based on the calculation results of the pump opening and closing points, analyzing the pump leakage, plunger loss and crude oil volume coefficient of the oil pump, and obtaining the CCUS well production; S5, comparing the obtained CCUS well production with the actual oil well production, analyzing the conversion of the pump dynamometer diagram and the picking of the pump opening and closing parts, if the accuracy of the pump dynamometer diagram conversion process is lower than the first threshold, returning to step S2, if the recognition effect of the opening and closing points of the pump dynamometer diagram is lower than the second threshold, returning to step S3; until the accuracy requirement of the oil well production metering is met.

[0106] This application proposes a production intelligent metering method for CCUS production wells. Based on the surface indicator diagram data of CCUS production wells, the downhole pump diagram is calculated to eliminate the influence of rod and pipe vibration in the indicator diagram. Then, the YOLO algorithm (You Only Look Once) is used to realize the intelligent picking of the opening and closing points of the fixed valve and the floating valve in the pump diagram, so as to calculate the effective filling degree in the oil pump, and finally realize the intelligent metering of the production well production.

[0107] Embodiment 2

[0108] like Figure 1-2 As shown, the CCUS production well production measurement method based on the oil well indicator diagram of the present application includes:

[0109] Step 1: CCUS production well data collection: Collect the suspension point load displacement data and oil well rod string structure data of CCUS production wells, and draw the suspension point displacement and load data into a closed dynamometer diagram image.

[0110] This step includes: collecting the load and displacement data at the suspension point of the pumping unit, plotting the load and displacement data as a closed curve, and obtaining the ground indicator diagram of the CCUS production well. Collect the basic data of the CCUS production well, including the well depth, rod column structure, and tubing column structure data of the oil well, and collect the production data of the oil well to test the effect of the intelligent oil measurement model, and calibrate the pump diagram conversion model and the pump diagram opening and closing point intelligent picking model.

[0111] Step 2: Solve the downhole pump diagram. Construct a wave equation to achieve the transformation from the uphole pump diagram to the downhole pump diagram.

[0112] This step includes: (1) Pump work diagram conversion wave equation establishment:

[0113] Combined with the structural parameters of the oil well tubing and rod string, the displacement and load at the suspension point of the pumping unit are converted into the displacement and load of the oil well pump using the Gibbs viscous damping wave equation.

[0114] First, the one-dimensional damped wave equation is constructed based on Gibbs:

[0115]

[0116] Among them, U(x,t) is the displacement of the rod string section x at time t; a is the propagation speed of the stress wave in the sucker rod string; C is the damping coefficient.

[0117] ① Initial conditions

[0118] The bottom dead point position of the plunger during the upward stroke of the rod pump is taken as the starting point of a stroke. The initial displacement and velocity of the plunger are both 0, so the initial condition of the wave equation can be expressed as:

[0119]

[0120] ② Boundary conditions

[0121] The dynamic load function D(t) and displacement function U(t) at a certain position of the column are the boundary conditions for solving the model, which can be expressed as:

[0122]

[0123] By expanding the load and displacement at the suspension point of the polished rod according to different Fourier series, we can get the equation:

[0124]

[0125] Where ω is the angular velocity in rad / s; T is the period in seconds. The Fourier coefficients σn, τn, υn, and δn can be obtained by step-by-step integration of the curves of D(t) and U(t).

[0126] ③Connection conditions between poles

[0127] The rod pumping system is often composed of a combination of rods of different levels. For a multi-level rod structure, assuming that the displacement and load of the connecting surface of two adjacent rods are continuous and unchanged, the connection condition between the rods can be expressed as:

[0128]

[0129] (2) Solution of the pump power diagram wave equation:

[0130] In this stage, based on the Fourier transform method, combined with the initial conditions, boundary conditions and connection conditions analyzed in the previous stage, the variables are separated and solved to obtain the following formula:

[0131]

[0132] From Hooke's law we can get:

[0133]

[0134] Then the relationship between the dynamic load and time at the depth x of the column can be expressed as:

[0135]

[0136] Where E is the elastic modulus of the sucker rod, kg / m 3 .

