A four-dimensional coupling regulation method for heavy non-ferrous metal smelting furnace structure, furnace three fields, furnace body thermal stress and injection system

Abstract

The application provides a four-dimensional coupling regulation method for a heavy non-ferrous metal bath smelting furnace structure-in-furnace three fields-furnace body thermal stress-injection system, which comprises collecting multi-source data in a smelting process, constructing a physical-data fusion model to predict key state variables; establishing a furnace structure, in-furnace temperature-flow field-concentration field, furnace body thermal stress and injection gas-liquid two-phase flow model and four-dimensional coupling, multi-objective optimization of structure and injection parameters; based on the optimization results, using model predictive control or reinforcement learning to dynamically regulate the injection system in real time, realizing efficient, safe and low-carbon operation in the smelting process. The application can realize efficient, energy-saving, safe and long-life operation of the heavy non-ferrous metal smelting process, and improve the intelligent level and comprehensive benefits of the smelting process.

Description

Technical Field

[0001] This invention relates to the field of heavy non-ferrous metal metallurgy technology, and more specifically to a four-dimensional coupled control method for heavy non-ferrous metal molten pool furnace structure, three fields inside the furnace, furnace body thermal stress, and injection system. Background Technology

[0002] Existing heavy non-ferrous metal molten pool smelting processes mainly include various furnace types such as top-blown, side-blown, and top-side combined blowing, which are core technologies for processing complex polymetallic sulfide ores and secondary resources. The technical principle is to achieve oxidation, combustion, and reduction reactions through the strong interaction between high-speed jets and high-temperature melt, accompanied by intense chemical reactions, interphase heat and mass transfer, and complex fluid dynamics. However, existing technologies have several problems: First, furnace structure design relies on experience and lacks linkage optimization with multi-field calculation results; second, the three-field calculations within the furnace often use a single CFD model with simplified boundary conditions, making it difficult to reflect furnace condition fluctuations in real time; third, furnace thermal stress analysis is usually decoupled from the furnace heat flow process and performed offline, failing to adapt to actual operating condition changes; finally, the jetting control system lacks a collaborative feedback mechanism with the furnace state, making it difficult to balance reaction efficiency, lance life, and furnace lining protection.

[0003] In implementing the embodiments of the present invention, the prior art has at least the following problems or defects: traditional methods generally suffer from problems such as the disconnect between furnace structure design and multi-field coupling optimization, the inconsistency between the simulation of three fields in the furnace and the actual working conditions, the decoupling between thermal stress analysis and heat flow process, and the lack of dynamic control mechanism in the injection system. These problems seriously limit the efficient, stable and low-carbon operation of the heavy non-ferrous metal molten pool smelting process. Summary of the Invention

[0004] This invention provides a four-dimensional coupled control method for the structure of a heavy non-ferrous metal molten pool furnace, the three fields inside the furnace, the thermal stress of the furnace body, and the injection system, including: S1: Process data acquisition step, which continuously samples multi-source process variable data of the molten pool smelting process through the data acquisition module; S2: Data-driven model construction step: Based on the process variable data, a data-driven model is constructed, and the data-driven model uses a physical-data fusion strategy to predict key state variables; S3: Four-dimensional coupling model construction and optimization steps: Construct a parameterized model of furnace structure, a coupled model of furnace temperature field-flow field-concentration field, a coupled model of furnace body thermal stress field, and a gas-liquid two-phase flow model of injection system, and couple the models in four dimensions; optimize the furnace structure parameters and injection regime through multi-objective optimization methods. S4: Dynamic control step, based on the data-driven model and the optimized injection regime, adopts model predictive control or reinforcement learning framework to perform real-time dynamic control of the injection system.

[0005] Furthermore, in step S1, the multi-source process variable data includes furnace front temperature, slag outlet temperature, copper outlet temperature, flue gas temperature, SO2 concentration in flue gas, O2 concentration in flue gas, CO concentration in flue gas, CO2 concentration in flue gas, multi-point temperature field obtained by the furnace shell infrared temperature measurement system, area array temperature field obtained by the furnace shell infrared temperature measurement system, spray gun cooling water flow rate, spray gun cooling water inlet and outlet temperature difference, tuyere cooling water flow rate, tuyere cooling water inlet and outlet temperature difference, blower outlet air volume, blower outlet pressure, oxygen concentration, fuel flow rate, current, voltage, and feeding speed of various silos.

[0006] Furthermore, in step S2, the data-driven model adopts a physical-data fusion strategy, specifically in the following form: in, These are the predicted values ​​for critical state variables. This represents the predicted value of the physical model under a given operating condition. For parameters Systemic biases in data-driven model learning For a moment The process variable vector; The data-driven model employs one of the following: multilayer feedforward neural network, convolutional neural network, long short-term memory network, Transformer, gradient boosting tree, or random forest. The key state parameters include the temperature at key locations inside the furnace, the local heat flux density of the inner wall of the furnace lining, the thermal stress safety factor, the splash intensity index of the free liquid surface area, the local bubble content, and the circulating flow intensity.

[0007] Furthermore, in step S3, the furnace structure parameterization model uses a furnace structure parameter vector. Describe the furnace body structure, the furnace structure parameter vector Represented as: in, The total length of the furnace body in the material flow direction. The width of the furnace chamber is lateral. This is the total height of the furnace body. The depth of the molten pool The height of the upper space, For the angle of the furnace wall, The vertical position of the nozzle or vent relative to the surface of the molten pool. Number of nozzles Nozzle diameter.

