Hot rolling force prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a hot rolling force prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: taking a rolling force influence parameter as network input, taking a rolling force correction coefficient as network output, pre-constructing a target neural network, collecting data corresponding to the rolling force influence parameter and current rolling data of a to-be-rolled pass, and obtaining a rolling force prediction result; calculating an initial rolling force predicted value of the to-be-rolled pass based on the current rolling data of the to-be-rolled pass, inputting the rolling force influence parameter corresponding data into the target neural network to obtain a rolling force correction coefficient corresponding estimated value of the to-be-rolled pass, and calculating the rolling force of the to-be-rolled pass according to the rolling force correction coefficient corresponding estimated value of the to-be-rolled pass. Correcting the initial rolling force predicted value of the to-be-rolled pass to obtain a target rolling force predicted value of the to-be-rolled pass; the accuracy of the rolling force prediction value can be effectively improved, then the rolling force prediction precision and the hot rolling product quality are improved, the high-speed rolling requirement is met, and the production efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of hot rolling, and particularly relates to a method, device, electronic device and storage medium for predicting hot rolling force. Background Art

[0002] The rolling force is an important process parameter in the hot rolling production process. The accurate prediction of the rolling force is of great significance for controlling product quality and improving production efficiency. Through the accurate prediction of the rolling force, the precise control of the rolling process can be realized, thereby ensuring the dimensional accuracy and surface quality of the product.

[0003] Traditional rolling force prediction mainly relies on mathematical models for calculation. These mathematical models are mostly empirical or semi-empirical formulas obtained through mechanism analysis under various assumptions and fitting of experimental data. Their theoretical calculation accuracy is limited, resulting in inaccurate predicted values of the rolling force, which may further reduce product quality and the finished product rate. Summary of the Invention

[0004] In view of the above-mentioned disadvantages of the prior art, the present application provides a method, device, electronic device and storage medium for predicting hot rolling force to solve the technical problem that traditional rolling force prediction mainly relies on mathematical models with limited calculation accuracy, resulting in inaccurate predicted values of the rolling force, which may further reduce product quality and the finished product rate.

[0005] The present application provides a method for predicting hot rolling force. The method includes: pre-constructing a target neural network with rolling force influence parameters as network inputs and rolling force correction coefficients as network outputs; collecting data corresponding to rolling force influence parameters and current rolling data of the to-be-rolled pass; based on the current rolling data of the to-be-rolled pass, calculating an initial rolling force prediction value of the to-be-rolled pass, and inputting the data corresponding to the rolling force influence parameters into the target neural network to obtain an estimated value of the rolling force correction coefficient corresponding to the to-be-rolled pass; and correcting the initial rolling force prediction value of the to-be-rolled pass according to the estimated value of the rolling force correction coefficient corresponding to the to-be-rolled pass to obtain a target rolling force prediction value of the to-be-rolled pass.

[0006] In an embodiment of the present application, calculating an initial rolling force prediction value for the to-be-rolled pass based on the current rolling data of the to-be-rolled pass includes: establishing a rolling force mathematical model for calculating the rolling force according to the workpiece width, roll diameter, workpiece reduction, rolling temperature, rolling strain, and rolling strain rate; calculating a target value of the workpiece reduction and an estimated value of the rolling strain for the to-be-rolled pass according to the current value corresponding to the workpiece inlet thickness and the target value corresponding to the workpiece outlet thickness of the to-be-rolled pass, and calculating an estimated value of the rolling strain rate for the to-be-rolled pass according to the target value of the workpiece reduction, the value corresponding to the roll diameter, and the target value of the rolling speed for the to-be-rolled pass, wherein the current rolling data of the to-be-rolled pass includes the current value corresponding to the workpiece inlet thickness, the target value corresponding to the workpiece outlet thickness, the value corresponding to the roll diameter, and the target value of the rolling speed of the to-be-rolled pass; inputting the current value corresponding to the workpiece width, the value corresponding to the roll diameter, the target value of the workpiece reduction, the target value of the rolling temperature, the estimated value of the rolling strain, and the estimated value of the rolling strain rate of the to-be-rolled pass into the rolling force mathematical model, so that the rolling force mathematical model calculates the rolling force to obtain the initial rolling force prediction value for the to-be-rolled pass, wherein the current rolling data of the to-be-rolled pass further includes the current value corresponding to the workpiece width and the target value of the rolling temperature of the to-be-rolled pass.

[0007] In an embodiment of the present application, a target neural network is pre-constructed with the rolling force influence parameter as the network input and the rolling force correction coefficient as the network output, including: selecting the parameters that affect the meta-dynamic recrystallization before rolling deformation and the rolling deformation as the rolling force influence parameters; establishing a nonlinear dynamic system with the historical and current rolling force influence parameters as the system input and the current rolling force correction coefficient as the system output, and introducing noise and the historical rolling force correction coefficient as variables into the nonlinear dynamic system; constructing a neural network structure based on the nonlinear dynamic system to obtain the target neural network.

[0008] In an embodiment of the present application, constructing a neural network structure based on the nonlinear dynamic system to obtain the target neural network includes: constructing an input layer based on the system input of the nonlinear dynamic system, constructing an output layer based on the system output of the nonlinear dynamic system, and constructing at least one reservoir according to the historical and current rolling force influence parameters and the historical rolling force correction coefficient; establishing the connection relationships between the input layer, the output layer, and each reservoir, and assigning weights to the output layer and each reservoir to obtain the target neural network.

[0009] In one embodiment of the present application, the method further includes: obtaining the estimated values and actual values of the rolling force correction coefficients corresponding to each rolled pass, wherein the estimated values of the rolling force correction coefficients corresponding to each rolled pass are obtained based on the output result of the target neural network, and the actual values of the rolling force correction coefficients corresponding to each rolled pass are determined based on the actual rolling data of each rolled pass; taking the difference between the estimated value of the rolling force correction coefficient corresponding to a rolled pass and the actual value of the rolling force correction coefficient corresponding to the rolled pass as a fitting error, and obtaining multiple fitting errors; online adjusting the weights of the output layer based on the multiple fitting errors, and offline adjusting the weights of each reservoir based on the multiple fitting errors to complete the parameter update of the target neural network.

[0010] In one embodiment of the present application, online adjusting the weights of the output layer based on multiple fitting errors includes:

[0011] After the rolling of each rolling pass is completed, forming an error matrix with multiple fitting errors, and forming a state variable matrix with the state variables corresponding to each reservoir; determining the new weights of the output layer according to the error matrix and the state variable matrix;

[0012] Or,

[0013] After the rolling of each rolling pass is completed, determining the minimum network fitting error based on multiple fitting errors, the weights of the output layer, and the weights of each reservoir; determining the new weights of the output layer according to the minimum network fitting error and the weights of the output layer.

