Design, Development and Application of Chemical Process Measurement Data Online Correction System (I)--Data Correction

Design, Development and Application of CPDRS for Chemical Process Measurement Data Online Correction System (I) * Data Correction Technology and Method for Calculating Physical Properties Zhou Chuanguang, Kong Ling 2, Zhao Wenjin, Si Yi, Zhang Qingrui, Xu Qingguo 1 (1. School of Chemical Engineering, Qingdao University of Science and Technology, Qingdao 266042, Shandong, China; 2. College of Life Sciences, Shandong University of Technology, Zibo 255049, Shandong, China) The device was applied and satisfactory results were obtained. The limitations of current error detection methods are discussed in detail. A combination test method combining process simulation with statistical verification is proposed. The characteristics of linear and non-linear data correction are analyzed and a combination correction method is proposed. For the first time, a time series analysis method for preprocessing measurement data is proposed, which not only increases the redundancy of the measurement data, but also improves the speed of the data correction and the accuracy of the correction values, and can make full use of the large amount of historical data of the DCS. The process of data correction; at the same time, it also discusses the property database and its calculation method.

The chemical process measurement data of the Journal of Qingdao University of Science and Technology is the basic information that reflects the characteristics of the device's operation status, and realizes computer process control monitoring, simulation optimization and production management. Due to the accuracy of the measuring instrument and the influence of the environment, these process-related in-situ measurement data are often not directly applied due to large errors. In addition, many process or control parameters are incomplete. Therefore, the process measurement data must be corrected to obtain high-quality data and correctly guide the actual operation and management of production.

Since the 1970s, with the application of computer information technology, data correction technology has gradually received people's attention, and made some progress in theory. Due to the complexity and particularity of chemical processes, it is difficult to achieve substantial results from theories of systems engineering and chemical engineering. The domestic correction algorithm for process data is the error vector, X is the correction value vector, U is the vector parameter to be estimated, and F is the function vector, which represents the material balance, energy balance, chemical reaction measurement relationship or other chemical and physical laws of the chemical process. Q is an (nXn)-order variance matrix of X that can be estimated by instrument accuracy or measurement samples.

The traditional statistical inspection and linearization methods use basic principles to detect negligence errors, data correction and parameter estimation. They have great limitations in practical use. First, there is a lack of reliable basis for detection of negligence errors, and the discriminative ability is not strong. Caused by the misdiagnosis; Second, data correction and parameter estimation rely too much on the process structure and spatial data, failed to make full use of the process of historical data; Third, the calculation time is long, it is difficult to meet the online correction of measurement data.

The research work of the system is relatively small, and most of them are limited to theoretical exploration. There have been no reports of successful general software development. The author developed special software for refining, ammonia, and urea plants based on the company's needs.

Based on the research and analysis of existing data correction techniques, this work explores in depth the detection of negligence errors and data correction, and proposes a new method of practical correction of process data.

1.1 The error of error detection The success of data correction technology depends on the effectiveness of the error detection method. If it cannot detect and eliminate On-line (CPDRS for short) in time, it is based on the actual production needs to integrate data correction technology with simulation optimization technology and develop application software on the computer. CPDRS can be used as a subsystem of CIMS to calibrate online measurement data with random errors, detect data with negligible errors, estimate unmeasured data, and use the corrected reliable data to perform real-time simulation optimization analysis of the device to determine the best operating parameters. , so as to achieve the purpose of reducing costs and saving energy and reducing consumption.

1 process data correction technology The basic idea of ​​measurement data correction technology is to assume that the process is steady state, the measuring instrument has a certain accuracy, the measurement data has a certain degree of redundancy, etc., so that the correction value of the measurement data meets the entire device and The balance of material balance, energy balance, etc. of the unit equipment also minimizes the sum of the squared differences of the measured values.

The basic model of data correction is: the error data contained in the quantity data may have a serious impact on the optimization control and management operations of the process.

