A data-driven intelligent identification method for plate shape control efficacy

Abstract

The present invention relates to the technical field of cold-rolled strip steel, and more particularly to a data-driven intelligent identification method for flatness control efficacy. The method comprises: acquiring industrial data from the production process of a rolling mill unit; preprocessing the industrial data; and establishing a data set based on the preprocessed industrial data, wherein the industrial data includes process record data and flatness meter data; performing noise reduction and decomposition and reconstruction on the flatness meter data in the data set; and intelligently identifying the control efficacy of the noise-reduced and decomposed and reconstructed data set using an adaptive moment estimation algorithm. The present invention can analyze and clarify the control capability of a control mechanism, providing more control information for subsequent calculation of control quantities for flatness control.

Description

Technical Field

[0001] The present invention relates to the technical field of cold-rolled strip steel, and in particular to a data-driven intelligent identification method for plate shape control efficacy. Background Art

[0002] The steel industry is a vital component of my country's industrial landscape, and its development, to a certain extent, represents a country's level of industrialization. Cold-rolled sheet and strip, characterized by high production output, rapid forming speed, and excellent performance, is widely used in industries such as automotive and electrical appliances. With the advancement of manufacturing, the precision requirements for sheet and strip are becoming increasingly stringent across various industries. The quality indicators of sheet and strip primarily focus on thickness and shape. While transverse thickness control currently largely meets these requirements, shape control still presents some unresolved challenges. Improving shape control accuracy is a crucial component in improving sheet quality.

[0003] The fifth stand in a six-high CVC mill uses a closed-loop feedback control strategy to fine-tune flatness, making it the most critical component of the flatness control system. Established during the stable rolling phase, this closed-loop feedback control system calculates the deviation between the measured flatness returned by the flatness meter and the target flatness. Using a closed-loop feedback control model, the control amount for each actuator is calculated based on its control efficiency. Control signals are then sent to the actuators, and new flatness deviations are detected. The actuators are continuously and dynamically adjusted in real time, ultimately achieving a stable, optimal flatness. Currently, the control mechanisms in a five-stand tandem cold mill include intermediate roll bending (IRB), work roll bending (WRB), intermediate roll shifting (IRS), tilting, and staged cooling. Tilting can eliminate primary and tertiary flatness deviations, while bending and shifting can eliminate secondary and quaternary flatness deviations. Higher-order flatness deviations require staged cooling. The control performance of each actuator is reflected in the control efficiency curve. Generally, segmented cooling cannot be solved by formulas and has a slow response speed. It is often used to eliminate plate shape deviations that cannot be eliminated by other actuators. Therefore, only the effects of pressing tilt, bending roll and shifting roll are considered here.

[0004] The control efficiency is the basis of plate shape control and can effectively reflect the control performance of the actuator on the plate shape. Due to its simple and fast calculation process, it can play a role in units with multiple control methods. Therefore, it can make full use of the rolling production process data to optimize it, thereby obtaining more accurate plate shape control. The control efficiency will affect the accuracy of plate shape control. In actual production, the control efficiency of each actuator varies with different process parameters, especially when the width-to-thickness ratio is large, it is more unstable. However, the control efficiency used on site is often used for multiple working conditions, and there are certain errors in specific working conditions, and the update speed is slow. Therefore, obtaining accurate control efficiency is the focus of plate shape control research. Summary of the Invention

[0005] The purpose of the present invention is to provide a data-driven intelligent identification method for plate shape control efficacy, clarify the control capabilities of each control mechanism, solve the problem of fixed control efficacy, improve the accuracy of plate shape control, and realize intelligent identification of control efficacy.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A data-driven intelligent identification method for plate shape control efficacy, comprising:

[0008] Acquiring industrial data during a production process of a rolling mill unit, preprocessing the industrial data, and establishing a data set based on the preprocessed industrial data, wherein the industrial data includes process record data and shape meter data;

[0009] The shape meter data in the data set is subjected to denoising and decomposition and reconstruction, and an adaptive moment estimation algorithm is used to intelligently identify the control efficacy of the denoised and decomposition-reconstructed data set.

[0010] Furthermore, preprocessing the industrial data includes: performing outlier processing on the process record data.

[0011] Furthermore, the dataset is established based on the preprocessed industrial data, including:

[0012] Based on the outlet strip length, the pre-processed process record data and the shape meter data are matched in time series to construct the data set.