[0014] In one embodiment of the present application, offline adjusting the weights of each reservoir based on multiple fitting errors includes: after the rolling of all rolling passes is completed, determining the minimum network fitting error based on multiple fitting errors, the weights of the output layer, and the weights of each reservoir; determining the new weights of each reservoir according to the minimum network fitting error and the weights of each reservoir.

[0015] In an embodiment of the present application, a hot rolling rolling force prediction device is further provided. The device includes: a data acquisition module for acquiring data corresponding to rolling force influence parameters and current rolling data of the to-be-rolled pass; a first information processing module for calculating an initial rolling force prediction value of the to-be-rolled pass based on the current rolling data of the to-be-rolled pass; a second information processing module for inputting the data corresponding to the rolling force influence parameters into a target neural network to obtain an estimated value corresponding to the rolling force correction coefficient of the to-be-rolled pass, wherein the target neural network is pre-constructed with the rolling force influence parameters as the network input and the rolling force correction coefficient as the network output; and a data correction module for correcting the initial rolling force prediction value of the to-be-rolled pass according to the estimated value corresponding to the rolling force correction coefficient of the to-be-rolled pass to obtain a target rolling force prediction value of the to-be-rolled pass.

[0016] In an embodiment of the present application, an electronic device is further provided. The electronic device includes: one or more processors; a storage device for storing one or more programs, which when executed by the one or more processors, cause the electronic device to implement the hot rolling rolling force prediction method as described above.

[0017] In an embodiment of the present application, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer is caused to execute the hot rolling rolling force prediction method as described above.

[0018] Advantages of the present invention: The present invention provides a hot rolling rolling force prediction method, device, electronic device and storage medium. The method constructs a target neural network with the rolling force influence parameters as the network input and the rolling force correction coefficient as the network output, and uses the target neural network to self-learn the influence of the rolling force influence parameters on the rolling force, estimates the rolling force correction coefficient of the to-be-rolled pass, and corrects the initial rolling force prediction value, which can better capture the non-linear relationship between the rolling force influence parameters and the rolling force correction coefficient, effectively improve the accuracy of the rolling force prediction value, and further improve the rolling force prediction accuracy and the quality of hot-rolled products. Moreover, the rolling force correction coefficient can be quickly generated through the target neural network to quickly correct the initial prediction value, which can well adapt to the high-speed requirements of the continuous rolling process, thereby improving the hot rolling production efficiency.

[0019] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Description of the Drawings

[0020] Figure 1 is a schematic diagram of the implementation environment of a hot rolling rolling force prediction method shown in an exemplary embodiment of the present application;

[0021] Figure 2 It is a flowchart of a hot rolling rolling force prediction method shown in an exemplary embodiment of the present application;

[0022] Figure 3 It is a schematic structural diagram of a target neural network shown in a specific embodiment of the present application;

[0023] Figure 4 It is a flowchart of the rolling force prediction of a rolling force calculation model shown in a specific embodiment of the present application;

[0024] Figure 5 It is a block diagram of a hot rolling rolling force prediction device shown in an exemplary embodiment of the present application;

[0025] Figure 6 It is a schematic diagram of a hot rolling rolling force prediction system shown in a specific embodiment of the present application;

[0026] Figure 7 is Figure 6 The working flowchart of the hot rolling rolling force prediction system in the specific embodiment shown;

[0027] Figure 8 It is a schematic structural diagram of an electronic device shown in an exemplary embodiment of the present application. Specific Embodiments

[0028] The following uses specific specific examples to illustrate the implementation manners of the present application. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0029] It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present application in a schematic manner. Therefore, only the components related to the present application are shown in the drawings, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0030] It should be noted that in the present application, "first", "second", etc. are only used to distinguish similar objects, and are not order limitations or sequence limitations on similar objects. The described "including", "having", etc. are deformed, indicating that the scope covered by the subject of the word does not exclude other examples except the examples shown by the word.

[0031] It is understood that the various numerical numbers, step numbers, etc. recorded in this application are for the convenience of description and do not limit the scope of this application. The size of the labels in this application does not imply the order of execution, and the execution order of each process should be determined by its function and internal logic.

[0032] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of this application. However, it is obvious to those skilled in the art that the embodiments of this application can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of this application difficult to understand.

[0033] Embodiments of this application respectively propose a hot rolling rolling force prediction method, a hot rolling rolling force prediction device, an electronic device, a computer-readable storage medium, and a computer program product. The following will describe these embodiments in detail.

[0034] Please refer to Figure 1 , Figure 1 which is a schematic diagram of the implementation environment of a hot rolling rolling force prediction method shown in an exemplary embodiment of this application.

[0035] As Figure 1 shown, the implementation environment may include a data acquisition device 110 and a computer device 120. Among them, the data acquisition device 110 may be various sensors installed on a hot rolling production line or a hot rolling automatic control system, or a data acquisition system for collecting data from various sensors on a hot rolling production line or a hot rolling automatic control system. The computer device 120 may be at least one of a microcomputer, an embedded computer, a neural network computer, etc., and no limitations are imposed here. The computer device 120 may be configured in a hot rolling automatic control system or may be a computer device independent of the hot rolling automatic control system, and no limitations are imposed here either. The data acquisition device 110 can be used to collect data corresponding to rolling force influence parameters and current rolling data of the to-be-rolled passes and provide them to the computer device 120, and the computer device 120 realizes hot rolling rolling force prediction.

[0036] Schematically, the computer device 120 pre - constructs a target neural network with the rolling force influencing parameter as the network input and the rolling force correction coefficient as the network output; the data acquisition device 110 acquires the data corresponding to the rolling force influencing parameter and the current rolling data of the to - be - rolled pass; based on the current rolling data of the to - be - rolled pass, calculates the initial rolling force prediction value of the to - be - rolled pass, and inputs the data corresponding to the rolling force influencing parameter into the target neural network to obtain the estimated value of the rolling force correction coefficient corresponding to the to - be - rolled pass; according to the estimated value of the rolling force correction coefficient corresponding to the to - be - rolled pass, corrects the initial rolling force prediction value of the to - be - rolled pass to obtain the target rolling force prediction value of the to - be - rolled pass. It can be seen that the technical solution of the embodiment of the present application constructs a target neural network with the rolling force influencing parameter as the network input and the rolling force correction coefficient as the network output, uses the target neural network to self - learn the influence of the rolling force influencing parameter on the rolling force, estimates the rolling force correction coefficient of the to - be - rolled pass, and corrects the initial rolling force prediction value, which can better capture the non - linear relationship between the rolling force influencing parameter and the rolling force correction coefficient, effectively improve the accuracy of the rolling force prediction value, and further improve the rolling force prediction accuracy and the quality of hot - rolled products. Moreover, through the target neural network, the rolling force correction coefficient can be quickly generated to realize the rapid correction of the initial prediction value, which can well adapt to the high - speed requirements of the continuous rolling process, thereby improving the hot - rolling production efficiency.