[0008] Furthermore, in step S3, the coupled model of the furnace temperature field-flow field-concentration field is numerically simulated using computational fluid dynamics, and the governing equations include: mass conservation equation:

[0009] Momentum conservation equation:

[0010] Energy conservation equation:

[0011] in, For fluid density, Let V be the volume of the liquid phase. For speed, For pressure, Molecular viscosity, It is the gravitational acceleration vector. For surface tension, For energy, For effective thermal conductivity, For temperature; The model uses a VOF model to track the gas-slag two-phase interface and employs... Two-equation model is used to simulate turbulence.

[0012] Furthermore, in step S3, the furnace body thermal stress field coupling model includes the furnace body heat conduction process and thermo-mechanical coupling stress calculation; The governing equation for the furnace body heat conduction process is: in, For solid temperature, For density, For specific heat, Thermal conductivity, As an internal heat source; The thermo-mechanical coupling stress calculation adopts linear elasticity theory, equivalent effect The calculation formula is: in, Principal stress; thermal stress safety factor Defined as: .in, Allowable stress.

[0013] Furthermore, in step S3, the gas-liquid two-phase flow model of the injection system uses the injection regime parameter vector. Describe the jetting system, the jetting regime parameter vector Represented as: in, This is the oxygen volumetric flow rate. This refers to the fuel volumetric flow rate. To dilute the gas volumetric flow rate, The static pressure in front of the spray gun or side air nozzle. The spray gun elevation angle, This refers to the azimuth angle of the spray gun; jet characteristic velocity The calculation formula is: in, The discharge coefficient is... The density of the injected gas, The pressure in the flow field near the surface of the molten pool; Bubble penetration depth The calculation formula is: in, The density of the liquid phase is... This represents the total inlet volumetric flow rate of the injected gas. This refers to the characteristic dimensions of nozzle spacing or furnace width. For local characteristic velocities in the liquid phase, The equivalent nozzle diameter, These are calibration coefficients.

[0014] Further, in step S4, the model prediction control framework includes: Construct discrete-time prediction equations: ,in For state vectors, For control vectors; Within each control cycle, based on the current estimated state Using these as initial conditions, we simulate and predict the future evolution of the system in the prediction time domain. A comprehensive performance index function is constructed based on melting strength deviation, energy consumption, thermal stress exceedance, and splashing risk indicators. And introduce a penalty term to control the increment; Solve the constrained rolling optimization problem to obtain the optimal control sequence. The constraints include thermal stress safety constraints, process constraints, and equipment constraints.

[0015] Furthermore, the state vector Including the temperature of hot spots inside the furnace Temperature at location represented by slag layer, temperature at location represented by metal layer, and maximum heat flux density of the furnace lining inner wall. thermal stress safety factor Splash risk indicators Smelting strength Flue gas SO2 concentration and oxygen utilization rate; The control vector Including total oxygen flow The flow distribution ratio between top and side blowing, and the adjustment amount of each spray gun. Adjustment amount of each air outlet Local oxygen enrichment rate, spray gun elevation angle, and spray gun extension.

[0016] Furthermore, in step S4, the reinforcement learning framework constructs a training environment based on a digital twin model, which integrates the furnace structure parameterization model, the furnace internal temperature field-flow field-concentration field coupling model, the furnace body thermal stress field coupling model, and the injection system gas-liquid two-phase flow model. In the reinforcement learning framework, the agent's state space includes the process variable data and the key state quantities, the action space includes the adjustable parameters of the control vector, and the reward function is based on a weighted combination of smelting intensity, energy consumption, thermal stress safety factor, and splash risk index. The agent learns through trial and error to select actions in a given state space to maximize cumulative rewards, thereby training a policy network. The reinforcement learning framework complements the model prediction control framework: when the model prediction accuracy is higher than a preset threshold, model prediction control is used for regulation; when the model prediction accuracy is lower than a preset threshold, the policy network is used for regulation. A safety filter layer is set before the control output of the reinforcement learning framework. The safety filter layer modifies the action based on physical constraints and empirical rules. The physical constraints include thermal stress safety constraints, process constraints and equipment constraints to ensure that the control action does not violate the safety boundary.

[0017] The embodiments of the present invention have at least the following beneficial effects: 1. By employing a four-dimensional coupled control method, the coordinated optimization and dynamic control of the furnace structure, three internal fields (temperature field, flow field, and concentration field), furnace thermal stress, and the injection system are achieved. This method effectively solves the problems of low efficiency and unstable operation caused by independent optimization of each system in traditional smelting processes, significantly improving smelting efficiency and product quality stability. Simultaneously, by dynamically controlling the injection system in real time, injection parameters can be adjusted promptly according to the furnace conditions, optimizing furnace reaction conditions, reducing splashing and lance burn-out, and extending equipment lifespan.

[0018] 2. A prediction method combining data-driven models and physical models is introduced to overcome the problem of insufficient prediction accuracy of traditional models under complex operating conditions. This method uses machine learning algorithms to learn from historical data, correcting the deviation between the physical model and actual observations, thereby improving the prediction accuracy of key state variables. This not only enhances the real-time monitoring and prediction capabilities of the smelting process but also provides a more reliable basis for optimized control, enabling the smelting process to operate under better conditions, reducing energy consumption and improving resource utilization.

[0019] 3. A control strategy combining model predictive control (MMC) and reinforcement learning effectively solves the problem of accurately modeling dynamic changes in complex smelting processes. MMC enables fine-grained control when model accuracy is high, while reinforcement learning provides robust control strategies when model prediction accuracy is insufficient, ensuring stable system operation under various conditions. This complementary control strategy not only improves the system's adaptability and robustness but also, through the design of a safety filter layer, ensures that control actions always conform to physical constraints and safety boundaries, thereby enhancing the safety and reliability of the smelting process. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The following drawings are not intentionally drawn to scale to actual size; their focus is on illustrating the main points of this disclosure.

[0021] Figure 1 This is a schematic diagram of a control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a molten pool furnace structure provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments.

