A method for adaptive control of roll eccentricity

Abstract

This invention discloses an adaptive roll eccentricity control method, comprising the following steps: acquiring a rolling force detection signal; collecting rolling force measurement values ​​via a rolling force measuring instrument, calculating the difference between the measured value and the set value to obtain a noisy rolling force fluctuation; obtaining a noisy eccentricity signal from the difference through plastic modulus (CM) conversion; representing the noisy eccentricity signal as an effective eccentricity signal; constructing an adaptive linear neural network model to identify the effective eccentricity component as the eccentricity control quantity; optimizing the signal input to the neural network; establishing an error function and optimizing the neural network model parameters; and calculating and executing the hydraulic roll gap adjustment. This invention can achieve good online approximation and elimination of roll eccentricity signals, enabling online roll eccentricity control.

Description

Technical Field

[0001] This invention relates to cold-rolled strip steel production control technology, and more particularly to an adaptive roll eccentricity control method. Background Technology

[0002] Plate thickness accuracy is a key quality indicator for strip steel and a crucial factor determining its market competitiveness. With industrial users continuously raising their requirements for plate thickness quality, plate thickness control technology has become one of the most core and complex technologies in the steel rolling industry, and a hot research topic worldwide.

[0003] Roll eccentricity is a significant factor affecting the quality of high-precision sheet and strip materials. Due to limitations imposed by various factors such as processing conditions and assembly processes, achieving completely eccentric rolls is impossible. Therefore, roll eccentricity compensation control has always been an important component of the AGC (Automatic Gain Control) system for cold-rolled sheet thickness control.

[0004] Roll eccentricity has the following characteristics: (1) Periodicity; the eccentricity of the roll is reflected in signals such as rolling force, roll gap, tension or thickness, and can be regarded as a superposition of a series of sinusoidal periodic waves with a frequency proportional to the roll speed. (2) Complexity; the roll eccentricity signal contains various random interferences caused by acquisition noise and changes in the hardness and thickness of the rolled workpiece, changes in oil film thickness, etc. (3) Variability; the frequency of eccentricity is variable, and the eccentricity frequency changes with the rolling speed. (4) Coupling; the eccentricity signal acquired during the rolling process is not the eccentricity signal of a single roll, but the coupled signal of all roll eccentricity signals. Taking a six-roll HC mill as an example, the eccentricity signal acquired during the rolling process, in addition to interference signals, is at least the superposition of the eccentricity signals of the upper and lower support rolls, upper and lower intermediate rolls, and upper and lower work rolls.

[0005] To address the periodicity and noise interference characteristics of roll eccentricity signals, and the fact that traditional neural networks require first using the FFT algorithm to obtain the eccentricity frequency before constructing an eccentricity signal identification model, resulting in slow adaptive speed during online learning and low operating efficiency, this invention proposes a roll eccentricity control method based on rolling force detection signals and an adaptive linear neural network. This method uses the noisy residual eccentricity signal acquired and converted by a rolling force measuring instrument and the Widrow-Hoff learning rule as the basis for online adjustment of the internal weights of the neural network model, and implements corresponding control measures according to online control requirements, achieving good online approximation and elimination of roll eccentricity signals. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an adaptive roll eccentricity control method to address the deficiencies in the prior art.

[0007] The technical solution adopted by this invention to solve its technical problem is: an adaptive roll eccentricity control method, comprising the following steps: 1) Acquire rolling force detection signal; The rolling force is measured by a rolling force measuring instrument, and the difference between the measured value and the set value is calculated to obtain the rolling force fluctuation with noise. 2) By converting the strip's plastic modulus CM, the difference in rolling force is obtained. Obtaining noisy eccentric signals ;