[0037] It should be noted that the hot - rolling rolling force prediction method provided by the embodiment of the present application can be specifically executed by the computer device 120. Correspondingly, the hot - rolling rolling force prediction device can be set in the computer device 120.

[0038] Please refer to Figure 2 , Figure 2 is a flowchart of a hot - rolling rolling force prediction method shown in an exemplary embodiment of the present application. This hot - rolling rolling force prediction method can be applied to Figure 1 the implementation environment shown, and is specifically executed by the computer device 120 in this implementation environment. It should be understood that this hot - rolling rolling force prediction method can also be applicable to other exemplary implementation environments and be specifically executed by devices in other implementation environments. The embodiment does not limit the implementation environment applicable to this hot - rolling rolling force prediction method. As Figure 2 shown, in an exemplary embodiment, this hot - rolling rolling force prediction method at least includes steps S210 to S240, which are introduced in detail as follows:

[0039] Step S210, pre - construct a target neural network with the rolling force influencing parameter as the network input and the rolling force correction coefficient as the network output.

[0040] In one embodiment of the present application, from the perspective of factors affecting the stress state, there are various interference factors at the hot-rolling production site. These interference factors may cause errors in the measurement of deformation zone parameters and rolling force and energy parameters. Changes in equipment state, steel plate shape, and surface state will also cause changes in friction conditions, all of which will change the stress state function. From the perspective of the properties of the metal itself, the initial microstructure morphology, strain, strain rate, temperature, etc. affect the dislocation evolution during the rolling process and determine the deformation resistance during the rolling process. These variables are simultaneously coupled with the rolling interval duration and rolling interval temperature, affecting the meta-dynamic recrystallization process during the rolling interval and determining the initial microstructure morphology of the next pass. It can be seen that the hot-rolling process exhibits the characteristics of multi-variables, strong coupling, and dynamic non-linearity. Therefore, one or more of these parameters can be selected according to requirements as the rolling force influence parameters. With the rolling force influence parameters as the network input and the rolling force correction coefficient as the network output, a target neural network is pre-constructed, and the target neural network self-learns the influence of the rolling force influence parameters on the rolling force to better capture the non-linear relationship between the rolling force influence parameters and the rolling force correction coefficient.

[0041] Among them, the rolling force influence parameters include at least one of the rolling interval duration, rolling interval temperature, rolling strain, rolling strain rate, rolling temperature, width of the rolled piece, roll diameter, deviation of the gold element content between the rolled piece and the standard part, ratio of the contact arc length in the deformation zone to the average height of the rolled piece, etc. The roll diameter includes at least one of the roll diameter and roll radius, etc.

[0042] In one embodiment of the present application, step S210 includes: selecting the parameters that affect the meta-dynamic recrystallization before rolling deformation and rolling deformation as the rolling force influence parameters; establishing a non-linear dynamic system with the historical and current rolling force influence parameters as the system input and the current rolling force correction coefficient as the system output, and introducing noise and the historical rolling force correction coefficient as variables into the non-linear dynamic system; constructing a neural network structure based on the characteristics of the non-linear dynamic system to obtain the target neural network.

[0043] In this embodiment, the initial microstructure morphology, rolling strain, rolling strain rate, rolling temperature, etc. affect the dislocation evolution during the rolling process and determine the deformation resistance during the rolling process. These variables are simultaneously coupled with the rolling interval duration and rolling interval temperature, which will affect the meta-dynamic recrystallization process before rolling deformation (i.e., the rolling interval). The meta-dynamic recrystallization process during the rolling interval determines the initial microstructure morphology of the next pass. Therefore, the parameters that affect the meta-dynamic recrystallization before rolling deformation and rolling deformation can be preferentially selected as the rolling force influence parameters, including rolling strain, rolling strain rate, rolling temperature, rolling interval duration, and rolling interval temperature.

[0044] After selecting the parameters reflecting the sub-dynamic recrystallization before rolling and the deformation conditions during the rolling process as the rolling force influence parameters, variables reflecting the internal microstructure state of the material and the equipment state are selected as the rolling force correction coefficients. Then, with the historical rolling force influence parameters and the current rolling force influence parameters as the system inputs and the current rolling force correction coefficient as the system output, a nonlinear dynamic system is constructed. In hot rolling, the rolling process of the subsequent pass is affected by the state of the historical pass. Therefore, to reflect the influence of the historical state at the current moment, the historical rolling force correction coefficient can be introduced as a variable of the function in this nonlinear dynamic system. In addition, there are some random factors in the hot rolling site that affect the rolling process. Therefore, noise can also be introduced as a variable of the function in this nonlinear dynamic system to characterize these random factors. Schematically, the mathematical model of the nonlinear dynamic system is as follows:

[0045] y(k + 1) = f[y(k), y(k - 1), …, y(k - p), u(k), u(k - 1), …, u(k - p), n(k)] Equation (1)

[0046] Wherein, y(k + 1) is the current rolling force correction coefficient, that is, the rolling force correction coefficient of the current pass (the pass to be rolled), y(k), y(k - 1), …, y(k - p) are the historical rolling force correction coefficients, that is, the rolling force correction coefficients of each historical pass (the rolled passes), u(k - 1), …, u(k - p) are the historical rolling force influence parameters, that is, the rolling force influence parameters of each historical pass (the rolled passes), u(k) is the current rolling force influence parameter, that is, the rolling force influence parameter of the current pass (the pass to be rolled), n(k) is the system input noise, and f is an unknown nonlinear function.

[0047] Since this nonlinear dynamic system is an unknown dynamic system, to meet the requirements of on-line control of hot rolling production, a fitting structure that reflects the characteristics of the dynamic system and can respond and update quickly on-line needs to be designed. Therefore, the characteristics and structure of this nonlinear dynamic system can be analyzed, the type of neural network can be determined according to its characteristics, and the network structure can be constructed according to its structure to obtain the target neural network.

