A data-driven based online control method for hot-rolled head buckling

Abstract

The application provides a kind of based on data-driven hot rolling buckling head online control method, belongs to the technical field of slab buckling head control.The method first acquires hot rolling slab buckling head related process parameters on site and carries out data preprocessing, and establishes slab buckling head regulation and control efficacy self-learning model;At the same time, a regression prediction model for slab buckling head based on data-driven is constructed;Then, the downstream pass slab buckling head prediction model is optimized, and the upstream and downstream pass slab buckling head control strategies are established respectively;Finally, the corresponding sled coefficient value is calculated by combining the slab buckling head control strategy and the regulation and control efficacy model and issued to the rough rolling basic automation control system to perform leveling control.The application can effectively reduce manual intervention and realize automatic control of hot rolling rough rolling slab buckling head.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of slab buckling head control, and particularly relates to a hot rolling buckling head online control method based on data driving. BACKGROUND

[0002] Slab buckling head is a common phenomenon in strip rolling process, and serious slab buckling head can easily cause "stacking steel" accidents, collide with the roller, and affect the rolling rhythm. However, there are many factors affecting the buckling head, and each factor is coupled with each other, the mechanism is complex, and there is a lack of mature mathematical model in the buckling head control. In the actual production process, the slab buckling head is manually regulated and controlled depending on the experience of the operator.

[0003] In the past few decades, with the introduction of power distribution control systems and new measurement methods, the recording and collection of industrial process data have become more common. At the same time, data mining and database technology provide important support for the development and application of database modeling technology in industrial processes. By viewing and analyzing the information contained in the data, different database models can be integrated into the industrial process and monitored.

[0004] As the mainstream direction of modern strip rolling field research, slab buckling head control has always been concerned. Due to many complex factors on site, it is difficult to model the precise control of slab buckling head. There are a large number of nonlinear influencing factors in the rolling process of the slab, and the existing buckling head mechanism control model cannot accurately describe its nonlinear characteristics. Therefore, a hot rolling buckling head online control method based on data driving is proposed. SUMMARY

[0005] The present application provides a hot rolling buckling head online control method based on data driving, which solves the problem that the manual control intervention of the slab buckling head is not timely under the condition of fast rolling rhythm and variable rolling steel grade in the hot rolling rough rolling production process, and the slab buckling head impacts the roller, and even causes the stacking steel accident. Through the built regression prediction model and optimization model of the slab buckling head based on data driving, and the proposed upstream and downstream pass slab buckling head control strategy, the sleigh coefficient adjustment value of each rolling pass is calculated in real time, and the sleigh coefficient adjustment value of the current pass is sent to the rough rolling basic automation control system, so as to realize the online control target of the hot rolling slab buckling head based on data driving.

[0006] To solve the above-mentioned application purposes, the technical solutions provided by the present application are as follows:

[0007] A hot rolling buckling head online control method based on data driving, comprising the following steps:

[0008] S1, collecting the related process parameters of the hot rolling rough rolling slab buckling head on site and performing data preprocessing;

[0009] S2, establishing a slab camber head regulation effect self-learning model;

[0010] S3, constructing a slab camber head regression prediction model based on data driving;

[0011] S4, optimizing the slab camber head regression prediction model of the downstream pass;

[0012] S5, respectively establishing the upstream and downstream pass slab camber head control strategies;

[0013] S6, combining the slab camber head control strategy and the regression prediction model to calculate the corresponding pass sledging coefficient value and issue it to the rough rolling basic automation control system, so as to realize the automatic control of the hot rolling rough rolling slab camber head.

[0014] The hot rolling rough rolling slab camber head related process parameters collected in the step S1 include: steel grade; heating furnace number; rolling pass number; slab camber height, unit mm; slab camber length, unit mm; slab inlet temperature, unit ℃; slab upper and lower surface temperature difference, unit ℃; snow camber coefficient set value, unit 1.

[0015] The data preprocessing in the step S1 is specifically as follows:

[0016] The collected hot rolling rough rolling slab camber head related process parameters are preprocessed by adopting the 3σ criterion, and the abnormal data are eliminated and normalized to the [-1, 1] interval.

