Double-objective optimization decision-making method for slag cooling process

Abstract

The invention provides a dual-objective optimization decision-making method for a slag cooling process, and belongs to the technical field of industrial production process optimization. According to the method, historical slag cooling data are collected and preprocessed, and the temperature and the cooling time after cooling are predicted through a multiple linear regression model. And a dynamic weighting objective function is further constructed, and optimal control parameters are solved by adopting a hybrid algorithm (combination of a genetic algorithm, a particle swarm algorithm and a reinforcement learning algorithm). And finally, integrating the solved optimal control parameters into a real-time control system, and dynamically adjusting the control parameters according to real-time monitoring data so as to realize dual-objective optimization of cooling time and hearth temperature fluctuation. According to the dual-objective optimization decision-making method for the slag cooling process, the efficiency and stability of the slag cooling process can be effectively improved, energy consumption is reduced, and remarkable economic benefits and environmental benefits are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production process optimization, and particularly to a dual-objective optimization decision-making method for the slag cooling process. Background Art

[0002] In modern industrial production, the slag cooling process is an indispensable link in high-temperature processes such as steel smelting. Slag cooling not only affects production efficiency but also is directly related to energy consumption, equipment life, and the stability of subsequent processes. However, the existing slag cooling technologies have the following deficiencies:

[0003] Low cooling efficiency: Traditional slag cooling methods usually adopt a single cooling strategy, such as fixed air volume or fixed cooling time, and cannot be adjusted in real time according to dynamic factors such as the initial temperature of the slag and the slag discharge amount. This results in a long cooling time, low production efficiency, and high energy consumption.

[0004] Unstable temperature control: During the cooling process, the furnace temperature fluctuates greatly and is difficult to accurately control. This unstable temperature change may cause uneven cooling of the slag, affect product quality, and even damage the cooling equipment.

[0005] Lack of optimization decision-making mechanism: In the existing technology, the setting of cooling parameters mostly depends on experience or fixed process parameters, lacking a scientific optimization decision-making method. This makes it impossible to achieve the best balance between cooling time and temperature stability during the cooling process, and it is difficult to meet the requirements of high-efficiency production and energy conservation and consumption reduction.

[0006] Insufficient real-time adjustment ability: In actual production, the properties of the slag and the cooling environment will change continuously, but the existing technology cannot monitor and dynamically adjust the cooling parameters in real time. This makes it difficult for the cooling process to adapt to complex production conditions and unable to effectively respond to emergencies.

[0007] Insufficient data utilization: Although a large amount of slag cooling data has been accumulated in modern industrial production, the existing technology has not fully utilized these data for in-depth analysis and optimization modeling. The value of the data has not been fully explored and cannot provide scientific guidance for the cooling process. Summary of the Invention

[0008] The object of the present invention is to provide a dual-objective optimization decision-making method for the slag cooling process, which can effectively improve the efficiency and stability of the slag cooling process, reduce energy consumption, and has significant economic and environmental benefits.

[0009] To achieve the above object, the present invention provides a dual-objective optimization decision-making method for the slag cooling process, including the following steps:

[0010] Step S1: Collect historical slag cooling data and preprocess it, and use a multiple linear regression model to predict the temperature T and cooling time t after cooling respectively;

[0011] Step S2: Construct a dynamic weighted objective function;

[0012] Step S3: Use a hybrid algorithm to solve for the optimal control parameters; the hybrid algorithm is a combination of a genetic algorithm, a particle swarm algorithm, and a reinforcement learning algorithm;

[0013] Step S4: Integrate the obtained optimal control parameters into the real-time control system, and dynamically adjust the control parameters according to the real-time monitoring data.

[0014] Preferably, in step S1, the historical slag cooling data includes: the steel belt speed v, the air damper opening d, the initial temperature T 0 , the slag discharge amount Q, the cooling air volume F, and the cooling air temperature T f .

[0015] Preferably, in step S1, the preprocessing includes: data cleaning and data normalization.

