Method for improving prediction precision of hot rolling pressure model

Abstract

The invention relates to a method for improving prediction precision of a hot rolling pressure model, and belongs to the technical field of plate and strip rolling methods. According to the technical scheme, hot rolling parameters are collected; the deviation between the measured value and the predicted value of the rolling force is counted, a correction coefficient is obtained through calculation, a mathematical model, related to rolling parameters, of the correction coefficient is established, an undetermined coefficient is regressed, an adaptive learning model is established through an exponential smoothing method, and a proper gain coefficient is determined; and integrating the mathematical regression model and the self-adaptive learning model in a weight distribution mode, and further correcting the rolling force prediction value of the hot rolling pressure model. The method has the beneficial effects that the coping capacity of the hot rolling pressure model to the condition of deviation of the learning effect of the self-adaptive model is enhanced, the prediction precision of the rolling force is improved, the method is beneficial to improving the production quality of strip steel, economic benefits can be brought to enterprise production, and the method has good popularization value.

Description

Technical Field

[0001] The invention relates to a method for improving the prediction accuracy of a hot rolling pressure model, and belongs to the technical field of plate and strip rolling methods. Background Art

[0002] The traditional method of obtaining the rolling force correction coefficient is simply to add the results of the mathematical regression model and the adaptive learning model for rolling force correction, or to ignore the mathematical regression model and only establish an adaptive learning model through related methods such as exponential smoothing to obtain the rolling force correction coefficient. The rolling force correction coefficient calculated by the mathematical regression model represents the correlation between the main rolling parameters and the degree of deviation of the rolling force prediction, so that the rolling force correction coefficient can adapt to the impact of changes in rolling conditions. The correction coefficient obtained by the adaptive learning model can make the model consider more realistic factors on this basis, thereby further improving the accuracy, which is more of a trend correction.

[0003] In practical applications, traditional methods occasionally encounter situations where the learning effect of the adaptive model deviates suddenly from the actual target value. In this case, the independent adaptive learning model often adjusts too slowly, and the simple addition of the results of the mathematical regression model and the adaptive learning model is often subject to the degree of deviation of the adaptive learning, resulting in poor results. Summary of the invention

[0004] The purpose of the present invention is to provide a method for improving the prediction accuracy of a hot rolling pressure model. The method comprises the following steps: comparing the two rolling force correction values ​​obtained by the mathematical regression model for rolling force correction and the adaptive learning model with the rolling force correction values ​​actually required, judging the prediction accuracy of the two, and thereby making the weight more biased towards the side with higher accuracy. At the same time, the weight can be automatically adjusted based on preset rules according to the deviation between the predicted rolling force value and the measured rolling force value. The method strengthens the ability of the hot rolling pressure model to cope with the situation where the learning effect of the adaptive model deviates from this situation, improves the prediction accuracy of the rolling force, is beneficial to improving the production quality of the strip steel, can bring economic benefits to the enterprise production, has good promotion value, and effectively solves the above-mentioned problems existing in the background technology.

[0005] The technical solution of the present invention is: a method for improving the prediction accuracy of a hot rolling pressure model, comprising the following steps:

[0006] S1, collecting hot rolling parameters;

[0007] S2. Calculate the deviation between the measured value and the predicted value of the rolling force, obtain the correction coefficient by calculation, establish a mathematical model related to the rolling parameters and regress the unknown coefficient;

[0008] S3, establish an adaptive learning model through exponential smoothing method and determine the appropriate gain coefficient;

[0009] S4. The mathematical regression model and the adaptive learning model are integrated by weight distribution, so as to correct the rolling force prediction value of the hot rolling pressure model, and the weight can be automatically adjusted based on preset rules according to the deviation between the rolling force prediction value and the measured value.

[0010] In the step S1, the hot rolling parameters are collected to collect the hot rolling parameters of a certain steel type or steel family, including: the measured rolling force and the predicted rolling force without considering the correction coefficient, the thickness of the strip at the outlet and entrance of the rolling mill, the strip width, the rolling speed and the rolling temperature.

