Roll quenching machine billet cooling process parameter optimization method based on hybrid model

Abstract

The invention relates to the technical field of roller type quenching machine cooling parameter optimization, and discloses a hybrid model-based roller type quenching machine billet cooling process parameter optimization method, which comprises the following steps of: performing parameter optimization on a flow set value and a roller way speed by adopting an improved particle swarm optimization algorithm to obtain a new flow set value of each section and an optimal roller way running speed; the new flow setting output by the improved PSO algorithm and the roller way speed serve as BP neural network input, and the water-cooling convective heat transfer coefficient is calculated; calculating boundary node temperature, solving an equation set through a TDMA numerical method, updating the position of a steel plate, calculating steel plate temperature distribution in the quenching process according to the position of the steel plate until the steel plate goes out of a quenching machine, constructing an optimized objective function of quenching cooling flow of the quenching machine, introducing a penalty function to perform constraint processing until iteration is terminated, and finally obtaining the quenching cooling flow of the quenching machine. And finally obtaining the set values of the regional flow and the roller speed corresponding to the minimum target function value of the steel plate, and outputting the corresponding temperature distribution of the steel plate.

Description

Technical Field

[0001] The invention relates to the technical field of optimization of cooling process parameters of steel plates in a quenching machine, and in particular to a method for optimizing cooling process parameters of steel plates in a roller quenching machine based on a hybrid model. Background Art

[0002] In recent years, many scholars have conducted in-depth research on plate shape modeling and control, and numerical simulation of plate quenching process. The production efficiency of the quenching process is closely related to its operating parameters. There are relatively few studies on working condition identification and operating parameter optimization in the quenching process, but they have been widely studied in some industrial processes and have achieved certain results, which has promoted the development of industrial processes.

[0003] Roller quenching machine is an important equipment for quenching heat treatment of medium and thick plates in steel plants. Quenching heat treatment refers to the rapid cooling of the steel plate heated by the quenching furnace, transforming its structure from austenite to martensite, and then tempering it at an appropriate temperature to obtain the required comprehensive properties. Due to the influence of the stability of supercooled austenite, cross-sectional dimensions and quenching cooling rate, the steel plate after quenching may not be able to obtain martensite completely, and there may be bainite and pearlite inside the steel plate.

[0004] In recent years, in response to market demand, many domestic and foreign steel companies, engineering companies, colleges and universities, and scientific research institutions have been conducting research on quenching equipment, mathematical models, and control systems. Most of the previous quenching machine operating parameters were set by manual experience, which could not meet the requirements of the critical cooling rate in the quenching project. The study of the quenching cooling process is related to the process formulation of actual production, and setting reasonable operating parameters has become an urgent need for steel companies. With the promotion and application of domestic roller quenching machine equipment, the research on the quenching process of carbon steel medium and thick plates has been further deepened. The numerical method is now used to comprehensively analyze the flow, heat transfer, and structure of metal parts during the quenching process. The phase change in the quenching process determines the material structure properties such as hardness and toughness of the product, and the important factor affecting the structure transformation is the cooling rate of the steel plate during the cooling process. At the same time, the uneven distribution of temperature during the cooling process and the different expansion amounts at each point will cause thermal stress and thermal deformation, thereby causing plate shape problems. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention adopts a method for optimizing the cooling process parameters of a steel billet based on a hybrid model roller quenching machine, which can set the optimal water flow rate, steel plate movement speed and other parameters in each water cooling area according to different steel plate specifications, so as to achieve precise control of the steel plate temperature and save water resources.

[0006] Taking the mathematical model of cooling steel plate in quenching machine as the framework, combined with BP neural network and particle swarm optimization algorithm, a hybrid modeling method is used to construct a temperature prediction model of steel plate in quenching machine with convection heat transfer coefficient as the core, and the quenching process is optimized and controlled, including the following steps: S1. Input the improved PSO algorithm parameters, including the population size, into the mathematical model of the quenching machine cooling steel plate N , dimension D , the maximum number of fitness evaluations maxFES , the learning probability of the particle P c , and input quenching machine parameters and billet information; S2, initializing the flow setting value and roller speed, and calculating the initial steel plate temperature distribution and the objective function value; S3, using improved particle swarm optimization algorithm to optimize the flow setting value and roller speed parameters, and obtain new flow setting values ​​for each section and the optimal roller speed; Specifically, according to the fitness value of the individual historical optimal position of the particle, the population is divided into an elite group and an ordinary group, and dynamic reorganization and mutual cooperation are used to search for optimization. In the early stage of evolution, the CL strategy with individual optimal position learning items is adopted for elite particles, and the CL strategy with global optimal solution is adopted for ordinary particles to complete the evolution of the current group; in the later stage of evolution, the individual historical optimal position stored in the historical optimal solution group is directly evolved; Among them, the individual historical optimal position of elite particles adopts the mutation strategy "DE / current-to-rand / 1", and the mutation of ordinary group particles adopts "DE / rand-to- p best” strategy; S4, inputting the steel plate position, time step and space step in the mathematical model of cooling the steel plate in the quenching machine; S5. Collect historical data of the quenching machine during the quenching process, select data under stable working conditions as the model input data set, train the BP neural network model, use the new flow setting and roller speed output by the improved PSO algorithm as the BP neural network input, and use the trained BP neural network model to calculate the water cooling convection heat transfer coefficient; S6. Based on the simulated steel plate position change, the BP neural network is used to obtain the water cooling convection heat transfer coefficient, the boundary node temperature is calculated, the equation group is solved by the TDMA numerical method, the internal node temperature distribution of the next time step is obtained, the steel plate position is continuously updated, and the steel plate temperature distribution during the quenching process is calculated by the steel plate position until the steel plate leaves the quenching machine; S7. Construct an optimization objective function for the quenching cooling flow rate of the quenching machine, introduce a penalty function to constrain it, and terminate the iteration until the set values ​​of the regional flow rate and roller speed corresponding to the minimum objective function value of the steel plate are finally obtained, and the corresponding steel plate temperature distribution is output.

