CN-121997154-A - Prediction method for rolling force of hot continuous rolling rough rolling area flat rolling mill in different times

CN121997154ACN 121997154 ACN121997154 ACN 121997154ACN-121997154-A

Abstract

The invention relates to the field of automatic control of metallurgical rolling, and discloses a rolling force prediction method for different passes of a hot continuous rolling rough rolling area flat rolling mill, which comprises the specific steps of collecting industrial data of different passes of the hot continuous rolling rough rolling area flat rolling mill, and preprocessing to form a complete data set; the method comprises the steps of constructing a rolling force theoretical model of a rough rolling mill in a rough rolling area based on a Ames formula to obtain theoretical rolling forces of different passes of the rough rolling mill in a hot continuous rolling rough rolling area, constructing a rolling force prediction model of the rough rolling mill in the rough rolling area based on the rolling force theoretical model of the rough rolling mill in the rough rolling area and a data driving model, training the constructed rolling force prediction model, and predicting the rolling forces of different passes of the rough rolling mill in the hot continuous rolling area. The method solves the problems of insufficient generalization of the traditional pure mechanism model and weak physical interpretability of the pure data model, forms a hybrid prediction frame with complementary advantages, lays a foundation for plate shape control of the subsequent finish rolling process, and improves rolling stability and production efficiency.

Inventors

SONG CHUNNING
GUO KAIMIN
CHEN YAFEI
WEI HAIYANG
SONG MENGXIAO
LIU CHENYANG
HU YUFA
ZHANG LI
LI DONG
GUO XIAOMING
Chen Senliu
WANG ZHEN
LIANG PEI

Assignees

河南钢铁集团有限公司
安阳钢铁集团有限责任公司

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (8)

