Hot galvanizing air knife control method and system

Abstract

The invention discloses a hot galvanizing air knife control method. The method comprises the following steps: S1, data acquisition; s2, data processing; s3, model screening; s4, numerical value derivation; s5, constructing a model; s6, air knife control; s7, adjusting an air knife; and S8, performing real-time feedback. The invention further discloses a hot galvanizing air knife control system. The invention solves the technical problem that the method for eliminating the thickness deviation of the plating layer by adjusting the distance of the air knife on one side has great difficulty because the central line of the strip does not coincide with the mechanical zero point of the air knife.

Description

Technical Field

[0001] The invention relates to an automatic control technology for hot-dip galvanizing, and more specifically to a hot-dip galvanizing air knife control method and system. Background Art

[0002] Hot-dip galvanizing, also known as hot-dip galvanizing or hot-dip galvanizing, is an effective anti-corrosion method used in the production and processing of strips. The thickness of the zinc layer and whether its distribution is uniform are important technical indicators for measuring the quality of the strip. Too thick a coating will not only increase the cost, but also affect the adhesion, spot weldability, anti-powdering and other performance of the strip; too thin a coating will make the strip's corrosion resistance insufficient. Therefore, the galvanizing control technology is directly related to the product quality, cost control and market competitiveness of hot-dip galvanized sheets.

[0003] The distance between the air knife and the strip is one of the main control variables that affect the thickness of the zinc layer in the hot-dip galvanizing process. The zinc layer thickness is a multivariate nonlinear model. It is difficult to accurately predict the zinc layer thickness. At present, the commonly used control method is a combination of feedforward and feedback. The air knife distance is judged by the feedback of the zinc layer weight by the thickness gauge. In the actual production process, due to the influence of factors such as strip shape, work roll wear, and mechanical errors, the thickness of the coating on the strip surface will fluctuate greatly even under stable working conditions. Since the center line of the strip does not coincide with the mechanical zero point of the air knife (that is, the measured distance of the side air knife is not equal to the actual distance between it and the center line of the strip), it is very difficult to eliminate the coating thickness deviation by adjusting the air knife distance. Summary of the invention

[0004] In view of the defects existing in the prior art, the purpose of the present invention is to provide a hot-dip galvanizing air knife control method and system to solve the technical problem that the method of eliminating the coating thickness deviation by adjusting the single-side air knife distance is very difficult because the center line of the strip and the mechanical zero point of the air knife do not coincide with each other.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A first aspect of the present invention provides a hot dip galvanizing air knife control method, comprising the following steps:

[0007] S1, data collection;

[0008] S2, data processing;

[0009] S3, model screening;

[0010] S4, numerical derivation;

[0011] S5, model construction;

[0012] S6, air knife control;

[0013] S7, air knife adjustment;

[0014] S8. Real-time feedback.

[0015] Preferably, the data collection in step S1 specifically includes:

[0016] The process parameters of the strip are measured by a data acquisition device, and then the measured data are stored in a remote processing host, which sets a prediction value according to the process parameters of the strip.

[0017] Preferably, the data processing in step S2 specifically includes:

[0018] The input data of the process parameters of the strip in the remote processing host is X∈Rn×m, and the output data is Y∈Rn×p, and some of the process parameters of the strip are randomly selected to form a training set and a test set, and finally the input data and output data of the strip are standardized;

[0019] N is the number of samples, m is the sample dimension, and p is the sample dimension.

[0020] Preferably, the model screening in step S3 specifically includes:

[0021] A PLS model is constructed, and the input data X and output data Y of the strip are subjected to PLS decomposition. The PLS model is solved according to a nonlinear iterative partial least squares model to obtain a score matrix, and the screened score matrix is ​​used as the input of a soft sensor model for data prediction.

[0022] Preferably, the numerical derivation in step S4 specifically includes:

[0023] The data acquisition device collects the observed values ​​of the strip and transmits them to the remote processing host. The observed values ​​and predicted values ​​in the remote processing host are combined with the prior distribution through the Bayesian formula, and the training set and the test set are added to form a covariance matrix to obtain the predicted value of Gaussian process regression.

