Rolling speed optimization method in strip steel hot continuous rolling and finish rolling process

Abstract

The invention discloses a rolling speed optimization method in a strip steel hot continuous rolling and finish rolling process, which comprises the following steps of: training by using historical data to obtain a finish rolling temperature and coiling temperature prediction model, and performing rolling optimization in stages; comprising the steps of determining input and output variables and loss functions of a finish rolling and finish rolling temperature prediction model and a coiling temperature prediction model in the strip steel hot continuous rolling process, and training the prediction models; constructing a rolling speed single-step optimization objective function; carrying out single-step multi-objective optimization solution, and carrying out multi-round iteration to obtain an optimal rolling speed solution set; and selecting the solution with the maximum numerical value, calculating to obtain the acceleration and the strip steel length, and performing the next optimization until the optimization is stopped. And finally, the strip steel length reaches the finished strip steel length. The precision rolling and finishing temperature precision and the coiling temperature precision in the strip steel hot continuous rolling process can be improved, the total rolling time is shortened, and the rolling efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of strip hot continuous rolling process optimization technology, and relates to a method for optimizing the rolling speed in the finishing rolling process. Specifically, it relates to a data-driven optimization method for the time-speed-distance curve (TVD curve) of the finishing rolling process of strip hot continuous rolling based on the NSGAⅡ multi-objective genetic algorithm, which can improve the accuracy of the finishing rolling temperature and coiling temperature, reduce the total rolling time, and improve the rolling efficiency in the strip hot continuous rolling process. Background Technology

[0002] The finishing rolling process is a crucial step in the hot continuous rolling of strip steel. Following the roughing rolling process and preceding the laminar flow cooling process, the finishing mill stand is the core component of the finishing rolling process. The roughing intermediate billet is continuously rolled through multiple finishing mill stands to produce strip steel of a specified precision. Properly setting the rolling speed in the finishing rolling process, i.e., the speed at which the strip steel passes through the finishing mill stands, is of great significance for improving product quality and stabilizing production.

[0003] Existing methods for setting the TVD curve in the hot strip rolling finishing process include: setting based on standard experience tables: Based on experience summarized in production, appropriate threading speeds, accelerations, and maximum rolling speeds are determined according to different steel grades and dimensions, and a standard experience table is created; manual setting via human-machine interface: operators manually adjust the settings of threading speed, acceleration, and maximum rolling speed based on actual production conditions and their production experience; setting using mechanistic formulas: the threading speed of the strip is inferred from the mechanistic model used to calculate the final rolling temperature; and setting using feedback control: the rolling speed is adjusted online based on the deviation between the measured final rolling temperature and the target temperature. Existing methods only consider the accuracy of the final rolling temperature; they rely too heavily on experience and mechanistic models, placing high demands on engineers and consuming significant effort; and the set TVD curves often deviate from actual applications, requiring substantial adjustments for online application. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this invention provides a TVD curve optimization method for the finishing process of hot strip rolling based on the NSGAⅡ (Non-dominated Sorting Genetic Algorithm II) multi-objective genetic algorithm. The method trains the finishing temperature prediction model and the coiling temperature prediction model using process data, and performs rolling optimization over time, thereby improving the accuracy of the finishing temperature and the coiling temperature in the hot strip rolling process and reducing the total rolling time.

[0005] This invention utilizes historical data to train a feedforward neural network prediction model for the finishing rolling temperature and coiling temperature during the acceleration and deceleration stages. It employs a phased, time-step rolling optimization method to optimize the rolling speed settings for the strip threading, acceleration, and deceleration stages. In each optimization step, a suitable time step is first selected, and an objective function (i.e., the mean square error of the finishing rolling temperature and coiling temperature within that time step) is established. Then, the NSGAII algorithm is used to solve the multi-objective optimization problem, minimizing the mean square error of the finishing rolling temperature and coiling temperature within that time step, with the strip rolling speed as the decision variable. Next, a solution with a higher rolling speed is selected from the obtained Pareto solution set to achieve a larger strip flow rate per second and shorten the total rolling time. Subsequently, the calculated rolling speed is recorded, the strip exit length is updated, and the next multi-objective optimization calculation is performed. Finally, the rolling speeds recorded at each step are connected to plot a vt graph, obtaining the optimized TVD curve for the finishing rolling process, thus achieving optimization of the rolling speed in the hot strip rolling finishing process.