1. A rolling force prediction method for different passes of a hot continuous rolling rough rolling zone flat rolling mill is characterized by comprising the following specific steps: industrial data of different passes of a hot continuous rolling rough rolling area plain rolling mill are collected, and pretreatment is carried out to form a complete data set; Constructing a rolling force theoretical model of the rough rolling area plain-barreled mill based on a Ames formula, and obtaining theoretical rolling forces of different passes of the hot continuous rolling rough rolling area plain-barreled mill; constructing a rolling force prediction model of the rough rolling area flat rolling mill based on the rolling force theoretical model and the data driving model of the rough rolling area flat rolling mill; And training the established rolling force prediction model by utilizing the acquired data set, and predicting the rolling forces of different passes of the hot continuous rolling rough rolling area plain rolling mill.
2. The method for predicting rolling force of different passes of hot continuous rolling rough rolling mill according to claim 1, wherein the industrial data of different passes of the hot continuous rolling rough rolling mill comprises roll radius, inlet width, outlet width, inlet thickness, outlet thickness, rolling temperature, rolling speed, roll elastic modulus, roll poisson ratio, friction coefficient, flow stress, roll gap value, deformation resistance and rolling force.
3. The method for predicting rolling force of different passes of hot continuous rolling rough rolling zone flat rolling mill according to claim 1, wherein the method is characterized by collecting industrial data of different passes of the hot continuous rolling rough rolling zone flat rolling mill and preprocessing the industrial data to form a complete data set, and comprises the following specific steps: the abnormal value of the acquired data is removed by adopting a 3 sigma criterion, and the formula is as follows: Wherein, the S x is standard deviation, n is the number of samples of the dataset, x i is the ith data; Normalizing the rolling data after the abnormal values are removed to reflect the inherent distribution characteristics of the rolling data, wherein the formula is as follows: Where x is the raw data, x min and x max are the minimum and maximum values of the data, respectively, and y is the normalized data.
4. The method for predicting the rolling force of different passes of the hot continuous rolling rough rolling mill according to claim 1, wherein the method for constructing the rolling force theoretical model of the rough rolling mill based on the Ames formula comprises the following specific steps: The theoretical model of the rolling force of the hot continuous rolling rough rolling area plain rolling mill adopts a Ames formula, and the following formula is adopted: wherein P is rolling force and Pa; K m is plane deformation resistance, B is plate width, mm, B= (B 0 +B 1 )/2,B 0 is plate width before rolling, B 1 is plate width after rolling, Q p is stress state coefficient, and l c ' is contact arc length, mm; plane deformation resistance K m was calculated using the following formula: Wherein sigma is deformation resistance, C is carbon content,%; T K is absolute temperature, T K =t+273, T is deformation temperature, °C, e is deformation degree, u is deformation speed, m/s, a 1 ～a 7 is deformation resistance model coefficient; The calculation formula of the stress state coefficient Q p after flattening the roller is as follows: Due to the influence of the roller elastic flattening on the contact arc length, the contact arc length l c ' of the roller elastic flattening is obtained as follows: Wherein p' m is the unit average pressure which is obtained by considering the elastic flattening of the roller, pa, R is the radius of the roller and mm, c is the calculation coefficient of the contact arc length which is obtained by considering the elastic flattening of the roller, R' is the elastic flattening radius of the roller, mm, E is the elastic modulus, and Deltah is the rolling reduction, mm.
5. The method for predicting the rolling force of different passes of the hot continuous rolling rough rolling zone flat rolling mill according to claim 1, wherein the method for constructing the rolling force prediction model of the rough rolling zone flat rolling mill based on the rolling force theoretical model and the data driving model comprises the following specific steps: Describing the overall error by using the absolute error, and establishing a support vector machine regression model; Optimizing a kernel function parameter sigma and a penalty factor gamma of a support vector machine regression model by adopting a gray wolf optimization algorithm to form a data driving model; and constructing a rolling force prediction model of the rough rolling area plain rolling mill based on the theoretical model by taking the theoretical rolling force calculated value as an input variable.
6. The method for predicting rolling force of different passes of hot continuous rolling rough rolling mill according to claim 5, wherein the specific steps of data driving model establishment are as follows: Setting the number N of the gray wolves, the maximum iteration number t, a punishment factor c to be optimized, a kernel function g, upper and lower boundaries of the wolf group position, and the position and fitness value of the initialized alpha wolves, beta wolves and delta wolves; the positions and objective functions of alpha wolves, beta wolves and delta wolves are updated by randomization according to the fitness function, the positions of the wolves are updated by continuous iteration, the positions of the wolves are tracked, approached, pursued, harassd and attacked, in each iteration process, the alpha wolves, the beta wolves and the delta wolves are respectively global first, second and third optimal solutions, the positions of other searched gray wolves are updated according to the position information of the three optimal solutions, and the search of the positions of the alpha wolves, the beta wolves and the delta wolves is updated according to the following formula: wherein t is the current iteration number; And Respectively representing the position vectors of alpha wolves, beta wolves and delta wolves in the current population; Respectively representing the distances between the current candidate gray wolves and the three optimal wolves; And Are coefficient vectors; The search for other gray wolf locations is updated as follows: wherein t is the current iteration number; is a coefficient vector; A position vector representing a prey; a position vector representing the current gray wolf; Is a random vector in [0,1], For the distance between the wolf's flock and the prey, The component of (2) decreases linearly from 2 to 0 in the iterative process.
7. The method for predicting rolling force of different passes of hot continuous rolling rough rolling area flat rolling mill according to claim 6, wherein the method for optimizing the kernel function parameter sigma and penalty factor gamma of the support vector machine regression model by adopting a gray wolf optimization algorithm is specifically as follows: When a support vector machine regression model is used for the sum of squares of errors as experience loss, for a sample set Wherein x i ∈R n is the input vector of training samples, y i ∈R n is the corresponding predicted output, N is the number of samples, N is the vector dimension, and nonlinear function is utilized Mapping the sample to a high-dimensional feature space to obtain a regression prediction model of the sample, wherein the regression prediction model comprises the following formula: Wherein w T is a weight vector in the feature space, b is a deviation, b ε R; When the support vector machine regression model is used for the regression task, the optimization model can be expressed as follows according to the principle of minimizing structural risk: Wherein gamma is punishment parameter, xi i is relaxation variable, For the mapped values of the original data, s.t. is a constraint condition, and then the regression function can be obtained by solving the Lagrangian multiplier method and the KKT condition as follows: Wherein α i is a lagrange multiplier, K (x, x i ) is a kernel function, K (x, x i )＝exp(-||x-x i || 2 /2σ 2 ), σ is a kernel width value, and x i are original sequence values; judging whether the support vector machine regression model reaches the maximum iteration number, and outputting a penalty factor c and a kernel function g which are optimal for the support vector machine regression model.
8. The method for predicting the rolling force of different passes of the hot continuous rolling rough rolling area flat rolling mill according to claim 1, wherein the training of the established rolling force prediction model by using the collected data set predicts the rolling force of different passes of the hot continuous rolling rough rolling area flat rolling mill, and comprises the following specific steps: taking production data and a rolling force theoretical calculation value as input, taking rough rolling forces of different passes as output, and training an established rolling force prediction model; Testing the trained rolling force prediction model; Evaluating the precision of a prediction result output by the rolling force model by adopting a preset evaluation standard; and predicting the rolling force of different passes of the hot continuous rolling rough rolling area plain rolling mill by using the rolling force prediction model after training.