[0011] In step S2, the specific steps include:

[0012] S2.1 Statistically calculate the deviation between the measured rolling force value and the predicted value, and obtain the rolling force correction coefficient by calculation. The calculation method of the rolling force correction coefficient is:

[0013]

[0014] In the formula, i is the rack number; n is the strip number in the rolling sequence, and f i,n is the rolling force correction coefficient; is the measured rolling force, kN; is the predicted rolling force without considering the correction factor, kN;

[0015] S2.2 Establish a mathematical model related to the correction coefficient and rolling parameters, referred to as the correction coefficient prediction model, and regress the unknown coefficients.

[0016] The correction coefficient prediction model is as follows:

[0017] f i,n =a0+a1·ε i,n +a2·v i,n +a3·T i,n +a4·B i,n +a5·ln(v i,n )+a6·ln(ε i,n ·v i,n )+

[0018] a7·ε i,n ·T i,n +a8·ε i,n ·v i,n ·T i,n

[0019] Where i is the rack number; n is the strip number in rolling order; a0, a1, a2, a3, a4, a5, a6, a7, a8 are the regression coefficients; ε i,n is the logarithmic strain of the strip in the i-stand; T i,nis the rolling temperature of the strip in stand i, ℃; v i,n is the rolling speed of the strip in stand i, m / s;

[0020] in

[0021] Where, h0 is the inlet thickness of the strip at the i-frame, mm; h1 is the outlet thickness of the strip at the i-frame, mm;

[0022] Based on the least squares regression method, the rolling force correction factor f is calculated by using the strip exit and entry thickness, strip width, rolling speed and rolling temperature data at the rolling mill. i,n Perform regression on the undetermined coefficients of the modified coefficient prediction model.

[0023] In step S3, the specific steps are as follows:

[0024] S3.1, sort the measured rolling forces of the same stand of the same steel grade or steel family collected on site and the predicted rolling forces without considering the correction coefficient according to the rolling time;

[0025] S3.2, bring the rolling force data sorted out in S3.1 into the exponential smoothing formula to establish a rolling force adaptive learning model,

[0026] The exponential smoothing formula is:

[0027]

[0028] In the formula, is the predicted value of the rolling force correction coefficient at the nth setting or control based on adaptive learning; is the measured value of the rolling force correction coefficient after the nth setting or control based on adaptive learning; is the predicted value of the rolling force correction coefficient set or controlled for the n+1th time based on adaptive learning; α i is the gain coefficient, 0≤α≤1;

[0029] S3.3, determine the appropriate gain factor α i , so that the deviation between the measured rolling force and the predicted rolling force meets a certain target value.

[0030] 5. The method for improving the prediction accuracy of the hot rolling pressure model according to claim 1, characterized in that: in the step S4, the specific steps are as follows:

[0031] S4.1 integrates the rolling force correction coefficient obtained by the mathematical regression model and the adaptive learning model by weight distribution:

[0032] The weight distribution method is:

[0033] m i，1 =k i，1 ·f i，1

[0034]

[0035] k i，n+1 =k i，n +s i，n，j , n=1,2,3,4…

[0036]

[0037] Where i is the rack number; n is the strip number in rolling order; m i，n It is the rolling force correction coefficient obtained by integrating the mathematical regression model and the adaptive learning model, which is referred to as the integrated rolling force correction coefficient in the following. i .n is the weight coefficient, 0≤k i，n ≤1, k i，1 =1;s i,n，j is the adjustment item of weight; j is the group number of different weight adjustment amounts;

[0038] S4.2 sets the weight adjustment rule and automatically adjusts the weight according to the deviation between the predicted value and the measured value of the rolling force.

[0039] Calculate the weight coefficient when the predicted rolling force value is theoretically equal to the measured value:

[0040]

[0041] In the formula, i is the rack number; n is the strip number in the rolling sequence; is the weight coefficient when there is no deviation between the predicted value and the measured value of rolling force, hereinafter referred to as the no-bias weight coefficient;

[0042] According to the unbiased weight coefficient With weight coefficient k i,n The weight adjustment amount is set for each deviation group.

[0043] The weight adjustment amount is set for each group. In group 0, the deviation amount The weight adjustment amount is s i,n,0 ; When the group is 1 to j-1, the deviation The weight adjustment amount is s i,n,j-1 ; In the jth group, the deviation The weight adjustment amount is s i,n,j ; where d j is the boundary value of the group;

[0044] Special case: If k i,n +si,n,j >1, then s i,n,j =1-k i,n ; if k i,n +s i,n,j <0, then s i,n,j =-k i,n .