In addition, the existence of negligence error is also a reflection of process failures. Measurement instrument faults and pipeline leakage are the main causes of fault errors. Therefore, the result of the detection of negligence error can guide the operator to repair the measuring instrument in a targeted manner and promptly eliminate the operation failure. Therefore, before the measurement data correction and parameter estimation are performed, the error detection must be performed first.

At present, there are many ways to detect fault errors. The commonly used methods are mostly based on mathematical statistics theory and have more limitations in practical use.

In the flow measurement data, the error errors are mostly caused by stable system deviations, such as the nature of the measurement medium has changed, and the flowmeter has not been calibrated in time. If such a group of data with large deviations is directly used to detect and correct the errors, the purpose of data correction is violated. However, if the measurement number min(2) of such a meter is removed, it may be due to insufficient data redundancy, and the number S.tF(X,U)=0(3) is affected by the accuracy of the correction or the correction.

Where X is a vector of measured values, which is a vector of real values, E is for a specific device, and the solution can be for each stream of weeklights, etc.: Design, development and application of a chemical process measurement data online correction system (CPDRS). Make a rough estimate of its deviation range. Prior to detection and correction, the raw measurement data is pre-processed. Then according to the actual operation and statistical experience, according to a certain coefficient to be corrected. This method is often arbitrariness and does not apply to general-purpose data correction software systems. To this end, the author proposed a combination test method, the specific steps are: (1) determine the standard value of the instrument. Generally, the set value of the device or the calculated value obtained by the system simulation can be selected.

(2) A modified method based on flow measurement mechanism analysis is used to determine the large system deviation of the instrument and calibrate it.

(3) Using the test data test method and the nodal test method to jointly detect the error of detection.

(4) If the measured value exceeds a certain range of the standard value, it is confirmed as a negligible error. After replacing it with the standard value, data correction and parameter estimation are performed.

(5) After the corrections and estimates have been obtained, compare them with analog values ​​or setpoints to determine if the resulting value is reasonable. If the error is still large, it will be included in the parameters to be estimated, and the correction and estimation will be performed again.

Repeat step 25 until the measured value's corrections and estimates are verified as reasonable data. The results of the example assessment show that the combined test method can accurately and effectively detect the system error error.

1.2 Correction of process measurement data After detecting the measurement error of the measurement data, the process measurement data correction model (2) can be solved according to the different forms of constraint equation (3).

If the constraint equation (13) is linear for both X and U, such as the material balance equation considering only the flow, the general solution can be deduced by the L method: the number matrix, C is a constant column vector.

If the constraint equation (3) is a nonlinear function of X and U, such as considering the material balance and energy balance equation constraints to solve the flow and temperature correction values, its data correction model is a set of nonlinear algebraic equations. Newton-Raphson method, Madron's linearization method and other iterative methods can be used to solve. This article uses the linearized iteration method to solve. The constraint equation (3) is linearly unwrapped: if x(,) and U1), then the i+1th iteration value x(, +0, and u(i+0.) is such that each iteration calculation constitutes a linear problem that can be directly Use the solution to the linear problem.

The iterative process until satisfying F(i(, +'U+4e, where e represents the accuracy required by the constraint equation. The initial value of the iterative process x, x, 0) can be selected empirically, in order to reduce the number of iterations, it is best to select The U can try to satisfy F(X(0), U(0)) = 0. In the past, the above two correction algorithms are usually used separately, if you simply use the linear method to correct the flow data, due to the number of constraint equations Less, so the accuracy is poor.

Considering energy balance, the number of constraint equations can be doubled, and the accuracy of temperature measurement values ​​can be fully utilized. However, when using the nonlinear iterative method for correction, the iterative calculation takes a long time and cannot meet the requirement for real-time correction of the device measurement data. For this reason, a combined correction algorithm is proposed, that is, the linear correction result of the flow data is used as the initial value of the flow for the nonlinear iteration. The application example shows that using this method improves the accuracy of the value of the flow data correction, but also can significantly reduce the number of iterations, reduce the calculation time, and meet the requirements of real-time data correction.