[0013] Furthermore, the process of performing the corresponding matching includes: respectively obtaining a first flatness deviation in the process record data and a second flatness deviation in the flatness meter data, and performing the corresponding matching based on comparative verification of the first flatness deviation and the second flatness deviation.

[0014] Furthermore, performing noise reduction processing on the shape meter data in the data set includes:

[0015] The wavelet transform algorithm is used to perform noise reduction on the shape meter data, and the formula is:

[0016] s(n)=f(n)+σe(n)

[0017] Where s(n) is the actual noisy signal; f(n) is the effective signal; n is the sampling time; σ is the noise intensity; and e(n) is the noise component.

[0018] Furthermore, performing noise reduction processing on the shape meter data using a wavelet transform algorithm includes:

[0019] The shape meter data is subjected to multi-layer decomposition of the original signal, and the max_level=log2(l data / l filter ) calculate the number of layers of the multi-layer decomposition; after determining the number of decomposition layers, optimize the threshold of the wavelet coefficient using a genetic algorithm; finally, reconstruct the signal using an inverse wavelet transform to obtain the denoised plate shape meter data.

[0020] Furthermore, decomposing and reconstructing the shape meter data in the data set includes:

[0021] Decomposing the shape meter data based on Legendre orthogonal polynomials to obtain coefficients λ1, λ2, λ3 and λ4 of each shape basis mode;

[0022] The shape meter data is reconstructed based on the coefficients λ1, λ2, λ3 and λ4 of each shape basis mode to obtain decomposed and reconstructed shape meter data.

[0023] Furthermore, the adaptive moment estimation algorithm is used to intelligently identify the control efficacy of the noise reduction and decomposition and reconstruction data set, including:

[0024] Based on the adaptive moment estimation algorithm, the first-order and second-order moment estimates of the gradient are used to automatically update the learning rate according to the data characteristics of the data set, and the first-order and second-order moment estimates are corrected for deviations. An updated calculation model for the variable x is defined based on the deviation corrections of the first-order and second-order moment estimates, and the variable x is updated and calculated based on the updated calculation model of the variable x, thereby realizing intelligent identification of the regulation efficacy.

[0025] Furthermore, the update calculation model of the variable x is:

[0026]

[0027] Where x t is the x variable at the current moment; x t-1 is the x variable at the previous moment; η∈(0,1] is the learning rate; ε is a constant; is the bias correction for the first-order moment estimate; is the bias correction for the second moment estimate.

[0028] The beneficial effects of the present invention are:

[0029] Starting from actual production data from industrial sites, this method performs data preprocessing operations such as outlier processing and data matching on process record data and shape meter data to create a data set. It then performs noise reduction on the shape meter data and uses an adaptive moment estimation algorithm to intelligently identify the control efficacy of the processed data set. This allows for analysis and clarification of the control mechanism's control capabilities, providing more control information for subsequent shape control calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0031] Figure 1 This is a flow chart of a data-driven intelligent identification method for plate shape control efficacy according to an embodiment of the present invention;

[0032] Figure 2 A comparison diagram of the flatness deviation between the process record data and the flatness meter data according to an embodiment of the present invention;

[0033] Figure 3 This is a diagram of the GA-WT noise reduction process according to an embodiment of the present invention;

[0034] Figure 4 Figure 1 is a comparison diagram of three denoising methods according to an embodiment of the present invention, wherein (a) is a denoising effect diagram of GA-WT, (b) is a denoising effect diagram of WT, and (c) is a denoising effect diagram of EEMD;

[0035] Figure 5 : This is a diagram representing the basis modes of various plate shape defects of the Legendre orthogonal polynomials according to an embodiment of the present invention, where Y1 is the right wave, Y2 is the left wave, Y3 is the double-sided wave, Y4 is the middle wave, Y5 is the left three-quarter wave, Y6 is the right three-quarter wave, Y7 is the edge-middle compound wave, and Y8 is the quarter wave;

[0036] Figure 6 This is a comparison chart of the control effects of the adaptive moment estimation algorithm recognition and on-site recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] This embodiment provides a data-driven intelligent identification method for plate shape control efficacy, such as Figure 1 As shown, including:

[0040] S1. Obtain industrial data during the actual production process of the rolling mill unit and pre-process the industrial data;

[0041] Industrial data includes process record data and flatness meter data. The flatness meter data includes the strip residual stress deviation and strip length signal in each measurement area. The strip residual stress deviation is processed and converted into flatness deviation; and the process record data is processed for outlier value processing using data cleaning.