[0017] The step S2 of establishing the slab camber head regulation effect self-learning model includes the following steps:

[0018] S21, establishing a regression model between the slab camber head amount and the process parameters:

[0019] Y = XB

[0020] B = X T V(U T XX T V) -1 U T Y

[0021] Wherein, X is a process parameter data matrix,

[0022] Y is a slab camber head amount data matrix,

[0023] U and V represent principal component matrices, and U and V are principal component matrices obtained by corresponding X and Y;

[0024] B is a coefficient matrix;

[0025] S22, solving the coefficient matrix B:

[0026]

[0027] wherein, X1 is a process parameter variation matrix,

[0028] B1 is a linear regression coefficient matrix of X1,

[0029] E is a regulation efficacy coefficient matrix:

[0030]

[0031] wherein, D x is a diagonal matrix composed of standard deviations of process parameter variation,

[0032] D y is a diagonal matrix composed of standard deviations of slab camber head value variation;

[0033] S23, constructing a slab camber head regulation efficacy self-learning model for each rolling pass according to real-time detection data on site:

[0034]

[0035]

[0036] wherein, η * is a self-learning coefficient update value, η' is a self-learning coefficient calculation value, η' = E * / E', E * , E' are slab camber head regulation efficacy values calculated according to detection data at adjacent detection moments on site, η is a self-learning coefficient of last learning, is a self-learning model smoothing coefficient,

[0037] The step S3 of constructing a slab camber head regression prediction model based on data driving includes two prediction models, i.e., a prediction model 1 and a prediction model 2, and the prediction model construction steps are as follows:

[0038] S31, determining model input and output parameters:

[0039] S311, the input parameters of the prediction model 1 are as follows: same furnace steel grade of previous slab; heating furnace number; rolling pass number; slab camber height of previous slab, unit: mm; slab camber length of previous slab, unit: mm; inlet temperature of previous slab, unit: ℃; temperature difference between upper and lower surfaces of previous slab, unit: ℃; snow camber coefficient set value of previous slab, unit: 1;

[0040] The output parameters of the prediction model 1 are as follows: slab camber height value of current slab in each pass, unit: mm;

[0041] S312, the input parameters of the prediction model 2 are: the steel grade of the current slab; the heating furnace number; the rolling pass number; the slab warping height of the adjacent previous pass of the current slab, in units of mm; the slab warping length of the adjacent previous pass of the current slab, in units of mm; the temperature difference between the current slab and the previous slab at the same pass, in units of ℃; the temperature difference between the current slab and the previous slab at the same pass, in units of ℃; the snow warping coefficient setting value of the adjacent previous pass of the current slab, in units of 1;

[0042] The output parameter of the prediction model 2 is: the slab warping height change of the current slab and the previous slab at the same pass, in units of mm;

[0043] S32, given the input and output data set S = {(x i ,y i )|i=1,2,…,m}, the regression prediction model objective function obj is represented as:

[0044]

[0045] Wherein, x i ∈R m is the m-dimensional input parameter,

[0046] y i ∈R is the output parameter,

[0047] k is the number of trees,

[0048] f k (xi) is the prediction value of the kth tree,

[0049] is the loss function,

[0050] is the regular term, ω j is the weight of the leaf node j, is the L2 norm square of ω j , rT r is the number of leaf nodes;

[0051] S33, the objective function can be obtained:

[0052]

[0053] Wherein, a i , b i are the first and second derivatives of the loss function, respectively;

[0054] S34, let I j be the sample set of the jth leaf node, and the objective function is described as:

[0055]

[0056] S35, the first derivative of the objective function described in step S34 is calculated and set to 0, and the weight of the leaf node j is obtained as:

[0057]

[0058] wherein A j and B j represent constants, and λ represents a penalty coefficient;

[0059] S36, the first derivative and the second derivative of each sample corresponding to each leaf node are calculated, and then all samples corresponding to each node are summed to obtain A j and B j , and all nodes of the tree are traversed to obtain the objective function.

[0060] The downstream pass plate blank camber head prediction model in step S4 is optimized, and the specific steps are as follows:

[0061] S41, a downstream pass plate blank camber head regression prediction optimization model 3 is established:

[0062] Pre_H t =z*Bar_H t +q*Pass_H t

[0063] wherein t is the rolling pass number, t = 3, 4, …, N,

[0064] z and q are coefficients of the optimization model, 0 ≤ z ≤ 1, 0 ≤ q ≤ 1,

[0065] Bar_H t is the prediction value of the first pass in the prediction model 1 in the data-driven plate blank camber head regression prediction model, with the unit of mm,

[0066] Pass_H t is the prediction value of the first pass in the prediction model 2 in the data-driven plate blank camber head regression prediction model, with the unit of mm,

[0067] Pre_H t is the downstream pass plate blank camber head prediction value, with the unit of mm;

[0068] S42, the following function is constructed to obtain the optimal solution of parameters z and q:

[0069]

[0070] wherein Ture_H t is the actual value of the downstream pass plate blank camber head, i.e. the collected plate blank camber height, with the unit of mm.

