RU2797753C1 - Method, system and computer-readable medium with a software product for predicting changes in layer-by-layer activity of a catalyst in a diesel fuel hydrotreating unit - Google Patents

Abstract

FIELD: computer engineering.

SUBSTANCE: invention concerns increasing the accuracy of predicting changes in the layered activity of catalysts to reduce energy consumption. It is achieved due to the fact that the prediction of changes in the layer-by-layer activity of the catalyst in the diesel fuel hydrotreatment unit (HT DF) is based on determining the rate of change of the linear approximation of the temperature drop in the catalyst layer in the reactor of the HT DF unit, determined using the temperature drop model in the catalyst bed. At the same time, the model is a polynomial regression model of arbitrary order with adjacent terms, the regressors of which are the parameters of the technological process, including: the boiling point R3 of 95% of the raw material at the inlet to the HT DF unit, the density R4 of the raw material at the inlet to the HT DF unit, the flow rate R5 of raw material to the HT DF unit, R6 consumption of make-up HBG for the HT DF unit, R13 content of sulphur in hydrotreated DF, R14 content of sulphur in feedstock, pressure difference between the reactor inlet and outlet, temperature difference between the reactor inlet and outlet, integral load F_sum.

EFFECT: increasing the accuracy of predicting changes in the layered activity of catalysts to reduce energy consumption.

15 cl, 4 dwg, 1 tbl

Description

FIELD OF TECHNOLOGY

[0001] The present invention generally relates to a system and method for controlling a process plant, such as, for example, a middle distillate refinery, and in particular, to a system and method for predicting changes in layer-by-layer catalyst activity in a diesel fuel hydrotreater (HD DF ).

[0002] In addition, the proposed invention relates to a computer-readable medium containing a software product, which, when executed by the processor, provides the ability to predict the change in layered activity of the catalyst in the installation of GO DF.

BACKGROUND OF THE INVENTION

[0003] At present, the oil refining industry has many unresolved problems associated with the introduction of more stringent requirements for gasoline, kerosene and diesel fuel to produce high-end environmentally friendly motor fuels. Comparatively rapidly changing requirements make it necessary to invest in the creation of new and modernization of existing plants.

[0004] One of the main methods of upgrading gasolines, diesel fuels, gas oils and other petroleum fractions are hydroprocesses. Upgrading in this field of technology refers, among other things, to the removal of compounds containing sulfur atoms from raw materials. Sulfur compounds worsen the quality of fuels, cause increased coke and soot formation in the engine, and increase emissions of sulfur oxides into the atmosphere. Hydrotreating (or hydrodesulfurization) is the process of upgrading raw materials on the active surface of a catalyst in a hydrogen-containing gas (HCG) environment.

[0005] The kinetics of the hydrotreating process is significantly affected by a number of factors, including, for example, temperature, pressure, feed space velocity, HCG cycle times, cycle gas purity, hydrogen partial pressure, contact time of the feed with the catalyst in the reaction zone, catalyst activity. It is extremely difficult for the plant operator to control some of these parameters, but these parameters must be taken into account in order to effectively control the plant. In this case, the process is controlled primarily by changing the temperature in the reaction zone.

[0006] Thus, the development of models for the hydrotreatment of petroleum fractions and the prediction of the parameters of these processes is an important task to ensure the possibility of predicting the yield of the final product, its quality and operating parameters of the process unit, as well as to plan further economic effect. Currently, there are several approaches to modeling and predicting various indicators for the components and stages of the production process. So, for example, in the article "In-line estimation of sulfur and nitrogen contents during hydrotreating of middle distillates" Pacheco M. E., Salim V. M., Pinto J. C. Brazilian Journal of Chemical Engineering (2009) describes a method for predicting the sulfur content in the final product based on the difference analysis between the specific gravity of the feedstock and the specific gravity of the final products. In the article “Digital twin. Modeling the Process of Hydro-Upgrading of Oil Fractions Using Machine Learning Methods” [Electronic resource] // URL: https://magazine.neftegaz.ra/articles/tsifrovizatsiya/504792-tsifrovoy-dvoynik-modelirovanie-protsessa-gidrooblagorazhivaniya-neftyanykh-fraktsiy-s -primeneniem-m/ (date of access: 04/22/2020)) for the analysis of residual sulfur content, it is proposed to use a process model using a neural network with the architecture of "Long short-term memory" (Long short-term memory, LSTM). The article "Catalyst Life Management with a Predictive Catalyst Deactivation Model", Robinson Paul, NPRA Plant Automation and Decision Support Conference (2004) describes a method for predicting catalyst deactivation based on a kinetic process model, the input parameters of which are: the chemical characteristics of the feedstock and product, data on the technological process of the previous run of the catalyst, a detailed description of the expected composition of the feedstock, a description of the expected modes and operating conditions. In the article "Development of a kinetic model of the diesel fuel hydrotreatment process", Afanasiev Yu.I., Krivtsova N.I. et al., Proceedings of TPU No. 3 (2012) to predict the activity of the catalyst, it is proposed to use a computer modeling system based on a kinetic model of the technological process that takes into account the transformation of sulfur compounds present in the feedstock of the hydrotreatment process.

[0007] Despite the fact that currently known approaches allow the design of models of petrochemical and oil refining processes, most of them are either extremely laborious (approx. physico-chemical modeling) or difficult in terms of understanding the final structure of the developed model (approx. machine learning) or do not provide sufficient prediction accuracy.

[0008] In this regard, there is currently a need to create a method, a system and a computer-readable medium with a software product that allows simple and high-precision prediction of process parameters in real time, in particular, predicting changes in layer-by-layer activity of catalysts used in technological installations of diesel fuel HE in general, in order to assess the residual resource and optimal use, without reference to the manufacturer, type and brand of catalyst, as well as to optimize energy consumption, expedient consumption of the catalyst resource and maintain the quality of the product at the required level.

SUMMARY OF THE INVENTION

[0009] The proposed method, system and computer-readable medium provide the ability to predict with the required accuracy the change in the layered activity of the catalyst in the GO DF unit in order to assess its residual resource and optimal use without reference to the manufacturer, type and brand of catalyst. The present invention allows real-time evaluation of the effect of a controlled process parameter on the predicted rate of layer-by-layer deactivation of a catalyst, as well as reducing the frequency of laboratory analyzes. Information about the activity of the catalyst by layers can help production technologists to consider the operation of reactors in more detail and more reasonably change the quench flow to control the temperature profile of the reactor.

[0010] According to the first aspect of the invention, a system is proposed for predicting changes in layer-by-layer activity of a catalyst in a diesel fuel hydrotreatment unit (HD DF), including: a parameter value receiving unit configured to receive hydrotreatment process parameter values, wherein the hydrotreatment process parameter values include laboratory data analyzes of raw materials at the inlet of the diesel fuel hydrotreatment unit and hydrotreated diesel fuel at the outlet of the specified unit and data from the sensors of the GO DT unit; a parameter value storage unit configured to store the received hydrotreating process parameter values as one or more time series; a block for generating and training a model, configured to: filter the stored values of time series based on Shewhart control charts, determine the values of the integral load F _sum of the reactor, which reflects the amount of raw material processed by the plant since the beginning of the use of the analyzed catalyst, form a training set based on the filtered values of time series , determining the size of the moving learning window based on the root-mean-square deviation of the temperature drop in the catalyst bed and the maximum correlation coefficient between the historical and predicted values of the temperature drop in the catalyst bed, forming a model of the temperature drop in the catalyst bed in the reactor of the GO DF unit, while the mentioned model is a polynomial regression an arbitrary-order model with adjacent terms, the regressors of which are the mentioned process parameters, including: the point R3 of boiling point of 95% of the feedstock at the inlet to the DT DH plant, the density R4 of the feedstock at the inlet to the DT DT unit, the flow rate R5 of the feedstock to the DH GO unit, the feed rate R6 WSG for the GO DF unit, R13 sulfur content in hydrotreated diesel fuel, R14 sulfur content in feedstock, pressure drop between the reactor inlet and outlet, temperature difference between the reactor inlet and outlet, integral loading F _sum , and training the model on the training set with obtaining regression coefficients , providing indicators of compliance of the simulated temperature drop in the catalyst layer with retrospective values above a predetermined level; a predictor configured to: determine the rate of change of the linear approximation of the temperature difference in the catalyst bed using the above model based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; evaluating the activity of the catalyst bed based on the rate of change of said linear approximation of temperature differences.

