A multi-level and multi-stage coordinated optimization control method for the entire rolling process

By employing hybrid neural networks and multi-level, multi-stage coordinated optimization control methods, the problems of real-time data integration and dynamic adjustment during hot continuous rolling were solved, thereby improving production efficiency and product quality.

CN118719820BActive Publication Date: 2024-11-19TAIYUAN UNIVERSITY OF SCIENCE AND TECHNOLOGY
View PDF 11 Cites 0 Cited by

Patent Information

Application Number
CN202411216660.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-11-19
Estimated Expiration
2044-09-02

AI Technical Summary

Technical Problem

Existing technologies fail to effectively integrate real-time monitoring data for immediate response and adjustment during hot continuous rolling, resulting in uneven resource allocation and low production efficiency, and making it impossible to achieve comprehensive optimization across multiple rolling stages.

Method used

A hybrid neural network model combined with a computer model of rolling process parameters is adopted. Rolling force data features are extracted through one-dimensional convolutional neural networks and recurrent neural networks. By combining Pareto front method and entropy weight decision technology, rolling parameters are dynamically adjusted to achieve multi-level and multi-stage coordinated optimization.

Benefits of technology

It improved production efficiency, reduced energy consumption, ensured or improved product quality, and enhanced system response speed and reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118719820B_ABST
    Figure CN118719820B_ABST
Patent Text Reader

Abstract

The present invention relates to a multi-level and multi-stage coordinated optimization control method for the entire rolling process, comprising: collecting target data in the rolling production process; inputting the target data into a rolling process parameter calculation model to obtain rolling force; inputting the target data into a hybrid neural network model to obtain a correction value; the hybrid neural network model comprises: a one-dimensional convolutional neural network model and a recurrent neural network model; the hybrid neural network model is obtained by training with a training set; according to the rolling force combined with the correction value, a corrected rolling force is obtained; a multi-level and multi-stage coordinated optimization model is constructed, the corrected rolling force is input into the multi-level and multi-stage coordinated optimization model, rolling parameters are dynamically adjusted, and a Pareto frontier method is introduced to weigh multiple conflicting objectives, and an entropy weight decision method and TOPSIS technology are further combined to determine a final optimization decision plan.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of automatic control in a steel manufacturing process, and in particular to a multi-level and multi-stage coordinated optimization control method for the entire rolling process. Background Art

[0002] As a key step in steel production, hot rolling has a decisive influence on the performance of the final product due to its operating efficiency and product quality. Traditional rolling control methods mostly focus on the optimization of a single stage, such as independently optimizing the rough rolling or finishing rolling process. This method often ignores the mutual influence between the various rolling stages, which may lead to uneven resource allocation and low production efficiency. In the prior art, although there have been attempts to use multi-objective optimization algorithms and swarm intelligence algorithms, such as differential algorithms and particle swarm optimization, to improve the efficiency of load distribution and plate quality control. However, these methods often fail to fully consider the dynamic coordination between the various rolling stages, and fail to effectively integrate real-time monitoring data for immediate response and adjustment. In addition, the methods in the prior art are usually unable to achieve highly integrated data processing and comprehensive optimization across multiple rolling stages. Summary of the invention

[0003] The purpose of the present invention is to solve the optimization problem of the rough rolling zone, finishing rolling zone and coiling stage in the hot rolling process, and propose a multi-level and multi-stage coordinated optimization control method for the entire rolling process, which is suitable for load distribution and coordinated control in the rough rolling zone, finishing rolling zone and coiling stage, improves the efficiency of the overall production process and product quality, and constructs an accurate rolling characteristic parameter calculation model through real-time acquisition and analysis of rolling force data, thereby realizing optimized control of the rolling process. The present invention pays special attention to improving the comprehensive management of the load distribution of the rolling mill and the quality of the plate and strip through advanced control strategies and coordinated optimization methods.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A multi-level and multi-stage coordinated optimization control method for the entire rolling process, comprising:

[0006] Collect target data during rolling production;

[0007] Inputting the target data into a rolling process parameter calculation model to obtain rolling force;

[0008] Input the target data into a hybrid neural network model to obtain a correction value; the hybrid neural network model includes: a one-dimensional convolutional neural network and a recurrent neural network model; the hybrid neural network model is obtained by training with a training set;

[0009] According to the rolling force and the correction value, a corrected rolling force is obtained;

[0010] A multi-level and multi-stage coordinated optimization model is constructed, the corrected rolling force is input into the multi-level and multi-stage coordinated optimization model, the rolling parameters are dynamically adjusted, and the Pareto frontier method is introduced to weigh multiple conflicting objectives. The entropy weight decision-making method and TOPSIS technology are further combined to determine the final optimization decision plan.