[0023] The following is for reference. Figure 1 , Figure 1 This is a schematic diagram of a four-dimensional coupled control method for a heavy non-ferrous metal molten pool smelting furnace structure, including three fields within the furnace, furnace body thermal stress, and injection system, provided as an embodiment of the present invention. Figure 1 As shown, a four-dimensional coupled control method for the structure of a heavy non-ferrous metal smelting furnace, including the three fields inside the furnace, the thermal stress of the furnace body, and the injection system, comprises: S1: Process data acquisition step, which continuously samples multi-source process variable data of the molten pool smelting process through the data acquisition module; S2: Data-driven model construction step: Based on the process variable data, a data-driven model is constructed, and the data-driven model uses a physical-data fusion strategy to predict key state variables; S3: Four-dimensional coupling model construction and optimization steps: Construct a parameterized model of furnace structure, a coupled model of furnace temperature field-flow field-concentration field, a coupled model of furnace body thermal stress field, and a gas-liquid two-phase flow model of injection system, and couple the models in four dimensions; optimize the furnace structure parameters and injection regime through multi-objective optimization methods. S4: Dynamic control step, based on the data-driven model and the optimized injection regime, adopts model predictive control or reinforcement learning framework to perform real-time dynamic control of the injection system.

[0024] By collecting multi-source data during the smelting process, a data-driven model is constructed, and the furnace structure, three internal fields (temperature, flow, and concentration), furnace thermal stress, and the injection system are coupled and optimized. Multi-source process variables refer to various parameters reflecting the furnace state during smelting, including temperature, concentration, and flow rate. This data is collected by sensors and used for subsequent model construction and optimization control. Four-dimensional coupling refers to the joint modeling and optimization of the furnace structure parameterization model, the furnace three-field coupling model, the furnace thermal stress field coupling model, and the gas-liquid two-phase flow model of the injection system to achieve comprehensive control of the smelting process. The dynamic control step, based on the optimized model and injection regime, uses model predictive control or reinforcement learning frameworks to adjust the injection system in real time to adapt to dynamic changes during the smelting process.

[0025] like Figure 2 As shown, the furnace body mainly includes the furnace wall, the injection system, the furnace structure, and the molten pool area inside the furnace. The molten pool is divided into a slag layer and a metallic manganese layer from top to bottom. The injection system is arranged on the side wall or top of the furnace body and is used to inject oxygen, fuel, or dilution gas into the molten pool to enhance the stirring, heat transfer, and chemical reaction within the molten pool.

[0026] Data-driven models can be constructed using machine learning or deep learning methods such as multi-layer feedforward neural networks, convolutional neural networks, long short-term memory networks, Transformers, gradient boosting trees, or random forests. These models can handle different types of data features. For example, convolutional neural networks are suitable for extracting spatial features from furnace shell temperature field images, while long short-term memory networks or Transformers are suitable for processing time-series information such as temperature and flow rate. During model construction, input features require preprocessing operations such as normalization, outlier removal, and feature selection to enhance model stability and generalization ability. The output can select appropriate loss functions for different physical quantities, such as mean squared error or weighted multi-task loss, to simultaneously consider multiple prediction objectives. During model training, an initial training set is constructed using historical simulation data and existing industrial operation data. Initial parameters are obtained through cross-validation, regularization, and hyperparameter optimization. Subsequently, during trial operation and long-term operation of the plant, newly collected data and actual production results are periodically added to the training set. Incremental learning or transfer learning methods are used to update the model parameters, enabling the model to adapt to changes that occur during long-term operation.

[0027] In some embodiments, the multi-source process variable data includes furnace front temperature, slag outlet temperature, copper outlet temperature, flue gas temperature, SO2 concentration in flue gas, O2 concentration in flue gas, CO concentration in flue gas, CO2 concentration in flue gas, multi-point temperature field obtained by the furnace shell infrared temperature measurement system, area array temperature field obtained by the furnace shell infrared temperature measurement system, spray gun cooling water flow rate, spray gun cooling water inlet and outlet temperature difference, tuyere cooling water flow rate, tuyere cooling water inlet and outlet temperature difference, blower outlet air volume, blower outlet pressure, oxygen concentration, fuel flow rate, current, voltage, and feeding speed of various silos.

[0028] The furnace front temperature refers to the temperature at the front end of the smelting furnace, usually measured by sensors such as thermocouples, reflecting the heat level of the initial reaction zone inside the furnace. The slag outlet temperature and copper outlet temperature indicate the temperature state of the slag and metal phases at discharge, respectively. These temperature data are crucial for controlling slag-metal separation and product quality. Flue gas temperature and the concentrations of SO2, O2, CO, and CO2 in the flue gas are monitored in real time by sensors installed in the flue gas channel. These data reflect the progress of redox reactions within the furnace and the completeness of fuel combustion. Multi-point and area array temperature field data acquired by the furnace shell infrared temperature measurement system are collected using equipment such as infrared thermal imagers. This data can display the temperature distribution on the furnace shell surface at high resolution, helping to promptly identify areas of localized overheating or insufficient cooling.

[0029] Parameters such as the flow rate and inlet / outlet temperature difference of the spray gun cooling water, and the flow rate and inlet / outlet temperature difference of the tuyeres cooling water, reflect the operating status of the cooling system and are crucial for protecting critical components such as the spray guns and tuyeres from high-temperature damage. Parameters such as the blower outlet air volume and pressure, oxygen concentration, fuel flow rate, current, and voltage relate to the energy input and equipment operating status during the smelting process, and are important bases for achieving efficient smelting and energy-saving control. The feeding rate of various hoppers directly affects the supply of materials in the furnace, and has a direct impact on maintaining a stable smelting process and product quality.

