A composite plate hot rolling speed-deformation-pressure coupling control system and method

Abstract

The application discloses a kind of composite board hot rolling speed-deformation-pressure coupling control system and method, based on multilayer composite board hot rolling data, using long short-term memory network establishes space-time sequence prediction model, and according to the deviation of prediction quality and target quality, using multi-objective balanced optimization algorithm to feed forward correction is carried out to roll gap.On the basis of feed forward correction, the residual between thickness actual value and set outlet thickness is further considered, and a roll gap feedback adjustment strategy is designed, realizing the coupling control of composite board hot rolling speed-deformation-pressure.The composite board hot rolling speed-deformation-pressure coupling control method proposed in the application has high control precision, makes up for the defects of traditional methods, improves the speed-deformation-rolling force dynamic coupling control, and can be widely used in multilayer composite board hot rolling.

Description

Technical Field

[0001] This invention belongs to the field of hot rolling technology of metal materials, specifically relating to a composite plate hot rolling speed-deformation-pressure coupling control system and method, which is suitable for improving the accuracy and stability of hot rolling forming of multi-layer composite plates. Background Technology

[0002] Hot rolling of composite plates is a core technology that achieves metallurgical bonding between dissimilar metal layers of the substrate and the composite material through hot rolling process. The substrate and the composite material have significant differences in material properties such as thermal expansion coefficient and deformation resistance. Effectively achieving coupled control of hot rolling speed, deformation amount and pressure of composite plates can ensure that the substrate and the composite material extend synchronously during rolling, avoiding quality problems such as warping, edge cracking or uneven thickness.

[0003] Currently, preliminary progress has been made in the hot rolling process control of multilayer composite plates. "CN113275381A" proposes a hot rolling composite method for preparing metal composite plates using large plastic deformation; "CN113627469A" proposes a method for predicting the crown of hot-rolled strip steel based on fuzzy inference algorithms, solving the technical problem of low temperature control accuracy in existing hot-rolled strip steel; "CN116371941B" proposes a method, device, and electronic equipment for predicting rolling force and layer thickness of metal composite plates. Based on the reduction rate of the soft metal slab, the exit thickness of the soft metal slab and the hard metal slab are determined, and the rolling force of the soft metal plate and the hard metal plate are calculated separately. The average of the two rolling forces is used as the target rolling force. The paper "CN115041529B" proposes a monitoring method, device, medium, and electronic equipment for hot-rolled thickness control models. It monitors the hot-rolled thickness control model for problems using self-learning coefficients generated by the self-learning module, allowing for timely adjustments. The paper "CN112845613A" proposes a thickness control method, device, and terminal equipment for hot-rolled strip steel. Based on the characteristics of the raw materials and rolling system during the rolling process, it optimizes the rolling signal. Through eccentricity compensation, the rolling signal avoids the influence of roll eccentricity, and through tail compensation, it eliminates the phenomenon of unstressed thickness jumps, effectively reducing the fluctuation of the finished steel strip thickness.

[0004] Currently, the hot rolling control methods for composite plates mentioned above have achieved some results, but certain shortcomings still exist. Existing control methods mostly employ a single-parameter independent control mode, lacking dynamic modeling of the coupling relationship between speed, deformation, and rolling force. This leads to defects such as interlayer misalignment, uneven thickness, and excessive residual stress during the rolling process. Furthermore, existing control methods exhibit lag in response to real-time changes in operating conditions, making high-precision closed-loop control impossible. Summary of the Invention

[0005] The purpose of this invention is to provide a composite plate hot rolling speed-deformation-pressure coupling control system and method to respond to changes in working conditions in real time and achieve high-precision closed-loop control.

[0006] The objective of this invention is achieved through the following technical means: a method for coupled control of hot rolling speed, deformation, and pressure in composite plates, comprising the following steps:

[0007] Step 1: Collect data on the hot rolling production process and product quality of multi-layer composite panels to construct the original dataset;

[0008] Step 2: Preprocess the original dataset;

[0009] Step 3: Based on the preprocessed original dataset, establish a convolutional-long short-term memory network spatiotemporal sequence prediction model to predict the exit thickness;

[0010] Step 4: Based on the prediction results of the prediction model, adopt the corresponding feedforward control strategy to correct the roll gap parameters;

[0011] Step 5: Based on the actual thickness of the hot-rolled multilayer composite plate, a feedback control strategy is adopted to adjust the roll gap.

