Rotary hearth furnace dynamic batching optimization method and system

Abstract

The application discloses a rotary hearth furnace dynamic batching optimization method and system, which is based on machine learning technology and element conservation mechanism to establish a rotary hearth furnace dynamic batching model. First, according to historical batching information, a constrained extreme gradient boosting XGBoost model is used to predict the expected batching scheme of each bin. Then, considering the bin state and batching stability, the constraint conditions and objective function of the batching model are constructed according to the element balance principle, and the model parameters are optimized by using the sensitivity analysis method to obtain the optimized dynamic batching model. Finally, the developed dynamic batching model is integrated with the existing process control system. According to the technical scheme, dynamic batching can be carried out according to the operating state of each bin of the rotary hearth furnace and the element content change, and the optimal batching scheme is given.

Description

Technical Field

[0001] This invention belongs to the field of metallurgical control technology, specifically relating to a dynamic batching optimization method and system for rotary hearth furnaces. Background Technology

[0002] A rotary hearth furnace is a production equipment in the metallurgical industry that uses a direct reduction process to reduce carbon-containing metal pellets. Over the past forty years, the direct reduction process using a rotary hearth furnace (RHF) has developed rapidly. It uses carbon-containing pellets as reactants, and the main reaction equipment is an annular heating furnace with an annular furnace chamber and a rotating hearth. Carbon-containing pellets are often prepared by mixing iron dust, coal powder, and carbon powder; alternatively, dust with a higher carbon content can be used to replace coal powder as the reaction raw material. After batching, mixing, pelletizing, and drying, the materials are formed into pellets, which are then heated at a high temperature of 1250~1350℃ by rotating with the furnace hearth. The iron oxides in the pellets react with carbon to produce a product—direct reduced iron (DRI). After cooling, the product can be used as raw material for electric furnaces, blast furnaces, converters, and smelting furnaces.

[0003] Scientific and rational raw material proportioning is a prerequisite for ensuring efficient and high-quality production in rotary hearth furnaces. Its accuracy directly affects the quality and stability of the product's direct reduction ratio (DRI), and indirectly impacts important indicators such as carbon emissions and production costs. Currently, rotary hearth furnace technology is mainly used for solid waste recycling, processing large amounts of iron-containing dust and sludge (100 kg / ton of iron) generated during steel production. Furthermore, the US, Germany, Japan, and other countries have applied rotary hearth furnace technology to the direct reduction of iron ore. China's Baosteel has also researched new processes for using rotary hearth furnaces to treat red mud, laterite nickel ore, and municipal solid waste. Rotary hearth furnaces can process a wide variety of raw materials, commonly including converter dust, blast furnace ash, electric arc furnace ash, cold-rolled waste, and iron concentrate powder. The wide range of raw material sources and the complexity of their composition increase the difficulty of rotary hearth furnace batching.

[0004] However, due to harsh production conditions and low levels of digitalization, existing rotary hearth furnace batching is usually based on experience, or batched according to specific proportions, or calculated manually. This is labor-intensive, inefficient, and fails to meet the current demand for intelligent rotary hearth furnace production. When the type of raw material changes or its composition fluctuates, multiple calculations are required, and it is difficult to find the optimal batching scheme, resulting in decreased production efficiency, reduced product quality, and increased labor costs. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for dynamic batching optimization in rotary hearth furnaces, enabling automatic batching and optimization of the furnace.

[0006] To achieve the above objectives, the basic solution of the present invention is: a dynamic batching optimization method for a rotary hearth furnace, comprising the following steps:

[0007] Based on machine learning models and historical data, predict the expected batching scheme for each silo;

[0008] Based on the element conservation mechanism and the process requirements of rotary hearth furnace, the constraints and objective function of the dynamic batching model are constructed.

[0009] The parameters of the dynamic batching model are optimized to obtain the optimized dynamic batching model;

[0010] Integrate the dynamic batching model with the process control system online.

[0011] The working principle and beneficial effects of this basic scheme are as follows: This technical scheme analyzes the main influencing components in the batching model and establishes an elemental balance model and a dynamic batching model for the rotary hearth furnace. For the elemental balance model, the relationship between the content of each component in the mixture and the proportion of each silo is derived. The model uses the upper and lower limits of the target product components set in actual production as constraints, and converts the derived relationships into linear constraints. Based on actual needs, the most stable batching scheme is set as the objective, and the objective is converted into a mathematical expression that is easy to calculate. A dynamic batching model is established to ensure that the material level always fluctuates within a safe range. The optimized model automatically obtains the optimized batching scheme, which is easy to use. It dynamically batches according to the operating status and elemental content changes of each silo in the rotary hearth furnace and provides the optimal batching scheme.