[0137] The analytical expressions On(x), Pn(x) are expressed as:

[0138] O n (x)=(K n chβn x+δ n shβ n x)sinα n x+(μ n shβ n x+v n chβ n x)cosα n xP n (x)

[0139] P n (x)=(K n shβ n x+δ n chβ n x)cosα n x-(μ n chβ n x+v n shβ n x)sinα n x

[0140] The damping during the pumping process of the sucker rod pump consists of two parts, namely viscous damping and non-viscous damping. Gibbs et al. believe that the non-viscous damping is small and can be ignored. Therefore, the present invention mainly considers the viscous friction between the sucker rod, the coupling and the liquid, and the viscous damping caused by the pressure loss of the fluid in the pump barrel when calculating the damping coefficient of the sucker rod pump. For CCUS production wells, there is a large amount of CO2 in the fluid in the wellbore. Using the conventional black oil model physical property calculation method to calculate the fluid viscosity will have a great impact on the final pump performance diagram calculation results. Therefore, the present invention uses the CO2-crude oil system viscosity calculation method (LBC method) summarized by Lohrenz et al. to calculate the viscosity of crude oil at different positions in the wellbore. This method can associate the viscosity under high pressure and high density with the low density viscosity and the relative density of the system. Due to its simple form and few variables, it is suitable for application under engineering conditions. Its equation form is as follows:

[0141]

[0142] Among them, the correlation coefficients a1-a5 are determined by experiments, and usually a1=0.1023, a2=0.023364, a3=0.058533, a4=-0.40758, and a5=0.0093324.

[0143] Therefore, the damping coefficient of the corresponding column is as follows:

[0144]

[0145] Among them, m, B1, B2 are as follows:

[0146]

[0147] For the Fourier coefficients of multi-stage poles, according to the principle of force continuity, the end point of the previous pole of the multi-stage pole is used as the starting point and connection condition of the next pole. Then the Fourier coefficients of the i-th pole are as follows:

[0148] σ n (i) = EA r O n '(x) (i-1)

[0149] τ n (i) = EA r P n '(x) (i-1)

[0150] v n (i) = O n (x) (i-1)

[0151] δ n (i) = P n (x) (i-1)

[0152] Finally, after step-by-step calculations, the load and displacement values ​​of the last stage of the sucker rod string are obtained, and the downhole pump work diagram of the corresponding CCUS production well is drawn.

[0153] Step 3: Establishment of the intelligent picking model of the opening and closing points of the pump power diagram. Based on the YOLO algorithm, the intelligent picking of the opening and closing points of the fixed valve and the floating valve of the oil well pump is realized.

[0154] This step includes:

[0155] (1) Dataset preparation. Since the pump performance diagram opening and closing point recognition task does not require color features, the pump performance diagram data size is a 416×416 grayscale image. The oil production process experts mark the ball valve opening and closing point positions for the converted pump performance diagrams, and use the LabelImg tool to prepare label data to generate files in the format required for YOLO network training to prepare the pump performance diagram dataset.

[0156] (2) Dataset division: The pump performance diagram dataset is randomly divided into training set, validation set and test set in a ratio of 6:2:2.

[0157] (3) Data enhancement. The training set images are enhanced. Conventional data enhancement methods include flipping, translation, cropping and other operations. However, translation and cropping will cause the loss of key information of the dynamometer diagram, and flipping will change the working conditions of the dynamometer diagram. Therefore, it is not recommended to use conventional data enhancement methods. In the present invention, a small angle rotation method is used to enhance the pump dynamometer diagram data.

[0158] (4) Pump opening and closing point picking model construction. The YOLO method is an efficient target detection model. The present invention realizes intelligent picking of pump opening and closing points based on the YOLO method. The network structure used includes:

[0159] ① Model input layer: The input layer of the pump power diagram open and close point recognition model receives a single-channel pump power diagram grayscale image with a size of 416x416 pixels. The pump power diagram will be normalized to the range of 0 to 1.

[0160] ② Feature extraction network: The pump power diagram opening and closing point feature extraction network consists of 6 convolutional layers and 5 bottleneck structures.

[0161] ③ Feature pyramid layer: The present invention uses a feature pyramid network (FPN) and a path aggregation network (PAN) structure for multi-scale feature fusion.

[0162] ④ Feature detection head: Create detection heads of different sizes and perform target detection on feature maps of different scales.

[0163] (5) Model training and effect evaluation. During the training of the pump opening and closing point intelligent picking model, the model will calculate the predicted value through forward propagation and use the back propagation algorithm to update the model weight so that the predicted value gradually approaches the true value. After the model training is completed, the trained model is evaluated using the validation set, and the precision, recall and F1 score are calculated to ensure that the model can accurately detect the opening and closing points.