[0006] The TVD curve optimization method for the hot strip rolling finishing process proposed in this invention, based on the NSGAII multi-objective genetic algorithm, is a completely data-driven approach. It does not require obtaining precise mechanistic formulas; it only needs to train and learn relevant patterns using historical data. This invention simultaneously considers the optimization of finishing rolling temperature accuracy and coiling temperature accuracy, representing a multi-objective optimization method, whereas existing methods often only consider a single indicator. This invention connects the finishing rolling, laminar cooling, and coiling processes, considering the correlation and coupling between multiple processes and systems during model training and optimization. This makes the trained model more accurate, achieves multi-process optimization, and better conforms to practical application standards.

[0007] The technical solution provided by this invention is:

[0008] A method for optimizing the rolling speed in the finishing process of hot strip rolling based on the NSGA II multi-objective genetic algorithm is presented. This is a fully data-driven optimization algorithm that uses historical data to train prediction models for the finishing rolling temperature and coiling temperature, and optimizes them in stages over time. Considering the coupling of multiple processes, it simultaneously optimizes the accuracy of the finishing rolling temperature and coiling temperature, reducing the total rolling time and increasing output. The main steps include:

[0009] 1) Collect data and train the finishing rolling temperature prediction model and coiling temperature prediction model for the acceleration and deceleration stages.

[0010] Numerous factors influence the finishing rolling temperature and coiling temperature, and their relationships are complex. This invention employs a feedforward neural network to establish a prediction model for these temperatures. The maximum number of training epochs is 1000, the training objective error is 1e-6, and the network learning rate is 0.01. The input and output variables of the model are determined by analyzing the mechanism of the hot strip rolling process. The input variables of the finishing rolling temperature prediction model are: strip rolling speed v, strip exit length l, cooling water flow rate ω, strip exit thickness h of the last stand, and finishing mill inlet temperature T. in The output variable is the finishing rolling temperature T. d The model structure is 5-50-1; the input variables for the coiling temperature prediction model are the strip rolling speed v, the strip exit thickness h of the last stand, and the finishing rolling temperature T. d The number of laminar flow cooling water sections n, and the output variable is the coiling temperature T. c The model structure is 4-50-1. This invention takes certain training samples (historical data, i.e., the data set of input and output variables corresponding to the prediction model over a period of time) from both the accelerated and decelerated rolling stages to train prediction models for the finishing rolling temperature and coiling temperature during the accelerated and decelerated stages. The loss function for training the finishing rolling temperature prediction model is:

[0011]

[0012] in, The predicted final rolling temperature is obtained by substituting the i-th set of input variable samples into the finishing rolling temperature prediction model. Let N be the i-th final rolling temperature sample value, and N be the number of samples.

[0013] The loss function for training the winding temperature prediction model is:

[0014]

[0015] in, Substitute the i-th set of input variable samples into the winding temperature prediction model to calculate the predicted winding temperature value. Let N be the temperature sample value of the i-th winding, and N be the number of samples.

[0016] 2) Construct a single-step optimization objective function for rolling speed;

[0017] In each optimization step, let the step size be s. Calculate the mean square error of the finishing rolling temperature and the mean square error of the coiling temperature within this step size, expressed as:

[0018]

[0019] Wherein, (Equation 3) represents the mean square error of the finishing rolling temperature, and (Equation 4) represents the mean square error of the coiling temperature.dtarget T is the target finishing rolling temperature. ctarget For the target winding temperature, Let be the finishing rolling temperature of the i-th sampling point. Let be the coiling temperature at the i-th sampling point, where 1 ≤ i ≤ s, and is an integer (assuming the sampling interval is γ seconds, then the time covered by this step size is s * γ seconds). The objective of this invention is to simultaneously minimize the mean square error of the finishing rolling temperature and the mean square error of the coiling temperature in each step of the rolling speed optimization, i.e., to solve the optimization problem as follows:

[0020]

[0021] The above formula is the single-step optimization objective function for rolling speed, where the decision variable v s This represents the rolling speed of the last sampling point within this step.

