Cold continuous rolling multi-rack edge drop control method and system based on machine learning

Abstract

The invention provides a cold continuous rolling multi-rack edge drop control method and system based on machine learning, and relates to the field of cold continuous rolling process control. The method comprises the following steps: acquiring key parameters of historical production and edge drop detection data of strip steel, introducing a unified timestamp mechanism, then performing preprocessing to obtain an edge drop control data set, and extracting a data pair set of corresponding sample points of each rack; constructing a rack recursion edge drop prediction model; training, verifying, testing and evaluating the model by adopting the data pair set; to-be-rolled strip steel is divided into a plurality of to-be-detected sections in the rolling direction, actual production key parameters of each to-be-detected section are obtained and input into the model, and the edge drop value, predicted at the fifth rack, of each to-be-detected section is output; calculating the qualified rate according to the section edge drop threshold value, and then judging whether the qualified rate is greater than the qualified rate threshold value; if not, rolling control parameters of the first rack, the second rack, the third rack and the third rack are optimized; if yes, the current actual production key parameters serve as the optimal parameters. According to the invention, the accuracy and precision of edge drop control are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of cold continuous rolling process control, and particularly relates to a cold continuous rolling multi-stand edge drop control method and system based on machine learning. BACKGROUND

[0002] The same plate difference reflects the shape accuracy of the plate from the horizontal and vertical dimensions, directly reflects the rolling technology level, and affects the magnetic properties and mechanical processing characteristics of high-end plate products such as silicon steel and medium-high carbon steel. In order to adapt to the upgrading of the downstream manufacturing industry, high-end plate products such as silicon steel and medium-high carbon steel generally need to have extremely low same plate differences, and the cold rolling preparation process needs to be strictly controlled. Unlike hot rolling crown control and cold rolling shape control, the cold continuous rolling control means for the same plate difference is less, mainly relying on the roll shifting optimization setting of the upstream large reduction rate stand, the dynamic compensation of the bending roll and the high-efficiency edge drop control roll shape; among them, the whole edge drop control process generally has the problems of imperfect control mechanism and large lag of control response. At the same time, cold rolling is the downstream process of hot rolling, and the hot rolling crown and hot rolling wedge are transmitted to the cold rolling process through the base material, which has a certain influence on the edge drop control effect of cold rolling.

[0003] In the prior art, for the edge drop control problem of the cold rolling stand, scholars and metallurgical equipment manufacturers in the industry have proposed two types of edge drop control optimization strategies under the existing rolling mill configuration conditions: one is to optimize the intermediate roll roll shifting preset value, combined with the dynamic adjustment of the work roll bending force and the design of the high-efficiency edge drop control roll shape, to realize the rapid response and active compensation of the edge drop change of the strip steel, significantly improve the edge drop control effect in the cold rolling process, and cooperate with the feedback information of the outlet edge drop instrument to build a closed-loop control system, further improve the cross-sectional size accuracy and stability of the strip steel; the second is a strategy based on the thickness difference control of the horizontal direction, which adjusts the work roll bending force, the intermediate roll transverse displacement and the rolling force distribution of the stand to realize the optimization of the thickness uniformity of the strip steel in the horizontal direction, thereby indirectly improving the edge drop characteristics. However, the above methods mostly rely on the rolling mill structure with work roll transverse shifting function, and in actual cold continuous rolling production, most production lines use UCM (Utility Crown Mill) six-high rolling mills, which are limited by structural design and do not have work roll transverse adjustment capability, making it difficult for high-efficiency roll shape design and work roll bending control strategies to fully play their control effect. SUMMARY

[0004] In order to solve the above problems, the present application provides a cold continuous rolling multi-stand edge drop control method and system based on machine learning, which keeps the overall geometric size of the strip steel crown and wedge within the permitted fluctuation range, optimizes the dynamic adjustment strategy of each actuator, maximizes the overall control capability of the unit, fully mobilizes the independent control capability of each stand of the cold continuous rolling unit, realizes the edge drop linkage control strategy across the stands and processes through global collaborative optimization of rolling parameters, and improves the edge drop control effect.

[0005] To achieve the above object, the technical scheme adopted by the embodiments of the present application is as follows: In a first aspect, the embodiments of the present application provide a cold continuous rolling multi-stand edge drop control method based on machine learning, which comprises the following steps: Step S1, continuously collecting key parameters of historical production of cold continuous rolling stands and edge drop detection data of the strip steel to obtain an original data set; Step S2, introducing a unified timestamp mechanism in the original data set to associate the edge drop detection data with the key parameters in time; Step S3, preprocessing the original data set to obtain an edge drop control data set; Step S4, setting a plurality of sample points according to a predetermined rolling length interval, and extracting a data pair set corresponding to the sample points of each stand from the edge drop control data set, wherein the data pair comprises independent variables including initial strip steel state parameters, process parameters and rolling control parameters, and dependent variables including edge drop data; Step S5, constructing a single-stand edge drop prediction model based on an XGBoost model, and further constructing a stand recursive edge drop prediction model; Step S6, training, verifying and testing the constructed stand recursive edge drop prediction model using the data pair set of the first to fifth stands, and evaluating the model using a K-fold cross-validation mechanism to obtain a mature stand recursive edge drop prediction model; Step S7, dividing the to-be-rolled strip steel into a plurality of to-be-tested sections along the rolling direction; obtaining the initial strip steel state parameters, rolling control parameters and process parameters of the to-be-rolled strip steel at the first stand, and the rolling control parameters and process parameters of the second to fifth stands as actual production key parameters; Step S8, inputting the actual production key parameters of all to-be-tested sections into the stand recursive edge drop prediction model to output the predicted edge drop value of each to-be-tested section at the fifth stand; Step S9, setting a section edge drop threshold and an eligibility rate threshold; if the predicted section edge drop value is less than or equal to the section edge drop threshold, the number of eligible sections is incremented by 1 and the total number of sections is incremented by 1; if the current edge drop value is greater than the section edge drop threshold, the number of eligible sections is incremented by 0 and the total number of sections is incremented by 1; the eligibility rate is calculated according to the number of eligible sections and the total number of sections; if the eligibility rate is greater than the eligibility rate threshold, the process proceeds to step S11; otherwise, the process proceeds to step S10; Step S10, based on the predetermined constraint condition, using an improved grey wolf optimization algorithm to optimize the rolling control parameters of the first to third stands; updating the actual production key parameters using the optimized rolling control parameters, and proceeding to step S8; Step S11, taking the current actual production key parameters as the optimal parameters for production to achieve edge drop control of the current strip steel.