[0008] 3) Noisy eccentric signals Represented as: ; For the effective eccentricity signal of the roll, This is a noise interference signal; Will Represented as: ; in, ; In the formula, Let be the amplitude of the i-th harmonic; The angular velocity of the roll; The i-th harmonic is the initial phase, i=1,2; t is time. 4) Construct an adaptive linear neural network model from noisy, eccentric signals. Identify effective eccentric components , as the eccentricity control quantity; The input to the adaptive linear neural network model is a noisy, eccentric signal; The output of the adaptive linear neural network model is: the eccentricity identification result; 5) To address the step response characteristics of the rolling force detection unit and the hydraulic roll gap adjustment circuit in actual rolling, Laplace transform and step response parameters are introduced into the neural network to incorporate equipment delay noise in the acquired signal into the model correction range, eliminate noise introduced by equipment dynamic characteristics, and optimize the signal input to the neural network. 6) Establish an error function and optimize the neural network model parameters: By optimizing the error function, the rolling force fluctuation at the stand caused by the effective component of roll eccentricity after eccentricity control is achieved. Rolling force fluctuations caused by eccentric adjustment The sum is minimized; Optimization objective: Minimize the error function, that is, minimize the residual error component that is not eliminated when the roll eccentricity dominant component occurs, and eliminate the influence of eccentricity on the rolling force; 7) Calculation and execution of hydraulic roller gap adjustment; Based on the identified eccentricity The hydraulic roller gap adjustment amount is calculated by combining the conversion factor K. The drive adjustment device performs correction.

[0009] According to the above scheme, in step 4), the neural network is an adaptive linear neural network with single input, single output, and 4 hidden layer nodes.

[0010] According to the above scheme, in step 4), the output of the neural network is: ; In the formula, For any roll eccentricity dominant component The results obtained through identification; For the eccentricity parameter The results obtained through identification; ω is the angular velocity of the roll; t is time.

[0011] According to the above scheme, in step 6), the error function is established as follows: min[ ] or min[ ];

[0012] ; In the formula, or It is an evaluation function for the residual error components of the roll eccentricity that have not been eliminated; For the eccentricity parameter The results obtained through identification; Rolling force measurement value The Laplace transform function; CM is the rolling force measurement value read from the controller CPU at the start of eccentric adjustment; CM is the plastic modulus of the strip. The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings besides (t). The Laplace transform function; The position setpoint read from the controller CPU at the start of eccentric adjustment; K is the conversion coefficient between the hydraulic roll gap position change and the eccentricity change in the cold-rolled strip mill. For the eccentricity adjustment amount or identification result of any roll The Laplace transform function; The fluctuation in stand rolling force caused by the dominant component of roll eccentricity. The Laplace transform function; The effective rolling force fluctuation of any roll External noise interference signals The Laplace transform function.

[0013] According to the above scheme, in step 6), the Widrow-Hoff (LMS) learning rule is adopted, and the network weights are corrected by gradient descent to minimize the error function.

[0014] According to the above scheme, in step 6), the Widrow-Hoff (LMS) learning rule is adopted, and the adjustment amount of the internal weights of the neural network model during the (k+1)th self-learning is: ; ; The new internal weights obtained by the neural network model after the (k+1)th self-learning iteration are... ; ; In the formula, J is the quadratic error evaluation function of the residual error component of the roll eccentricity dominant component that has not been eliminated. ; It is an evaluation function for the residual error components of the roll eccentricity that have not been eliminated; yes The inverse Laplace transform; For learning speed, k represents the number of times the eccentric signal is collected during the rolling process, k = 0, 1, 2, ...; subscript n = 1, 2; Rolling force measurement value The Laplace transform function; The rolling force measurement value read from the controller CPU at the start of eccentric adjustment; The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings except for The Laplace transform function; The position setting value read from the controller CPU at the start of the eccentric adjustment; The result is obtained by identifying the dominant component of eccentricity of any roll using a neural network model; Represents the function Perform the inverse Laplace transform; K is the conversion coefficient between the change in the hydraulic roll gap position and the change in eccentricity in the cold-rolled strip mill.

[0015] According to the above scheme, in step 6), the roll performs a self-learning calculation once every time it passes through a set angle.

[0016] According to the above scheme, in step 7), the conversion coefficient K is calculated as follows: Based on the frame stiffness coefficient CG and the strip plastic modulus CM, we get K=CG / (CG+CM).

