Method and equipment for optimizing steelmaking process in steel industry and storage medium

Abstract

The invention relates to the technical field of low-carbon process optimization in the iron and steel industry. The invention provides a steel industry steelmaking process optimization method and device and a storage medium. The method comprises the following steps that a parameterization process of a steelmaking technology is established, materials, energy consumption, carbon emission coefficients and cost parameters are defined, and a steelmaking process parameterization model is constructed; based on the steelmaking process parameter model, defining a state space and an action space of the steelmaking process, and setting a multi-target reward function and a corresponding constraint violation penalty term to obtain a reinforcement learning model; the reinforcement learning model is trained based on a Q-learning algorithm, a Q value is updated through a state-action-reward interactive cycle, and a selection strategy is dynamically adjusted; and performing simulation verification on the selection strategy output by the reinforcement learning model. Therefore, the problems of poor dynamic adaptability, single optimization target and unstable technical transition in the existing low-carbon path research method in the iron and steel industry are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of low-carbon process optimization in the steel industry, and in particular to a steelmaking process optimization method, equipment and storage medium in the steel industry. Background Art

[0002] As a core component of the global industrial system, the steel industry has long faced challenges of high energy consumption and high carbon emission intensity.

[0003] At present, the research methods for the low-carbon path of the steel industry can be divided into three categories: 1) Bottom-up models (such as MESSAGE, AIM / Enduse models) optimize technology paths based on the principle of cost minimization, but they are not adaptable enough to market dynamics; 2) Material flow analysis methods reveal emission reduction potential by tracking the flow of materials throughout the life cycle, but lack the ability to dynamically predict technology diffusion; 3) Top-down models (such as computable general equilibrium models) focus on the interaction of macroeconomic systems, but have difficulty dealing with multi-objective optimization problems.

[0004] The limitations of existing technologies are mainly reflected in: Poor dynamic adaptability: Traditional models rely on preset parameters and explicit constraint equations, making it difficult to cope with dynamic environments such as policy adjustments and market fluctuations; Single optimization target: Most models focus on optimizing only one target, namely cost or carbon emission, and lack the ability to coordinate optimization of multiple targets (such as cost, carbon emission reduction, and market share). Unsmooth technology transition: Failure to consider system stability when switching technology paths may lead to production fluctuations or waste of resources. Summary of the invention

[0005] The purpose of the present invention is to provide a steelmaking process optimization method, equipment and storage medium for the steel industry, aiming to solve the problems of poor dynamic adaptability, single optimization target and unstable technology transition existing in the existing low-carbon path research methods in the steel industry.

[0006] The embodiments of the present invention are implemented by the following technical solutions: A method for optimizing steelmaking process in the steel industry comprises the following steps: Establish a parameterized process including three steelmaking technologies: blast furnace-converter steelmaking, scrap steel steelmaking, and reduced iron steelmaking. Define materials, energy consumption, carbon emission coefficients, and cost parameters, and build a parameterized model for steelmaking processes. Based on the steelmaking process parameter model, the state space and action space of the steelmaking process are defined, and a multi-objective reward function and corresponding constraint violation penalty items are set to obtain a reinforcement learning model. The state space includes market share, total cost, carbon emissions, output, and technical stability; the action space is the selection strategy of the three steelmaking technologies. The reinforcement learning model is trained based on the Q-learning algorithm, and the Q value is updated through the interactive cycle of state-action-reward to dynamically adjust the selection strategy; The selection strategy of the reinforcement learning model output is simulated and verified. If the simulation result does not meet the expected optimization goal, the corresponding parameters of the reinforcement learning model are adjusted through the offline gradient descent method and the reinforcement learning model is retrained until the optimization goal is met.

[0007] Optionally, market price fluctuation factors are introduced to dynamically adjust energy unit prices and material unit prices, optimize equipment efficiency parameters based on historical data, and correct the energy and material consumption coefficients of each process link.

[0008] Optionally, the specific process of introducing market price fluctuation factors, dynamically adjusting energy unit prices and material unit prices, optimizing equipment efficiency parameters based on historical data, and correcting energy and material consumption coefficients of each process link is as follows: Establish a market database to collect historical price data and supply and demand relationship data of energy and materials; Use time series analysis or machine learning algorithms to establish a price fluctuation prediction model to quantify the fluctuation range and frequency of energy unit prices and material unit prices; Generate energy unit price fluctuation factors and material unit price fluctuation factors based on the price fluctuation prediction model, and update the fluctuation factors through the sliding window method to ensure that the price parameters are synchronized with the real-time market dynamics; Construct an equipment efficiency evaluation index system, including energy conversion efficiency, material utilization rate, and equipment failure rate, and use particle swarm optimization algorithm or genetic algorithm to optimize equipment efficiency parameters with historical operation data as input; Based on the optimized equipment efficiency parameters, the energy consumption coefficient and material consumption coefficient of each process link are recalculated, and the uncertainty of the revised parameters is quantified using the Bayesian inference method, and the confidence interval is constructed. The revised parameter model is verified through the process simulation platform to ensure that the prediction error remains ≤5% within the ±10% market fluctuation range; Set the parameter update cycle, trigger instant updates when market fluctuations exceed the threshold, and establish a parameter version management system to record the fluctuation factors, equipment efficiency parameters and correction basis for each adjustment.