[0011] According to one embodiment of the invention, the parameter receiving unit may be configured to obtain new values of the hydrotreating process parameters, and the model generation and training unit may be configured to automatically retrain the models based on the received new values of the hydrotreating process parameters.

[0012] According to another embodiment, the predictor may be configured to communicate the resulting estimate of catalyst bed activity to the plant operator.

[0013] According to another embodiment, the predictor may be configured to transmit the obtained estimate of catalyst bed activity to the plant control unit.

[0014] According to another embodiment, the aforementioned determination of the values of the integral load Fsum of the reactor is carried out according to the formula:

where t _i - i-th point in time, F(t) - load at time t.

[0015] According to another embodiment, said construction of a linear approximation of the integral load Fsum is carried out according to the formula:

where t is time, a and b are linear coefficients found by the least squares method.

[0016] According to the second aspect of the invention, a system is proposed for predicting changes in layer-by-layer activity of a catalyst in a diesel fuel hydrotreatment unit (HD DF), including: a hydrotreatment process parameter values receiving unit configured to receive hydrotreatment process parameter values, including data from laboratory analyzes of raw materials at the inlet installations for hydrotreatment of diesel fuel and hydrotreated diesel fuel at the outlet of the specified installation and data from the sensors of the GO DT installation; a parameter value storage unit configured to store the received parameter values of the hydrotreating process; a predictor configured to: determine the rate of change of the linear approximation of the temperature difference in the catalyst bed using the trained polynomial regression model with adjacent terms based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; estimating the activity of the catalyst bed based on the rate of change of said linear approximation of the temperature difference across the catalyst bed; at the same time, the regressors of the mentioned model are the mentioned parameters of the hydrotreatment process, including: the point R3 of boiling point of 95% of the feedstock at the inlet to the DT DT unit, the density R4 of the feedstock at the inlet to the DT DT unit, the flow rate R5 of the feedstock to the DT DT unit, the flow rate R6 of make-up HSG to the unit GO DF, R13 sulfur content in hydrotreated DF, R14 sulfur content in feedstock, pressure drop between the reactor inlet and outlet, temperature difference between the reactor inlet and outlet, integral loading F _sum , reflecting the amount of feedstock processed by the unit from the beginning of the use of the analyzed catalyst.

[0017] According to the third aspect of the invention, a method is proposed for predicting changes in layer-by-layer activity of a catalyst in a diesel fuel hydrotreatment unit (HD DF), including: receiving and storing the values of the hydrotreating process parameters in the form of one or more time series, while the values of the hydrotreating process parameters include laboratory data analyzes of raw materials at the inlet of the unit and hydrotreated diesel fuel at the outlet of the unit and data from the sensors of the GO DT unit; filtering time series values using Shewhart control charts; determining the values of the integral load F _sum of the reactor, reflecting the amount of raw materials processed by the installation from the beginning of the use of the analyzed catalyst; formation of a training set based on filtered time series; determining the size of the moving learning window based on the standard deviation of the temperature difference in the catalyst bed and the maximum of the correlation coefficient between the historical and predicted values of the temperature difference in the catalyst bed; formation of a temperature drop model in the catalyst bed in the reactor of the GO DF unit, while the mentioned model is a polynomial regression model of arbitrary order with adjacent terms, the regressors of which are the mentioned process parameters, including: R4 of feedstock at the inlet to the DF HE unit, consumption R5 of feedstock to the DT DH unit, flow R6 of make-up HSG to the DF HT unit, R13 content of sulfur in hydrotreated diesel fuel, R14 content of sulfur in the feedstock, pressure drop between the reactor inlet and outlet, temperature difference between input and output of the reactor, the integral load F _sum ; training the model on the training set with obtaining regression coefficients that provide indicators of compliance of the simulated temperature drop in the catalyst layer with retrospective values above a predetermined level; determining the rate of change of the linear approximation of the temperature difference in the catalyst bed using the above model based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; evaluating the activity of the catalyst bed based on the rate of change of said linear approximation of temperature differences.

[0018] According to one embodiment, the method may further include the step of obtaining new hydrotreating process parameter values and automatically retraining the model based on said new hydrotreating process parameter values.

[0019] According to another embodiment, the method may further include the step of communicating the obtained estimate of catalyst bed activity to the plant operator.

[0020] According to another embodiment, the method may further include the step of transmitting the obtained catalyst bed activity estimate to the plant control unit.

[0021] According to another embodiment, the aforementioned determination of the values of the integral load Fsum of the reactor is carried out according to the formula:

where t _i - i-th point in time, F(t) - load at time t.

[0022] According to another embodiment, said construction of a linear approximation of the integral load Fsum is carried out according to the formula:

where t is time, a and b are linear coefficients found by the least squares method.

[0023] According to the fourth aspect of the invention, a method is proposed for predicting changes in the layer-by-layer activity of a catalyst in a diesel fuel hydrotreatment unit (HD DF), including: receiving and storing the values of the hydrotreating process parameters in the form of one or more time series, while the values of the hydrotreating process parameters include laboratory data analyzes of raw materials at the inlet of the unit and hydrotreated diesel fuel at the outlet of the unit and data from the sensors of the GO DT unit; determining the rate of change of the linear approximation of the temperature difference in the catalyst bed using the trained polynomial regression model with adjacent terms based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; evaluating the activity of the catalyst bed based on the rate of change of said linear approximation of the temperature difference across the catalyst bed; at the same time, the regressors of the mentioned model are the mentioned parameters of the hydrotreatment process, including: the point R3 of boiling point of 95% of the feedstock at the inlet to the DT DT unit, the density R4 of the feedstock at the inlet to the DT DT unit, the flow rate R5 of the feedstock to the DT DT unit, the flow rate R6 of make-up HSG to the unit GO DF, R13 sulfur content in hydrotreated DF, R14 sulfur content in feedstock, pressure drop between the reactor inlet and outlet, temperature difference between the reactor inlet and outlet, integral loading F _sum , reflecting the amount of feedstock processed by the unit from the beginning of the use of the analyzed catalyst.

[0024] According to a fifth aspect of the invention, a computer-readable medium is provided that stores a program product with program instructions that, when executed by a processor, perform one of the above methods.

BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The following describes non-limiting examples of preferred embodiments of the present invention with reference to the drawings, which illustrate the disclosed examples of preferred embodiments and do not limit the scope of the invention. On the drawings:

[0026] in FIG. 1 is a schematic representation of a typical diesel hydrotreater in accordance with the prior art;

[0027] in FIG. 2 schematically depicts a diesel fuel hydrotreater with a system for predicting the change in layered activity of the catalyst, made in accordance with the proposed invention;

[0028] in FIG. 3A, 3B, 4A and 4B are graphs of the results of a test run of the developed models.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0029] FIG. 1 shows an example of a typical diesel fuel hydrotreatment (HDF) unit 11 including, without limitation, two reactors 12 that can be used simultaneously. This configuration of unit 11 provides for hydrotreatment in two independent parallel feed streams, which makes it possible to carry out repair work without a complete shutdown of production. For a person skilled in the art it is obvious that the installation 11 GO DF may contain a different number of reactors, for example, one or more than two.

[0030] The raw material from the buffer tanks 13 of the tank farm is fed by means of the pump 14 to the mixing tee 15, in which the raw material is mixed with circulating hydrogen-containing gas (CVG). The raw materials can be, for example, light gas oil and straight-run diesel fractions. After mixing, the gas-raw mixture is heated in the HSS heating means 16, which may include furnaces and/or heat exchangers, and is fed to the inlet of the reactor 12.

[0031] The CVSG pressure in the hydrogen-containing gas circulation system is provided by the CVSG compressor 17, while to maintain the required hydrogen concentration, hydrogen can be supplied to the hydrogen-containing gas circulation system through the make-up WASH compressor 18, obtained, for example, at hydrogen production and / or catalytic reforming plants naphtha.

[0032] The hydrotreating reactor 12 is the main equipment of the hydrotreating process and is a vertical cylindrical vessel with a convex bottom of a spherical or close to elliptical shape, the height of which is greater than its diameter. According to the number of catalyst beds, the reactor 12 can be one-, two- or multi-sectional: reactors with one or two catalyst beds are typical for distillate hydrodesulfurization plants, and for hydrocracking plants with four to five. Cold recycle HSG (quench) may be fed into the reactor as an additional lever to control the temperature in the reactor 12. Above and below the catalyst beds may be limited by layers of porcelain beads and/or protective layers larger than the catalyst particles.

[0033] During operation, the catalyst layers become less permeable. This is especially true for the upper part. Therefore, the pressure drop in the reactor 12 at the end of the operating run is greater than at the beginning. In this case, the deactivation of the catalyst layers occurs unevenly; over time, the load on isothermal reactions can be transferred from one layer to another. This effect is explained by a number of factors, for example, the accumulation of corrosion products and coke in the layer, a decrease in the strength of catalyst particles, sintering, etc. With an increase in pressure drop, the costs of circulating hydrogen-containing gas increase.

[0034] Deactivation of the catalyst leads to a decrease in the activity of the catalyst and, as a consequence, to a decrease in the degree of desulfurization. To reduce the impact of the catalyst deactivation factor on the hydrotreatment process, the consumption of make-up HSG or the pressure of the gas-feed mixture at the inlet to the reactor is increased, or the partial pressure of hydrogen in the mixture is increased. One of the main methods for leveling catalyst deactivation is to increase the temperature of the feed stream at the reactor inlet. An increase in the temperature of the feed mixture at the inlet to the reactor due to the deactivation of the catalyst leads to an increase in the yield of by-products of gas and gasoline, and therefore reduces the efficiency of the diesel fuel hydrotreatment unit.

[0035] There are three main causes of catalyst deactivation: sintering or thermal deactivation, poisoning and blocking of active sites by coke.

[0036] Catalyst sintering leads to a decrease in the surface of the support, as well as to "coalescence" or loss of fineness of the metal crystallites. The loss of dispersity leads to a sharp decrease in activity. Impurity poisoning proceeds under the influence of adsorption on active centers of small amounts of a substance called poison and specific for this catalyst.

[0037] In the reactor 12, the feedstock is fed through a nozzle, which may be located in the upper part of the reactor 12, and is evenly distributed over the entire cross section of the reactor 12. In addition, the reactor 12 can be provided with a radial input of feedstock. In order to clean the raw material from mechanical impurities, mesh baskets immersed in the top layer of the catalyst can be used. Mesh baskets are not only a filtering device, but also serve to uniformly distribute raw materials with gases over the horizontal section of the reactor 12.

[0038] Many types of catalysts are known in the art for use in hydrotreating. In the majority of world and Russian enterprises, aluminum-cobalt-molybdenum (AKM), aluminum-nickel-molybdenum (ANM), and mixed aluminum-nickel-cobalt-molybdenum (ANKM) catalysts are most widely used. In the processes of deep hydrogenation of compounds containing nitrogen, as well as aromatic compounds, paraffins and oil fractions, aluminum-nickel or aluminum-cobalt-tungsten catalysts (ANV or AKV) are used. The choice of a particular catalyst depends on the physical and chemical characteristics required in the production, the shape and physical strength of the particles, selectivity, hydrogenation activity, desulfurization activity, operating pressure ranges, etc.

[0039] A detailed description of the reactions is not given in this application, since hydrotreating reactions depend on the composition of the industrial feedstock. Oil fractions, depending on their boiling range, may contain from several hundred to several thousand different compounds. This implies a huge variety of sequential and parallel reactions. Due to the high degree of polydispersity of hydrocarbon mixtures, the chemical activity of the reagents involved in such reactions varies over a wide range.

[0040] The main factors in the hydrotreatment process in the reactor 12 are, for example, temperature, pressure, feed space velocity, HCG cycle times, cycle gas purity, catalyst physicochemical characteristics.

[0041] Temperature. The reaction temperature is one of the most important process parameters. The depth and selectivity of hydrotreatment and hydrocracking reactions are very sensitive to it, since their rate constants increase exponentially with increasing temperature. On the other hand, an increase in the reaction temperature inevitably accelerates coke formation due to an increase in the rate of condensation of unstable cracking products. Therefore, to achieve the required selectivity, the temperature must be selected in accordance with the chemistry of the process. The control of the reaction temperature can be carried out, for example, by changing the temperature of the GSS entering the reactor 12, using means 16 for heating the GSS. For example, it may be possible to change the temperature in the furnace by increasing the supply of fuel gas to the furnace nozzles. In addition, the reaction temperature can be changed by changing other parameters, for example, by changing the volume of feed to the reactor 12 using appropriate valves, compressors, pumps, and the like.

[0042] The hydrotreating of middle distillates is carried out in a temperature range high enough for almost complete hydrodesulfurization reactions (330-390°C). At higher temperatures, splitting of light hydrocarbons is possible due to thermal cracking and, in the case of hydrotreating, an unfavorable shift in the equilibrium of the aromatic hydrogenation reaction. An increase in temperature produces several side effects that need to be carefully evaluated. Even a slight increase in temperature above acceptable values leads to a loss of selectivity and excessive activation of the catalyst. Therefore, for each individual case, there is a certain maximum permissible temperature. An increase in the reaction temperature inevitably accelerates coke formation due to an increase in the rate of condensation of unstable cracking products. High temperatures increase the degree of desulfurization, but at the same time accelerate the irreversible loss of activity due to metal precipitation. Temperatures above 410°C promote thermal cracking of valuable hydrocarbon components with the formation of significant amounts of low molecular weight liquids and gases. In addition, cracking of residues under harsh conditions can cause the formation of residues that are prone to fouling equipment of all kinds.