[0048] It should be understood that the selection of system input and output is not fixed. The artificial experience method or machine learning can be used.

[0049] In some embodiments, the ratio of the contact arc length in the deformation zone to the average height of the rolled piece, the width of the rolled piece, and the roll diameter will affect the deformation zone. Therefore, these parameters affecting the deformation zone can also be used as the rolling force influence parameters.

[0050] In some embodiments, the traditional rolling force calculation formula is obtained based on standard parts in the laboratory, and there are certain differences between the chemical compositions of the standard parts and those of the rolled pieces in actual industrial production, which will also lead to calculation deviations. Therefore, the deviation of the gold element content between the rolled piece and the standard part can also be used as a rolling force influence parameter.

[0051] In some embodiments, the rolling interval duration, rolling interval temperature, rolling strain, rolling strain rate, rolling temperature, width of the rolled piece, roll diameter, deviation of the gold element content between the rolled piece and the standard part, and the ratio of the contact arc length in the deformation zone to the average height of the rolled piece can be used together as rolling force influence parameters.

[0052] In an embodiment of the present application, a neural network structure is constructed based on a nonlinear dynamics system to obtain a target neural network, including: constructing an input layer based on the system input of the nonlinear dynamics system, constructing an output layer based on the system output of the nonlinear dynamics system, and constructing at least one reservoir according to historical rolling force influence parameters and historical rolling force correction coefficients; establishing connection relationships between the input layer, the output layer, and each reservoir, and performing weight allocation on the output layer and each reservoir to obtain the target neural network.

[0053] In this embodiment, it can be seen from analyzing Equation (1) that the current rolling force correction coefficient y(k + 1) output by the nonlinear dynamics system is affected by historical rolling force correction coefficients y(k), y(k - 1), …, y(k - p), which reflects the influence of the historical state on the current moment of rolling. Based on this characteristic, a network structure that can reflect the influence of the historical state on the current moment can be selected as the target neural network. Schematically, an echo network can be selected as the target neural network.

[0054] The system input of the nonlinear dynamics system is used as the input parameter of the target neural network to establish an input layer, the system output of the nonlinear dynamics system is used as the output parameter of the target neural network to establish an output layer, and reservoirs are established according to historical rolling force influence parameters, current rolling force influence parameters, and historical rolling force correction coefficients. The number of each layer and the connection method between each layer are determined to connect each layer to complete the construction of the target neural network. In addition, the weights of the output layer and each reservoir can be obtained by initializing network parameters and performing weight allocation on the output layer and each reservoir respectively.

[0055] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of the target neural network shown in a specific embodiment of the present application. As Figure 3 shown, the target neural network includes an input layer u, an output layer y, and s reservoirs (i.e., Reservoir 1 to Reservoir s). The input layer u, Reservoir 1, …, Reservoir s, and the output layer y are connected in sequence. Among them, x1 ~x s respectively represent the state variables of reservoirs 1 to s. Schematically, the functional expression form of the target neural network is as follows:

[0056]

[0057] where x l (k + 1) is the state variable of reservoir l at time step k + 1, and α l is the neuron signal attenuation rate of reservoir l, x l (k) is the state variable of reservoir l at time step k, g is the activation function, is the weight of reservoir l, z l (k) is the input vector of reservoir l at time step k, u(k) is the input vector of the input layer at time step k, that is, the external input vector of the target neural network at time step k, y(k) is the output vector at time step k, x l-1 (k + 1) is the state variable of reservoir l - 1 at time step k + 1, is the output vector of the output layer at time step k, that is, the output vector of the target neural network at time step k, w out (k) is the weight of the output layer, x s (k) is the state variable of reservoir s at time step k, l = 1, 2, 3, ……, s, s is the number of reservoirs, and the time step represents each pass.

[0058] Step S220: Collect the data corresponding to the rolling force influence parameters and the current rolling data of the to-be-rolled pass.

[0059] In an embodiment of the present application, the data corresponding to the rolling force influence parameters includes at least one of the data corresponding to the rolling force influence parameters of the rolled passes and the data corresponding to the rolling force influence parameters of the to-be-rolled pass. The hot rolling process of a rolled piece includes multiple passes. The rolled passes refer to the passes that have been completed, and the to-be-rolled passes refer to the passes that are about to be rolled. The data corresponding to the rolling force influence parameters of the rolled passes refers to the specific values of the rolling force influence parameters collected at the rolling moment of the rolled passes. The data corresponding to the rolling force influence parameters of the to-be-rolled passes refers to the specific values of the rolling force influence parameters estimated or preset at the rolling moment of the to-be-rolled passes. The current rolling data of the to-be-rolled passes includes the true values of some rolling parameters (such as the entrance thickness of the rolled piece, the width of the rolled piece, etc.) before the start of rolling of the to-be-rolled passes, and the target values of some rolling parameters (such as the rolling temperature, the exit thickness of the rolled piece, etc.) during and after the rolling process of the to-be-rolled passes that are expected or estimated. The data corresponding to the rolling force influence parameters and the current rolling data of the to-be-rolled passes can be collected by various sensors or data acquisition systems installed on the hot rolling production line or the hot rolling automatic control system.

[0060] Step S230: Calculate the predicted value of the initial rolling force for the to-be-rolled pass based on the current rolling data of the to-be-rolled pass, and input the corresponding data of the rolling force influence parameters into the target neural network to obtain the corresponding estimated value of the rolling force correction coefficient for the to-be-rolled pass.

[0061] In an embodiment of the present application, a traditional hot rolling mathematical model can be used to calculate the rolling force based on the current rolling data of the to-be-rolled pass as the predicted value of the initial rolling force for the to-be-rolled pass. Through self-learning of the target neural network based on the corresponding data of the rolling force influence parameters, the output result of the target neural network is used as the corresponding estimated value of the rolling force correction coefficient for the to-be-rolled pass, improving the accuracy of the rolling force correction coefficient.