[0009] Optionally, the specific process of defining the state space and action space of the steelmaking process based on the steelmaking process parameter model, setting the multi-objective reward function and the corresponding constraint violation penalty term, and obtaining the reinforcement learning model is: Taking market share, total cost, carbon emissions, output and technical stability as state variables, a five-tuple state vector is constructed, as shown in the following formula (1):

[0010] in, Display Technology In time market share; Indicates time Total steelmaking cost; Indicates time carbon emissions; Indicates time of steel production; represents technological stationarity, i.e., the absolute value of the change in market share between adjacent time steps; The market share distribution ratio of the three steelmaking technologies is taken as the action, as shown in the following formula (2):

[0011] in, Indicates that the agent is Optional actions in the state; , and Respectively represent the market share distribution ratios corresponding to the three steelmaking technologies; and Respectively represent the upper and lower limits of market share; The reward function is shown in the following formula (3):

[0012] , and The expressions of are shown in equations (4), (5) and (6):

[0013]

[0014]

[0015] in, , , and Respectively represent the weight coefficients of the corresponding state vectors; , and Respectively represent the penalty coefficients of the corresponding penalty items; represents the penalty term of carbon emission constraint; represents the penalty term of market share constraint; represents the penalty term for the steel production constraint.

[0016] Optionally, the specific process of training the reinforcement learning model based on the Q-learning algorithm, updating the Q value through the state-action-reward interactive cycle, and dynamically adjusting the selection strategy is: Initialize the Q-values ​​of all state-action pairs to random values ​​or zero; Based on the current state and ε-greedy strategy selects actions , where the ε-greedy strategy randomly explores the action space with ε probability and selects the action with the largest current Q value with 1-ε probability; Execute the selected action , triggering the environment state from Transfer to , and calculate the instant reward according to the reward function, wherein the reward function includes a market share reward term, a cost penalty term, a carbon emission penalty term, a technical stability penalty term, and a constraint violation penalty term; The Q value is updated using formula (7), which is as follows:

[0017] in, represents the Q value; represents the Q value learning rate; represents the discount factor used to control the impact of future rewards; The maximum Q value in the next state, indicating the reward after the agent chooses the best action; The behavior strategy is iteratively updated through the following formula (8), as shown in formula (8):

[0018] in, Indicates that the agent is in state Select the steelmaking process below.

[0019] Optionally, gradually increase the penalty coefficient in the later stages of training , and , ensuring that the technology path selected by the intelligent agent meets the carbon emission cap, market share constraints and output constraints.

[0020] Optionally, the selection strategy for the output of the reinforcement learning model is simulated and verified. If the simulation result does not achieve the expected optimization goal, the specific process of adjusting the corresponding parameters of the reinforcement learning model by the offline gradient descent method and retraining the reinforcement learning model is: Collect historical data, including market share, total cost, carbon emissions and market share changes of various steelmaking processes; Construct the objective function as shown in the following formula (9):

[0021] Among them, the reward function It includes market share bonus, cost penalty, carbon emission penalty and technical stability penalty; Calculate the objective function for each weight coefficient , , and The gradient of is shown in the following formula (10):

[0022] in, , , and Respectively represent the weight coefficient , , and Calculate partial derivatives; update the weight coefficients using the following formula (11):

[0023] in, represents the weight learning rate; , , and Respectively represent The weight coefficients corresponding to market share, cost, carbon emissions and technological stability in the reward function at the iteration; , , and Respectively represent the corresponding weight coefficients after update; Indicates The discount factor of times; Represents the number of time steps of historical data.

[0024] Optionally, the specific process of establishing an accounting model for equipment depreciation cost, maintenance cost, environmental protection cost, energy cost, material cost and financial subsidy, constructing a carbon emission coefficient matrix for each process link, and calculating the total carbon emissions in combination with energy consumption data is as follows: Assume that the total cost includes equipment depreciation cost, equipment maintenance cost, environmental protection equipment investment cost, financial subsidies, and the energy cost and material cost required in different steelmaking processes, as shown in the following formula (12):

[0025] in, represents the total cost of steelmaking; represents the depreciation cost of equipment; represents the equipment maintenance cost; represents the investment cost of environmental protection equipment; represents the energy cost; The material cost expressed; It means financial subsidy; During the steelmaking process, the carbon emission expression is shown in the following formula (13):

[0026] in, represents the total amount of carbon emissions; Indicates the steelmaking process; Indicates steelmaking equipment; Indicates the type of energy; Represents the carbon emission factor.