[0043] In commercial fixed bed reactors, the temperature increases as one moves down the bed. For this reason, the main problem in hydrotreatment operations is temperature control, for example, by setting a certain temperature of the GSS at the inlet to the reactor. Usually, in order to limit heat generation to smaller and safe portions, the total catalyst volume is distributed over several layers with intermediate cooling between them. For the hydrotreatment of gasoline fraction and kerosene, one layer is usually sufficient, since the heat release is relatively small, but in the case of heavier feedstock, a single-layer reactor may not be practical due to excessive temperature rise. In such cases, the catalytic beds are arranged so as to achieve a more favorable temperature distribution.

[0044] The temperature distribution in the reactor 12 is also affected by the loss of catalyst activity. During the cycle, it is compensated by a periodic increase in temperature, as a result of which the amplitude of the temperature profile gradually shifts upward. When the upper temperature limit reaches the maximum allowable for the reactor construction material, the cycle is terminated. At the end of the working run, the average temperature in the reactor 12 may exceed the initial temperature, for example, by 20-60°C. If the temperature along the axis of the reactor 12 is not correctly distributed, premature forced shutdown is possible - especially with a rapid loss of activity, as in the hydro-refrigeration of residues. In such cases, it is desirable to choose the lowest possible temperature rise in the layer in order to delay the moment of reaching the maximum allowable temperature. This means an increase in the number of layers and, accordingly, the dimensions of the reactor, which is necessary to accommodate additional equipment for intermediate cooling.

[0045] Thus, hydrogenation desulfurization of petroleum fractions is an exothermic process. Therefore, the temperature of the gas-raw mixture increases as it passes through the catalyst bed. In general, the higher the hydrogen consumption for the reaction, the more heat is released. To enable temperature control along the height of the reactor 12, it is possible to introduce cold hydrogen-containing gas (quench) into the zones between the catalyst layers.

[0046] Pressure. The hydrotreating process is affected by both the total pressure in the reactor 12 and the partial pressure of hydrogen. The degree of desulfurization increases with an increase in the total pressure in the reactor 12, or more precisely, the partial pressure of hydrogen. This slows down the reactions of dehydrogenation of naphthenic hydrocarbons, reduces the coking of the catalyst, accelerates the reactions of hydrogen saturation of unsaturated hydrocarbons and hydrogenation of aromatic hydrocarbons. The total consumption of hydrogen increases with increasing pressure.

[0047] Although the increase in the partial pressure of hydrogen plays a positive role, for him there are certain practical limitations. The allowable pressure is limited by the technical specifications of the equipment. Therefore, in order to bring the partial pressure of hydrogen as close as possible to the design one, it is important to keep the purity of the circulating hydrogen as high as possible. Another factor limiting the pressure is a strong increase in the cost of the reactor: in order to maintain a high operating pressure, it is necessary to increase the thickness of its wall.

[0048] With a significant increase in the total pressure, part of the raw material, even relatively light, for example, diesel fuel distillate, enters the reactor 12 in a liquid state, which negatively affects the efficiency of the process, since the rate of hydrogen diffusion through liquid hydrocarbons is low, and active centers catalysts in liquid-filled pores practically do not participate in reactions.

[0049] The volumetric feed rate of raw materials. Reducing the duration of contact as a result of increasing the space velocity of the feedstock (the ratio of the volume of liquid feedstock entering in 1 hour to the volume of the catalyst, based on bulk density) reduces the depth of desulfurization. As a result, the consumption of hydrogen and the degree of coking of the catalyst are reduced.

[0050] WASH circulation rate. In industrial practice, hydrotreatment processes are carried out with an excess amount of hydrogen, given that with an increase in its partial pressure, the reaction rates increase. The circulation rate is selected based on the goals of the process and economic considerations. If its value is lower than the calculated value, coke formation increases and the degree of conversion decreases. The circulation rate is particularly important for prolonging the catalyst cycle, which for this reason is usually carried out at very high hydrogen to feed ratios (1000-2000 Nm ³ /m ³ ). At the same time, maintaining a high circulation rate requires an increase in the flow rate of heat carriers and a more powerful compressor.

[0051] Gas circulation affects the balance of the liquid and gas phases in the reactor. Most hydrotreaters operate with partially vaporized hydrocarbon feeds. This circumstance affects the composition of the gas and the reaction rates. By increasing the ratio of hydrogen to the feedstock, one can achieve the concentration of the heaviest and most stable compounds in the liquid phase and increase the time of their contact with the catalyst. But, on the other hand, an increase in the gas circulation rate can lead to the fact that some substances in the evaporated fraction will not have access to the active centers of the catalyst particles.

[0052] With the passage of each layer in the reactor, the ratio of hydrogen to feed is reduced due to its chemical consumption. Intermediate cooling plays the role of a source of hydrogen replenishment at the entrance to the next layer. The distribution of the ratio of hydrogen to feed along the axis of the reactor is inversely proportional to the temperature distribution.

[0053] There is an optimal WASH circulation rate. The low degree of desulfurization of raw materials at low circulation ratio is explained by the insufficient supply of molecular hydrogen to the reactor 12 . The decrease in the degree of desulfurization at a gas flow rate above the optimum, but with the same capacity of the reactor 12 for raw materials, is associated with a decrease in the duration of its contact with the catalyst. The multiplicity of gas circulation for various conditions can be 220-700 nm ³ /m ³ liquid raw materials. The energy costs for CVSG compression by compressor 17 increase with an increase in the ratio of hydrogen to feedstock, as well as the hydraulic resistance of the CVSG circulation system, estimated from the pressure difference of the CVSG at the outlet of the compressor 17 and at its inlet.

[0054] Ceteris paribus, with an increase in the ratio of hydrogen-containing gas to the distillate feedstock, the amount of unevaporated feedstock decreases, and when a sufficiently high ratio is reached, a single-phase gas-vapor mixture enters the reactor 12.

[0055] The purity of the circulating gas. Hydrotreating often uses hydrogen-containing gas from catalytic reformers, and specialized hydrogen production units can also be used. Obviously, the concentration of hydrogen in the make-up gas can vary depending on the requirements. In CVSG, the hydrogen content is usually somewhat lower, since gaseous hydrocarbons formed in reactor 12 are added to the ballast gases supplied from outside (as part of fresh gas). in block 19 cleaning.

[0056] In the purification unit 19, the reaction products coming from the outlet of the reactor 12 are separated in the form of a gas-vapor mixture into hydrogen sulfide-enriched CVSG, hydrotreated gasoline and / or GO diesel fuel, as well as purification of hydrogen sulfide-enriched CVSG from reaction products with raw materials. To simplify the understanding of the present invention, the corresponding devices known in the prior art (refrigerators, heat exchangers, condensers, separators, stabilization columns, washing sections, etc.) that are part of the purification unit 19 are not shown in the drawings and are not disclosed in this description. .

[0057] In accordance with the invention, a prediction system 20 is provided that is capable of predicting the change in layer-by-layer activity of the catalyst in the DT DF unit 11 using prediction models that are mathematical regression models. The key parameters of the technological process and indicators of the quality of raw materials and product flows act as regressors that are part of the models.

[0058] In one embodiment shown in FIG. 2, the prediction system 20 includes a hydrotreating process parameter value receiving unit 21, a parameter value storage unit 22, a filtering unit 23, a model generation and training unit 24, and a prediction unit 25.

[0059] The parameter value receiving unit 21 is configured to receive the values of the physical and/or chemical parameters of the hydrotreating process from sensors located on the installation 11. The parameter value receiving unit 21 together with the sensors can be, for example, an industrial data acquisition system (DSS) , that is, a set of hardware or firmware that collects, selects, converts, stores and initially processes various input analog and / or digital signals. The sensors can also be structurally and functionally separate devices containing one or more primary measuring transducers that generate a signal with information about the measurement in a form that is compatible with the block 21 for receiving parameter values. The signal from each of the sensors can be transmitted to the parameter value receiving unit 21 via, for example, a wired, fiber optic, wireless connection, or a combination thereof.

