A Method and System for Coking Sensing and Prediction of Furnace Tubes in a Tubular Cracking Furnace

Abstract

The present invention discloses a method and system for sensing and predicting furnace tube coking of a tubular cracking furnace, including: constructing a digital twin of the cracking furnace; performing interaction on the virtual-real space data between the cracking furnace and the digital twin of the cracking furnace; based on the interacted virtual-real space data, combining deep learning of multi-source data, reconstructing the spatio-temporal dynamic change process of the temperature field, pressure field and flow field inside the furnace tubes of the cracking furnace, and realizing the prediction of the best timing for furnace tube decoking or burning to coke for furnace tube coking statistical inference and trend prediction. The present invention is of great significance for ensuring the safety of petrochemical production.

Description

Technical Field

[0001] The present invention belongs to the technical fields of safe and stable operation of petrochemical production equipment, process safety control, information engineering, data diagnosis, etc., and specifically relates to a method and system for sensing and predicting coking of furnace tubes in a tubular cracking furnace. Background Art

[0002] The ethylene cracking furnace is an important component of the ethylene plant and plays a leading role in ethylene and even the entire petrochemical production. The main function of the cracking furnace is to process raw materials into cracked gas and transport it through pipelines to subsequent production equipment in the process flow for processing into ethylene, propylene, and various by-products. The production capacity and technical level of the cracking furnace determine the production scale, production capacity, and product quality of the ethylene plant.

[0003] As is well known, the cracking device has many potential accident hazards, high safety risks, and is difficult to prevent and control online. Among them, the most typical ones are abnormal furnace conditions (such as burner flashback / flameout, flame detachment, etc.) and abnormal furnace tube conditions (such as coking, blockage, cracks, damage, deformation, etc.). Whether it is the furnace chamber or the furnace tubes, once an abnormality or a failure occurs, it will affect production at least, and may even induce catastrophic accidents in severe cases. In view of this, it is necessary and meaningful to establish a sensing method for safety hazards such as abnormal furnace chamber flames, local coking of furnace tubes, and corrosion damage of pipelines around the ethylene plant, especially the cracking furnace.

[0004] The cracking furnace is the core equipment of the ethylene cracking device and is also a disaster-prone area for potential hazards to induce catastrophes. Taking the tubular cracking furnace commonly used by domestic and foreign enterprises as an example, after the ethylene production raw materials are preheated to about 6000°C in the convection chamber furnace tubes of the cracking furnace, they quickly flow through the furnace tubes in the furnace chamber, and the materials are quickly heated to about 8000°C during the process. Chemical reactions occur during this period to complete the cracking process of breaking large molecules into small molecules. During the cracking process of hydrocarbons, due to the occurrence of secondary reactions such as polymerization and condensation, it is inevitable to form coke or carbon deposit and accumulate on the inner wall of the furnace tubes to form coking. And as the operation cycle lengthens, the cracking depth deepens, and the production materials become heavier, the degree of coking will become more and more serious, and even block the furnace tubes, affecting the safety of the device. Currently, the typical method is to carry out coke burning or decoking according to the change of the furnace tube outlet temperature COT, or to regularly carry out online coke burning and shutdown decoking. The above practices have obvious limitations: one is that the coking of furnace tubes is complex and diverse, and COT is mainly related to the coking near the tube mouth and is difficult to reflect the coking situation of the entire furnace tube; the other is that premature or late decoking, too long or too short coke burning time will not only affect production but also damage the furnace tubes or accelerate the aging of the furnace tubes.

[0005] Coking is an important factor affecting the production safety of ethylene plants. How to accurately perceive the coking situation and reasonably arrange the coking removal method and timing is a technical problem that has long troubled the "stable, long-term, full-load, high-quality" production. Considering that the coking part of the furnace tube is not easy to conduct heat, the cracking efficiency decreases, local heat release appears in the furnace tube, and even burns through; moreover, as the amount of coke accumulated in the furnace tube increases, it will block the pipeline and affect production safety. In case of serious coking, it is easy to cause furnace tube blockage and even pipe explosion accidents. The current methods used by enterprises are mainly based on the pressure difference between the inlet and outlet of the furnace tube, the temperature of the pipe mouth and the appearance of the pipe surface, which have obvious limitations. In view of this, the present invention establishes a full-furnace-tube coking perception and coking removal or burning timing auxiliary decision-making method starting from key parameters such as cracking temperature, reaction pressure, and residence time. Summary of the Invention

[0006] In view of the deficiencies of the prior art, the present invention proposes a furnace tube coking perception and prediction method and system for tubular cracking furnaces, which is of great significance for ensuring petrochemical production safety.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A furnace tube coking perception and prediction method for a tubular cracking furnace includes the following steps:

[0009] Construct a digital twin of the cracking furnace;

[0010] Interact the virtual-real space data between the cracking furnace and the digital twin of the cracking furnace;

[0011] Based on the interacted virtual-real space data, combined with deep learning of multi-source data, reconstruct the spatio-temporal dynamic change process of the temperature field, pressure field and flow field in the furnace tube, and realize the prediction of the best timing of furnace tube coking removal or burning for coking statistical inference and trend prediction.