[0045] The beneficial effects of the present invention are as follows: by comparing the two rolling force correction values ​​obtained by the mathematical regression model used for rolling force correction and the adaptive learning model with the actually required rolling force correction value, the prediction accuracy of the two is judged, and the weight is more biased towards the side with higher accuracy. At the same time, the weight can be automatically adjusted based on preset rules according to the deviation between the predicted value and the measured value of the rolling force; the ability of the hot rolling pressure model to cope with the situation where the learning effect of the adaptive model deviates from this situation is enhanced, the prediction accuracy of the rolling force is improved, which is beneficial to improving the production quality of strip steel, can bring economic benefits to the production of enterprises, and has good promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is the overall flow chart of the present invention;

[0047] Figure 2 This is a rolling force prediction deviation rate diagram before the adaptive model is applied in the embodiment of the present invention;

[0048] Figure 3 A rolling force prediction deviation rate diagram after the adaptive model is applied in the embodiment of the present invention;

[0049] Figure 4 The rolling force prediction deviation rate diagram of the embodiment of the present invention is integrated with the mathematical regression model and the adaptive learning model in a weight allocation manner. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the invention implementation cases clearer, the technical solutions in the implementation cases of the present invention will be clearly and completely described below in conjunction with the drawings in the implementation cases. Obviously, the implementation cases described are only a small part of the implementation cases of the present invention, rather than all the implementation cases. Based on the implementation cases in the present invention, all other implementation cases obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] A method for improving the prediction accuracy of a hot rolling pressure model comprises the following steps:

[0052] S1, collecting hot rolling parameters;

[0053] S2. Calculate the deviation between the actual rolling force value and the predicted value, obtain the correction coefficient by calculation, establish a mathematical model related to the rolling parameters and regress the undetermined coefficient;

[0054] S3, establish an adaptive learning model through exponential smoothing method and determine the appropriate gain coefficient;

[0055] S4. The mathematical regression model and the adaptive learning model are integrated by weight distribution, so as to correct the rolling force prediction value of the hot rolling pressure model, and the weight can be automatically adjusted based on preset rules according to the deviation between the rolling force prediction value and the measured value.

[0056] In the step S1, the hot rolling parameters are collected to collect the hot rolling parameters of a certain steel type or steel family, including: the measured rolling force and the predicted rolling force without considering the correction coefficient, the thickness of the strip at the outlet and entrance of the rolling mill, the strip width, the rolling speed and the rolling temperature.

[0057] In step S2, the specific steps include:

[0058] S2.1 Statistically calculate the deviation between the measured rolling force value and the predicted value, and obtain the rolling force correction coefficient by calculation. The calculation method of the rolling force correction coefficient is:

[0059]

[0060] In the formula, i is the rack number; n is the strip number in the rolling sequence, and f i,n is the rolling force correction coefficient; is the measured rolling force, kN; is the predicted rolling force without considering the correction factor, kN;

[0061] S2.2 Establish a mathematical model related to the correction coefficient and rolling parameters, referred to as the correction coefficient prediction model, and regress the unknown coefficients.

[0062] The correction coefficient prediction model is as follows:

[0063] f i,n =a0+a1·ε i,n +a2·v i,n +a3·T i,n +a4·B i,n +a5·ln(v i,n )+a6·ln(ε i,n ·v i,n )+

[0064] a7·ε i,n ·T i,n +a8·ε i,n ·v i,n ·T i,n

[0065] Where i is the rack number; n is the strip number in rolling order; a0, a1, a2, a3, a4, a5, a6, a7, a8 are the regression coefficients; ε i,n is the logarithmic strain of the strip in the i-stand; T i,n is the rolling temperature of the strip in stand i, ℃; v i,n is the rolling speed of the strip in stand i, m / s;

[0066] in

[0067] Where, h0 is the inlet thickness of the strip at the i-frame, mm; h1 is the outlet thickness of the strip at the i-frame, mm;

[0068] Based on the least squares regression method, the rolling force correction factor f is calculated by using the strip exit and entry thickness, strip width, rolling speed and rolling temperature data at the rolling mill. i,n Perform regression on the undetermined coefficients of the modified coefficient prediction model.