[0024] Preferably, the model building in step S5 specifically includes:

[0025] The remote processing host establishes a GPR model between the score matrix and the output data Y of the strip, constructs different Gaussian process regression models by selecting and combining different covariance functions, and performs standardization processing. The partial derivative of the hyperparameters is obtained by the maximum likelihood method, and the conjugate gradient method is used to obtain the optimal solution of the hyperparameters. After the hyperparameters are obtained, the root mean square error corresponding to different models is calculated according to the predicted value and the true value of the Gaussian process regression.

[0026] Preferably, the air knife control in step S6 specifically includes:

[0027] The remote processing host determines whether the latest root mean square error of the process parameters of the strip exceeds the standard deviation of the deviation correction. If not, the change in the air knife distance is set to 0.

[0028] Preferably, the step S7 of adjusting the air knife specifically includes:

[0029] If the remote processing host determines that the latest root mean square error of the process parameters of the strip exceeds the standard deviation of the deviation correction, the remote processing host optimizes and calculates the optimal air knife distance in real time based on the root mean square error of the zinc layer thickness prediction, thereby controlling the change in the air knife distance from the strip based on the real-time optimization result.

[0030] Preferably, the real-time feedback in step S8 specifically includes:

[0031] The remote processing host performs prediction and feedforward correction based on the root mean square error and pre-parameter real-time optimization.

[0032] Preferably, in the data collection of step S1, the remote processing host cleans, processes missing values ​​and performs time alignment on the process parameters of the strip.

[0033] Preferably, in the step S3 model screening, solving the PLS model according to the nonlinear iterative partial least squares model is specifically as follows:

[0034] Perform PLS decomposition on the input data X and output data Y, solve the PLS model according to the nonlinear iterative least squares method, and obtain T∈Rn×d. The screened score matrix is ​​used as the input of the model for data prediction. The partial least squares model decomposes the input data X and output data Y as follows:

[0035]

[0036] Where T∈Rn×d is the score matrix; P∈Rm×d and Q∈Rp×d are the loading matrices of X and Y respectively; E and F are the residual matrices of X and Y respectively, and d is the number of PLS ​​latent variables.

[0037] Preferably, in the model construction of step S5, the covariance function includes a square exponential covariance function SE, a linear covariance function L, and a periodic covariance function P.

[0038] A second aspect of the present invention provides a hot dip galvanizing air knife control system, comprising:

[0039] A data acquisition device for measuring the process parameters of the strip;

[0040] A remote processing host, used to store the process parameters of the strip, and to divide the information of the strip surface obtained by the data acquisition device into regions, and to detect the roughness of the strip information in different regions;

[0041] The hot-dip galvanizing air knife control system realizes the hot-dip galvanizing air knife control method.

[0042] Preferably, the data acquisition device comprises a distance sensor, a cold zinc layer measurement and a strip thickness gauge which are sequentially arranged along the advancing direction of the strip;

[0043] The distance sensor is used to sense the surface position of the distance sensor and the strip material, and transmit the data to the remote processing host to evaluate the plate shape of the strip material;

[0044] The cold zinc layer measuring instrument is used to detect the zinc layer thickness of the strip and transmit the data information to the remote processing host;

[0045] The strip thickness gauge is used to measure the thickness of the strip.

[0046] Preferably, the remote processing host comprises:

[0047] A processing module, for optimizing the measured values ​​of the process parameters of the strip, and calculating in real time whether the root mean square error exceeds the standard deviation of the deviation correction;

[0048] The control module is used to receive the standard deviation of the deviation correction fed back by the processing module, and adjust the process parameters of the strip by correcting the standard deviation.