[0011] Optionally, the method for constructing the rolling process parameter calculation physical model is:

[0012]

[0013] Where: is the rolling force; is bandwidth; To consider the horizontal projection length of the contact arc between the roller and the workpiece after flattening; is the influence coefficient of considering the stress state in the deformation zone; is the metal deformation resistance, is the influence coefficient of the front and rear tensile stress on the rolling force;

[0014] Consider the influence coefficient of the stress state in the deformation zone:

[0015]

[0016] Where: is the roller radius; , is the inlet and outlet thickness of each rack; is the pressing amount of the rack;

[0017] Rolling power model:

[0018]

[0019] Where: is the rolling torque; is the rolling power; is the roller speed; is the rolling force; is the force arm; is the force arm coefficient.

[0020] Optionally, constructing the hybrid neural network model includes:

[0021]

[0022] in, Output for the hybrid model; Output of CNN model; Output of the RNN model; and is the fusion coefficient; is the bias term; is the activation function.

[0023] Optionally, obtaining the hybrid neural network model by training with a training set includes:

[0024] The historical rolling force data in the training set is input into the one-dimensional convolutional neural network, and the local features of the sequence data in the historical rolling force data are extracted. The recurrent neural network model is used to capture the long-term dependencies in the local features, so as to identify the dynamic change trends occurring in the continuous rolling process, and an adaptive genetic algorithm is used to optimize the hyperparameters of the one-dimensional convolutional neural network and the recurrent neural network model, and the optimal network parameter configuration is automatically found to obtain the hybrid neural network model.

[0025] Optionally, the method for obtaining the corrected rolling force is:

[0026]

[0027] in, It is the rolling force calculated by the traditional physical model; is the corrected rolling force; represents the rolling force predicted by the hybrid network model based on historical data; is the correction value predicted by the hybrid neural network model; is the correction factor.

[0028] Optionally, the multi-level and multi-stage coordinated optimization model includes:

[0029] The rough rolling load distribution coordination optimization sub-model is used to ensure the quality of the strip in the final rolling pass and make load distribution decisions to control the thickness and crown of the strip:

[0030]

[0031] in, for the goal; is the rolling force of the i-th pass; is the average rolling force of all passes; is the number of rolling passes; is the actual measured convexity of the intermediate billet; is the target convexity value; is the total energy consumption in the rough rolling stage; For the Power of the pass;

[0032] The finishing rolling load distribution sub-model is used to receive the data of the rough rolling load distribution coordination optimization sub-model, and perform load balancing and plate shape optimization based on the data and the real-time monitoring data of the finishing rolling process:

[0033]

[0034] in, for the goal; For the The actual exit thickness of the pass; is the target thickness; is the rolling pass; is the rolling line speed; is the total energy consumed in this phase; The total amount of production.

[0035] Optionally, the multi-level and multi-stage coordinated optimization model further includes:

[0036] The coiling stage sub-model is used to adjust the coiling tension according to the optimization results of the finishing load distribution sub-model:

[0037]

[0038] in, for the goal; is the actual measured tension; To set the tension; is the actual measured coil diameter; is the target roll diameter.

[0039] Optionally, the Pareto frontier method is introduced to balance multiple conflicting objectives, including:

[0040] The Pareto frontier method is introduced to balance the minimization of rolling force, reduction of energy consumption, improvement of production efficiency and assurance of product quality:

[0041]

[0042] in, and is a decision variable. If there is no other solution , so that all objective functions are The value at is no greater than , and at least one objective function is The value is better than The value of is considered to be Pareto optimal. is the objective function defined in the decision space.

[0043] Optionally, the method for determining the final optimization decision solution is:

[0044]

[0045] in, is the comprehensive decision score; is the number of decision criteria; For the The entropy weight of each decision criterion; For the Rating of decision criteria; For the The relative proximity of the standards; and is the distance between the goal and the positive and negative ideals; For the The decision in Ratings on the criteria; Provide a specific score for each decision; For the A specific score for a decision.