[0006] As a preferred embodiment of the present application, the key parameters of the historical production in step S1 include strip state parameters, rolling control parameters and process parameters; wherein the strip state parameters include strip thickness and strip width; the rolling control parameters include work roll bending force WRB, intermediate roll bending force IRB and intermediate roll transverse shift IRS; the process parameters include reduction ratio, rolling speed, pre-reduction tension and post-reduction tension.

[0007] As a preferred embodiment of the present application, the strip edge drop data in step S1 includes edge drop values of the left / right edge region δ eo 、 δ ed and center thickness δ c .

[0008] As a preferred embodiment of the present application, step S2 introduces a unified timestamp mechanism, including: Step S21, during the rolling control parameter acquisition process, simultaneously record the data acquisition time, and bind the time with the rolling control parameters; Step S22, correspond the timestamp in the edge drop detection data with the time data bound in the rolling control parameters, associate the rolling control parameters corresponding to the time with the edge drop detection data, and ensure the time sequence consistency of the data; Step S23, based on the collected edge drop detection data, trace back to match the strip state parameters and process parameters at the corresponding time point, and ensure the completeness and representativeness of the data.

[0009] As a preferred embodiment of the present application, the data pair in step S4 includes independent variables with initial strip state parameters, process parameters and rolling control parameters as contents, and dependent variables with edge drop data as contents.

[0010] As a preferred embodiment of the present application, when data extraction is performed, the initial strip state parameters, end strip state parameters, process parameters, rolling control parameters of the first stand and the corresponding edge drop data of each sample point are extracted, the end strip state parameters, process parameters, rolling control parameters of the second to fifth stands and the corresponding edge drop data are extracted, and the end strip state parameters of the previous stand are taken as the initial strip state parameters of the next stand; for each stand, the initial strip state parameters, process parameters and rolling control parameters are taken as independent variables, and the edge drop data is taken as dependent variable, to construct data pairs, and obtain the data pair set of the first to fifth stands.

[0011] As a preferred embodiment of the present application, the segment edge drop threshold aThe particle size is 5-7 μm, and the qualified rate threshold q% is 95-98%.

[0012] As a preferred embodiment of the present application, the constraints involved in the optimization of the rolling control parameters in step S10 include rolling process constraints, strip state constraints and process consistency constraints.

[0013] As a preferred embodiment of the present application, the rolling process constraints include:

[0014] The strip state constraints include: (a) crown constraint:

[0015] (b) thickness variation rate constraint:

[0016] (c) edge drop variation smoothness constraint:

[0017] (d) wedge constraint:

[0018] wherein, represents the work roll bending force of the i-th stand, represents the intermediate roll bending force of the i-th stand, represents the intermediate roll transverse displacement of the i-th stand; represents the strip crown value at the exit of the i-th stand, represents the upper limit of the crown allowed for the i-th sample section; represents the thickness variation rate of the i-th stand from the entry to the exit, represents the allowed range of the thickness variation rate; represents the edge drop value at the exit of the i-th stand, represents the edge drop value at the exit of the i-th stand, represents the maximum value of the edge drop variation allowed between two adjacent sections of measurement; represents the thickness variation amount of the left side relative to the center, represents the thickness variation amount of the right side relative to the center, represents the allowed threshold value of the absolute value of the left-right wedge difference; ​​​​​​​​The process consistency constraint includes: the strip thickness change rate and width change trend predicted by each rack need to be consistent with the metal material volume conservation relationship, to ensure the rationality in the recursive prediction process.

[0019] In a second aspect, the embodiments of the present application also provide a cold continuous rolling multi-rack edge drop control system based on a machine learning method, which comprises a data acquisition module, a timestamp mechanism introduction module, a data preprocessing module, a data pair set extraction module, a model construction and training module, an edge drop prediction module, a threshold comparison module, a rolling control parameter optimization module and a production parameter output module. The data acquisition module is configured to continuously acquire key parameters of historical production of the cold continuous rolling rack and edge drop detection data of the strip, to obtain an original data set; and is further configured to divide the to-be-rolled strip into a plurality of to-be-tested sections along the rolling direction; and to acquire initial strip state parameters, rolling control parameters and process parameters of the to-be-rolled strip at the first rack, and rolling control parameters and process parameters of the second to fifth racks, as actual production key parameters. The timestamp mechanism introduction module is configured to introduce a unified timestamp mechanism in the original data set, to associate the edge drop detection data with the key parameters in time. The data preprocessing module is configured to preprocess the original data set to obtain an edge drop control data set. The data pair set extraction module is configured to set a plurality of sample points according to a predetermined rolling length interval, and extract a data pair set of each rack corresponding to the sample points from the edge drop control data set, wherein the data pair comprises independent variables including the initial strip state parameters, the process parameters and the rolling control parameters, and dependent variables including the edge drop data. The model construction and training module is configured to construct a single-rack edge drop prediction model based on an XGBoost model, and further construct a rack recursive edge drop prediction model, and train, verify and test the constructed rack recursive edge drop prediction model using the data pair sets of the first to fifth racks, and evaluate using a K-fold cross-validation mechanism, to obtain a mature rack recursive edge drop prediction model. The edge drop prediction module is configured to input the actual production key parameters of all to-be-tested sections into the rack recursive edge drop prediction model, and output the predicted edge drop value of each to-be-tested section at the fifth rack. The threshold comparison module is configured to set a section edge drop threshold and a qualified rate threshold and make a judgment, if the predicted section edge drop value is less than or equal to the section edge drop threshold, then the qualified section +1, and the total section number +1; if the current edge drop value is greater than the section edge drop threshold, then the qualified section +0, and the total section number +1; calculate the qualified rate according to the number of qualified sections and the total section number, if the qualified rate is greater than the qualified rate threshold, then start the production parameter output module; otherwise, start the rolling control parameter optimization module. The rolling control parameter optimization module is configured to optimize the rolling control parameters of the first to third stands based on preset constraint conditions by using an improved grey wolf optimization algorithm; and the actual production key parameters are updated by using the optimized rolling control parameters. The production parameter output module is configured to take the current actual production key parameters as optimal parameters for production to achieve edge drop control of the current strip steel.