[0012] Step 6: Input the hot rolling production process data and product quality data of the multilayer composite plate into the convolutional-long short-term memory network spatiotemporal sequence prediction model to predict the exit thickness. If the prediction result does not match the actual required exit thickness, adjust the roll gap parameters and predict again until the prediction result matches the actual required exit thickness.

[0013] The roll gap parameters are predicted using a convolutional long short-term memory network spatiotemporal sequence prediction model for production, and the feedback control strategy in step 5 is used to adjust the roll gap in real time during the production process.

[0014] The data on the hot rolling process and product quality of the multi-layer composite plate mentioned in step 1 specifically include: inlet thickness, set outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, rolling force, and outlet thickness.

[0015] Step 2 specifically includes the following steps:

[0016] Step 2.1: Use the K-nearest neighbor interpolation algorithm to impute missing values ​​in the original dataset, as shown in the following formula:

[0017]

[0018]

[0019]

[0020] In the formula, For the first One sample to distance, It is the first The coordinates of each sample For the first The value of each sample These are the coordinates of the missing values ​​in the sample. For missing values ​​in the sample, For the first The weight of the nearest neighbor;

[0021] Step 2.2: The Savitzky-Golay filtering method is used to filter the padded data to obtain the noise-reduced data. The formula is as follows:

[0022]

[0023]

[0024]

[0025] In the formula, Let be the degree of the one-sided fitting. It is about of order polynomial, That is, the integer index within the window. The coefficients of the polynomial, , For index Enter the value of the data at the field. To fit the sum of squares of the residuals between the data points and the original data points, The data is after noise reduction;

[0026] Step 2.3: Normalize the denoised data using max-min scaling to obtain standardized data, as shown in the following formula:

[0027]

[0028] in This represents the data after noise reduction. Represents standardized data. and These represent the maximum and minimum values ​​of the original dataset after denoising.

[0029] Step 2.4: Extract 80% of the standardized data as training data to build a prediction model, and use the remaining data as test data to evaluate the model performance.

[0030] Step 3 specifically includes the following steps:

[0031] Step 3.1: Build a convolutional-long short-term memory network spatiotemporal sequence pre-training model. The principle formula is as follows:

[0032]

[0033] In the formula, Indicates the first The weights corresponding to the convolution kernel at each position. Indicates the first The bias corresponding to the convolution kernel at the given position. This represents the feature mapping of the previous layer. This represents the feature mapping of the current layer. This represents the activation function. Represents the feature mapping set,

[0034]

[0035]

[0036] In the formula, For the pooling result, The pixel values ​​are the input features. and These are the row and column indices of the feature, respectively. and These represent the row and column dimensions of the pooling window, respectively.

[0037]

[0038]

[0039]

[0040]

[0041] In the formula, To represent the weight matrix of the input gate, This is the bias of the input gate. For the sigmoid function, Here is the weight matrix for the forget gate. Indicates the vector sum vector The concatenation forms a new vector. The hidden state at time step t-1. The input at time step t, For the offset of the forget gate, This is the weight matrix of the output gate. For the output gate bias, This is a temporary hidden state value. This is the weight matrix used for feature selection with the forget gate. This is the bias used when using the forget gate for feature selection;

[0042] Step 3.2: Use the inlet thickness, set the outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, and rolling force as model inputs, and the outlet thickness as model output;

[0043] Step 3.3: Use grid search strategy and cross-validation to determine the optimal hyperparameters of the spatiotemporal series prediction model to ensure that the model has the best prediction performance;

[0044] Step 3.4: Using the hyperparameters of the above model, construct a quality prediction model for hot-rolled multilayer composite plates.

[0045] Step 4 specifically includes the following steps:

[0046] Step 4.1: Based on the deviation between the predicted exit thickness and the target exit thickness, a multi-objective balance optimization algorithm is used to perform feedforward correction on the roll gap;

[0047] Step 4.2: Determine whether the adjustment amount of the last rolling parameter exceeds the limit; if it exceeds the limit, repeat step 4.1; if it does not exceed the limit, use the corrected roll gap for rolling.