[0012] Furthermore, the XGBoost prediction model is introduced, and the objective is constrained through a penalty function. This constraint is then incorporated into the objective function to predict the ingredient ratio. The specific steps are as follows:

[0013] exist At any given moment, select past moments from the historical database. This is a collection of ingredient formulas. , , Indicates a specific moment in the past; Indicates the first Individual material silos; Indicates the past The corresponding silo in the time-based batching plan The ratio; assuming any point in the past Ingredient formulation and time window If there is a mapping relationship between the data within, the following dataset can be constructed. :

[0014]

[0015] in, The time delay window can be manually set to reduce the amount of data and improve the model learning speed. Indicates the past The corresponding silo in the time-based batching plan The ratio; n represents the total ingredient ratio;

[0016] because and If a correlation exists between them, then the following time series model is constructed:

[0017]

[0018] in, express The predicted value; express and There is a function mapping relationship between them.

[0019] Furthermore, a constrained XGBoost prediction model is constructed:

[0020] The basic loss function of XGBoost is defined as follows:

[0021]

[0022] Based on the principle of keeping the total weight of the configuration constant, the prediction output is constrained within a specific range. Within, that is:

[0023]

[0024] By introducing the constraints into XGBoost in the form of penalty functions, a new loss function is obtained:

[0025]

[0026] In the dataset Based on this, gradient descent is used to adjust the loss function of XGBoost. Optimize to obtain a trained XGBoost model, and then obtain the time step. Ingredient formulation Ultimately, we obtain the result at the current moment. Predictive ingredient scheme dataset :

[0027] .

[0028] Because industrial processes are characterized by significant noise fluctuations, the XGBoost model is introduced to reduce prediction errors.

[0029] Furthermore, the silo proportioning decision variables include the proportion of dry quenching coke. %, electric furnace ash ratio %, proportion of ash from the iron tapping site %, Proportion of No. 1 dry-wet mixed sludge %, Proportion of No. 2 dry-wet mixed sludge %, Proportion of sludge recovered from No. 1 electric furnace %, Proportion of sludge recovered from No. 2 electric furnace %, proportion of secondary ash from blast furnace %, cold-rolled slurry proportion %, environmental gray ratio %, adhesive ratio %.

[0030] Setting decision variables facilitates subsequent use.

[0031] Furthermore, the specific method for establishing the elemental balance model is as follows:

[0032] Assume that the product after mixing only has requirements for the content of four elements: C, Cl, Zn, and O.

[0033] Determine the content of the main elements in the internal ash mixture, and then calculate the element content in the mixture based on this. The content of each component in the internal ash mixture is the sum of the products of the mass fraction of different elements in the corresponding silo and their proportions, expressed as:

[0034]

[0035]

[0036]

[0037]

[0038] in, For silos The content of element C, i.e., its mass fraction. This refers to the total content of carbon (C) in the internal ash mixture. For silos The content of element Cl, i.e., its mass fraction. This refers to the total content of the internal gray element Cl; For silos The content of element Zn, i.e., its mass fraction. This refers to the total content of the internal ash element Zn; For silos The content of element Fe, i.e., its mass fraction. This refers to the total content of Fe, an internal ash element. For silos The proportions;

[0039] Considering only the combination of oxygen with iron and zinc, calculate the oxygen content combined with iron. and the oxygen content bound to zinc :

[0040]

[0041]

[0042] in, Indicates silo For the corresponding Content, i.e., mass fraction;

[0043] Total oxygen content in internal ash It is the sum of the two, which simplifies to:

[0044]

[0045] Similarly, the oxygen content in the return silo and environmental ash silo can be derived. , They are respectively:

[0046]

[0047]

[0048] Therefore, the element content in the mixture (including internal ash, environmental ash, binder, and recycled material) is as follows:

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] in, It represents the total content of a certain element in the mixture.

[0056] Furthermore, during the rotary hearth furnace production process, there are restrictive requirements on two key indicators: the chloride-zinc ratio and the carbon-oxygen ratio, expressed as follows:

[0057]

[0058]

[0059] in, The ratio of chlorine to zinc in the mixture; The carbon-to-oxygen ratio in the mixture; given a total material quantity Under the premise of [specific conditions], the total feed rate of internal ash (silos 1-9) for:

[0060]

[0061] Dry basis feed rate of each material in the internal ash mixing for:

[0062]

[0063] Wet basis feed rate of each material in the internal ash mixing for:

[0064] .

[0065] Based on the principle of element conservation, an elemental balance model for rotary hearth furnace batching is established for ease of use.