[0060] In one embodiment, the sensors may include one or more of the following: a raw material flow sensor R1 per thread, which can be installed on the production line before the mixing tee 15, a raw material flow sensor R5 on the installation 11, which can be installed on the production line between the buffer tank 13 and the mixing tee 15, the sensor R6 for the make-up WASH flow rate to the unit 11, which can be installed at the inlet to the WASH circulation system, the sensor R7 for the CVSG flow into the mixing tee 15, which can be installed at the CVSG inlet to the mixing tee 15, sensors R8 for the consumption of diesel fuel and R9 of gasoline at the outlet of the unit 11, which can be installed at the outlets of the purification unit 19, the sensor R10 for the flow of light gas oil (LG) at the unit 11, the sensor R11 for the temperature of the GSS at the inlet to the reactor 12, which can be installed on the line feedstock before entering the reactor 12, a pressure sensor R12 at the outlet of the heating means 16, which can be installed at the outlet of the HSS heating furnace, pressure sensors R15 in the reactor 12, which can be installed between the catalyst layers in the reactor 12, temperature sensors R16 in reactor, which can be thermocouples installed between the catalyst layers in the reactor 12.

[0061] In addition, the parameter value receiving unit 21 can be configured to receive laboratory analysis data, for example, data on the point R2 of the boiling point of 50% of the raw material at the entrance to the installation 11, data on the point R3 of the boiling point of 95% of the raw material at the entrance to the installation 11 , data on the density R4 of the raw material at the entrance to the installation 11, data on the content R13 of sulfur in the final product, data on the content of R14 sulfur in the raw material.

[0062] The sequences of values received by the parameter value receiving unit 21 are transferred to the parameter value storage unit 22 for indexing these values by the time of measurement or by the time of acquisition and storage in the form of time series. At the same time, the block 22 for storing parameter values can be configured to store the values of various parameters in the form of a plurality of one-dimensional time series, each of which reflects the development of only one process over time, or in the form of one or a plurality of multidimensional time series, each of which contains observations for changing more than one parameter. The values of the time series are obtained by registering the corresponding parameter of the process under study at certain time intervals. In this case, depending on the nature of the data and the nature of the tasks being solved, either the current value or the sum of the values accumulated over a certain time interval can be recorded.

[0063] In view of the fact that the data of the sensors and the data of laboratory analyzes can be recorded at unequal time intervals, the parameter value storage unit 22 can also be configured to aggregate the data.

[0064] In general, the parameter value storage unit 22 may be configured to record, store, process data, and provide access to data. The parameter value storage unit 22 may be implemented using any kind of volatile or non-volatile memory, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disk, disk array or other storage device, or any other medium capable of storing the required data and which can be accessed using a computing device. In addition, combinations and combinations of any of the above devices can be used to implement the block 22 for storing parameter values. To perform data processing and indexing, the parameter value storage unit 22 may also include a specialized computing device and/or a general purpose computing device configured to perform the required operations.

[0065] To improve the accuracy of the prediction model, the original data stored in the form of time series can be filtered in the filtering unit 23 . The filter unit 23 may be implemented with one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers. , microprocessors or other electronic components. The filtering unit 23 can also be implemented on the basis of a personal or industrial computer with sufficient computing power or a distributed network of such computing facilities. The filter unit 23 may also be equipped with an input/output interface providing an interface between the filter unit 23 and peripheral devices such as a keyboard, display, and the like. Peripheral devices can be used, for example, to make changes to the program code under which the filter unit 23 performs its functions.

[0066] In the present invention, the source data filtering algorithms are based on the principle of Shewhart's control charts. Before the formation of the retraining matrix of the model, the data in the "raw" form for the period determined by the number of laboratory analysis points (or the retraining window) are received at the input of the filtering unit 23. In this case, the incoming data are sets of vectors of different lengths, from which the retraining matrix will be compiled. Next comes the work with each vector in the loop. In one iteration, one vector for the retraining matrix is processed. For each vector, a partition into N subgroups occurs (for example, N=5), and a matrix of subgroups and a range vector of each subgroup are formed. After that, the average ranges for all subgroups are determined, as well as the upper and lower control limits, the coefficients of the borders are selected in accordance with a table of predetermined coefficients for calculating the lines of the control charts. Further, in the current vector, groups that go beyond the control boundaries are determined and removed. These groups do not participate in retraining.

[0067] Performing filtering of the initial data allows to increase the accuracy of the model due to the fact that the model is trained only on the correct set of initial data, from which outliers, “broken” and erroneous values are excluded.

[0068] The filter unit 23 can also be configured to further filter and transform the original data (time series) by methods known in the art to obtain a training set.

[0069] In one embodiment, prediction system 20 also includes a model generation and training unit 24 . The model generation and training unit 24 may be implemented using one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), field programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components. Block 24 of the formation and training of the model can also be implemented on the basis of a personal or industrial computer with sufficient computing power or a distributed network of such computing facilities. The model building and learning unit 24 may also be equipped with an input/output interface providing an interface between the model building and learning unit 24 and peripheral devices such as a keyboard, display, and the like. Peripheral devices can be used, for example, to make changes to the program code under the control of which the block 24 for generating and training the model performs its functions. The present invention uses regression analysis to predict the temperature drop for one or more catalyst beds, which the inventors have found to be optimal for the task at hand and to find correlations between input and output variables. At the same time, each input parameter affects the resulting value with its own specific weight - the regression coefficient, which makes it possible to assess the physical reliability of the model.

[0070] The model generation and training unit 24 may be configured to determine the optimal size of the moving learning window. In this case, the calculation is performed on the basis of the root-mean-square deviation of the simulated temperature difference between the beginning and end of the catalyst layer from the predicted values and the maximum correlation coefficient between the historical (retrospective) and predicted values of the temperature difference between the beginning and end of the catalyst layer. For each of a predetermined set of moving learning window sizes, correlation coefficients are calculated and the size that provides the highest correlation coefficient is selected. Defining a learning window allows you to highlight the most significant data in the process history. From the entire time sample, a certain number of recent points are selected, used for training to exclude old and weakly influencing data.

[0071] In addition, the block 24 of the formation and training of the model can be made in the form of a software and hardware complex that allows the operator to manually set the size of the moving learning window.

[0072] The model building and training unit 24 can also be configured to evaluate the relative importance of regressors, search for multicollinear regressors, calculate regression coefficients, assess the model's fit with historical data, add and remove regression coefficients, and perform other functionality that allows you to improve performance. model quality.

[0073] Combining retraining polynomial regression models with the optimal size of the moving model training window can significantly increase the accuracy of the generated model. At the same time, a larger number of degrees of freedom makes it possible to build more accurate models due to the complexity of the computational process.

[0074] In one embodiment, the block 24 generation and training of the model allows you to generate polynomial models of arbitrary order with adjacent members to build models that are more closely approximated to real non-linear systems. Polynomials of more than the first order include the products of regressors and their degrees of more than the first. Block 24 of the formation and training of the model can additionally be performed with the ability to build predictive models to reflect the impact of current readings of process parameters on possible subsequent changes in the final product.