[0016] Preferably, in step S1, the multiple linear regression model includes a temperature regression model and a cooling time regression model;

[0017] The temperature regression model is as follows:

[0018] T = β 0T + β 1T v + β 2T d + β 3T T 0 + β 4T Q + β 5T F + β 6T T f + ε T ;

[0019] The cooling time regression model is as follows:

[0020] t = β 0t + β 1t v + β 2t d + β 3t T 0 + β 4t Q + β 5t F + β 6t T f + ε t ;

[0021] Among them, β 0T , β 1T , β 2T , β 3T , β 4T , β5T and β 6T both represent the regression coefficients of the temperature regression model; β 0t and β 1t and β 2t and β 3t and β 4t and β 5t and β 6t all represent the regression coefficients of the cooling time regression model; ε T and ε t represent the error term.

[0022] Preferably, in step S2, the dynamic weighted objective function is as follows:

[0023]

[0024] where α and β represent the dynamic weight coefficients; t(v, d) represents the cooling time; ΔT(v, d) represents the furnace temperature fluctuation.

[0025] Preferably, the calculation formula of the dynamic weight coefficient is as follows:

[0026]

[0027] where α max and α min represent the maximum value and the minimum value of α respectively; β max and β min represent the maximum value and the minimum value of β respectively; T max represents the maximum number of iterations in the optimization process.

[0028] Preferably, in step S3, the genetic algorithm is as follows:

[0029] Use the dynamic weighted objective function as the fitness function to calculate the fitness value of each individual;

[0030] Select two individuals for crossover according to the adaptive crossover rate to generate new individuals;

[0031] Mutate the individuals according to the adaptive mutation rate to increase the population diversity;

[0032] Use the tournament selection method to select the individuals with higher fitness to enter the next generation;

[0033] Add the global optimal position g of the particle swarm algorithm as an elite individual directly to the next generation population of the genetic algorithm:

[0034] GAPopulation new = GAPopulation selected ∪ {g};

[0035] Among them, GAPopulation new represents the new generation population of the genetic algorithm; GAPopulation selected represents a subset composed of individuals with higher fitness selected from the current population through the tournament selection method.

[0036] Preferably, in step S3, the particle swarm optimization algorithm is as follows:

[0037] Randomly generate an initial population, where each individual represents a set of control parameters (v, d), and the population size is N;

[0038] Initialize the position and velocity of the particle swarm;

[0039] According to the update formula of the particle swarm optimization algorithm, adjust the velocity and position of the particles;

[0040] Dynamically adjust the inertia weight w according to the number of iterations:

[0041]

[0042] Among them, w max and w min represent the maximum and minimum values of the inertia weight respectively; S max represents the maximum number of iterations; s represents the current number of iterations;

[0043] Compare the position of the current particle with the individual optimal position p i , and feedback the individual solution with the highest fitness in the genetic algorithm to the particle swarm optimization algorithm to update the individual optimal position;

[0044] Then compare the global optimal position and update the global optimal position g:

[0045]

[0046]

[0047] Among them, represents the position of the i-th particle at the (s + 1)-th iteration; represents the fitness function value corresponding to the individual optimal position p i of the i-th particle; represents the fitness function value corresponding to the global optimal position.

[0048] Preferably, in step S3, the reinforcement learning algorithm is as follows:

[0049] Select an action a according to the current state c;

[0050] Execute the action a, adjust the control parameters v and d, and calculate the reward value R:

[0051]

[0052] Update the Q value according to the Bellman equation;

[0053] Update the state c to the new state c', repeat the above process, and feedback the optimized control parameters (v * , d * ) obtained by reinforcement learning to the genetic algorithm and the particle swarm optimization algorithm to update the population and the particle positions:

[0054]

[0055] wherein, v * represents the optimal steel strip speed; d * represents the optimal air damper opening.