[0069] In step S3, the specific steps are as follows:

[0070] S3.1 Sort the measured rolling forces of the same stand of the same steel grade or steel family collected on site and the predicted rolling forces without considering the correction coefficient according to the rolling time;

[0071] S3.2 brings the rolling force data sorted out in S3.1 into the exponential smoothing formula to establish a rolling force adaptive learning model.

[0072] The exponential smoothing formula is:

[0073]

[0074] In the formula, is the predicted value of the rolling force correction coefficient at the nth setting or control based on adaptive learning; is the measured value of the rolling force correction coefficient after the nth setting or control based on adaptive learning; is the predicted value of the rolling force correction coefficient set or controlled for the n+1th time based on adaptive learning; α i is the gain coefficient, 0≤α≤1;

[0075] S3.3 Determine the appropriate gain factor α i , so that the deviation between the measured rolling force and the predicted rolling force meets a certain target value.

[0076] 5. The method for improving the prediction accuracy of the hot rolling pressure model according to claim 1, characterized in that: in the step S4, the specific steps are as follows:

[0077] S4.1 integrates the rolling force correction coefficient obtained by the mathematical regression model and the adaptive learning model by weight distribution:

[0078] The weight distribution method is:

[0079] m i，1 =k i，1 ·f i，1

[0080]

[0081] k i，n+1 =k i，n +s i，n，j , n=1,2,3,4…

[0082]

[0083] Where i is the rack number; n is the strip number in rolling order; m i，n It is the rolling force correction coefficient obtained by integrating the mathematical regression model and the adaptive learning model, which is referred to as the integrated rolling force correction coefficient in the following. i.n is the weight coefficient, 0≤k i，n ≤1, k i，1 =1;s i,n，j is the adjustment item of weight; j is the group number of different weight adjustment amounts;

[0084] S4.2 sets the weight adjustment rule and automatically adjusts the weight according to the deviation between the predicted value and the measured value of the rolling force.

[0085] Calculate the weight coefficient when the predicted rolling force value is theoretically equal to the measured value:

[0086]

[0087] In the formula, i is the rack number; n is the strip number in the rolling sequence; is the weight coefficient when there is no deviation between the predicted value and the measured value of rolling force, hereinafter referred to as the no-bias weight coefficient;

[0088] According to the unbiased weight coefficient With weight coefficient k i,n The weight adjustment amount is set for each deviation group.

[0089] The weight adjustment amount is set for each group. In group 0, the deviation amount The weight adjustment amount is s i,n,0 ; When the group is 1 to j-1, the deviation The weight adjustment amount is s i,n,j-1 ; In the jth group, the deviation The weight adjustment amount is si,n,j ; where d j is the boundary value of the group;

[0090] Special case: If k i,n +s i,n,j >1, then s i,n,j =1-k i,n ; if k i,n +s i,n,j <0, then s i,n,j =-k i,n .

[0091] Example:

[0092] The purpose of this embodiment is to evaluate and verify that the present invention improves the rolling force prediction accuracy of a certain stand compared to the rolling force correction method using only adaptive learning, and specifically includes the following steps in combination:

[0093] 1. Collect the rolling parameters of 650CL hot rolling of a certain stand, including: the measured rolling force and the predicted rolling force without considering the correction coefficient, the thickness of the strip at the exit and entrance of the rolling mill, the strip width, the rolling speed, and the rolling temperature. The specific contents are as follows:

[0094]

[0095]

[0096]

[0097]

[0098]

[0099] 2. Statistically calculate the deviation between the measured value and the predicted value of rolling force, obtain the correction coefficient by calculation, establish a mathematical model related to the rolling parameters and regress the undetermined coefficients. The specific contents are as follows:

[0100] 1) Calculate the deviation between the actual rolling force value and the predicted value, and obtain the correction coefficient by calculation;