[0027] Based on the same inventive concept, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned steel industry steelmaking process optimization method.

[0028] Based on the same inventive concept, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned photovoltaic power generation energy storage grid-connected optimization system and method are implemented.

[0029] The technical solution of the embodiment of the present invention has at least the following advantages and beneficial effects: Enhanced dynamic adaptability: By building a coupling system of parameterized process models and reinforcement learning models, real-time response capabilities to dynamic environments such as market fluctuations and policy adjustments are achieved; the Q-learning algorithm can autonomously update technology selection strategies through cyclic iterations of state-action-reward, breaking through the limitations of traditional models that rely on preset parameters.

[0030] Multi-objective collaborative optimization: The five-dimensional state space including market share, cost, carbon emissions, output and technical stability is introduced, and the collaborative optimization of economic and environmental benefits is achieved in combination with a multi-objective reward function. Compared with the traditional single-objective optimization model, it can simultaneously pursue cost minimization and market competitiveness improvement under the constraint of carbon emission reduction.

[0031] Stability assurance for technology transition: Incorporate technology stability into the state space, and effectively suppress drastic switching of technology paths through a constraint violation penalty mechanism; combined with parameter calibration of the offline gradient descent method, it can ensure that the production system maintains stable operation during the technology iteration process, avoiding resource waste and production capacity fluctuations.

[0032] Improved model generalization capability: The parameterized process model uniformly models the three mainstream steelmaking processes of blast furnace-converter, scrap steel, and reduced iron. The reinforcement learning framework has the ability to optimize strategies across technical paths. Through simulation verification and parameter fine-tuning mechanisms, it can quickly adapt to the energy structure, policy environment, and market demand of different regions.

[0033] Intelligent decision support: It breaks through the static analysis mode of traditional models and can dynamically generate the optimal technology combination plan; by quantifying the marginal emission reduction costs and market benefits of each technical path, it provides scientific decision-making basis for steel companies to formulate medium- and long-term low-carbon transformation plans.

[0034] Prediction of low-carbon technology diffusion: Reinforcement learning models can simulate the dynamic evolution of technology selection strategies, reveal the diffusion laws of different technology paths in market competition, and provide theoretical support for promoting the large-scale application of low-carbon technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic diagram of a process for optimizing a steelmaking process in the steel industry according to an embodiment of the present invention; Figure 2 It is a schematic diagram of a blast furnace-converter steelmaking process flow in a steel industry steelmaking process optimization method according to an embodiment of the present invention; Figure 3 It is a schematic diagram of a scrap steelmaking process flow in a steelmaking process optimization method for the steel industry according to an embodiment of the present invention; Figure 4 It is a schematic diagram of the reduced iron steelmaking process flow in the steel industry steelmaking process optimization method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following is a specific implementation method in conjunction with the drawings.

[0037] Reference Figure 1 , a steelmaking process optimization method for the steel industry, comprising the following steps: Step 1: Establish a parametric process including three steelmaking technologies: blast furnace-converter steelmaking, scrap steel steelmaking, and reduced iron steelmaking, define materials, energy consumption, carbon emission coefficient and cost parameters, and construct a parametric model of the steelmaking process.

[0038] In some embodiments, the specific process of constructing the steelmaking process parameterized model is: Reference Figure 2, blast furnace-converter steelmaking parameterization: Define the consumption of coking coal, electricity, and coal. Define the consumption of coking coal. High-precision material metering equipment, such as belt scales, can be installed in the coking process of blast furnace-converter steelmaking to monitor the quality of coking coal entering the coking oven in real time and record the consumption per unit time. Electricity consumption can be measured by installing an electric meter at the power input port of the steelmaking equipment to accurately measure the amount of electricity consumed during the operation of the equipment and classify and count it according to different equipment or equipment groups. The definition of coal consumption is similar to coking coal. For coal used in other links (such as heating, etc.), its consumption is recorded through appropriate metering devices.

[0039] Establish a material balance equation for the conversion of coking coal into coke. Based on the principle of chemical reaction, analyze the composition changes of coking coal during the coking process, determine the conversion ratio of various elements (such as carbon, hydrogen, oxygen, etc.) in coking coal during the conversion into coke; collect relevant data in the coking process, such as the initial mass of coking coal, component analysis data, and the mass and composition data of the output coke. Through the law of conservation of materials, establish a mathematical relationship between the mass of coking coal and the mass of coke to form a material balance equation. For example, assuming that most of the carbon element in coking coal is converted into the carbon element in coke, part of the equation can be constructed based on the conservation of mass of carbon element.