[0056] Preferably, in step S4, integrate the obtained optimal control parameters into the real-time control system, and dynamically adjust the control parameters according to the real-time monitoring data. The specific operations are as follows:

[0057] Record the temperature and the cooling time obtained in real time during the cooling process by the sensor as X real-time ;

[0058] Predict the current cooling time and temperature fluctuation according to the temperature regression model and the cooling time regression model in step S1:

[0059]

[0060] wherein, represents the predicted temperature after cooling; represents the predicted cooling time; f T , f t represent the functions of the prediction model;

[0061] According to the predicted cooling time and temperature fluctuation, and in combination with the optimal control parameters in step S3, dynamically adjust the steel strip speed and the air damper opening:

[0062] v new = v current + Δv;

[0063] d new = d current + Δd;

[0064] wherein, v current , d current represent the current control parameters; v new , d new represent the adjusted control parameters; Δv, Δd represent the adjustment amounts calculated according to the prediction model and the optimal control parameters;

[0065] Δv = k v (v * - v current );

[0066] Δd = k d (d * - d current );

[0067] where k v and k d represent the adaptive control coefficients.

[0068] Therefore, the present invention adopts the above-mentioned dual-objective optimization decision-making method for the slag cooling process, and the beneficial technical effects are as follows:

[0069] (1) Significantly improve the cooling efficiency:

[0070] By dynamically adjusting the cooling parameters through the multiple linear regression model and the hybrid optimization algorithm, the present invention significantly shortens the cooling time, significantly improves the production efficiency, and significantly reduces the energy consumption compared with the traditional cooling method with fixed parameters, having significant economic and environmental benefits.

[0071] (2) Achieve high-precision stable control of the furnace temperature:

[0072] The present invention adopts a dynamic weighted objective function, which can effectively balance the cooling time and the furnace temperature fluctuation during the cooling process, significantly improving the temperature stability. Compared with the large temperature fluctuation in the prior art, the present invention ensures a uniform and efficient cooling process, further improving the product quality.

[0073] (3) Enhance the real-time dynamic adjustment ability:

[0074] The present invention integrates the optimized control parameters into the real-time control system and dynamically adjusts the cooling parameters in combination with the real-time monitoring data, enabling a rapid response to changes in the production process. Compared with the defect of lacking real-time adjustment ability in the prior art, the present invention significantly improves the adaptability and flexibility of the cooling process.

[0075] (4) Innovation and intelligence of the optimization decision-making mechanism:

[0076] The present invention adopts a hybrid optimization strategy combining genetic algorithm, particle swarm algorithm and reinforcement learning algorithm, overcomes the limitations of traditional empirical settings, and realizes the scientific optimization and dynamic adjustment of cooling parameters. This intelligent decision-making mechanism significantly improves the intelligent level of the cooling process, having wide applicability and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1Flow chart of a dual - objective optimization decision - making method for the slag cooling process of the present invention;

[0078] Figure 2 It is the flow chart optimized by the hybrid algorithm. Specific implementation mode

[0079] The technical solution of the present invention will be further described below through the drawings and embodiments.

[0080] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the field to which the present invention belongs.

[0081] Embodiment 1

[0082] The slag cooling system of a certain steel plant has problems such as low cooling efficiency, high energy consumption, and large temperature fluctuations under the traditional cooling method. To solve these problems, the dual - objective optimization decision - making method for the slag cooling process of the present invention is adopted to optimize the cooling process, improve production efficiency, and reduce energy consumption.

[0083] As Figure 1 shown, it is the flow chart of a dual - objective optimization decision - making method for the slag cooling process of the present invention, including:

[0084] Step S1: Collect historical data of the slag cooling process in the past year, including the following parameters:

[0085] Steel belt speed (unit: m / min);

[0086] Air damper opening (unit: %);

[0087] Initial temperature (unit: °C);

[0088] Slag discharge amount (unit: t / h);

[0089] Cooling air volume (unit: m 3 / min);

[0090] Cooling air temperature (unit: °C);

[0091] Cooling time (unit: min);

[0092] Temperature after cooling (unit: °C);

[0093] The data is sourced from the production records and sensor monitoring systems of the factory.

[0094] The pre - processing includes: data cleaning and data normalization.

[0095] Data cleaning: Remove missing values and outliers. For example, eliminate data points with extremely short or long cooling times, and records with initial temperatures that are significantly inconsistent with production conditions.

[0096] Data normalization: Normalize all parameters to the interval [0, 1] for subsequent modeling and optimization.