[0030] In some embodiments, in step S2, the data-driven model employs a physical-data fusion strategy, specifically in the form of: in, These are the predicted values ​​for critical state variables. This represents the predicted value of the physical model under a given operating condition. For parameters Systemic biases in data-driven model learning For a moment The process variable vector; The data-driven model employs one of the following: multilayer feedforward neural network, convolutional neural network, long short-term memory network, Transformer, gradient boosting tree, or random forest. The key state parameters include the temperature at key locations inside the furnace, the local heat flux density of the inner wall of the furnace lining, the thermal stress safety factor, the splash intensity index of the free liquid surface area, the local bubble content, and the circulating flow intensity.

[0031] Specifically, the predicted values ​​of the physical model It is calculated based on known physical laws and chemical reaction mechanisms, reflecting the expected state of the system under given operating conditions. System bias in data-driven model learning. These parameters are learned from historical data using machine learning algorithms and are used to correct discrepancies between physical models and actual observations. These are the adjustable parameters of the data-driven model, determined through the training process. (Process variable vector) This includes various data collected during the smelting process, such as temperature, flow rate, and concentration, which reflect the real-time status of the system. Key state variables include temperatures at critical locations within the furnace, local heat flux density on the furnace lining, and thermal stress safety factors; these variables are crucial for assessing the stability and safety of the smelting process. Data-driven models can employ various machine learning methods, such as multilayer feedforward neural networks and convolutional neural networks, with the specific choice depending on the type and characteristics of the data.

[0032] The process of building a data-driven model includes: collecting historical and real-time data; preprocessing the data, including normalization and outlier removal, to improve the model's stability and generalization ability. Next, a suitable machine learning algorithm is selected to build the data-driven model; for example, a Long Short-Term Memory (LSTM) network can be used for time-series data, and a Convolutional Neural Network (CNN) can be used for image data. During model training, historical data and the physical model's predictions are used as inputs, and the actual observed values ​​are used as the target for training. The model parameters are adjusted through algorithm optimization. This allows the model to learn the deviations between the physical model and actual observations. Finally, the trained data-driven model is combined with the physical model to form a fusion model for real-time prediction of key state variables. In practical applications, the model's input parameters and structure can be adjusted according to the specific needs of the smelting process to improve prediction accuracy and control performance.

[0033] In some embodiments, in step S3, the furnace structure parameterization model uses a furnace structure parameter vector. Describe the furnace body structure, the furnace structure parameter vector Represented as: in, The total length of the furnace body in the material flow direction. The width of the furnace chamber is lateral. This is the total height of the furnace body. The depth of the molten pool The height of the upper space, For the angle of the furnace wall, The vertical position of the nozzle or vent relative to the surface of the molten pool. Number of nozzles The nozzle diameter is [value].

[0034] The parameterized model of furnace structure is based on a set of parameter vectors. This parameterization method describes the overall structural characteristics of the molten pool furnace. It simplifies complex furnace structures into an adjustable set of parameters, facilitating flexible adjustments to the furnace structure to adapt to different smelting requirements during subsequent optimization. The parameterized furnace structure model not only covers the basic dimensions of the furnace body but also includes the geometric features of the injection system; these parameters collectively determine the furnace's performance and efficiency.

[0035] Specifically, and It determines the horizontal dimensions of the furnace, affecting the area of ​​the molten pool and the distribution of materials; , and These parameters determine the total height of the furnace body, the depth of the molten pool, and the height of the upper space, respectively. These height parameters have a significant impact on the heat distribution and gas-liquid flow during the smelting process. The angle of inclination of the furnace wall affects the flow of materials and heat transfer within the furnace. , and The geometric parameters of the injection system determine the distribution and impact position of the injected gas, which plays a key role in the stirring and reaction efficiency of the molten pool.

[0036] When constructing a parametric model of the furnace structure, computer-aided design (CAD) and computational fluid dynamics (CFD) software can be used to generate parameter vectors. The corresponding 3D geometric model is generated. First, initial parameter values ​​are set according to process requirements and design goals. Then, the geometric model of the furnace body is generated in CAD software using a parameter-driven approach. Next, CFD software is used to mesh the generated geometric model, ensuring mesh quality in complex flow regions, such as near the nozzles and on the molten pool surface. During model optimization, parameter vectors can be adjusted. The system automatically updates the geometric model and re-simulates the various parameter values ​​to evaluate the impact of different furnace structures on the smelting process. For example, by changing... and The value of can be used to study the influence of furnace size on the molten pool circulation flow; by adjusting and This allows for the optimization of the injection system layout to improve stirring efficiency. Through this parametric modeling and optimization method, various furnace design schemes can be rapidly evaluated during the design phase, thereby finding the optimal combination of furnace structural parameters to achieve a highly efficient smelting process.

[0037] In some embodiments, in step S3, the furnace temperature field-flow field-concentration field coupled model is numerically simulated using computational fluid dynamics, and the governing equations include: mass conservation equation:

[0038] Momentum conservation equation:

[0039] Energy conservation equation: in, For fluid density, Let V be the volume of the liquid phase. For speed, For pressure, Molecular viscosity, It is the gravitational acceleration vector. For surface tension, For energy, For effective thermal conductivity, For temperature; The model uses a VOF model to track the gas-slag two-phase interface and employs... Two-equation model is used to simulate turbulence.

[0040] The mass conservation equation describes the fluid density and fluid velocity The relationship between these factors ensures that the mass of the fluid remains constant at all times and in all spaces. The momentum conservation equation, on the other hand, takes into account the velocity of the fluid. ,pressure Viscosity and external forces, such as gravity and surface tension The effect on fluid motion. The energy conservation equation describes the energy of the fluid. and temperature The change in effective thermal conductivity This is a key parameter that determines the efficiency of heat transfer in the fluid. Furthermore, the model employs a volume fraction (VOF) model to track the gas-slag two-phase interface, ensuring accurate capture of the distribution of the gas and liquid phases. The turbulence model uses... Two-equation model, by solving turbulent kinetic energy Sum of dissipation Equations are used to describe the characteristics of turbulence. These equations and models were chosen to more accurately simulate the complex multiphase flow and heat transfer processes within the furnace.