[0066] Furthermore, considering the predicted batching ratio of the silos and the opening and closing status of the silos, a dynamic batching model is established:

[0067] The total proportion of the ingredients in the specified internal batches (bins 1-9) is 100%, that is:

[0068] Establish constraints for chlorine (Cl), zinc (Zn), carbon (C), carbon-oxygen ratio (C / O), and chlorine-zinc ratio (Cl / Zn):

[0069] (1)

[0070] in, and These are the upper and lower limits of each element or element ratio in the mixture; This refers to the content of each element in the mixture, i.e., its mass fraction. This represents the ratio of carbon to oxygen. This represents the ratio of chlorine to zinc.

[0071] Consider the current open / closed state variables of the silo. 0 indicates that the silo is closed, and 1 indicates that the silo is open. The material distribution constraints for each silo are as follows:

[0072] (2)

[0073] Where bigM represents a sufficiently large constant; and These represent the lower and upper limits of the material mixing ratio in the silo, respectively.

[0074] Consider the current silo The feed rate of the belt is limited and the following feed constraints must be met:

[0075] (3)

[0076] The method for constructing the ingredient objective function is as follows:

[0077] During the batching process, the rotary hearth furnace batching scheme is continuously adjusted to meet constraints as the raw material composition changes. The adjusted formula should be as close as possible to the predicted batching scheme to reduce fluctuations. Furthermore, frequent opening and closing of the silos should be avoided as much as possible during this process. The objective function is:

[0078]

[0079] in, For silos The proportions are the proportion decision variables; To predict the proportions of each material in the batching scheme; For the decision of opening and closing the silo, This indicates the current actual open / closed status of the silo. The material level control function is proportional; The weighting coefficient set by the user is used to adjust the material level; The larger the value, the greater the proportion of the material level adjustment function in the model, and the greater the impact of the material level on the model; an appropriate value should be set. The value is used to balance the two model objectives of material level adjustment and batching stability; although and Both can adjust the proportion of the material level control function, but This involves adjusting the material level function as a whole, treating it as a variable quantity. When the material level is within the normal range, it ensures that the values ​​of the two target functions are roughly on the same order of magnitude, preventing the material level adjustment function from becoming dominant. It is only related to the actual material level of each silo, and changes with the material level, so that the model will prioritize adjusting the material level of a silo when the material level of a certain silo is abnormal.

[0080] The dynamic model of ingredient proportioning is expressed as the following mathematical programming model:

[0081]

[0082] By introducing auxiliary variables, the above model can be transformed into a linear programming model, which can then be solved directly to obtain the optimal proportion of each silo.

[0083] Setting limits on the content of ingredients in the product composition facilitates subsequent solution.

[0084] The present invention also provides a dynamic batching optimization system for rotary hearth furnaces, including a processing unit, wherein the processing unit executes the method described in the present invention to optimize the automatic batching of rotary hearth furnaces.

[0085] This system enables automated ingredient dispensing, making it easy to use. Attached Figure Description

[0086] Figure 1 This is a schematic diagram of the dynamic batching model of the rotary hearth furnace dynamic batching optimization method of the present invention;

[0087] Figure 2 This invention differs from the dynamic batching optimization method for rotary hearth furnaces. Material level change diagram of the feed hopper (I);

[0088] Figure 3 This invention differs from the dynamic batching optimization method for rotary hearth furnaces. Material level change diagram of the hopper (II);

[0089] Figure 4 This invention differs from the dynamic batching optimization method for rotary hearth furnaces. A schematic diagram of the sum of squared differences between adjacent ingredient batching schemes under a given value;

[0090] Figure 5 This is a flowchart illustrating the XGBoost prediction model of the rotary hearth furnace dynamic batching optimization method of the present invention. Detailed Implementation

[0091] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0092] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0093] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0094] This invention discloses a method for optimizing dynamic batching in a rotary hearth furnace, comprising the following steps:

[0095] Based on machine learning models and historical data, predict the expected batching scheme for each silo;

[0096] Based on the element conservation mechanism and the process requirements of the rotary hearth furnace, the constraints and objective function of the dynamic batching model are constructed. The main mechanism of the batching model is element conservation; the raw materials supplied to each bin should ensure that the mixed product meets the target composition requirements. The target product composition mainly includes elements such as carbon, chlorine, oxygen, and zinc, which have a significant impact on actual production. Based on the element conservation mechanism and decision variables, an element balance model is established; the bin batching decision variables include the proportion of dry quenching coke. %, electric furnace ash ratio %, proportion of ash from the iron tapping site %, Proportion of No. 1 dry-wet mixed sludge %, Proportion of No. 2 dry-wet mixed sludge %, Proportion of sludge recovered from No. 1 electric furnace %, Proportion of sludge recovered from No. 2 electric furnace %, proportion of secondary ash from blast furnace %, cold-rolled slurry proportion %, environmental gray ratio %, adhesive ratio %.