[0020] The scheme of the embodiment of the present application has the following beneficial effects: The cold continuous rolling multi-stand edge drop control method and system based on machine learning provided by the embodiment of the present application constructs a multi-stand recursive nonlinear prediction model of "rolling parameters-edge drop response" based on machine learning technology, and establishes a dynamic regulation and control mechanism with minimization of edge drop as the core optimization objective and convexity and wedge at the outlet of the last stand as process state constraint conditions, to realize accurate adjustment of edge drop characteristics in the cold continuous rolling process. The high-precision edge drop detection device is configured at the outlet of the last stand, and the time and data backtracking method are combined to realize dynamic tracking of rolling parameters and accurate identification of edge drop. Based on the XGBoost recursive prediction method, the edge drop is predicted from stand to stand, to solve the chain reaction and nonlinear mutation problems in the rolling process, and ensure high accuracy and high robustness of model prediction. The present application establishes a multi-dimensional mapping relationship between the control parameters such as work roll bending force (WRB), intermediate roll bending force (IRB) and intermediate roll transverse displacement (IRS) and edge drop characteristics by taking the XGBoost model as a proxy function, and realizes global optimization and dynamic correction of multi-stand control parameters by combining the black-box optimization algorithm based on the adaptive search strategy, to effectively solve the technical problems such as response lag, insufficient control accuracy and parameter coordination mismatch between stands in the traditional control method. By constructing an optimization evaluation model with minimization of edge drop as the core objective and combining the control mechanism based on recursive prediction and closed-loop feedback correction, the stability of the whole steel strip edge drop control is ensured. The present application takes the outlet convexity and wedge deviation as process state constraint conditions, and forms a set of whole-process edge drop accurate regulation and control strategy and optimization method suitable for the cold continuous rolling process, which significantly improves the responsiveness and accuracy of multi-stand parameter coordination regulation and control under complex working conditions, and promotes the high-precision and intelligent application development of edge drop control technology in cold continuous rolling intelligent manufacturing.

[0021] Of course, implementing any product or method of the present application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description only show some of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0023] Figure 1 The flow chart of the cold continuous rolling multi-stand edge reduction control method based on machine learning described in the embodiments of the present application. DETAILED DESCRIPTION

[0024] After discovering the above problems, the present inventors carefully studied the existing cold rolling stand edge reduction control problem. It was found that some scholars proposed a material incoming cross-section shape roll shifting compensation method based on big data. Although this method has certain data-driven characteristics, it uses an exhaustive recursive search optimization strategy, which has low calculation efficiency and is difficult to adapt to high-frequency switching industrial scenarios. At the same time, this method mainly relies on static historical data for grouping analysis, lacks prediction and adaptive adjustment capability based on real-time trends, and when facing steel variety diversification and production rhythm fluctuation, the control effect is easy to lose stability. Some scholars also provided a cold continuous rolling strip breaking control method. Although the control flow coverage is wide, the control strategy mainly relies on experience parameter setting, lacks adjustment mechanism based on dynamic modeling or data mining, and is difficult to cope with process fluctuation problems in complex disturbance environment. At the same time, each control unit operates independently, and a unified control target coordination mechanism is not established, the systematic linkage adjustment is insufficient, and it is difficult to realize the control strategy with optimal overall performance.

[0025] It should be noted that the defects in the above prior art solutions are the result of the inventors' practice and careful study, therefore, the discovery process of the above problems and the solutions proposed by the embodiments of the present application to solve the above problems should be the contribution of the inventors to the present application.

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0027] It should be noted that similar reference numerals and letters refer to like items throughout the accompanying drawings, and once an item is defined in one drawing, it is not necessary to further define and explain it in subsequent drawings. In the description of the present application, the terms "first", "second", "third", "fourth" and the like are used only to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0028] Based on the above in-depth analysis, the embodiment of the present application provides a cold continuous rolling multi-stand edge drop control method and system based on machine learning. By collecting key process parameters in different stands of the cold continuous rolling production line and fifth stand outlet edge drop detection data, a multi-stand dynamic prediction model containing parameters such as strip thickness (mm), strip width (mm), reduction rate (%), pre-reduction tension (MPa), and post-reduction tension (MPa) is established. The XGBoost recursive modeling method is used to accurately depict the nonlinear coupling relationship between the control parameters and the edge drop response. In view of the response lag, insufficient control accuracy, and parameter coordination mismatch problems existing in the traditional edge drop control, adaptive search optimization is carried out based on process parameter constraints to realize global optimization and dynamic correction of key control parameters such as multi-stand bending force and transverse displacement, and to improve the accuracy and process adaptability of edge drop control.

[0029] As shown in Figure 1 , the cold continuous rolling multi-stand edge drop control method based on machine learning specifically includes the following steps: Step S1, continuously collecting key parameters of historical production of cold continuous rolling stands and edge drop detection data of the strip to obtain an original data set.

[0030] In this step, the cold continuous rolling stand is a five-stand, and when data collection is performed, a high-precision edge drop detection device is set at the outlet of the fifth stand to collect strip edge drop data in real time, and key parameters of the first to fifth stands in the historical production of the cold continuous rolling stand are collected synchronously, including strip state parameters, rolling control parameters, and process parameters. Among them, the strip edge drop data includes the edge drop values of the left / right edge region δ eo 、 δ ed and the center thickness δ c ; the strip state parameters include strip thickness (mm) and strip width (mm); the rolling control parameters include work roll bending force (WRB), intermediate roll bending force (IRB), and intermediate roll transverse displacement (IRS); and the process parameters include reduction rate (%), rolling speed (m / min), pre-reduction tension (MPa), and post-reduction tension (MPa) and other parameters to comprehensively reflect the stress and deformation state of the strip in each stand.