Description

Prediction method for rolling force of hot continuous rolling rough rolling area flat rolling mill in different times Technical Field The invention relates to the field of automatic control of metallurgical rolling, in particular to a rolling force prediction method for different passes of a hot continuous rolling rough rolling zone flat rolling mill. Background As a key device for rolling a plate blank into an intermediate blank, the hot continuous rolling rough rolling area flat rolling mill is used for accurately predicting rolling force, and is a core technical support for realizing dynamic setting of a roll gap, stable control of a plate shape and smooth production flow. The current prediction of rough rolling force still has a remarkable technical bottleneck, and the requirements of modern hot rolling production on high precision and strong robustness are difficult to meet. At present, the traditional method mainly depends on a mechanism model (such as an Orowan formula and a Sims formula) based on a rolling theory, and the mechanism model can reflect physical essence and has definite theoretical basis, but because of nonlinear, time-varying and strong coupling relations among various factor variables in the rolling process, a larger error exists between a result of calculating the rolling force and an actual field working condition by adopting the mechanism model, so that the accuracy of the mechanism model is difficult to meet actual production requirements. Unlike theoretical analysis method, the pure data driving model depends on the historical data to train the mapping relation of input and output, and can realize higher precision when the data are sufficient and the working condition is stable, but the generalization capability and the interpretability of the model have inherent defects, the adaptability to new steel types and new processes is poor, the inherent physical mechanism of the rolling force change cannot be revealed, and the process parameter optimization is difficult to guide. The existing rolling force prediction method has obvious defects in the aspects of complex working condition adaptability, physical mechanism and data fusion, so that the prediction precision is low, the generalization capability is weak, the interpretation is poor, and the requirements of high-precision and intelligent hot rolling production cannot be met. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a rolling force prediction method for different passes of a hot continuous rolling rough rolling area flat rolling mill, which can integrate rolling mechanism and real-time data and give consideration to pass characteristic difference, and has important significance for improving the production stability of the rough rolling area, reducing equipment load fluctuation and improving the dimensional accuracy of products. In order to achieve the above purpose, the invention adopts the following technical scheme: The rolling force prediction method for different passes of hot continuous rolling rough rolling zone flat rolling mill comprises the following specific steps: industrial data of different passes of a hot continuous rolling rough rolling area plain rolling mill are collected, and pretreatment is carried out to form a complete data set; Constructing a rolling force theoretical model of the rough rolling area plain-barreled mill based on a Ames formula, and obtaining theoretical rolling forces of different passes of the hot continuous rolling rough rolling area plain-barreled mill; constructing a rolling force prediction model of the rough rolling area flat rolling mill based on the rolling force theoretical model and the data driving model of the rough rolling area flat rolling mill; And training the established rolling force prediction model by utilizing the acquired data set, and predicting the rolling forces of different passes of the hot continuous rolling rough rolling area plain rolling mill. Further, the industrial data of different passes of the hot continuous rolling rough rolling zone flat rolling mill comprise roll radius, inlet width, outlet width, inlet thickness, outlet thickness, rolling temperature, rolling speed, roll elastic modulus, roll poisson ratio, friction coefficient, flow stress, roll gap value, deformation resistance and rolling force. Further, the method for collecting industrial data of different passes of the hot continuous rolling rough rolling zone plain rolling mill and preprocessing the industrial data to form a complete data set comprises the following specific steps: the abnormal value of the acquired data is removed by adopting a 3 sigma criterion, and the formula is as follows: Wherein, the S x is standard deviation, n is the number of samples of the dataset, x i is the ith data; Normalizing the rolling data after the abnormal values are removed to reflect the inherent distribution characteristics of the rolling data, wherein