[0097] Construct a multiple linear regression model including a temperature regression model and a cooling time regression model;

[0098] The temperature regression model is as follows:

[0099] T = β 0T + β 1T v + β 2T d + β 3T T 0 + β 4T Q + β 5T F + β 6T T f + ε T ;

[0100] The cooling time regression model is as follows:

[0101] t = β 0t + β 1t v + β 2t d + β 3t T 0 + β 4t Q + β 5t F + β 6t T f + ε t ;

[0102] Among them, β 0T , β 1T , β 2T , β 3T , β 4T , β 5T , β 6T all represent the regression coefficients of the temperature regression model; β 0t , β 1t , β 2t , β 3t , β 4t , β 5t , β 6t all represent the regression coefficients of the cooling time regression model; ε T , ε t represent the error terms.

[0103] Step S2: Construct a dynamic weighted objective function to balance the cooling time and the furnace temperature fluctuation;

[0104] The dynamic weighted objective function is as follows:

[0105]

[0106] Among them, α and β represent dynamic weight coefficients; t(v, d) represents the cooling time; ΔT(v, d) represents the furnace temperature fluctuation.

[0107] The calculation formula for the dynamic weight coefficient is as follows:

[0108]

[0109] Among them, α max and α min represent the maximum and minimum values of α respectively; β max and β min represent the maximum and minimum values of β respectively; T max represents the maximum number of iterations in the optimization process.

[0110] Step S3, as Figure 2 shown, use a hybrid algorithm to solve the optimal control parameters.

[0111] The hybrid algorithm is a combination of the genetic algorithm (GA), the particle swarm optimization algorithm (PSO), and the reinforcement learning algorithm.

[0112] Genetic algorithm.

[0113] The parameter settings are as follows:

[0114] Population size: 100;

[0115] Crossover rate: 0.8;

[0116] Mutation rate: 0.05;

[0117] Fitness function: dynamic weighted objective function;

[0118] Selection method: tournament selection method;

[0119] Number of iterations: 50.

[0120] Use the dynamic weighted objective function as the fitness function to calculate the fitness value of each individual;

[0121] Select two individuals for crossover according to the adaptive crossover rate to generate new individuals;

[0122] Mutate the individuals according to the adaptive mutation rate to increase the population diversity;

[0123] Use the tournament selection method to select the individuals with higher fitness to enter the next generation;

[0124] Add the global optimal position g of the particle swarm optimization algorithm as an elite individual directly to the next generation population of the genetic algorithm:

[0125] GAPopulation new = GAPopulation selected ∪ {g};

[0126] Wherein, GAPopulation new represents the new generation population of the genetic algorithm; GAPopulation selected represents a subset composed of individuals with higher fitness selected from the current population by the tournament selection method.

[0127] Particle swarm optimization algorithm.

[0128] The parameter settings are as follows:

[0129] Population size: 50;

[0130] Individual learning factor: 2.0;

[0131] Social learning factor: 2.0;

[0132] Number of iterations: 50.

[0133] Randomly generate the initial population, each individual represents a set of control parameters (v, d), and the population size is N;

[0134] Initialize the position and velocity of the particle swarm;

[0135] According to the update formula of the particle swarm optimization algorithm, adjust the velocity and position of the particles;

[0136] Dynamically adjust the inertia weight w according to the number of iterations:

[0137]

[0138] Wherein, w max , w min represent the maximum and minimum values of the inertia weight respectively; S max represents the maximum number of iterations; s represents the current number of iterations;

[0139] Compare the position of the current particle with the individual optimal position p i , and feedback the individual solution with the highest fitness in the genetic algorithm to the particle swarm optimization algorithm to update the individual optimal position;

[0140] Then compare the global optimal position and update the global optimal position g:

[0141]

[0142] Wherein, represents the position of the i-th particle at the (s + 1)-th iteration; Denote the individual optimal position \(p\) of the \(i\)-th particle i and the corresponding fitness function value; Denote the fitness function value corresponding to the global optimal position.

[0143] Reinforcement learning algorithm.

[0144] Set as follows:

[0145] State space: steel belt speed and damper opening;

[0146] Action space: adjustment amounts of steel belt speed and damper opening;

[0147] Reward function: calculated according to cooling time and temperature fluctuation;

[0148] Learning rate: 0.1;

[0149] Discount factor: 0.9;

[0150] Number of iterations: 100.