[0012] Preferably, the method for constructing a digital twin of the cracking furnace includes:

[0013] Taking the cracking device as the core, construct a digital twin conceptual model around the physical entity PE, virtual entity VE, digital twin data DD, safety service Ss for orientation and the connection CN between entities, and integrate information data and physical data;

[0014] Based on the integrated information data and physical data, establish a digital twin system reference architecture in a bottom-up manner;

[0015] Based on the digital twin system reference architecture, establish a digital model or information transfer model of PE, build a two-way data communication between VE and entities, form a digital shadow model of the ethylene plant process, and then form a complete digital twin model to construct a digital twin of the cracking process.

[0016] Preferably, the method for interacting virtual-real space data between the cracking furnace and the digital twin of the cracking furnace includes:

[0017] Based on the 3D virtual simulation of the ethylene cracking system, combined with the digital twin model, the digital main line dredges the data generation, exchange, and transfer relationships of each link in the ethylene process, realizes the dynamic reproduction and deduction of the production process and safety risks in the virtual environment, as well as the seamless flow and process traceability of the data of each production process.

[0018] Preferably, the method for reconstructing the spatio-temporal dynamic change process of the temperature field, pressure field, and flow field in the cracking furnace tubes includes:

[0019] Construct an extensible model set S for the temperature, pressure, flow rate changes, and material cracking reactions in the convection section of the cracking furnace;

[0020] Based on the extensible model set S, through numerical simulation, quantitatively analyze the coupling relationship between the changes in the feed characteristic quantities of the furnace tubes, the changes in the internal and external environments of the furnace tubes, and the changes in the characteristic quantities of the materials at the outlet of the convection section;

[0021] Use a 3D grid to divide the convection section of the cracking furnace into grids D, and based on the extensible model set S, calculate the temperature, pressure, flow velocity, and flow rate data of each node in the grid D;

[0022] Utilize the coupling relationship between the changes in the feed characteristic quantities of the furnace tubes, the changes in the internal and external environments of the furnace tubes, and the changes in the characteristic quantities of the materials at the outlet of the convection section, combined with the temperature, pressure, flow velocity, and flow rate data of each node in the grid, to train a multi-layer convolutional neural network to obtain a deep learning model;

[0023] Use a field description container to describe the temperature, pressure, flow velocity, and radiation characteristic attributes at any point in the transfer pipeline, and obtain the simulation data of the temperature, pressure, flow velocity, and radiation characteristic attributes of the tensor field at the grid nodes;

[0024] Use the digital twin spatio-temporal deduction simulation data to obtain a tensor field function fitting model based on deep learning, and gradient, divergence, and curl function models;

[0025] Combine the temperature, pressure, flow velocity, and radiation data collected on-site in the ethylene plant to update the tensor field function fitting model based on deep learning and the gradient, divergence, and curl function models; combine the updated models and the spatio-temporal interpolation method to obtain the soft measurement data of the temperature, pressure, flow velocity, and radiation characteristic attributes at any point in the ethylene cracking furnace tubes changing with time and space.

[0026] Preferably, the method for predicting the best timing of furnace tube decoking or burning for furnace tube coking statistical inference and trend prediction includes:

[0027] Adopt the method of using online measured data to drive the digital twin, and visually display the coking parts, coking degree and change process of the furnace tubes;

[0028] Utilize the historical data of the actual production process of multiple groups of cracking furnaces of the same type to reveal the basic characteristics and mutual influence laws of the flow, heat transfer, mass transfer and cracking reaction processes in the cracking furnace tubes;

[0029] Adopt the method of combining data mining analysis and machine learning of the cracking furnace to construct a causal relationship tree with coking and creep as the core, as well as a multi-factor regression analysis model for the coking thickness of the furnace tubes and a statistical inference model for the coking parts and thickness of the furnace tubes based on local vortex changes in the temperature field and pressure field;

[0030] Adopt the measured working condition data and the deep learning models of the temperature field and pressure field to calibrate and soft-sense the temperature field and pressure field inside the furnace tubes, and realize the online prediction of the time-space changes of the temperature field and pressure field;

[0031] Adopt the prediction data-driven digital twin system, combine the multi-factor regression analysis model of the coking thickness and the statistical inference model of the coking parts to realize the prediction of the coking thickness and parts of the furnace tubes; combine the knowledge of domain experts to determine the optimal coking disposal countermeasures and the best disposal timing, and provide auxiliary decision-making for coking disposal.

[0032] The present invention also provides a furnace tube coking perception and prediction system for a tubular cracking furnace, including: a construction module, an interaction module and a prediction module;

[0033] The construction module is used to construct a digital twin of the cracking furnace;

[0034] The interaction module is used to interact with the virtual-real space data between the cracking furnace and the digital twin of the cracking furnace;

[0035] The prediction module is used to reconstruct the time-space dynamic change process of the temperature field, pressure field and flow field inside the cracking furnace tubes based on the interacted virtual-real space data, and combine with multi-source data deep learning to realize the prediction of the best timing for furnace tube decoking or burning for coking statistical inference and trend prediction.

[0036] Preferably, the construction module includes: a data integration unit, a reference architecture construction unit and a digital twin construction unit;

[0037] The data integration unit is used to take the cracking device as the core, and build a digital twin conceptual model around the physical entity PE, virtual entity VE, digital twin data DD, safety service Ss for facing and the connection CN between entities, and integrate information data and physical data;

[0038] The reference architecture construction unit is used to establish a digital twin system reference architecture in a bottom-up manner based on the integrated information data and physical data;

[0039] The digital twin construction unit is used to establish a digital model or information transfer model of the PE based on the digital twin system reference architecture, build a two-way data communication between the VE and the entity, form a digital shadow model of the ethylene plant process, and then form a complete digital twin model to construct a digital twin of the cracking process.