[0062] In an embodiment of the present application, calculating the predicted value of the initial rolling force for the to-be-rolled pass based on the current rolling data of the to-be-rolled pass includes: establishing a rolling force mathematical model for calculating the rolling force according to the width of the rolled piece, the roll diameter, the reduction of the rolled piece, the rolling temperature, the rolling strain, and the rolling strain rate; calculating the corresponding target value of the reduction of the rolled piece and the corresponding estimated value of the rolling strain for the to-be-rolled pass according to the current value corresponding to the entrance thickness of the rolled piece and the target value corresponding to the exit thickness of the rolled piece for the to-be-rolled pass, and calculating the corresponding estimated value of the rolling strain rate for the to-be-rolled pass according to the corresponding target value of the reduction of the rolled piece, the corresponding value of the roll diameter, and the corresponding target value of the rolling speed for the to-be-rolled pass, where the current rolling data of the to-be-rolled pass includes the current value corresponding to the entrance thickness of the rolled piece, the target value corresponding to the exit thickness of the rolled piece, the corresponding value of the roll diameter, and the corresponding target value of the rolling speed for the to-be-rolled pass; inputting the current value corresponding to the width of the rolled piece, the corresponding value of the roll diameter, the corresponding target value of the reduction of the rolled piece, the corresponding target value of the rolling temperature, the corresponding estimated value of the rolling strain, and the corresponding estimated value of the rolling strain rate for the to-be-rolled pass into the rolling force mathematical model to enable the rolling force mathematical model to perform rolling force calculation and obtain the predicted value of the initial rolling force for the to-be-rolled pass, where the current rolling data of the to-be-rolled pass further includes the current value corresponding to the width of the rolled piece and the corresponding target value of the rolling temperature for the to-be-rolled pass.

[0063] In this embodiment, the expression of the rolling force mathematical model can be as follows:

[0064]

[0065] where F is the predicted value of the initial rolling force, W is the width of the rolled piece, R′ is the roll radius, Δh is the reduction of the rolled piece, Q P is the stress state parameter, σ is the physical property parameter of the rolled piece, K1 and K2 are empirical coefficients, is the rolling strain rate, ε is the rolling strain. K1 and K2 can be preset, Q PIt can be preset or calculated according to existing formulas such as Shida Mao's formula. Schematically, Q P can be calculated as follows:

[0066]

[0067] where Q P is the stress state parameter, ε is the rolling strain, R′ is the roll radius, and h0 is the workpiece inlet thickness.

[0068] By inputting the current value corresponding to the workpiece width of the to-be-rolled pass, the corresponding value of the roll radius, the target value corresponding to the workpiece reduction, the target value corresponding to the rolling temperature, the estimated value corresponding to the rolling strain, and the estimated value corresponding to the rolling strain rate into the rolling force mathematical model shown in Equation (3), the predicted value of the initial rolling force for the to-be-rolled pass can be obtained. Among them, the calculation method of the workpiece reduction can be as follows:

[0069] Δh = h0 - h f Equation (5)

[0070] where Δh is the workpiece reduction, h0 is the workpiece inlet thickness, and h f is the workpiece outlet thickness. The calculation method of the rolling strain can be as follows:

[0071] ε = Δh / h0 Equation (6)

[0072] where ε is the rolling strain, Δh is the workpiece reduction, and h0 is the workpiece inlet thickness.

[0073] The calculation method of the rolling strain rate can be as follows:

[0074]

[0075] where is the rolling strain rate, v is the rolling speed, R′ is the roll radius, and Δh is the workpiece reduction.

[0076] The current value corresponding to the workpiece inlet thickness of the to-be-rolled pass and the target value corresponding to the workpiece outlet thickness can be substituted into Equation (5) to obtain the target value corresponding to the workpiece reduction of the to-be-rolled pass. The target value corresponding to the workpiece reduction of the to-be-rolled pass and the current value corresponding to the workpiece inlet thickness can be substituted into Equation (6) to obtain the estimated value corresponding to the rolling strain of the to-be-rolled pass. The target value corresponding to the rolling speed of the to-be-rolled pass, the corresponding value of the roll radius, and the target value corresponding to the workpiece reduction can be substituted into Equation (7) to obtain the estimated value corresponding to the rolling strain rate of the to-be-rolled pass.

[0077] In one embodiment of the present application, inputting the data corresponding to the rolling force influence parameters into the target neural network to obtain the estimated value corresponding to the rolling force correction coefficient for the to-be-rolled pass, including: inputting the data corresponding to the rolling force influence parameters of the already-rolled passes and the data corresponding to the rolling force influence parameters of the to-be-rolled pass into the target neural network to obtain the estimated value corresponding to the rolling force correction coefficient for the to-be-rolled pass. Among them, the data corresponding to the rolling force influence parameters of the to-be-rolled pass can be predicted through other modules of the hot rolling control system, such as the temperature calculation module, the rolling deformation parameter calculation module, and the rolling time calculation module, or obtained through presetting.

[0078] Step S240, correcting the predicted value of the initial rolling force for the to-be-rolled pass according to the estimated value corresponding to the rolling force correction coefficient for the to-be-rolled pass to obtain the predicted value of the target rolling force for the to-be-rolled pass.

[0079] In one embodiment of the present application, the estimated value corresponding to the rolling force correction coefficient for the to-be-rolled pass and the predicted value of the initial rolling force can be multiplied to calculate the predicted value of the target rolling force for the to-be-rolled pass. Schematically, the calculation method of the predicted value of the target rolling force can be as follows:

[0080] F’ = γF Equation (8)

[0081] Wherein, F’ is the predicted value of the target rolling force, γ is the estimated value corresponding to the rolling force correction coefficient, and F is the predicted value of the initial rolling force.

[0082] In one embodiment of the present application, the method further includes: obtaining the estimated value corresponding to the rolling force correction coefficient and the actual value corresponding to the rolling force correction coefficient for each already-rolled pass, wherein the estimated value corresponding to the rolling force correction coefficient for each already-rolled pass is obtained based on the output result of the target neural network, and the actual value corresponding to the rolling force correction coefficient for each already-rolled pass is determined based on the actual rolling data of each already-rolled pass; taking the difference between the estimated value corresponding to the rolling force correction coefficient of an already-rolled pass and the actual value corresponding to the rolling force correction coefficient as a fitting error to obtain a plurality of fitting errors; online adjusting the weights of the output layer based on the plurality of fitting errors, and offline adjusting the weights of each reservoir based on the plurality of fitting errors to complete the parameter update of the target neural network.

[0083] The basic echo state network is randomly initialized and the reservoir weights are fixed. Only the weights of the output layer are trained (adjusted), enabling the rapid training of the neural network. Therefore, it has a fast calculation speed and great potential in online control, and can be well applied to the rolling process with high requirements for speed. However, the randomly initialized weights may not be suitable for a specific nonlinear system. Since the basic echo state network only trains the weights of the output layer while fixing the reservoir weights, its fitting performance is not good. Moreover, when there is input noise, only adjusting the weights of the output layer while fixing the reservoir weights may also cause the neural network to be sensitive to input noise. Therefore, the basic echo state network has poor adaptability to the hot rolling production line with many unknown interference factors. In this embodiment, by adopting the method of online adjusting the weights of the output layer and offline adjusting the weights of the reservoir, not only can the calculation speed of the target neural network be ensured, but also the fitting performance of the target neural network can be improved, the influence of noise on the neural network can be reduced, and the estimation accuracy of the rolling force correction coefficient of the target neural network can be further improved, thereby further improving the accuracy of rolling force prediction and better adapting to the hot rolling production line with complex interference factors.

