Robust PSFA-based oil field water injection system metering error prediction model construction method and correction system

Abstract

The invention provides an oilfield water injection system metering error prediction model construction method and correction system based on robust PSFA, and belongs to the technical field of oilfield water injection metering monitoring correction. Dividing test data into auxiliary variables and key variables and then establishing a data set; carrying out dimensionless processing on the data set; inputting the data set after dimensionless processing into a robust PSFA model, performing model parameter set learning training to obtain a model parameter set, and further obtaining a metering error prediction model of the oilfield water injection system; the system comprises a test data acquisition module, a model construction module, a field data acquisition module, a metering error prediction module, an error judgment module and a flow correction module. According to the method, the metering error of the oilfield water injection system can be predicted and corrected, the field working condition can be accurately restored, real-time monitoring and online compensation of the metering error of the flowmeter are achieved, and the fine management level of the oilfield water injection system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oilfield water injection metering and monitoring, and in particular to a method for constructing a metering error prediction model of an oilfield water injection system based on robust PSFA and a correction system. Background Art

[0002] During oilfield development and production, as the production time increases, the energy of the oil layer itself is continuously consumed, causing the oil layer pressure to continuously decrease, underground crude oil to degas, viscosity to increase, and oil well production to significantly decrease, or even stop flowing and stop production. Therefore, it is necessary to inject water into the oil layer to promote oilfield production. Currently, water injection development has become a widely used oil production method in oil fields both at home and abroad, and a variety of water injection schemes have been developed.

[0003] However, numerous challenges exist during oilfield waterflooding, hindering improvements in refined management. As injection time increases, factors such as large pores, cross-channeling, tubing and casing leakage, and high pressure and low permeability contribute to significant uncertainty in oilfield waterflooding. This in turn affects the system's injection volume, leading to injection anomalies and the need for precise metering of injection volumes.

[0004] Flowmeters (such as electromagnetic flowmeters, ultrasonic flowmeters, and orifice differential pressure flowmeters) are crucial instruments for monitoring water volume, ensuring adequate water injection, and ensuring precise water injection. They are widely used in all production links of water injection systems. Typically, flowmeter accuracy measurements require disassembly and completion at a testing center. However, many flowmeters cannot be frequently disassembled for inspection due to on-site production conditions and various objective factors, making it difficult to guarantee flowmeter accuracy. Due to the high oil and impurity content of produced water from oilfields, as well as the wear and tear and failure caused by the constant changes in the forces on the flowmeter during use, metering errors can easily occur during operation. If the flowmeter is not disassembled for inspection and correction in a timely manner, the metering error will persist or even increase. According to statistics, in existing oilfield water injection development, the metering error in the water injection system exceeds 10%.

[0005] CN105203189A discloses a self-calibration method for an online detection device of a liquid flow meter. The online detection device of the liquid flow meter calibrated by the self-calibration method includes a standard stainless steel straight pipe section, a portable ultrasonic flow meter to be tested, and a flow measurement and monitoring component. The self-calibration method for the online detection device of the liquid flow meter includes: step 1, before the online detection device of the liquid flow meter performs on-site flow detection, the flow measurement and monitoring component is verified or calibrated; step 2, starting the online detection device of the liquid flow meter, and simultaneously reading the flow signals of the flow measurement and monitoring component and the portable ultrasonic flow meter to be tested; and step 3, correcting the portable ultrasonic flow meter to be tested according to the flow value of the flow measurement and monitoring component. The technical solution of this method uses a portable ultrasonic flowmeter as a standard flowmeter. Before performing on-site flow measurement and comparing it with the value of the flowmeter to be measured to obtain the metering error of the flowmeter to be measured, in order to ensure the accuracy and reliability of the on-site measurement data, the standard flowmeter is first calibrated. Then, the calibrated standard flowmeter is brought to the production site for metering error detection of the flowmeter to be measured. This method has the advantages of not having to disassemble the flowmeter to be measured and not affecting the normal operation of the flowmeter to be measured, and can realize online detection of the metering error of the flowmeter.