[0042] In exported process log data files, null values ​​often appear due to inconsistent sensor sampling times or electrical failures. This can cause errors in numerical statistics and calculations, so null values ​​need to be filled. If the value preceding the null value is valid, its value replaces the null value. If the null value is in the first row, the null value row is filled with the first non-null value found in the subsequent query.

[0043] S2, perform data matching and establish a data set;

[0044] To analyze the influence of the process parameters of the cold rolling mill on the plate shape, it is necessary to match the process parameters and the plate shape meter data in time series, that is, to match the process recording data and the plate shape meter data one by one according to the time series.

[0045] The only signal in the process recording data that corresponds to the flatness is the length of the outlet strip. Therefore, the outlet strip length signal is used to match the process recording data and the flatness meter data. Then, the flatness deviation is calculated from the flatness meter data and compared with the flatness deviation signal in the process recording data to verify the degree of matching. If the degree of matching meets the preset degree, a data set is established.

[0046] Take the time series matching of the process record data and the flatness meter data of a steel coil as an example. The matching standard is the length of the exported strip, and the verification standard is the first flatness deviation in the process record data and the second flatness deviation in the flatness meter data. The flatness deviation verification diagram after splicing is as follows: Figure 2 As shown, the unit of flatness deviation is IU.

[0047] S3, using a wavelet transform algorithm based on a genetic algorithm to perform noise reduction on the shape meter data;

[0048] The wavelet transform is a time-frequency localized analysis method that overcomes the short-time Fourier transform's drawback of using a fixed sliding window function, which prevents changes in temporal resolution. While the area of ​​the wavelet window remains constant, its shape can be adaptively modified. When analyzing low-frequency signals, the wavelet transform uses a longer time window, capturing more valid signals; when analyzing high-frequency signals, the window becomes shorter, refining the local signal.

[0049] Wavelet is a time function with a very short duration. When the Fourier transform of the integrable function ψ(t) satisfies the following formula, it is called the basic wavelet, as shown in the following formula:

[0050]

[0051] The set of basic wavelets {ψ a,b (t)} can be obtained by scaling and translating the wavelet ψ(t), as shown in the following formula:

[0052]

[0053] Where b is the translation factor, which determines the position of the function translation; a is the scale factor, which determines the scale of the function expansion and contraction.

[0054] For a signal f(t)∈L 2 (R), its continuous wavelet transform is shown as follows:

[0055]

[0056] Where W f (a, b) are the coefficients obtained by wavelet transform of signal f(t).

[0057] The inverse transform of f(t) can restore the signal as shown below:

[0058]

[0059] Discretize the wavelet and take the scale factor The translation factor is j is the power of the control scale factor. The larger j is, the smaller the scale is and the higher the frequency is. k is the position parameter of the control translation factor. a0 is the step size. is the change in scale of the wavelet function; b0 is the basic scale factor, which is used to determine the basic scale of the wavelet. The step value is a fixed value that is not 1, usually a0>1, then It can be written as follows:

[0060]

[0061] The corresponding discrete wavelet transform can be written as follows:

[0062]

[0063] Wavelet transform can effectively identify signals of various frequencies, remove noise signals and retain the effective information and mutations of the signals. The formula for removing noise using wavelet transform is as follows:

[0064] s(n)=f(n)+σe(n)

[0065] Where s(n) is the actual noisy signal; f(n) is the effective signal; n is the sampling time; σ is the noise intensity; and e(n) is the noise component.

[0066] Using wavelet transform to remove noise requires multiple layers of decomposition of the original signal, thresholding of the wavelet coefficients, and reconstruction using the inverse wavelet transform. The more layers of decomposition, the easier it is to remove the noise signal, but the error after reconstruction will be greater. Therefore, choosing the right number of layers is crucial. The following formula can be used to calculate the right number of layers:

[0067] max_level=log2(l data / l filter )

[0068] Where max_level is the number of decomposition layers; l data is the length of the data; l filter is the length of the filter.