[0040] Preferably, the process of interacting the virtual-real space data between the cracking furnace and the digital twin of the cracking furnace includes:

[0041] Based on the 3D virtual simulation of the ethylene cracking system and combined with the digital twin model, the digital main line dredges the data generation, exchange and transfer relationships of each link in the ethylene process, realizes the dynamic reproduction and deduction of the production process and safety risks in the virtual environment, and the seamless flow and process traceability of the data of each production process.

[0042] Preferably, the prediction module includes: an expandable model set S construction unit, a coupling relationship analysis unit, a change amount calculation unit, a deep learning model construction unit, a grid node simulation data calculation unit, a time-space vector field diagram drawing unit, and a soft measurement data calculation unit;

[0043] The expandable model set S construction unit is used to construct an expandable model set S for the temperature, pressure, flow rate changes and material cracking reactions in the convection chamber of the cracking furnace;

[0044] The coupling relationship analysis unit is used to quantitatively analyze the coupling relationship between the changes in the inlet characteristics of the furnace tubes, the changes in the internal and external environments of the furnace tubes, and the changes in the outlet material characteristics of the convection chamber based on the expandable model set S through numerical simulation;

[0045] The change amount calculation unit is used to divide the grid D of the convection chamber of the cracking furnace by using a 3D grid, and calculate the temperature, pressure, flow velocity, and flow rate data of each node of the grid D based on the expandable model set S;

[0046] The deep learning model construction unit is used to train a multi-layer convolutional neural network by using the coupling relationship between the changes in the inlet characteristics of the furnace tubes, the changes in the internal and external environments of the furnace tubes, and the changes in the outlet material characteristics of the convection chamber, combined with the temperature, pressure, flow velocity, and flow rate data of each node of the grid, to obtain a deep learning model;

[0047] The grid node simulation data calculation unit is used to obtain the simulation data of the temperature, pressure, flow velocity, and radiation characteristic attributes of the tensor field at the grid nodes by using the field description container and the temperature, pressure, flow velocity, and radiation characteristic attributes of any point in the transmission pipeline;

[0048] The spatio-temporal vector field diagram plotting unit is used to obtain a tensor field function fitting model based on deep learning, as well as gradient, divergence, and curl function models, using digital twin spatio-temporal deduction simulation data;

[0049] The soft measurement data calculation unit is used to update the tensor field function fitting model based on deep learning and the gradient, divergence, and curl function models by combining the temperature, pressure, flow rate, and radiation data collected on-site in the ethylene plant; and obtain the soft measurement data of the temperature, pressure, flow rate, and radiation characteristic attributes of any point in the ethylene cracking furnace tube changing with time and space by combining the updated model and the spatio-temporal interpolation method.

[0050] Preferably, the prediction module further includes: a visualization display unit, a law revelation unit, an inference model construction unit, an online prediction unit, and a decision-making unit;

[0051] The visualization display unit is used to visually display the coking part, coking degree, and change process of the furnace tube by using the method of driving the digital twin with online measured data;

[0052] The law revelation unit is used to reveal the basic characteristics and mutual influence laws of the flow, heat transfer, mass transfer, and cracking reaction processes in the cracking furnace tube by using the historical data of the actual production processes of multiple groups of cracking furnaces of the same model;

[0053] The inference model construction unit is used to construct a causal relationship tree with coking and creep as the core, as well as a multi-factor regression analysis model of the coking thickness of the furnace tube and a statistical inference model of the coking part and thickness of the furnace tube based on local vorticity changes in the temperature field and pressure field by using the method of combining cracking furnace data mining analysis and machine learning;

[0054] The online prediction unit is used to calibrate and soft measure the internal temperature field and pressure field of the furnace tube by using the measured working condition data and the deep learning models of the temperature field and pressure field, and realize the online prediction of the spatio-temporal changes of the temperature field and pressure field;

[0055] The decision-making unit is used to drive the digital twin system with the prediction data, combine the multi-factor regression analysis model of the coking thickness and the statistical inference model of the coking part, and realize the prediction of the coking thickness and part of the furnace tube; combine the knowledge of domain experts to determine the optimal coking disposal countermeasures and the best disposal timing, and provide auxiliary decision-making for coking disposal.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] This application has invented a set of methods for furnace tube coking perception, coking change trend prediction, and condition-based maintenance auxiliary decision-making with strong pertinence and practicability. By reasonably constructing a digital twin of the cracking furnace and realizing the virtual-real space data interaction between the cracking furnace and the twin, combined with deep learning of multi-source data, the spatio-temporal dynamic change process of the temperature field, pressure field, and flow field inside the cracking furnace tubes is reconstructed, so as to realize the statistical inference of the coking situation of the furnace tubes, trend prediction, and prediction of the best timing for furnace tube decoking or burning. The method is original, novel, and advanced, and is of great significance for ensuring petrochemical production safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0059] Figure 1 Schematic diagram of the digital twin construction and virtual-real space data interaction model for the embodiments of the present invention;

[0060] Figure 2 Schematic diagram of the field reconstruction and spatio-temporal soft measurement process for the embodiments of the present invention;

[0061] Figure 3 Schematic diagram of the furnace tube coking prediction, early warning, and auxiliary decision-making process for the embodiments of the present invention;

[0062] Figure 4 Schematic diagram of a method for furnace tube coking perception and prediction of a tubular cracking furnace for the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0064] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0065] Embodiment 1

[0066] As Figure 4As shown in the figure, the present invention focuses on the key problem of furnace tube coking, which affects production safety. Starting from the construction of the digital twin of the cracking furnace, through virtual-real and data-driven approaches, it realizes the auxiliary decision-making of coking perception and predictive maintenance in four steps.