[0101] Strip steel serial number Correction factor Strip steel serial number Correction factor Strip steel serial number Correction factor Strip steel serial number Correction factor 1 1.034948 29 1.045357 57 1.044431 85 1.032916 2 1.048076 30 1.048482 58 1.066837 86 1.045186 3 1.037205 31 1.051603 59 1.043282 87 1.042414 4 1.040461 32 1.050342 60 1.059602 88 1.022234 5 1.044499 33 1.052848 61 1.049233 89 1.033359 6 1.045438 34 1.030744 62 1.049651 90 0.995241 7 1.044193 35 1.050663 63 1.059581 91 0.952994 8 1.046922 36 1.051672 64 1.054208 92 0.947188 9 1.045226 37 1.034232 65 1.002622 93 0.953647 10 1.041553 38 1.05276 66 0.996814 94 0.967376 11 1.038336 39 1.053464 67 0.961297 95 0.966302 12 1.044921 40 1.037874 68 1.006428 96 0.977028 13 1.044007 41 1.043729 69 0.937676 97 0.940708 14 1.052296 42 1.050303 70 0.993483 98 0.947819 15 1.051697 43 1.042039 71 0.979523 99 0.95055 16 1.040702 44 1.045771 72 0.976445 100 0.945567 17 0.962021 45 1.048249 73 0.980849 101 0.967213 18 0.96242 46 1.052054 74 0.954714 102 0.951483 19 0.96828 47 1.044009 75 0.99738 103 0.966574 20 0.965566 48 1.048338 76 1.002177 104 0.952306 21 0.954757 49 1.056184 77 0.976123 105 1.00161 22 0.975012 50 1.048296 78 0.964444 106 0.962796 23 0.929524 51 1.045783 79 0.943209 107 0.965566 24 0.958088 52 1.037755 80 0.972478 108 0.994223 25 1.044338 53 1.042753 81 1.004249 109 0.953976 26 1.052709 54 1.041072 82 1.033104 110 0.962164 27 1.046025 55 1.052373 83 1.052222 111 0.952449 28 1.046458 56 1.039335 84 1.047521 112 0.949934

[0102] 2) Establish a mathematical model related to rolling parameters and regress the unknown coefficients.

[0103] The mathematical model related to the rolling force correction coefficient and rolling parameters is (hereinafter referred to as the correction coefficient prediction model):

[0104] f i,n =a0+a1·ε i,n +a2·v i,n +a3·Ti,n +a4·B i,n +a5·ln(v i,n )+a6·ln(ε i,n ·v i,n )+

[0105] a7·ε i,n ·T i,n +a8·ε i,n ·v i,n ·T i,n

[0106] Where i is the rack number; n is the strip number in rolling order; a0, a1, a2, a3, a4, a5, a6, a7, a8 are the regression coefficients; ε i,n is the logarithmic strain of the strip in the i-stand; T i,n is the rolling temperature of the strip in stand i, ℃; v i,n is the rolling speed of the strip in stand i, m / s; B i,n is the strip width, mm.

[0107] The undetermined coefficients in the correction coefficient prediction model obtained through data regression are as follows:

[0108] <![CDATA[a0]]> -24.96473433 <![CDATA[a1]]> 46.2742493 <![CDATA[a2]]> 0.695402858 <![CDATA[a3]]> 1.87E-02 <![CDATA[a4]]> -3.36E-04 <![CDATA[a5]]> 2.631852329 <![CDATA[a6]]> -4.182263579 <![CDATA[a7]]> -3.95E-02 <![CDATA[a8]]> 1.57E-03

[0109] 3. Establish an adaptive learning model through exponential smoothing method and determine the appropriate gain coefficient. The specific contents are as follows:

[0110] 1) The measured rolling force of the same stand of the same steel grade or steel family collected on site and the predicted rolling force without considering the correction coefficient are sorted according to the rolling time. Since the aforementioned data have been sorted according to the rolling time, they will not be repeated;

[0111] 2) The sorted rolling force data is brought into the exponential smoothing formula to establish a rolling force adaptive learning model;

[0112] The exponential smoothing formula is:

[0113]

[0114] 3) Determine a suitable gain coefficient α so that the average deviation between the measured rolling force and the predicted rolling force of the data set meets a certain target value. When the average deviation between the measured rolling force and the predicted rolling force of the data set is less than 3%, it can be considered that the target is met. After verification and calculation, when the gain coefficient is 0.2, the target requirement is met. In order to compare the effect of improving the rolling force prediction accuracy of a certain stand using only the adaptive learning rolling force correction method, the rolling force prediction deviation rate data before the adaptive model is applied and the rolling force prediction deviation rate data after the adaptive model is applied are statistically analyzed and plotted, as shown in the following figure. Figure 2 and Figure 3 shown.