[0031] Preferably, the edge drop detection device has an automatic zero drift compensation function, and combines with a periodic displacement sensor calibration mechanism to ensure the long-term stability and measurement accuracy of the edge drop data.

[0032] In the actual production data collection, this step can also include: establishing a periodic calibration system for the detection device related to the roll bending and roll shifting control system to ensure the measurement accuracy of the displacement sensor of each actuator, preventing the edge drop detection accuracy from being affected by drift or error accumulation; for the zero drift problem that may occur in the detection device, an automatic compensation mechanism is implemented, and combined with periodic zero correction operation, the long-term stability of the edge drop data is improved.

[0033] At the same time, in this step, the data is classified and stored according to the rack number, steel grade and process parameter dimensions, ensuring that the data has good structure and traceability, providing ordered data support for subsequent modeling and feature analysis.

[0034] Step S2, introduce a unified timestamp mechanism in the original data set, associate the edge drop detection data with the key parameters in time, and ensure the time sequence consistency and integrity of the data.

[0035] This step specifically includes: Step S21, record the data collection time during the rolling control parameter collection process, and bind the time with the rolling control parameters; Step S22, correspond the timestamp in the edge drop detection data with the time data bound in the rolling control parameters, associate the rolling control parameters corresponding to the time with the edge drop detection data, and ensure the time sequence consistency of the data; Step S23, based on the collected edge drop detection data, trace back the strip state parameters and process parameters at the corresponding time point to ensure the integrity and representativeness of the data.

[0036] Step S3, pre-process the original data set to obtain the edge drop control data set.

[0037] In this step, the pre-processing includes: smoothing the original detection data using a moving average filtering algorithm; eliminating abnormal values outside the process permission range, etc. Specifically, it includes:

[0038] Step S31, use a moving average filtering algorithm to smooth the original detection data, eliminate sudden points and measurement singular points, and improve the data stability and credibility.

[0039] Step S32, set data validity check rules to automatically eliminate abnormal data points outside the process permission range, avoiding extreme value interference in subsequent modeling and optimization calculation. By setting a process threshold, abnormal data points outside the permitted range are automatically eliminated.

[0040] Step S4, set a plurality of sample points at predetermined rolling length intervals, and extract a set of data pairs corresponding to each stand sample point from the edge drop control data set, wherein the data pair includes independent variables containing initial strip state parameters, process parameters, rolling control parameters, and dependent variables containing edge drop data; the "initial" is relative to the production process of the stand.

[0041] In this step, when extracting the set of data pairs for each stand, the initial strip state parameters, end strip state parameters, process parameters, rolling control parameters of the first stand and the corresponding edge drop data of each sample point are extracted, the end strip state parameters, process parameters, rolling control parameters of the second to fifth stands and the corresponding edge drop data are extracted, and the end strip state parameters of the previous stand are taken as the initial strip state parameters of the next stand; for each stand, the initial strip state parameters, process parameters, rolling control parameters are taken as independent variables, and the edge drop data is taken as dependent variables to construct data pairs, and the set of data pairs of the first to fifth stands is obtained.

[0042] Among them, the rolling control parameters include WRB, IRB, IRS, the initial strip state parameters of each stand include strip thickness and strip width, and the process parameters include reduction rate, rolling speed, pre-reduction tension and post-reduction tension. In fact, the fourth and fifth stands cannot directly adjust the rolling control parameters, and in the set of data pairs of the fourth and fifth stands, the rolling control parameters are constant; the rolling control parameters of the first to third stands can be adjusted as the rolling progress changes.

[0043] Step S5, based on the XGBoost model, a single-stand edge drop prediction model is constructed, and a stand recursive edge drop prediction model is further constructed; the single-stand edge drop prediction model takes the corresponding set of data pairs of each stand as training data, and the input includes the initial strip state parameters, process parameters, rolling control parameters of the current stand corresponding to the sample point, and the output is the edge drop at the exit of the stand.

[0044] In this step, due to the strong nonlinearity and local mutation characteristics between the edge drop and the control parameters, the traditional linear model is difficult to effectively fit the complex process response relationship, therefore, the XGBoost is used to construct a recursive prediction model for each stand. The single-stand edge drop prediction model is applicable to all stands; for each stand, there are the same input and output parameters, but different values, for example, the rolling control parameters of the first to third stands are variables, while the rolling control parameters of the fourth to fifth stands are constants. The model of all stands is combined to construct a recursive edge drop prediction model for the stands; here, the recursion refers to the strip state parameters after rolling in the previous stand, which are the initial strip state parameters of the next stand and part of the input values of the next stand, thereby forming the transmission of part of the parameters between the stands and generating the recursive information.

[0045] Preferably, the single-stand edge drop prediction model uses the root mean square error (RMSE), mean absolute error (MAE), mean relative error (MAPE), mean square error (MSE), and determination coefficient (R²) as loss functions for convergence optimization of the model. The single-stand edge drop prediction model sets the importance coefficients of the three rolling control parameters WRB, IRB, and IRS to the edge drop, and continuously optimizes the sensitivity of the rolling control parameters of each stand to the edge drop change in the model training to obtain the optimal importance coefficient combination, thereby better meeting the actual production.

[0046] Step S6, the data of the first to fifth stands are used to train, verify, and test the constructed recursive edge drop prediction model for the stands, and the K-fold cross-validation mechanism is used to evaluate the generalization ability and stability of the recursive edge drop prediction model for the stands. After training, verification, testing, and evaluation, a mature recursive edge drop prediction model for the stands is obtained.

[0047] Step S7, the to-be-rolled strip is divided into a plurality of to-be-tested segments along the rolling direction; the initial strip state parameters, rolling control parameters, and process parameters of the to-be-rolled strip at the first stand, and the rolling control parameters and process parameters of the second to fifth stands are obtained as the actual production key parameters.

[0048] Step S8, the actual production key parameters of all to-be-tested segments are input into the recursive edge drop prediction model for the stands, and the predicted edge drop value of each to-be-tested segment at the fifth stand is output.