[0067] Step 1: Construct the digital twin of the cracking furnace and realize the data interaction between the virtual and real spaces. The specific construction method is as follows:

[0068] S1-1. Taking the cracking unit as the core, starting from the cracking unit and its process flow, around its physical entity PE, virtual entity VE, digital twin data DD, safety service-oriented Ss, and the connection CN between entities, construct its digital twin conceptual model (PE, VE, Ss, DD, CN), integrate information data and physical data, and provide data support for the whole elements / whole process;

[0069] S1-2. Adopting a bottom-up approach, establish a reference architecture of the digital twin system with five-layer structures including the real physical domain, measurement and control entities, digital twin, user domain, and cross-domain functional entities; on the basis of abstracting and modeling the ethylene unit, process, and safety risks, establish a digital model or information transfer model of PE, build a two-way data communication between VE and the entity, form a digital shadow model of the ethylene unit process, and then form a complete digital twin model; among them, the digital shadow model of the ethylene unit process is actually a model that can run on a digital computer, which is the mapping of the PE entity model in the VE environment, that is, the model form of VE. The complete digital twin model is the digital representation of the entity mechanism model and also the relationship mapping of the entity essence in the VE environment.

[0070] S1-3. Based on the 3D virtual simulation of the ethylene cracking system, combined with the digital twin model, use the digital main line to dredge the data generation, exchange, and transfer relationships of each link in the ethylene process, form a digital twin model of the high-temperature cracking process in the tube and a model-based virtual environment, ensure its digital coexistence, and realize the real-time data interaction between the real space and the virtual space of the cracking process. Among them, the construction of the digital twin and the data interaction model between the virtual and real spaces are as Figure 1 shown, realizing the dynamic reproduction and deduction of the production process and safety risks in the virtual environment, as well as the seamless flow of data between each production process and the process traceability. Among them, the digital simulation of the ethylene process in the VE environment, the data of each link in the dynamic system simulation process are generated by the digital twin model, the data exchange and transfer relationship between each link of VE is the same as the data exchange process between each unit in the PE environment, but it is more concise, just the data transfer in the digital computer simulation environment.

[0071] Step 2: Reconstruct the temperature field, pressure field, and flow field of the time-space change in the cracking furnace tube through data driving. The specific construction method is as follows:

[0072] S2-1. Based on the historical data of the cracking furnace, especially the measured historical data of different measuring points during the in-service process of furnace tubes with different coking conditions, fuse the soft sensor of spatio-temporal variation characteristics and the multi-layer convolutional neural network to reconstruct the spatio-temporal evolution models of the temperature field, pressure field and flow field inside the tubes; use classical models such as the one-way pressure drop and heat transfer in the convection section of the ethylene cracking furnace, the heat transfer of the flue gas through the bare tubes outside the tubes, the temperature change and pressure transfer models inside the tubes, the two-phase flow pressure drop model inside the tubes, the two-phase convective heat transfer model inside the tubes, the flue gas pressure drop model outside the tubes, the heat transfer rate in the radiation chamber, the heat balance equation and the material cracking reaction process model inside the tubes, etc. to construct an expandable model set S for the temperature, pressure, flow rate change and material cracking reaction in the convection chamber of the cracking furnace; among them, the expandable model set, that is, the basic model set formed by using the above models, can expand the model set online.

[0073] S2-2. Based on the expandable model set S, through numerical simulation, quantitatively analyze the coupling relationships between different sources and different elements such as the changes in the feed characteristic quantities (such as temperature, pressure, flow velocity, flow rate, etc.) of the furnace tubes, the changes in the internal and external environments of the furnace tubes, and the changes in the outlet material characteristic quantities (such as temperature, pressure, flow velocity, flow rate, etc.) of the convection chamber; among them, to quantitatively analyze the coupling relationships between various elements, multivariate analysis, variance analysis and correlation analysis, etc. can be used and selected according to the actual situation.

[0074] S2-3. Use a 3D grid to divide the convection chamber of the cracking furnace into grids D. Based on the above expandable model set S, use the mechanism model to calculate the temperature change amount, the three direction components of the pressure vector, the three direction components of the flow velocity vector, the flow rate change amount, etc. of each node D(x, y, z; t) in the grid D over time.

[0075] S2-4. Use a large amount of different feed data of the furnace tubes in S2-2, different internal and external environment data, and the outlet material data obtained from simulation calculations, and combine the temperature, pressure, flow velocity, flow rate, etc. data of each node D(x, y, z; t) in the grid in S2-3 to train the multi-layer convolutional neural network to obtain a deep learning model (CNN), denoted as CNNa.