[0006] CN112798082A discloses an online calibration method and online calibration equipment for a swirl flowmeter, and CN112833998A discloses an online calibration method for a pipeline liquid flowmeter, which respectively provide technical solutions for online detection of the flowmeter to be tested at an oil field production site.

[0007] Although the above scheme can detect the metering error of the flow meter, it requires the use of special equipment and can only be tested regularly at the production site. When a large number of flow meters are deployed in the oil field water injection system, it requires huge manpower costs and cannot achieve real-time monitoring of the metering error of the flow meter, resulting in a correction lag problem. Summary of the Invention

[0008] In response to the above problems, the present invention provides a method for constructing a metering error prediction model and a correction system for an oilfield water injection system based on robust PSFA, which can realize real-time online monitoring of flow meter metering errors and obtain metering error results in a timely manner without the need for additional special equipment for detection near the flow meter to be tested, thereby improving the efficiency of metering error detection and saving labor costs.

[0009] In a first aspect, the present invention provides a method for constructing a metering error prediction model for an oilfield water injection system based on robust PSFA, the method comprising:

[0010] Obtain test data of standard flowmeters and test flowmeters under different working conditions;

[0011] Dividing the test data into auxiliary variables and key variables and establishing a data set based on the auxiliary variables and the key variables, wherein the auxiliary variables are operating parameters of the standard flow meter and the test flow meter, and the key variables are flow errors of the standard flow meter and the test flow meter;

[0012] performing dimensionless processing on the data set established based on the auxiliary variables and the key variables;

[0013] The dimensionless data set is input into the robust PSFA model, and the model parameter set is trained to obtain the model parameter set, and then the metering error prediction model of the oilfield water injection system is obtained.

[0014] As a further improvement of the present invention, the operating parameters include the pressure P, pH, oil content O, Reynolds number Re, and magnetic flux V of the on-site environment of the medium.

[0015] As a further improvement of the present invention, the auxiliary variable is defined as x, the key variable is defined as y, x∈R D , where D represents the number of auxiliary variables x;

[0016] Define the dataset as T,

[0017]

[0018] t n =(x n ,y n )

[0019] Among them, t n represents the sample obtained at time n, N represents the number of samples, x n 、y n They represent the auxiliary variable values ​​and key variable values ​​collected at time n respectively.

[0020] As a further improvement of the present invention, the dimensionless processing of the data set established based on the auxiliary variables and the key variables includes converting the sample variances of the auxiliary variable samples and the key variable samples into unit variances.

[0021] As a further improvement of the present invention, the model parameter set is defined as Θ,

[0022] Θ={A,Λ,H,Φ,v}

[0023] Where A represents the state transfer matrix, Λ represents the variance matrix of the transfer noise, H represents the emission matrix, Φ represents the variance matrix of the observation noise, and v represents the degree of freedom parameter.

[0024] As a further improvement of the present invention, the robust PSFA model includes:

[0025] In the state space, the conditional probability density function of the latent variable is:

[0026] p(s n |s n-1 )=St(s n |As n-1 , Λ -1 , v)

[0027] p(s1)=St(s1|0,I K , v)

[0028] where s n 、s n-1 Represent the hidden variables at time n and n-1 respectively, A represents the state transfer matrix, Λ -1 represents the precision matrix, and v represents the degree of freedom parameter.

[0029] The probability density function of the observed variable in the observation space is:

[0030] p(d n |s n )=St(d n |Hs n , Φ -1 , v)

[0031] Among them, d n represents the observed variable at time n, s n represents the latent variable at time n, H represents the emission matrix, Φ represents the variance matrix of the observation noise, and v represents the degree of freedom parameter.

[0032] As a further improvement of the present invention, according to the properties of Student's t distribution, the robust PSFA model is further transformed into:

[0033] p(s n |s n-1 ,θ)=N(s n |As n-1 ,θ -1 Λ)

[0034] p(s1|θ)=N(s1|0,θ -1 I K )

[0035] p(d n |s n )=N(d n |Hs n ,θ -1 Φ)

[0036]

[0037] Where θ represents the precision parameter and Gam(·) represents the probability density function of the gamma distribution.