[0007] The steel billet information in step S1 includes: steel billet specifications, types, thermophysical parameters, and initial temperature.

[0008] Step S3 specifically includes the following steps: S3.1. After the particles are grouped, they evolve the current group and the historical optimal solution group in turn. For each particle i , the current position d dimension X i d Update as follows: (4) S3.2. For elite particles i ( i = 1, 2, …, N / 2 ), update and calculate each dimension according to the following formula d Speed ​​on: (5) In the formula, c represents the acceleration coefficient; rn d Represents a random number that follows a uniform distribution in the range [0, 1]; f i ( d ) represents the index vector f i =[ f i (1) f i (2) ... , f i ( D )] in d dimensional element, which indicates the i The particle d Which particle's d Dimensional learning; Represents particles i The learning model is based on the learning probability PC choose; For the learning probability used in the improved algorithm PC For example, the process parameters are calculated PC As shown below: (6) In the formula, PC i For each particle i No. d dimension, representing a random number between 0 and 1. Then with the elite particle i The corresponding learning probability PC i Value comparison: If the random number is less than PC i Value, then the elite particles i No. d The dimension needs to be the individual historical optimal solution corresponding to the first d dimensional learning, if the random number is greater than / equal to its PC i value, then the particle i No. d The dimension will move to the first corresponding to its own individual historical optimal solution. d Dimensional learning; S3.3 For each ordinary particle, its velocity d The dimension is updated using the CL strategy with a global optimal solution. The specific update formula is as follows: (7) In the formula, rn 1 d and rn 2 d Both represent random numbers in the range [0, 1].

[0009] In order to fully develop the area with promising solutions, continue to update the global optimum and improve the solution accuracy: S3.4. For the individual historical optimal position of elite particles, the mutation strategy "DE / current-to-rand / 1" is adopted to strengthen the information exchange between elite particles: (12) In the formula, r 1. r 2 and r 3 means in [1, N / 2] to find three different integers. PV i Represents elite particles i The corresponding mutation vector; F i Represents elite particles i The corresponding adaptive coefficient of variation, adaptive coefficient of variation F i Based on the mean μ F, generated by a Cauchy distribution with a standard deviation of 0.1; S3.5. For each ordinary particle i ( i = N / 2+1, N / 2+2, … , N ) of the individual historical optimal solution, using “DE / rand-to- p The "best / 1" mutation strategy enhances the guidance of the top optimal particles to ordinary individuals: (13) In the formula, Indicates that from the top p %Select the individual historical optimal solution from all historical optimal solutions, r 1. r 2 means in [1, N ] and find two i Integers that are not equal to each other, Represents the individual historical optimal solution and external archive of all particles A A solution vector randomly selected from the union of ; S3.6. Perform a crossover operation on each individual to generate their own experimental vector PU i,g = [ u i,1,g , … , u i,j,g , … , u i,D,g ], which comes from X i,g and V i,g Combination of: (14) In the formula, CR i ∈[0, 1] represents each particle i The corresponding adaptive crossover probability; d rand Indicates that in [1, D ]; S3.7. Each particle i The individual historical optimal solution P i The corresponding experimental vector PU i For comparison: (15) In the formula,f ( P i )and f ( PU i ) respectively represent P i and PU i The corresponding fitness function value is finally used to obtain the new flow setting values ​​of each section of the quenching machine and the optimal operating speed of the roller.

