RU2706578C1 - Method for filtration and clustering modes of complex rectification column - Google Patents

Abstract

FIELD: computer equipment.

SUBSTANCE: invention relates to computer engineering. Technical result is achieved by a method which includes forming from initial set of subsets of samples in form of initial clusters, initial set is formed by identification of each classified object by its parameter specifying coordinate of classified object in source set, and considered as training sample, data on clusters obtained at training stage, are entered into corresponding elements of memory unit, which are used at further sequential accumulation of measurement information in them. Objects to be classified are pressure and temperature, based on the measured pressure values, top and bottom temperatures and CRC sampling trays values of these temperatures are calculated for conditionally constant pressure, according to the recalculated temperatures, temperature profiles of CRC are constructed, boundary values at each temperature are assigned, which divide a plurality of process modes into subsets of temperature profiles.

EFFECT: technical result consists in obtaining a homogeneous set of data, which are used to obtain situational models, providing possibility of prediction and automatic control of production.

1 cl, 1 dwg

Description

The invention relates to the field of computer technology and can be used to control various processes of an oil refining complex with complex distillation columns (IBS) with several side screenings to provide the required quality criteria for production products through the use of predictive process models.

There is a method of collecting information about the ecological state of the region and an automated system for emergency and ecological monitoring of the environment of the region, including the placement of stationary and mobile control posts equipped with measuring equipment, registration and analysis of various environmental parameters, the received information is transmitted to documentation devices and simulation is performed, in the process modeling data is divided into a number of volumes, for each of which are a material balance model and noznuyu model (RF patent 2443001 №, G01W 1/00, op. 20.02.2012, BI № 5).

The disadvantage of this method is the lack of a data filtering mechanism according to the criterion of their reliability and uniformity.

Closest to the proposed invention is a method of automatic clustering of objects, comprising forming from the original set of classified objects samples in the form of initial clusters, the initial set being formed by identifying each classified object by its parameter that defines the coordinate of the classified object in the original set, and is considered as a training sample, which is formed according to the exponential distribution law, and cluster data obtained at the training stage are recorded in the corresponding elements of the memory block, which are used for subsequent sequential accumulation of measurement information in them, at the training stage, they also determine the Ki cluster model with the number of Ni elements satisfying the minimum risk RMi (α) of forming the cluster model (RF patent No. 2586025, G06F 17 / 30, op. 02/10/2016, BI No. 4).

This method uses a priori information only in choosing the exponential distribution law of cluster centers and intervals of permissible risks of classification errors, but does not use heuristic information about the clustered set. Since statistical methods are used for which the data homogeneity requirements of the clustered set must be fulfilled, in the absence of homogeneity information, additional regularization mechanisms are introduced to ensure the stability of solutions. To do this, first find a model of the first approximation of the location of clusters in a limited sample, and then carry out a dual procedure for classifying newly incoming data, i.e. simultaneously carry out training and clustering procedures.

Due to the fact that preliminary clustering is carried out according to a limited sample, which during the execution of the algorithm is supplemented with data with unpredictable properties, the process of clustering and training (tuning) can lead to incorrect identification of newly incoming data. At the same time, the mechanism for assigning newly received data to clusters is carried out only by probabilistic characteristics (confidence intervals), which cannot guarantee the correctness of the solution.

The technical result of the invention is to obtain a homogeneous set of data that are used to obtain situational models that provide forecasting capabilities and automatic production control by product quality indicators, which are estimated by calculating their values by models.

The specified technical result is achieved by the fact that in the method of filtering and clustering the modes of a complex distillation column, which includes forming from the initial set of subsets of samples in the form of initial clusters, the initial set is formed by identifying each classified object by its parameter defining the coordinate of the classified object in the initial set, and consider as a training sample, data on clusters obtained at the training stage are recorded in the corresponding elements of the memory block, the cat For the subsequent sequential accumulation of measurement information in them, according to the invention, the classified objects are pressure and temperature; on the basis of the measured values of pressure, top, bottom and bottom plates of complex distillation columns, the values of these temperatures for conditionally constant pressure are calculated from In this way, the temperature profiles are constructed of the temperature profiles of a complex distillation column, boundary values are assigned for each temperature, which They divide many technological modes into subsets of temperature profiles, technological modes for which temperature profiles do not satisfy the described conditions are excluded from the sample, and for each of these samples situational models for calculating the quality indicators of production products are built, and if the predicted accuracy of the obtained models turns out to be below the specified one, the filtering and clustering procedure is repeated for each of the subsets of temperature profiles obtained in the previous step.

The first difference between the proposed method and the prototype is that a priori information is used on the relationship of the technological parameters of a complex distillation column with side sampling, on the basis of which data are clustered. The second difference is that clustering from the very beginning is conducted across the entire data set. The classified set is formed by the measured values of the temperatures of the lateral takeoffs and the pressure obtained from the measuring devices, which is filtered and clustered to filter out the transient modes, and the modes that are a priori set by signs (criteria) differ significantly. Such classification features are, for example, the load of production, the characteristics of raw materials or the quality of the resulting products.

When selecting homogeneous data and clustering many states of technological modes to obtain situational models, we use the feature that, in a stationary steady state of the process, the relationship between the temperatures of the side sampling, which form the temperature profile of a complex distillation column, changes slightly. This relationship is used as a formal attribute for the separation of technological situations as a filtering rule:

(1)

(one)

– оператор преобразования множества технологических параметров

в множество признаков – температурных профилей СРК;Where

- operator of the conversion of many technological parameters

in many signs - temperature profiles of IBS;

. J is a criterion by which a generalized sign or a filtering rule into a set

.