[0049] The hot-dip galvanizing air knife control method and system provided by the present invention have the following beneficial effects:

[0050] 1) The present invention simultaneously extracts features of input data and output data through PLS, so that the quality-related characteristics of the output end are maintained while the input data is screened. Gaussian process models are established by selecting and combining covariance functions, and a reasonable air knife distance is calculated to obtain air knife pressure and air knife distance setting values ​​that meet the current control accuracy and have an adjustment margin.

[0051] 2) The present invention can quickly respond to the process parameter deviation of the strip caused by the working condition deviation of the production line or other interference factors, ensure the uniformity of the front and rear coatings, and quickly optimize the air knife distance, eliminate the process parameter deviation of the strip caused by the working condition deviation and the change of strip thickness, ensure the adjustment margin of the air knife pressure, and ensure the longitudinal uniformity of the coating thickness. While improving the production line capacity, it can still ensure that the product quality meets the index requirements and reduce the production of unqualified products.

[0052] 3) The remote processing host used in the present invention can accurately optimize the air knife distance when the front and rear coating thicknesses deviate, so as to quickly and effectively eliminate the deviation in the front and rear coating thicknesses, thereby ensuring the uniformity of the coating thickness. At the same time, it can effectively ensure that the air knife pressure is away from the saturation zone, increase the adjustment space of the air knife pressure, ensure the surface quality of the coating, and ensure that the product quality meets the index requirements while improving the production line capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic flow chart of the hot dip galvanizing air knife control method of the present invention;

[0054] Figure 2 It is a schematic flow diagram of model screening in the hot dip galvanizing air knife control method of the present invention;

[0055] Figure 3 It is a schematic diagram of the framework structure of the hot dip galvanizing air knife control system of the present invention. DETAILED DESCRIPTION

[0056] In order to better understand the above technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0057] Combination Figure 1 As shown, a hot dip galvanizing air knife control method provided by the present invention comprises the following steps:

[0058] S1. Data acquisition: The process parameters of the strip are measured by a data acquisition device, and then the measured data are stored in a remote processing host, which sets the prediction value according to the process parameters of the strip.

[0059] S2, data processing, the input data of the process parameters of the strip in the remote processing host is X∈Rn×m, the output data is Y∈Rn×p, and the process parameters of some strips are randomly selected to form a training set and a test set, and finally the input data and output data of the strip are standardized;

[0060] N is the number of samples, m is the sample dimension, and p is the sample dimension.

[0061] S3, model screening, constructing the PLS model, performing PLS decomposition on the input data X and output data Y of the strip, solving the PLS model according to the nonlinear iterative partial least squares model, and obtaining the score matrix. The screened score matrix is ​​used as the input of the soft sensor model for data prediction.

[0062] S4, numerical derivation, the data acquisition device collects the observed values ​​of the strip and transmits them to the remote processing host. The observed values ​​and predicted values ​​in the remote processing host are combined with the prior distribution through the Bayesian formula, and the training set and the test set are added to form a covariance matrix to obtain the predicted value of Gaussian process regression.

[0063] S5. Model construction: The remote processing host establishes a GPR model between the score matrix and the output data Y of the strip. Different Gaussian process regression models are constructed by selecting and combining different covariance functions, and standardized. The partial derivative of the hyperparameters is obtained by the maximum likelihood method, and the conjugate gradient method is used to obtain the optimal solution of the hyperparameters. After obtaining the hyperparameters, the root mean square error corresponding to different models is calculated based on the predicted value and the true value of the Gaussian process regression, such as Figure 2 shown.

[0064] S6, air knife control, the remote processing host determines whether the root mean square error of the process parameters of the latest strip exceeds the standard deviation of the deviation correction. If not, the air knife distance change is set to 0.

[0065] S7, air knife adjustment. If the remote processing host determines that the root mean square error of the process parameters of the latest strip exceeds the standard deviation of the deviation correction, the remote processing host will optimize and calculate the optimal air knife distance in real time based on the root mean square error of the zinc layer thickness prediction, thereby controlling the change in the air knife distance from the strip according to the real-time optimization result.

[0066] S8, real-time feedback, the remote processing host performs prediction and feedforward correction based on the root mean square error and pre-parameter real-time optimization.