[0010] Step S5 specifically includes the following steps: S5.1: preprocessing the historical data during the quenching process, including data cleaning and data normalization operations, to improve the training efficiency and prediction accuracy of the BP neural network; wherein the historical data during the quenching process includes water flow rate, nozzle height from the billet, nozzle water pressure, billet moving speed, injection angle, and corresponding water cooling convection heat transfer coefficient; S5.2: Construct a BP neural network model, the number of neurons in the input layer, hidden layer and output layer, determine the water flow rate and billet movement speed as the input layer, the water cooling convection heat transfer coefficient as the output layer, select appropriate activation function and loss function, input the preprocessed historical data into the BP neural network model for training, and minimize the error between the predicted value and the true value by continuously adjusting the model parameters to obtain the optimal water cooling convection heat transfer coefficient.

[0011] Calculating the boundary node temperature in step S6 specifically includes: A one-dimensional unsteady-state heat conduction equation is established, and the specific mathematical description is as follows: (16) In the formula, C p represents the specific heat of the steel plate; T Indicates the temperature of the steel plate; ρ Indicates the density of steel plate; τ Indicates time; Indicates thermal conductivity of steel plate. The above thermal physical parameters change with temperature. Discretize the steel plate section and use the heat balance method to establish the differential equation of each unit cell: Mathematical description and numerical solution of steel plate quenching cooling process: upper and lower surface nodes: (17) (18) Simplifying, we get: (19) (20) Solve for the internal nodal temperatures: (twenty one) in , .

[0012] In step S6, the equation group is solved by the TDMA numerical method to obtain the temperature distribution of the quenched steel plate: Convert formula (21) into a (n-1) × (n-1) matrix form: (twenty two) (twenty three) (twenty four) The temperature distribution of the quenched steel plate is obtained by solving the triangular matrix and the one-dimensional unsteady boundary equation.

[0013] Step S7 constructs an optimization function with the water consumption of the quenching machine as the target, taking into account that the cooling of the steel plate in the quenching machine needs to meet additional target constraints, that is, the temperature after quenching needs to be controlled within a reasonable range and the cooling rate of the center of the steel plate is less than the critical cooling rate, otherwise the solution is not feasible; the penalty function method is introduced for constraint processing, wherein the mathematical expression of the optimization objective function of the quenching cooling water flow is: (1) In the formula, f represents the objective function; Q i represents the flow rate of the i-th nozzle; represents the ratio of the upper and lower flow rates of the mth nozzle (water ratio); P1 and P2 represent the penalty term for the post-quenching temperature and the penalty term for the cooling rate at the center of the steel plate, respectively; Among them, P1 and P2 represent the penalty term of temperature after quenching and the penalty term of cooling rate at the center of the steel plate, respectively; (2) In the formula, K1 represents the first penalty coefficient; and They represent the average temperatures of the center and surface of the steel plate after it has completely left the quenching machine. represents the temperature difference threshold, (℃); (3) Wherein, K2 represents the second penalty coefficient.

[0014] Compared with the prior art, the present invention has the following beneficial technical effects and advantages: 1) By constructing a temperature prediction model for the steel plate in the quenching machine, the temperature distribution of the steel plate during the quenching process can be accurately predicted. The quenching convection cooling coefficient is calculated using the BP neural network model. A variety of influencing factors are considered, and the functional relationship between the operating parameters and the convection heat transfer coefficient is constructed, which improves the accuracy of the temperature prediction model and lays the foundation for the subsequent optimization of the steel plate quenching water cooling process; 2) The present invention optimizes and controls the quenching process through a particle swarm optimization algorithm, takes the total water consumption as the objective function, and the cooling rate of the steel plate and the final temperature difference as the constraint conditions, thereby improving the cooling efficiency and quenching quality, while reducing the water consumption. The improved particle swarm optimization algorithm is used for optimization, which improves the algorithm performance and convergence speed, and integrates and optimizes the operating parameters of the quenching machine, so that it can quickly and stably find the set values ​​suitable for the production conditions, which has significant economic and environmental benefits; 3) Based on the hybrid modeling, a high-precision hybrid prediction model of billet temperature distribution is obtained. By collecting actual operation data, a one-dimensional non-steady-state temperature field cooling mathematical model of steel plate is established. The present invention uses an improved particle swarm algorithm to achieve the optimal setting of quenching machine operating parameters. The model can simulate the temperature drop curve of steel plate under different operating conditions; the improved particle swarm optimization algorithm is used to optimize the operating parameters required by the model, and it is continuously iterated to find the optimal objective function value; by comparing the model prediction and the measured data, the parameters are adjusted to achieve the best cooling effect and realize the intelligent process; by introducing a hybrid model, it can not only overcome the limitations of traditional physical modeling, but also predict through a neural network model. This method is particularly suitable for complex and nonlinear industrial environments, and is an ideal choice for optimizing the heat exchange process of the quenching machine, ensuring the effectiveness and accuracy of the steel plate quenching water cooling temperature prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Picture 1 It is a schematic diagram of the overall process of the method for optimizing the cooling process parameters of a quenching machine steel billet of the present invention; Picture 2 It is a schematic diagram of the optimization simulation flow of the mathematical model for predicting the temperature of the steel plate of the quenching machine of the present invention; Picture 3 A flow chart showing the optimal water flow rate and steel plate moving speed for the improved PSO algorithm of the present invention; DETAILED DESCRIPTION

[0016] The present invention is described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited by the accompanying drawings.