[0041] Preferably, when constructing the coupled model of the temperature field, flow field, and concentration field within the furnace, it is first necessary to establish the computational domain based on the specific geometry of the smelting furnace and then mesh it. During mesh generation, special attention needs to be paid to complex flow regions such as the vicinity of the nozzle and the surface of the molten pool to ensure computational accuracy. Then, the mass conservation equation, momentum conservation equation, and energy conservation equation are input into the CFD software as governing equations, and corresponding boundary conditions are set, such as the temperature and pressure of the furnace wall, and the flow rates at the inlet and outlet.

[0042] The VOF model requires defining the volume fractions of the gas and liquid phases to accurately track the location of the two-phase interface. In the turbulence model, turbulence parameters, such as initial values ​​for turbulent kinetic energy and specific dissipation rate, need to be set according to actual operating conditions. Through these steps, a coupled model that accurately reflects the physical phenomena within the furnace can be constructed. During the model solution process, an iterative algorithm can be used to solve the governing equations step by step until convergence. Finally, post-processing software can extract the distribution of temperature, flow rate, and concentration within the furnace, providing detailed data support for optimizing the smelting process.

[0043] In some embodiments, the furnace thermal stress field coupling model in step S3 includes the furnace heat conduction process and the calculation of thermo-mechanical coupling stress. The governing equation for the furnace body heat conduction process is: in, For solid temperature, For density, For specific heat, Thermal conductivity, As an internal heat source; The thermo-mechanical coupling stress calculation adopts linear elasticity theory, equivalent stress. The calculation formula is: in, Principal stress; thermal stress safety factor Defined as: in, Allowable stress.

[0044] The governing equations for the heat conduction process in the furnace body describe the temperature variation in the solid region of the furnace body, where... It indicates the solid temperature, reflecting the temperature distribution at different locations within the furnace body; and These represent the density and specific heat of a solid, respectively. They are thermophysical property parameters of a material that determine its ability to absorb or release heat when the temperature changes. Thermal conductivity is the coefficient of performance in which a material conducts heat. This is an internal heat source term, which is usually negligible during the smelting process. Thermo-mechanical coupled stress calculations use linear elasticity theory, with equivalent stress... Through principal stress , and The calculated overall stress level is used to assess the material's stress level under multiaxial stress conditions. Thermal stress safety factor. It is achieved by comparing the equivalent stress with the allowable stress of the material. The results are obtained through comparison and are used to determine the safety margin of the furnace body under the current operating conditions.

[0045] When constructing the coupled thermal stress field model of the furnace body, it is first necessary to divide the furnace body into multiple subdomains based on its actual structure, such as the working layer refractory bricks, the permanent layer refractory material, the insulation layer, the steel shell, and the water-cooled walls, each with different thermophysical properties. Then, based on the furnace body's geometry and material properties, a three-dimensional heat conduction model is established and discretized using the finite element method. During the solution process, the wall heat flux density calculated from the three-field model within the furnace needs to be considered. The internal boundary conditions are input into the thermal conductivity model, while the external boundary conditions are set based on the environmental conditions of the furnace exterior surface. In the thermo-mechanical coupling calculation, the temperature field obtained from the thermal conductivity model is used as a known load, and stress-strain analysis is performed in conjunction with the material's elastic constants, such as elastic modulus and Poisson's ratio. Finally, the equivalent stress is calculated. and thermal stress safety factor This can assess the thermal stress level and safety status of the furnace body under different operating conditions. In practical applications, the furnace body structural design can be adjusted or the injection system optimized based on the calculation results to reduce thermal stress risks and extend the service life of the furnace body.

[0046] In some embodiments, in step S3, the gas-liquid two-phase flow model of the injection system uses the injection regime parameter vector. Describe the jetting system, the jetting regime parameter vector Represented as: in, This is the oxygen volumetric flow rate. This refers to the fuel volumetric flow rate. To dilute the gas volumetric flow rate, The static pressure in front of the spray gun or side air nozzle. The spray gun elevation angle, Spray gun azimuth angle; jet characteristic velocity The calculation formula is: in, The discharge coefficient is... The density of the injected gas, The pressure in the flow field near the surface of the molten pool; Bubble penetration depth The calculation formula is: in, The density of the liquid phase is... This represents the total inlet volumetric flow rate of the injected gas. This refers to the characteristic dimensions of nozzle spacing or furnace width. For local characteristic velocities in the liquid phase, The equivalent nozzle diameter, These are calibration coefficients.

[0047] The gas-liquid two-phase flow model of the injection system is obtained through the injection regime parameter vector. This model describes the operating status and characteristics of the injection system. The core of this model lies in simplifying the complex operating conditions of the injection system into a set of adjustable parameters through parameterization, thereby facilitating flexible adjustment of the system's operating conditions during subsequent optimization and control. The injection system plays a crucial role in the smelting process, determining not only the supply methods of oxygen, fuel, and dilution gases, but also directly affecting the bubble plume structure, stirring intensity, and heat release method within the molten pool, ultimately impacting smelting efficiency and furnace lining life.

[0048] Oxygen volume flow rate The amount of oxygen supplied directly affects the oxidation reaction rate during the smelting process; fuel volumetric flow rate. and dilution gas volume flow rate These controls the fuel supply and dilution gas flow rate, respectively, to regulate the combustion process and the properties of the injected gas; the static pressure before the spray gun or side-blowing nozzle... Affecting the velocity and penetration ability of the injected gas; spray gun elevation angle and azimuth This determines the jet direction, affecting the distribution and area of ​​influence of the bubble plume in the molten pool. Furthermore, the jet characteristic velocity... and bubble penetration depth These parameters are calculated using relevant formulas and further describe the behavior and effects of the injected gas in the molten pool.