[0164] ①The accuracy calculation formula is:

[0165]

[0166] ②The recall rate calculation formula is:

[0167]

[0168] ③The F1 score calculation formula is:

[0169]

[0170] Among them, TP: positive samples predicted by the model as positive, FP: negative samples predicted by the model as positive, FN: positive samples predicted by the model as negative.

[0171] Step 4: Based on the calculation results of the pump opening and closing points, calculate the effective stroke of the pump and the degree of pump filling, analyze the pump leakage, plunger loss and crude oil volume coefficient of the oil well pump, and obtain the CCUS well production.

[0172] This step includes:

[0173] (1) Division of opening and closing points of pump ball valve

[0174] From step 3, the positions of the opening and closing points of the four ball valves in the pump performance diagram can be obtained, which correspond to the closing point of the floating valve (recorded as point A in the pump performance diagram), the opening point of the fixed valve (recorded as point B in the pump performance diagram), the closing point of the fixed valve (recorded as point C in the pump performance diagram), and the opening point of the floating valve (recorded as point D in the pump performance diagram).

[0175] (2) Calculation of effective plunger stroke

[0176] The effective stroke of the plunger represents the displacement that actually produces the pumping effect within one stroke, which can be expressed as:

[0177] S pd =X d -X a

[0178] Among them, X a , X d is the horizontal coordinate of points A and D.

[0179] (3) Calculation of rod pump output

[0180] In addition to the effective stroke of the pump, the factors that affect the production of the rod pump also include the elastic coefficient of the pipe rod, the volume coefficient, and the leakage coefficient.

[0181] ① Elastic coefficient of pipe rod:

[0182]

[0183] S pu =X c -X b

[0184] ②The influence of filling coefficient

[0185]

[0186] S pd =X d -X a

[0187] ③Influence of volume coefficient

[0188]

[0189] ④The influence of leakage coefficient

[0190]

[0191] The pump efficiency of the rod pump can be expressed as:

[0192] η=η λ β η B ·η L

[0193] The final rod pump output can be expressed as:

[0194] Q t ′=1440SNA p η

[0195] Among them, Q t ′ is the theoretical production of CCUS production wells; S is the stroke of the rod pump, N is the number of rod pump strokes, A p is the pump area, B1 is the crude oil volume coefficient, S pu S is the distance between the opening and closing points of the fixed valve; pd is the distance between the opening and closing points of the floating valve; X a , X b , X c , X d The horizontal coordinates of the four points are A, B, C, and D respectively.

[0196] Since the production of the rod pump needs to be determined first in the leakage volume, the pump efficiency η is first assumed to be 100%, and then the production of the rod pump is iteratively calculated.

[0197] Step 5: Compare with the actual oil well production, analyze the pump performance diagram conversion and the pump opening and closing parts picking situation. If the pump performance diagram conversion process has low accuracy, return to step 2. If the pump performance diagram opening and closing point recognition effect is poor, return to step 3.

[0198] This step includes:

[0199] (1) Comparison of actual oil well production Q t The theoretical liquid production Q calculated in the present invention is t ′, calculate the difference in production. If the difference in production is small, it means that the work diagram conversion model and the pump opening and closing point intelligent picking model are effective and the model can be applied on site. If the difference in production is large, proceed to the next step.

[0200] (2) Change the Fourier series K used in the Fourier transform method in the power diagram conversion model, and then retrain the pump opening and closing point intelligent picking model until the model accuracy meets the accuracy requirements of oil well production measurement.

[0201] Step 6: Deployment of intelligent metering model for CCUS production well output.

[0202] Compared with the prior art, the present invention has the following characteristics:

[0203] (1) The automated pump opening and closing point identification method is adopted to reduce the time of manual processing and analysis of the performance diagram. Currently, the acquisition of the surface performance diagram of oil wells has been realized in real time. The method of the present invention can be used to quickly analyze the production of oil wells, saving a lot of manpower and equipment costs for the oil field.

[0204] (2) Compared with conventional oil wells, the fluid of CCUS production wells has the characteristic of high CO2 content. The present invention uses the LBC method in the work diagram construction work diagram conversion model, corrects the viscosity calculation model of CO2-containing crude oil, improves the conversion accuracy of the pump work diagram, and can more accurately analyze the effective filling degree of the rod pump.