[0151] Select action \(a\) according to the current state \(c\);

[0152] Execute action \(a\), adjust the control parameters \(v\) and \(d\), and calculate the reward value \(R\):

[0153]

[0154] Update the \(Q\)-value according to the Bellman equation;

[0155] Update the state \(c\) to the new state \(c'\), repeat the above process, and feedback the control parameters \((v\) * , \(d\) * ) optimized by reinforcement learning to the genetic algorithm and the particle swarm algorithm to update the population and particle positions:

[0156] GAPopulation new = GAPopulation new ∪ \(\{(v\) * , \(d\) * )\};

[0157]

[0158] where \(v\) * represents the optimal steel belt speed; \(d\) * represents the optimal damper opening.

[0159] Step S4: Integrate the obtained optimal control parameters into the real-time control system, and dynamically adjust the control parameters according to the real-time monitoring data. The specific operations are as follows:

[0160] The temperature and cooling time during the cooling process obtained in real time through the sensor are denoted as X real-time ;

[0161] Predict the current cooling time and temperature fluctuation according to the temperature regression model and cooling time regression model in step S1:

[0162]

[0163] Among them, represents the predicted temperature after cooling; represents the predicted cooling time; f T 、f t represents the function of the prediction model, and its specific form is the same as that in step S1;

[0164] According to the predicted cooling time and temperature fluctuation, combined with the optimal control parameters in step S3, dynamically adjust the steel strip speed and damper opening:

[0165] v new = v current + Δv;

[0166] d new = d current + Δd;

[0167] Among them, v current 、d current represent the current control parameters; v new 、d new represent the adjusted control parameters; Δv, Δd represent the adjustment amounts calculated according to the prediction model and optimal control parameters;

[0168] Δv = k v (v * - v current );

[0169] Δd = k d (d * - d current );

[0170] Among them, k v 、k d represent the adaptive control coefficients.

[0171] By using the method proposed in the present invention, the performance improvement results are shown in Table 1.

[0172] Table 1 Performance Improvement Results

[0173]

[0174]

[0175] It should be noted that the content not elaborated in detail in the present invention is all prior art and is well known to those skilled in the art.

[0176] Therefore, by adopting the above-mentioned dual-objective optimization decision-making method for the slag cooling process, the present invention can effectively improve the efficiency and stability of the slag cooling process, reduce energy consumption, and has significant economic and environmental benefits.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dual-objective optimization decision method for slag cooling process, characterized in that: The following steps are involved: Step S1, collecting and preprocessing historical slag cooling data, and using a multivariate linear regression model to predict the temperature T and cooling time t after cooling respectively; Step S2, constructing a dynamic weighted objective function; Step S3, using a hybrid algorithm to solve the optimal control parameters; the hybrid algorithm is a combination of a genetic algorithm, a particle swarm algorithm, and a reinforcement learning algorithm; Step S4: Integrate the optimal control parameters obtained by solving into the real-time control system, and dynamically adjust the control parameters according to the real-time monitoring data.

2. A dual-objective optimization decision method for slag cooling process according to claim 1, characterized in that: In step S1, the historical slag cooling data includes: steel belt speed v, air door opening d, initial temperature T0, slag discharge volume Q, cooling air volume F and cooling air temperature T f .

3. The dual-objective optimization decision method for slag cooling process according to claim 1, characterized in that: In step S1, preprocessing includes: data cleaning and data normalization.

4. A dual-objective optimization decision method for slag cooling process according to claim 2, characterized in that: In step S1, the multiple linear regression model includes a temperature regression model and a cooling time regression model; The temperature regression model is as follows: T=β 0T +β 1T v+β 2T d+β 3T T0+β 4T Q+β 5T F+β 6T T f +ε T ; The cooling time regression model is as follows: t=β 0t +β 1t v+β 2t d+β 3t T0+β 4t Q+β 5t F+β 6t T f +ε t ; Among them, β 0T , β 1T , β 2T , β 3T , β 4T , β 5T , β 6T Both represent the regression coefficients of the temperature regression model; β 0t , β 1t , β 2t , β 3t , β 4t , β 5t , β 6t Both represent the regression coefficients of the cooling time regression model; ε T , ε t represents the error term.