[0084] In this embodiment, after the rolling of each pass is completed, the actual rolling data of this pass (at this time, this pass is the rolled pass) real-time feedback by the rolling mill can be collected, including the actual rolling force of the rolling mill, the actual values corresponding to the rolling temperature, the actual values corresponding to the reduction of the rolled piece, the actual values corresponding to the rolling speed, etc. By calculation, the actual values corresponding to the rolling strain and the rolling strain rate can be obtained, and then substituting them into the rolling force mathematical model, the rolling force output by the rolling force mathematical model can be obtained. There is a deviation between this rolling force and the actual rolling force of the rolling mill, and the actual rolling force correction coefficient, that is, the actual value corresponding to the rolling force correction coefficient, can be calculated based on the two.

[0085] The estimated value corresponding to the rolling force correction coefficient is the output result of the target neural network. The results output by the target neural network for each pass can be stored as the estimated value corresponding to the rolling force correction coefficient of this pass.

[0086] The calculation method of the fitting error can be as follows:

[0087]

[0088] Among them, e(k) is the fitting error, is the estimated value corresponding to the rolling force correction coefficient, that is, the output result of the target neural network, y r (k) is the actual value corresponding to the rolling force correction coefficient, and k is the time step, representing each pass.

[0089] The estimated value corresponding to the rolling force correction coefficient and the actual value corresponding to the rolling force correction coefficient of each rolled pass can be substituted into Equation (9) to obtain the fitting error of this rolled pass and store it.

[0090] After the rolling of each pass is completed, the weights of the output layer can be updated based on the multiple fitting errors obtained, so as to realize the online adjustment of the weights of the output layer. After the rolling of all passes of the rolled piece is completed, or when the rolling mill stops operating, the weights of each reservoir can be updated based on the multiple fitting errors, so as to realize the offline adjustment of the reservoir weights.

[0091] In an embodiment of the present application, the online adjustment of the weights of the output layer based on multiple fitting errors includes:

[0092] After the rolling of each rolling pass is completed, form an error matrix from the multiple fitting errors, and form a state variable matrix from the state variables corresponding to each reservoir; determine the new weights of the output layer according to the error matrix and the state variable matrix;

[0093] Or,

[0094] After the rolling of each rolling pass is completed, determine the minimum network fitting error based on the multiple fitting errors, the weights of the output layer, and the weights of each reservoir; determine the new weights of the output layer according to the minimum network fitting error and the weights of the output layer.

[0095] In some embodiments, the weight update method of the output layer can be as follows:

[0096]

[0097] Among them, w out (k + 1) is the new weight of the output layer, X l is the state variable matrix, λ is the regularization parameter, λ ∈ R, I is the identity matrix, E is the error matrix, e(k), e(k - 1), …, e(k - p) are the fitting errors at each time step (pass), x l (k), x l (k - 1), …, x l (k - p) are the state vectors of reservoir l at each time step (pass), l = 1, 2, 3, ……, s, and s is the number of reservoirs.

[0098] In other embodiments, the weight update method of the output layer can be as follows:

[0099]

[0100] Among them, w out (k + 1) is the new weight of the output layer, w out (k) is the weight of the output layer, δ is the network learning step size, C(w) is the minimum network fitting error, e(k) is the fitting error, λ is the regularization parameter, and w is the weight parameter to be optimized. is the weight of reservoir l, where l = 1, 2, 3, ……, s, and s is the number of reservoirs.

[0101] In one embodiment of the present application, the weights of each reservoir are adjusted offline based on multiple fitting errors, including: after the rolling of all rolling passes is completed, determining the minimum network fitting error based on multiple fitting errors, the weights of the output layer, and the weights of each reservoir; and determining the new weights of each reservoir according to the minimum network fitting error and the weights of each reservoir.

[0102] In this embodiment, the weight update method of the reservoir can be as follows:

[0103]

[0104] where is the new weight of reservoir l, is the weight of reservoir l, δ is the network learning step size, C(w) is the minimum network fitting error, e(k) is the fitting error, λ is the regularization parameter, w is the weight parameter to be optimized, w out (k) is the weight of the output layer, where l = 1, 2, 3, ……, s, and s is the number of reservoirs.

[0105] In some embodiments, the target neural network can be connected to the rolling force mathematical model in a multiplication form to form a model-data hybrid-driven rolling force calculation model.

[0106] Please refer to Figure 4 , Figure 4 is the flowchart of the rolling force prediction of the rolling force calculation model shown in a specific embodiment of the present application. As Figure 4 shown, the rolling force calculation model is composed of a target neural network and a traditional hot rolling mathematical model in a multiplication form, where the traditional hot rolling mathematical model is an example of the rolling force mathematical model. The rolling force prediction process of the rolling force calculation model is as follows:

[0107] 1. Collect the measured data of the hot rolling process, including the corresponding data of the rolling force influence parameters of the rolled passes and the to-be-rolled passes, and the current rolling data of the to-be-rolled passes;

[0108] 2. Input the current rolling data of the to-be-rolled passes, including the entrance thickness, exit thickness, rolling temperature, roll diameter, etc., into the traditional hot rolling mathematical model for rolling force calculation to obtain the initial rolling force prediction value; at the same time, input the corresponding data of the rolling force influence parameters of the rolled passes and the to-be-rolled passes, including the rolling interval duration, rolling interval temperature, rolling strain, rolling strain rate, rolling temperature, workpiece width, roll diameter, the deviation of the gold element content between the workpiece and the standard part, the ratio of the contact arc length in the deformation zone to the average height of the workpiece, etc., into the target neural network for self-learning to obtain the estimated value of the rolling force correction coefficient;

[0109] 3. Calculate the initial rolling force prediction value output by the traditional hot rolling mathematical model and the corresponding estimated value of the rolling force correction coefficient output by the target neural network in a multiplicative form to obtain the target rolling force prediction value.

[0110] 4. Online adjust the weights of the output layer of the target neural network and offline adjust the reservoir weights of the target neural network based on the actual rolling data fed back by the rolling mill and the output results of the target neural network.