1.3 The time series analysis method has been used before data correction. One of the prerequisites for process measurement data correction is that the data has a certain degree of redundancy. Redundant data can be divided into two categories: The connection (balance equation) inside the process network causes the redundant data to be spatially redundant; the data that is repeatedly measured by the same precision meter at the same measurement point is time-redundant.

Current data correction methods and examples are mostly based on spatial redundancy. For a chemical plant that implements DCS automatic control, its database management system can collect and store a large amount of real-time measurement data. Obviously there is a certain link between the historical data of these same instruments and the current value, such as process changes, rate of change, etc. Therefore, starting from the principle of data correction and the effective use of information resources, these time-redundant data and spatial redundancy should be taken into consideration. The data is combined for data correction. In order to improve the accuracy of the data correction value, a new strategy of correcting the process data using the modified time series analysis method is proposed from the timing characteristics of the research time redundant data.

1.3.1 The basic idea of ​​the time series analysis method Time series analysis is a branch of probability statistics. It is the analysis and study of various average values ​​of dynamic data in the time domain. The use of time series analysis method for chemical process data correction can make full use of a large amount of historical measurement data information, to a great extent eliminate the influence of random errors, thereby significantly improve the accuracy of the correction value.

The superscript indicates the number of iterations. The objects studied in the known time series analysis of the ith iteration are a series of digital sequences (dynamic data) that have changed and are independent of each other over time in the Journal of Qingdao University of Science and Technology. The expression of the q-order moving average process model MA(q) is as follows: From equation (8), it can be seen that the output variable is a linear combination of the current and limited range of historical values ​​of the process, and xk only affects the future values ​​of Xk. That is, the input contained in Xk is completely refreshed after more than n sampling periods.

The number of measurement data in the time domain.

13.2 Data Correction with Random Error Equation (9) is a measurement data model with negligible errors, where j is the spatial position number of the measurement, / is the time series number of the measurement, x, V is the measurement, x(,) For true values, e, j are random errors, and b, ju, and j are errors. Assume that the data used for data correction does not contain negligible error terms b, ju, v, so the model (9) can be simplified to the following formula: (2) The time-series average model is given by (9), 10) +* v = e, which can be seen from the random error term *, = 1 e, v becomes e, v, since e, j obeys the normal distribution N (0, 12), it can be proved that the random error of e, V will be far Far smaller than e, j (because the positive and negative random errors cancel each other out), the basic steps of obeying the positive (3) time series correction method can be obtained by mathematical derivation (a) Determining an appropriate time domain value, which is to ensure the accuracy of the timing correction method The key lies. In general, the larger the time domain value is, the smaller the variance of the time sequence method is, and the better the correction result is. But to ensure that the premise of the timing method does not involve fluctuations in the process, the probability of fluctuations in a relatively small time domain interval is small. In practical applications, the requirements of the above two aspects must be taken into consideration. The time domain should not generally be too large. The time series average should be obtained based on the number of measurements (b) to perform data correction processing. The calculation results show that when the domain value *=10, the variance of the correction value is about one-tenth of the correction value not corrected by the time-series method.

1.3.3 Preprocessing of measurement data Since there is a gross error in the measurement process, a very large value of the deviation will appear in the measurement data. If the average method of the time series method (9) is directly applied, it is likely that the error will be averaged. Apportioning the measured values ​​in the time domain causes the average value to deviate significantly from the true value, rather than the original real-time measurement value. In addition, due to the large deviation of the timing method, the false detection rate is large. Therefore, for the first time, a new idea for preprocessing the measurement data should be proposed when applying the sequential method. That is, based on the Gaussian error normal distribution theory, a statistical discriminant method is used to remove a mixed value of bad values ​​from a series of measurement values. In order to make the results more in line with the actual situation, this work refers to the time sequence method with preprocessing as the modified time sequence method. Of course, the number of bad values ​​to be eliminated by statistical discrimination should be equal to the time domain value (capacity). Very few, there are currently many ways to use statistical discrimination to eliminate bad values, such as (a) the Grubbs criterion, (b) the Raita method, (c) the t-test criterion, and (d) Dixon criteria, etc. The CPDRS has special requirements on the physical subsystems. First, the material to be processed has a wide range of applications, such as gas, liquid and solid phases; various inorganic substances, organic substances, oil products, electrolyte solutions, etc. The second is that the method of estimation is simple and practical, and the reliability is high, so the accurate and rapid calculation of the physical property is a guarantee for the successful development of the process measurement data correction system, and it is also a prerequisite for realizing on-line data correction. Based on the previous work, this study mainly analyzed the petroleum distillates and the physical properties of electrolytes.