[0038] As a further improvement of the present invention, the performing model parameter set learning and training to obtain the model parameter set includes learning the model parameters using an expectation maximization algorithm.

[0039] As a further improvement of the present invention, the update formula of the expectation maximization algorithm is:

[0040]

[0041]

[0042]

[0043] Among them, a k3 、a k2 、a k1 、a k0 represents the equation coefficient, λ k represents the polynomial equation variable, k represents the iteration step, represents the error estimate of the data point, C d Represents the covariance matrix associated with the data points.

[0044] And there are:

[0045] a k3 =N-1

[0046]

[0047]

[0048]

[0049] Among them, s nk Indicates the k-th step of the n-th sample, s n-1，k represents the k-th step of the n-1th sample, N represents the number of samples, and <·> represents the mathematical expectation.

[0050] As a further improvement of the present invention, the solution formula for the degree of freedom parameter v is:

[0051]

[0052] where ψ represents the logarithmic derivative of the gamma function.

[0053] As a further improvement of the present invention, the statistic <θs n >、 The expression is:

[0054] <θs n >=<θ>μ n

[0055]

[0056]

[0057] where μ n 、V n 、J n-1 Obtained by forward and backward learning process.

[0058] As a further improvement of the present invention, the model parameter set learning and training to obtain the model parameter set also includes using the convergence condition of the expected value maximization algorithm iteration as a basis for judging whether the learning and training is completed.

[0059] As a further improvement of the present invention, whether the convergence condition is met is determined based on the calculation of the likelihood function.

[0060] The expression of the likelihood function after completing k iterations is as follows:

[0061]

[0062] The convergence conditions are:

[0063]

[0064] Among them, ε is the artificially set convergence threshold.

[0065] As a further improvement of the present invention, the method also includes verifying the metering error prediction model of the oilfield water injection system, inputting the auxiliary variable data under the specific test conditions into the metering error prediction model of the oilfield water injection system to obtain predicted key variable values, and comparing the key variable values ​​with the flow errors of the standard flow meter and the test flow meter measured under the corresponding working conditions.

[0066] In a second aspect, the present invention provides a metering error correction system for an oilfield water injection system, the system comprising:

[0067] Test data acquisition module, used to obtain test data of standard flowmeter and test flowmeter under different working conditions;

[0068] A model building module, configured to classify key variables and auxiliary variables according to the test data acquired by the test data acquisition module and then build a metering error prediction model for the oilfield water injection system;

[0069] On-site data acquisition module, used to obtain auxiliary variable data of all flow meters to be tested in real time;

[0070] a metering error prediction module, configured to calculate based on the auxiliary variable data acquired by the field data acquisition module and the prediction model constructed by the model construction module to obtain a predicted metering error of the flow meter to be measured;

[0071] an error judgment module, configured to compare the predicted metering error calculated by the metering error prediction module with a preset metering error threshold, and issue a correction instruction when the predicted metering error exceeds the metering error threshold; and continue monitoring the metering error when the predicted metering error does not exceed the metering error threshold;

[0072] The flow correction module is used to correct the flow meter according to the correction instruction given by the error judgment module after comparing the measurement error.

[0073] As a further improvement of the present invention, the model building module includes:

[0074] A data set establishment unit, configured to establish a data set after classifying the data acquired by the test data acquisition module into key variables and auxiliary variables;

[0075] a data set conversion unit, configured to perform dimensionless processing on the data set established by the data set establishment unit;

[0076] The model parameter learning and training unit is used to input the dimensionless data set obtained by the data set conversion unit into the robust PSFA model to perform model parameter learning and training, and determine the model parameters according to the training results.

[0077] As a further improvement of the present invention, the system further includes:

[0078] An error storage and processing module, used for recording and storing the calculated error data of the flow meter obtained by the metering error prediction module;

[0079] and an alarm module, which is used to issue an alarm message to notify the management personnel after the error judgment module gives a correction instruction.