[0111] In this specific embodiment, for the detailed process of rolling force prediction, please refer to the descriptions in the foregoing various embodiments, and details will not be repeated here. By establishing a model-data hybrid-driven rolling force calculation model and adopting an online-offline combined method to adaptively adjust the rolling force calculation process, this specific embodiment can improve the rolling force prediction accuracy and the quality of hot-rolled products. Moreover, the target neural network has strong versatility, can be embedded in various existing models, and can be transplanted to different production lines.

[0112] Please refer to Figure 5 , Figure 5 which is a block diagram of a hot rolling force prediction device shown in an exemplary embodiment of the present application. This device can be applied to Figure 1 the implementation environment shown in

[0113] and is specifically configured in the computer device 120. This device can also be applicable to other exemplary implementation environments and can be specifically configured in other devices. This embodiment does not limit the implementation environment applicable to this device. Figure 5 As shown in

[0114] In this embodiment, the data acquisition module 510 may be various sensors installed on a hot rolling production line or a hot rolling automatic control system, or may be a data acquisition system for collecting data from various sensors on the hot rolling production line or the hot rolling automatic control system, or may be a data receiver or a network card; the first information processing module 520 and the second information processing module 530 may be a computer, a computing cluster, a microcomputer, an embedded computer, a neural network computer, a processor, etc., and the data correction module 540 may be a multiplier, a DSP (Digital Signal Processor), etc., and no limitations are imposed here. The first information processing module 520 and the second information processing module 530 may be integrated in the same computer device, or they may be independent computer devices, and no limitations are imposed here either.

[0115] It should be noted that the hot rolling rolling force prediction device provided in the above embodiment and the hot rolling rolling force prediction method provided in the above embodiment belong to the same concept. The specific manners in which each module and unit perform operations have been described in detail in the method embodiment, and will not be elaborated here. In actual application, the hot rolling rolling force prediction device provided in the above embodiment may, according to needs, allocate the above functions to different functional modules, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and no limitations are imposed here either.

[0116] Please refer to Figure 6 and Figure 7 , as Figure 6 shown, in a specific embodiment of the present application, a hot rolling rolling force prediction system is further provided. The hot rolling rolling force prediction system includes a network function management module, a data conversion module, a data management module, an algorithm module, and a memory space. Among them, the network function management module is used to set the hyperparameters of the neural network, generate a network of a specified type, and manage the calls between modules according to usage requirements, and is the core trigger; the data management module contains a custom data structure, performs memory management, and provides appropriate data for the calculation of the neural network; the algorithm module provides a calculation function, contains a variety of algorithms, can perform online and offline operations, and provides the best fitting effect; the data conversion module performs the read / write and content conversion functions for the outside, and provides data in the format required for the neural network operation to the data management module.

[0117] As Figure 7 shown, the working process of the hot rolling rolling force prediction system is as follows:

[0118] 1. Network initialization: Select the input and output parameters of the neural network according to the constructed non-linear dynamic system; Initialize the neural network through the network function management module, including the number of neurons in the input layer, the number of neurons in the output layer, the type of activation function, the network depth, the number of neurons in each layer of the network, the neuron signal attenuation rate, the learning step size, the regularization parameter, the reservoir weight, the output layer weight, etc., and select the method of online optimizing the output layer weight and offline optimizing the reservoir weight; Call the data management module through the network function management module to allocate memory space for the use of the network.

[0119] 2. Data acquisition and data processing: The traditional hot rolling automatic control system is responsible for collecting the measured data of the hot rolling production line, including the furnace outlet temperature, the roller table transportation time, the temperature of the steel plate during roller table transportation, the rolling temperature of each pass, the inlet thickness, the outlet thickness, the rolling speed, the preset rolling force, the actual rolling force, the rolling interval time (rolling break time) between passes, the rolling interval temperature (rolling break temperature), the work roll diameter, etc., and is responsible for forwarding these contents; The network function management module triggers the data conversion module to convert the received raw data, form a network input subset from the converted data, and hand it over to the data management module for management. This network input subset will form a set D with the network output subset, and the network output subset is composed of the output of the neural network each time.

[0120] 3. Online adjustment of network parameters: After each pass is completed, the network function management module triggers the algorithm module to read the data in set D from the data management module and update the output layer weight of the network based on the neural network parameter configuration.

[0121] 4. Prediction value correction: Based on the data collected and processed in step 2, the network function management module calls the algorithm module to calculate the initial rolling force prediction value of the subsequent passes to be rolled using the traditional hot rolling mathematical model, then outputs the rolling force correction coefficient through the neural network, and then calculates and outputs the target rolling force prediction value by multiplication.

[0122] 5. Offline adjustment of network parameters: After all passes are completed, the network function management module triggers the algorithm module to read set D from the data management module and update the reservoir weight.

[0123] 6. Data saving: After the reservoir weight is updated, save the output layer weight, the reservoir weight, and other hyperparameters to the data management module for future use.

[0124] 7. Close the network: The network function management module triggers the data conversion module to save the data to an external unit, triggers the data management module to recycle the memory, and closes the neural network to stop the service.

[0125] This embodiment also provides an electronic device, including: one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the hot rolling force prediction method provided in each of the above embodiments.

[0126] Please refer to Figure 8 , Figure 8 which is a schematic structural diagram of an electronic device shown in an exemplary embodiment of the present application. It should be noted that Figure 8 the electronic device 800 shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present application.

[0127] As Figure 8 shown, the electronic device 800 includes a processor 801, a memory 802, and a communication bus 803; the communication bus 803 is used to connect the processor 801 and the memory 802; the processor 801 is used to execute the computer program stored in the memory 802 to implement one or more of the methods in the above embodiments.

[0128] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer is caused to execute the hot rolling force prediction method as described above. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist alone without being assembled into the electronic device.

[0129] This embodiment also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to execute the hot rolling force prediction method provided in each of the above embodiments.

[0130] The electronic device provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication therebetween. The memory is used to store a computer program, the communication interface is used for communication, and the processor and the transceiver are used to run the computer program to cause the electronic device to execute each step of the above method.

[0131] In this embodiment, the memory may include a random access memory (Random Access Memory, abbreviated as RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0132] The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU for short), a Network Processor (NP for short), etc.; it may also be a Digital Signal Processor (DSP for short), an Application Specific Integrated Circuit (ASIC for short), a Field-Programmable Gate Array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0133] For the computer-readable storage medium in this embodiment, those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to a computer program. The foregoing computer program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM (Read Only Memory), RAM (Random Access Memory), magnetic disk, or optical disc that can store program codes.

[0134] The above embodiments only exemplarily illustrate the principles and effects of the present application, rather than limiting the present application. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present application should still be covered by the claims of the present application.