[0067] In the above step S1 data collection, the remote processing host cleans the process parameters of the strip, processes missing values, and performs time alignment.

[0068] In the above step S3 model screening, the PLS model is solved according to the nonlinear iterative partial least squares model as follows:

[0069] Perform PLS decomposition on the input data X and output data Y, solve the PLS model according to the nonlinear iterative least squares method, and obtain T∈Rn×d. The screened score matrix is ​​used as the input of the model for data prediction. The partial least squares model decomposes the input data X and output data Y as follows:

[0070]

[0071] Where T∈Rn×d is the score matrix; P∈Rm×d and Q∈Rp×d are the loading matrices of X and Y respectively; E and F are the residual matrices of X and Y respectively, and d is the number of PLS ​​latent variables.

[0072] In the above step S5 of model construction, the covariance function includes the square exponential covariance function SE, the linear covariance function L, and the periodic covariance function P.

[0073] Combination Figure 3 As shown, the present invention also provides a hot dip galvanizing air knife control system, comprising:

[0074] A data acquisition device 1, used to measure the process parameters of the strip;

[0075] The remote processing host 2 is used to store the process parameters of the strip, and to divide the information of the strip surface into regions through the data acquisition device 1, and to detect the roughness of the strip information in different regions;

[0076] The hot-dip galvanizing air knife control system of the present invention realizes the hot-dip galvanizing air knife control method of the present invention.

[0077] The data acquisition device 1 includes a distance sensor, a cold zinc layer measurement and a strip thickness gauge which are sequentially arranged along the advancing direction of the strip;

[0078] The distance sensor is used to sense the surface position of the distance sensor and transmit the data to the remote processing host 2 for evaluating the strip shape;

[0079] The cold zinc layer measuring instrument is used to detect the zinc layer thickness of the strip and transmit the data information to the remote processing host 2;

[0080] Strip thickness gauge is used to measure the thickness of strip.

[0081] The remote processing host 2 includes:

[0082] Processing module 201 is used to optimize the measured values ​​of the process parameters of the strip and calculate in real time whether the root mean square error exceeds the standard deviation of the deviation correction;

[0083] The control module 202 is used to receive the standard deviation of the deviation correction fed back by the processing module 201, and adjust the process parameters of the strip by correcting the standard deviation.

[0084] The Gaussian process regression soft sensor modeling method based on partial least squares is adopted in the present invention. Partial least squares PLS is used to complete the small range selection of the number of measured variables, so that the variables have a higher correlation with the dominant variables while reducing the dimension. Then, a Gaussian process regression model GPR is constructed in combination with different covariance functions, so as to provide a comparison of different models and achieve optimal prediction.

[0085] Example

[0086] When setting up the measuring instrument for measuring the strip, a distance sensor, a cold zinc layer measurement and a strip thickness gauge are sequentially set along the forward direction of the strip. The distance sensor measures the data of the strip before galvanizing, the cold zinc layer measurement measures the data of the air knife adjustment after galvanizing, and the strip thickness gauge measures the thickness of the strip. By starting the distance sensor, the cold zinc layer measurement and the strip thickness gauge in sequence, the measured values ​​are sequentially transmitted to the remote processing host, so that the remote processing host can obtain the position information and surface information of the strip in real time. The detected data is processed in real time by the processing module in the remote processing host, a regression model is established by Gaussian, and the Gaussian regression model is trained, so that the processing module can optimize the hysteresis of the data measuring device in actual use and reduce the error in actual use.