[0049] During implementation, firstly, initial injection parameters, such as oxygen volumetric flow rate and fuel volumetric flow rate, are set according to actual working conditions, and the jet characteristic velocity is calculated using relevant formulas. and bubble penetration depth Jet characteristic velocity Through static pressure of the spray gun and molten pool surface pressure The calculated value reflects the velocity of the injected gas as it enters the molten pool; the bubble penetration depth. The penetration capability of bubbles in the molten pool is calculated using parameters such as the density ratio of gas to liquid, the flow rate of the injected gas, the nozzle spacing or furnace width characteristic scale, the local characteristic velocity of the liquid phase, and the equivalent nozzle diameter. During model building, the interaction between the injected gas and the molten pool liquid also needs to be considered, such as bubble breakage, coalescence, and their impact on the flow field. Through these detailed modeling and calculation steps, the behavior and effects of the injection system in the smelting process can be accurately described, providing a theoretical basis for optimizing the injection regime.

[0050] In some embodiments, in step S4, the model prediction control framework includes: Construct discrete-time prediction equations: ,in For state vectors, For control vectors; Within each control cycle, based on the current estimated state Using these as initial conditions, we simulate and predict the future evolution of the system in the prediction time domain. A comprehensive performance index function is constructed based on melting strength deviation, energy consumption, thermal stress exceedance, and splashing risk indicators. And introduce a penalty term to control the increment; Solve the constrained rolling optimization problem to obtain the optimal control sequence. The constraints include thermal stress safety constraints, process constraints, and equipment constraints.

[0051] Indicates the system at time... The state vector reflects important state information such as temperature, heat flux density, and thermal stress safety factor at key locations inside the furnace; It is the control vector, which includes adjustable parameters such as total oxygen flow rate, flow rate adjustment for each spray nozzle, and spray nozzle elevation angle. Within each control cycle, it is based on the current estimated state. Using these as initial conditions, we simulate and predict the future evolution of the system in the prediction time domain. (Comprehensive performance index function) It is constructed using indicators such as melting strength deviation, energy consumption, thermal stress exceedance, and splashing risk to evaluate control effectiveness, and introduces a control increment penalty term to avoid overly drastic control actions. Constraints include thermal stress safety constraints, process constraints, and equipment constraints, which ensure that control actions do not violate physical safety boundaries and equipment operating limits.

[0052] The input parameters of the discrete-time prediction equation include the current furnace state vector. and control vector These parameters can be acquired in real time by sensors or predicted through models. During the prediction process, numerical simulation methods are used to predict the future behavior of the system, and the prediction is based on a comprehensive performance index function. The optimal control sequence is determined by considering constraints. For example, linear or nonlinear programming algorithms can be used to optimize control parameters such as oxygen flow rate and lance position, provided that the thermal stress safety factor does not exceed the allowable value, the furnace temperature is maintained within the process's permissible range, and the equipment operation does not exceed limits. In practical applications, the optimized control sequence can be gradually applied to the injection system, and the control strategy can be adjusted based on real-time feedback to ensure the dynamic stability and safety of the smelting process.

[0053] In some embodiments, the state vector Including the temperature of hot spots inside the furnace Temperature at location represented by slag layer, temperature at location represented by metal layer, and maximum heat flux density of the furnace lining inner wall. thermal stress safety factor Splash risk indicators Smelting strength Flue gas SO2 concentration and oxygen utilization rate; The control vector Including total oxygen flow The flow distribution ratio between top and side blowing, and the adjustment amount of each spray gun. Adjustment amount of each air outlet Local oxygen enrichment rate, spray gun elevation angle, and spray gun extension.

[0054] Reinforcement learning trains agents in a virtual environment built on a digital twin model, enabling them to select optimal actions based on current furnace conditions and injection status to maximize long-term cumulative rewards. The digital twin model integrates multiple models, including furnace structure, internal furnace fields, furnace thermal stress, and the injection system, reflecting the dynamic characteristics of the smelting process in real time. The reinforcement learning framework complements the model predictive control framework; when model prediction accuracy is insufficient, the policy network trained through reinforcement learning is used for adjustment, ensuring the robustness and adaptability of the control system.

[0055] The agent's state space includes process variable data and key state quantities, reflecting the real-time state of the smelting process, such as furnace temperature, heat flux density, and thermal stress safety factor. The action space includes adjustable parameters of the control vector, such as oxygen flow rate and lance elevation angle, which directly affect the operating state of the injection system. The reward function is constructed based on a weighted combination of smelting intensity, energy consumption, thermal stress safety factor, and splashing risk index, and is used to evaluate the effectiveness of the agent's actions. The agent learns through trial and error, selecting actions within a given state space to maximize cumulative rewards. A safety filter layer modifies actions based on physical constraints and empirical rules, ensuring that control actions do not violate fundamental safety boundaries such as thermal stress safety constraints, process constraints, and equipment constraints.