[0022] 3) Use the NSGAⅡ algorithm to perform single-step multi-objective optimization to obtain the optimal rolling speed solution set;

[0023] First, a primary population (parent population) of 20 individuals is randomly generated, representing 20 possible rolling speeds (decision variables) v. s The solution is then used for crossover and genetics to generate offspring populations. The parent and offspring populations are then merged into a population of 40 individuals. Let the optimization step size be s, the data sampling interval be γ seconds, the previous optimization result be the rolling speed v0 (i.e., the rolling speed of the last sampling point in the previous step), and the strip length calculated in the previous step be L0. Then the acceleration a corresponding to the current optimization step size is:

[0024] a=(v s -v0) / (s*γ) (Equation 6)

[0025] For each sampling point i, 1 ≤ i ≤ s, and i is an integer, the strip rolling speed v i and the corresponding strip length L i for:

[0026] v i =v0+a*i*γ (Equation 7)

[0027]

[0028] Based on equations (6)-(8), calculate the rolling speed and strip length for each sampling point corresponding to each individual in the population, and then substitute them into the finishing rolling temperature prediction model and the coiling temperature prediction model for calculation. The other input variables are set to default values ​​and are not optimized.

[0029] Next, calculate the performance index corresponding to each individual in the population according to (Equation 3)-(Equation 4), namely, the mean square error of finishing rolling temperature F1 and the mean square error of coiling temperature F2 within the optimization step size.

[0030] Then, a non-dominated ranking is performed based on the two performance metrics (mean square error of finishing rolling temperature F1 and mean square error of coiling temperature F2) corresponding to each individual in the population. For the solution v of the two decision variables... a v b (i.e., two individuals in the population with different rolling speeds), if F1(v a )≤F1(v b ), F2(v a )≤F2(v b ), and there exist j∈1,2 such that F j (v a ) <F j (v b If the condition is met, then the solution is called v. a Dominate v b If a solution to a decision variable is not dominated by other solutions, then it is called a non-dominated solution. In a population, all non-dominated solutions form a set called a Pareto surface or Pareto set, with a Pareto rank of 1. Removing non-dominated solutions from the population, the remaining solutions that can be non-dominated form another Pareto surface with a Pareto rank of 2; and so on, resulting in a series of Pareto surfaces with different ranks.

[0031] Next, the crowding distance of each individual in the population is calculated, and natural selection is performed using an elitist mechanism. Specifically, individuals from the Pareto surfaces with smaller Pareto ranks are preferentially retained. If the total number of individuals in the first n-1 Pareto surfaces is less than 20, while the total number of individuals in the first n Pareto surfaces is greater than 20, then the crowding distances of individuals in the Pareto surface of rank n are compared, and individuals with larger crowding distances are preferentially retained, until the total number of retained individuals equals 20. This results in a next-generation population with a total of 20 individuals.

[0032] Then, a new round of crossover mutation, performance index calculation, non-dominated sorting, and natural selection is performed, for a total of 100 iterations, to obtain the optimal rolling speed solution set. In multi-objective optimization problems, the Pareto optimal solution refers to a solution that can no longer improve the value of one objective function without harming the values ​​of other objective functions.

[0033] 4) Select the solution with the largest value from the optimal rolling speed solution set and proceed to the next optimization step until the requirements for stopping optimization at each stage are met. Ultimately, this ensures that the strip length reaches the specified finished strip length.

[0034] In the Pareto optimal solution set for rolling speed, a subset of solutions with moderate mean square errors in both final rolling temperature and coiling temperature is selected. This involves eliminating "extreme solutions" (one with excellent performance and the other with poor performance). The solution with the largest value in this subset is chosen to achieve the goal of shortening total rolling time and increasing output. This solution is denoted as v. * Next, calculate v. * The corresponding acceleration a * and strip length L * :

[0035] a * =(v * -v0) / (s*γ) (Equation 9)

[0036]

[0037] v * and L * Substitute this into the next optimization step, i.e., repeat step 3).