[0075] In one of the embodiments, the block 24 of the formation and training of the model can be made with the possibility of correcting the effects of individual technological parameters for the most accurate processing of possible disturbances. This functionality can be implemented, for example, by adding to the final result the product of the change in the parameter by the modifying coefficient when a significant perturbation in the modifying parameter is accumulated.

[0076] The following regressors can be included in the model of the temperature difference between the beginning and end of the catalyst bed: consumption R1 of feedstock per string, boiling point R2 of 50% of feedstock at the inlet to unit 11, point R3 of boiling up of 95% of feedstock at the inlet of hydrotreating unit 11, density R4 feedstock at the inlet to the hydrotreatment unit 11, feedstock R5 consumption to the hydrotreatment unit 11, flow rate R6 make-up HSG to the hydrotreatment unit 11, flow rate R7 CFSG to the mixing tee 15, flow rate R8 DF at the outlet of the unit 11, consumption R9 of gasoline at the outlet of the unit 11, LH flow rate R10 at installation 11, HSS temperature R11 at the inlet to reactor 12, pressure R12 at the outlet of the HSS heating means 16, sulfur content R13 in hydrotreated diesel fuel, sulfur content R14 in the feedstock, pressure drop R15 between the reactor inlet and outlet, differential R16 temperature between the inlet and outlet of the reactor.

[0077] The authors of the present invention found that in order to obtain an acceptable correlation coefficient and standard (root mean square) deviation, the following regressors should preferably be included in the model: the point R3 of the boiling point of 95% of the feedstock at the inlet to the GO DF unit, the density R4 of the feedstock at the inlet to the DF HE unit, feedstock consumption R5 to the DF HT unit, feed rate R6 of make-up HSG to the DF HT unit, sulfur content R13 in the hydrotreated diesel fuel, sulfur content R14 in the feedstock, pressure drop R15 between the inlet and outlet of the reactor, temperature difference R16 between the inlet and reactor outlet.

[0078] In addition, in order to take into account the influence of the astronomical time elapsed since the installation of the catalyst on the directly considered object on the chemical and physical processes, an additional regressor “integral load” should be introduced into the model, reflecting the amount of raw material used by the 11 GO DT unit from the beginning of the use of the catalyst . The integral load at a particular point in time can be calculated by the block 24 for the formation and training of the model, for example, based on the known consumption of raw materials per thread and the time elapsed since the installation of the catalyst.

[0079] Thus, preferably, the regressor integral load F _sum should also be included in the model. In this case, the equation for calculating the integral load F _sum can be as follows:

where t _i - i-th point in time, F(t) - load at time t.

Thus, at each moment of time F _sum (t) - reflects the amount of raw materials that have passed through the block since a given moment t ₀ .

[0080] In one of the embodiments of the invention, the calculation of the integral load F _sum is carried out in the block 24 of the formation and training of the model, in addition, the calculation of the integral load F _sum can also be carried out in the prediction block 25 or performed using other software or firmware and be fed ready-made to the block for receiving parameter values.

[0081] In one embodiment, the regression equation of a mathematical model for predicting the temperature drop across a particular catalyst bed may be, for example, as follows:

where ΔT _model [j] - temperature difference model in layer j, C _i - coefficients of the model, R _i - regressors, F _sum - integral load with the corresponding coefficient C _Fsum .

[0082] For convenience, the list of regressors included in the above model of the temperature difference in the catalyst bed is presented in the following table:

[0083] The block 24 for generating and training the model can also be configured to retrain the models with further accumulation of the historical sample, for example, with the arrival of new data from laboratory analyzes, with a change in the regression coefficients C _i . This allows you to correct the approximate functions of the models in the vicinity of the current equilibrium point of the real nonlinear process, thereby increasing their accuracy. In this case, retraining can be carried out automatically upon receipt of new retrospective data to block 21 for receiving parameter values.

[0084] Preferably, all process data entering the model, with the exception of laboratory analyzes, is averaged per hour.

[0085] The parameter value receiving unit 21 may also be configured to receive a second set of values, which may be hydrotreating process parameter values that are not included in the training set, i.e., are not retrospective. The parameter value receiving unit 21 can also be configured to receive individual parameter values from the operator, for example, to predict the effect of changing one of the parameters included in the model on the rate of deactivation of a particular catalyst layer.

[0086] Formed and trained by the block 24 of the formation and training of the model, as well as the data received from the block 21 for receiving parameter values, are fed to the block 25 of prediction.

[0087] The predictor 25 may be implemented with one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), field programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components. The prediction block 25 can also be implemented on the basis of a personal or industrial computer with sufficient computing power or a distributed network of such computing facilities. The prediction unit 25 may also be equipped with an input/output interface providing an interface between the prediction unit 25 and peripheral devices such as a keyboard, display, and the like. Peripheral devices can be used, for example, to make changes to the program code under which the predictor 25 performs its functions.

[0088] As described above, the process of catalytic hydrotreating is accompanied by the release of heat during the exothermic reactions caused by the catalyst. Over time, the catalyst loses its properties, and therefore the amount of heat released during the hydrotreatment process decreases. In this case, when a certain value of the temperature difference between the catalyst layers is reached, the layer is considered exhausted and has exhausted its resource, since exothermic processes have ceased to occur in it. Thus, the rate of change in the temperature difference across the catalyst bed directly indicates the rate of deactivation of this catalyst bed. This principle can be used by predictor 25 to predict the stratified rate of catalyst deactivation.

[0089] The predictor 25 may be configured to determine a temperature linear approximation with respect to time (indirectly via a linear approximation of the integral load) based on the generated model, the linearly approximated integral load, and the average values of the regressors over a predetermined period of time:

where avg(R _i ) - the average values of the regressors for a predetermined period of time with their respective coefficients C _i , F _sum1 - linearly approximated integral load with the corresponding coefficient C _Fsum .

[0090] In this case, the linearly approximated integral load F _sum1 can be calculated by the predictor 25, for example, using the least squares method:

where t is time, a and b are linear coefficients found by the least squares method.

[0091] Next, in predictor 25, the linear velocity of the obtained temperature approximation may be determined. In one embodiment, the linear velocity of the resulting temperature approximation can be determined using the following formula:

where ΔT _model is a linear approximation of the temperature, t _end is the current time, t ₀ is the start time.

[0092] In one embodiment of the present invention, the time difference can be counted in days.

[0093] The results of predicting the change in layered activity of the catalyst and other process parameters can be sent to the terminal 26 of the operator and/or process engineer for display on the screen of the terminal 26 of the operator and/or process engineer, for example, through the construction of appropriate graphs. At the same time, the visualization of the technological process on the screen of the terminal 26 of the operator and/or process engineer can be carried out in real time. The terminal 26 of the operator and/or process engineer may also provide the ability to select additional objects and parameters for analysis and monitoring using appropriate user interface elements.

[0094] Examples of information displayed on the screen of the operator's and/or process engineer's terminal 26 are shown in FIG. 3A, 3B, 4A and 4B. In particular, in FIG. 3A shows a graph of the temperature difference for the 1st catalyst layer based on real historical data and a graph of the temperature difference for the same layer predicted by the proposed prediction system 20 using the trained model. In FIG. 3B is a plot of the deactivation rate for the 1st catalyst bed.

[0095] FIG. 4A shows a plot of the temperature drop for the 3rd catalyst layer based on real historical data and a plot of the temperature drop for the same bed predicted by the proposed prediction system 20 using the trained model. In FIG. 4B is a plot of the deactivation rate for the 3rd catalyst bed.