[0040] Consumption models for coke, iron ore, electricity, and coal are established to correlate the proportional relationship between sinter output and raw material input. The consumption models for coke, iron ore, electricity, and coal can be established through long-term production data recording and analysis. During the production process, the input of these raw materials and energy in each batch of production, as well as the corresponding product output (such as molten iron output, etc.) are recorded simultaneously; statistical analysis methods, such as regression analysis, are used to find the intrinsic relationship between these consumption and production indicators, and a consumption model is constructed. For example, the average amount of coke, iron ore, electricity, and coal required for each production of a certain mass of molten iron and their fluctuation range are analyzed; for the proportional relationship between sinter output and raw material input, in the sintering process, the input of iron ore, flux (such as limestone, etc.), fuel (such as coke powder, etc.), and the quality of the output sinter are recorded. Through the analysis of material balance and actual production data, the proportional relationship between them is determined, such as how much mass of sinter can be produced for each input of a certain mass of iron ore and other raw materials.

[0041] Based on the consumption of coke, sintered ore, electricity, and coal, the mass conservation equation of molten iron output is constructed. Combined with the consumption model and related data established previously, according to the principle of mass conservation, it is analyzed how the input of coke, sintered ore, electricity, coal, etc. is converted into molten iron output during the blast furnace ironmaking process; considering some material losses that may exist in the ironmaking process (such as part of the iron elements taken away by the slag, etc.), these losses are quantified and corrected through detailed research and data statistics of the production process. Thus, a mass conservation equation with the consumption of coke, sintered ore, electricity, and coal as independent variables and the output of molten iron as the dependent variable is established.

[0042] Quantify the consumption parameters of molten iron, scrap steel, electricity, coal, natural gas, and coke, and establish a linear relationship between crude steel output and material input. In the converter steelmaking process, the actual consumption of molten iron, scrap steel, electricity, coal, natural gas, coke, etc. is accurately measured through metering devices installed on various equipment and material transportation pipelines; detailed records are kept for each steelmaking production, including the amount of various materials input, energy consumption, and the quality of crude steel output; these data are analyzed and processed using mathematical methods such as linear regression to find out the linear relationship between crude steel output and the amount of material input such as molten iron, scrap steel, electricity, coal, natural gas, coke, etc., and establish a corresponding mathematical model. By defining each parameter and establishing related equations and models, we can gain an in-depth understanding of the flow and transformation relationship between materials and energy in the blast furnace-converter steelmaking process, providing a solid theoretical basis and data support for subsequent process optimization. For example, accurate material balance equations and mass conservation equations can help technicians clearly see the changes in materials at each link, so as to make targeted improvements.

[0043] Reference Figure 3 , parameterization of scrap steelmaking: Based on the molten iron flow rate produced by blast furnace-converter steelmaking, a model for the ratio of molten iron to scrap steel is established. Collect the molten iron flow rate data produced by blast furnace-converter steelmaking under different production conditions, such as recording the output of molten iron under different production shifts and different equipment operating conditions. These data can be obtained in real time through flow monitoring equipment installed in the blast furnace-converter steelmaking production line. Analyze and process the collected molten iron flow rate data, and establish a mathematical model to describe the reasonable ratio relationship between molten iron and scrap steel, taking into account the influence of factors such as the quality and specifications of scrap steel on the ratio of molten iron and scrap steel. For example, a linear regression model can be used to determine the linear relationship between the molten iron flow rate and the amount of scrap steel added by fitting historical production data, that is, assuming that the molten iron flow rate is , the amount of scrap steel added is , we can get of the form, where and It is a coefficient obtained through data fitting. In actual production, according to the real-time monitoring of the molten iron flow produced by blast furnace-converter steelmaking, the corresponding amount of scrap steel added is calculated using the established proportioning model to ensure the reasonable input of materials in the scrap steelmaking process.

[0044] Define the energy consumption coefficients of electricity, coal, and natural gas, as well as the proportional relationship between crude steel output and total material input. For the determination of the energy consumption coefficient of electricity, install an electricity monitoring device on the scrap steelmaking equipment to record the power consumption at different stages of the scrap steelmaking process (such as scrap steel preheating, smelting, etc.) in real time. By statistically analyzing the power consumption data of a large number of production batches, combined with the output of crude steel, calculate the electricity consumed per unit of crude steel production, and thus determine the energy consumption coefficient of electricity. For the energy consumption coefficient of coal, record the amount of coal used in the scrap steelmaking process, including information such as the type and quality of coal. Similarly, by analyzing the coal usage and crude steel output of multiple production batches, calculate the amount of coal consumed per unit of crude steel production, and obtain the energy consumption coefficient of coal. For the energy consumption coefficient of natural gas, install natural gas flow monitoring equipment to monitor the amount of natural gas used in the scrap steelmaking process. Calculate the amount of natural gas consumed per unit of crude steel production in a similar way to electricity and coal, and determine the energy consumption coefficient of natural gas. Clarify that the total material input includes molten iron, scrap steel, and other possible auxiliary materials. During the production process, the quality of each input of molten iron, scrap steel and auxiliary materials, as well as the quality of the final crude steel output, are accurately recorded. By analyzing multiple sets of production data, the proportional relationship between crude steel output and total material input is established. For example, of the form, where k It represents the proportional coefficient obtained through data statistics and analysis, which reflects the material conversion efficiency in the process of scrap steel making; represents the crude steel output, Represents the total material input. By establishing a ratio model for molten iron and scrap steel, the amount of scrap steel added can be reasonably adjusted according to the molten iron flow rate produced by blast furnace-converter steelmaking, avoiding material waste or shortage, thereby improving material utilization efficiency and reducing production costs. For example, when the molten iron flow rate is low, the proportion of scrap steel can be appropriately increased to ensure the continuity and stability of production, while reducing excessive dependence on molten iron.