[0115] 4. The mathematical regression model and the adaptive learning model are integrated by weight distribution to correct the rolling force prediction value of the hot rolling pressure model. The weight can be automatically adjusted based on the preset rules according to the deviation between the rolling force prediction value and the measured value. The specific contents are as follows:

[0116] 1) Integrate mathematical regression model and adaptive learning model through weight distribution;

[0117] m i,1 =k i，1 ·f i，1

[0118]

[0119] k i，n+1 =k i，n +s i，n，j ,n=1,2,3,4…

[0120]

[0121] 2) Set the weight adjustment rule and automatically adjust the weight according to the deviation between the predicted value and the measured value of the rolling force. Calculate the weight coefficient when there is no deviation between the predicted value and the measured value of the rolling force in theory:

[0122]

[0123] In the formula, i is the rack number; n is the strip number in the rolling sequence; It is the weight coefficient when there is no deviation between the predicted value and the measured value of rolling force, hereinafter referred to as the no-bias weight coefficient.

[0124] According to the unbiased weight coefficient With weight coefficient k i，n The weight adjustment amount is set for each deviation group. The weight adjustment rules of this embodiment are as follows:

[0125] According to the unbiased weight coefficient With weight coefficient k i，n The deviation between groups is divided into different weight adjustment groups:

[0126]

[0127]

[0128] The deviation boundary settings and weight adjustments are as follows:

[0129] Group 0 1 2 3 4 5 Deviation Boundary Setting <![CDATA[d0=-1]]> <![CDATA[d1=-0.5]]> <![CDATA[d2=0]]> <![CDATA[d3=0.5]]> <![CDATA[d4=1]]> <![CDATA[d4=1]]> Weight adjustment <![CDATA[s i,n,0 =-0.1]]> <![CDATA[s i,n,1 =-0.05]]> <![CDATA[s i,n,2 =-0.01]]> <![CDATA[s i,n,3 =0.01]]> <![CDATA[s i,n,4 =0.05]]> <![CDATA[s i,n,5 =0.1]]>

[0130] In order to compare the improvement effect of the rolling force prediction accuracy of a certain stand using only the adaptive learning rolling force correction method, the rolling force prediction deviation rate data under the present invention are statistically analyzed and plotted, as shown in FIG. Figure 4 shown.

[0131] based on Figure 2 , Figure 3 and Figure 4 The results shown are that the mean rolling force prediction deviation rate before the adaptive model is applied is 3.97%, and the standard deviation is 1.453; the mean rolling force prediction deviation rate after the adaptive model is applied is 2.57%, and the standard deviation is 1.312; the mean rolling force prediction deviation rate under the present invention is 1.51%, and the standard deviation is 1.104.

[0132] By comparing the rolling force prediction accuracy of the present invention which integrates the mathematical regression model and the adaptive learning model by means of weight distribution with that of the rolling force correction method which only adopts adaptive learning, it can be obviously found that the present invention has higher rolling force prediction accuracy and the overall rolling force prediction deviation rate is smaller.

Claims

1. A method for improving the prediction accuracy of hot rolling pressure model, characterized in that The following steps are involved: S1, collecting hot rolling parameters; S2. Calculate the deviation between the measured value and the predicted value of the rolling force, obtain the correction coefficient by calculation, establish a mathematical model related to the rolling parameters and regress the unknown coefficient; S3, establish an adaptive learning model through exponential smoothing method and determine the appropriate gain coefficient; S4. The mathematical regression model and the adaptive learning model are integrated by weight distribution, so as to correct the rolling force prediction value of the hot rolling pressure model, and the weight can be automatically adjusted based on preset rules according to the deviation between the rolling force prediction value and the measured value.

2. The method for improving the prediction accuracy of hot rolling pressure model according to claim 1, characterized in that: In the step S1, the hot rolling parameters are collected to collect the hot rolling parameters of a certain steel type or steel family, including: the measured rolling force and the predicted rolling force without considering the correction coefficient, the thickness of the strip at the outlet and entrance of the rolling mill, the strip width, the rolling speed and the rolling temperature.