[0076] S2-5. Considering that the characteristic variables such as the temperature, pressure, flow velocity and radiation of the material inside the cracking furnace tubes change with time, space and the technological process, use the field to describe the temperature, pressure, flow velocity, radiation and other characteristic attributes at any point in the container and the transmission pipeline, take the multi-order tensor field composed of the spatio-temporal changes and their attributes inside the convection chamber and the transmission pipeline as the object, and combine the 3D simulation data-driven digital twin to obtain the simulation data of the temperature, pressure, flow velocity, radiation and other characteristic attributes of the tensor field at the grid nodes (the data of the tensor field changing with time); the field reconstruction and spatio-temporal soft sensor process are as Figure 2 shown.

[0077] S2-6. Use the spatio-temporal deduction data of the digital twin to train the multi-dimensional DeepAR and Spacetimeformer, obtain a tensor field function fitting model based on deep learning, and its gradient, divergence, and curl function models, and respectively draw the time-space vector field diagrams of temperature, pressure, flow rate, and radiation;

[0078] S2-7. Use the historical data of temperature, pressure, flow rate, radiation, etc. in the cracking furnace data warehouse to train and perform transfer learning on the models obtained in S2-4 and S2-6; combine the data of temperature, pressure, flow rate, radiation, etc. collected on-site in the ethylene plant to update the tensor field deep learning model and the gradient, divergence, and curl fitting models; combine the updated models and the time-space interpolation method to obtain the soft measurement data of the characteristic attributes such as temperature, pressure, flow rate, radiation, etc. at any point in the ethylene cracking furnace tube changing with time and space.

[0079] Step 3: Infer the coking of the furnace tubes and predict the best timing for furnace tube decoking or burning. The inference and auxiliary decision-making methods are as follows:

[0080] S3-1. Start from the reconstruction of the in-tube temperature field and pressure field based on measured data, combine the digital twin system, and use the method of driving the digital twin with online measured data to visually display the coking location, coking degree, and change process of the furnace tubes;

[0081] S3-2. Use the historical data of the actual production process of multiple groups of cracking furnaces of the same model to train, test, verify, and iteratively optimize the deep learning model CNNa to obtain a more accurate neural network model CNNb; compare and analyze the changes in the simulation results and measured data to obtain an understanding of the distribution of material velocity, temperature, pressure, and reaction rate in the cracking furnace, and reveal the basic characteristics and mutual influence laws of the flow, heat transfer, mass transfer, and cracking reaction processes in the cracking furnace tubes;

[0082] S3-3. Adopt the method of combining cracking furnace data mining analysis and machine learning to construct a causal relationship tree with coking and creep as the core, as well as a multi-factor regression analysis model for the coking thickness of the furnace tubes and a statistical inference model for the coking location and thickness of the furnace tubes based on local vorticity changes in the temperature field and pressure field;

[0083] S3-4. Use the measured working condition data and the deep learning models (CNNb) of the temperature field and pressure field to calibrate and perform soft measurement on the in-tube temperature field and pressure field of the furnace tubes; use the soft measurement data of the temperature and pressure changes and the recursive neural network (DA-RNN) based on the dual-stage attention mechanism to achieve online prediction of the time-space changes of the temperature field and pressure field;

[0084] S3-5. Adopt a prediction data-driven digital twin system, combine a multi-factor regression analysis model of coking thickness and a statistical inference model of coking location, perform high-fidelity adaptive fault-tolerant exponential smoothing prediction through characteristic parameters, and conduct decision-level fusion of multiple prediction modes to achieve the prediction of the coking thickness and location of furnace tubes; combine domain expert knowledge to determine the optimal coking disposal countermeasures and the best disposal timing, where the countermeasures are to select on-line coke blowing, burning or furnace shutdown for coke cleaning according to the coking situation; the best treatment timing is the most appropriate time; provide auxiliary decision-making for coking disposal. The specific prediction and early warning and auxiliary decision-making process is as Figure 3 shown.

[0085] Step 4: Test and verification. Adopt the method of replaying historical receipts afterwards to conduct experiments on instrument data security monitoring and cause tracing, and verify the effectiveness and security of the monitoring method and tracing algorithm. The specific test and verification process is as follows:

[0086] S4-1. Use specific data and experimental results, combine fault cases and domain expert knowledge to analyze and verify the reasons for coking abnormalities in the cracking furnace; on this basis, evaluate the performance and effect of the monitoring algorithm, and through feedback and iteration, achieve the improvement and optimization of the scheme, method and adjustable parameters.

[0087] S4-2. Utilize the opportunities such as furnace shutdown and coke burning of the cracking furnace to collect the actual data inside the furnace and in the tubes, combine the monitoring data before furnace shutdown, conduct experimental verification on the method of equipment local hidden danger safety perception and prediction and early warning, compare with the actual measurement results, verify the coincidence degree with the inference results based on the tensor field, and optimize the tensor field model and improve the trend prediction algorithm through evaluation and iteration.

[0088] S4-3. Connect the laboratory device and the software platform, read the DCS data, and conduct tests and verifications of the physical-in-the-loop based on the Matlab / Simulink simulation environment and experimental platform, and comprehensively evaluate and examine the process safety.