[0048] The feedback control strategy in step 5 specifically includes the following steps:

[0049] Step 5.1: Calculate the residual between the actual thickness of the composite plate at the current moment and the set exit thickness. Based on the residual value at the current moment, calculate the roll gap feedback adjustment amount at the next moment.

[0050] Step 5.2: Determine whether the roll gap feedback adjustment exceeds the limit; if it does, repeat step 5.1; if it does not exceed the limit, add the feedforward correction roll gap to the roll gap feedback adjustment and roll.

[0051] A composite plate hot rolling speed-deformation-pressure coupling control system includes,

[0052] The export thickness prediction system is used to predict the export thickness of finished multilayer composite boards based on production and process parameters. When the predicted export thickness does not meet the set export thickness requirements, the production and process parameters are modified until the predicted export thickness matches the set export thickness.

[0053] The roll gap parameter feedback control system employs a feedback control strategy to control the roll gap parameters of the production equipment to ensure that the actual production results meet the requirements.

[0054] The beneficial effects of this invention are as follows: For the constant roll gap control in hot rolling of multilayer composite plates, a dual-loop control method of "feedforward plus feedback" is proposed. Based on hot rolling data of multilayer composite plates, a spatiotemporal sequence prediction model is established using a long short-time memory network to predict rolling speed, rolling force, and deformation. Based on the deviation between the predicted quality and the target quality, a multi-objective balance optimization algorithm is used to perform feedforward correction on the roll gap. On the basis of feedforward correction, the residual between the actual thickness and the set exit thickness is further considered, and a roll gap feedback adjustment strategy is designed to achieve coupled control of hot rolling speed, deformation, and pressure for composite plates. The coupled control method of hot rolling speed, deformation, and pressure proposed in this invention has high control accuracy, overcomes the shortcomings of traditional methods, and improves the dynamic coupling control of speed, deformation, and rolling force. It can be widely applied to the hot rolling of multilayer composite plates. Attached Figure Description

[0055] Figure 1 This is a flowchart of the composite plate hot rolling speed-deformation-pressure coupling control method of the present invention;

[0056] Figure 2 This is a performance comparison chart between the present invention and a single prediction model in an embodiment of the invention;

[0057] Figure 3 This is a comparison diagram of hot rolling force control before and after in an embodiment of the present invention;

[0058] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Detailed Implementation

[0059] A method for coupled control of hot rolling speed, deformation, and pressure in composite plates includes the following steps:

[0060] Step 1: Collect data on the hot rolling production process and product quality of multi-layer composite panels to construct the original dataset;

[0061] The hot rolling process data and product quality data of the multilayer composite plate mentioned in step 1 specifically include: inlet thickness, set outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, rolling force, and outlet thickness. The set outlet thickness and outlet thickness are the final product quality data, while the rest are production process data.

[0062] Historical production process data and product quality data are used for subsequent predictive model training.

[0063] Step 2: Preprocess the original dataset;

[0064] Step 2 specifically includes the following steps:

[0065] Step 2.1: Use the K-nearest neighbor interpolation algorithm to impute missing values ​​in the original dataset, as shown in the following formula:

[0066]

[0067]

[0068]

[0069] In the formula, For the first One sample to distance, It is the first The coordinates of each sample For the first The value of each sample These are the coordinates of the missing values ​​in the sample. For missing values ​​in the sample, For the first The weight of the nearest neighbor;

[0070] Step 2.2: The Savitzky-Golay filtering method is used to filter the padded data to obtain the noise-reduced data. The formula is as follows:

[0071]

[0072]

[0073]

[0074] In the formula, Let be the degree of the one-sided fitting. It is about of order polynomial, That is, the integer index within the window. The coefficients of the polynomial, , For index Enter the value of the data (i.e., the data after filling in step 2.1). To fit the sum of squares of the residuals between the data points and the original data points, The data is after noise reduction;

[0075] Step 2.3: Normalize the denoised data using max-min scaling to obtain standardized data, as shown in the following formula:

[0076]

[0077] in This represents the data after noise reduction. Represents standardized data. and These represent the maximum and minimum values ​​of the original dataset after denoising.

[0078] Step 2.4: Extract 80% of the standardized data as training data to build a prediction model, and use the remaining data as test data to evaluate the model performance.