[0071] The step S5 of establishing the upper and lower stream sub-slab camber head control strategy respectively includes the following steps:

[0072] S51, establish the upper stream sub-slab camber head control strategy, and the calculation formula of the ski coefficient is:

[0073]

[0074] Wherein, t is the rolling pass number, t = 1, 2,

[0075] The current block steel t-pass ski coefficient set value is 1,

[0076] The same furnace t-pass ski coefficient set value of the previous block steel is mm,

[0077] The current block steel t-pass ski coefficient adjustment amount in the upper stream control strategy is 1;

[0078] S52, calculate the t-pass ski coefficient adjustment amount

[0079]

[0080] Wherein, Bar_H t The t-pass mill outlet slab camber head prediction value in the prediction model 1 of the slab camber head regression prediction model based on data driving is mm,

[0081] And The upper limit and the lower limit of the t-pass slab camber value threshold interval n of the current block steel are mm,

[0082] The upper stream t-pass control efficacy coefficient is

[0083] N is the slab camber head control threshold interval;

[0084] S53, establish the lower stream control strategy, and the calculation formula of the ski coefficient is:

[0085]

[0086] Wherein, t is the rolling pass number, t = 3, 4, …, N,

[0087] The current block steel t-pass ski coefficient set value is 1,

[0088] This is the set value for the t-th pass of the previous batch of lumpy steel in the same heat, in units of 1.

[0089] ΔQ t This represents the adjustment amount of the t-th pass sled coefficient for the current block steel, in units of 1;

[0090] S54. Calculate the downstream t-th ski coefficient adjustment ΔQ t :

[0091]

[0092] Among them, Pre_H t This is the predicted value of the t-th pass of the current block steel, in mm.

[0093] The target range for adjusting the tilt and overhang of the slab in the tth pass of the same heat batch, in mm. Let t be the downstream regulation efficiency coefficient.

[0094] Specifically, step S6 involves:

[0095] By using the data-driven slab tilting regression prediction model established in S3 and the optimized model obtained in S4, combined with the upstream and downstream slab tilting control strategy proposed in S5, the sled coefficient adjustment value of each rolling pass is calculated in real time, and the sled coefficient adjustment value of this pass is sent to the basic automated control system of the roughing mill to realize the automatic control of hot-rolled slab tilting.

[0096] As mentioned above, prediction model 1 predicts the target value of all tracks; prediction model 2 only predicts the target value of downstream tracks; prediction model 3 is obtained by optimizing the predicted values ​​of downstream tracks in prediction model 1 and prediction model 2; prediction model 1 and prediction model 2 are only different in their input and output parameters, and the modeling process is carried out according to steps S32-S36.

[0097] The above technical solution has at least the following advantages compared with the existing technology:

[0098] 1. Improve slab quality: Based on the data-driven online control method for hot-rolled slab tilting head, combined with the slab tilting head control strategy, the sled coefficient adjustment value for each rolling pass is automatically given, which can more accurately predict the slab tilting head and perform online control.

[0099] 2. Reduce labor intensity: Through data-driven methods, the sled coefficient leveling value is automatically provided, reducing the occurrence of slab warping and operator intervention, thereby reducing labor intensity and achieving automatic control of hot-rolled rough slab warping. Attached Figure Description

[0100] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort based on these drawings.

[0101] Figure 1 A flow chart of a data-driven hot rolling buckling head online control method of the present application. DETAILED DESCRIPTION

[0102] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of protection of the present application.

[0103] Unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meanings understood by those skilled in the art in the field of the present application.

[0104] The present application provides a data-driven hot rolling buckling head online control method.