[0069] Threshold processing of wavelet coefficients is the key to improving the denoising effect of wavelet transform, and mainly uses hard threshold function or soft threshold function. Since the soft threshold function can shrink the coefficients that are too large, it can reduce the discontinuity problem that occurs in the hard threshold denoising process. At the same time, it has good adaptability and better denoising effect. Therefore, the soft threshold method is selected, and the expression is shown in the following formula:

[0070]

[0071] Where λ is the threshold; w is the original wavelet coefficient; w λ are the new wavelet coefficients.

[0072] To measure noise reduction performance, the signal-to-noise ratio (SNR) and root mean square error (RMSE) are generally used as evaluation criteria. The signal-to-noise ratio (SNR) is defined as the ratio of the effective component of the reconstructed signal to the noise after noise removal. The larger the ratio, the better the noise reduction performance. The calculation formula is shown below:

[0073]

[0074] Where y(n) is the original signal; x(n) is the signal after noise reduction; and N is the signal length.

[0075] The root mean square error (RMSE) reflects the similarity between the reconstructed signal and the original signal. The smaller the value, the closer it is to the original signal. Its definition is as follows:

[0076]

[0077] When denoising actual data, using wavelet transform alone can lead to signal distortion or limited denoising effects due to improper threshold selection. Therefore, a genetic algorithm can be combined to optimize the threshold of the wavelet coefficients to remove as much noise as possible while retaining the signal characteristics. Selecting an appropriate threshold is the key to wavelet denoising. To obtain the optimal wavelet coefficients, this embodiment combines a genetic algorithm to optimize the threshold and proposes a genetic algorithm-based wavelet transform algorithm (GA-WT). After wavelet transform decomposition obtains wavelet coefficients, the genetic algorithm is used to select an appropriate threshold to remove high-frequency noise and retain useful information. The signal is then reconstructed using an inverse wavelet transform to obtain denoised data. Figure 3 is the denoising process of GA-WT, where cD j is the high-frequency wavelet coefficient obtained by decomposition, cA j is the low-frequency wavelet coefficient obtained by decomposition.

[0078] The commonly used EEMD (Ensemble Empirical Mode Decomposition) method and the wavelet transform method without genetic algorithm optimization were compared with the GA-WT method proposed in this embodiment, with SNR and RMSE as evaluation indicators.

[0079] To build a GA-WT model, we first need to select a suitable wavelet basis function. In this embodiment, we select the db8 basis function, and max_level=log2(l data / l filter ) calculated, the number of decomposition layers is 6. Low-frequency wavelet coefficients are retained, and a genetic algorithm is used to select the optimal threshold for high-frequency wavelet coefficients. The genetic algorithm has an initial population of 200, a crossover probability of 0.8, 50 iterations, and a mutation probability of 0.003.

[0080] Figure 4 The following table compares the denoising effects of the three methods. Figure (a) shows the denoising effect of GA-WT, Figure (b) shows the denoising effect of WT, and Figure (c) shows the denoising effect of EEMD. It can be seen that all three methods can basically restore the signal trend, but the genetic algorithm-based wavelet transform performs better in detail, with smoother curve transitions. The unoptimized wavelet transform and EEMD, on the other hand, have larger fluctuations, fail to remove more noise, and also remove some effective information. The denoising effects of the three methods are shown in Table 1. The SNR of GA-WT is 16.76dB, which is higher than the 11.64dB of the ordinary wavelet transform and 13.46dB of EEMD. The RMSE of GA-WT is 0.51IU, which is lower than the 0.92IU of the ordinary wavelet transform and 0.76IU of EEMD. It can be seen that GA-WT has better denoising performance than the other two methods. Therefore, GA-WT was selected for denoising the measured flatness data.

[0081] Table 1

[0082]

[0083] S4. Decomposing and reconstructing the shape meter data based on Legendre orthogonal polynomials;

[0084] The shape meter data after the noise reduction process in step S3 is decomposed based on the first wave basic mode, the second wave basic mode, the third wave basic mode and the fourth wave basic mode by using Legendre orthogonal polynomials. The first wave basic mode, the second wave basic mode, the third wave basic mode and the fourth wave basic mode are represented by δ1, δ2, δ3 and δ4 respectively. The formulas are as follows: δ1(x)=x, After decomposition, the coefficients of each basis mode are obtained as λ1, λ2, λ3, and λ4.