[0049] Step S9, the segment edge drop threshold and the qualified rate threshold are set; if the predicted segment edge drop value is less than or equal to the segment edge drop threshold, the qualified segment is incremented by 1, and the total number of segments is incremented by 1; if the current edge drop value is greater than the segment edge drop threshold, the qualified segment is incremented by 0, and the total number of segments is incremented by 1; the qualified rate is calculated according to the number of qualified segments and the total number of segments; if the qualified rate is greater than the qualified rate threshold, step S11 is entered; otherwise, step S10 is entered.

[0050] In this step, when the average edge drop of a certain segment ≤ μm, this segment is considered a qualified segment, and the segment edge drop threshold a is 5 - 7 μm. Count the number of all qualified segments in the entire coil and calculate the ratio with the total number of segments. When the proportion of qualified segments with edge drop in the entire coil of steel reaches % or more (that is, the number of qualified sample segments accounts for ≥ % of the total number of segments), it is determined that the overall edge drop control quality of this coil of steel is qualified. That is:

[0051]

[0052] In the formula, is the number of segments with qualified edge drop in the entire coil, is the total number of sample segments.

[0053] If the target is not achieved or the target is achieved but the process parameters exceed the constraint conditions, and the proportion < %, it is considered that there is a risk of edge drop deviation in this coil, and the model parameters need to be adaptively adjusted and iteratively optimized. Preferably, q takes a value of 95 - 98.

[0054] Step S10, based on the preset constraint conditions, use the improved grey wolf optimization algorithm (IGWO) to optimize the rolling control parameters of the first to the third stands; update the actual production key parameters with the optimized rolling control parameters, and transfer to step S8; In this step, the constraint conditions involved in optimizing the rolling control parameters include: rolling process constraints, strip state constraints, and process consistency constraints.

[0055] Among them, the rolling process constraint conditions include:

[0056] The strip state constraint conditions include: (a) Crown constraint:

[0057] (b) Thickness change rate constraint:

[0058] (c) Edge drop change smoothness constraint: ]>

[0059] (d) Wedge constraint:

[0060] Among them, represents the work roll bending force of the th stand, indicates the bending roll force of the intermediate roll of the indicates the transverse displacement of the intermediate roll of the indicates the strip crown value at the exit of the indicates the upper limit of the crown allowed for the indicates the thickness change rate of the indicates the allowed range of the thickness change rate; indicates the edge drop value at the exit of the indicates the edge drop value at the exit of the indicates the maximum value of the edge drop change allowed between two adjacent measurement sections; indicates the thickness change amount of the left edge relative to the center, indicates the thickness change amount of the right edge relative to the center, indicates the absolute value threshold of the allowed left-right wedge difference.

[0061] The process consistency constraint conditions include: The strip thickness change rate and width change trend predicted by each stand need to be consistent with the metal material volume conservation relationship, to ensure the rationality in the recursive prediction process; This step controls the exit crown fluctuation within the process permitted range and stabilizes the wedge deviation within the target tolerance interval by setting the above constraint conditions, forming a global optimization problem with edge drop optimization as the core and crown and wedge deviation as the constraint conditions, to realize the precise adjustment of the strip edge drop and the dynamic balance of the shape control under the premise of meeting the crown and wedge state constraint conditions.

[0062] In the optimization process, the IGWO algorithm dynamically adjusts the search direction and step factor of the whole body based on the above constraint conditions, to realize the adaptive correction of the rolling control parameters of the first to third stands.

[0063] In step S11, the current actual production key parameters are taken as the optimal parameters, and production is performed to realize the edge drop control of the current strip.

[0064] Based on the same idea, the embodiment of the present application also provides a cold continuous rolling multi-stand edge drop control system based on a machine learning method, which comprises a data acquisition module, a timestamp mechanism introduction module, a data preprocessing module, a data pair set extraction module, a model construction and training module, an edge drop prediction module, a threshold comparison module, a rolling control parameter optimization module and a production parameter output module.​​​​​​​ The data acquisition module is configured to continuously acquire key parameters of historical production of the cold rolling mill stand and edge drop detection data of the strip steel, to obtain an original data set; and is further configured to divide the to-be-rolled strip steel into a plurality of to-be-tested segments along a rolling direction; to acquire initial strip steel state parameters, rolling control parameters and process parameters of the to-be-rolled strip steel at the first mill stand, and rolling control parameters and process parameters of the second to fifth mill stands, as actual production key parameters; The timestamp mechanism introduction module is configured to introduce a unified timestamp mechanism in the original data set, to associate the edge drop detection data with the key parameters in time; The data preprocessing module is configured to preprocess the original data set to obtain an edge drop control data set; The data pair set extraction module is configured to set a plurality of sample points according to a predetermined rolling length interval, and to extract, from the edge drop control data set, a data pair set of each mill stand corresponding sample point, wherein the data pair includes independent variables with the initial strip steel state parameters, the process parameters and the rolling control parameters as contents, and dependent variables with the edge drop data as contents; The model construction and training module is configured to construct a single mill stand edge drop prediction model based on an XGBoost model, and to further construct a mill stand recursive edge drop prediction model, and to train, validate and test the constructed mill stand recursive edge drop prediction model using the data pair sets of the first to fifth mill stands, and to evaluate using a K-fold cross-validation mechanism, to obtain a mature mill stand recursive edge drop prediction model; The edge drop prediction module is configured to input actual production key parameters of all to-be-tested segments into the mill stand recursive edge drop prediction model, and to output predicted edge drop values of each to-be-tested segment at the fifth mill stand; The threshold value comparison module is configured to set a segment edge drop threshold value and an eligibility rate threshold value and to make a judgment, if the predicted segment edge drop value is less than or equal to the segment edge drop threshold value, then the eligible segment + 1, and the total segment number + 1; if the current edge drop value is greater than the segment edge drop threshold value, then the eligible segment + 0, and the total segment number + 1; to calculate an eligibility rate according to the number of eligible segments and the total segment number, if the eligibility rate is greater than the eligibility rate threshold value, then to start a production parameter output module; otherwise, to start a rolling control parameter optimization module; The rolling control parameter optimization module is configured to optimize the rolling control parameters of the first to third mill stands based on a preset constraint condition using an improved grey wolf optimization algorithm; and to update the actual production key parameters using the optimized rolling control parameters; The production parameter output module is configured to use the current actual production key parameters as optimal parameters for production, to achieve edge drop control of the current strip steel.