[0089] Embodiment 2

[0090] The present invention also provides a coking perception and prediction system for furnace tubes of a tubular cracking furnace, including: a construction module, an interaction module and a prediction module;

[0091] The construction module is used to construct a digital twin of the cracking furnace;

[0092] The interaction module is used to interact the virtual-real space data between the cracking furnace and the digital twin of the cracking furnace;

[0093] The prediction module is used to reconstruct the spatio-temporal dynamic change process of the temperature field, pressure field and flow field in the furnace tubes of the cracking furnace based on the interacted virtual-real space data, combined with multi-source data deep learning, and achieve the prediction of the best timing for furnace tube coke cleaning or burning through coking statistical inference and trend prediction.

[0094] In this embodiment, the construction module includes: a data integration unit, a reference architecture construction unit, and a digital twin construction unit;

[0095] The data integration unit is used to take the cracking unit as the core, start from the cracking unit and the corresponding process flow, and build a digital twin conceptual model (PE, VE, Ss, DD, CN) around the physical entity PE, virtual entity VE, digital twin data DD, safety service Ss for facing, and the connection CN between entities, and integrate information data and physical data;

[0096] The reference architecture construction unit is used to establish a digital twin system reference architecture with a five-layer structure of the real physical domain, measurement and control entities, digital twin, user domain, and cross-domain functional entities in a bottom-up manner based on the integrated information data and physical data;

[0097] The digital twin construction unit is used to establish a digital model or information transfer model of PE based on the digital twin system reference architecture, on the basis of abstracting and modeling the ethylene unit, process, and safety risks, build a two-way data communication between VE and the entity, form a digital shadow model of the ethylene unit process, and then form a complete digital twin model, and construct a digital twin of the cracking process.

[0098] In this embodiment, the process of interacting the virtual-real space data between the cracking furnace and the digital twin of the cracking furnace includes:

[0099] Based on the 3D virtual simulation of the ethylene cracking system, combined with the digital twin model, the digital main line dredges the data generation, exchange, and flow relationship of each link in the ethylene process, realizes the dynamic reproduction and deduction of the production process and safety risks in the virtual environment, and the seamless flow and process traceability of the data of each production process.

[0100] In this embodiment, the prediction module includes: an expandable model set S construction unit, a coupling relationship analysis unit, a change amount calculation unit, a deep learning model construction unit, a grid node simulation data calculation unit, a time-space vector field map drawing unit, and a soft measurement data calculation unit;

[0101] The expandable model set S construction unit is used to construct an expandable model set S of the temperature, pressure, flow rate change, and material cracking reaction in the convection chamber of the cracking furnace by using the classical models of the single-phase pressure drop and heat transfer in the convection section of the ethylene cracking furnace, the heat transfer of the flue gas outside the tube, the temperature change and pressure transfer models inside the tube, the two-phase flow pressure drop model inside the tube, the two-phase convective heat transfer model inside the tube, the flue gas pressure drop model outside the tube, the heat transfer rate in the radiation chamber, the heat balance equation, and the material cracking reaction process model inside the tube;

[0102] The coupling relationship analysis unit is used to quantitatively analyze the coupling relationship among the changes in the furnace tube feed characteristic quantities, the changes in the internal and external environments of the furnace tubes, and the changes in the characteristic quantities of the materials at the convection chamber outlet based on the expandable model set S through numerical simulation;

[0103] The change quantity calculation unit is used to divide the grid of the convection chamber of the cracking furnace by using a 3D grid D, and calculate the temperature change quantity, the three-direction components of the pressure vector, the three-direction components of the flow velocity vector, and the flow rate change quantity of each node D(x, y, z; t) of the grid D over time based on the expandable model set S;

[0104] The deep learning model construction unit is used to train a multi-layer convolutional neural network by using the coupling relationship among the changes in the furnace tube feed characteristic quantities, the changes in the internal and external environments of the furnace tubes, and the changes in the characteristic quantities of the materials at the convection chamber outlet, and combine the temperature, pressure, flow velocity, and flow rate data of each node D(x, y, z; t) of the grid to obtain a deep learning model, denoted as CNNa;

[0105] The grid node simulation data calculation unit is used to describe the temperature, pressure, flow velocity, and radiation characteristic attributes of any point in the field description container and the transmission pipeline, take the multi-order tensor field composed of the spatio-temporal changes and attributes inside the convection chamber and the transmission pipeline as the object, and combine the 3D simulation data to drive the digital twin to obtain the simulation data of the temperature, pressure, flow velocity, and radiation characteristic attributes of the tensor field at the grid nodes;

[0106] The spatio-temporal vector field map drawing unit is used to train multi-dimensional DeepAR and Spacetimeformer by using the spatio-temporal deduction data of the digital twin, obtain a deep learning-based tensor field function fitting model, and gradient, divergence, and curl function models, and draw the spatio-temporal vector field maps of temperature, pressure, flow velocity, and radiation respectively;

[0107] The soft measurement data calculation unit is used to train and perform transfer learning on the model by using the temperature, pressure, flow velocity, and radiation historical data in the cracking furnace data warehouse; update the deep learning-based tensor field function fitting model and the gradient, divergence, and curl function models by combining the temperature, pressure, flow velocity, and radiation data collected on-site in the ethylene plant; and obtain the soft measurement data of the temperature, pressure, flow velocity, and radiation characteristic attributes of any point in the ethylene cracking furnace tube changing with time and space by combining the updated model and the spatio-temporal interpolation method.