[0046] The beneficial effects of the present invention are:

[0047] The present invention can integrate real-time data and advanced optimization algorithms to achieve real-time dynamic optimization in the production process, thereby improving production efficiency, reducing energy consumption, and ensuring or improving product quality.

[0048] The present invention optimizes the dynamic data processing and multi-stage decision-making process during the rolling process, predicts and adjusts production parameters through advanced data analysis and machine learning algorithms, reduces waste in production, and improves resource utilization efficiency. In addition, the multi-stage coordinated optimization characteristics of the method enable each subsystem to perform real-time decision processing locally, thereby reducing the need for central processing and improving the response speed and reliability of the system.

[0049] In order to adapt to different industrial environments and specific needs, the present invention can have multiple variations. For example, the data acquisition and processing modules can be customized according to specific production equipment and sensor technology, and the optimization model and algorithm can be adjusted according to different production goals and constraints. In addition, the architecture of the entire system can be modularized and easily upgraded and expanded to adapt to rapidly changing technology and market needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0051] Figure 1 A flow chart of a multi-level and multi-stage coordinated optimization control method for the entire rolling process according to an embodiment of the present invention;

[0052] Figure 2 A flowchart of the part of constructing a hybrid neural network to realize rolling force prediction and real-time correction of the calculation model in an embodiment of the present invention;

[0053] Figure 3 This is a flowchart of the part of constructing a multi-level and multi-stage coordinated optimization model in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] like Figure 1 As shown, this embodiment discloses a multi-level and multi-stage coordinated optimization control method for the entire rolling process, including: collecting target data in the rolling production process; constructing a rolling process parameter calculation physical model, inputting the target data into the rolling process parameter calculation physical model, and obtaining the rolling force; constructing a hybrid neural network model; the hybrid neural network model includes: a one-dimensional convolutional neural network model and a recurrent neural network model; the hybrid neural network model is obtained by training with a training set; inputting the target data into the rolling process parameter calculation physical model to obtain a correction value; obtaining the corrected rolling force according to the rolling force combined with the correction value; constructing a multi-level and multi-stage coordinated optimization model, inputting the corrected rolling force into the multi-level and multi-stage coordinated optimization model, dynamically adjusting the rolling parameters, and introducing the Pareto frontier method to weigh multiple conflicting objectives, and further combining the entropy weight decision-making method and TOPSIS technology to determine the final optimization decision plan.

[0057] The method for constructing the rolling process parameter calculation physical model is:

[0058]

[0059] Where: is the rolling force; is bandwidth; To consider the horizontal projection length of the contact arc between the roller and the workpiece after flattening; is the influence coefficient of considering the stress state in the deformation zone; is the metal deformation resistance, is the influence coefficient of the front and rear tensile stress on the rolling force;

[0060] Consider the influence coefficient of the stress state in the deformation zone:

[0061]

[0062] Where: is the roller radius; , is the inlet and outlet thickness of each rack; is the pressing amount of the rack;

[0063] Rolling power model:

[0064]

[0065] Where: is the rolling torque; is the rolling power; is the roller speed; is the rolling force; is the force arm; is the force arm coefficient.

[0066] Building a hybrid neural network model includes:

[0067]

[0068] in, Output for the hybrid model; Output of CNN model; Output of the RNN model; and is the fusion coefficient; is the bias term; is the activation function.

[0069] The hybrid neural network model is obtained by training with a training set, including: inputting the historical rolling force data in the training set into a one-dimensional convolutional neural network, extracting the local features of the sequence data in the historical rolling force data, capturing the dynamic characteristics of the sequence data in the local features using a recurrent neural network model, and optimizing the hyperparameters of the one-dimensional convolutional neural network and the recurrent neural network model using an adaptive genetic algorithm, automatically finding the optimal network parameter configuration, and thus obtaining a hybrid neural network model.

[0070] The method to obtain the corrected rolling force is:

[0071]

[0072] in, It is the rolling force calculated by the traditional physical model; is the corrected rolling force; represents the rolling force predicted by the hybrid network model based on historical data; is the correction value predicted by the hybrid neural network model; is the correction factor.