[0017] The beneficial effects of this invention are: 1. This invention can achieve good online approximation and elimination of roll eccentricity signals. Addressing the periodicity and noise interference characteristics of roll eccentricity signals, and the fact that traditional neural networks require first using FFT algorithms to obtain the eccentricity frequency before constructing an eccentricity signal identification model, or suffer from slow adaptive speed during online learning, resulting in low operating efficiency and unsuitability for online rolling control requirements, this invention proposes a roll eccentricity control method based on rolling force detection signals and an adaptive linear neural network. This method uses the noisy eccentricity signal collected by the rolling force measuring instrument as the basis for online adjustment of the internal weights of the neural network model, achieving good online approximation and elimination of roll eccentricity signals. 2. The impact of the step response characteristics of the rolling force and pressure detection unit and the hydraulic roll gap position adjustment circuit on the control system was considered; the step response characteristics of the rolling force and pressure detection unit and the hydraulic roll gap position adjustment circuit were considered, so that the control system could more realistically reflect the online rolling state. 3. During the rolling process, after the eccentricity compensation function is put into operation, it automatically tracks and identifies the eccentricity signal and optimizes the amount of correction required by the correction equipment. Attached Figure Description The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a neural network structure diagram according to an embodiment of the present invention; Figure 3 This is a neural network structure diagram of an embodiment of the present invention, considering the step response characteristics of the rolling force pressure detection unit and the hydraulic roll gap position adjustment circuit; Figure 4 This is a schematic diagram of the roll eccentricity control method according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] like Figure 1 As shown, an adaptive roll eccentricity control method includes the following steps: 1) Acquire rolling force detection signal; The rolling force is measured by a rolling force measuring instrument, and the difference between the measured value and the set value is calculated to obtain the rolling force fluctuation with noise. 2) By converting the strip's plastic modulus CM, the difference in rolling force is obtained. Obtaining noisy eccentric signals ;

[0020] 3) Noisy eccentric signals Represented as: ; For the effective eccentricity signal of the roll, This is a noise interference signal; Will Represented as: ; in, ; In the formula, Let be the amplitude of the i-th harmonic; The angular velocity of the roll; Let t be the initial phase of the i-th harmonic, i=1,2; t is time.

[0021] 4) Construct an adaptive linear neural network model from noisy, eccentric signals. Identify effective eccentric components And use an eccentricity correction device to correct it; The input to the adaptive linear neural network model is a noisy, eccentric signal; The output of the adaptive linear neural network model is: the eccentricity identification result; like Figure 2 As shown in the diagram, in the neural network structure, The weight matrix of the hidden layer input is D=[1, 1, 1, 1] T The weight matrix of the output layer input is W=[ , , , ] T The matrix of hidden layer nodes is C=[ , , , ] T The eccentricity adjustment amount of the neural network output is ; The results obtained by identifying the dominant component of roll eccentricity using a neural network model:

[0022] In the formula, The result is obtained by identifying the dominant component of eccentricity of any roll using a neural network model; It is both the result obtained by the neural network model in identifying the actual eccentricity parameters and the internal weights of the neural network model; Let ω be the angular velocity of any roll; t be time.

[0023] By constructing an adaptive linear neural network with a single input, a single output, and four hidden nodes, the hidden nodes strictly correspond to the dominant component of roll eccentricity, eliminating interference from non-periodic noise (such as random mechanical vibration), making the model sensitive only to eccentricity signals.

[0024] 5) To address the step response characteristics of the rolling force detection unit and the hydraulic roll gap adjustment circuit in actual rolling, the step response time constant T2 of the rolling force detection unit and the step response time constant T1 of the hydraulic roll gap adjustment circuit are incorporated into the model; for example... Figure 3 ; ; ; in, Let be the transfer function of the hydraulic roll gap position adjustment loop. The transfer function of the rolling force and pressure detection unit; like Figure 4 As shown, considering the step response characteristics of the rolling force pressure detection unit and the hydraulic roll gap position adjustment circuit, the measured rolling force value is expressed as follows:

[0025] In the formula, Rolling force measurement value The Laplace transform function; CM is the rolling force measurement value read from the controller CPU at the start of eccentric adjustment; CM is the plastic modulus of the strip. The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; such as Figure 4 As shown, To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings besides (t). The Laplace transform function; The position setting value read from the controller CPU at the start of the eccentric adjustment; For the eccentricity adjustment amount or identification result of any roll The Laplace transform function; The fluctuation in stand rolling force caused by the dominant component of roll eccentricity. The Laplace transform function; The effective rolling force fluctuation of any roll External noise interference signals The Laplace transform function; K is the conversion coefficient between the change in the hydraulic roll gap position and the change in eccentricity of the cold-rolled strip mill; K=CG / (CG+CM), where CG is the frame stiffness coefficient.