[0017] The present invention provides a method for optimizing the cooling process parameters of a steel plate of a roller quenching machine based on a hybrid model. The method takes the optimization of the operating parameters of a medium and thick plate roller quenching machine of a steel plant as the background, aims to improve the accuracy of a temperature prediction model of the steel plate during the cooling process of the quenching machine, adjusts and sets the water consumption and the roller running speed of the steel plate during the quenching process on the premise of meeting the quenching cooling speed of the steel plate, and optimizes the key operating parameters of the quenching machine with this as the goal.

[0018] The present invention uses a novel meta-inspired intelligent optimization algorithm, takes the mathematical model of cooling steel plate of quenching machine as the framework, combines BP neural network and particle swarm optimization algorithm, and adopts a hybrid modeling method to construct a steel plate temperature prediction model with convection heat transfer coefficient as the core, conducts in-depth analysis on the cooling process of steel plate, and constructs a one-dimensional non-steady-state thermal conduction quenching mathematical model. Using BP neural network, the relationship between the operating parameters of quenching machine and the quenching water cooling convection coefficient is constructed. On this basis, a more accurate quenching machine steel plate temperature prediction model can be obtained.

[0019] The basic idea of ​​the particle swarm optimization algorithm is that particles approach or reach the global optimal solution through mutual cooperation and information sharing. In the algorithm, all particles form a population, which corresponds to a flock of birds or a school of fish in nature. Each particle in the population corresponds to an individual bird in a flock of birds or an individual fish in a school of fish. The optimization process of the algorithm is that the particles in the population simulate the foraging behavior of individual birds or fish. Specifically, all individuals such as birds or fish in the group are randomly distributed in an area, which corresponds to the random initialization process of particles in the PSO algorithm. Assuming that there is only one place with food in this area, the food is the global optimal solution to the specific optimization problem.

[0020] In order to improve the performance of the algorithm-optimized steel plate temperature prediction model, the present invention makes improvements based on the original PSO. The traditional PSO algorithm has the problem of early convergence and is prone to losing population diversity when solving high-dimensional or multiple local optima. Through the study of the existing PSO improved algorithm, it is found that there are few studies on the direct evolution of the historical optimal solution group. Considering that the optimal solution group has a significant guiding effect on the evolution of the current population, the present invention uses this idea to improve and optimize on the basis of the original PSO to further improve the algorithm performance.

[0021] The present invention takes the mathematical model of cooling steel plate of quenching machine as the framework, combines BP neural network and particle swarm optimization algorithm, and adopts hybrid modeling method to construct the temperature prediction model of steel plate of quenching machine with convection heat transfer coefficient as the core, so as to optimize and control the quenching process. Picture 1~2 As shown: The following steps are included: S1. Input the improved PSO algorithm parameters, including the population size, into the mathematical model of the quenching machine cooling steel plate N , dimension D , the maximum number of fitness evaluations maxFES, the learning probability of the particle P c , and input quenching machine parameters and billet information, where the billet information includes billet specifications, types, thermophysical parameters, and initial temperature.

[0022] S2, initializing the flow setting value and roller speed, and calculating the initial steel plate temperature distribution and the objective function value; S3, using improved particle swarm optimization algorithm to optimize the flow setting value and roller speed parameters, and obtain new flow setting values ​​for each section and the optimal roller speed; like Picture 3 As shown in the figure, specifically: according to the fitness value of the individual historical optimal position of the particle, the population is divided into an elite group and an ordinary group, and dynamic reorganization and mutual cooperation are used to search for optimization. In the early stage of evolution, the CL strategy with individual optimal position learning items is adopted for elite particles, and the CL strategy with global optimal solution is adopted for ordinary particles to complete the evolution of the current group; in the later stage of evolution, the individual historical optimal position stored in the historical optimal solution group is directly evolved; Among them, the individual historical optimal position of elite particles adopts the mutation strategy "DE / current-to-rand / 1", and the mutation of ordinary group particles adopts "DE / rand-to- p best” strategy.