[0085] like Figure 5 As shown in the figure, the basic modes of Legendre orthogonal polynomials are used to represent the common wave shapes of plate defects, where Y1 is the right wave, Y2 is the left wave, Y3 is the double-sided wave, Y4 is the middle wave, Y5 is the left three-point wave, Y6 is the right three-point wave, Y7 is the side-middle compound wave, and Y8 is the quarter wave. The basic modes of various common wave shapes include:

[0086] Right wave: Y1=δ1(x)=x;

[0087] Left wave: Y2=-δ1(x)=-x;

[0088] Double wave:

[0089] Middle Wave:

[0090] Left three-point wave:

[0091] Right three-point wave:

[0092] Side-middle compound wave:

[0093] Quarter Wave:

[0094] Based on the coefficients λ1, λ2, λ3, and λ4 of each basic mode obtained by decomposition, the plate shape is reconstructed through Δflat=λ1δ1+λ2δ2+λ3δ3+λ4δ4 to realize filtering processing of the plate shape meter data.

[0095] S5. Using an adaptive moment estimation algorithm to intelligently identify the control efficacy of the control mechanism, and using a curve graph to intuitively describe the control efficacy;

[0096] In this example, the objective function for intelligent identification of plate shape control efficacy is defined as follows:

[0097]

[0098] Where w j is the control quantity of the jth actuator; eff ji is the control efficiency of the jth actuator in the i-th section of the bandwidth direction; n is the number of flatness deviation values; Δflat i is the flatness deviation of the i-th segment in the bandwidth direction.

[0099] In order to minimize the objective function J(eff), the gradient matrix g at the current moment is calculated by the following formula: t :

[0100]

[0101] The adaptive moment estimation algorithm is a variant of the gradient descent method, which makes up for the deficiency of the stochastic gradient descent method that the learning rate cannot be changed. The adaptive moment estimation algorithm uses the first-order and second-order moment estimation of the gradient to automatically update the learning rate according to the characteristics of the data. The current gradient calculated by the above formula is g t , then the first-order moment estimate M t As shown in the following formula:

[0102] M t =β1·M t-1 +(1-β1)·g t

[0103] Where t is the current moment; β1∈[0,1) is the exponential decay rate of the moment, which is generally 0.9; M t-1 is the first-order moment estimate of the previous moment; g t is the gradient at the current moment.

[0104] Second-order moment estimate V t As shown in the following formula:

[0105] V t =β2·V t-1 +(1-β2)·g t 2

[0106] Where V t-1 is the second-order moment estimate of the previous moment; β2∈[0,1) is the exponential decay rate of the second-order moment, which is generally taken as 0.999.

[0107] When the first-order and second-order moment estimates are initialized to zero vectors, the initialization step size will be too large, so they need to be corrected for the deviation. Take the second-order moment as an example to calculate the deviation correction term. Since Vt Approaching 0, the second-order moment estimation formula can be expanded to obtain the following formula:

[0108]

[0109] Where, is the exponential decay rate of the second-order moment at the previous moment; g i 2 is the gradient at the i-th moment.

[0110] Taking the expectation on both sides of the above equation, we can get the following equation:

[0111]

[0112] The second-order moment V is estimated according to the following formula t To perform bias correction:

[0113]

[0114] Where, β2 t is the exponential decay rate of the second-order moment at the current moment.

[0115] The deviation correction process is the same as that of the second-order moment estimation. The following formula is used to estimate the first-order moment M: t To perform bias correction:

[0116]

[0117] Where, β1 t is the first-order moment exponential decay rate at the current moment.

[0118] In this embodiment, the variable x is defined as the control efficacy matrix, and the update of the variable x is as follows:

[0119]

[0120] Where η∈(0,1] is the learning rate; ε is a very small constant, which can be taken as 1×10 -8 , used to avoid the second-order moment estimation V in the calculation process t When x is zero; t =eff t represents the updated regulatory efficacy matrix, x t-1 =eff t-1 It represents the control efficacy matrix before updating. The above content can be used to realize the iterative update of the control efficacy matrix.

[0121] Since the adjustment amount of the unit's roller shifting is very small and the response is slow, it is impossible to solve the control effect of roller shifting. For the processed data, the adaptive moment estimation algorithm is used to calculate the control effect of Tilt, WRB and IRB respectively, and compared with the control effect provided on site. The identification results are as follows: Figure 6 As shown in the figure, the control efficiency obtained by adaptive moment estimation and identification are similar to the on-site control efficiency curve in terms of trend and value range. The control performance of each mechanism is consistent with empirical knowledge, but there are slight differences in details.