[0080] The present invention provides a method for constructing a metering error prediction model for an oilfield water injection system based on a robust PSFA and a correction system, thereby establishing a reliable prediction model for calculating the metering error of the oilfield water injection system. The prediction model can be used to realize online real-time monitoring of the metering error of the flow meter to be measured based on auxiliary variable data automatically collected at the oilfield site. After receiving feedback, the correction system can promptly correct the flow meter with excessive metering error, which significantly improves the efficiency of monitoring and correction. At the same time, there is no need for manual on-site inspection, which saves manpower and material costs and is beneficial to reducing costs and increasing efficiency in oilfields. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a flow chart of a method for constructing a metering error prediction model for an oilfield water injection system according to a first embodiment of the present invention.

[0082] Figure 2 It is a schematic diagram of a simulation platform for analyzing sensitive factors of water injection system metering errors in the method for constructing a metering error prediction model for an oilfield water injection system according to the first embodiment of the present invention.

[0083] Figure 3 It is a structural diagram of a metering error correction system for an oilfield water injection system according to a third embodiment of the present invention.

[0084] Figure 4 It is a schematic diagram of the metering error correction process of the metering error correction system of the oilfield water injection system according to the third embodiment of the present invention. DETAILED DESCRIPTION

[0085] The following is a detailed description of the embodiments and the accompanying drawings. Figure 1-4 The invention is described in detail so that those skilled in the art can fully understand the purpose, features and effects of the invention.

[0086] In current oilfield production, the production data of the water injection system has basically been collected fully automatically. Various instruments and meters collect large amounts of data in real time every day and transmit them to the headquarters for storage, generating massive amounts of data related to water injection.

[0087] The present invention provides a method for constructing a metering error prediction model for an oilfield water injection system based on a robust PSFA and a correction system, thereby making corresponding improvements to the metering error monitoring of the existing oilfield water injection system. Based on various data automatically collected by the oilfield every day, the constructed prediction model can accurately judge the metering error of the flow meter, and the flow meter can be corrected in time after the metering error prediction value exceeds the error threshold. This can greatly reduce the metering error of the oilfield water injection system, make full and effective use of the large amount of data collected by various instruments and meters in oilfield production, and realize the refined management of the oilfield water injection system.

[0088] Example 1

[0089] As a specific embodiment of the present invention, this embodiment provides a method for constructing a metering error prediction model for an oilfield water injection system based on robust PSFA, wherein PSFA is a probabilistic slow feature analysis. Figure 1 ,include:

[0090] S100, obtain flow meter test sample data

[0091] Two groups of flow meters are set up, one of which is a standard flow meter and the other is a test flow meter. Flow tests are carried out on the water injection system measurement error sensitivity factor analysis simulation platform. The simulation platform is as follows: Figure 2 As shown, sensors are installed on the simulation platform for measuring parameters.

[0092] Based on the varying media pressures P, pH values, oil contents O, Reynolds numbers Re, and magnetic flux V in the field environment, various operating conditions are designed. Real-time flow rates Q and pressure data p measured by sensors in the simulation platform, as well as the actual flow rate Q1 of the standard flowmeter and the observed flow rate Q2 of the test flowmeter, are collected under these different operating conditions. In alternative embodiments, other parameter sets that can be automatically collected on-site in the oilfield can also be collected.

[0093] S200, establish a data set

[0094] The flow error ΔQ is calculated based on the flow values ​​of the standard flow meter and the test flow meter under different working conditions obtained in S100.

[0095] ΔQ=|Q1-Q2|

[0096] Define the key variable y, and take the flow error ΔQ as the key variable y.

[0097] Define auxiliary variables x associated with key variables, x∈R D , where D represents the number of auxiliary variables x, which include the medium pressure P, pH, oil content O, Reynolds number Re, and magnetic flux V of the field environment.

[0098] Define a dataset T consisting of key variables y and auxiliary variables x.

[0099]

[0100] t n =(x n ,y n )

[0101] Among them, t n Represents the sample obtained at time n, N represents the number of samples, n represents the time, x n 、y nThey represent the auxiliary variable values ​​and key variable values ​​collected at time n respectively.