[0108] In this embodiment, the prediction module further includes: a visualization display unit, a law revelation unit, an inference model construction unit, an online prediction unit, and a decision-making unit;

[0109] The visualization unit starts from reconstructing the in-tube temperature field and pressure field based on measured data, combines with the digital twin system, and uses the method of driving the digital twin body with online measured data to visually display the coking parts, coking degree and change process of the furnace tubes;

[0110] The law revelation unit is used to train, test, verify and iteratively optimize the deep learning model CNNa by using the historical data of the actual production process of multiple groups of cracking furnaces of the same model, and obtain a more accurate neural network model CNNb; compare and analyze the simulation results and the changes in measured data, and obtain an understanding of the distribution of material velocity, temperature, pressure and reaction rate in the cracking furnace, and reveal the basic characteristics and mutual influence laws of the flow, heat transfer, mass transfer and cracking reaction processes in the cracking furnace tubes;

[0111] The inference model construction unit is used to construct a causal relationship tree with coking and creep as the core, a multi-factor regression analysis model of the coking thickness of the furnace tubes, and a statistical inference model of the coking parts and thickness of the furnace tubes based on local vorticity changes in the temperature field and pressure field by using the method of combining data mining analysis and machine learning of the cracking furnace;

[0112] The online prediction unit is used to calibrate and soft-sense the in-tube temperature field and pressure field by using the measured working condition data and the deep learning models of the temperature field and pressure field; based on the soft-sensed data of temperature and pressure changes, use the recursive neural network DA-RNN based on the two-stage attention mechanism to realize the online prediction of the time-space changes of the temperature field and pressure field;

[0113] The decision-making unit is used to drive the digital twin system with prediction data, combine the multi-factor regression analysis model of the coking thickness and the statistical inference model of the coking parts, and through the high-fidelity adaptive fault-tolerant exponential smoothing prediction of the characteristic parameters, and conduct decision-level fusion of multiple prediction modes, to realize the prediction of the coking thickness and parts of the furnace tubes; combine with the knowledge of domain experts to determine the optimal coking disposal countermeasures and the best disposal time, and provide auxiliary decision-making for coking disposal.

[0114] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for sensing and predicting coking in a tubular cracking furnace, characterized in that: The following steps are involved: Build a digital twin of the cracking furnace; Interact virtual-real space data between the cracking furnace and its digital twin; Based on interactive virtual-real space data and combined with deep learning of multi-source data, the spatiotemporal dynamic changes of temperature, pressure, and flow fields in the cracking furnace tubes are reconstructed to achieve statistical inference and trend prediction of furnace tube coking and the optimal timing for furnace tube de-coking or burning. The method for reconstructing the spatiotemporal dynamic change process of the temperature field, pressure field and flow field in the cracking furnace tube includes: Using the classic model of one-way pressure drop and heat transfer in the convection section of the ethylene cracking furnace, the heat transfer model of the bare tube flue gas outside the tube, the temperature change and pressure transmission model inside the tube, the two-phase flow pressure drop model inside the tube, the two-phase convection heat transfer model inside the tube, the flue gas pressure drop model outside the tube, the heat transfer rate of the radiation chamber, the heat balance equation and the material cracking reaction process model inside the tube, an expandable model set S for the temperature, pressure and flow changes in the convection chamber of the cracking furnace and the material cracking reaction was constructed; Based on the scalable model set S, numerical simulation is used to quantitatively analyze the coupling relationship between the changes in the characteristic quantities of the furnace tube feed, the changes in the internal and external environments of the furnace tube, and the changes in the characteristic quantities of the material at the convection chamber outlet. The convection chamber of the cracking furnace is meshed using a 3D grid. Based on the scalable model set S, the temperature, pressure, velocity, and flow rate data of each grid node are calculated. By utilizing the coupling relationship between the changes in the characteristic quantities of the furnace tube feed, the changes in the internal and external environments of the furnace tube, and the changes in the characteristic quantities of the material at the convection chamber outlet, and combining the temperature, pressure, flow rate, and flow rate data of each grid node, a multi-layer convolutional neural network is trained to obtain a deep learning model. The field is used to describe the temperature, pressure, flow rate, and radiation characteristic attributes of any point in the container and transmission pipeline. The multi-order tensor field composed of the spatiotemporal changes and attributes inside the convection chamber and transmission pipeline is used as the object. The digital twin is driven by 3D simulation data to obtain simulation data of the temperature, pressure, flow rate, and radiation characteristic attributes of the tensor field at the grid nodes. Using digital twin space-time deduction simulation data, we obtain a tensor field function fitting model based on deep learning, as well as gradient, divergence, and curl function models; Combined with the temperature, pressure, flow rate, and radiation data collected on-site in the ethylene plant, the tensor field function fitting model based on deep learning and the gradient, divergence, and curl function models are updated; combined with the updated model and the time-space interpolation method, the soft measurement data of the temperature, pressure, flow rate, and radiation characteristic attributes of any point in the ethylene cracking furnace tube that change with time and space are obtained.