5. A dual-objective optimization decision method for slag cooling process according to claim 4, characterized in that: In step S2, the dynamic weighted objective function ζ(v,d) is as follows: ζ(v,d)=α×t(v,d)+β×ΔT(v,d); Among them, α and β represent dynamic weight coefficients; t(v, d) represents cooling time; ΔT(v, d) represents furnace temperature fluctuation.

6. A dual-objective optimization decision method for slag cooling process according to claim 5, characterized in that: The calculation formula of dynamic weight coefficient is as follows: Among them, α max , α min Respectively represent the maximum and minimum values ​​of α; β max , β min Respectively represent the maximum and minimum values ​​of β; T max Indicates the maximum number of iterations of the optimization process.

7. A dual-objective optimization decision method for slag cooling process according to claim 6, characterized in that: In step S3, the genetic algorithm is as follows: Use the dynamic weighted objective function ζ(v,d) as the fitness function to calculate the fitness value of each individual; Select two individuals for crossover according to the adaptive crossover rate to generate new individuals; Mutate individuals according to the adaptive mutation rate to increase population diversity; Use tournament selection to select individuals with higher fitness to enter the next generation; The global optimal position g of the particle swarm algorithm is directly added to the next generation population of the genetic algorithm as an elite individual: GAPopulation new =GAPopulation selected ∪{g}; Among them, GAPopulation new Represents the new generation population of genetic algorithm; GAPopulation selected Represents a subset of individuals with higher fitness selected from the current population through the tournament selection method.

8. A dual-objective optimization decision method for slag cooling process according to claim 7, characterized in that: In step S3, the particle swarm algorithm is as follows: Randomly generate the initial population, each individual represents a set of control parameters (v, d), and the population size is N; Initialize the position and velocity of the particle swarm; According to the update formula of the particle swarm algorithm, adjust the speed and position of the particles; Dynamically adjust the inertia weight w according to the number of iterations: Among them, w max 、w min Respectively represent the maximum and minimum values ​​of the inertia weight; S max Indicates the maximum number of iterations; s indicates the current number of iterations; Compare the current particle position and the individual optimal position p i , the individual solution with the highest fitness in the genetic algorithm is fed back to the particle swarm algorithm to update the optimal position of the individual; Then compare the global optimal position and update the global optimal position g: in, represents the position of the ith particle at the s+1th iteration; ζ(p i ) represents the individual optimal position p of the i-th particle i The corresponding fitness function value; ζ(g) represents the fitness function value corresponding to the global optimal position.

9. A dual-objective optimization decision method for slag cooling process according to claim 8, characterized in that: In step S3, the reinforcement learning algorithm is as follows: Select action a according to the current state c; Execute action a, adjust control parameters v and d, and calculate reward value R: R = -ζ(v,d); Update the Q value according to the Bellman equation; Update state c to the new state c', repeat the above process, and use the control parameters (v * ,d * ) is fed back to the genetic algorithm and particle swarm algorithm to update the population and particle positions: Among them, v * Indicates the optimal steel belt speed; d * Indicates the optimal air door opening.

10. A dual-objective optimization decision method for slag cooling process according to claim 9, characterized in that: In step S4, the optimal control parameters obtained by solving are integrated into the real-time control system, and the control parameters are dynamically adjusted according to the real-time monitoring data. The specific operations are as follows: The temperature and cooling time obtained in real time by the sensor during the cooling process are recorded as X real-time ; Predict the current cooling time and temperature fluctuation based on the temperature regression model and cooling time regression model in step S1: in, represents the predicted temperature after cooling; represents the predicted cooling time; f T 、f t A function representing a prediction model; According to the predicted cooling time and temperature fluctuation, combined with the optimal control parameters of step S3, the steel belt speed and air door opening are dynamically adjusted: v new =v current +Δv; d new =d current +Δd; Among them, v current d current Indicates the current control parameters; v new d new represents the adjusted control parameters; Δv and Δd represent the adjustment amounts calculated based on the prediction model and the optimal control parameters; Δv=k v (v * -v current ); Δd=k d (d * -d current ); Among them, k v , k d represents the adaptive control coefficient.

Images