[0079] In this embodiment, a hot-rolled strip steel dataset is constructed as shown in Table 1. Missing values ​​are filled and normalized in the dataset. 80% of the data is randomly extracted as training data and the remaining data is used as test data to evaluate the model performance.

[0080] Table 1

[0081]

[0082] The parameters omitted in the columns of Table 1 are the set exit thickness, deformation, reduction rate, inlet width, and rolling temperature, which are omitted because they are not used in the subsequent feedforward process. The rows omit data sets with different numbers, as they are omitted due to their large number.

[0083] Step 3: Based on the preprocessed original dataset, establish a convolutional-long short-term memory network spatiotemporal sequence prediction model to predict the exit thickness;

[0084] Step 3 specifically includes the following steps:

[0085] Step 3.1: Build a CNN-LSTM pre-trained model (i.e., a convolutional long short-term memory network spatiotemporal sequence prediction model). The principle formula is as follows:

[0086]

[0087] In the formula, Indicates the first The weights corresponding to the convolution kernel at each position. Indicates the first The bias corresponding to the convolution kernel at the given position. This represents the feature mapping of the previous layer. This represents the feature mapping of the current layer. This represents the activation function. Represents the feature mapping set,

[0088]

[0089]

[0090] In the formula, For the pooling result, The pixel values ​​are the input features. and These are the row and column indices of the feature, respectively. and These represent the row and column dimensions of the pooling window, respectively.

[0091]

[0092]

[0093]

[0094]

[0095] In the formula, To represent the weight matrix of the input gate, This is the bias of the input gate. For the sigmoid function, Here is the weight matrix for the forget gate. Indicates the vector sum vector The concatenation forms a new vector. The hidden state at time step t-1. The input at time step t, For the offset of the forget gate, This is the weight matrix of the output gate. For the output gate bias, This is a temporary hidden state value. This is the weight matrix used for feature selection with the forget gate. This is the bias used when using the forget gate for feature selection;

[0096] Step 3.2: Use the inlet thickness, set the outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, and rolling force as model inputs, and the outlet thickness as model output;

[0097] Step 3.3: Use grid search strategy and cross-validation to determine the optimal hyperparameters of the spatiotemporal sequence prediction model to ensure that the model has the best prediction performance; when the results of grid search strategy and cross-validation meet the design requirements, the corresponding hyperparameters are the optimal hyperparameters.

[0098] Step 3.4: Using the hyperparameters of the above model, construct a quality prediction model for hot-rolled multilayer composite plates.

[0099] The optimal hyperparameters are then fed into the CNN-LSTM pre-trained model, which is the prediction model used for subsequent predictions.

[0100] In this embodiment, R is used. 2The RMSE evaluation metric is used for hyperparameter tuning and performance testing of the fusion model, namely the convolutional-long short-term memory network spatiotemporal sequence prediction model in step 3, and the formula is as follows:

[0101]

[0102] In the formula The coefficient of determination These are measured values. For predicted values, For the sample size, This represents the average value of the measured variables. 'i' is the sample index symbol used in mathematical summation.

[0103]

[0104] In the formula The root mean square error, These are measured values. For predicted values, This represents the number of measured values. 'i' is the symbol for the sample index in mathematical summation.

[0105] In this embodiment, for the dataset (the original dataset constructed by collecting multilayer composite plate hot rolling production process data and product quality data in step 1), cross-validation and grid search strategies are used to determine the main hyperparameters of all candidate models in the model candidate pool, as shown in Table 2.

[0106] Table 2

[0107]

[0108] In this embodiment, the established CNN-LSTM spatiotemporal sequence model is tested on the real-time acquired dataset (the original dataset from step 1), and compared with other spatiotemporal sequence models. The performance comparison of the best model corresponding to the data is shown in Table 3. Figure 2 As shown, the final prediction result R of CNN-LSTM is obtained. 2 The value is 0.978, and the RMSE value is 2.933. Compared with the results of testing other spatiotemporal prediction strategy models directly on the original dataset, the CNN-LSTM model performs better.

[0109] Table 3

[0110]

[0111] Step 4: Based on the prediction results of the prediction model, adopt the corresponding feedforward control strategy to correct the roll gap parameters; the roll gap parameters specifically refer to the gap height between the two rolls of the rolling equipment, which is a control quantity in the rolling process. The thickness of the finished product, that is, the thickness of the composite plate at the exit, is controlled by setting the roll gap.