[0073] The multi-level and multi-stage coordinated optimization model includes:

[0074] The rough rolling load distribution coordination optimization sub-model is used to ensure the quality of the strip in the final rolling pass and make load distribution decisions to control the thickness and crown of the strip:

[0075]

[0076] in, for the goal; is the rolling force of the i-th pass; is the average rolling force of all passes; is the number of rolling passes; is the actual measured convexity of the intermediate billet; is the target convexity value; is the total energy consumption in the rough rolling stage; For the Power of the pass;

[0077] The finishing rolling load distribution sub-model is used to receive the data of the rough rolling load distribution coordination optimization sub-model, and perform load balancing and plate shape optimization based on the data and the real-time monitoring data of the finishing rolling process:

[0078]

[0079] in, for the goal; For the The actual exit thickness of the pass; is the target thickness; is the rolling pass; is the rolling line speed; is the total energy consumed in this phase; The total amount of production.

[0080] The multi-level and multi-stage coordinated optimization model also includes:

[0081] The coiling stage submodel is used to adjust the coiling tension according to the optimization results of the finishing load distribution submodel:

[0082]

[0083] in, for the goal; is the actual measured tension; To set the tension; is the actual measured coil diameter; is the target roll diameter.

[0084] The Pareto frontier method is introduced to balance the minimization of rolling force, reduction of energy consumption, improvement of production efficiency and product quality assurance:

[0085]

[0086] in, and is a decision variable. If there is no other solution , so that all objective functions are The value at is no greater than , and at least one objective function is The value is better than The value of is considered to be Pareto optimal. is the objective function defined in the decision space.

[0087] The method to determine the final optimization decision plan is:

[0088]

[0089] in, is the comprehensive decision score; is the number of decision criteria; For the The entropy weight of each decision criterion; For the Rating of decision criteria; For the The relative proximity of the standards; and is the distance between the goal and the positive and negative ideals; For the The decision in Ratings on the criteria; Provide a specific score for each decision; For the A specific score for a decision.

[0090] This embodiment discloses a multi-level and multi-stage coordinated optimization control method for the entire rolling process, including:

[0091] Data acquisition, computational model construction and hybrid neural network model construction:

[0092] In one embodiment of the present invention, the sensor system deployed at each key position of the hot rolling mill begins to collect data in real time, including but not limited to rolling force, rolling speed and material temperature. These data are transmitted to the central processing unit in real time through the industrial Internet, specifically including the following steps:

[0093] Step S1: The data sets acquired in real time are transmitted to the data processing center, and these data will be used for subsequent analysis and model building.

[0094] Step S2: Construct a rolling force model, a rolling torque model and a rolling power model in combination with the acquired real-time rolling data. The specific implementation method is as follows:

[0095] Rolling force model:

[0096]

[0097] Where: is the rolling force; is bandwidth; To consider the horizontal projection length of the contact arc between the roller and the workpiece after flattening; is the influence coefficient of considering the stress state in the deformation zone; It is the metal deformation resistance, which is determined by the chemical composition of the metal material and the physical conditions of deformation; is the influence coefficient of the front and rear tensile stress on the rolling force.

[0098]

[0099] Where: is the roller radius; , is the inlet and outlet thickness of each rack; is the amount of rack depression.

[0100] Rolling power model:

[0101]

[0102] Where: is the rolling torque; is the rolling power; is the roller speed; is the rolling force; is the force arm; is the force arm coefficient.

[0103] like Figure 2As shown, step S3: in the central processing unit, the rolling characteristic parameter calculation model constructed in step S2 is initialized and the hybrid neural network model is trained. This model combines the rolling force model, the rolling torque model and the rolling power model to adjust and optimize for specific rolling conditions and material properties.

[0104] This embodiment uses a hybrid neural network that combines convolutional neural network (CNN) and traditional recurrent neural network (RNN) technology to process and analyze multidimensional data collected during the rolling process. The hybrid model aims to make full use of the advantages of both networks to improve the ability and accuracy of processing complex rolling data.

[0105] First, refine feature extraction. The convolutional neural network is responsible for extracting spatial features from real-time monitoring data. By applying multiple layers of convolution and pooling operations, CNN can effectively identify and extract key features that affect rolling quality and efficiency.