[0087] refer to Figures 1 to 3 As shown, the hot dip galvanizing air knife control method of this embodiment specifically includes the following steps:

[0088] S01, data acquisition, the process parameters of the strip are measured by the data acquisition device 1, and then the measured data are stored in the remote processing host 2. The remote processing host 2 sets the prediction number according to the process parameters of the strip. The remote processing host 2 pre-sets the pre-parameters as the corrected standard deviation range, whose mean is 0 and the standard deviation is 1. The remote processing host 2 includes a processing module 201 and a control module 202, and the remote processing host 2 receives the data transmitted by the data acquisition device 1, divides the acquired information on the surface of the strip by the data acquisition device 1, and detects the roughness of the strip information in different areas. The data acquisition device 1 includes a distance sensor, a cold zinc layer measurement and a strip thickness gauge arranged in sequence along the forward direction of the strip. The distance sensor is used to sense its position with the strip surface and transmit the data to the remote processing host 2 to evaluate the strip plate shape; the cold zinc layer measurement instrument is used to measure the zinc layer thickness of the strip and transmit the data information to the remote processing host 2; the strip thickness gauge is used to measure the thickness of the strip. The processing module 201 is used to optimize the coating thickness measurement value in the remote processing host 2, calculate in real time whether the root mean square error exceeds the standard deviation of the deviation correction, and transmit the processing result to the control module 202. The processing module 201 also includes cleaning of the coating thickness measurement value in the remote processing host 2, missing value processing and timing alignment; the control module 202 is used to receive the standard deviation of the deviation correction fed back by the processing module 201, and adjust the air knife by correcting the standard deviation.

[0089] S02, data processing, the processing module 201 inputs the process parameter data of the strip in the remote processing host 2 as X∈Rn×m, and outputs the process parameter data of the strip as Y∈Rn×p, and randomly selects part of the data to form a training set and a test set, and finally completes the standardization of the input data and the output data. The formula is X∈Rn×m, n represents the number of samples, m represents the sample dimension, and the formula is Y∈Rn×p, p represents the sample dimension.

[0090] S03, model screening, constructing a PLS model, performing PLS decomposition on the input data X and output data Y, solving the PLS model according to the nonlinear iterative least squares model, and obtaining a score matrix. The screened score matrix is ​​used as the input of the soft sensor model for data prediction. The partial least squares model decomposes the input data X and output data Y as follows:

[0091] Perform PLS decomposition on the input data X and output data Y, solve the PLS model according to the nonlinear iterative least squares method, and obtain T∈Rn×d. The screened score matrix is ​​used as the input of the model for data prediction. The partial least squares model decomposes the input data X and output data Y as follows:

[0092]

[0093] Where T∈Rn×d is the score matrix; P∈Rm×d and Q∈Rp×d are the loading matrices of X and Y respectively; E and F are the residual matrices of X and Y respectively, and d is the number of PLS ​​latent variables.

[0094] S04, numerical derivation, the data acquisition device 1 collects the observed values ​​of the strip process parameters and transmits them to the remote processing host 2, the processing module 201 combines the observed values ​​and predicted values ​​in the remote processing host 2 with the prior distribution through the Bayesian formula, and adds the training set and the test set to form a covariance matrix to obtain the predicted value of Gaussian process regression.

[0095] S05, model construction, processing module 201 establishes a GPR model between the score matrix and the output data Y: different Gaussian process regression models are constructed by selecting and combining the square exponential covariance function SE, the linear covariance function L, and the periodic covariance function P, and standardized, the partial derivatives of the hyperparameters are obtained by the maximum likelihood method, and the optimal solution of the hyperparameters is obtained by the conjugate gradient method. After the hyperparameters are obtained, the root mean square error corresponding to different models is calculated according to the predicted value and the true value of the Gaussian process regression, and different Gaussian process regression models are constructed by selecting and combining different covariance functions. By obtaining the predicted value of the Gaussian process regression, the Gaussian process regression model is constructed by different covariance functions, and the covariance functions are combined in an cumulative manner. The set of hyperparameters is obtained by the maximum likelihood method.