[0097] The parameters of the dynamic batching model are optimized to obtain the optimized dynamic batching model;

[0098] The dynamic batching model is integrated online with the process control system, and the model is encapsulated as a dynamic link library (DLL) file and integrated into the PLC of the process control system.

[0099] In a preferred embodiment of the present invention, because industrial processes are characterized by significant noise fluctuations, simply comparing the material proportions from the previous batching plan with the existing batching plan can easily lead to large errors. To address this problem, such as... Figure 5 As shown, the XGBoost prediction model is introduced, and the objective is constrained by a penalty function. The constraint is added to the objective function to predict the silo batching scheme. The specific steps are as follows:

[0100] exist At any given moment, select past moments from the historical database. This is a collection of ingredient formulas. , , Indicates a specific moment in the past; Indicates the first Individual material silos; Indicates the past The corresponding silo in the time-based batching plan The ratio; assuming any point in the past Ingredient formulation and time window If there is a mapping relationship between the data within, the following dataset can be constructed. :

[0101]

[0102] in, The time delay window can be manually set to reduce the amount of data and improve the model learning speed. Indicates the past The corresponding silo in the time-based batching plan The ratio; n represents the total ingredient ratio;

[0103] because and If a correlation exists between them, then the following time series model is constructed:

[0104]

[0105] in, express The predicted value; express and There is a function mapping relationship between them.

[0106] More preferably, a constrained XGBoost prediction model is constructed:

[0107] The basic loss function of XGBoost is defined as follows:

[0108]

[0109] Based on the principle of keeping the total weight of the configuration constant, the prediction output is constrained within a specific range. Within, that is:

[0110]

[0111] By introducing the constraints into XGBoost in the form of penalty functions, a new loss function is obtained:

[0112] In the dataset Based on this, gradient descent is used to adjust the loss function of XGBoost. Optimize to obtain a trained XGBoost model, and then obtain the time step. Ingredient formulation Ultimately, we obtain the result at the current moment. Predictive ingredient scheme dataset :

[0113] .

[0114] In a preferred embodiment of the present invention, the specific method for establishing the elemental balance model is as follows:

[0115] Assuming the mixed product only has requirements for the content of four elements: C, Cl, Zn, and O, and that O in the raw materials mainly combines with Zn and Fe, other elements are not considered; the process from single raw materials to the mixed product involves only physical changes, and the various chemical components are additive; the element content in the binder is ignored; considering the compositional limitations of Cl, Zn, C, and O, and since O in the raw materials combines with the main elements Fe and Zn, the relevant parameters of Fe should also be introduced into the model; the proportion of the return silo... (%) is a value given by the user;

[0116] Determine the content of the main elements in the internal ash mixture, and then calculate the element content in the mixture based on this. The content of each component in the internal ash mixture is the sum of the products of the mass fraction of different elements in the corresponding silo and their proportions, expressed as:

[0117]

[0118]

[0119]

[0120]

[0121] in, For silos The content of element C, i.e., its mass fraction. This refers to the total content of carbon (C) in the internal ash mixture. For silos The content of element Cl, i.e., its mass fraction. This refers to the total content of the internal gray element Cl; For silos The content of element Zn, i.e., its mass fraction. This refers to the total content of the internal ash element Zn; For silos The content of element Fe, i.e., its mass fraction. This refers to the total content of Fe, an internal ash element. For silos The proportions;

[0122] Considering only the combination of oxygen with iron and zinc, calculate the oxygen content combined with iron. and the oxygen content bound to zinc :

[0123]

[0124]

[0125] in, Indicates silo For the corresponding Content, i.e., mass fraction;

[0126] Total oxygen content in internal ash It is the sum of the two, which simplifies to:

[0127]

[0128] Similarly, the oxygen content in the return silo and environmental ash silo can be derived. , They are respectively:

[0129]

[0130]

[0131] Therefore, the element content in the mixture (including internal ash, environmental ash, binder, and recycled material) is as follows:

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138] in, The total content of a certain element in the mixture;

[0139] In the rotary hearth furnace production process, there are restrictive requirements on two key indicators: the chloride-zinc ratio and the carbon-oxygen ratio, expressed as follows:

[0140]

[0141]

[0142] in, The ratio of chlorine to zinc in the mixture; The carbon-to-oxygen ratio in the mixture; given a total material quantity Under the premise of [specific conditions], the total feed rate of internal ash (silos 1-9) for:

[0143]

[0144] Dry basis feed rate of each material in the internal ash mixing for:

[0145]

[0146] Wet basis feed rate of each material in the internal ash mixing for:

[0147] .