[0045] Reference Figure 4, parameterization of direct reduced iron steelmaking: establish the relationship between the output of iron ore particles and the consumption of iron ore, electricity, and coal. Collect a large amount of direct reduced iron steelmaking production data, covering different production conditions, equipment operating parameters, etc. Organize and analyze the data to determine the relationship between the input and output of iron ore in the process of converting it into iron ore particles. At the same time, monitor the real-time consumption of electricity and coal in the process, and establish a mathematical model based on production data. For example, multiple linear regression and other methods can be used to determine the functional relationship between the output of iron ore particles and the input of iron ore, electricity, and coal, that is: ,in, represents the output of iron ore particles, represents the iron ore input, Indicates the power consumption, Indicates coal consumption.

[0046] Quantify the consumption of natural gas, electricity, and hydrogen, and construct a material balance between direct reduced iron output and raw materials. During the production of direct reduced iron, install high-precision metering equipment to accurately measure the real-time consumption of natural gas, electricity, and hydrogen. Based on the principles of chemical reactions and the law of conservation of matter, establish a material balance equation between direct reduced iron output and these raw material inputs. For example, assuming that the main chemical reaction in direct reduced iron production is: xF 2 O 3 +yCH 4 +zH 2 +mO 2 →nFe +pCO 2 +qH 2 O (The stoichiometric coefficients here are x , y , z , m , n , p , q According to the actual reaction), based on this reaction formula and the actual consumption of natural gas, hydrogen, etc., a material balance equation can be established, such as: ,in, represents the output of direct reduced iron, represents the natural gas consumption, Indicates the power consumption, Indicates hydrogen consumption.

[0047] Define the ratio parameters of direct reduced iron and scrap steel, as well as the energy consumption models of natural gas, electricity and coal, and establish a linear equation for crude steel output. In the actual production process, set different direct reduced iron and scrap steel ratio schemes, accurately control the input of the two through the automatic control system, and record the corresponding crude steel output and natural gas, electricity and coal consumption data. Define the ratio parameters of direct reduced iron and scrap steel ,in, is the quality of direct reduced iron, The quality of scrap steel is obtained by using grey correlation analysis, principal component analysis and other methods to find out the key influencing factors between the energy consumption of natural gas, electricity, and coal and the output of crude steel, and the ratio of direct reduced iron to scrap steel, and to construct an energy consumption model. For example, the natural gas energy consumption model is ,in, is the crude steel output, , , as model parameters; similarly, energy consumption models for electricity and coal are constructed. Finally, through the analysis and fitting of a large amount of production data, a linear equation is established between crude steel output and direct reduced iron input, scrap steel input, natural gas consumption, electricity consumption, coal consumption and other factors. The accurate consumption relationship and material balance model enable producers to predict the demand for raw materials and energy in the production process in advance, adjust the input in time, avoid production interruptions or inefficiencies caused by raw material shortages or energy waste, and greatly improve the controllability and stability of the production process.

[0048] Cost and carbon emission parameterization: Establish accounting models for equipment depreciation costs, maintenance costs, environmental protection costs, energy costs, material costs and financial subsidies; construct a carbon emission coefficient matrix for each process link, and calculate total carbon emissions based on energy consumption data.

[0049] In some embodiments, the calculation model of equipment depreciation cost, maintenance cost, environmental protection cost, energy cost, material cost and financial subsidy is established, the carbon emission coefficient matrix of each process link is constructed, and the specific process of calculating the total carbon emissions in combination with energy consumption data is as follows: Assume that the total cost includes equipment depreciation cost, equipment maintenance cost, environmental protection equipment investment cost, financial subsidies, and the energy cost and material cost required in different steelmaking processes, as shown in the following formula (12):

[0050] in, represents the total cost of steelmaking; represents the depreciation cost of equipment; represents the equipment maintenance cost; represents the investment cost of environmental protection equipment; represents the energy cost; The material cost expressed; It means financial subsidy; During the steelmaking process, the carbon emission expression is shown in the following formula (13):

[0051] in, represents the total amount of carbon emissions; Indicates the steelmaking process; Indicates steelmaking equipment; Indicates the type of energy; Represents the carbon emission factor.