[0102] S300, transform dataset

[0103] Based on the mean and standard deviation of the data, the data set T obtained by S200 is dimensionlessly processed, and the sample variances of the auxiliary variable samples and the key variable samples are converted to unit variance. The converted data conform to the standard normal distribution with a mean of 0 and a standard deviation of 1.

[0104] S400, learning and training model parameters

[0105] The data set T obtained in S300 is input into the robust PSFA model to perform learning and training of the model parameter set Θ.

[0106] Θ={A,Λ,H,Φ,v}

[0107] Where A represents the state transfer matrix, Λ represents the variance matrix of the transfer noise, H represents the emission matrix, Φ represents the variance matrix of the observation noise, and v represents the degree of freedom parameter.

[0108] According to the learning results, the model parameter set Θ is determined, and then the metering error prediction model of the oilfield water injection system based on robust PSFA is obtained.

[0109] To further verify the accuracy of the prediction model, after S400 , S500 is further included to verify the prediction model.

[0110] S500, prediction model verification

[0111] Obtain auxiliary variable data values ​​under a certain number of test conditions, including the medium pressure P, pH, oil content O, Reynolds number Re, and magnetic flux V of the field environment, as well as the actual flow rate Q1 of the standard flowmeter and the observed flow rate Q2 of the test flowmeter under the corresponding conditions, and calculate the key variable values.

[0112] The auxiliary variable data under each operating condition are dimensionless according to S300. Then, the model parameter set Φ obtained by S400 learning is used to predict the key variables based on the robust PSFA. Finally, the predicted values ​​of the key variables are compared with the measured values ​​obtained using the standard flowmeter and the test flowmeter.

[0113] When monitoring flowmeter metering errors, it's sufficient to collect auxiliary variable data from the flowmeter under test at a specific moment at the water injection and oil production site. After preprocessing, this data is input into a prediction model to determine the metering error of the corresponding flowmeter under test. Based on the resulting metering error, a decision can be made as to whether the flowmeter should be calibrated. The data required for auxiliary variables, such as the medium's pressure P, pH, oil content O, Reynolds number Re, and the field's magnetic flux V, are directly or indirectly measured by appropriate sensors. The data collected by these sensors can be transmitted in real time, enabling fully automated prediction of metering errors. Pressure P can be obtained using a pressure transmitter, pH using a pH concentration meter sensor, oil content O indirectly calculated using process technology, the Reynolds number indirectly calculated from the fluid pipe diameter, viscosity, and density, and the magnetic flux V using a flux sensor.

[0114] The method for constructing a metering error prediction model for an oilfield water injection system based on robust PSFA of the present invention can establish a reliable prediction model for calculating the metering error of the oilfield water injection system. The metering errors of all flow meters to be tested can be obtained based on the various parameter data automatically collected at the oilfield production site. This can realize real-time online monitoring of the metering error of the oilfield water injection system. The monitoring is accurate and efficient, and there is no need to arrange special personnel to carry special equipment to the oilfield production site for separate testing, which significantly reduces labor costs.

[0115] Example 2

[0116] As another specific embodiment of the present invention, this embodiment provides a method for constructing a metering error prediction model for an oilfield water injection system based on robust PSFA. On the basis of the first embodiment, the following further features are provided:

[0117] In S400, the robust PSFA model for the data set T input includes:

[0118] In the state space, the conditional probability density function of the latent variable is:

[0119] p(s n |s n-1 )=St(s n |As n-1 , Λ -1 , v)

[0120] p(s1)=St(s1|0,I K , v)

[0121] In the mathematical expression of probability density function, p(·) represents the probability density function, and p(W|Q) represents the probability density function of Q under the condition of W.

[0122] In the above formula, s n 、s n-1Represent the hidden variables at time n and n-1 respectively, A represents the state transfer matrix, Λ -1 represents the precision matrix, and v represents the degree of freedom parameter.

[0123] In mathematical expression, St(X|μ, Λ, v) represents the probability density function of the Student's t distribution, where X is a random variable and the parameters are the mean vector μ, the precision matrix Λ, and the degree of freedom parameter v.