[0096] The ability to set arbitrary values of controlled parameters using terminal 26 of the operator and / or process engineer, for example, the flow rate R5 of raw materials for the installation, the flow rate R6 of make-up HSG for the installation, the temperature R11 of the GSS at the inlet to the reactor, the flow rate of the quench, etc. allows provide real-time decision support. For example, the forecasting system provides the ability to build real-time forecasts of changes in the process when one or more parameters change. Thus, the operator and / or process engineer has the opportunity to assess the impact of proposed changes on the process before they are directly implemented. After evaluating and selecting the required change in parameters, the corresponding commands can be transmitted from the terminal 26 of the operator and / or process engineer to the controller that controls the pump 14, the compressor 18, the valves installed on the production lines and / or the means for heating the gas-feed mixture at the inlet to the reactor 2. The temperature in the furnace of the heating means 16 is, for example, changed by increasing the supply of fuel gas to the furnace nozzles. Other temperatures can be changed by adjusting other parameters.

[0097] Commands from the terminal 26 of the operator and/or process engineer can also be sent to the appropriate controller and without the participation of the operator. For example, predefined commands may be sent to the controller when the model predicts a deviation in the catalyst deactivation rate from a predetermined value.

[0098] Operator and/or process engineer terminal 26 may be a computing device (such as a personal computer, industrial computer, server, laptop, or mobile computing device) or an appliance (such as a workstation (AWS)) . Terminal 26 of the operator and/or process engineer may be equipped with a user interface and configured to receive, process and transmit data.

[0099] In accordance with the invention, a method is also provided for predicting the change in the layered activity of the catalyst in the installation 11 GO DF based on prediction models, which are mathematical regression models, generated and trained, for example, using the prediction system 20, which is described above.

[0100] In accordance with the proposed method for predicting changes in the layered activity of the catalyst in the installation 11 GO DF in the block 21 for receiving parameter values are received, and in the block 22 for storing parameter values, the values of the parameters of the hydrotreatment process are stored in the form of one or more time series. The values of the parameters of the hydrotreatment process include data from laboratory analyzes of raw materials at the inlet of the unit and hydrotreated diesel fuel at the outlet of the unit and data from the sensors of the GO DT unit.

[0101] Next, filtering the values in the mentioned time series using Shewhart's control charts.

[0102] After that, based on the filtered time series, a training set is formed and the size of the moving training window is determined based on the standard deviation and the maximum of the correlation coefficient between the historical and predicted values.

[0103] Next, a temperature drop prediction model or models are generated for one or more catalyst beds based on an arbitrary-order polynomial regression model with adjacent terms, the regressors of which are the mentioned process parameters, including: the 95% feed point R3 at the inlet to the GO unit DF, density R4 of feedstock at the inlet to the DF DH plant, flow rate R5 of feedstock to the DH DH plant, flow rate R6 of make-up HSG to the DF DH plant, sulfur content R13 in hydrotreated diesel fuel, sulfur content R14 in the feedstock, pressure drop R15 between the inlet and outlet of the reactor , temperature difference R16 between the inlet and outlet of the reactor and the integral load F _sum .

[0104] After the model is formed, it is trained on the training set to obtain regression coefficients that provide indicators of the compliance of the predicted reactor inlet temperature with historical values above a predetermined level.

[0105] Next, using the prediction unit 25, a linear approximation of the temperature difference with reference to time is determined based on the generated model, the linearly approximated integral load, and the average values of the regressors over a predetermined period of time.

[0106] Thereafter, in predictor 25, the linear rate of the resulting approximation of the temperature drop can be determined and an estimate of the rate of deactivation of one or more catalyst beds can be determined. The resulting value is the rate of change in the temperature difference of the reactor layer by layer (for example, in °C/month), when a certain value of the temperature difference between the catalyst layers is reached, the layer is considered exhausted and has exhausted its resource, since exothermic processes have ceased to occur in it.

[0107] The present invention makes it possible to facilitate and provide support in decision-making by personnel when changing the parameters of technological processes, with the functions of online analysis of data arrays, calculation of the temperature in the reactor; increase the flexibility of planning and process control; to increase the loading of technological installations due to a more complete use of the resource of the catalytic system and better controllability of technological processes; increase the yield of more valuable products due to better process control and mode softening; reduce the time to involve the operating personnel of process units to perform manual calculations in order to analyze the state of catalytic systems; increase the yield of target products due to milder processing conditions; increase the amount of processed raw materials on one catalyst; reduce the operating costs of the catalyst; reduce the risk of unscheduled downtime to replace catalytic systems due to premature deactivation; increase the accuracy of planning, increase the amount of processed feedstock on the catalyst, or increase the weight of the processed feedstock due to operational forecasting of the remaining catalyst life.

[0108] The above description is intended solely to illustrate the proposed invention, and not to limit it. The described embodiments and their aspects can be used, for example, in combination with each other. In addition, the proposed invention may be modified to suit specific situations and conditions, and such modifications are also within the scope of protection. Many other embodiments will be apparent to those skilled in the art upon reading the present disclosure. Therefore, the scope of protection of the proposed invention is determined by the appended claims, along with all the equivalents to which such claims give the right.

[0109] For example, the division into blocks used in the present description is only a division according to logical functions. Thus, one or more functional blocks can be implemented, for example, in a single set of firmware (eg, a general purpose signal processor, microcontroller, random access memory, hard disk, etc.). In addition, several blocks or components can be combined or integrated into another system. Alternatively, some functions may be omitted or not performed. In addition, an interconnection or direct connection or communication connection shown or discussed may be an indirect connection or communication connection via interfaces, devices or blocks, or may also be an electrical, mechanical or other type of connection. The functionality of one or more units may be implemented, for example, by means of a computer-readable medium on which program code is recorded, when executed by the processor of the computing device, the computing device performs the corresponding functions. If the blocks are implemented as separate software and hardware systems, communication between them, as well as with sensors at the 11 GO DT unit, can be carried out using wireless communication tools, for example, through the Industrial Internet of Things (IIoT) system, technologies Wireless Local Area Network (WiFi), Long-Term Evolution (LTE), Broadcast Channels, Near Field Communication (NFC), Bluetooth (BT), and other wired and/or wireless technologies.

[0110] The blocks shown in figure 2 system 20 prediction can be implemented on the basis of an industrial computer or distributed network.

[0111] Used in the present description, the element set forth in the singular, should not be understood as excluding the plural of the mentioned elements, unless such an exception is explicitly stated. In addition, references to "one embodiment" of the proposed invention should not be interpreted as excluding the existence of additional embodiments that also include these distinctive features. Moreover, unless explicitly stated otherwise, embodiments of the invention "comprising" or "comprising" an element or a plurality of elements having a particular property may further include those elements that do not have that property.