[0205] (3) Compared with the conventional pump opening and closing point identification method, the present invention adopts the YOLO model in the field of target detection to identify the opening and closing point positions of the fixed valve and the floating valve in the pump performance diagram, which reduces the influence of noise, vibration and other external interference on the production measurement, improves the accuracy of production measurement, and improves the digitization, intelligence and automation of the measurement process.

[0206] (4) By comparing the theoretical liquid production with the actual liquid production, the Fourier series K in the pump power diagram conversion model and the pump opening and closing point picking model can be further optimized, thereby further improving the accuracy and availability of the power diagram oil measurement.

[0207] Embodiment 3

[0208] The principle of this application is to use the Gibbs method to calculate the downhole pump diagram based on the surface indicator diagram data of the CCUS production well, thereby eliminating the influence of the rod and pipe vibration in the indicator diagram. In this process, the LBS method is used to correct the influence of CO2 on crude oil viscosity and improve the accuracy of the CCUS production well diagram conversion model. Then, the YOLO algorithm (You Only Look Once) is used to realize the intelligent picking of the opening and closing points of the fixed valve and the floating valve in the pump diagram, so as to calculate the effective filling degree in the oil pump, and finally realize the intelligent measurement of the production well output. Figure 2 As shown in the figure, a CCUS production well production intelligent metering method based on oil well dynamometer diagram is provided. It mainly includes the following processes:

[0209] Step 1: CCUS production well data collection and dynamometer image drawing.

[0210] (1) Collect the load and displacement data at the suspension point of the pumping unit and calculate the load P (t1, t2, t3, ... t n ) and displacement data X(t1, t2, t3, ...t n ) is drawn as a closed curve to obtain the surface performance diagram of the CCUS production well.

[0211] (2) Collect basic data of CCUS production wells, including well depth, rod string structure, and tubing string structure data, as well as oil well production data, to verify the effectiveness of the intelligent oil measurement model and calibrate the pump performance diagram conversion model and the pump performance diagram opening and closing point intelligent picking model.

[0212] Step 2: Pump performance diagram conversion model considering CO2 viscosity correction.

[0213] (1) Pump power diagram conversion wave equation establishment:

[0214] Combined with the structural parameters of the oil well tubing and rod string, the displacement and load at the suspension point of the pumping unit are converted into the displacement and load of the oil well pump using the Gibbs viscous damping wave equation.

[0215] First, the one-dimensional damped wave equation is constructed based on Gibbs:

[0216]

[0217] Among them, U(x,t) is the displacement of the rod string section x at time t; a is the propagation speed of the stress wave in the sucker rod string; C is the damping coefficient.

[0218] ① Initial conditions

[0219] The bottom dead point position of the plunger during the upward stroke of the rod pump is taken as the starting point of a stroke. The initial displacement and velocity of the plunger are both 0, so the initial condition of the wave equation can be expressed as:

[0220]

[0221] ② Boundary conditions

[0222] The dynamic load function D(t) and displacement function U(t) at a certain position of the column are the boundary conditions for solving the model, which can be expressed as:

[0223]

[0224] By expanding the load and displacement at the suspension point of the polished rod according to different Fourier series, we can get the equation:

[0225]

[0226] Where ω is the angular velocity in rad / s, and T is the period in seconds. The Fourier coefficients σn, τn, υn, and δn can be obtained by step-by-step integration of the curves of D(t) and U(t).

[0227] ③Connection conditions between poles

[0228] The rod pumping system is often composed of a combination of rods of different levels. For a multi-level rod structure, assuming that the displacement and load of the connecting surface of two adjacent rods are continuous and unchanged, the connection condition between the rods can be expressed as:

[0229]

[0230] (2) Solution of the pump power diagram wave equation:

[0231] In this stage, based on the Fourier transform method, combined with the initial conditions, boundary conditions and connection conditions analyzed in the previous stage, the variables are separated and solved to obtain the following formula:

[0232]

[0233] From Hooke's law we can get:

[0234]

[0235] Then the relationship between the dynamic load and time at the depth x of the column can be expressed as:

[0236]

[0237] Where E is the elastic modulus of the sucker rod, kg / m 3 .