[0052] Introduce market price fluctuation factors, dynamically adjust energy unit prices and material unit prices, optimize equipment efficiency parameters based on historical data, and correct the energy and material consumption coefficients of each process link.

[0053] In some embodiments, the specific process of introducing market price fluctuation factors, dynamically adjusting energy unit prices and material unit prices, optimizing equipment efficiency parameters based on historical data, and correcting energy and material consumption coefficients of each process link is as follows: Establish a market database to collect historical price data and supply and demand relationship data of energy and materials; Use time series analysis or machine learning algorithms to establish a price fluctuation prediction model to quantify the fluctuation range and frequency of energy unit prices and material unit prices; Generate energy unit price fluctuation factors based on price fluctuation prediction model and material price fluctuation factor , the expression is shown in the following formula (14):

[0054] in, and Respectively represent energy and materials The base price of and Indicates time The corresponding dynamic adjustment price; the volatility factor is updated through the sliding window method to ensure that the price parameters are synchronized with the real-time market dynamics; Construct an equipment efficiency evaluation index system, including energy conversion efficiency, material utilization rate, and equipment failure rate, and use particle swarm optimization algorithm or genetic algorithm to optimize equipment efficiency parameters with historical operation data as input; establish a mapping relationship between equipment efficiency parameters and process parameters. For example: ,in, Indicates the corrected energy consumption; represents the baseline energy consumption; Represents energy conversion efficiency.

[0055] Based on the optimized equipment efficiency parameters, the energy consumption coefficient and material consumption coefficient of each process link are recalculated, and the uncertainty of the revised parameters is quantified using the Bayesian inference method, and the confidence interval is constructed. The revised parameter model is verified through the process simulation platform to ensure that the prediction error remains ≤5% within the ±10% market fluctuation range; Set the parameter update cycle, trigger instant updates when market fluctuations exceed the threshold, and establish a parameter version management system to record the fluctuation factors, equipment efficiency parameters and correction basis for each adjustment.

[0056] Step 2: Based on the steelmaking process parameter model, define the state space and action space of the steelmaking process, set the multi-objective reward function and the corresponding constraint violation penalty items, and obtain the reinforcement learning model; among which, the state space includes market share, total cost, carbon emissions, output and technical stability; the action space is the selection strategy of the three steelmaking technologies.

[0057] In some embodiments, based on the steelmaking process parameter model, the state space and action space of the steelmaking process are defined, a multi-objective reward function and corresponding constraint violation penalty items are set, and the specific process of obtaining the reinforcement learning model is as follows: Taking market share, total cost, carbon emissions, output and technical stability as state variables, a five-tuple state vector is constructed, as shown in the following formula (1):

[0058] in, Display Technology In time market share; Indicates time Total steelmaking cost; Indicates time carbon emissions; Indicates time of steel production; represents technological stationarity, i.e., the absolute value of the change in market share between adjacent time steps; The market share distribution ratio of the three steelmaking technologies is taken as the action, as shown in the following formula (2):

[0059] in, Indicates that the agent is Optional actions in the state; , and Respectively represent the market share distribution ratios corresponding to the three steelmaking technologies; and Respectively represent the upper and lower limits of market share; The reward function is shown in the following formula (3):

[0060] , and The expressions of are shown in equations (4), (5) and (6):

[0061]

[0062]

[0063] in, , , and Respectively represent the weight coefficients of the corresponding state vectors; , and Respectively represent the penalty coefficients of the corresponding penalty items; represents the penalty term of carbon emission constraint; represents the penalty term of market share constraint; represents the penalty term for the steel production constraint.

[0064] Step 3: Train the reinforcement learning model based on the Q-learning algorithm, update the Q value through the state-action-reward interaction cycle, and dynamically adjust the selection strategy.

[0065] In some embodiments, the reinforcement learning model is trained based on the Q-learning algorithm, and the Q value is updated through the state-action-reward interactive cycle. The specific process of dynamically adjusting the selection strategy is: Initialize the Q-values ​​of all state-action pairs to random values ​​or zero; Based on the current state and ε-greedy strategy selects actions , where the ε-greedy strategy randomly explores the action space with ε probability and selects the action with the largest current Q value with 1-ε probability; Execute the selected action , triggering the environment state from Transfer to , and calculate the instant reward according to the reward function, wherein the reward function includes a market share reward term, a cost penalty term, a carbon emission penalty term, a technical stability penalty term, and a constraint violation penalty term; The Q value is updated using formula (7), which is as follows:

[0066] in, represents the Q value; represents the Q value learning rate; represents the discount factor used to control the impact of future rewards; The maximum Q value in the next state, indicating the reward after the agent chooses the best action; The behavior strategy is iteratively updated through the following formula (8), as shown in formula (8):

[0067] in, Indicates that the agent is in state Select the steelmaking process below.