3. The method for improving the prediction accuracy of hot rolling pressure model according to claim 1, characterized in that: In step S2, the specific steps include: S2.1 Statistically calculate the deviation between the measured rolling force value and the predicted value, and obtain the rolling force correction coefficient by calculation. The calculation method of the rolling force correction coefficient is: In the formula, i is the rack number; n is the strip number in the rolling sequence, and f i,n is the rolling force correction coefficient; is the measured rolling force, kN; is the predicted rolling force without considering the correction factor, kN; S2.2 Establish a mathematical model related to the correction coefficient and rolling parameters, referred to as the correction coefficient prediction model, and regress the unknown coefficients. The correction coefficient prediction model is as follows: f i,n s0+a1·ε i,n +a2·v i,n +a3·T i,n +a4·B i,n +a5·ln(v i,n )+a6·ln(ε i,n ·v i,n )+a7·ε i,n ·T i,n +a8·ε i,n ·v i,n ·T i,n Where i is the rack number; n is the strip number in rolling order; a0, a1, a2, a3, a4, a5, a6, a7, a8 are the regression coefficients; ε i,n is the logarithmic strain of the strip in the i-stand; T i,n is the rolling temperature of the strip in stand i, ℃; v i,n is the rolling speed of the strip in stand i, m / s; Where, h0 is the inlet thickness of the strip at the i-frame, mm; h1 is the outlet thickness of the strip at the i-frame, mm; Based on the least squares regression method, the rolling force correction factor f is calculated by using the strip exit and entry thickness, strip width, rolling speed and rolling temperature data at the rolling mill. i,n Perform regression on the undetermined coefficients of the modified coefficient prediction model.

4. The method for improving the prediction accuracy of hot rolling pressure model according to claim 1, characterized in that: In step S3, the specific steps are as follows: S3.1 Sort the measured rolling forces of the same stand of the same steel grade or steel family collected on site and the predicted rolling forces without considering the correction coefficient according to the rolling time; S3.2 brings the rolling force data sorted out in S3.1 into the exponential smoothing formula to establish a rolling force adaptive learning model. The exponential smoothing formula is: In the formula, is the predicted value of the rolling force correction coefficient at the nth setting or control based on adaptive learning; is the measured value of the rolling force correction coefficient after the nth setting or control based on adaptive learning; is the predicted value of the rolling force correction coefficient set or controlled for the n+1th time based on adaptive learning; α i is the gain coefficient, 0≤α≤1; S3.3 Determine the appropriate gain factor α i , so that the deviation between the measured rolling force and the predicted rolling force meets a certain target value.

5. The method for improving the prediction accuracy of hot rolling pressure model according to claim 1, characterized in that: In step S4, the specific steps are as follows: S4.1 integrates the rolling force correction coefficient obtained by the mathematical regression model and the adaptive learning model by weight distribution: The weight distribution method is: Where i is the rack number; n is the strip number in rolling order; m i,n It is the rolling force correction coefficient obtained by integrating the mathematical regression model and the adaptive learning model, which is referred to as the integrated rolling force correction coefficient in the following. i.n is the weight coefficient, 0≤k i,n ≤1, k i,1 =1;s i,n,j is the adjustment item of weight; j is the group number of different weight adjustment amounts; S4.2 sets the weight adjustment rule and automatically adjusts the weight according to the deviation between the predicted value and the measured value of the rolling force. Calculate the weight coefficient when the predicted rolling force value is theoretically equal to the measured value: In the formula, i is the rack number; n is the strip number in the rolling sequence; is the weight coefficient when there is no deviation between the predicted value and the measured value of rolling force, hereinafter referred to as the no-bias weight coefficient; According to the unbiased weight coefficient With weight coefficient k i,n The weight adjustment amount is set for each deviation group.

6. A method for improving the prediction accuracy of hot rolling pressure model according to claim 5, characterized in that: The weight adjustment amount is set for each group. In group 0, the deviation amount The weight adjustment amount is s i,n,0 ; When the group is 1 to j-1, the deviation The weight adjustment amount is s i,n,j-1 ; In the jth group, the deviation The weight adjustment amount is s i,n,j ; where d j is the boundary value of the group; Special case: If k i,n +s i,n,j >1, then s i,n,j =1-k i,n ; if k i,n +s i,n,j <0, then s i,n,j =-k i,n .

Images