[0023] More specifically: the following steps are included: The population is randomly initialized within a given search range, and the steel plate temperature prediction model is called in the program to calculate the quenching water cooling temperature of the steel plate. Then, the fitness value of the objective function is calculated, the optimal solution is recorded, and iteration begins. In the improved PSO, all particles in the population are divided into elite groups and ordinary groups, and optimization is performed through dynamic reorganization and mutual cooperation between them. According to the fitness value of their individual historical optimal position, the population particles are divided into elite groups and ordinary groups; S3.1. After the particles are grouped, they evolve the current group and the historical optimal solution group in turn. For each particle i , its speed d dimension V i d and the current position d dimension X i d Update according to the following formulas: (4) S3.2. For elite particles i ( i ≤0.5*N), update and calculate each dimension according to the following formula d Speed ​​on; (5) In the formula, c represents the acceleration coefficient; rn d Represents a random number that follows a uniform distribution in the range [0, 1]; f i ( d ) represents the index vector f i =[ f i (1) f i (2) ... , f i ( D )] in d dimensional element, which indicates the i The particle d Which particle's d Dimensional learning; Represents particles i The learning model is based on the learning probability PC choose; For the learning probability used in the improved algorithm PC For example, the process parameters are calculated PC As shown below: (6) In the formula, PC i For each particle i No. d dimension, representing a random number between 0 and 1. Then with the elite particle i The corresponding learning probability PC i Value comparison: If the random number is less than PC i Value, then the elite particles i No. d The dimension needs to be the individual historical optimal solution corresponding to the first d dimensional learning, if the random number is greater than / equal to its PC i value, then the particle i No. d The dimension will move to the first corresponding to its own individual historical optimal solution. d Dimensional learning; S3.3. For each ordinary particle, i >0.5*N, the speed of dThe dimension is updated using the CL strategy with a global optimal solution. The specific update formula is as follows: (7) In the formula, rn 1 d and rn 2 d Both represent random numbers in the range [0, 1].

[0024] In the improved PSO algorithm, an adaptive strategy is adopted to maintain population diversity, inertia weight coefficient, acceleration factor c , c 1. c 2 Use the linear decreasing strategy, from large to small, according to the needs of algorithm optimization: (8) (9) (10) (11) In the formula, and Respectively represent the minimum and maximum values ​​of the inertia weight in the algorithm. FES max Indicates the maximum number of fitness value evaluations in the algorithm; FES Indicates the number of times the current fitness value is evaluated in the algorithm; In order to fully develop the area with promising solutions, continue to update the global optimum and improve the solution accuracy: S3.4. For the individual historical optimal position of elite particles, the mutation strategy "DE / current-to-rand / 1" is adopted to strengthen the information exchange between elite particles: (12) In the formula, r 1. r 2 and r 3 means in [1, N / 2] to find three different integers. PV i Represents elite particles i The corresponding mutation vector; F i Represents elite particles i The corresponding adaptive coefficient of variation, adaptive coefficient of variation F i Based on the mean μ F , generated by a Cauchy distribution with a standard deviation of 0.1; S3.5. For each ordinary particle i ( i = N / 2+1, N / 2+2, … , N ) of the individual historical optimal solution, using “DE / rand-to- p The "best / 1" mutation strategy enhances the guidance of the top optimal particles to ordinary individuals: (13) In the formula, Indicates that from the top p %Select the individual historical optimal solution from all historical optimal solutions, r 1. r 2 means in [1, N ] and find two i Integers that are not equal to each other, Represents the individual historical optimal solution and external archive of all particles A A solution vector randomly selected from the union of ; S3.6, as with the differential evolution algorithm, a crossover operation is also required to generate each individual's own experimental vector PU i,g = [ u i,1,g , … , u i,j,g , … , u i,D,g ], which comes from X i,g and V i,g Combination of: (14) In the formula, CR i ∈[0, 1] represents each particle i The corresponding adaptive crossover probability; d rand Indicates that in [1, D ]; S3.7, then continue to perform the selection operation to determine which individual with a better fitness value can survive to the next generation, and each particle i The individual historical optimal solution P i The corresponding experimental vector PU i For comparison: (15) In the formula, f ( P i )and f ( PU i ) respectively represent P i and PU i The corresponding fitness function value is finally used to obtain the new flow setting values ​​of each section of the quenching machine and the optimal operating speed of the roller.

[0025] S4, collecting historical data of the quenching machine during the quenching process, selecting data under stable working conditions as the model input data set, training the BP neural network model, using the new flow setting and roller speed output by the improved PSO algorithm as the BP neural network input, and using the trained BP neural network model to calculate the water cooling convection heat transfer coefficient; Specifically, the steel plate position, time step and space step are input into the mathematical model of cooling steel plate in the quenching machine; S5. Collect historical data of the quenching machine during the quenching process, select data under stable working conditions as the model input data set, train the BP neural network model, use the new flow setting and roller speed output by the improved PSO algorithm as the BP neural network input, and use the trained BP neural network model to calculate the water cooling convection heat transfer coefficient; Specifically: S5.1: preprocess the historical data during the quenching process, including data cleaning, data normalization and other operations, to improve the training efficiency and prediction accuracy of the BP neural network; wherein, the historical data during the quenching process include water flow rate, height of the nozzle from the billet, nozzle water pressure, billet moving speed, injection angle and corresponding water cooling convection heat transfer coefficient.