[0026] To address the step response characteristics of the rolling force and pressure detection unit and the hydraulic roll gap adjustment circuit in actual rolling, Laplace transform and step response parameters are introduced into the neural network. Equipment delay noise in the acquired signal (such as signal distortion caused by sensor response lag) is included in the model correction range, and noise introduced by the dynamic characteristics of the equipment is eliminated, so that the signal input to the neural network is closer to the real eccentric state. 6) Establish an error function and continuously adaptively optimize the neural network model parameters: By optimizing the error function, the rolling force fluctuation at the stand caused by the dominant component of roll eccentricity after eccentricity control is achieved. Rolling force fluctuations caused by eccentric adjustment Minimize the sum of the components; under the most ideal adjustment effect, the rolling force fluctuation at the stand caused by the dominant component of roll eccentricity. Rolling force fluctuations caused by eccentric adjustment The sum is zero, meaning the influence of the dominant component of roll eccentricity on the product strip is completely eliminated. Optimization objective: Minimize the error function, that is, minimize the residual error component that is not eliminated when the adjustment occurs; eliminate the influence of eccentricity on rolling force.

[0027] By using Laplace transform and inverse transform, the error function J is converted into a form that can be directly obtained by the controller, thus avoiding... , Signals that cannot be directly measured.

[0028] min[ ] or min[ ]; ; like Figure 4 As shown, ; ;

[0029]

[0030] ; ; ; make ; for and In other words, For variables, and Since it is a constant term, therefore, it makes Satisfy min[ Equivalent to making Satisfy min[ Therefore, the purpose of the roll eccentricity control method based on rolling force detection signal and adaptive linear neural network can be described as: by calculating a suitable... to make Minimize, that is, make Satisfy min[ ].

[0031] In the formula, or It is an evaluation function for the residual error components of the roll eccentricity that have not been eliminated; For the eccentricity parameter The results obtained through identification; Rolling force measurement value The Laplace transform function; CM is the rolling force measurement value read from the controller CPU at the start of eccentric adjustment; CM is the plastic modulus of the strip. The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; such as Figure 4 As shown, To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings besides (t). The Laplace transform function; The position setting value read from the controller CPU at the start of the eccentric adjustment; For the eccentricity adjustment amount or identification result of any roll The Laplace transform function; The fluctuation in stand rolling force caused by the dominant component of roll eccentricity. The Laplace transform function; The effective rolling force fluctuation of any roll External noise interference signals The Laplace transform function; K is the conversion coefficient between the change in the hydraulic roll gap position and the change in eccentricity of the cold-rolled strip mill, K=CG / (CG+CM), where CG is the frame stiffness coefficient.

[0032] In this embodiment, the Widrow-Hoff (Least Mean Square Error) learning rule is adopted, and the network weights are corrected using gradient descent to minimize the error function. Therefore, the error evaluation function of the adaptive neural network can be set as follows:

[0033]

[0034] In the formula, J is the square error evaluation function of the remaining error component of the roll eccentricity dominant component that has not been eliminated; It is an evaluation function for the residual error components of the roll eccentricity that have not been eliminated; yes The inverse Laplace transform; For the eccentricity parameter The results obtained through identification; Rolling force measurement value The Laplace transform function; CM is the rolling force measurement value read from the controller CPU at the start of eccentric adjustment; CM is the plastic modulus of the strip. The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; such as Figure 4 As shown, To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings besides (t). The Laplace transform function; The position setting value is read from the controller CPU at the start of the eccentric adjustment; K is the conversion coefficient between the change in the hydraulic roll gap position and the change in eccentricity of the cold-rolled strip mill, K=CG / (CG+CM), where CG is the frame stiffness coefficient.