[0056] A digital twin model is constructed by integrating a parameterized model of the furnace structure, a three-field coupling model within the furnace, a coupled model of the furnace body's thermal stress field, and a gas-liquid two-phase flow model of the injection system to achieve high-precision simulation of the smelting process. The state space and action space of the agent need to be defined in detail according to the actual smelting process. For example, the state space can include key state variables such as the temperature of hot spots in the furnace and the temperatures of representative locations in the slag and metal layers, while the action space can include adjustable parameters such as the total oxygen flow rate and the adjustment amount of the flow rate of each injection lance. During training, the agent continuously adjusts its action strategy to maximize the reward function through interaction with the digital twin model. The weights of the reward function can be adjusted according to actual production needs; for example, when smelting intensity is more important, the weight of the smelting intensity index can be increased. The implementation of the safety filter layer needs to be combined with specific physical constraints; for example, when the predicted thermal stress exceeds the allowable value, the injection intensity is automatically adjusted to reduce the risk of thermal stress. Finally, the trained policy network can be used in conjunction with the model predictive control framework during the actual smelting process to form an adaptive, highly robust dynamic control system.

[0057] In some embodiments, in step S4, the reinforcement learning framework constructs a training environment based on a digital twin model, wherein the digital twin model integrates the furnace structure parameterization model, the furnace internal temperature field-flow field-concentration field coupling model, the furnace body thermal stress field coupling model, and the injection system gas-liquid two-phase flow model. In the reinforcement learning framework, the agent's state space includes the process variable data and the key state quantities, the action space includes the adjustable parameters of the control vector, and the reward function is based on a weighted combination of smelting intensity, energy consumption, thermal stress safety factor, and splash risk index. The agent learns through trial and error to select actions in a given state space to maximize cumulative rewards, thereby training a policy network. The reinforcement learning framework complements the model prediction control framework: when the model prediction accuracy is higher than a preset threshold, model prediction control is used for regulation; when the model prediction accuracy is lower than a preset threshold, the policy network is used for regulation. A safety filter layer is set before the control output of the reinforcement learning framework. The safety filter layer modifies the action based on physical constraints and empirical rules. The physical constraints include thermal stress safety constraints, process constraints and equipment constraints to ensure that the control action does not violate the safety boundary.

[0058] The agent's state space in the reinforcement learning framework includes process variable data and key state quantities, reflecting the real-time state of the smelting process, such as furnace temperature, heat flux density, and thermal stress safety factor. The action space includes adjustable parameters of the control vector, such as oxygen flow rate and lance elevation angle, which directly affect the operating state of the injection system. The reward function is constructed based on a weighted combination of smelting intensity, energy consumption, thermal stress safety factor, and splashing risk index, and is used to evaluate the effectiveness of the agent's actions.

[0059] The agent learns through trial and error, selecting actions within a given state space to maximize cumulative rewards. A safety filter layer modifies actions based on physical constraints and empirical rules, ensuring that control actions do not violate fundamental safety boundaries such as thermal stress safety constraints, process constraints, and equipment constraints. These constraints are crucial for ensuring the safe operation of the smelting process; for example, thermal stress safety constraints ensure the safety of the furnace structure, process constraints guarantee the efficiency and quality of the smelting process, and equipment constraints prevent equipment from being overloaded.

[0060] The above embodiments of the present invention have the following beneficial effects: 1. By employing a four-dimensional coupled control method, the coordinated optimization and dynamic control of the furnace structure, three internal fields (temperature field, flow field, and concentration field), furnace thermal stress, and the injection system are achieved. This method effectively solves the problems of low efficiency and unstable operation caused by independent optimization of each system in traditional smelting processes, significantly improving smelting efficiency and product quality stability. Simultaneously, by dynamically controlling the injection system in real time, injection parameters can be adjusted promptly according to the furnace conditions, optimizing furnace reaction conditions, reducing splashing and lance burn-out, and extending equipment lifespan.

[0061] 2. A prediction method combining data-driven models and physical models is introduced to overcome the problem of insufficient prediction accuracy of traditional models under complex operating conditions. This method uses machine learning algorithms to learn from historical data, correcting the deviation between the physical model and actual observations, thereby improving the prediction accuracy of key state variables. This not only enhances the real-time monitoring and prediction capabilities of the smelting process but also provides a more reliable basis for optimized control, enabling the smelting process to operate under better conditions, reducing energy consumption and improving resource utilization.

[0062] 3. A control strategy combining model predictive control (MMC) and reinforcement learning effectively solves the problem of accurately modeling dynamic changes in complex smelting processes. MMC enables fine-grained control when model accuracy is high, while reinforcement learning provides robust control strategies when model prediction accuracy is insufficient, ensuring stable system operation under various conditions. This complementary control strategy not only improves the system's adaptability and robustness but also, through the design of a safety filter layer, ensures that control actions always conform to physical constraints and safety boundaries, thereby enhancing the safety and reliability of the smelting process.

[0063] The foregoing description is illustrative of the invention and should not be construed as limiting it. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Therefore, all such modifications are intended to be included within the scope of the invention as defined in the claims. It should be understood that the foregoing description is illustrative of the invention and should not be construed as limiting it to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.