[0148] In a preferred embodiment of the present invention, the predicted mixing ratio of the silo and the opening and closing status of the silo are taken into consideration, such as... Figure 1 As shown, a dynamic ingredient mixing model is established:

[0149] The total proportion of the ingredients in the specified internal batches (bins 1-9) is 100%, that is:

[0150] Establish constraints for chlorine (Cl), zinc (Zn), carbon (C), carbon-oxygen ratio (C / O), and chlorine-zinc ratio (Cl / Zn):

[0151] (1)

[0152] in, and These are the upper and lower limits of each element or element ratio in the mixture; This refers to the content of each element in the mixture, i.e., its mass fraction. This represents the ratio of carbon to oxygen. This represents the ratio of chlorine to zinc.

[0153] Consider the current open / closed state variables of the silo. 0 indicates that the silo is closed, and 1 indicates that the silo is open. The material distribution constraints for each silo are as follows:

[0154] (2)

[0155] Where bigM represents a sufficiently large constant; and These represent the lower and upper limits of the material mixing ratio in the silo, respectively.

[0156] Consider the current silo The feed rate of the belt is limited and the following feed constraints must be met:

[0157] (3)

[0158] The method for constructing the ingredient objective function is as follows:

[0159] During the batching process, the rotary hearth furnace batching scheme is continuously adjusted to meet constraints as the raw material composition changes. The adjusted formula should be as close as possible to the predicted batching scheme to reduce fluctuations. Furthermore, frequent opening and closing of the silos should be avoided as much as possible during this process. The objective function is:

[0160]

[0161] in, For silos The proportions are the proportion decision variables; To predict the proportions of each material in the batching scheme; For the decision of opening and closing the silo, This indicates the current actual open / closed status of the silo. The material level control function is proportional; The weighting coefficient set by the user is used to adjust the material level; The larger the value, the greater the proportion of the material level adjustment function in the model, and the greater the impact of the material level on the model; an appropriate value should be set. The value is used to balance the two model objectives of material level adjustment and batching stability; although and Both can adjust the proportion of the material level control function, but This involves adjusting the material level function as a whole, treating it as a variable quantity. When the material level is within the normal range, it ensures that the values ​​of the two target functions are roughly on the same order of magnitude, preventing the material level adjustment function from becoming dominant. It is only related to the actual material level of each silo, and changes with the material level, so that the model will prioritize adjusting the material level of a silo when the material level of a certain silo is abnormal.

[0162] The dynamic model of ingredient proportioning is expressed as the following mathematical programming model:

[0163]

[0164] By introducing auxiliary variables, the above model can be transformed into a linear programming model, which can then be solved directly to obtain the optimal proportion of each silo.

[0165] For production safety reasons, the material levels in each silo need to fluctuate within a safe range as much as possible. To meet this condition, the discharge ratio of the current batching plan needs to be adjusted based on the material levels in each silo in the predicted batching plan. The predicted batching plan is generated by the XGBoost prediction model using data from previous batching plans. In actual production, the current material level in the silo can be read online before batching begins. Material level control can be achieved by controlling the opening and closing of the silos and adjusting the parameters before the decision variables.

[0166] This technical solution uses Python as the programming language, and the model solver is written and run on the PyCharm platform. The effectiveness of the element balance model is verified separately using comparative analysis and simulation methods. Ten sets of data were selected from historical batching schemes. With all other input quantities remaining identical, the advantages and disadvantages of the scheme calculated by the program were compared with the original scheme, as shown in Table 1.

[0167] Table 1. Elemental Balance Model Batching Scheme and Historical Batching Scheme

[0168] Historical Plan 1 0.245 0.472 0.534 no Improvement Plan 0.056 0.310 0.876 yes 2.38s Historical Option 2 0.181 0.478 0.595 no Improvement Plan 0.069 0.320 0.895 yes 3.83s Historical Option 3 0.212 0.475 0.783 no Improvement Plan 0.037 0.291 0.998 yes 2.99s Historical Option 4 0.18 0.392 0.667 no Improvement Plan 0.029 0.288 1 yes 3.53s Historical Plan 5 0.1844 0.475 0.767 no Improvement Plan 0.051 0.293 1 yes 3.68s Historical Plan 6 0.08 0.392 0.668 no Improvement Plan 0.029 0.288 0.99 yes 3.31s Historical Plan 7 0.218 0.263 0.272 no Improvement Plan 0.048 0.250 0.725 yes 2.54s Historical Plan 8 0.32 0.127 0.504 no Improvement Plan 0.026 0.128 0.721 yes 2.73s Historical Plan 9 0.1 0.305 0.694 no Improvement Plan 0.013 0.261 0.7 yes 2.34s Historical Plan 10 0.095 0.271 0.722 yes Improvement Plan 0.014 0.198 0.801 yes 2.40s Average change Reduced by 79.5% Decrease by 28.0% An increase of 40.2%