[0105] As shown in the figure, the method comprises the following steps: Figure 1

[0106] S1, collecting hot rolling rough rolling slab buckling head related process parameters on site and performing data preprocessing;

[0107] S2, establishing a slab buckling head control efficacy self-learning model;

[0108] S3, constructing a slab buckling head regression prediction model based on data driving;

[0109] S4, optimizing the slab buckling head regression prediction model of the downstream pass;

[0110] S5, establishing an upper and lower stream slab buckling head control strategy respectively;

[0111] S6, combining the slab buckling head control strategy and the regression prediction model to calculate the corresponding pass sled coefficient value and issue it to the rough rolling basic automation control system, to realize hot rolling rough rolling slab buckling head automatic control.

[0112] ​The process parameters related to the buckling head of the hot-rolled rough-rolled slab collected in the step S1 include: steel grade; heating furnace number; rolling pass number; slab buckling height, unit: mm; slab buckling length, unit: mm; slab inlet temperature, unit: ℃; slab upper and lower surface temperature difference, unit: ℃; snow buckling coefficient set value, unit: 1.

[0113] The data preprocessing in the step S1 is specifically:

[0114] The collected process parameters related to the buckling head of the hot-rolled rough-rolled slab are preprocessed by using the 3σ criterion, and abnormal data are eliminated and normalized to the interval [-1, 1].

[0115] The step S2 of establishing the slab buckling head control effect self-learning model includes the following steps:

[0116] S21, a regression model between the slab buckling head amount and the process parameters is established:

[0117] Y = XB

[0118] B = X T V(U T XX T V) -1 U T Y

[0119] Wherein, X is a process parameter data matrix,

[0120] Y is a slab bucking head amount data matrix,

[0121] U and V represent principal component matrices, U and V are principal component matrices obtained by corresponding X and Y respectively;

[0122] B is a coefficient matrix;

[0123] S22, solving the coefficient matrix B:

[0124]

[0125] Wherein, X1 is a process parameter change amount matrix,

[0126] B1 is a linear regression coefficient matrix of X1,

[0127] E is a control effect coefficient matrix:

[0128]

[0129] Wherein, D x is a diagonal matrix composed of process parameter change amount standard deviations,

[0130] D y is a diagonal matrix composed of slab buckling head value change amount standard deviations;

[0131] S23, constructing a slab camber control efficacy self-learning model for each rolling pass according to real-time detection data on site:

[0132]

[0133]

[0134] wherein η * is a self-learning coefficient update value, η' is a self-learning coefficient calculation value, η' = E * / E', E * and E' are slab camber control efficacy values calculated according to detection data at adjacent detection times on site, η is a self-learning coefficient learned last time, is a self-learning model smoothing coefficient,

[0135] The slab camber regression prediction model based on data driving is constructed in step S3, and the specific model includes two prediction models, namely prediction model 1 and prediction model 2. The prediction model construction steps are as follows:

[0136] S31, determining model input and output parameters:

[0137] S311, the input parameters of prediction model 1 are: the same furnace steel grade before the block steel; the heating furnace number; the rolling pass number; the slab camber height of the same furnace block steel before, unit: mm; the slab camber length of the same furnace block steel before, unit: mm; the inlet temperature of the same furnace block steel before, unit: ℃; the temperature difference between the upper and lower surfaces of the same furnace block steel before, unit: ℃; the snow camber coefficient set value of the same furnace block steel before, unit: 1;

[0138] The output parameters of prediction model 1 are: the slab camber height value of the current block steel in each pass, unit: mm;

[0139] S312, the input parameters of prediction model 2 are: the current block steel grade; the heating furnace number; the rolling pass number; the slab camber height of the adjacent last pass of the current block steel odd / even pass, unit: mm; the slab camber length of the adjacent last pass of the current block steel odd / even pass, unit: mm; the inlet temperature difference of the same furnace current block steel and the previous block steel in the same pass, unit: ℃; the temperature difference between the upper and lower surfaces of the same furnace current block steel and the previous block steel in the same pass, unit: ℃; the snow camber coefficient set value of the adjacent last pass of the current block steel odd / even pass, unit: 1;

[0140] The output parameters of prediction model 2 are: the slab camber height change amount of the same furnace current block steel and the previous block steel in the same pass, unit: mm;

[0141] S32, given input and output data set S = {(x i , yi}, regression prediction model objective function obj is represented as:

[0142]

[0143] wherein x i ∈R m is an m-dimensional input parameter,

[0144] y i ∈R is an output parameter,

[0145] k is the number of trees,

[0146] f k (x i ) is the prediction value of the kth tree,

[0147] is a loss function,

[0148] is a regular term, ω j is the weight of the leaf node j, is the L2 norm square of ω j , and rT r is the number of leaf nodes;