As a criterion J, the use of the temperature profile of a certain region limited by the limiting values of the temperature of the lateral taps is used, while the temperature profile must completely belong to a given region. The fact that any temperature of the temperature profile is related to adjacent regions is a sign that for the regime under consideration the predetermined relationship between the temperatures of the side withdrawals is violated, and the regime is not stationary.

The filtering and clustering of complex distillation column modes is as follows.

For all modes presented in the statistical sample, based on the measured pressure values, top, bottom and bottom plates of a complex distillation column, the values of these temperatures for conditionally constant pressure are calculated. Based on the temperatures thus recalculated, the temperature profiles of SRS are built. Boundary values are assigned for each temperature, which divide the set of technological regimes into subsets of temperature profiles. Technological modes for which temperature profiles do not satisfy the described conditions are excluded from the sample. This procedure simultaneously filters and clusters (segmentes) data into subsets of statistically more homogeneous samples. For each of these samples, individual (situational) models for calculating the quality indicators of production products are further constructed, and if the predicted accuracy of the obtained models is lower than the specified one, the filtering and clustering procedure is repeated for each of the subsets of temperature profiles obtained at the previous stage.

An example of a specific implementation of the proposed method of filtering and clustering.

The measurement data of technological parameters were checked for gaps and random emissions under the condition:

T ^- <T <T ⁺ ,

where T ^- , T ⁺ - lower and upper permissible temperature (T) of lateral selection of complex distillation columns.

Formed an array of temperatures of the side withdrawals reduced to the base pressure based on Charles's law:

P / T = const (reduced temperatures); for this, the cast pressure P was assigned _. and temperatures of side sampling were calculated according to the formula:

,

,

,

– температура и давление соответствующего i-го бокового отбора;Where

,

- temperature and pressure of the corresponding i-th side selection;

– приведенная температура соответствующего i-го бокового отбора.

- reduced temperature of the corresponding i-th side selection.

After this, the values of the reduced temperatures of the side samplings were received in the unit for constructing and processing temperature profiles. The number of temperature profiles corresponds to the number of observations in the initial sample, which characterize the technological mode at the time the measurements were taken.

в этом случае – это температурный профиль сложной ректификационной колонны, определяемый по значениям температур верха СРК (цифра 1 по оси абсцисс), на тарелке отбора первого погона (цифра 2), второго погона (цифра 3), третьего погона (цифра 4) и температура низа СРК (цифра 5). На фиг. 1 показана кривая температурного профиля некоторого режима (пунктирная линия), который не соответствует данному критерию, поэтому значения параметров такого режима будут отфильтрованы.In FIG. 1 shows the separation of multiple temperature profiles into two clusters: upper, lower. Operator $$\text{Ψ}\{•\}$$

in this case, this is the temperature profile of the complex distillation column, determined by the temperature values of the top of the discharge system (figure 1 along the abscissa axis), on the selection plate of the first shoulder strap (figure 2), second shoulder strap (figure 3), third shoulder strap (figure 4) and temperature bottom of the IBS (number 5). In FIG. Figure 1 shows the temperature profile curve of a certain mode (dashed line), which does not meet this criterion, therefore, the values of the parameters of this mode will be filtered.

With ^about

Figure 1

Data that satisfy condition (1) form a cluster of homogeneous technological regimes.

The values of the technological modes corresponding to the clusters are highly likely to correspond to stationary modes and are homogeneous, and for them the known methods of regression analysis can be used to obtain models for calculating the quality indicators of production products that are used in the control device to calculate control actions according to specified quality indicators.

The proposed method for filtering and clustering modes of a complex distillation column can be used to identify a situational model for calculating quality indicators.

Using the proposed invention allows to solve the problem of obtaining situational models for calculating production quality indicators, which generally simplifies the structure of the regression model and reduces its adaptation time. At the same time, the requirements of a given accuracy are provided for the calculation of the product quality indicator. In the process of work, the proposed method of filtering and clustering data is used to identify the technological situation and situational models of calculation by a personal computer.

The method provides the possibility of obtaining situational models for calculating the quality indicators of production products with a given accuracy and automatic control of the separation of mixtures in complex distillation columns.

Claims (1)

A method for filtering and clustering the regimes of a complex distillation column, which includes generating from the initial set of subsets of samples in the form of initial clusters, the initial set is formed by identifying each classified object by its parameter setting the coordinate of the classified object in the initial set, and the cluster data is considered as a training sample, obtained at the training stage, recorded in the corresponding elements of the memory block, which are used for further sequential storage measurement information in them, characterized in that the classified objects are pressure and temperature, on the basis of the measured values of pressure, top, bottom and bottom plates of a complex distillation column, the values of these temperatures for a conditionally constant pressure are calculated, the temperature is calculated from the temperatures profiles of complex distillation columns, assign boundary values for each temperature, which divide the set of technological modes into subsets temperature profiles, technological conditions for which temperature profiles do not satisfy the described conditions are excluded from the sample, and for each of these samples situational models for calculating the quality indicators of production products are built, and if the predicted accuracy of the obtained models is lower than the specified one, the filtering and clustering procedure repeat for each of the subsets of temperature profiles obtained in the previous step.

Images