[0038] Requirements for stopping optimization at each stage:

[0039] In the TVD curve optimization process of the finishing rolling process, optimization is performed sequentially for the strip threading, acceleration, and deceleration stages. The finishing rolling temperature prediction models for the acceleration stage and coiling temperature prediction models trained in step 1) are used uniformly during the optimization calculations for the strip threading and acceleration stages. Similarly, the finishing rolling temperature prediction models for the deceleration stage and coiling temperature prediction models trained in step 1) are used during the optimization calculations for the deceleration stage. First, the TVD curve optimization for the strip threading stage is performed. At this point, the acceleration a corresponding to each optimization step size in step 3) is 0, when the strip length L... * Optimization stops when the rolling speed is ≥50m; the next stage, the acceleration stage, begins when the rolling speed v... * When the pre-defined maximum rolling speed is reached, acceleration stops and the movement remains constant. When the strip length L... * Optimization stops when the predetermined starting deceleration length is reached; then, the optimization proceeds to the next stage, the deceleration stage, when the rolling speed v... * Stop decelerating when the specified strip throwing speed is reached, and maintain a constant speed until the strip length L is reached. * When the specified length of finished strip is reached (at which point the last stand discards the strip), optimization stops.

[0040] 5) Plot the optimized TVD curve;

[0041] The optimization results of each step in the belt-threading, acceleration, and deceleration phases are v * Record the data, combine it with the optimization step size for each step, plot the points on the velocity-time (vt) coordinate system, and connect the points to obtain the optimized TVD curve.

[0042] Through the above steps, the rolling speed of the hot strip finishing process is optimized based on the NSGA II multi-objective genetic algorithm.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention provides a TVD curve optimization method for the hot strip rolling finishing process based on the NSGAII multi-objective genetic algorithm. It trains a finishing temperature prediction model and a coiling temperature prediction model using process data, and performs rolling optimization over time, improving the accuracy of the finishing temperature and coiling temperature in the hot strip rolling process while reducing the total rolling time. This invention considers the inter-process and multi-system coupling between finishing, laminar cooling, and coiling operations, making the trained model more accurate, achieving multi-process optimization, and better conforming to the standards of practical applications in hot strip rolling finishing. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the time-speed-distance curve (TVD curve) of the finishing rolling process studied in this invention.

[0046] Figure 2 A schematic diagram of the feedforward neural network prediction model for finishing rolling temperature and coiling temperature established in this invention.

[0047] Among them, (a) is the feedforward neural network prediction model for finishing rolling temperature; (b) is the feedforward neural network prediction model for coiling temperature.

[0048] Figure 3 This is a flowchart illustrating the algorithm for a specific implementation of the method of the present invention.

[0049] Figure 4 The following is an example of the TVD curve optimization results for the finishing rolling process provided by this invention;

[0050] (a) shows the comparison results with L1 data; (b) shows the comparison results with L2 data. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings and examples, but this does not limit the scope of the invention in any way.

[0052] This invention provides a TVD curve optimization method for the hot strip rolling finishing process based on the NSGAⅡ multi-objective genetic algorithm. The method uses process data to train a finishing rolling temperature prediction model and a coiling temperature prediction model, and performs rolling optimization over time.

[0053] In its specific implementation, this invention utilizes historical data to learn and train a prediction model for the finishing rolling temperature and coiling temperature, and then optimizes it in stages over time; including the following steps:

[0054] 1) Collect data to determine the input, output variables, and loss function of the finishing and coiling temperature prediction models for the hot strip rolling process, and train the prediction models. The input variables of the finishing temperature prediction model are strip rolling speed, strip exit length, cooling water flow rate, strip exit thickness of the last stand, and finishing inlet temperature, and the output variable is finishing temperature. The input variables of the coiling temperature prediction model are strip rolling speed, strip exit thickness of the last stand, finishing temperature, and number of laminar cooling water sections, and the output variable is coiling temperature.