[0149] S33, the objective function can be obtained:

[0150]

[0151] wherein a i and b i are the first and second derivatives of the loss function, respectively;

[0152] S34, let I j be the sample set of the jth leaf node, and the objective function is described as:

[0153]

[0154] S35, the first derivative of the objective function described in step S34 is obtained, and it is equal to 0, then the weight of the leaf node j is:

[0155]

[0156] wherein A j and B j represent constants, and λ represents a penalty coefficient;

[0157] S36, the first and second derivatives of each sample corresponding to each leaf node are obtained, and then the sum of all samples corresponding to each node is obtained to obtain A j and B jThen, the objective function can be obtained by traversing all nodes of the tree.

[0158] The step S4 of optimizing the prediction model of the buckle head of the slab in the downstream pass is specifically as follows:

[0159] S41, a regression prediction optimization model 3 of the buckle head of the slab in the downstream pass is established:

[0160] Pre_H t =z*Bar_H t +q*Pass_H t

[0161] Wherein, t is the pass number, t = 3, 4, …, N,

[0162] z, q are coefficients of the optimization model, 0≤z≤1, 0≤q≤1,

[0163] Bar_H t is the prediction value of the prediction model 1 in the slab buckle head regression prediction model based on data driving, unit: mm,

[0164] Pass_H t is the prediction value of the prediction model 2 in the slab buckle head regression prediction model based on data driving, unit: mm,

[0165] Pre_H t is the prediction value of the slab buckle head in the downstream pass, unit: mm;

[0166] S42, the following function is constructed to obtain the optimal solution of the parameters z and q:

[0167]

[0168] Wherein, Ture_H t is the actual value of the slab buckle head in the downstream pass, unit: mm.

[0169] The step S5 of establishing the control strategy of the slab buckle head in the upstream and downstream passes respectively includes the following steps:

[0170] S51, the control strategy of the slab buckle head in the upstream pass is established, and the calculation formula of the sledging coefficient is:

[0171]

[0172] Wherein, t is the pass number, t = 1, 2,

[0173] is the set value of the sledging coefficient of the current block steel in the t pass, unit: 1,

[0174] The sled coefficient of the tth pass of the same block steel is set to a value, unit: mm,

[0175] The sled coefficient adjustment amount of the tth pass of the current block steel in the upstream pass control strategy, unit: 1;

[0176] S52, calculate the tth pass sled coefficient adjustment amount

[0177]

[0178] Wherein, Bar_H t The tth pass mill outlet slab camber prediction value in the prediction model 1 of the slab camber prediction model based on data driving, unit: mm,

[0179] And The upper limit and lower limit of the n-th pass slab camber value threshold interval of the current block steel, unit: mm,

[0180] The upstream tth pass control efficacy coefficient,

[0181] N is the slab camber control threshold interval;

[0182] In this embodiment, the slab camber control interval is specifically divided into (-∞, -100), [-100, -50), [-50, 0), [0, 50), [50, 100), [100, 150), [150, +∞) 7 segments, so the camber control target interval value n = 3;

[0183] S53, establish the downstream pass control strategy, the sled coefficient calculation formula is:

[0184]

[0185] Wherein, t is the pass number, t = 3, 4, …, N,

[0186] The tth pass sled coefficient set value of the current block steel, unit: 1,

[0187] The tth pass sled coefficient set value of the same block steel, unit: 1,

[0188] ΔQ t The tth pass sled coefficient adjustment amount of the current block steel, unit: 1;

[0189] S54, calculate the tth pass sled coefficient adjustment amount ΔQ t :

[0190]

[0191] Pre_H t Pre_H is a prediction value of the first t pass of the current block steel, unit: mm,

[0192] Pre_H is a control target interval value of the first t pass of the same batch of slab, unit: mm, is the control efficacy coefficient of the downstream t pass.

[0193] The step S6 is specifically:

[0194] Through the regression prediction model of the slab camber based on data driving built in S3 and the optimization model obtained in S4, combined with the upstream and downstream pass slab camber control strategy proposed in S5, the real-time calculation of the sleigh coefficient adjustment value of each rolling pass is realized, and the sleigh coefficient adjustment value of the current pass is issued to the rough rolling basic automatic control system, so as to realize the automatic control of the hot rolling slab camber.