[0106] Subsequently, a recurrent neural network (RNN) was used to process the features extracted from the CNN to resolve the sequential dependencies in the data. Although the data here is not a strict time series, the use of RNN is still intended to identify and understand the patterns and dependencies that emerge during the continuous rolling process. The network is trained through the back-propagation algorithm, using cross entropy as the loss function, and the Adam optimizer is used for parameter optimization.

[0107] The specific method of building a rolling force regression prediction model is as follows: a hybrid prediction model that introduces multiple neural networks is built, and the model hyperparameters are optimized using an adaptive genetic algorithm (AGA), and the model is saved to ensure the accuracy of the real-time output rolling force and to correct the rolling force mechanism calculation model in real time. The specific method of building the hybrid model is as follows:

[0108]

[0109] Where: Output for the hybrid model; Output of CNN model; Output of the RNN model; and is the fusion coefficient; is the bias term; is the activation function.

[0110] The core of step S3 is to introduce a hybrid correction formula based on physics and data:

[0111]

[0112] Where: It is the rolling force calculated by the traditional physical model; is the corrected rolling force; represents the rolling force predicted by the hybrid network model based on historical data; is the correction value predicted by the hybrid neural network model; It is the correction factor, which balances the deviation between prediction and reality.

[0113] In order to maximize the network performance, the present invention uses an adaptive genetic algorithm to systematically optimize the structure and hyperparameters of the hybrid neural network. This process includes dynamic adjustment of key hyperparameters such as the number of network layers, the number of neurons, the connection weights, and the learning rate. This optimization strategy aims to find the network configuration that best adapts to the changing needs of the current rolling process.

[0114] When the adaptive genetic algorithm is started, an initial population is first generated, and each individual represents a possible network configuration. The fitness evaluation of each individual is based on its performance in predicting rolling force accuracy and computational efficiency. Individuals with high fitness have a higher chance of being selected for crossover and mutation operations, which mimics the principle of survival of the fittest in natural selection. The crossover operation allows two network configurations to mix their parameters to create a new configuration that may have higher performance. The mutation operation randomly changes certain parameters of the individual, increasing the diversity of the population, helping the algorithm to jump out of the local optimum and explore a wider parameter space.

[0115] In addition, to prevent premature convergence to the local optimal solution, the present invention also introduces a diversity maintenance mechanism, fitness sharing and elite retention strategy. Fitness sharing maintains diversity by promoting less common configurations in the population, and the elite retention strategy ensures that the best individuals can be directly passed from one generation to the next without being affected by crossover or mutation.

[0116] The iterative optimization process of the adaptive genetic algorithm continues until a predetermined termination condition is met, such as reaching the maximum number of iterations or the fitness improvement is below a threshold. In this way, we can scientifically determine the optimal configuration of the hybrid neural network for a specific rolling process, thereby ensuring that the model achieves the best prediction performance in practical applications.

[0117] Through this machine learning technology, step S3 can not only significantly improve the accuracy of rolling force prediction, but also ensure that the model can quickly adapt to various dynamic changes in the production process. This makes the control of the rolling process more precise, thereby optimizing production efficiency and product quality.

[0118] like Figure 3 As shown, a multi-level and multi-stage coordinated optimization model is constructed:

[0119] Step S4: In terms of real-time control of the rolling process, the multi-stage coordinated optimization subsystem uses the output of the prediction model optimized in step S3 to dynamically adjust the rolling parameters. This system includes coordinated optimization control of the roughing, finishing and coiling stages, and each subsystem independently completes the adjustment decision based on its own real-time data and optimization results. The advanced optimization controller integrates the decision outputs of all subsystems to ensure the coordination of the entire rolling process and optimize overall production efficiency and product quality.

[0120] The subsystem objectives for the rough rolling stage are:

[0121]

[0122] Where: for the goal; is the rolling force of the i-th pass; is the average rolling force of all passes; is the number of rolling passes; is the actual measured convexity of the intermediate billet; is the target convexity value; is the total energy consumption in the rough rolling stage; For the The power of the pass.

[0123] The subsystem objectives of the finishing rolling stage are:

[0124]

[0125] Where: for the goal; For the The actual exit thickness of the pass; is the target thickness; is the rolling pass; is the rolling line speed; is the total energy consumed in this phase; The total amount of production.