[0124] The probability density function of the observed variable in the observation space is:

[0125] p(d n |s n )=St(d n |Hs n , Φ -1 , v)

[0126] Among them, d n represents the observed variable at time n, s n represents the latent variable at time n, H represents the emission matrix, Φ represents the variance matrix of the observation noise, and v represents the degree of freedom parameter.

[0127] Furthermore, according to the properties of Student's t distribution, the robust PSFA model can be further transformed into:

[0128] p(s n |s n-1 ,θ)=N(s n |As n-1 ,θ -1 Λ)

[0129] p(s1|θ)=N(s1|0,θ -1 I K )

[0130] p(d n |s n )=N(d n |Hs n ,θ -1 Φ)

[0131]

[0132] Where θ represents the precision parameter and Gam(·) represents the probability density function of the gamma distribution.

[0133] In mathematical expression, N(X|a, b) represents the probability density function of the normal distribution, where X is a random variable and the parameters are the mean vector a and the covariance matrix b.

[0134] Furthermore, when learning and training the model parameter set Θ, the expectation maximization algorithm is used to learn the model parameters, and the update formula is as follows:

[0135]

[0136]

[0137]

[0138]

[0139] Among them, a k3 、a k2 、a k1 、a k0 represents the equation coefficient, λ k represents the polynomial equation variable, k represents the iteration step, represents the error estimate of the data point, C d Represents the covariance matrix associated with the data points.

[0140] And there are:

[0141] a k3 =N-1

[0142]

[0143]

[0144]

[0145] Among them, s nk Indicates the k-th step of the n-th sample, s n-1，k It represents the kth step of the n-1th sample, N represents the number of samples, and <·> represents the mathematical expectation of the variable distribution.

[0146] Furthermore, the degree of freedom parameter v is obtained by solving the following nonlinear equation.

[0147]

[0148] where ψ represents the logarithmic derivative of the gamma function.

[0149] Furthermore, the statistic <θs n >、 The expression is:

[0150] <θs n >=<θ>μ n

[0151]

[0152]

[0153] Among them, μ n 、V n 、J n-1 Obtained by forward and backward learning process.

[0154] Furthermore, in forward learning, the conditional probability distribution p(s n |t 1：n ) is calculated as follows:

[0155]

[0156]

[0157]

[0158] And there are:

[0159] K n =P n-1 H T (Φ+HP n-1 H T ) -1

[0160]

[0161] Forward recursive message passing starts with the conditional probability distribution The calculation based on Bayesian principle is as follows:

[0162]

[0163]

[0164] K1=H T (Φ+HH T ) -1

[0165] Furthermore, based on the forward recursion, the posterior distribution update formula of the latent variable of the backward recursion is:

[0166] p(s n |T)=N(s n |μ n , V n )

[0167]

[0168]

[0169]

[0170] The backward recursive message passing begins with:

[0171]

[0172] Furthermore, the statistics <θ>,<lnθ> The expression is:

[0173] <θ>=e N / f N

[0174] <lnθ> =ψ(e N )-lnf N

[0175] have:

[0176] e N =(v+ND) / 2

[0177]

[0178] Where D represents the number of auxiliary variables x.

[0179] g n and G n From the following formula:

[0180] c n (θ)=p(t n |t 1：n-1 ,θ)=N(t n |g n ,θ -1 G n )

[0181] And there are:

[0182] g n =HAμ n

[0183] G n =Φ+HP n-1 H T

[0184] Furthermore, the convergence condition of the expectation maximization algorithm iteration determines whether the learning and training process is completed. The expression of the observation data likelihood function after completing k iterations is as follows:

[0185]

[0186] The convergence conditions are:

[0187]

[0188] Among them, ε is the artificially set convergence threshold.

[0189] The robust PSFA model based on Student's t distribution not only has the ability to handle the dynamic characteristics of data, but also can handle the noise interference caused by outliers, further improving the prediction accuracy of the model.