[0195] In this embodiment, compared with the prediction model 1, the hit rate of the optimized prediction model 3 in the downstream pass slab camber control effect error within ± 30mm is increased by 2.03%, reaching 96.7%.

[0196] After the control method is applied to the industrial field, the statistical results show that the proportion of the slab with the camber range of 0mm to 50mm is significantly increased, which is increased by 18.56% compared with the manual control, the serious camber phenomenon is significantly reduced, which is reduced by 11.12% compared with the manual control, and the proposed online control method of the hot rolling camber based on data driving can well meet the requirements of the field and effectively improve the slab quality.

[0197] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A data-driven based online control method for hot rolled crooked head, characterized in that, The steps include the following: S1, collecting hot-rolled rough-rolled slab buckling head related process parameters on site and performing data preprocessing; S2, establishing a slab buckling head control efficiency self-learning model; S3, constructing a data-driven slab buckling head regression prediction model; S4, optimizing the downstream pass slab buckling head regression prediction model; S5, respectively establishing the upstream and downstream pass slab buckling head control strategies; S6, combining the slab buckling head control strategy and the regression prediction model to calculate the corresponding pass sledging coefficient value and issue it to the rough rolling basic automation control system to realize automatic control of the hot-rolled rough-rolled slab buckling head. The step S3 of constructing a data-driven slab buckling head regression prediction model includes two prediction models, prediction model 1 and prediction model 2. The prediction model construction steps are as follows: S31, determine the model input and output parameters: S311, the input parameters of prediction model 1 are: the same furnace steel grade; the heating furnace number; the rolling pass number; the same furnace slab buckling height, unit mm; the same furnace slab buckling length, unit mm; the same furnace slab inlet temperature, unit ℃; the same furnace slab upper and lower surface temperature difference, unit ℃; the same furnace snow buckling coefficient set value, unit 1; The output parameters of prediction model 1 are: the current slab buckling height value of each pass, unit mm; S312, the input parameters of prediction model 2 are: the current slab steel grade; the heating furnace number; the rolling pass number; the current slab odd or even pass adjacent last pass slab buckling height, unit mm; the current slab odd or even pass adjacent last pass slab buckling length, unit mm; the same furnace current slab and the previous slab same pass inlet temperature difference, unit ℃; the same furnace current slab and the previous slab same pass slab upper and lower surface temperature difference, unit ℃; the current slab odd or even pass adjacent last pass snow buckling coefficient set value, unit 1; The output parameters of prediction model 2 are: the same furnace current slab and the previous slab same pass buckling height change, unit mm; S32、Given input-output data set S = {(x i ,y i )|i = 1,2,…,m}, the regression prediction model objective function obj is represented as: where x i ∈ R m is an m-dimensional input parameter, y i ∈R is an output parameter, k is the number of trees, f k (x i ) is the predicted value of the kth tree, is a loss function, is a regular term, ω j is a weight of a leaf node j, is ω j L2 norm squared of rT r is the number of leaf nodes; S33, the target function is solved: where a i , b i are the first and second derivatives of the loss function, respectively; S34, set I j Let Sj be the set of samples for the jth leaf node. The objective function is described as: S35, take the first derivative of the target function in step S34 and set it equal to 0, then the weight of the leaf node j is: where A j and B j denote constants, and l denotes a penalty coefficient; S36. Find the first and second derivatives of each sample for each leaf node, and then sum all the samples corresponding to each node to obtain A. j and B j Then, by traversing all nodes of the tree, the objective function can be obtained.

2. The data-driven based online control method of hot rolled cambered head as claimed in claim 1, wherein, The hot-rolled rough-rolled slab buckling head related process parameters collected in step S1 include: steel grade; heating furnace number; rolling pass number; slab buckling height, unit mm; slab buckling length, unit mm; slab inlet temperature, unit ℃; slab upper and lower surface temperature difference, unit ℃; snow buckling coefficient set value, unit 1.

3. The data driven based online control method of hot rolled cambered head as claimed in claim 1 wherein, The data preprocessing in step S1 is as follows: The collected hot-rolled rough-rolled slab buckling head related process parameters are preprocessed using the 3σ criterion to eliminate abnormal data and normalized to the [-1, 1] interval.