[0055] 2) Construct a single-step optimization objective function for rolling speed;

[0056] In each optimization step, let the step size be s. Calculate the mean square error of the finishing rolling temperature and the mean square error of the coiling temperature within this step size, expressed as follows:

[0057]

[0058] In Equation 3, F1 represents the mean square error of the finishing rolling temperature, and in Equation 4, F2 represents the mean square error of the coiling temperature; T dtarget T is the target finishing rolling temperature. ctarget For the target winding temperature, Let be the finishing rolling temperature of the i-th sampling point. Let be the winding temperature of the i-th sampling point, where 1≤i≤s and is an integer; let the sampling interval be γ seconds, then the time covered by this step size is s*γ seconds;

[0059] The optimization objective is to simultaneously minimize the mean square error of the finishing rolling temperature and the mean square error of the coiling temperature in each step of the rolling speed optimization. In other words, the single-step optimization objective function for the rolling speed is expressed as:

[0060]

[0061] Where the decision variable v s This refers to the rolling speed of the last sampling point within this step.

[0062] 3) Perform single-step multi-objective optimization solution, and obtain the optimal rolling speed solution set by performing multiple rounds of crossover mutation, performance index calculation, non-dominated sorting, and natural selection iteration;

[0063] 4) Select the solution with the largest value from the optimal rolling speed solution set and proceed to the next optimization step until optimization stops;

[0064] Through the above steps, the rolling speed of the hot strip finishing process is optimized based on the NSGAⅡ multi-objective genetic algorithm.

[0065] Figure 1 The TVD curve of the finishing rolling process studied in this invention is shown, depicting the change in strip speed. Modern hot strip rolling finishing processes all employ a speed-increasing rolling mechanism. Figure 1 As shown, section S1 is the strip-threading speed operation section, during which the strip enters stands F1 to F7 until its head is within approximately 50m of stand F7 (precision instruments such as thickness gauges, width gauges, and temperature gauges are arranged on both sides of the roller conveyor approximately 50m from the exit of the last stand F7). Sections S2-S4 are the acceleration operation section, which is used to compensate for the temperature drop along the length of the strip, improve the accuracy of the final rolling temperature, and increase output. At time 4, the strip speed reaches the maximum rolling speed, and acceleration ceases, maintaining the highest speed (section S4). Sections S5-S6 are the strip-deceleration operation section. At time 5, the tail of the strip leaves the stand where deceleration began, and the strip begins to decelerate. At time 6, it decelerates to the strip-throwing speed, and deceleration ceases, maintaining the strip-throwing speed (section S6). Strip-throwing is completed at time 7. Subsequently, the stand continues to decelerate to the strip-threading speed of the next strip (section S7), and then maintains the strip-threading speed of the next strip (section S8).

[0066] The following embodiments of the present invention select a slab produced by a steel mill. The slab width is 1785 mm, the target thickness is 3.795 mm, and the total finishing rolling time is 66.1 s. The target finishing rolling temperature of this slab is 930℃, and the target coiling temperature is 750℃. The strip rolling speed v, strip exit length l, cooling water flow rate ω, strip exit thickness h at the last stand, and finishing mill inlet temperature T during the acceleration and deceleration stages are extracted using a sampling frequency of 0.02 s. in Finishing rolling temperature T d Number of laminar flow cooling water sections n, winding temperature T c The measurement data were used to train the prediction models for the finishing rolling temperature and coiling temperature at each stage, with 2275 samples in the acceleration stage and 378 samples in the deceleration stage.

[0067] Figure 2 The structure of the feedforward neural network prediction model for finishing rolling temperature and coiling temperature established in this invention is shown. The input and output variables of the model are shown. The model structures are 5-50-1 and 4-50-1, respectively (as shown in the figure, both models contain a hidden layer, and the number of neurons in the hidden layer is 50).