2. The method for sensing and predicting coking in a tubular cracking furnace according to claim 1, wherein: The method of building a digital twin of a cracking furnace includes: With the cracking device as the core, a digital twin conceptual model is constructed around the physical entity PE, virtual entity VE, digital twin data DD, security-oriented service Ss and entity connection CN to integrate information data and physical data; Based on integrated information data and physical data, a bottom-up approach is used to establish a digital twin system reference architecture; Based on the digital twin system reference architecture, a digital model or information transfer model of PE is established, and two-way data communication between VE and entities is established to form a digital shadow model of the ethylene plant process, and then a complete digital twin model is formed to construct a digital twin of the cracking process.

3. The method for sensing and predicting coking in a tubular cracking furnace according to claim 2, wherein: The method for interacting virtual-real space data between the cracking furnace and the cracking furnace digital twin includes: Based on the 3D virtual simulation of the ethylene cracking system and combined with the digital twin model, the digital main line clears the data generation, exchange and flow relationship in each link of the ethylene process, realizes the dynamic reproduction and deduction of the production process and safety risks in a virtual environment, as well as the seamless flow of data and process traceability of each production process.

4. A tube coking sensing and prediction system for a tubular cracking furnace, characterized in that: include: Building modules, interaction modules, and prediction modules; The building module is used to build a digital twin of a cracking furnace; The interaction module is used to interact with the virtual-real space data between the cracking furnace and the cracking furnace digital twin; The prediction module is used to reconstruct the spatiotemporal dynamic change process of the temperature field, pressure field and flow field in the cracking furnace tube based on interactive virtual-real space data and combined with multi-source data deep learning, so as to achieve the prediction of the optimal time for furnace tube coking or burning by statistical inference and trend prediction of furnace tube coking; The prediction module includes: an expandable model set S construction unit, a coupling relationship analysis unit, a change calculation unit, a deep learning model construction unit, a grid node simulation data calculation unit, a time-space vector field drawing unit and a soft measurement data calculation unit; The expandable model set S construction unit is used to construct an expandable model set S of the cracking furnace convection chamber temperature, pressure, flow changes and material cracking reaction using the one-way pressure drop and heat transfer model of the ethylene cracking furnace convection section and the classic model of light tube flue gas heat transfer outside the tube, the temperature change and pressure transmission model inside the tube, the two-phase flow pressure drop model inside the tube, the two-phase convection heat transfer model inside the tube, the flue gas pressure drop model outside the tube, the radiation chamber heat transfer rate, the heat balance equation and the material cracking reaction process model inside the tube; The coupling relationship analysis unit is used to quantitatively analyze the coupling relationship between the change of the characteristic quantity of the furnace tube feed, the change of the internal and external environment of the furnace tube, and the change of the characteristic quantity of the material at the convection chamber outlet through numerical simulation based on the expandable model set S; The change calculation unit is used to grid the cracking furnace convection chamber using a 3D grid, and calculate the temperature, pressure, flow rate and flow rate data of each grid node based on the expandable model set S; The deep learning model construction unit is used to utilize the coupling relationship between the change of the characteristic quantity of the furnace tube feed, the change of the internal and external environment of the furnace tube, and the change of the characteristic quantity of the material at the convection chamber outlet, combined with the temperature, pressure, flow rate, and flow rate data of each node of the grid, to train a multi-layer convolutional neural network to obtain a deep learning model; The grid node simulation data calculation unit is used to use fields to describe the temperature, pressure, flow rate, and radiation characteristic attributes of any point in the container and the transmission pipeline. The multi-order tensor field composed of the time-space changes and attributes inside the convection chamber and the transmission pipeline is used as the object. In combination with 3D simulation data, the digital twin is driven to obtain simulation data of the temperature, pressure, flow rate, and radiation characteristic attributes of the tensor field at the grid node; The time-space vector field map drawing unit is used to use the digital twin space-time deduction simulation data to obtain a tensor field function fitting model based on deep learning, as well as gradient, divergence and curl function models; The soft measurement data calculation unit is used to update the tensor field function fitting model based on deep learning and the gradient, divergence and curl function models in combination with the temperature, pressure, flow rate and radiation data collected on-site in the ethylene plant; and combines the updated model with the time-space interpolation method to obtain the soft measurement data of the temperature, pressure, flow rate and radiation characteristic attributes of any point in the ethylene cracking furnace tube that change with time and space.

5. A furnace tube coking sensing and prediction system for a tubular cracking furnace according to claim 4, characterized in that: The building modules include: a data integration unit, a reference architecture building unit and a digital twin building unit; The data integration unit is used to build a digital twin concept model with the cracking device as the core, centering around the physical entity PE, the virtual entity VE, the digital twin data DD, the security service Ss and the entity connection CN, and integrate information data and physical data; The reference architecture construction unit is used to establish a digital twin system reference architecture based on integrated information data and physical data in a bottom-up manner; The digital twin construction unit is used to establish a digital model or information transfer model of PE based on the digital twin system reference architecture, build two-way data communication between VE and entities, form a digital shadow model of the ethylene unit process, and then form a complete digital twin model to construct a digital twin of the cracking process.

6. A tube coking sensing and prediction system for a tubular cracking furnace according to claim 5, characterized in that: The process of interacting virtual-real space data between the cracking furnace and its digital twin includes: Based on the 3D virtual simulation of the ethylene cracking system and combined with the digital twin model, the digital main line clears the data generation, exchange and flow relationship in each link of the ethylene process, realizes the dynamic reproduction and deduction of the production process and safety risks in a virtual environment, as well as the seamless flow of data and process traceability of each production process.

Images