[0112] Step 4 specifically includes the following steps:

[0113] Step 4.1: Based on the deviation between the predicted exit thickness and the target exit thickness, a multi-objective balance optimization algorithm is used to perform feedforward correction on the roll gap;

[0114]

[0115]

[0116]

[0117]

[0118] In the formula, To predict thickness deviation, The predicted thickness of the spatiotemporal sequence prediction model for convolutional-long short-term memory networks. For the target thickness, Let $\frac{ ... For a multi-objective equilibrium optimization function, To constrain the roll gap adjustment amount, This is a set of process constraints, including process limitations such as rolling force, temperature, and tension. The feedforward correction roll gap after the k-th iteration. For the initial roll gap, The thickness prediction result is after k iterations. This is a quality prediction model for hot-rolled multilayer composite plates (i.e., a spatiotemporal sequence prediction model using convolutional-long short-term memory networks).

[0119] Step 4.2: Determine whether the adjustment of the last rolling parameters (motor speed, roll gap, hydraulic reduction) exceeds the limit; if it exceeds the limit, repeat step 4.1; if it does not exceed the limit, use the corrected roll gap for rolling.

[0120]

[0121] In the formula, To allow for thickness tolerance, the absolute value of the difference between the predicted thickness and the target thickness must be less than or equal to this threshold, i.e., it must not exceed the limit.

[0122] Before actual production, the exit thickness is predicted using the multilayer composite plate hot-rolled product quality prediction model (i.e., the convolutional-long short-term memory network spatiotemporal sequence prediction model in step 3) based on the anticipated process parameters and other data. The prediction result is then checked to see if it meets the required exit thickness. The process parameters, excluding the exit thickness, are the multilayer composite plate hot-rolling production process data and product quality data mentioned in step 1, specifically including: inlet thickness, set exit thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, and rolling force.

[0123] If the result meets the requirements, production proceeds. If not, a multi-objective balance optimization algorithm is used to modify the roll gap size, and the prediction is repeated until the prediction result meets the requirements.

[0124] Step 5: Based on the actual thickness of the hot-rolled multilayer composite plate, a feedback control strategy is adopted to adjust the roll gap.

[0125] The feedback control strategy in step 5 specifically includes the following steps:

[0126] Step 5.1: Calculate the residual between the actual thickness of the composite plate at the current moment and the set exit thickness. Based on the residual value at the current moment, calculate the roll gap feedback adjustment amount at the next moment.

[0127]

[0128] In the formula, for The thickness residual at any given moment for The actual thickness of a moment.

[0129] The roll gap feedback adjustment is converted based on its own rigidity coefficient, so that the converted value is equal to the residual value at the current moment.

[0130]

[0131] In the formula, For the next moment ( The roll gap feedback adjustment amount, This is the mill rigidity coefficient.

[0132] The roll gap feedback adjustment is converted based on its own rigidity coefficient, so that the converted value is equal to the residual value at the current moment.

[0133] For example, if the residual value is 0.01m thicker than the standard thickness, after conversion based on the rigidity coefficient, the roll gap can be reduced by 0.005m, which can reduce the thickness of the multilayer composite board by 0.01m.

[0134] Step 5.2: Determine whether the roll gap feedback adjustment exceeds the limit; if it does, repeat step 5.1; if it does not exceed the limit, add the feedforward correction roll gap to the roll gap feedback adjustment and roll.

[0135]

[0136]

[0137] In the formula, This represents the absolute value of the maximum allowable adjustment in feedback control. for The final roll gap setting value at any given time. for The feedforward correction roll gap value at any given time.

[0138] Taking the first sample in the original dataset as an example, its predicted strip thickness is 3.43 mm, and the target thickness is 3.40 mm. Based on the feedback of the deviation between the target thickness and the predicted thickness, a multi-objective balance optimization algorithm is used to perform feedforward control of the roll gap for each pass. The roll gap adjustment is calculated based on the deviation of the previous sample to obtain the final roll gap value for each pass, and then the corrected rolling force for each pass is calculated. Figure 3 As shown, the adjustment limit was not exceeded. The output thickness after control was 3.41 mm, and the residual difference from the target thickness was only 0.01 mm, achieving a good strip thickness control effect. This demonstrates that the present invention can accurately and effectively realize the intelligent control of hot-rolled multilayer composite plate products.