[0238] The analytical expressions On(x), Pn(x) are expressed as:

[0239] O n (x)=(K n chβ n x+δ n shβ n x)sinα n x+(μ n shβ n x+v n chβ n x)cosα n xP n (x)

[0240] P n (x)=(K n shβ n x+δ n chβ n x)cosα n x-(μ n chβ n x+v n shβ n x)sinα n x

[0241] The damping of the sucker rod pump during the pumping process consists of two parts, namely viscous damping and non-viscous damping. Gibbs et al. believe that the non-viscous damping is small and can be ignored. Therefore, the present invention mainly considers the viscous friction between the sucker rod, the coupling and the liquid, and the viscous damping caused by the pressure loss of the fluid in the pump barrel when calculating the damping coefficient of the sucker rod pump.

[0242] For CCUS production wells, there is a large amount of CO2 in the fluid in the wellbore. Using the conventional black oil model physical property calculation method to calculate the fluid viscosity will have a significant impact on the final pump performance diagram calculation results. Therefore, the present invention uses the CO2-crude oil system viscosity calculation method (LBC method) summarized by Lohrenz et al. to calculate the crude oil viscosity at different locations in the wellbore:

[0243]

[0244] Among them, the correlation coefficients a1-a5 are determined by experiments, and usually a1=0.1023, a2=0.023364, a3=0.058533, a4=-0.40758, and a5=0.0093324.

[0245]

[0246] Among them, m, B1, B2 are as follows:

[0247]

[0248] For the Fourier coefficients of multi-stage poles, according to the principle of force continuity, the end point of the previous pole of the multi-stage pole is used as the starting point and connection condition of the next pole. Then the Fourier coefficients of the i-th pole are as follows:

[0249] σ n (i) = EA r O n '(x) (i-1)

[0250] τ n (i) = EA r P n '(x) (i-1)

[0251] v n (i) = O n (x) (i-1)

[0252] δ n (i) = P n (x) (i-1)

[0253] The final work map conversion model effect is as follows Figure 3 shown.

[0254] Step 3: Establish the intelligent picking model of the opening and closing points of the pump power diagram.

[0255] (1) Dataset preparation. Since the pump performance diagram opening and closing point recognition task does not require color features, the pump performance diagram data size is a 416×416 grayscale image. The oil production process experts mark the ball valve opening and closing point positions for the converted pump performance diagrams, and use the LabelImg tool to prepare label data to generate files in the format required for YOLO network training to prepare the pump performance diagram dataset.

[0256] (2) Dataset division: The pump performance diagram dataset is randomly divided into training set, validation set and test set in a ratio of 6:2:2.

[0257] (3) Data enhancement. The training set images are enhanced. Conventional data enhancement methods include flipping, translation, cropping and other operations. However, translation and cropping will cause the loss of key information of the dynamometer diagram, and flipping will change the working conditions of the dynamometer diagram. Therefore, it is not recommended to use conventional data enhancement methods. In the present invention, a small angle rotation (-15 degrees to 15 degrees) is used to enhance the pump dynamometer diagram data.

[0258] (4) Pump opening and closing point picking model construction. The YOLO method is an efficient target detection model. The present invention realizes the intelligent picking of pump opening and closing points based on the YOLO method. The network structure used is as follows: Figure 4 shown, including:

[0259] ① Model input layer: The input layer of the pump power diagram open and close point recognition model receives a single-channel pump power diagram grayscale image with a size of 416x416 pixels. The pump power diagram will be normalized to the range of 0 to 1.

[0260] ② Feature extraction network: The pump power diagram opening and closing point feature extraction network consists of 6 convolutional layers and 5 bottleneck structures. In the present invention, the detailed structure of the feature extraction network is:

[0261] The first convolution layer: 32 3x3 convolution kernels, stride 1, and output feature map size is 416x416.

[0262] The second convolution layer: 64 3x3 convolution kernels, stride 2, and output feature map size is 208x208.

[0263] The first bottleneck structure: contains two 3x3 convolution kernels, and the feature map size is kept at 208x208.

[0264] The third convolutional layer: 128 3x3 convolution kernels, stride 2, and output feature map size is 104x104.

[0265] The second bottleneck structure: contains four 3x3 convolution kernels, and the feature map size is kept at 104x104.

[0266] The 4th convolutional layer: 256 3x3 convolution kernels, stride 2, and output feature map size 52x52.

[0267] The third bottleneck structure: contains eight 3x3 convolution kernels, and the feature map size is kept at 52x52.

[0268] The fifth convolutional layer: 512 3x3 convolution kernels, stride 2, and output feature map size 26x26.