[0096] S06, air knife control, the control module 202 determines whether the root mean square error of the latest process parameters of the strip exceeds the standard deviation range pre-set by the pre-parameters. If it does not exceed the standard deviation range of the process parameter deviation correction of the strip, the air knife distance change is set to 0. The air knife distance, air knife pressure, production line speed and coating thickness deviation of the zinc layer thickness prediction value are optimized and calculated in real time to obtain the optimal air knife distance, thereby determining the air knife distance change, wherein the deviation amount of the front coating thickness that needs to be changed, the deviation amount of the rear coating thickness that needs to be changed, and the coating thickness that needs to be compensated are: CWm=NN(D(t), P(t), S(t))-NN(D(t), Pini, S(t)) wherein D(t) is the air knife distance at time t, P(t) is the air knife pressure at time t, S(t) is the production line speed at time t, and Pini is the air knife pressure after adding the standard deviation; if the standard deviation of the deviation correction is not exceeded, the change in the air knife distance is set to 0. The coating deviation optimization method adopted is not possessed by the traditional feedback control system which mainly uses air knife pressure and supplemented by total air knife distance. Specifically, this embodiment takes into account the problem of strip coating deviation caused by working condition offset or other interference factors, and adopts air knife distance optimization to control the air knife distance, which can ensure the consistency of the coating thickness on the strip surface and improve the control quality.

[0097] S07, air knife adjustment, the control module 202 determines that if the standard deviation range of the pre-parameter is pre-set, then the optimal air knife distance is calculated in real time according to the root mean square error of the zinc layer thickness prediction, so as to control the process parameter change amount according to the real-time optimization result to correct the standard deviation. When the optimal air knife distance is optimized and calculated, the optimal air knife distance is optimized and calculated using the real-time optimization module. The real-time optimization module takes minimizing the deviation between the coating thickness prediction value and the control target as the goal, takes the air knife distance operation range as the constraint, and iteratively searches for the optimal air knife distance under the given air knife pressure, production line speed and coating thickness setting value. The traditional feedback PID needs to be controlled multiple times according to the feedback deviation. Due to the large measurement hysteresis of the galvanizing production system, the adjustment process of the coating thickness deviation is very slow and the control effect is poor. The present invention takes the deviation between the coating thickness and the control target as the goal, takes the process specification of the air knife distance as the constraint, and uses the neural network prediction model to iteratively optimize the air knife distance, which can eliminate the deviation between the coating thickness and the control target at one time, and can quickly and accurately ensure the uniformity of the coating thickness.

[0098] S08, real-time feedback: the processing module 201 performs real-time optimization based on the root mean square error and pre-parameters, and transmits the data to the remote processing module 202 to carry out prediction and feedforward correction.

[0099] Those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A hot dip galvanizing air knife control method, characterized in that: The following steps are involved: S1, data collection; S2, data processing; S3, model screening; S4, numerical derivation; S5, model construction; S6, air knife control; S7, air knife adjustment; S8. Real-time feedback.

2. The hot dip galvanizing air knife control method according to claim 1, characterized in that: The step S1 data collection specifically includes: The process parameters of the strip are measured by a data acquisition device, and then the measured data are stored in a remote processing host, which sets a prediction value according to the process parameters of the strip.

3. The hot dip galvanizing air knife control method according to claim 2, characterized in that: The data processing in step S2 specifically includes: The input data of the process parameters of the strip in the remote processing host is X∈Rn×m, and the output data is Y∈Rn×p, and some of the process parameters of the strip are randomly selected to form a training set and a test set, and finally the input data and output data of the strip are standardized; N is the number of samples, m is the sample dimension, and p is the sample dimension.

4. The hot dip galvanizing air knife control method according to claim 3, characterized in that: The model screening step S3 specifically includes: A PLS model is constructed, and the input data X and output data Y of the strip are subjected to PLS decomposition. The PLS model is solved according to a nonlinear iterative partial least squares model to obtain a score matrix, and the screened score matrix is ​​used as the input of a soft sensor model for data prediction.

5. The hot dip galvanizing air knife control method according to claim 4, characterized in that: The numerical derivation in step S4 specifically includes: The data acquisition device collects the observed values ​​of the strip and transmits them to the remote processing host. The observed values ​​and predicted values ​​in the remote processing host are combined with the prior distribution through the Bayesian formula, and the training set and the test set are added to form a covariance matrix to obtain the predicted value of Gaussian process regression.