[0122] Starting from actual production data from industrial sites, this method performs data preprocessing operations such as outlier processing and data matching on process record data and shape meter data to create a data set. It then performs noise reduction on the shape meter data and uses an adaptive moment estimation algorithm to intelligently identify the control efficacy of the processed data set. This allows for analysis and clarification of the control mechanism's control capabilities, providing more control information for subsequent shape control calculations.

[0123] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A data-driven intelligent identification method for plate shape control efficacy, characterized in that: include: Acquiring industrial data during a production process of a rolling mill, preprocessing the industrial data, and establishing a data set based on the preprocessed industrial data, wherein the industrial data includes process record data and shape meter data, wherein the shape meter data includes a strip residual stress deviation and a strip length signal in each measurement area; Performing noise reduction and decomposition and reconstruction on the shape meter data in the data set, and performing intelligent identification of control efficacy on the noise reduction and decomposition and reconstruction data set using an adaptive moment estimation algorithm; Preprocessing the industrial data includes: processing the residual stress deviation of the strip steel and converting it into a plate shape deviation; processing the process record data for abnormal values ​​and filling the null values. If the value before the null value is valid, the null value is replaced by the value before the null value; if the null value is in the first row, the first non-null value queried backward is filled into the null value row; Building a dataset based on preprocessed industrial data includes: Based on the outlet strip length, the pre-processed process record data and the shape meter data are matched in time series to construct the data set; The process of performing the corresponding matching includes: respectively obtaining a first flatness deviation in the process record data and a second flatness deviation in the flatness meter data, and performing the corresponding matching based on a comparison and verification of the first flatness deviation and the second flatness deviation; The noise reduction processing of the shape meter data in the data set includes: The wavelet transform algorithm is used to perform noise reduction on the shape meter data, and the formula is: ； Where, is the actual noisy signal; is a valid signal; is the sampling time; is the noise intensity; is the noise component; The noise reduction process of the shape meter data using the wavelet transform algorithm includes: The shape meter data is subjected to multi-layer decomposition of the original signal, using the db8 basis function, and the Calculate the number of layers of the multi-layer decomposition, where max_level is the number of decomposition layers, l data is the length of the data, l filter is the length of the filter; after determining the number of decomposition layers, the low-frequency wavelet coefficients are retained, and the threshold of the high-frequency wavelet coefficients is optimized using a genetic algorithm. The initial population of the genetic algorithm is 200, the crossover probability is 0.8, the number of iterations is 50, and the mutation probability is 0.003; finally, the inverse wavelet transform is used to reconstruct the signal to obtain the denoised flatness meter data.

2. The data-driven intelligent identification method for plate shape control efficacy according to claim 1 is characterized in that: Decomposing and reconstructing the shape meter data in the data set includes: The shape meter data is decomposed based on Legendre orthogonal polynomials to obtain the coefficients of each shape basis mode , , and ; Based on the coefficients of each plate-shaped basis mode , , and The shape meter data is reconstructed to obtain decomposed and reconstructed shape meter data.

3. The data-driven intelligent identification method for plate shape control efficacy according to claim 1, characterized in that: The intelligent identification of the control effect of the noise reduction and decomposition and reconstruction data set using the adaptive moment estimation algorithm includes: Based on the adaptive moment estimation algorithm, the first-order and second-order moment estimates of the gradient are used to automatically update the learning rate according to the data characteristics of the data set, and the first-order and second-order moment estimates are corrected for deviations. An updated calculation model for the variable x is defined based on the deviation corrections of the first-order and second-order moment estimates, and the variable x is updated and calculated based on the updated calculation model of the variable x to achieve intelligent identification of the regulation efficacy, wherein the variable x is a regulation efficacy matrix.

4. The data-driven intelligent identification method for plate shape control efficacy according to claim 3 is characterized in that: The update calculation model of the variable x is: ； Where x t is the x variable at the current moment; x t-1 is the x variable at the previous moment; is the learning rate; is a constant; is the bias correction for the first-order moment estimate; is the bias correction for the second moment estimate.

Images