[0169] The results in the table show that the elemental balance model reduces the difference between adjacent batching schemes, improves the stability of the batching schemes, lowers the chloride-zinc ratio, and increases the carbon-oxygen ratio. The batching schemes solved by the model are more beneficial to actual production compared to historical batching schemes. Furthermore, in actual factory batching processes, Excel is often used for batching calculations, which is cumbersome, time-consuming, and requires manual operation. This technical solution, however, uses automatic control and can complete batching in just 2-3 seconds, greatly improving production efficiency.

[0170] In a preferred embodiment of the present invention, the method for optimizing the parameters of the dynamic batching model is as follows:

[0171] When the material level in the silo approaches or exceeds the upper limit of the material level. When the material level approaches or exceeds the lower limit, When the material level is high, but far from the safe level, In other cases, The material level control function of the corresponding silo will not affect the batching model:

[0172]

[0173] in, Indicates the material level in the silo. Indicates the upper limit of the safe material level. This indicates that the material level is relatively high, but not exceeding the safe material level value. Indicates the lower limit of the safe material level;

[0174] The proportion α of the level control function significantly influences the importance and priority of level control in the batching model. The value of α greatly affects the batching scheme solved by the model. Studying the value of α helps the model balance the two major objectives of batching stability and level safety. To better observe changes in the level of each silo and evaluate the level control function under different α values, the initial level is set as the upper limit of the level of each silo. This initial condition setting allows us to explore the impact of different α values ​​on the model's level and proportion under extreme conditions, when level adjustment is required in each silo. Furthermore, this can provide a reference for selecting an appropriate α value when multiple silo level anomalies occur in actual production.

[0175] like Figure 2 As shown in the graph, the material level changes in four silos (202, 204, 207, and 208) under different α values ​​demonstrate that the value of α does indeed affect the material level in the silos. Specifically, when α > 0, the material level is more prone to "fluctuations" and "turning points," stabilizing within a certain range. When α = 0, the material level tends to change continuously according to a certain trend until it reaches the safety level limit. For example, in silo 202, when α = 0, the silo level continuously decreases until it reaches the lower limit of the safety level before gradually rising again. This is because the opening and closing control functions of the silos and feeding equipment in the model still exist, and all silo levels will promptly return to the normal range when they exceed the safety level.

[0176] like Figure 3 As shown, At times, the material levels in some silos decreased more slowly, with a flatter curve. The material levels in silos 206, 209, and 210 remained close to the upper limit of the safe material level, only slowly decreasing after multiple batching operations. This situation may lead to prolonged shutdown of the feeding equipment, affecting the continuity of raw material transportation and batching production.

[0177] different The sum of squares of the differences between adjacent ingredient plans is as follows: Figure 4 As shown, The larger the value of α, the larger the overall function value for the objective of stable batching schemes. When α = 0.001 and α = 0, the objective function value is close to 0, and the change lines of the objective function values ​​for both are basically overlapping, indicating a certain marginal effect. When the material level control function ratio α is smaller, the overall difference between adjacent batching schemes is smaller, but a certain marginal effect still exists. In actual production, a reasonable value of α should be set according to the main objective of the model; an α value between 10 and 20 is preferable. -3 Up to 10 -2 Between these two orders of magnitude;

[0178] In the formulation, chlorine is a harmful element in the production process, while zinc is a valuable element that needs to be recycled; therefore, the chlorine-zinc ratio needs to be as low as possible. Different chlorine-zinc ratio limits were set, and the success rate of the model solution under different limits was investigated, i.e., the percentage of times the model had a solution out of the total number of solutions, as shown in Table 2.

[0179] Table 2 Results of the Chlorine-Zinc Ratio Study Procedure

[0180]

[0181] As shown in the table, for stable model solutions, the chlorine-zinc ratio should be set above 0.4. If production requirements for the chlorine-zinc ratio are more stringent, its limiting range can be appropriately narrowed. Furthermore, the data in the table shows that under the current parameter conditions, the chlorine-zinc ratio mostly fluctuates within the range of 0-0.2.