[0026] S5.2: Construct a BP neural network model, determine the number of neurons in the input layer, hidden layer, and output layer, select appropriate activation functions and loss functions, and use the back propagation algorithm for model training; among them, the number of neurons in the input layer is 2, corresponding to the water flow rate and the moving speed of the steel plate respectively; the number of neurons in the output layer is 1, corresponding to the quenching convection cooling coefficient, the number of neurons in the hidden layer can be adjusted according to the actual situation, generally 5-10 neurons are selected; the activation function can choose the commonly used activation function such as Sigmoid; the loss function can choose the commonly used loss function such as mean square error (MSE), input the preprocessed historical data into the BP neural network model for training, and minimize the error between the predicted value and the true value by continuously adjusting the model parameters. The error back propagation algorithm is used for learning, and the neural network is corrected layer by layer according to the error to obtain the optimal water-cooled convection heat transfer coefficient.

[0027] S6. Based on the simulated steel plate position change, the BP neural network is used to obtain the water cooling convection heat transfer coefficient, the boundary node temperature is calculated, the equation group is solved by the TDMA numerical method, the internal node temperature distribution of the next time step is obtained, the steel plate position is continuously updated, and the steel plate temperature distribution during the quenching process is calculated by the steel plate position until the steel plate leaves the quenching machine; Specifically: Considering that the cooling of the steel plate in the quenching machine needs to meet additional target constraints, that is, the temperature after quenching needs to be controlled within a reasonable range and the cooling rate at the center of the steel plate is less than the critical cooling rate, otherwise the solution is not feasible; considering that there is no internal heat source term inside the steel plate, and the width and length of the quenched steel plate are much larger than the thickness, the temperature gradient in the length and width directions is much smaller than the temperature gradient in the thickness direction ; In addition, the process requires that the temperature of the steel plate is uniform in length and width to reduce deformation, so the cooling process of the steel plate in the quenching machine can be simplified to a one-dimensional non-steady-state heat conduction problem: 1) A one-dimensional non-steady-state mathematical model is adopted and the following assumptions are made to simplify the model.

[0028] a) Assume that the steel plate is a one-dimensional flat plate with no internal heat source; b) Ignore the effect of iron scale on the steel plate during water cooling; c) Quenching mainly relies on the upper and lower nozzles to strengthen the cooling of the quenched steel plate, assuming that the upper and lower nozzles are symmetrically distributed; d) The steel plate moves at a constant speed during the quenching and cooling process; 2) Establish a one-dimensional unsteady-state heat conduction equation. The specific mathematical description is as follows: (16) In the formula, C p represents the specific heat of the steel plate, ; T Indicates the temperature of the steel plate, °C; ρ Indicates the density of steel plate, Kg / m 3 ; τ Indicates time, s; represents the thermal conductivity of the steel plate, , the above thermophysical parameters change with temperature; 3) Discretize the steel plate cross section and use the heat balance method to establish the differential equation of each unit cell.

[0029] Mathematical description and numerical solution of steel plate quenching cooling process: Upper and lower surface nodes: (17) (18) The upper and lower surface nodes are simplified to obtain: (19) (20) Solve for the internal nodal temperatures: (twenty one) in , , transform the above linear equations into a (n-1) × (n-1) matrix form.

[0030] (twenty two) (twenty three) (twenty four) 4) The temperature distribution of the quenched steel plate is obtained by solving the triangular matrix and the one-dimensional non-steady-state boundary equation.

[0031] S7, construct the optimization objective function of the quenching cooling flow of the quenching machine, introduce the penalty function to constrain it, until the iteration is terminated, and finally obtain the set values ​​of the regional flow and roller speed corresponding to the minimum objective function value of the steel plate, and output the corresponding steel plate temperature distribution, wherein the mathematical expression of the optimization objective function of the quenching cooling water flow is: (1) In the formula, f represents the objective function; Q i represents the flow rate of the i-th nozzle; represents the ratio of the upper and lower flow rates of the mth nozzle (water ratio); P1 and P2 represent the penalty term for the post-quenching temperature and the penalty term for the cooling rate of the center of the steel plate, respectively.

[0032] Among them, P1 and P2 represent the penalty term of temperature after quenching and the penalty term of cooling rate at the center of the steel plate, respectively; (2) In the formula, K1 represents the first penalty coefficient; and They represent the average temperatures of the center and surface of the steel plate after it has completely left the quenching machine. represents the temperature difference threshold, (℃); (3) Wherein, K2 represents the second penalty coefficient.