[0269] The fourth bottleneck structure: contains eight 3x3 convolution kernels, and the feature map size is kept at 26x26.

[0270] The sixth convolutional layer: 1024 3x3 convolution kernels, stride 2, and output feature map size 13x13.

[0271] The fifth bottleneck structure: contains four 3x3 convolution kernels, and the feature map size is kept at 13x13.

[0272] ③ Feature pyramid layer: such as Figure 5 As shown, the present invention introduces a feature pyramid network (FPN) and a path aggregation network (PAN) structure for multi-scale feature fusion.

[0273] Upsampling layer: Use the upsampling layer to upsample the low-resolution feature map to a higher resolution. The 26x26 and 13x13 feature maps are fused through upsampling and convolution to generate a 13x13 feature map.

[0274] Feature fusion layer: Concatenate and convolve feature maps of different scales to generate a fused feature map. The 52x52 and 26x26 feature maps are fused through upsampling and convolution to generate a 26x26 feature map.

[0275] ④ Feature detection head: Create detection heads of different sizes, and perform target detection on feature maps of different scales. Three types of feature detection heads are set in the present invention.

[0276] Detection head 1: Perform convolution operation on the 13x13 feature map to generate prediction results for large objects.

[0277] Detection head 2: Perform convolution operation on the 26x26 feature map to generate the prediction result of the target.

[0278] Detection head 3: Perform convolution operation on the 52x52 feature map to generate the prediction result of the target.

[0279] (5) Model training and effect evaluation. During the training of the pump opening and closing point intelligent picking model, the model will calculate the predicted value through forward propagation, and use the back-propagation algorithm to update the model weight so that the predicted value gradually approaches the true value. In this example, the number of training rounds of the model is 500, the step size is 0.0001, the optimizer is the Adam optimizer, and the Batch_Size is 32. After the model training is completed, the trained model is evaluated using the validation set, and the precision, recall, and F1 score are calculated to ensure that the model can accurately detect the opening and closing points.

[0280] ①The accuracy calculation formula is:

[0281]

[0282] ②The recall rate calculation formula is:

[0283]

[0284] ③The F1 score calculation formula is:

[0285]

[0286] Step 4: Calculate the production of CCUS production wells.

[0287] (1) Division of opening and closing points of pump ball valve

[0288] From step 3, the positions of the opening and closing points of the four ball valves in the pump performance diagram can be obtained, which correspond to the closing point of the floating valve (recorded as point A in the pump performance diagram), the opening point of the fixed valve (recorded as point B in the pump performance diagram), the closing point of the fixed valve (recorded as point C in the pump performance diagram), and the opening point of the floating valve (recorded as point D in the pump performance diagram).

[0289] 2) Calculation of effective plunger stroke

[0290] The effective stroke of the plunger represents the displacement that actually produces the pumping effect within one stroke, which can be expressed as:

[0291] S pd =X d -X a

[0292] Among them, X a , X d is the horizontal coordinate of points A and D.

[0293] 3) Calculation of rod pump output

[0294] like Figure 6 As shown in the figure, the factors affecting the production of the rod pump include the effective stroke of the pump and the elastic coefficient of the pipe rod, the volume coefficient, and the leakage coefficient.

[0295] ① Elastic coefficient of pipe rod:

[0296]

[0297] S pu =X c -X b

[0298] ②The influence of filling coefficient

[0299]

[0300] S pd =X d -X a

[0301] ③Influence of volume coefficient

[0302]

[0303] ④The influence of leakage coefficient

[0304]

[0305] The pump efficiency of the rod pump can be expressed as:

[0306] η=η λ β η B ·η L

[0307] The final rod pump output can be expressed as:

[0308] Q t ′=1440SNA p η

[0309] Among them, Q t ′ is the theoretical production of CCUS production wells; S is the stroke of the rod pump, N is the number of rod pump strokes, A p is the pump area, B1 is the crude oil volume coefficient, S pu S is the distance between the opening and closing points of the fixed valve; pd is the distance between the opening and closing points of the floating valve; X a , X b , X c , X d Are the horizontal coordinates of the four points A, B, C, and D.

[0310] Since the production of the rod pump needs to be determined first in the leakage volume, the pump efficiency η is first assumed to be 100%, and then the production of the rod pump is iteratively calculated.

[0311] Step 5: CCUS production well production comparison and model correction.