[0126] The subsystem objectives for the coiling phase are:

[0127]

[0128] Where: for the goal; is the actual measured tension; To set the tension; is the actual measured coil diameter; is the target roll diameter.

[0129] Multi-objective coordinated optimization and decision support.

[0130] This example demonstrates how to use the Pareto frontier method in combination with the entropy weight decision method and TOPSIS technology to handle multi-objective optimization problems and achieve the optimal balance of multiple key performance indicators in the production process.

[0131] Step S5: To further enhance the decision support capability of the system, the Pareto frontier method is introduced in the embodiment of the present invention to deal with the multi-objective optimization trade-off problem. This method achieves an effective balance between minimizing rolling force, reducing energy consumption, improving production efficiency and ensuring product quality by identifying the optimal trade-off point between different production objectives. Pareto frontier optimization analyzes the interaction of each objective function and selects a solution set with balanced performance in all aspects, providing a scientific basis for production scheduling:

[0132]

[0133] Where: and is a decision variable. If there is no other solution , so that all objective functions are The value at is no greater than , and at least one objective function is The value is better than The value of It is considered to be Pareto optimal; is the objective function defined in the decision space.

[0134] In order to assist the decision-making process of the Pareto frontier, the present invention also combines the entropy weight decision-making method and TOPSIS technology to construct a global advanced coordinated optimization controller for further screening:

[0135] Entropy weight method: This method assigns weights by analyzing the entropy value of each decision target, reflecting the importance of each target in the overall decision.

[0136] TOPSIS method: This technique ranks each solution by evaluating its relative closeness to the ideal solution (best) and the negative ideal solution (worst) to determine the final decision solution. The specific method is as follows:

[0137] Data Standardization:

[0138]

[0139] Where: is the decision matrix; is the standardized matrix; and is the maximum and minimum value of the indicator.

[0140] Calculate entropy and weight:

[0141]

[0142] Where: is the probability distribution; ; For the The weight of the item; For the The entropy value of the term; is a constant used to calculate the normalized entropy value; For the The sample in Probability distribution under each indicator; For the The sample in The actual value of each indicator; is the number of samples; Is the number of indicators.

[0143] Weighted normalization matrix:

[0144]

[0145] Where: is the weighted normalization matrix; is the element in the weighted normalization matrix; For the The weight of the item; For the The sample in The actual value of the indicator.

[0146] Determine the positive ideal solution and the negative ideal solution and calculate the distance between each solution and the positive and negative ideal solutions:

[0147]

[0148] Where: For the positive ideal; Negative ideal.

[0149]

[0150] Where: and is the distance between the positive ideal solution and the negative ideal solution.

[0151] Compute relative proximity and sort:

[0152]

[0153] Where: is the relative proximity value.

[0154] The data transmission and decision feedback between the rough rolling load distribution subsystem, the finishing rolling load distribution subsystem and the coiling stage subsystem are optimized and sorted using the entropy weight decision method and TOPSIS similar order preference technology to obtain the final rolling plan. The specific decision method is as follows:

[0155]

[0156] Where: is the comprehensive decision score; is the number of decision criteria; For the The entropy weight of each decision criterion; For the Rating of decision criteria; For the The relative proximity of the standards; and is the distance between the goal and the positive and negative ideals; For the The decision in Ratings on the criteria; Provide a specific score for each decision; For the A specific score for a decision.

[0157] By integrating these advanced decision-making technologies into the optimal control of the rolling process, the present invention can effectively balance and optimize the production process, thereby achieving higher production efficiency and better product quality while reducing energy consumption and material consumption.

[0158] The above embodiments are only descriptions of the preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the protection scope determined by the claims of the present invention.