[0190] The method for constructing a metering error prediction model for oilfield water injection systems based on robust PSFA, provided by this invention, uses a machine learning algorithm to build a mathematical model based on dynamic outlier data. This model accurately predicts the magnitude of flowmeter errors over time, providing a basis for correcting flowmeter metering errors and achieving precise metering. The constructed prediction model provides real-time flow data for timely correction of flowmeter accuracy, enabling refined management of oilfield injection and production throughout the entire injection and production process.

[0191] Example 3

[0192] As a specific embodiment of the present invention, this embodiment provides a metering error correction system for an oilfield water injection system, which can use the metering error prediction model constructed in the first or second embodiment to predict the metering error and correct the metering error. Figure 3 and Figure 4 ,include:

[0193] The test data acquisition module is used to obtain different medium pressures P, different pH values, different oil contents O, different Reynolds numbers Re, different magnetic fluxes V in the field environment, as well as real-time flow rates Q and pressure data p under different working conditions during the test, the actual flow rate Q1 of the standard flowmeter, and the observed flow rate Q2 of the test flowmeter.

[0194] The model building module is used to build a metering error prediction model for the oilfield water injection system based on the test data obtained by the test data acquisition module, and includes a data set building unit, a data set conversion unit and a model parameter learning and training unit.

[0195] in:

[0196] The data set establishment unit is used to classify the data acquired by the test data acquisition module into key variables and auxiliary variables and then establish a data set. The medium pressure P, pH, oil content O, Reynolds number Re, and the magnetic flux V of the field environment are defined as auxiliary variables, and the differential flow error between Q1 and Q2 is defined as the key variable;

[0197] A data set conversion unit, used for dimensionless processing of the data set created by the data set creation unit;

[0198] A model parameter learning and training unit is used to input the dimensionless data set obtained by the data set conversion unit into the robust PSFA model to perform model parameter learning and training, and determine the model parameters according to the training results;

[0199] On-site data acquisition module, used to obtain auxiliary variable data of all flow meters to be tested in real time;

[0200] The metering error prediction module is used to calculate the key variable value, that is, the predicted metering error of the flow meter to be measured, based on the auxiliary variable data obtained by the field data acquisition module and the prediction model constructed by the model construction module.

[0201] The error judgment module is used to compare the predicted metering error calculated by the metering error prediction module with the preset metering error threshold. When the predicted metering error exceeds the metering error threshold, a correction instruction is given; when the predicted metering error does not exceed the metering error threshold, the metering error monitoring continues.

[0202] The flow correction module is used to correct the flow meter according to the correction instruction given by the error judgment module after comparing the measurement error.

[0203] After determining that the predicted metering error of the flow meter exceeds the metering error threshold, management personnel can also be arranged to go to the oil field production site to manually calibrate the flow meter, with online automatic calibration being preferred to improve calibration efficiency and reduce labor costs.

[0204] Furthermore, it also includes an error storage and processing module for recording and storing the calculated error data of the flow meter obtained according to the metering error prediction module, so as to facilitate subsequent analysis and use.

[0205] Furthermore, it also includes an alarm module, which is used to issue an alarm message to notify the management personnel after the error judgment module gives a correction instruction, so that the management personnel can promptly know which flow meters have been calibrated and can choose manual on-site correction. When the correction frequency of some flow meters is too high, the management personnel need to focus on these flow meters to find the cause.

[0206] The metering error correction system of the oilfield water injection system of the present application makes full use of various data related to water injection collected in the oilfield information management, and calculates the metering error of the corresponding flow meter through the metering error prediction module using the metering error prediction model constructed by the model construction module through the selected auxiliary variable parameter data obtained by the field data acquisition module, and decides whether to calibrate the flow meter based on the result obtained by comparison with the error judgment module, thereby ensuring that the flow meter can be calibrated in time and improving the correction efficiency of the flow meter.

[0207] The oilfield water injection system metering error correction system of the present invention can make full use of the massive data resources in the oilfield production process, quickly and accurately compensate for the metering error of the flow meter in real time, improve production efficiency and quality, and provide strong support for the stable production and development of the oilfield.