[0035] Using the Widrow-Hoff (Least Mean Square Error) learning rule, the adjustment amount of the internal weights of the neural network model during the (k+1)th self-learning iteration is: ; ; The new internal weights obtained by the neural network model after the (k+1)th self-learning iteration are... ; ; In the formula, For learning speed, k represents the number of times the eccentric signal is collected during the rolling process, k = 0, 1, 2, ...; subscript n = 1, 2; Rolling force measurement value The Laplace transform function; The rolling force measurement value read from the controller CPU at the start of eccentric adjustment; The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings except for The Laplace transform function; The position setting value read from the controller CPU at the start of the eccentric adjustment; The result is obtained by identifying the dominant component of eccentricity of any roll using a neural network model; Represents the function Perform the inverse Laplace transform; CM is the plastic modulus of the strip; K is the conversion coefficient between the change in the hydraulic roll gap position and the change in eccentricity of the cold-rolled strip mill, K=CG / (CG+CM), where CG is the frame stiffness coefficient.

[0036] 7) Calculation and execution of hydraulic roller gap adjustment; Based on the identified eccentricity The hydraulic roller gap adjustment amount is calculated by combining the conversion factor K. The drive adjustment device performs correction. Calculate the conversion factor K: Based on the frame stiffness coefficient CG and the strip plastic modulus CM, we get K=CG / (CG+CM); Derivation of hydraulic roll gap adjustment amount: That is, based on the identified eccentricity, the required hydraulic roller gap adjustment range is calculated; Execute corrective measures: The signal is sent to the hydraulic roll gap position adjustment device on the frame to adjust the roll gap in real time and counteract the rolling force fluctuations caused by roll eccentricity. Closed-loop feedback: During the next signal acquisition, the adjusted rolling force change is incorporated to continuously optimize the weights and adjustment amounts, forming closed-loop control.

[0037] Simulation experiments and results analysis.

[0038] To verify the feasibility and reliability of the roll eccentricity control method based on rolling force detection signal and adaptive linear neural network in online rolling applications through simulation experiments, it is assumed that the dominant eccentricity component of any roll in the cold rolling mill stand is...

[0039] In the formula, NEU represents the rotational speed of the eccentric roll, measured in revolutions per second (rpm).

[0040] Within the eccentric signal, add one low-frequency perturbation signal and one high-frequency perturbation signal.

[0041] Thus, the noisy eccentric signal obtained by the rolling force measuring instrument is...

[0042] In addition, assuming that the change in the position setpoint output by all functions other than the eccentricity compensation function, such as AGC, is...

[0043] Simultaneously, it is assumed that the time constant of the hydraulic roll gap position adjustment circuit of the cold-rolled strip mill is... The time constant of the rolling force pressure detection unit The conversion coefficient K = 0.5 between the change in the hydraulic roll gap position and the change in eccentricity in a cold-rolled strip mill.

[0044] Increase learning speed The internal weights of the neural network model are adaptively calculated every 1.5° rotation of the roll.

[0045] Simulation experiments demonstrate that, under conditions of superimposed low-frequency and high-frequency noise disturbances, this adaptive neural network control method can: reduce the length of the eccentricity defect during the unit startup phase; when the roll speed is between 0.3 rpm and 4 rpm, it eliminates over 95% of the dominant eccentricity; when the roll speed is between 4 rpm and 8.5 rpm, it eliminates over 80% of the dominant eccentricity; if the set stable roll speed is between 0.3 rpm and 8.5 rpm, it can still effectively reduce the adverse effects of the dominant eccentricity during rolling speed switching, and can quickly adapt to the new eccentricity frequency after the speed stabilizes at the end of the rolling speed switching process. This indicates that the method has good roll eccentricity elimination capability, noise interference resistance capability, and online rolling application and adaptability.

[0046] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. An adaptive roll eccentricity control method, characterized in that, Includes the following steps: 1) Acquire rolling force detection signal; The rolling force is measured by a rolling force measuring instrument, and the difference between the measured value and the set value is calculated to obtain the rolling force fluctuation with noise. 2) By converting the strip's plastic modulus CM, the difference in rolling force is obtained. Obtaining noisy eccentric signals ; 3) Noisy eccentric signals Represented as: ； This is the effective eccentricity signal for the rolls. This is a noise interference signal; Will Represented as: ； in, ; In the formula, Let be the amplitude of the i-th harmonic; The angular velocity of the roll; The i-th harmonic is the initial phase, i=1,2; t is time. 4) Construct an adaptive linear neural network model from noisy, eccentric signals. Identify effective eccentric components , as the eccentricity control quantity; The input to the adaptive linear neural network model is a noisy, eccentric signal; The output of the adaptive linear neural network model is: the eccentricity identification result; 5) To address the step response characteristics of the rolling force detection unit and the hydraulic roll gap adjustment circuit in actual rolling, Laplace transform and step response parameters are introduced into the neural network to incorporate equipment delay noise in the acquired signal into the model correction range, eliminate noise introduced by equipment dynamic characteristics, and optimize the signal input to the neural network. 6) Establish an error function and optimize the neural network model parameters: By optimizing the error function, the rolling force fluctuation at the stand caused by the effective component of roll eccentricity after eccentricity control is achieved. With the rolling force fluctuation caused by the eccentricity adjustment The sum is minimized; Optimization objective: Minimize the error function, that is, minimize the residual error component that is not eliminated when the roll eccentricity dominant component occurs, and eliminate the influence of eccentricity on the rolling force; 7) Calculation and execution of hydraulic roller gap adjustment; Based on the identified eccentricity The hydraulic roller gap adjustment amount is calculated by combining the conversion factor K. The drive adjustment device performs correction.