2.1 Calculation of Physical Properties of Petroleum Distillates When data correction is performed on an oil refining unit, calculation of the specific heat or enthalpy of crude oil and various fractions is required. This paper uses the frequency of oil and the stability of the system. Fractionation of virtual components. Is the use of the measured weeks of light, etc.: Design and development of chemical process measurement data online correction system CPDRS (I), such as the entire fraction of the density and Engelz distillation (ASTM) or boiling point (TBP) distillation data, with Proper steps and correlations determine the cut range and properties of each component (density, boiling point, relative molecular mass, critical temperature, critical pressure, eccentricity, etc.). The system uses the distillation profile data (ASTM or TBP) of petroleum fractions and the overall density for virtual component processing.

The physical properties such as relative molecular mass, critical temperature, critical pressure, and eccentricity factor required for the virtual component are estimated using the following empirical correlations.

Temperature, Critical Pressure, and Eccentricity Factors; Boundary Temperature, Critical Pressure); Critical Pressure); The above empirical correlations are all functions of the boiling point and density of petroleum fractions, and the results of various correlation calculations are slightly different. The actual calculations show that choosing different virtual component properties to calculate the correlation formula will affect the thermodynamic properties of the whole fraction and the prediction effect of the gas-liquid equilibrium. 2.2 Physical data including weak electrolytes are estimated and a small amount of biuret and air are calculated. The calculation of physical properties mainly includes the physical properties of these materials, the basic thermodynamic properties, and the calculation of thermodynamic gas-liquid balance. Based on a large number of manuals, the author selected a representative mathematical model based on expert recommendations and practical experience. For certain characteristics of certain substances, due to the large errors in each model, they are obtained by regression analysis of experimental data.

Gas-liquid equilibrium calculation is a difficult point in physical property calculation. The urea production process involves the calculation of the gas-liquid equilibrium of the NH3-H2O binary system, the NH3-CO2-H2O ternary system and the NH3-C2-H2-urea quaternary system. Since NH3 and C2 are weak electrolytes, and the normal operating conditions of urea synthesis are often above the critical conditions of the elemental components of the reactants, the phase balance of the urea system has some characteristics that are different from the general water and salt systems, while the characteristics of the The phase equilibrium is also accompanied by chemical equilibrium, which makes the gas-liquid balance of the system more complicated. This work mainly selects the Edward model to process the NH3-H2O binary system and the NH3-H2 ternary system's gas-liquid equilibrium calculation. For the NH3-CO2-H2O-urea quaternary system, the method for determining the liquid activity coefficient of the larvae was mainly used.

3 conclusions have good practicality.

(1) Based on the limitations of the current method of detecting negligence error, the combination of process simulation and statistical test combined test can effectively detect negligence error.

(2) The proposed combined correction method not only increases the redundancy of the measurement data, but also improves the speed of the data correction and the accuracy of the correction values, and can be used in the online data correction process.

(3) In order to make full use of the large amount of historical data of DCS, the time series analysis method for preprocessing the measurement data was first proposed for real-time process data correction. The modified timing method makes full use of the time redundancy of the measurement data and improves the accuracy of data correction.

(4) Established a database of physical properties of commonly used compounds; studied the virtual component processing methods for petroleum fractions, laying the foundation for the design and implementation of a general-purpose chemical process online measurement data correction and simulation optimization system.