Claims (54)

1. A system for predicting changes in the layer-by-layer activity of a catalyst in a diesel fuel hydrotreater (HD DF), including: a parameter value receiving unit configured to receive parameter values of the hydrotreatment process, wherein the parameter values of the hydrotreatment process include data from laboratory analyzes of raw materials at the inlet of the GO DT unit and hydrotreated diesel fuel at the outlet of said unit and data from the sensors of the GO DT unit; a parameter value storage unit configured to store the received hydrotreating process parameter values as one or more time series; a filtering unit configured to filter the stored time series values using Shewhart's control charts; block of formation and training of the model, made with the possibility of: determining the values of the integral load F _sum of the reactor, which reflects the amount of raw materials processed by the installation from the beginning of the use of the analyzed catalyst, formation of a training set based on the filtered values of the time series, determining the size of the moving learning window based on the standard deviation of the temperature drop in the catalyst bed and the maximum of the correlation coefficient between the historical and predicted values of the temperature drop in the catalyst bed, formation of a temperature drop model in the catalyst bed in the reactor of the GO DF unit, while the mentioned model is a polynomial regression model of arbitrary order with adjacent terms, the regressors of which are the mentioned process parameters, including: R4 of feedstock at the inlet to the DF HE unit, consumption R5 of feedstock to the DT DH unit, flow R6 of make-up HSG to the DF HT unit, R13 content of sulfur in hydrotreated diesel fuel, R14 content of sulfur in the feedstock, pressure drop between the reactor inlet and outlet, temperature difference between the inlet and outlet of the reactor, the integral load F _sum , and training the model on the training set with obtaining regression coefficients that provide indicators of compliance of the simulated temperature drop in the catalyst layer with retrospective values above a predetermined level; prediction unit configured to: determining the rate of change of the linear approximation of the temperature difference in the catalyst bed using the above model based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; estimating the activity of the catalyst bed based on the rate of change of said linear approximation of temperature differences. 2. The system of claim. 1, in which the block receiving parameters is configured to receive new values of the parameters of the hydrotreating process, and the block of formation and training of the model is configured to automatically retrain the models based on the above-mentioned obtained new values of the parameters of the hydrotreating process. 3. A system according to any one of the preceding claims, wherein the predictor is configured to transmit the resulting estimate of catalyst bed activity to an operator and/or process engineer terminal. 4. The system according to any one of the preceding claims, in which the prediction unit is configured to transmit the obtained assessment of the activity of the catalyst layer to the control unit of the GO DT unit. 5. The system according to any one of the preceding claims, wherein said determination of the values of the integral load F _sum of the reactor is carried out according to the formula:

where t _i - i-th point in time, F(t) - load at time t. 6. The system according to any of the preceding claims, wherein said construction of a linear approximation of the integral load F _sum is carried out according to the formula:

where t is time, a and b are linear coefficients found by the least squares method. 7. System for predicting changes in the layer-by-layer activity of the catalyst in the diesel fuel hydrotreatment unit (HD DF), including: a block for receiving values of the parameters of the hydrotreating process, configured to receive values of the parameters of the hydrotreating process, including data from laboratory analyzes of raw materials at the inlet of the GO DT unit and hydrotreated diesel fuel at the outlet of the said unit and data from the sensors of the GO DT unit; a parameter value storage unit configured to store the received parameter values of the hydrotreating process; prediction unit configured to: determining the rate of change of the linear approximation of the temperature difference in the catalyst bed using the trained polynomial regression model with adjacent terms based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; evaluating the activity of the catalyst bed based on the rate of change of said linear approximation of the temperature difference across the catalyst bed; at the same time, the regressors of the mentioned model are the mentioned parameters of the hydrotreatment process, including: the point R3 of boiling point of 95% of the feedstock at the inlet to the DT DT unit, the density R4 of the feedstock at the inlet to the DT DT unit, the flow rate R5 of the feedstock to the DT DT unit, the flow rate R6 of make-up HSG to the unit GO DF, R13 sulfur content in hydrotreated DF, R14 sulfur content in feedstock, pressure drop between the reactor inlet and outlet, temperature difference between the reactor inlet and outlet, integral loading F _sum , reflecting the amount of feedstock processed by the unit from the beginning of the use of the analyzed catalyst. 8. A method for predicting changes in the layer-by-layer activity of a catalyst in a diesel fuel hydrotreater (HD DF), including: receiving and storing the values of the parameters of the hydrotreating process in the form of one or more time series, while the values of the parameters of the hydrotreating process include data from laboratory analyzes of raw materials at the inlet of the GO DT unit and hydrotreated diesel fuel at the outlet of the said unit and data from the sensors of the GO DT unit; filtering time series values using Shewhart control charts; determining the values of the integral load F _sum of the reactor, reflecting the amount of raw materials processed by the GO DT unit from the beginning of the use of the analyzed catalyst; formation of a training set based on filtered time series; determining the size of the moving learning window based on the standard deviation of the temperature difference in the catalyst bed and the maximum of the correlation coefficient between the historical and predicted values of the temperature difference in the catalyst bed; formation of a temperature drop model in the catalyst bed in the reactor of the GO DF unit, while the mentioned model is a polynomial regression model of arbitrary order with adjacent terms, the regressors of which are the mentioned process parameters, including: R4 of feedstock at the inlet to the DF HE unit, consumption R5 of feedstock to the DT DH unit, flow R6 of make-up HSG to the DF HT unit, R13 content of sulfur in hydrotreated diesel fuel, R14 content of sulfur in the feedstock, pressure drop between the reactor inlet and outlet, temperature difference between input and output of the reactor, the integral load F _sum ; training the model on the training set with obtaining regression coefficients that provide indicators of compliance of the simulated temperature drop in the catalyst layer with retrospective values above a predetermined level; determining the rate of change of the linear approximation of the temperature difference in the catalyst bed using the above model based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; evaluating the activity of the catalyst bed based on the rate of change of said linear approximation of temperature differences. 9. The method of claim 8, which further comprises the step of obtaining new hydrotreating process parameter values and automatically retraining the model based on said new hydrotreating process parameter values. 10. The method according to any one of paragraphs. 8, 9, which further includes the step of transmitting the obtained catalyst bed activity estimate to the terminal of the operator and/or process engineer. 11. The method according to any one of paragraphs. 8-10, which further includes the step of transmitting the resulting catalyst bed activity estimate to the plant control unit. 12. The method according to any one of paragraphs. 8-11, in which the said determination of the values of the integral load F _sum of the reactor is carried out according to the formula:

where t _i - i-th point in time, F(t) - load at time t. 13. The method according to any one of paragraphs. 8-12, in which the said construction of a linear approximation of the integral load F _sum is carried out according to the formula:

where t is time, a and b are linear coefficients found by the least squares method. 14. A method for predicting changes in the layer-by-layer activity of a catalyst in a diesel fuel hydrotreater (HD DF), including: receiving and storing the values of the parameters of the hydrotreating process in the form of one or more time series, while the values of the parameters of the hydrotreating process include data from laboratory analyzes of raw materials at the inlet of the GO DT unit and hydrotreated diesel fuel at the outlet of the said unit and data from the sensors of the GO DT unit; determining the rate of change of the linear approximation of the temperature difference in the catalyst bed using the trained polynomial regression model with adjacent terms based on the average values of the regressors over a predetermined period of time and linearly approximated values of the integral load F _sum ; evaluating the activity of the catalyst bed based on the rate of change of said linear approximation of the temperature difference across the catalyst bed; at the same time, the regressors of the mentioned model are the mentioned parameters of the hydrotreatment process, including: the point R3 of boiling point of 95% of the feedstock at the inlet to the DT DT unit, the density R4 of the feedstock at the inlet to the DT DT unit, the flow rate R5 of the feedstock to the DT DT unit, the flow rate R6 of make-up HSG to the unit GO DF, R13 sulfur content in hydrotreated DF, R14 sulfur content in feedstock, pressure drop between the reactor inlet and outlet, temperature difference between the reactor inlet and outlet, integral loading F _sum , reflecting the amount of feedstock processed by the unit from the beginning of the use of the analyzed catalyst. 15. A computer-readable medium that stores a software product with program instructions, the execution of which by the processor ensures the execution of the method according to any one of paragraphs. 8-14.

Images