Claims

1. A hot rolling force prediction method, characterized in that: The method comprises: The target neural network is pre-built with the rolling force influencing parameters as network input and the rolling force correction coefficient as network output; Collect the corresponding data of rolling force influencing parameters and the current rolling data of the waiting rolling pass; Based on the current rolling data of the to-be-rolled pass, the initial rolling force prediction value of the to-be-rolled pass is calculated, and the corresponding data of the rolling force influencing parameter is input into the target neural network to obtain the corresponding estimated value of the rolling force correction coefficient of the to-be-rolled pass; According to the estimated value corresponding to the rolling force correction coefficient of the to-be-rolled pass, the initial rolling force prediction value of the to-be-rolled pass is corrected to obtain the target rolling force prediction value of the to-be-rolled pass.

2. The hot rolling force prediction method according to claim 1, characterized in that: Calculating the predicted value of the initial rolling force of the to-be-rolled pass based on the current rolling data of the to-be-rolled pass, comprising: A rolling force mathematical model for calculating rolling force is established according to the width of the rolled piece, the diameter of the roll, the reduction of the rolled piece, the rolling temperature, the rolling strain and the rolling strain rate; According to the current value corresponding to the thickness of the rolled piece at the entrance of the to-be-rolled pass and the target value corresponding to the thickness of the rolled piece at the exit of the to-be-rolled pass, the target value corresponding to the reduction amount of the rolled piece and the estimated value corresponding to the rolling strain of the to-be-rolled pass are calculated, and according to the target value corresponding to the reduction amount of the rolled piece at the to-be-rolled pass, the corresponding numerical value of the roll diameter and the corresponding target value of the rolling speed, the estimated value corresponding to the rolling strain rate of the to-be-rolled pass is calculated, wherein the current rolling data of the to-be-rolled pass include the current value corresponding to the thickness of the rolled piece at the entrance of the to-be-rolled pass, the target value corresponding to the thickness of the rolled piece at the exit of the to-be-rolled pass, the corresponding numerical value of the roll diameter and the corresponding target value of the rolling speed; The current value corresponding to the width of the workpiece to be rolled, the corresponding value of the roll diameter, the target value corresponding to the workpiece reduction, the target value corresponding to the rolling temperature, the estimated value corresponding to the rolling strain and the estimated value corresponding to the rolling strain rate of the pass to be rolled are input into the rolling force mathematical model, so that the rolling force mathematical model calculates the rolling force to obtain the initial rolling force prediction value of the pass to be rolled, wherein the current rolling data of the pass to be rolled also includes the current value corresponding to the width of the workpiece to be rolled and the target value corresponding to the rolling temperature of the pass to be rolled.

3. The hot rolling force prediction method according to claim 1, characterized in that: The target neural network is pre-built with the rolling force influencing parameters as network input and the rolling force correction coefficient as network output, including: Selecting parameters that affect subdynamic recrystallization and rolling deformation before rolling deformation as the rolling force influencing parameters; A nonlinear dynamic system is established with historical and current rolling force influencing parameters as system input and the current rolling force correction coefficient as system output, and noise and the historical rolling force correction coefficient are introduced as variables into the nonlinear dynamic system; A neural network structure is constructed based on the nonlinear dynamic system to obtain the target neural network.

4. The hot rolling force prediction method according to claim 3, characterized in that: The neural network structure is constructed based on the nonlinear dynamic system to obtain the target neural network, including: constructing an input layer based on the system input of the nonlinear dynamic system, constructing an output layer based on the system output of the nonlinear dynamic system, and constructing at least one reservoir layer according to historical and current rolling force influencing parameters and historical rolling force correction coefficients; A connection relationship between the input layer, the output layer and each reservoir is established, and weights are assigned to the output layer and each reservoir to obtain the target neural network.

5. The hot rolling force prediction method according to claim 4, characterized in that: The method further comprises: Obtaining an estimated value corresponding to a rolling force correction coefficient of each rolled pass and an actual value corresponding to the rolling force correction coefficient, wherein the estimated value corresponding to the rolling force correction coefficient of each rolled pass is obtained based on an output result of the target neural network, and the actual value corresponding to the rolling force correction coefficient of each rolled pass is determined based on actual rolling data of each rolled pass; The difference between the estimated value corresponding to the rolling force correction coefficient of a rolling pass and the actual value corresponding to the rolling force correction coefficient is taken as a fitting error to obtain a plurality of fitting errors; The weight of the output layer is adjusted online based on a plurality of fitting errors, and the weight of each reservoir is adjusted offline based on a plurality of fitting errors, thereby completing the parameter update of the target neural network.

6. The hot rolling force prediction method according to claim 5, characterized in that: The weight of the output layer is adjusted online based on a plurality of fitting errors, including: After each rolling pass is completed, multiple fitting errors are formed into an error matrix, and state variables corresponding to each reservoir are formed into a state variable matrix; a new weight of the output layer is determined according to the error matrix and the state variable matrix; or, After each rolling pass is completed, the minimum network fitting error is determined based on multiple fitting errors, the weight of the output layer and the weight of each reservoir; and the new weight of the output layer is determined according to the minimum network fitting error and the weight of the output layer.

7. The hot rolling force prediction method according to claim 5, characterized in that: The weights of each reservoir are adjusted offline based on multiple fitting errors, including: After all rolling passes are completed, a minimum network fitting error is determined based on a plurality of fitting errors, a weight of the output layer, and a weight of each reservoir layer; The new weight of each reservoir is determined according to the minimum network fitting error and the weight of each reservoir.

8. A hot rolling force prediction device, characterized in that: The device comprises: A data acquisition module is used to collect data corresponding to rolling force influencing parameters and current rolling data of the to-be-rolled passes; A first information processing module, configured to calculate a predicted value of an initial rolling force of the to-be-rolled pass based on current rolling data of the to-be-rolled pass; A second information processing module is used to input the rolling force influencing parameter corresponding data into a target neural network to obtain an estimated value corresponding to the rolling force correction coefficient of the to-be-rolled pass, wherein the target neural network is pre-constructed by taking the rolling force influencing parameter as the network input and the rolling force correction coefficient as the network output; The data correction module is used to correct the initial rolling force prediction value of the to-be-rolled pass according to the rolling force correction coefficient corresponding to the estimated value of the to-be-rolled pass, so as to obtain the target rolling force prediction value of the to-be-rolled pass.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the electronic device to implement the hot rolling force prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the hot rolling force prediction method according to any one of claims 1 to 7.

Images