[0312] (1) Comparison of actual oil well production Q tThe theoretical liquid production Q calculated in the present invention is t ′, calculate the difference in output. If the output difference is small, it means that the work diagram conversion model and the pump opening and closing point intelligent picking model are good. If the difference is large, go to the next step.

[0313] (2) Figure 7 As shown, the Fourier series K used in the Fourier transform method in the power diagram conversion model is changed. In the present invention, the value range of the Fourier series K is 10 to 20. The lower the Fourier series, the lower the fluctuation of the pump power diagram curve, and the smoother the image curve; the higher the Fourier series, the greater the fluctuation of the pump power diagram, and the steeper the image. Different downhole pump power diagrams are obtained by trying different Fourier series. To maximize the recognition effect of the opening and closing points of the pump ball valve. Then retrain the pump opening and closing point intelligent picking model. Until the model accuracy meets the accuracy requirements of oil well production measurement.

[0314] The method of the present application eliminates the influence of rod-tube vibration in the surface indicator diagram by calculating the downhole pump performance diagram, and then adopts the YOLO (You Only Look Once) model, a deep learning method in the field of target detection, to intelligently pick up the opening and closing points of the fixed valve and the floating valve in the pump performance diagram, and analyzes the fullness of the oil pump in combination with the position of the opening and closing points of the pump ball valve, and finally realizes the intelligent measurement of the output of CCUS production wells.

[0315] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for measuring the production of CCUS production wells based on oil well dynamometer diagrams, characterized in that: include: S1, collect CCUS production well data; S2, solving the downhole pump performance diagram based on the collected data; S3, establishing an intelligent picking model for opening and closing points of downhole pump performance diagram; S4, based on the calculation results of the pump opening and closing points, calculate the effective stroke of the pump and the degree of pump filling, analyze the pump leakage, plunger loss and crude oil volume coefficient of the oil pump, and obtain the CCUS well production; S5, compare the obtained CCUS well production with the actual oil well production, analyze the pump performance diagram conversion and the pump opening and closing parts picking situation, if the accuracy of the pump performance diagram conversion process is lower than the first threshold, return to step S2, if the pump performance diagram opening and closing point recognition effect is lower than the second threshold, return to step S3; until the accuracy requirement of oil well production measurement is met.

2. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 1, characterized in that: Step S1 includes: S11, collecting load and displacement data at the suspension point of the pumping unit, plotting the load and displacement data as a closed curve, and obtaining a surface indicator diagram of the CCUS production well; S12, collecting basic data of CCUS production wells, the basic data including well depth, rod string structure, and tubing string structure data of the oil wells, and collecting production data of the oil wells.

3. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 1 or 2, characterized in that: Step S2 includes: S21, the pump work diagram is transformed into a wave equation: Combining the structural parameters of the oil well pipe string and rod string, and using the Gibbs viscous damping wave equation, the displacement and load at the pumping unit suspension point are converted into the displacement and load of the oil well pump. The one-dimensional damped wave equation based on Gibbs is: Where U(x,t) is the displacement of the rod string section x at time t; a is the propagation speed of the stress wave in the sucker rod string; C is the damping coefficient; ① Initial conditions The bottom dead point position of the plunger during the upward stroke of the rod pump is taken as the starting point of a stroke. The initial displacement and velocity of the plunger are both 0, so the initial condition of the wave equation can be expressed as: ② Boundary conditions The dynamic load function D(t) and displacement function U(t) at a certain position of the column are the boundary conditions for solving the model, which can be expressed as: By expanding the load and displacement at the suspension point of the polished rod according to different Fourier series, we can get the equation: Where ω is the angular velocity; T is the period; σ n , τ n ,υ n , δ n is the Fourier coefficient; ③Connection conditions between poles The rod pumping system is often composed of a combination of rods of different levels. For a multi-level rod structure, assuming that the displacement and load of the connecting surface of two adjacent rods are continuous and unchanged, the connection condition between the rods can be expressed as: S22, pump power diagram wave equation solution: In this stage, based on the Fourier transform method, combined with the initial conditions, boundary conditions and connection conditions analyzed in the previous stage, the variables are separated and solved to obtain the following formula: From Hooke's law we can get: Then the relationship between the dynamic load and time at the depth x of the column can be expressed as: Where E is the elastic modulus of the sucker rod string, kg / m 3 ; The analytical expressions On(x), Pn(x) are expressed as: O n (x)=(K n chβ n x+δ n shβ n x)sinα n x+(μ n shβ n x+v n chβ n x)cosα n xP n (x) P n (x)=(K n shβ n x+δ n chβ n x)cosα n x-(μ n chβ n x+v n shβ n x)sinα n x Calculate the viscosity of the crude oil at different locations in the wellbore: In the formula, the correlation coefficients a1-a5 are determined by experiments; The corresponding damping coefficient of the rod column is as follows: Where m, B1, and B2 are as follows: For the Fourier coefficients of the multi-stage column, according to the principle of force continuity, the end point of the previous stage column of the multi-stage column is used as the starting point and connection condition of the next stage column, and the corresponding relationship of the Fourier coefficients of the i-th stage column is as follows: σ n (i)=EA r You n '(x) (i-1) τ n (i)=HE r P n '(x) (i-1) v n (i)=O n (x) (i-1) δ n (i)=P n (x) (i-1) Finally, after step-by-step calculations, the load and displacement values ​​of the last stage of the sucker rod string are obtained, and the downhole pump work diagram of the corresponding CCUS production well is drawn.

4. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 1 or 2, characterized in that: Step S3 includes: S31, the pump performance diagram data is a grayscale image of a preset size, the ball valve opening and closing point positions are marked for the converted pump performance diagram, the LabelImg tool is used to make label data, a file in a format suitable for YOLO network training is generated, and a pump performance diagram data set is made; S32, randomly dividing the pump performance diagram data set into a training set, a validation set, and a test set according to a preset ratio; S33, performing data enhancement on the pump performance diagram data by rotating the pump performance diagram data at a preset angle; S34, building a pump opening and closing point picking model based on the YOLO network; S35, after the model training is completed, the trained model is evaluated using the validation set, and the precision, recall and F1 score are calculated to ensure that the model can accurately detect the opening and closing points; The accuracy calculation formula is: The recall calculation formula is: The F1 score calculation formula is:

5. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 4 is characterized in that: In step S31, the pump performance diagram data is a grayscale image of 416×416 pixels.

6. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 4, characterized in that: In step S32, the pump performance diagram data set is randomly divided into a training set, a validation set and a test set in a ratio of 6:2:

2.

7. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 1 or 2, characterized in that: Step S4 includes: S41, from step S3, the four ball valve opening and closing point positions A, B, C, and D in the pump power diagram can be obtained, point A corresponds to the closing point of the floating valve, point B corresponds to the opening point of the fixed valve, point C corresponds to the closing point of the fixed valve, and point D corresponds to the opening point of the floating valve; S42, calculation of plunger effective stroke: The effective stroke of the plunger represents the displacement that actually produces the pumping effect within one stroke, which can be expressed as S pd =X d -X a Where, X a , X d is the horizontal coordinate of points A and D; S43, rod pump output calculation: The factors that affect the production of rod pumps include the effective stroke of the pump, the elastic coefficient of the pipe rod, the volume coefficient, and the leakage coefficient. Elastic coefficient of pipe rod: S pu =X c -X b The influence of filling coefficient: S pd =X d -X a The influence of volume coefficient: The influence of leakage coefficient: The pump efficiency of the rod pump can be expressed as: the=the λ ·b·h B ·or L The final rod pump output can be expressed as: Q t ′=1440SNA p or In the formula, Q t ′ is the theoretical production of CCUS production wells; S is the stroke of the rod pump, N is the number of rod pump strokes, A p is the pump area, B1 is the crude oil volume coefficient, S pu S is the distance between the opening and closing points of the fixed valve; pd X is the distance between the opening and closing points of the floating valve; a , X b , X c , X d The horizontal coordinates of the four points are A, B, C, and D respectively; First assume that the pump efficiency η is 100%, and then iteratively calculate the output of the rod pump.

8. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 1 or 2, characterized in that: Step S5 includes: S51, compared with the actual liquid production of the oil well Q t The theoretical liquid production Q calculated in step S4 t ', calculate the output difference, if the output difference is greater than the preset value, go to step S52; S52, changing the Fourier series K used in the Fourier transform method in the power diagram conversion model, and then retraining the pump opening and closing point intelligent picking model until the model accuracy meets the accuracy requirement of oil well production measurement.

9. The CCUS production well production measurement method based on the oil well dynamometer diagram according to claim 1 or 2, characterized in that: After step S5, the method further includes: deploying a CCUS production well production intelligent metering model.

Images