Claims

1. A multi-level and multi-stage coordinated optimization control method for the entire rolling process, characterized in that: include: Collect target data during rolling production; Inputting the target data into a rolling process parameter calculation model to obtain rolling force; Inputting the target data into a hybrid neural network model to obtain a correction value; The hybrid neural network model includes: a one-dimensional convolutional neural network and a recurrent neural network model; the hybrid neural network model is obtained by training with a training set; Constructing the hybrid neural network model includes: in, Output for the hybrid model; Output of CNN model; Output of the RNN model; and is the fusion coefficient; is the bias term; is the activation function; The hybrid neural network model is obtained by training with a training set, including: Inputting the historical rolling force data in the training set into the one-dimensional convolutional neural network, extracting local features of the sequence data in the historical rolling force data, using the recurrent neural network model to capture the long-term dependencies in the local features, thereby identifying the dynamic change trend that occurs during the continuous rolling process, and using an adaptive genetic algorithm to optimize the hyperparameters of the one-dimensional convolutional neural network and the recurrent neural network model, automatically finding the optimal network parameter configuration, thereby obtaining the hybrid neural network model; According to the rolling force and the correction value, a corrected rolling force is obtained; The method for obtaining the corrected rolling force is: in, It is the rolling force calculated by the traditional physical model; is the corrected rolling force; represents the rolling force predicted by the hybrid network model based on historical data; is the correction value predicted by the hybrid neural network model; is the correction factor; Constructing a multi-level and multi-stage coordinated optimization model, inputting the corrected rolling force into the multi-level and multi-stage coordinated optimization model, dynamically adjusting the rolling parameters, and introducing the Pareto frontier method to weigh multiple conflicting objectives, and further combining the entropy weight decision method and TOPSIS technology to determine the final optimization decision plan; The multi-level and multi-stage coordinated optimization model includes: The rough rolling load distribution coordination optimization sub-model is used to ensure the quality of the strip in the final rolling pass and make load distribution decisions to control the thickness and crown of the strip: in, for the goal; is the rolling force of the i-th pass; is the average rolling force of all passes; is the number of rolling passes; is the actual measured convexity of the intermediate billet; is the target convexity value; is the total energy consumption in the rough rolling stage; For the Power of the pass; The finishing rolling load distribution sub-model is used to receive the data of the rough rolling load distribution coordination optimization sub-model, and perform load balancing and plate shape optimization based on the data and the real-time monitoring data of the finishing rolling process: in, for the goal; For the The actual exit thickness of the pass; is the target thickness; is the rolling pass; is the rolling line speed; is the total energy consumed in this phase; is the total amount of production; The multi-level and multi-stage coordinated optimization model also includes: The coiling stage sub-model is used to adjust the coiling tension according to the optimization results of the finishing load distribution sub-model: in, for the goal; is the actual measured tension; To set the tension; is the actual measured coil diameter; is the target roll diameter; The Pareto frontier method is introduced to balance multiple conflicting objectives: The Pareto frontier method is introduced to balance the minimization of rolling force, reduction of energy consumption, improvement of production efficiency and assurance of product quality: in, and is a decision variable. If there is no other solution , so that all objective functions are The value at is no greater than , and at least one objective function is The value is better than The value of is considered to be Pareto optimal. is the objective function defined in the decision space; The method for determining the final optimization decision plan is: in, is the comprehensive decision score; is the number of decision criteria; For the The entropy weight of each decision criterion; For the Rating of decision criteria; For the The relative proximity of the standards; and is the distance between the goal and the positive and negative ideals; For the The decision in Ratings on the criteria; Provide a specific score for each decision; For the A specific score for a decision.

2. The multi-level and multi-stage coordinated optimization control method for the entire rolling process according to claim 1 is characterized in that: The method for constructing the rolling process parameter calculation physical model is: Where: is the rolling force; is bandwidth; To consider the horizontal projection length of the contact arc between the roller and the workpiece after flattening; is the influence coefficient of considering the stress state in the deformation zone; is the metal deformation resistance, is the influence coefficient of the front and rear tensile stress on the rolling force; Consider the influence coefficient of the stress state in the deformation zone: Where: is the roller radius; , is the inlet and outlet thickness of each rack; is the pressing amount of the rack; Rolling power model: Where: is the rolling torque; is the rolling power; is the roller speed; is the rolling force; is the force arm; is the force arm coefficient.

Citation Information

Patent Citations

  • Hot continuous rolling whole-process load distribution method

    CN109848221A

  • Aero-engine thrust estimation method based on deep CNN-RNN

    CN113255955A

  • Plate rolling thickness control method, device and system and storage medium

    CN113569343A

  • Multi-objective decision optimization method, terminal device and storage medium

    CN114819293A

  • Hot continuous rolling schedule multi-objective optimization method based on improved NSGA-III

    CN115292931A