[0208] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent variation based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A method for constructing a metering error prediction model for an oilfield water injection system based on robust PSFA, characterized in that: The method comprises: Obtain test data of standard flowmeters and test flowmeters under different working conditions; Dividing the test data into auxiliary variables and key variables and establishing a data set based on the auxiliary variables and the key variables, wherein the auxiliary variables are operating parameters of the standard flow meter and the test flow meter, and the key variables are flow errors of the standard flow meter and the test flow meter; performing dimensionless processing on the data set established based on the auxiliary variables and the key variables; The dimensionless data set is input into the robust PSFA model, and the model parameter set is trained to obtain the model parameter set, and then the metering error prediction model of the oilfield water injection system is obtained.

2. The method for constructing a measurement error prediction model for an oilfield water injection system based on robust PSFA according to claim 1, characterized in that: The operating parameters include the pressure P, pH, oil content O, Reynolds number Re, and magnetic flux V of the on-site environment of the medium.

3. The method for constructing a measurement error prediction model for an oilfield water injection system based on robust PSFA according to claim 1, characterized in that: The dimensionless processing of the data set established based on the auxiliary variables and the key variables includes converting the sample variances of the auxiliary variable samples and the key variable samples into unit variances.

4. The method for constructing a measurement error prediction model for an oilfield water injection system based on robust PSFA according to claim 1, characterized in that: The performing model parameter set learning and training to obtain the model parameter set includes learning the model parameters using an expectation maximization algorithm.

5. The method for constructing a measurement error prediction model for an oilfield water injection system based on robust PSFA according to claim 4, characterized in that: The method of performing model parameter set learning and training to obtain the model parameter set also includes using the convergence condition of the expected value maximization algorithm iteration as a basis for judging whether the learning and training is completed.

6. The method for constructing a measurement error prediction model for an oilfield water injection system based on robust PSFA according to claim 5, characterized in that: Whether the convergence condition is met is determined based on the calculation of the likelihood function.

7. The method for constructing a measurement error prediction model for an oilfield water injection system based on robust PSFA according to claim 1, characterized in that: The method also includes verifying the metering error prediction model of the oilfield water injection system, inputting the auxiliary variable data under specific test conditions into the metering error prediction model to obtain predicted key variable values, and comparing the key variable values ​​with the flow errors of the standard flow meter and the test flow meter measured under corresponding conditions.

8. A measurement error correction system for an oilfield water injection system, characterized in that: The system comprises: Test data acquisition module, used to obtain test data of standard flowmeter and test flowmeter under different working conditions; A model building module, configured to classify key variables and auxiliary variables according to the test data acquired by the test data acquisition module and then build a metering error prediction model for the oilfield water injection system; On-site data acquisition module, used to obtain auxiliary variable data of all flow meters to be tested in real time; a metering error prediction module, configured to calculate based on the auxiliary variable data acquired by the field data acquisition module and the prediction model constructed by the model construction module to obtain a predicted metering error of the flow meter to be measured; an error judgment module, configured to compare the predicted metering error calculated by the metering error prediction module with a preset metering error threshold, and issue a correction instruction when the predicted metering error exceeds the metering error threshold; and continue monitoring the metering error when the predicted metering error does not exceed the metering error threshold; The flow correction module is used to correct the flow meter according to the correction instruction given by the error judgment module after comparing the measurement error.

9. The oilfield water injection system measurement error correction system according to claim 8, characterized in that: The model building module includes: A data set establishment unit, configured to establish a data set after classifying the data acquired by the test data acquisition module into key variables and auxiliary variables; a data set conversion unit, configured to perform dimensionless processing on the data set established by the data set establishment unit; The model parameter learning and training unit is used to input the dimensionless data set obtained by the data set conversion unit into the robust PSFA model to perform model parameter learning and training, and determine the model parameters according to the training results.

10. The oilfield water injection system measurement error correction system according to claim 8, characterized in that: The system further comprises: An error storage and processing module, used for recording and storing the calculated error data of the flow meter obtained by the metering error prediction module; and an alarm module, which is used to issue an alarm message to notify the management personnel after the error judgment module gives a correction instruction.

Images