[0065] The modules in the embodiment are implemented by a processor, and a memory is appropriately increased when storage is required. The processor can be, but is not limited to, a microprocessor (MPU), a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, and the like. The memory can include a random access memory (RAM) and can also include a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0066] In the above embodiment, all or part of the implementation can be achieved by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the implementation can be achieved in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiment of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner.

[0067] In addition, it should be noted that the cold rolling multi-stand edge reduction control system based on the machine learning method and the cold rolling multi-stand edge reduction control method based on the machine learning method described in the embodiment are corresponding. The description and limitation of the method are also applicable to the system, and will not be repeated here. The cold rolling multi-stand edge reduction control method and system based on machine learning in the embodiment of the present application are applied to specific production practice. The following will be illustrated by two different cold rolling strip steels.

[0068] Embodiment 1: In this embodiment, the cold rolling multi-stand edge reduction control method based on machine learning is used to control the edge reduction of the cold rolling multi-stand under the condition that the incoming material has small convexity or stable initial shape.

[0069] The involved tandem cold rolling mill is five stands, wherein the work roll diameter is 390 mm, the length is 1700 mm, the intermediate roll diameter is 470 mm, the length is 2300 mm, the backup roll diameter is 1210 mm, and the length is 1730 mm; the initial thickness of the strip steel is 2.00 mm, and the target outlet thickness is 1.32 mm.

[0070] A high-precision edge drop detection device is installed on the outlet side of the fifth stand of the tandem cold rolling line. In the historical production process, the edge drop data are collected in real time, and the strip state parameters, rolling control parameters and process parameters of the first to fifth stands of the UCM rolling mill are collected synchronously. A unified time stamp mechanism is introduced in the collected raw data set, time sequence consistent registration is performed, and the integrity of the data is ensured; then, the mutation and abnormal values are eliminated through sliding average filtering, the data validity verification rules are set, and the abnormal data points beyond the process permission range are removed. The preprocessed data are stored according to the process dimensions such as stand number, steel grade, strip thickness, width, reduction rate, deformation resistance, etc., forming a structured data set. Among them, the intermediate roll bending force is 38 kN, the work roll bending force is 29 kN, and the unit rolling pressure is 9.1 kN / mm.

[0071] To verify the robustness of the model under various incoming material states and working conditions, simulation experiment parameters are set and compared with the measured data. Specifically, the first stand reduction rate is 36.12%, the front tension is 145.25 MPa, the rear tension is 57.42 MPa, the work roll bending force (WRB) is 140 kN, the intermediate roll bending force (IRB) is 95.10 kN, and the intermediate roll displacement (IRS) is 20 mm; the second stand reduction rate is 38.98%, the front tension is 142.94 MPa, the rear tension is 138.23 MPa, the WRB is 171.23 kN, the IRB is 71.36 kN, and the IRS is 20 mm; the third stand reduction rate is 34.85%, the front tension is 137.18 MPa, the rear tension is 149.67 MPa, the WRB is 165.80 kN, the IRB is 78.45 kN, and the IRS is 18 mm. After the strip steel rolling is completed, the sample strips are cut at the entrances and exits of the first to third stands, and the average thickness and edge drop changes are measured to reduce the interference of accidental errors on the verification results, and the adaptability and prediction accuracy of the model under the conditions of multiple steel grades and multiple specifications of strip steel are evaluated.

[0072] The initial strip state parameters, process parameters and rolling control parameters (WRB, IRB and IRS) are taken as inputs, and the edge drop detection values at the outlets of the first to fifth stands are taken as outputs to train the XGBoost-based recursive edge drop prediction model of the stands in sequence. After the training is completed, the actual production key parameters are input into the recursive edge drop prediction model of the stands, and an optimization algorithm is used to optimize the input values under the preset constraint conditions, and finally the edge drop value at the outlet of the fifth stand is predicted. In the triggering condition of the optimization algorithm, the segment edge drop threshold is set to 7 μm, and the qualified rate is 98%.

[0073] The model prediction results are compared with the actual measured data after rolling, and the results show that the model prediction trend is basically consistent with the actual edge drop change, and the consistency is high. Through the comparison and calculation of the strip samples, it is known that the relative error between the measured edge drop value and the predicted value in multiple samples is less than 3%; among them, the maximum deviation of the edge drop appears on the sample at the outlet of the second stand, the measured value is 8 μm, the predicted value is 5 μm, and the difference between the two is 3 μm. Overall, the prediction model can control the edge drop error within ±1 μm under most working conditions, the edge drop control effect is stable, and the XGBoost recursive prediction model of the stands has good adaptability and generalization performance in the matching of multiple stand parameter inputs and the modeling of nonlinear edge drop response relationship. Although multiple parameters such as the bending roll force and the roll displacement of each stand need to be coordinated in the modeling process, the optimization target is always focused on the accurate control of the edge drop characteristics, and the crown and wedge deviation are taken as process state constraint conditions to ensure that the product shape quality requirements are met and the optimization and regulation effect of the edge drop standard is realized. The model has high accuracy and practicality when applied to edge drop prediction in the UCM rolling mill.

[0074] In the optimization process of the input values, the improved grey wolf optimization algorithm is used, and the optimization constraint conditions are set according to the actual production conditions. The optimization results show that: the first stand has the most significant control effect on the edge drop of the 75-100 mm area from the edge; the second stand has the strongest deviation regulation and control ability on the edge drop of the 45-70 mm area from the edge; and the third stand focuses on the fine correction of the edge drop within 45 mm from the edge.

[0075] After cold rolling production is performed by using the embodiment, compared with the original production process, the predicted edge drop value is close to the actual production value, and the accuracy of edge drop prediction and regulation is effectively improved.

[0076] Embodiment 2 This embodiment is based on the complex working conditions of large incoming material crown or abnormal and uneven initial shape in a certain cold continuous rolling mill train, and uses the above-mentioned cold continuous rolling multi-stand edge drop control method based on machine learning to control the edge drop of the cold continuous rolling multi-stand, and verifies the robustness of the edge drop prediction and control model under extreme shape conditions.