[0068] In some embodiments, the penalty coefficient is gradually increased in the later stages of training. , and , ensuring that the technology path selected by the intelligent agent meets the carbon emission cap, market share constraints and output constraints.

[0069] Step 4: Perform simulation verification on the selection strategy output by the reinforcement learning model. If the simulation result does not meet the expected optimization goal, adjust the corresponding parameters of the reinforcement learning model through the offline gradient descent method and retrain the reinforcement learning model until the optimization goal is met.

[0070] In some embodiments, the selection strategy output by the reinforcement learning model is simulated and verified. If the simulation result does not reach the expected optimization goal, the corresponding parameters of the reinforcement learning model are adjusted by the offline gradient descent method and the specific process of retraining the reinforcement learning model is as follows: Collect historical data, including market share, total cost, carbon emissions and market share changes of various steelmaking processes; Construct the objective function as shown in the following formula (9):

[0071] Among them, the reward function It includes market share bonus, cost penalty, carbon emission penalty and technical stability penalty; Calculate the objective function for each weight coefficient , , and The gradient of is shown in the following formula (10):

[0072] in, , , and Respectively represent the weight coefficient , , and Calculate partial derivatives; update the weight coefficients using the following formula (11):

[0073] in, Represents the weight learning rate; , , and Respectively represent The weight coefficients corresponding to market share, cost, carbon emissions and technological stability in the reward function at the iteration; , , and Respectively represent the corresponding weight coefficients after update; Indicates The discount factor of times; Represents the number of time steps of historical data.

[0074] Based on the same inventive concept, corresponding to the steelmaking process optimization method for the steel industry in any of the above-mentioned embodiments, the present invention provides an electronic device including a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to enable the electronic device to execute the steelmaking process optimization method for the steel industry in the embodiment.

[0075] Optionally, the above-mentioned electronic device may be a server.

[0076] In addition, this embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steel industry steelmaking process optimization method of the embodiment is implemented.

[0077] It is understandable that the processor in the embodiments of the present invention may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0078] The method steps in the embodiments of the present invention can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0079] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted through the storage medium. The computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)), etc.

Claims

1. A method for optimizing steelmaking process in the steel industry, characterized in that: The following steps are involved: Establish a parameterized process including three steelmaking technologies: blast furnace-converter steelmaking, scrap steel steelmaking, and reduced iron steelmaking. Define materials, energy consumption, carbon emission coefficients, and cost parameters, and build a parameterized model for steelmaking processes. Based on the steelmaking process parameter model, the state space and action space of the steelmaking process are defined, and a multi-objective reward function and corresponding constraint violation penalty items are set to obtain a reinforcement learning model. The state space includes market share, total cost, carbon emissions, output, and technical stability; the action space is the selection strategy of the three steelmaking technologies. The reinforcement learning model is trained based on the Q-learning algorithm, and the Q value is updated through the interactive cycle of state-action-reward to dynamically adjust the selection strategy; The selection strategy of the reinforcement learning model output is simulated and verified. If the simulation result does not meet the expected optimization goal, the corresponding parameters of the reinforcement learning model are adjusted through the offline gradient descent method and the reinforcement learning model is retrained until the optimization goal is met.

2. The method for optimizing steelmaking process in the steel industry according to claim 1, characterized in that: Introduce market price fluctuation factors, dynamically adjust energy unit prices and material unit prices, optimize equipment efficiency parameters based on historical data, and correct the energy and material consumption coefficients of each process link.

3. The method for optimizing steelmaking process in the steel industry according to claim 2, characterized in that: The specific process of introducing market price fluctuation factors, dynamically adjusting energy unit prices and material unit prices, optimizing equipment efficiency parameters based on historical data, and correcting energy and material consumption coefficients of each process link is as follows: Establish a market database to collect historical price data and supply and demand relationship data of energy and materials; Use time series analysis or machine learning algorithms to establish a price fluctuation prediction model to quantify the fluctuation range and frequency of energy unit prices and material unit prices; Generate energy unit price fluctuation factors and material unit price fluctuation factors based on the price fluctuation prediction model, and update the fluctuation factors through the sliding window method to ensure that the price parameters are synchronized with the real-time market dynamics; Construct an equipment efficiency evaluation index system, including energy conversion efficiency, material utilization rate, and equipment failure rate, and use particle swarm optimization algorithm or genetic algorithm to optimize equipment efficiency parameters with historical operation data as input; Based on the optimized equipment efficiency parameters, the energy consumption coefficient and material consumption coefficient of each process link are recalculated, and the uncertainty of the revised parameters is quantified using the Bayesian inference method, and the confidence interval is constructed. The revised parameter model is verified through the process simulation platform to ensure that the prediction error remains ≤5% within the ±10% market fluctuation range; Set the parameter update cycle, trigger instant updates when market fluctuations exceed the threshold, and establish a parameter version management system to record the fluctuation factors, equipment efficiency parameters and correction basis for each adjustment.