[0033] The present invention obtains a high-precision hybrid prediction model of billet temperature distribution on the basis of hybrid modeling. By collecting actual operation data, a one-dimensional non-steady-state temperature field cooling mathematical model of steel plate is established. The present invention uses an improved particle swarm algorithm to realize the optimization setting of quenching machine operation parameters. The model can simulate the temperature drop curve of steel plate under different operating conditions; the improved particle swarm optimization algorithm is used to optimize the operation parameters required by the model, and it is continuously iterated to find the best objective function value; by comparing the model prediction with the measured data, the parameters are adjusted to achieve the best cooling effect, and the process is intelligentized; by introducing a hybrid model, it can not only overcome the limitations of traditional physical modeling, but also be predicted by a neural network model. This method is particularly suitable for complex and nonlinear industrial environments, and is an ideal choice for optimizing the heat exchange process of the quenching machine, ensuring the effectiveness and accuracy of the steel plate quenching water cooling temperature prediction model.

[0034] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model, characterized in that: Taking the mathematical model of cooling steel plate in quenching machine as the framework, combined with BP neural network and particle swarm optimization algorithm, a hybrid modeling method is used to construct a temperature prediction model of steel plate in quenching machine with convection heat transfer coefficient as the core, and the quenching process is optimized and controlled, including the following steps: S1. Input the improved PSO algorithm parameters, including the population size, into the mathematical model of the quenching machine cooling steel plate N , dimension D , the maximum number of fitness evaluations maxFES , the learning probability of the particle P c , and input quenching machine parameters and billet information; S2, initializing the flow setting value and roller speed, and calculating the initial steel plate temperature distribution and the objective function value; S3, using improved particle swarm optimization algorithm to optimize the flow setting value and roller speed parameters, and obtain new flow setting values ​​for each section and the optimal roller speed; Specifically, according to the fitness value of the individual historical optimal position of the particle, the population is divided into an elite group and an ordinary group, and dynamic reorganization and mutual cooperation are used to search for optimization. In the early stage of evolution, the CL strategy with individual optimal position learning items is adopted for elite particles, and the CL strategy with global optimal solution is adopted for ordinary particles to complete the evolution of the current group; in the later stage of evolution, the individual historical optimal position stored in the historical optimal solution group is directly evolved; Among them, the individual historical optimal position of elite particles adopts the mutation strategy "DE / current-to-rand / 1", and the mutation of ordinary group particles adopts "DE / rand-to- p best” strategy; S4, inputting the steel plate position, time step and space step in the mathematical model of cooling the steel plate in the quenching machine; S5. Collect historical data of the quenching machine during the quenching process, select data under stable working conditions as the model input data set, train the BP neural network model, use the new flow setting and roller speed output by the improved PSO algorithm as the BP neural network input, and use the trained BP neural network model to calculate the water cooling convection heat transfer coefficient; S6. Based on the simulated steel plate position change, the BP neural network is used to obtain the water cooling convection heat transfer coefficient, the boundary node temperature is calculated, the equation group is solved by the TDMA numerical method, the internal node temperature distribution of the next time step is obtained, the steel plate position is continuously updated, and the steel plate temperature distribution during the quenching process is calculated by the steel plate position until the steel plate leaves the quenching machine; S7. Construct an optimization objective function for the quenching cooling flow rate of the quenching machine, introduce a penalty function to constrain it, and terminate the iteration until the set values ​​of the regional flow rate and roller speed corresponding to the minimum objective function value of the steel plate are finally obtained, and the corresponding steel plate temperature distribution is output.

2. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 1, characterized in that: The steel billet information in step S1 includes: steel billet specifications, types, thermophysical parameters, and initial temperature.

3. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 1, characterized in that: Step S3 specifically includes the following steps: S3.

1. After the particles are grouped, they evolve the current group and the historical optimal solution group in turn. For each particle i , the current position d dimension X i d Update as follows: ； S3.

2. For elite particles i ( i = 1, 2, …, N / 2 ), update and calculate each dimension according to the following formula d Speed ​​on: ； In the formula, c represents the acceleration coefficient; rn d Represents a random number that follows a uniform distribution in the range [0, 1]; f i ( d ) represents the index vector f i =[ f i (1) f i (2) ... , f i ( D )] in d dimensional element, which indicates the i The particle d Which particle's d Dimensional learning; Represents particles i The learning model is based on the learning probability Pc choose; For the learning probability used in the improved algorithm Pc For example, the process parameters are calculated Pc As shown below: ； In the formula, Pc i For each particle i No. d dimension, representing a random number between 0 and 1, and then with the elite particles i The corresponding learning probability Pc i Value comparison: If the random number is less than Pc i Value, then the elite particles i No. d The dimension needs to be the individual historical optimal solution corresponding to the first d dimensional learning, if the random number is greater than / equal to its Pc i value, then the particle i No. d The dimension will move to the first corresponding to its own individual historical optimal solution. d Dimensional learning; S3.3 For each ordinary particle, its velocity d The dimension is updated using the CL strategy with a global optimal solution. The specific update formula is as follows: ； In the formula, rn 1 d and rn 2 d Both represent random numbers in the range [0, 1].

4. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 3, characterized in that: In step S3, in order to fully develop the area with promising solutions, the global optimum is continuously updated to improve the solution accuracy: S3.

4. For the individual historical optimal position of elite particles, the mutation strategy "DE / current-to-rand / 1" is adopted to strengthen the information exchange between elite particles: ； In the formula, r 1. r 2 and r 3 means in [1, N / 2] to find three different integers. PV i Represents elite particles i The corresponding mutation vector; F i Represents elite particles i The corresponding adaptive coefficient of variation, adaptive coefficient of variation F i Based on the mean μ F , generated by a Cauchy distribution with a standard deviation of 0.1; S3.

5. For each ordinary particle i ( i = N / 2+1, N / 2+2, … , N ) of the individual historical optimal solution, using "DE / rand-to- p The "best / 1" mutation strategy enhances the guidance of the top optimal particles to ordinary individuals: ； In the formula, Indicates that from the top p %Select the individual historical optimal solution from all historical optimal solutions, r 1. r 2 means in [1, N ] and find two i Integers that are not equal to each other, Represents the individual historical optimal solution and external archive of all particles A A solution vector randomly selected from the union of ; S3.

6. Perform a crossover operation on each individual to generate their own experimental vector PU i,g = [ u i,1,g , … , u i,j,g , … , u i,D,g ], which comes from X i,g and V i,g Combination of: ； In the formula, CR i ∈[0, 1] represents each particle i The corresponding adaptive crossover probability; d rand Indicates that in [1, D ]; S3.

7. Each particle i The individual historical optimal solution P i The corresponding experimental vector PU i For comparison: ； In the formula, f ( P i )and f ( PU i ) respectively represent P i and PU i The corresponding fitness function value is finally used to obtain the new flow setting values ​​of each section of the quenching machine and the optimal operating speed of the roller.

5. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 1, characterized in that: Step S5 specifically includes the following steps: S5.1: preprocessing the historical data during the quenching process, including data cleaning and data normalization operations, to improve the training efficiency and prediction accuracy of the BP neural network; wherein the historical data during the quenching process includes water flow rate, nozzle height from the billet, nozzle water pressure, billet moving speed, injection angle, and corresponding water cooling convection heat transfer coefficient; S5.2: Construct a BP neural network model, the number of neurons in the input layer, hidden layer and output layer, determine the water flow rate and billet movement speed as the input layer, the water cooling convection heat transfer coefficient as the output layer, select appropriate activation function and loss function, input the preprocessed historical data into the BP neural network model for training, and minimize the error between the predicted value and the true value by continuously adjusting the model parameters to obtain the optimal water cooling convection heat transfer coefficient.

6. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 1, characterized in that: Calculating the boundary node temperature in step S6 specifically includes: A one-dimensional unsteady-state heat conduction equation is established, and the specific mathematical description is as follows: ； In the formula, C p Indicates the specific heat of the steel plate; T Indicates the temperature of the steel plate; ρ Indicates the density of steel plate; τ Indicates time; Indicates thermal conductivity of steel plate. The above thermal physical parameters change with temperature. Discretize the steel plate section and use the heat balance method to establish the differential equation of each unit cell: Mathematical description and numerical solution of steel plate quenching cooling process: upper and lower surface nodes: ； ； Simplifying, we get: ； ； Solve for the internal nodal temperatures: ； in , .

7. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 6, characterized in that: In step S6, the equation group is solved by the TDMA numerical method to obtain the temperature distribution of the quenched steel plate: Convert formula (21) into a (n-1) × (n-1) matrix form: ； ； ； The temperature distribution of the quenched steel plate is obtained by solving the triangular matrix and the one-dimensional unsteady boundary equation.

8. The method for optimizing cooling process parameters of a steel billet in a roller quenching machine based on a hybrid model according to claim 1, characterized in that: Step S7 constructs an optimization function with the water consumption of the quenching machine as the target, taking into account that the cooling of the steel plate in the quenching machine needs to meet additional target constraints, that is, the temperature after quenching needs to be controlled within a reasonable range and the cooling rate of the center of the steel plate is less than the critical cooling rate, otherwise the solution is not feasible; the penalty function method is introduced for constraint processing, wherein the mathematical expression of the optimization objective function of the quenching cooling water flow is: ； In the formula, f represents the objective function; Q i represents the flow rate of the i-th nozzle; represents the ratio of the upper and lower flow rates of the mth nozzle; P1 and P2 represent the penalty term for the post-quenching temperature and the penalty term for the cooling rate at the center of the steel plate, respectively; Among them, P1 and P2 represent the penalty term of temperature after quenching and the penalty term of cooling rate at the center of the steel plate, respectively; ； In the formula, K1 represents the first penalty coefficient; and They represent the average temperatures of the center and surface of the steel plate after it has completely left the quenching machine. represents the temperature difference threshold; ； Wherein, K2 represents the second penalty coefficient.

Images