6. The hot dip galvanizing air knife control method according to claim 5, characterized in that: The step S5 model building specifically includes: The remote processing host establishes a GPR model between the score matrix and the output data Y of the strip, constructs different Gaussian process regression models by selecting and combining different covariance functions, and performs standardization processing. The partial derivative of the hyperparameters is obtained by the maximum likelihood method, and the conjugate gradient method is used to obtain the optimal solution of the hyperparameters. After the hyperparameters are obtained, the root mean square error corresponding to different models is calculated according to the predicted value and the true value of the Gaussian process regression.

7. The hot dip galvanizing air knife control method according to claim 6, characterized in that: The step S6 of air knife control specifically includes: The remote processing host determines whether the latest root mean square error of the process parameters of the strip exceeds the standard deviation of the deviation correction. If not, the change in the air knife distance is set to 0.

8. The hot dip galvanizing air knife control method according to claim 7, characterized in that: The step S7 of air knife adjustment specifically includes: If the remote processing host determines that the latest root mean square error of the process parameters of the strip exceeds the standard deviation of the deviation correction, the remote processing host optimizes and calculates the optimal air knife distance in real time based on the root mean square error of the zinc layer thickness prediction, thereby controlling the change in the air knife distance from the strip based on the real-time optimization result.

9. The hot dip galvanizing air knife control method according to claim 8, characterized in that: The step S8 real-time feedback specifically includes: The remote processing host performs prediction and feedforward correction based on the root mean square error and pre-parameter real-time optimization.

10. The hot dip galvanizing air knife control method according to claim 2, characterized in that: In the data collection step S1, the remote processing host cleans, processes missing values, and performs time alignment on the process parameters of the strip.

11. The hot dip galvanizing air knife control method according to claim 4, characterized in that: In the step S3 model screening, solving the PLS model according to the nonlinear iterative partial least squares model is specifically as follows: Perform PLS decomposition on the input data X and output data Y, solve the PLS model according to the nonlinear iterative least squares method, and obtain T∈Rn×d. The screened score matrix is ​​used as the input of the model for data prediction. The partial least squares model decomposes the input data X and output data Y as follows: Where T∈Rn×d is the score matrix; P∈Rm×d and Q∈Rp×d are the loading matrices of X and Y respectively; E and F are the residual matrices of X and Y respectively, and d is the number of PLS ​​latent variables.

12. The hot dip galvanizing air knife control method according to claim 6, characterized in that: In the model construction of step S5, the covariance function includes a square exponential covariance function SE, a linear covariance function L, and a periodic covariance function P.

13. A hot dip galvanizing air knife control system, characterized in that: include: A data acquisition device for measuring the process parameters of the strip; A remote processing host, used to store the process parameters of the strip, and to divide the surface information of the strip into regions through the data acquisition device, and to detect the roughness of the strip information in different regions; The hot-dip galvanizing air knife control system implements the hot-dip galvanizing air knife control method as described in any one of claims 1-12.

14. The hot dip galvanizing air knife control system according to claim 13, characterized in that: The data acquisition device includes a distance sensor, a cold zinc layer measurement and a strip thickness gauge which are sequentially arranged along the advancing direction of the strip; The distance sensor is used to sense the surface position of the distance sensor and the strip material, and transmit the data to the remote processing host to evaluate the plate shape of the strip material; The cold zinc layer measuring instrument is used to detect the zinc layer thickness of the strip and transmit the data information to the remote processing host; The strip thickness gauge is used to measure the thickness of the strip.

15. The hot dip galvanizing air knife control system according to claim 13, characterized in that: The remote processing host comprises: A processing module, used for optimizing the measured values ​​of the process parameters of the strip, and calculating in real time whether the root mean square error exceeds the standard deviation of the deviation correction; The control module is used to receive the standard deviation of the deviation correction fed back by the processing module, and to adjust the process parameters of the strip by correcting the standard deviation.

Images