[0139] Step 6: Input the hot rolling production process data and product quality data of the multilayer composite plate into the convolutional-long short-term memory network spatiotemporal sequence prediction model to predict the exit thickness. If the prediction result does not match the actual required exit thickness, adjust the roll gap parameters and predict again until the prediction result matches the actual required exit thickness.

[0140] The roll gap parameters are predicted using a convolutional long short-term memory network spatiotemporal sequence prediction model for production, and the feedback control strategy in step 5 is used to adjust the roll gap in real time during the production process.

[0141] After inputting the parameters from step 1 into the convolutional-long short-term memory network spatiotemporal sequence prediction model for preliminary prediction, it is ensured that the final exit thickness can theoretically meet the design requirements. Then, the roll gap parameters are corrected in step 4 so that the final actual exit thickness meets the design requirements.

[0142] Then, production is carried out using the corrected roll gap parameters, as well as other preset process parameters (i.e., the parameters in step 1, as well as the hydraulic pressing and motor speed). During production, the feedback control strategy in step 5 is used to adjust the roll gap in real time, so as to obtain the required exit thickness of the multilayer composite board.

[0143] A composite plate hot rolling speed-deformation-pressure coupling control system includes,

[0144] The export thickness prediction system is used to predict the export thickness of finished multilayer composite boards based on production and process parameters. When the predicted export thickness does not meet the set export thickness requirements, the production and process parameters are modified until the predicted export thickness matches the set export thickness.

[0145] Production and process parameters, namely production process data and product quality data excluding exit thickness, specifically include: inlet thickness, set exit thickness, deformation amount, reduction rate, inlet width, rolling temperature, rolling speed, and rolling force.

[0146] The roll gap parameter feedback control system employs a feedback control strategy to control the roll gap parameters of the production equipment, ensuring that the actual production results meet requirements. Meeting requirements means that the actual output thickness of the composite board meets the set output thickness requirement.

[0147] like Figure 1 As shown, the original set values, namely the pre-set production parameters and process parameters (i.e., inlet thickness, set outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed and rolling force), are used by the prediction model to obtain the prediction result, which is the theoretically predicted outlet thickness.

[0148] The predicted export thickness is compared with the required export thickness. Based on the feedforward control strategy in step 4 of the method, the production and process parameters are feedforward corrected so that the predicted export thickness matches the set export thickness. In other words, it is theoretically possible to produce composite boards with qualified thickness.

[0149] Production is carried out using the corrected parameters. The actual process parameters such as speed, deformation, and rolling force are detected (where speed corresponds to motor speed, deformation corresponds to roll gap opening, and rolling force corresponds to hydraulic pressing). The parameters are compared with the target parameters. According to the feedback control strategy in step 5 of the method, the roll gap opening, motor speed (motor speed is used to control the rolling speed) and hydraulic pressing are corrected.

[0150] The final product is a multi-layer composite board that meets the requirements.