Claims

1. A four-dimensional coupled control method for the structure of a heavy non-ferrous metal molten pool furnace, including three fields within the furnace, thermal stress in the furnace body, and the injection system, characterized in that... Includes the following steps: S1: Process data acquisition step, which continuously samples multi-source process variable data of the molten pool smelting process through the data acquisition module; S2: Data-driven model construction step: Based on the process variable data, a data-driven model is constructed, and the data-driven model uses a physical-data fusion strategy to predict key state variables; S3: Four-dimensional coupling model construction and optimization steps: Construct a parameterized model of furnace structure, a coupled model of furnace temperature field-flow field-concentration field, a coupled model of furnace body thermal stress field, and a gas-liquid two-phase flow model of injection system, and couple the models in four dimensions; optimize the furnace structure parameters and injection regime through multi-objective optimization methods. S4: Dynamic control step, based on the data-driven model and the optimized injection regime, adopts model predictive control or reinforcement learning framework to perform real-time dynamic control of the injection system; In step S2, the data-driven model adopts a physical-data fusion strategy, specifically in the following form: in, These are the predicted values ​​for critical state variables. The physical model predicts the value under a given operating condition. For parameters Systemic biases in data-driven model learning For a moment The process variable vector; The data-driven model employs one of the following: multilayer feedforward neural network, convolutional neural network, long short-term memory network, Transformer, gradient boosting tree, or random forest. The key state parameters include the temperature at key locations inside the furnace, the local heat flux density of the inner wall of the furnace lining, the thermal stress safety factor, the splash intensity index of the free liquid surface area, the local bubble content, and the circulating flow intensity. In step S4, the reinforcement learning framework constructs a training environment based on a digital twin model. The digital twin model integrates the furnace structure parameterization model, the furnace internal temperature field-flow field-concentration field coupling model, the furnace body thermal stress field coupling model, and the injection system gas-liquid two-phase flow model. In the reinforcement learning framework, the agent's state space includes the process variable data and the key state quantities, the action space includes the adjustable parameters of the control vector, and the reward function is based on a weighted combination of smelting intensity, energy consumption, thermal stress safety factor, and splash risk index. The agent learns through trial and error to select actions in a given state space to maximize cumulative rewards, thereby training a policy network. The reinforcement learning framework complements the model prediction control framework: when the model prediction accuracy is higher than a preset threshold, model prediction control is used for regulation; when the model prediction accuracy is lower than a preset threshold, the policy network is used for regulation. A safety filter layer is set before the control output of the reinforcement learning framework. The safety filter layer modifies the action based on physical constraints and empirical rules. The physical constraints include thermal stress safety constraints, process constraints and equipment constraints to ensure that the control action does not violate the safety boundary. In step S3, the gas-liquid two-phase flow model of the injection system uses the injection regime parameter vector. Describe the jetting system, the jetting regime parameter vector Represented as: in, This is the oxygen volumetric flow rate. This refers to the fuel volumetric flow rate. To dilute the gas volumetric flow rate, The static pressure in front of the spray gun or side air nozzle. The spray gun elevation angle, This refers to the azimuth angle of the spray gun; jet characteristic velocity The calculation formula is: in, The discharge coefficient is... The density of the injected gas, The pressure in the flow field near the surface of the molten pool; Bubble penetration depth The calculation formula is: in, The density of the liquid phase is... This represents the total inlet volumetric flow rate of the injected gas. This refers to the characteristic dimensions of nozzle spacing or furnace width. For local characteristic velocities in the liquid phase, The equivalent nozzle diameter, These are calibration coefficients; The state vector Including the temperature of hot spots inside the furnace Temperature at location represented by slag layer, temperature at location represented by metal layer, and maximum heat flux density of the furnace lining inner wall. Thermal stress safety factor Splash risk indicators Smelting strength SO2 concentration and oxygen utilization rate in flue gas; The control vector Including total oxygen flow The flow distribution ratio between top and side blowing, and the adjustment amount of each spray gun. Adjustment amount of each air outlet Local oxygen enrichment rate, spray gun elevation angle, and spray gun extension; In step S4, the model prediction control framework includes: Construct discrete-time prediction equations: ,in For state vectors, For control vectors; Within each control cycle, based on the current estimated state Using these as initial conditions, we simulate and predict the future evolution of the system in the prediction time domain. A comprehensive performance index function is constructed based on melting strength deviation, energy consumption, thermal stress exceedance, and splashing risk indicators. And introduce a penalty term to control the increment; Solve the constrained rolling optimization problem to obtain the optimal control sequence. The constraints include thermal stress safety constraints, process constraints, and equipment constraints.

2. The method according to claim 1, characterized in that, In step S1, the multi-source process variable data includes furnace front temperature, slag outlet temperature, copper outlet temperature, flue gas temperature, SO2 concentration in flue gas, O2 concentration in flue gas, CO concentration in flue gas, CO2 concentration in flue gas, multi-point temperature field obtained by the furnace shell infrared temperature measurement system, area array temperature field obtained by the furnace shell infrared temperature measurement system, spray gun cooling water flow rate, spray gun cooling water inlet and outlet temperature difference, tuyere cooling water flow rate, tuyere cooling water inlet and outlet temperature difference, blower outlet air volume, blower outlet pressure, oxygen concentration, fuel flow rate, current, voltage, and feeding speed of various silos.

3. The method according to claim 1, characterized in that, In step S3, the furnace structure parameterization model uses a furnace structure parameter vector. Describe the furnace body structure, the furnace structure parameter vector Represented as: in, The total length of the furnace body in the material flow direction. The width of the furnace chamber is lateral. This is the total height of the furnace body. The depth of the molten pool The height of the upper space, For the angle of the furnace wall, The vertical position of the nozzle or vent relative to the surface of the molten pool. Number of nozzles The nozzle diameter is [value].

4. The method according to claim 1, characterized in that, In step S3, the coupled model of the furnace temperature field-flow field-concentration field is numerically simulated using computational fluid dynamics, and the governing equations include: mass conservation equation: Momentum conservation equation: Energy conservation equation: in, For fluid density, Let V be the volume of the liquid phase. For speed, For pressure, Molecular viscosity, It is the gravitational acceleration vector. For surface tension, For energy, For effective thermal conductivity, The temperature is used; the model uses a VOF model to track the gas-slag two-phase interface, and employs... Two-equation model is used to simulate turbulence.

5. The method according to claim 1, characterized in that, In step S3, the furnace body thermal stress field coupling model includes the furnace body heat conduction process and thermo-mechanical coupling stress calculation; The governing equation for the furnace body heat conduction process is: in, For solid temperature, For density, For specific heat, Thermal conductivity, As an internal heat source; The thermo-mechanical coupling stress calculation adopts linear elasticity theory, equivalent stress. The calculation formula is: in, Principal stress; thermal stress safety factor Defined as: in, Allowable stress.

Images