[0068] Figure 3 This is a flowchart illustrating the algorithm for optimizing the TVD curve in the hot strip finishing process based on the NSGAⅡ algorithm proposed in this invention. The specific algorithm is shown below:

[0069]

[0070]

[0071] After training the prediction models for the finishing rolling temperature and coiling temperature at each stage, multi-objective optimization of the TVD curve was performed. The optimization results are as follows: Figure 4 As shown, L1 level data refers to basic automation level data, i.e., data measured on the rolling mill floor, specifically the rolling speed measured on-site; L2 level data refers to process automation level data, i.e., pre-calculated and set data, specifically the calculated and set rolling speed, i.e., TVD information. Before optimization (i.e., actual production conditions), the mean square error of the final rolling temperature was 4.9955℃, and after optimization, it was 4.8955℃, an improvement of 2%; before optimization, the mean square error of the coiling temperature was 4.6048℃, and after optimization, it was 3.5733℃, an improvement of 22.4%; before optimization, the total rolling time was 66.1s, and after optimization, it was 64.62s, an improvement of 2.24%.

[0072] The embodiments disclosed above are intended to help further understand the present invention. However, those skilled in the art will understand that various substitutions and modifications are possible without departing from the scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the scope of protection of the present invention is defined by the claims.

Claims

1. A method for optimizing the rolling speed in the hot continuous strip finishing process, characterized in that, A prediction model for the finishing rolling temperature and coiling temperature is obtained by learning and training using historical data, and then optimized in stages over time; including the following steps: 1) Collect data, determine the input, output variables and loss function of the finishing rolling temperature prediction model and the coiling temperature prediction model for the hot strip rolling process, and train the prediction model. The input variables for the finishing and final rolling temperature prediction model are strip rolling speed, strip exit length, cooling water flow rate, strip exit thickness at the last stand, and finishing inlet temperature; the output variable is finishing and final rolling temperature. The input variables for the coiling temperature prediction model are strip rolling speed, strip exit thickness at the last stand, finishing and final rolling temperature, and number of laminar cooling water sections; the output variable is coiling temperature. 2) Construct a single-step optimization objective function for rolling speed; In each optimization step, let the step size be s. Calculate the mean square error of the finishing rolling temperature and the mean square error of the coiling temperature within this step size, expressed as follows: In Equation 3, F1 represents the mean square error of the finishing rolling temperature, and in Equation 4, F2 represents the mean square error of the coiling temperature; T dtarget T is the target finishing rolling temperature. ctarget For the target winding temperature, Let be the finishing rolling temperature of the i-th sampling point. Let be the winding temperature of the i-th sampling point, where 1≤i≤s and is an integer; let the sampling interval be γ seconds, then the time covered by this step size is s*γ seconds; The optimization objective is to simultaneously minimize the mean square error of the finishing rolling temperature and the mean square error of the coiling temperature in each step of the rolling speed optimization. In other words, the single-step optimization objective function for the rolling speed is expressed as: Where the decision variable v s This refers to the rolling speed of the last sampling point within this step. 3) Perform single-step multi-objective optimization solution, and obtain the optimal rolling speed solution set by performing multiple rounds of crossover mutation, performance index calculation, non-dominated sorting, and natural selection iteration; 4) Select the solution with the largest value from the optimal rolling speed solution set, calculate the corresponding acceleration and strip length, and repeat step 3) to perform the next optimization until optimization stops; Ultimately, this allows the strip length to reach the finished strip length; The above steps optimize the rolling speed in the hot strip finishing process.

2. The method for optimizing the rolling speed in the hot continuous strip finishing process as described in claim 1, characterized in that, In step 1), the loss function P1 for training the finishing rolling temperature prediction model is expressed as: in, The predicted final rolling temperature is obtained by substituting the i-th set of input variable samples into the finishing rolling temperature prediction model. Let N be the i-th final rolling temperature sample value, and N be the number of samples; The loss function P2 for training the temperature prediction model is expressed as: in, Substitute the i-th set of input variable samples into the winding temperature prediction model to calculate the predicted winding temperature value. Let N be the temperature sample value of the i-th winding, and N be the number of samples.

3. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 2, characterized in that, Step 3) specifically involves using the NSGA II algorithm to perform single-step multi-objective optimization to obtain the optimal rolling speed solution set; including: First, a primary population, or parent population, is randomly generated, consisting of multiple individuals, representing multiple possible rolling speeds v. s The solution; Next, crossover and genetics are performed to produce offspring populations; Then the parent population and the child population are merged to generate a population with twice the number of individuals; Let the optimization step size be s, the data sampling interval be γ seconds, the rolling speed v0 of the last sampling point in the previous step, and the strip length calculated in the previous step be L0. The acceleration a corresponding to the current optimization step size is expressed as: a=(v s -v0) / (s*γ) (equation 6) For each sampling point i, 1 ≤ i ≤ s, and i is an integer, the strip rolling speed v i and the corresponding strip length L i Represented as: v i =v0+a*i*γ (Expression 7) Based on equations (6)-(8), calculate the rolling speed and strip length for each sampling point corresponding to each individual in the population; then substitute these values ​​into the finishing rolling temperature prediction model and the coiling temperature prediction model to obtain the results.

4. The method for optimizing the rolling speed in the hot continuous strip finishing process as described in claim 3, characterized in that, Calculate the performance index corresponding to each individual in the population according to (Equation 3)-(Equation 4), that is, obtain the mean square error of finishing rolling temperature and the mean square error of coiling temperature within the optimization step size.

5. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 4, characterized in that, Non-dominated sorting is performed based on two performance indicators for each individual in the population: mean square error of finishing rolling temperature and mean square error of coiling temperature; including: For the solution v of two decision variables a v b That is, the two rolling speeds in the population: If F1(v) a )≤F1(v b ), F2(v a )≤F2(v b ), and there exist j∈1,2 such that F j (v a ) <F j (v b If ) holds true, then the solution v a Dominate v b A solution to a decision variable is not dominated by other solutions; it is a non-dominated solution. All non-dominated solutions in the population are combined into a set, which is a Pareto surface or Pareto set, and the Pareto level is 1. Remove non-dominated solutions from the population; the Pareto surface formed by the non-dominated solutions in the remaining solutions has a Pareto rank of 2. This results in a series of Pareto surfaces of different levels.

6. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 5, characterized in that, Calculate the crowding distance for each individual in the population and use an elite mechanism for natural selection; that is, prioritize individuals in the Pareto surface with small Pareto rank values ​​to obtain the next generation population.

7. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 6, characterized in that, If the total number of individuals in the current n-1 Pareto surfaces is less than 20, while the total number of individuals in the first n Pareto surfaces is greater than 20, the crowding distance of individuals in the Pareto surface of level n is compared, and individuals with larger crowding distances are selected to be retained first, until the total number of retained individuals is equal to 20, thus obtaining the next generation population with a total number of 20 individuals.

8. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 5, characterized in that, In step 4), the acceleration and strip length corresponding to the solution with the largest selected value are calculated and expressed as: a * =(v * -v0) / (s*γ) (equation 9) Among them, a * For acceleration; L * This refers to the length of the strip steel.

9. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 8, characterized in that, In step 4), the requirements for stopping optimization include: In the process of optimizing the TVD curve of the finishing rolling process, the optimization of the strip threading, acceleration and deceleration stages are carried out in sequence. When optimizing the strip threading and acceleration stages, the finishing rolling temperature prediction model and coiling temperature prediction model of the acceleration stage trained in step 1) are used. When optimizing the deceleration stage, the finishing rolling temperature prediction model and coiling temperature prediction model of the deceleration stage trained in step 1) are used. When optimizing the TVD curve during the strip threading stage, in step 3), the acceleration a = 0 for each optimization step size. When the strip length L * Optimization stops when the rolling speed is ≥50m; the next stage, the acceleration stage, begins; when the rolling speed v... * When the preset maximum rolling speed is reached, acceleration stops and the movement remains constant; when the strip length L... * Optimization stops when the length at which deceleration begins is reached; then, optimization proceeds to the next stage, the deceleration stage, when the rolling speed v... * Stop decelerating when the steel throwing speed is reached, and maintain a constant speed until the strip length L is reached. * Optimization stops when the finished strip length is reached.

10. The method for optimizing rolling speed in the hot continuous strip finishing process as described in claim 9, characterized in that, Further, the optimized TVD curve is plotted by recording the optimized rolling speed results for each step of the threading, acceleration, and deceleration stages. Based on the optimization step size for each step, points are plotted and connected on the speed-time coordinate system to obtain the optimized TVD curve.

Images