Claims

1. A method for coupled control of hot rolling speed, deformation, and pressure in composite plates, characterized in that, Includes the following steps: Step 1: Collect data on the hot rolling production process and product quality of multi-layer composite panels to construct the original dataset; Step 2: Preprocess the original dataset; Step 3: Based on the preprocessed original dataset, establish a convolutional-long short-term memory network spatiotemporal sequence prediction model to predict the exit thickness; Step 4: Based on the prediction results of the prediction model, adopt the corresponding feedforward control strategy to correct the roll gap parameters; Step 5: Based on the actual thickness of the hot-rolled multilayer composite plate, a feedback control strategy is adopted to adjust the roll gap. Step 6: Input the hot rolling production process data and product quality data of the multilayer composite plate into the convolutional-long short-term memory network spatiotemporal sequence prediction model to predict the exit thickness. If the prediction result does not match the actual required exit thickness, adjust the roll gap parameters and predict again until the prediction result matches the actual required exit thickness. The roll gap parameters are predicted using a convolutional long short-term memory network spatiotemporal sequence prediction model for production, and the feedback control strategy in step 5 is used to adjust the roll gap in real time during the production process. Step 3 specifically includes the following steps. Step 3.1: Build a convolutional-long short-term memory network spatiotemporal sequence pre-training model. The principle formula is as follows: In the formula, Indicates the first The weights corresponding to the convolution kernel at each position. Indicates the first The bias corresponding to the convolution kernel at the given position. This represents the feature mapping of the previous layer. This represents the feature mapping of the current layer. This represents the activation function. Represents the feature mapping set, In the formula, For the pooling result, The pixel values ​​are the input features. and These are the row and column indices of the feature, respectively. and These represent the row and column dimensions of the pooling window, respectively. In the formula, To represent the weight matrix of the input gate, This is the bias of the input gate. For the sigmoid function, Here is the weight matrix for the forget gate. Indicates the vector sum vector The concatenation forms a new vector. The hidden state at time step t-1. The input at time step t, For the offset of the forget gate, This is the weight matrix of the output gate. For the output gate bias, This is a temporary hidden state value. This is the weight matrix used for feature selection with the forget gate. This is the bias used when using the forget gate for feature selection; Step 3.2: Use the inlet thickness, set the outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, and rolling force as model inputs, and the outlet thickness as model output; Step 3.3: Use grid search strategy and cross-validation to determine the optimal hyperparameters of the spatiotemporal series prediction model to ensure that the model has the best prediction performance; Step 3.4: Using the hyperparameters of the above model, construct a quality prediction model for hot-rolled multilayer composite plates; The feedback control strategy in step 5 specifically includes the following steps. Step 5.1: Calculate the residual between the actual thickness of the composite plate at the current moment and the set exit thickness. Based on the residual value at the current moment, calculate the roll gap feedback adjustment amount at the next moment. Step 5.2: Determine whether the roll gap feedback adjustment exceeds the limit; if it does, repeat step 5.1; if it does not exceed the limit, add the feedforward correction roll gap to the roll gap feedback adjustment and roll.

2. The composite plate hot rolling speed-deformation-pressure coupling control method according to claim 1, characterized in that, The data on the hot rolling process and product quality of the multi-layer composite plate mentioned in step 1 specifically include: inlet thickness, set outlet thickness, deformation, reduction rate, inlet width, rolling temperature, rolling speed, rolling force, and outlet thickness.

3. The composite plate hot rolling speed-deformation-pressure coupling control method according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Use the K-nearest neighbor interpolation algorithm to impute missing values ​​in the original dataset, as shown in the following formula: In the formula, For the first One sample to distance, It is the first The coordinates of each sample For the first The value of each sample These are the coordinates of the missing values ​​in the sample. For missing values ​​in the sample, For the first The weight of the nearest neighbor; Step 2.2: The Savitzky-Golay filtering method is used to filter the padded data to obtain the noise-reduced data. The formula is as follows: In the formula, Let be the degree of the one-sided fitting. It is about of order polynomial, That is, the integer index within the window. The coefficients of the polynomial, , For index Enter the value of the data at the field. To fit the sum of squares of the residuals between the data points and the original data points, The data is after noise reduction; Step 2.3: Normalize the denoised data using max-min scaling to obtain standardized data, as shown in the following formula: in This represents the data after noise reduction. Represents standardized data. and These represent the maximum and minimum values ​​of the original dataset after denoising. Step 2.4: Extract 80% of the standardized data as training data to build a prediction model, and use the remaining data as test data to evaluate the model performance.

4. The composite plate hot rolling speed-deformation-pressure coupling control method according to claim 1, characterized in that, Step 4 specifically includes the following steps: Step 4.1: Based on the deviation between the predicted exit thickness and the target exit thickness, a multi-objective balance optimization algorithm is used to perform feedforward correction on the roll gap; Step 4.2: Determine whether the adjustment amount of the last rolling parameter exceeds the limit; if it exceeds the limit, repeat step 4.1; if it does not exceed the limit, use the corrected roll gap for rolling.

5. A composite plate hot rolling speed-deformation-pressure coupling control system according to any one of claims 1-4, characterized in that: include, The export thickness prediction system is used to predict the export thickness of finished multilayer composite boards based on production and process parameters. When the predicted export thickness does not meet the set export thickness requirements, the production and process parameters are modified until the predicted export thickness matches the set export thickness. The roll gap parameter feedback control system employs a feedback control strategy to control the roll gap parameters of the production equipment to ensure that the actual production results meet the requirements.

Images