[0077] The involved cold continuous rolling mill is five stands, the initial strip thickness is 2.30 mm, the target outlet thickness is 1.35 mm, the maximum entry crown is up to 54 μm, the strip width uneven fluctuation range is ±1.5 mm, there are significant edge wave and transverse thickness protrusion and other problems. The equipment parameters are as follows: work roll diameter 390 mm, length 1700 mm; intermediate roll diameter 470 mm, length 2300 mm; backup roll diameter 1210 mm, length 1730 mm. A high-precision edge drop detection device is installed on the outlet side of the fifth stand of the cold continuous rolling production line, which collects strip edge drop data in real time, synchronizes the strip state parameters, rolling control parameters and process parameters of the first to fifth stands. Among them, the initial rolling parameters are intermediate roll bending force 42 kN, work roll bending force 35 kN, unit rolling pressure 9.6 kN / mm, reduction 32%, rolling speed 120 m / min, front tension 100 MPa, and rear tension 60 MPa. A unified time stamp mechanism is introduced in the collected raw data set, which is time-sequentially consistent and ensures the integrity of the data. Considering the severe fluctuation of the incoming material shape, the raw data set is processed by a sliding average filtering algorithm to smooth the intensity and suppress abnormal peaks and measurement mutations. Strict data validity verification rules are introduced to eliminate outlier data that exceeds the process permission range, ensuring the stability and reasonableness of the data set. The preprocessed valid data is classified and stored according to the stand number, steel grade, strip thickness, width, reduction, deformation resistance and other process state parameters to form an ordered structured sample library.

[0078] A multi-stand simulation input configuration under high crown conditions is constructed to verify the nonlinear adaptability and prediction accuracy of the model. The first stand reduction is 39.40%, the front tension is 151.86 MPa, the rear tension is 68.12 MPa, the WRB is 160 kN, the IRB is 102.25 kN, and the IRS is 22 mm; the second stand reduction is 41.87%, the front tension is 146.97 MPa, the rear tension is 142.84 MPa, the WRB is 176.31 kN, the IRB is 77.20 kN, and the IRS is 20 mm; the third stand reduction is 36.15%, the front tension is 141.76 MPa, the rear tension is 155.91 MPa, the WRB is 169.92 kN, the IRB is 85.67 kN, and the IRS is 19 mm. After rolling, the inlet and outlet sample strips are cut according to the stands, and the transverse thickness and edge drop measured values are measured for model verification.

[0079] After the model training is completed, the actual production key parameters are input into the rack recursive edge drop model. Although the initial condition is complex, the model prediction trend is highly consistent with the measured edge drop change; through comparison of multiple sample bands, the relative error of edge drop is controlled within 4%; the maximum deviation appears at the first rack outlet position, the measured value is 12 μm, the model prediction value is 8 μm, and the difference is 4 μm. Under most working conditions, the edge drop prediction error is concentrated within ±2 μm, and the model shows strong non-linear mapping ability and fault tolerance. The actual results show that the first rack has the highest efficiency in controlling the edge drop deviation of 75-100 mm from the edge, the second rack has the highest efficiency in controlling the edge drop deviation of 45-70 mm from the edge, and the third rack is limited to controlling the edge drop deviation of 45 mm from the edge.

[0080] The optimized control parameters are introduced into the production line control system and verified by combining detection data. The edge drop value is stably controlled within 7 μm, the crown fluctuation is within the process requirement range; the geometric size is good; the qualified section ratio of the whole steel coil edge drop reaches 98.1%, which meets the high-quality edge drop control standard. The embodiment verifies that the method of the present application still has good edge drop prediction and regulation ability under the condition of high crown and complex initial appearance, and the cooperative mechanism of the model and the optimization algorithm shows excellent adaptability and engineering application value under extreme working conditions.

[0081] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles, and is not intended to limit the scope of the claimed application, but only represents the preferred embodiment of the present application. Those skilled in the art should understand that the scope of the application involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

Claims

1. A machine learning-based method for controlling the edge drop of multi-stand cold rolling mills, characterized in that, The method includes the following steps: Step S1: Continuously collect key parameters from historical production of the cold rolling mill stand and edge drop detection data of the strip to obtain the raw dataset; Step S2: Introduce a unified timestamp mechanism into the original dataset to correlate the edge descent detection data with key parameters over time. Step S3: Preprocess the original dataset to obtain the edge descent control dataset; Step S4: Set several sample points according to the predetermined rolling length interval, and extract the data pair set corresponding to each stand from the edge drop control dataset. The data pair includes independent variables containing initial strip state parameters, process parameters, and rolling control parameters, and dependent variables containing edge drop data. Step S5: Construct a single rack edge drop prediction model based on the XGBoost model, and further construct a rack recursive edge drop prediction model. Step S6: The rack recursive edge drop prediction model is trained, validated, and tested using the data sets of racks 1 to 5, and evaluated using the K-segment cross-validation mechanism to obtain a mature rack recursive edge drop prediction model. Step S7: Divide the strip steel to be rolled into several test sections along the rolling direction; obtain the initial strip steel state parameters, rolling control parameters and process parameters of the strip steel to be rolled in the first stand of each test section, as well as the rolling control parameters and process parameters of the second to fifth stands, as key parameters for actual production. Step S8: Input all actual production key parameters of the test sections into the frame recursive edge drop prediction model, and output the predicted edge drop value of each test section at the 5th frame. Step S9: Set the segment edge reduction threshold and the pass rate threshold; if the predicted segment edge reduction value is less than or equal to the segment edge reduction threshold, then the number of qualified segments is increased by 1, and the total number of segments is increased by 1; if the current edge reduction value is greater than the segment edge reduction threshold, then the number of qualified segments is increased by 0, and the total number of segments is increased by 1; calculate the pass rate based on the number of qualified segments and the total number of segments; if the pass rate is greater than the pass rate threshold, then proceed to step S11. Otherwise, proceed to step S10; Step S10: Based on preset constraints, the improved gray wolf optimization algorithm is used to optimize the rolling control parameters of the first to third stands; the optimized rolling control parameters are used to update the key parameters of actual production, and then proceed to step S8. Step S11: Use the current actual production key parameters as the optimal parameters to carry out production, so as to achieve edge drop control of the current strip steel.