4. The method for optimizing steelmaking process in the steel industry according to claim 1, characterized in that: The specific process of defining the state space and action space of the steelmaking process based on the steelmaking process parameter model, setting the multi-objective reward function and the corresponding constraint violation penalty term, and obtaining the reinforcement learning model is as follows: Taking market share, total cost, carbon emissions, output and technical stability as state variables, a five-tuple state vector is constructed, as shown in the following formula (1): in, Display Technology In time market share; Indicates time Total steelmaking cost; Indicates time carbon emissions; Indicates time of steel production; represents technological stationarity, i.e., the absolute value of the change in market share between adjacent time steps; The market share distribution ratio of the three steelmaking technologies is taken as the action, as shown in the following formula (2): in, Indicates that the agent is Optional actions in the state; , and Respectively represent the market share distribution ratios corresponding to the three steelmaking technologies; and Respectively represent the upper and lower limits of market share; The reward function is shown in the following formula (3): , and The expressions of are shown in equations (4), (5) and (6): in, , , and Respectively represent the weight coefficients of the corresponding state vectors; , and Respectively represent the penalty coefficients of the corresponding penalty items; represents the penalty term of carbon emission constraint; represents the penalty term of market share constraint; represents the penalty term for the steel production constraint.

5. The method for optimizing steelmaking process in the steel industry according to claim 4, characterized in that: The specific process of training the reinforcement learning model based on the Q-learning algorithm, updating the Q value through the interactive cycle of state-action-reward, and dynamically adjusting the selection strategy is as follows: Initialize the Q-values ​​of all state-action pairs to random values ​​or zero; Based on the current state and ε-greedy strategy selects actions , where the ε-greedy strategy randomly explores the action space with ε probability and selects the action with the largest current Q value with 1-ε probability; Execute the selected action , triggering the environment state from Transfer to , and calculate the instant reward according to the reward function, wherein the reward function includes a market share reward term, a cost penalty term, a carbon emission penalty term, a technical stability penalty term, and a constraint violation penalty term; The Q value is updated using formula (7), which is as follows: in, represents the Q value; represents the Q value learning rate; represents the discount factor used to control the impact of future rewards; The maximum Q value in the next state, indicating the reward after the agent chooses the best action; The behavior strategy is iteratively updated through the following formula (8), as shown in formula (8): in, Indicates that the agent is in state Select the steelmaking process below.

6. The method for optimizing steelmaking process in the steel industry according to claim 5, characterized in that: Gradually increase the penalty coefficient in the later stages of training , and , ensuring that the technology path selected by the intelligent agent meets the carbon emission cap, market share constraints and output constraints.

7. The method for optimizing steelmaking process in the steel industry according to claim 4, characterized in that: The selection strategy for the output of the reinforcement learning model is simulated and verified. If the simulation result does not reach the expected optimization goal, the specific process of adjusting the corresponding parameters of the reinforcement learning model by the offline gradient descent method and retraining the reinforcement learning model is as follows: Collect historical data, including market share, total cost, carbon emissions and market share changes of various steelmaking processes; Construct the objective function as shown in the following formula (9): Among them, the reward function It includes market share bonus, cost penalty, carbon emission penalty and technical stability penalty; Calculate the objective function for each weight coefficient , , and The gradient of is shown in the following formula (10): in, , , and Respectively represent the weight coefficient , , and Calculate partial derivatives; update the weight coefficients using the following formula (11): in, Represents the weight learning rate; , , and Respectively represent The weight coefficients corresponding to market share, cost, carbon emissions and technological stability in the reward function at the iteration; , , and Respectively represent the corresponding weight coefficients after update; Indicates The discount factor of times; Represents the number of time steps of historical data.

8. The method for optimizing steelmaking process in the steel industry according to claim 4, characterized in that: The specific process of establishing the accounting model of equipment depreciation cost, maintenance cost, environmental protection cost, energy cost, material cost and financial subsidy, constructing the carbon emission coefficient matrix of each process link, and calculating the total carbon emissions in combination with energy consumption data is as follows: Assume that the total cost includes equipment depreciation cost, equipment maintenance cost, environmental protection equipment investment cost, financial subsidies, and the energy cost and material cost required in different steelmaking processes, as shown in the following formula (12): in, represents the total cost of steelmaking; represents the depreciation cost of equipment; represents the equipment maintenance cost; represents the investment cost of environmental protection equipment; represents the energy cost; The material cost expressed; It means financial subsidy; During the steelmaking process, the carbon emission expression is shown in the following formula (13): in, represents the total amount of carbon emissions; Indicates the steelmaking process; Indicates steelmaking equipment; Indicates the type of energy; Represents the carbon emission factor.

9. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the steelmaking process optimization method for the steel industry according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for optimizing the steelmaking process in the steel industry described in any one of claims 1 to 8 is implemented.

Images