2. The adaptive roll eccentricity control method according to claim 1, characterized in that, In step 4), the neural network is a single-input, single-output, adaptive linear neural network containing 4 hidden layer nodes.

3. The adaptive roll eccentricity control method according to claim 2, characterized in that, In step 4), the output of the neural network is: ； In the formula, For any roll eccentricity dominant component The results obtained through identification; For the eccentricity parameter The results obtained through identification; The angular velocity of the roll; t represents time.

4. The adaptive roll eccentricity control method according to claim 1, characterized in that, In step 6), the error function is established as follows: min[ ] or min[ ]; ； In the formula, or It is an evaluation function for the residual error components of the roll eccentricity that have not been eliminated; For the eccentricity parameter The results obtained through identification; Rolling force measurement value The Laplace transform function; CM is the rolling force measurement value read from the controller CPU at the start of eccentric adjustment; CM is the plastic modulus of the strip. The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings besides (t). The Laplace transform function; The position setpoint read from the controller CPU at the start of the eccentricity adjustment; K is the conversion coefficient between the hydraulic roll gap position change and the eccentricity change in the cold-rolled strip mill. For the eccentricity adjustment amount or identification result of any roll The Laplace transform function; The fluctuation in stand rolling force caused by the dominant component of roll eccentricity. The Laplace transform function; The effective rolling force fluctuation of any roll External noise interference signals The Laplace transform function.

5. The adaptive roll eccentricity control method according to claim 4, characterized in that... In step 6), the Widrow-Hoff learning rule is adopted, and the network weights are corrected by gradient descent to minimize the error function.

6. The adaptive roll eccentricity control method according to claim 5, characterized in that, In step 6), the Widrow-Hoff learning rule is adopted, and the adjustment amount of the internal weights of the neural network model during the (k+1)th self-learning is: Using the Widrow-Hoff learning rule, the adjustment amount of the internal weights of the neural network model during the (k+1)th self-learning is: ； ； The new internal weights obtained by the neural network model after the (k+1)th self-learning iteration are... ； ； Where J is the quadratic error evaluation function. ; In the formula, For learning speed, k represents the number of times the eccentric signal is collected during the rolling process, k = 0, 1, 2, ...; subscript n = 1, 2; Rolling force measurement value The Laplace transform function; The rolling force measurement value read from the controller CPU at the start of eccentric adjustment; The step response time constant of the hydraulic roll gap position adjustment circuit; The step response time constant of the rolling force and pressure detection unit; To exclude the eccentricity compensation position setting value The sum of all other roll gap position settings except for The Laplace transform function; The position setting value read from the controller CPU at the start of the eccentric adjustment; The result is obtained by identifying the dominant component of eccentricity of any roll using a neural network model; This indicates that the inverse Laplace transform is performed on the function; K is the conversion coefficient between the change in the hydraulic roll gap position and the change in eccentricity in the cold-rolled strip mill; CM is the plastic modulus of the strip.

7. The adaptive roll eccentricity control method according to claim 6, characterized in that, In step 6), the roll performs a self-learning calculation once every time it passes through a set angle.

8. The adaptive roll eccentricity control method according to claim 1, characterized in that, In step 7), the conversion coefficient K is calculated as follows: Based on the frame stiffness coefficient CG and the strip plastic modulus CM, we get K=CG / (CG+CM).

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images