2. The method according to claim 1, characterized in that, The key parameters of historical production in step S1 include strip condition parameters, rolling control parameters, and process parameters. Among them, strip condition parameters include strip thickness and strip width; rolling control parameters include work roll bending force WRB, intermediate roll bending force IRB, and intermediate roll lateral displacement IRS; and process parameters include reduction rate, rolling speed, pre-reduction tension, and post-reduction tension.

3. The method according to claim 1, characterized in that, The strip edge drop data in step S1 includes the edge drop values ​​for the left / right side regions. δ eo , δ ed and center thickness δ c .

4. The method according to claim 1, characterized in that, Step S2 introduces a unified timestamp mechanism, including: Step S21: During the rolling control parameter acquisition process, the data acquisition time is recorded simultaneously, and the time is linked to the rolling control parameters; Step S22: Match the timestamp in the edge drop detection data with the time data bound in the rolling control parameters, and associate the rolling control parameters corresponding to the time with the edge drop detection data to ensure the consistency of the data sequence. Step S23: Based on the collected edge drop detection data, backtrack and match the strip steel state parameters and process parameters at the corresponding time points to ensure the integrity and representativeness of the data.

5. The method according to claim 1, characterized in that, The data pairs in step S4 include independent variables consisting of initial strip state parameters, process parameters, and rolling control parameters, and dependent variables consisting of edge drop data.

6. The method according to claim 5, characterized in that, When extracting data, the initial strip state parameters, the final strip state parameters, process parameters, rolling control parameters and the corresponding edge drop data of the first stand of each sample point are extracted. The final strip state parameters, process parameters, rolling control parameters and the corresponding edge drop data of the second to fifth stands are extracted, and the final strip state parameters of the previous stand are used as the initial strip state parameters of the next stand. For each stand, data pairs are constructed using initial strip state parameters, process parameters, and rolling control parameters as independent variables, and edge drop data as dependent variables, to obtain the data pair set for stands 1 to 5.

7. The method according to claim 1, characterized in that, Segment edge threshold reduction a The thickness is 5~7μm; the pass rate threshold q% is 95~98%.

8. The method according to claim 1, characterized in that, The constraints involved in optimizing the rolling control parameters in step S10 include: rolling process constraints, strip condition constraints, and process consistency constraints.

9. The method according to claim 8, characterized in that, The rolling process constraints include: ； The strip condition constraints include: (a) Convexity constraint: ； (b) Thickness change rate constraint: ； (c) Smoothness constraint of edge descent change: ； (d) Wedge constraint: ； in, Indicates the first The bending force of the working rolls on each frame Indicates the first The bending force of the intermediate roll of each frame Indicates the first The transverse displacement of the intermediate rollers of each frame; Indicates the first Strip crown value at the exit of each frame. Indicates the first The upper limit of convexity allowed for each sample segment; Indicates the first Thickness change rate from rack inlet to outlet Indicates the allowable range of thickness variation rate; Indicates the first Edge drop at the rack exit Indicates the first -1 Edge drop value at the rack exit This indicates the maximum allowable change in slope between two adjacent measurement segments; This indicates the change in thickness of the left side relative to the center. This indicates the change in thickness of the right side relative to the center. This represents the threshold for the absolute value of the allowed left and right wedge differences; The process consistency constraints include: the rate of change of strip thickness and the trend of change of width of the predicted output of each rack must be consistent with the volume conservation relationship of the metal material to ensure the rationality of the recursive prediction process.

10. A cold continuous rolling mill multi-stand edge drop control system based on machine learning methods, characterized in that, The system includes: a data acquisition module, a timestamp mechanism introduction module, a data preprocessing module, a data pair set extraction module, a model building and training module, an edge drop prediction module, a threshold comparison module, a rolling control parameter optimization module, and a production parameter output module; The data acquisition module is used to continuously collect key parameters of historical production of cold rolling mill stands and edge drop detection data of strip steel to obtain raw datasets; it is also used to divide the strip steel to be rolled into several test sections along the rolling direction; and to obtain the initial strip steel state parameters, rolling control parameters and process parameters of the strip steel to be rolled in each test section at the first stand, as well as the rolling control parameters and process parameters of the second to fifth stands, as key parameters for actual production. The timestamp mechanism introduction module is used to introduce a unified timestamp mechanism into the original dataset, and to correlate the edge descent detection data with key parameters in time. The data preprocessing module is used to preprocess the original dataset to obtain the edge descent control dataset; The data pair set extraction module is used to set a number of sample points at predetermined rolling length intervals and extract the data pair set corresponding to each stand from the edge drop control dataset. The data pair includes independent variables containing initial strip state parameters, process parameters, and rolling control parameters, and dependent variables containing edge drop data. The model building and training module is used to build a single rack edge drop prediction model based on the XGBoost model, and further build a rack recursive edge drop prediction model. The constructed rack recursive edge drop prediction model is trained, validated and tested using the data set of racks 1 to 5, and evaluated using the K-segmented cross-validation mechanism to obtain a mature rack recursive edge drop prediction model. The edge drop prediction module is used to input the actual production key parameters of all test sections into the frame recursive edge drop prediction model and output the predicted edge drop value of each test section at the 5th frame. The threshold comparison module is used to set the segment edge reduction threshold and the pass rate threshold and make judgments. If the predicted segment edge reduction value is less than or equal to the segment edge reduction threshold, the number of qualified segments is increased by 1, and the total number of segments is increased by 1. If the current edge reduction value is greater than the segment edge reduction threshold, the number of qualified segments is increased by 0, and the total number of segments is increased by 1. The pass rate is calculated based on the number of qualified segments and the total number of segments. If the pass rate is greater than the pass rate threshold, the production parameter output module is activated. Otherwise, the rolling control parameter optimization module is activated. The rolling control parameter optimization module is used to optimize the rolling control parameters of the first to third stands based on preset constraints and using an improved gray wolf optimization algorithm; the optimized rolling control parameters are then used to update the key parameters of actual production. The production parameter output module is used to take the current actual production key parameters as the optimal parameters for production, so as to achieve edge drop control of the current strip steel.

Images