[0182] when When the concentration is ≥1, reduction can proceed more thoroughly, and excess carbon can also cause carburization of the pellets. Therefore, it is advisable to first investigate... The solution process of the model when ≥1. Assume 0≤ ≤1、1≤ ≤1.2, 1.2≤ ≤1.4, 1.4≤ ≤1.6、 The model was investigated for five carbon-oxygen ratio ranges ≥1. The proportion of successful solutions (model with solutions) out of the total number of solutions was calculated. The specific results are shown in Table 3.

[0183] Table 3 Simulation results of carbon-oxygen ratio (Part 1)

[0184]

[0185] The data in the table shows that it is difficult to stably solve the model when the carbon-oxygen ratio is greater than 1 using the existing parameters. This may be due to the excessive oxides and low carbon content in the raw materials. Furthermore, for economic reasons, the cost of dry quenching is higher than that of solid waste in the plant, making a carbon-oxygen ratio of less than 1 more conducive to complete carbon reaction, thereby achieving cost savings. However, to ensure the reduction reaction is as complete as possible... It also needs to be within a reasonably high range. Therefore, the experiment was further subdivided into 0≤ ≤1, exploring when When ≤1, the model's execution results are shown in Table 4:

[0186] Table 4 Simulation Results of Carbon-Oxygen Ratio (Part 2)

[0187]

[0188] As shown in the table, the parameter should be set within 0.2 ≤ When the value is ≤1, the model can be solved stably, and the C / O ratio can reach a reasonably high value, so that the oxygen in the pellets can be reduced as much as possible.

[0189] To explore the optimal combination of carbon-oxygen ratio and chlorine-zinc ratio, different ranges of carbon-oxygen ratio and chlorine-zinc ratio were arranged and combined. The success rate of the model solution is shown in Table 5.

[0190] Table 5 Success Rate Table

[0191]

[0192] Since the chloride-zinc ratio needs to be as small as possible, while the carbon-oxygen ratio needs to be appropriately large, 0 ≤ [missing value] is chosen to ensure stable solution of the model. ≤0.4, 0.2≤ The combination of constraints ≤1 is quite suitable. However, in actual production, to cope with more complex production conditions and ensure production stability, the range of constraints selected for the model can be appropriately expanded.

[0193] The present invention also provides a dynamic batching optimization system for rotary hearth furnaces, including a processing unit, wherein the processing unit executes the method described in the present invention to optimize the automatic batching of rotary hearth furnaces.

[0194] This system enables automated ingredient dispensing, making it easy to use.

[0195] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0196] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for optimizing dynamic batching in a rotary hearth furnace, characterized in that, Includes the following steps: Based on machine learning models and historical data, predict the expected batching scheme for each silo; Based on the element conservation mechanism and the process requirements of rotary hearth furnace, the constraints and objective function of the dynamic batching model are constructed. The parameters of the dynamic batching model are optimized to obtain the optimized dynamic batching model; Integrate the dynamic batching model with the process control system online; Considering the predicted batching ratio of the silos and the opening and closing status of the silos, a dynamic batching model is established: The specified proportions of ingredients are as follows: the sum of the proportions in bins 1-9 is 100%, that is: ， Establish constraints for chlorine (Cl), zinc (Zn), carbon (C), carbon-oxygen ratio (C / O), and chlorine-zinc ratio (Cl / Zn): （1）， in, and These are the upper and lower limits of each element or element ratio in the mixture; This refers to the content of each element in the mixture, i.e., its mass fraction. This represents the ratio of carbon to oxygen. This represents the ratio of chlorine to zinc. Consider the current open / closed state variables of the silo. 0 indicates that the silo is closed, and 1 indicates that the silo is open. The material distribution constraints for each silo are as follows: （2）， Where bigM represents a constant; and These represent the lower and upper limits of the material mixing ratio in the silo, respectively. Consider the current silo The feed rate of the belt is limited and the following feed constraints must be met: （3）， The method for constructing the ingredient objective function is as follows: During the batching process, the rotary hearth furnace batching scheme is continuously adjusted to meet constraints as the raw material composition changes. The adjusted formula should be close to the predicted batching scheme to reduce fluctuations. The objective function is: ， in, For silos The proportions are the matching decision variables; To predict the proportions of each material in the batching scheme; For the decision of opening and closing the silo, This indicates the current actual open / closed status of the silo. The material level control function is proportional; The weighting coefficient set by the user is used to adjust the material level; The larger the value, the greater the proportion of the material level adjustment function in the model, and the greater the impact of the material level on the model; set The value is used to balance the two model objectives of material level adjustment and batching stability; when the material level is within the normal range, the values ​​of the two objective functions are on the same order of magnitude to avoid the material level adjustment function from becoming dominant. It is only related to the actual material level of each silo and changes with the material level, so that the model will prioritize adjusting the material level of that silo when the material level is abnormal. The dynamic model of ingredient proportioning is expressed as the following mathematical programming model: ， By introducing auxiliary variables, the mathematical programming model is transformed into a linear programming model, which can be solved directly to obtain the optimal ratio of each silo.