4. The data driven based online control method of hot rolled cambered head as claimed in claim 1 wherein, The step S2 of establishing a slab buckling head control efficiency self-learning model includes the following steps: S21, establish a regression model between the slab buckling head quantity and the process parameters: Y = XB B = X T V(U T XX T V) -1 U T Y Where, X is the process parameter data matrix, Y is the slab buckling head quantity data matrix, U, V represent principal component matrix, U and V are principal component matrix obtained by corresponding X and Y respectively; B is a coefficient matrix; S22, solving the coefficient matrix B: Wherein, X1 is the process parameter variation matrix, B1 is the linear regression coefficient matrix of X1, E is the control efficacy coefficient matrix: where D x is a diagonal matrix of standard deviations of process parameter variations, D y is a diagonal matrix composed of the standard deviation of the change amount of the slab camber at the head; S23, according to the real-time detection data of the field to build each rolling pass slab buckling head control efficacy self-learning model: wherein η * is a self-learning coefficient update value, η' is a self-learning coefficient calculated value, η' = E * / E', E * E' and E are respectively the slab camber head control efficacy values calculated according to the detection data of the adjacent detection time on site, η is the last learning self-learning coefficient, to smooth the coefficients of the self-learning model, 5. The data driven based online control method of hot rolled cambered head as claimed in claim 1 wherein, The step S4 in the downstream pass slab buckling head prediction model is optimized, and the specific steps are as follows: S41, establish downstream pass slab buckling head regression prediction optimization model 3: Pre_H t = z * Bar_H t + q * Pass_H t Wherein, t is the rolling pass number, t = 3, 4, …, N, z, q are the coefficients of the optimization model, 0 ≤ z ≤ 1, 0 ≤ q ≤ 1, Bar_H t For the prediction model 1 of the data-driven-based slab head camber regression prediction model, the prediction value of the t-th pass, unit: mm, Pass_H t Pass_H t Pass_H t Pass_H t Pass_H t Pass_H t Pass_H t Pass_H t Pass_H t Pass_H t Pass_H <000003 Pre_H t For the downstream pass slab buckling head prediction value, unit mm; S42, construct the following function to obtain the optimal solution of parameters z, q: Wherein, Ture_H t is the actual value of the downstream pass slab camber head, that is, the collected slab camber height, in mm.

6. The data driven based online control method of hot rolled cambered head as claimed in claim 1 wherein, The step S5 respectively establishes the upstream and downstream pass slab buckling head control strategy, including the following steps: S51, establish the upstream pass slab buckling head control strategy, the calculation formula of the sledging coefficient is: Wherein, t is the rolling pass number, t = 1, 2, Set the value of the tth pass sledging coefficient for the current block steel, in units of 1, Set the sledging coefficient value for the same furnace front block steel t pass, unit mm, is the current block steel tthh strip pass sleigh coefficient adjustment amount in the upstream pass control strategy, unit 1; S52, calculate the tth pass sled coefficient adjustment amount Bar_H t Bar_H is the prediction value of the slab head camber at the tth pass mill exit in the prediction model 1 of the data-driven slab head camber regression prediction model, in mm, and are respectively the upper limit and the lower limit of the threshold interval n of the current block steel t-pass slab camber value, unit: mm, to regulate the effectiveness coefficient for the upstream tth pass, n is the slab buckling head control threshold interval; S53, establish downstream pass control strategy, the calculation formula of the sledging coefficient is: Wherein, t is the rolling pass number, t = 3, 4, …, N, Set the value of the tth pass sledging coefficient for the current block steel, unit 1, Set value for sledging coefficient for the tth pass of the preceding block steel for the same heat, unit 1, ΔQ t is the current block steel t pass sled coefficient adjustment amount, unit 1; S54, calculate downstream tth pass sledging coefficient adjustment amount ΔQ t : Pre_H = Pre_H + (H - Pre_H) * (1 - e) where Pre_H t is the predicted value of the first pass buckling head of the current block steel, unit: mm, Target interval value for the first pass slab head camber of the same previous block steel, unit: mm, Regulate the efficacy coefficient for the downstream tth pass.

7. The data driven based online control method of hot rolled cambered head as claimed in claim 1 wherein, The step S6 is specifically: Through the slab buckling head regression prediction model based on data driving built in S3 and the optimization model obtained in S4, combined with the upstream and downstream pass slab buckling head control strategy proposed in S5, the sledging coefficient adjustment value of each rolling pass is calculated in real time, and the sledging coefficient adjustment value of this pass is issued to the rough rolling basic automation control system, to realize the automatic control of hot rolling slab buckling head.

Images