2. The dynamic batching optimization method for rotary hearth furnace as described in claim 1, characterized in that, The XGBoost prediction model is introduced, and the objective is constrained by a penalty function. The constraint is added to the objective function to predict the ingredient ratio. The specific steps are as follows: exist At any given moment, select past moments from the historical database. This is a collection of ingredient formulas. , , Indicates a specific moment in the past; Indicates the first Individual material silos; Indicates the past The corresponding silo in the time-based batching plan The ratio; assuming any point in the past Ingredient formulation and time window If there is a mapping relationship between the data, then construct the following dataset. : ， in, A time-delay window is used to reduce the amount of data and improve the model's learning speed; Indicates the past The corresponding silo in the time-based batching plan The ratio; n represents the total ingredient ratio; because and If a correlation exists between them, then the following time series model is constructed: ， in, express Predicted values ​​for the ingredient batching plan at any given time; express and There is a function mapping relationship between them.

3. The dynamic batching optimization method for rotary hearth furnace as described in claim 2, characterized in that, Construct a constrained XGBoost prediction model: The loss function for XGBoost is defined as follows: ， Based on the principle of keeping the total weight of the configuration constant, the prediction output is constrained within a certain range. Within, that is: ， By introducing the constraints into XGBoost in the form of penalty functions, a new loss function is obtained: ， In the dataset Based on this, gradient descent is used to adjust the loss function of XGBoost. Optimize to obtain a trained XGBoost model and get the time step. Ingredient formulation Ultimately, we obtain the result at the current moment. Predictive ingredient scheme dataset : 。 4. The dynamic batching optimization method for rotary hearth furnace as described in claim 1, characterized in that, The silo proportioning decision variables include the proportion of dry quenching coke. %, electric furnace ash ratio %, proportion of ash from the iron tapping site %, Proportion of No. 1 dry-wet mixed sludge %, Proportion of No. 2 dry-wet mixed sludge %, Proportion of sludge recovered from No. 1 electric furnace %, Proportion of sludge recovered from No. 2 electric furnace %, proportion of secondary ash from blast furnace %, cold-rolled slurry proportion %, environmental gray ratio %, adhesive ratio %.

5. The dynamic batching optimization method for rotary hearth furnace as described in claim 4, characterized in that, The specific method for establishing an elemental balance model is as follows: The product after mixing is set to have requirements only for the content of four elements: C, Cl, Zn, and O. Determine the content of the main elements in the internal ash mixture, and then calculate the element content in the mixture based on this. The content of each component in the internal ash mixture is the sum of the products of the mass fraction of different elements in the corresponding silo and their proportions, expressed as: ， ， ， ， in, For silos The content of element C, i.e., its mass fraction. This refers to the total content of carbon (C) in the internal ash mixture. For silos The content of element Cl, i.e., its mass fraction. This refers to the total content of the internal ash element Cl; For silos The content of element Zn, i.e., its mass fraction. This refers to the total content of the internal ash element Zn; For silos The content of element Fe, i.e., its mass fraction. This refers to the total content of Fe, an internal ash element. For silos The proportions; Considering only the combination of oxygen with iron and zinc, calculate the oxygen content combined with iron. and the oxygen content bound to zinc : ， ， in, Indicates silo For the corresponding Content, i.e., mass fraction; Total oxygen content in internal ash It is the sum of the two, which simplifies to: ， Oxygen content in return silos and environmental ash silos , They are respectively: ， ， Therefore, the content of each element in the mixture is: ， ， ， ， ， ， in, It represents the total content of a certain element in the mixture.

6. The dynamic batching optimization method for rotary hearth furnace as described in claim 5, characterized in that, In the rotary hearth furnace production process, there are restrictive requirements on two key indicators: the chloride-zinc ratio and the carbon-oxygen ratio, expressed as follows: ， ， in, The ratio of chlorine to zinc in the mixture; The carbon-to-oxygen ratio in the mixture; given a total material quantity Under the premise of internal ash distribution: total feed rate of silos 1-9 for: ， Dry basis feed rate of each material in the internal ash mixing for: ， Wet basis feed rate of each material in the internal ash mixing for: 。 7. A dynamic batching optimization system for a rotary hearth furnace, characterized in that, The system includes a processing unit that executes the method described in any one of claims 1-6 to control the discharge rate of each hopper and achieve automatic optimized batching of the rotary hearth furnace.

Images