A hot continuous rolling temperature control method, system and electronic device

Abstract

The application provides a hot continuous rolling temperature control method, system and electronic equipment, relates to the technical field of metallurgical industry process control, and the method comprises the following steps: acquiring multi-modal data of a hot continuous rolling production process at a current time; performing feature extraction on the multi-modal data to obtain heat exchange characteristics related to a strip steel in the hot continuous rolling production process, and constructing a state vector based on the heat exchange characteristics; performing collaborative operation on the state vector through mechanism rules in an equivalent heat exchange coefficient dynamic model and a machine learning correction term to obtain an equivalent heat exchange coefficient at the current time; generating a composite control strategy containing rack control and temperature control at the current time according to the equivalent heat exchange coefficient and a preset temperature target; and controlling an actuator to operate according to the composite control strategy. The application uses the equivalent heat exchange coefficient dynamic model to guarantee the stability and consistency of the strip steel temperature under a dynamic rolling condition and improve the temperature control effect.

Description

Technical Field

[0001] This invention relates to the field of process control technology in the metallurgical industry, and more specifically, to a hot continuous rolling temperature control method, system, and electronic equipment. Background Technology

[0002] In hot strip rolling production, precise temperature control is crucial for ensuring the quality of strip steel products, and the core of temperature control lies in the accurate description of the complex heat exchange process between the strip steel and the external environment. Therefore, equivalent heat transfer coefficient models are typically used to obtain the heat transfer coefficient, thereby achieving temperature control. However, equivalent heat transfer coefficient models are mostly steady-state empirical formulas based on specific operating conditions, making it difficult to adapt to the dynamically changing rolling process.

[0003] In related technologies, to address the poor adaptability of equivalent heat transfer coefficient models to dynamic changes, dynamic heat transfer coefficient models are typically used to adapt to the rolling process. These models include black-box models based on transfer functions and numerical models based on computational fluid dynamics. While black-box models offer high computational efficiency, they lack physical mechanism support, resulting in insufficient extrapolation capabilities. Numerical models offer high accuracy but are time-consuming, making them difficult to apply to real-time temperature control in hot strip rolling production. Therefore, neither black-box models nor numerical models can meet the model requirements of hot strip rolling production. Summary of the Invention

[0004] The problem addressed by this invention is how to improve the temperature control effect of hot continuous rolling based on the heat transfer coefficient.

[0005] To address the above problems, this invention provides a hot rolling temperature control method, system, and electronic device.

[0006] In a first aspect, the present invention provides a hot rolling temperature control method, comprising: Acquire multimodal data of the hot strip rolling production process at the current moment; Feature extraction is performed on the multimodal data to obtain the heat transfer characteristics related to strip steel during the hot continuous rolling process, and a state vector is constructed based on the heat transfer characteristics. Based on the state vector, the equivalent heat transfer coefficient at the current moment is obtained through collaborative calculation using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient. Based on the equivalent heat transfer coefficient and combined with the preset temperature target, a composite control strategy including rack control and temperature control is generated at the current moment. The actuator is controlled to operate according to the composite control strategy.

[0007] Optionally, acquiring the multimodal data of the hot rolling mill production process at the current moment includes: Real-time acquisition of rolling process parameters, strip body parameters, environmental parameters, and measured temperature data during the hot continuous rolling production process; The multimodal data at the current moment are obtained by combining the rolling process parameters, strip body parameters, environmental parameters, and measured temperature data.

[0008] Optionally, the step of extracting features from the multimodal data to obtain heat transfer features related to strip steel during the hot continuous rolling process, and constructing a state vector based on the heat transfer features, includes: The multimodal data is subjected to data standardization processing to obtain the effective data in the multimodal data; Key features directly related to the strip heat exchange process are extracted from the effective data. These key features include process features, physical features, and environmental features that affect the heat transfer effect between the strip and the rolls, cooling medium, and environment. The process features, the ontology features, and the environmental features are dimensionally optimized to obtain the core effective features corresponding to the process features, the ontology features, and the environmental features, respectively. The core effective features are then arranged and combined in a preset order to obtain the state vector.

[0009] Optionally, the step of obtaining the equivalent heat transfer coefficient at the current moment by performing a joint operation based on the state vector using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient includes: The state vector is input into the dynamic model of the equivalent heat transfer coefficient, and a dynamic correction factor that changes with time is determined based on the current moment corresponding to the state vector. Based on the aforementioned mechanism rules, determine the velocity influence factor, water cooling mode influence factor, and water flow density influence factor; Based on the velocity influence factor, the water cooling mode influence factor, and the water flow density influence factor, combined with the dynamic correction factor, the theoretical value of the heat transfer coefficient at the current moment is obtained. The machine learning correction term corresponding to the state vector is generated using the correction function based on the state vector. The theoretical value of the heat transfer coefficient is corrected according to the machine learning correction term to obtain the equivalent heat transfer coefficient at the current moment.

[0010] Optionally, the step of generating a composite control strategy that includes rack control and temperature control at the current moment based on the equivalent heat transfer coefficient and a preset temperature target includes: Based on the equivalent heat transfer coefficient at the current moment, combined with the heat conduction law of strip steel and the heat transfer model of the rolling process, the predicted temperature value of the strip steel at the current moment is determined. The predicted temperature value is compared with the preset temperature target to determine the temperature deviation of the strip at the current moment; The rack-level control quantity and the cooling-level control quantity are simultaneously solved based on the temperature deviation using a composite control algorithm. The frame-level control quantities include mill speed setting values ​​and acceleration compensation values, and the cooling-level control quantities include water volume setting values ​​for each cooling section and valve opening correction values. The rack-level control variables and cooling-level control variables are coupled and optimized according to preset priorities and safety constraints to generate the composite control strategy.

[0011] Optionally, controlling the operation of the actuator according to the composite control strategy includes: The composite control strategy is parsed into low-level executable instructions, which include mill main drive speed instructions, looper tension adjustment instructions, and cooling valve opening instructions. The executable commands are sent in real time through the industrial communication network to the actuators corresponding to the main drive speed command of the rolling mill, the looper tension adjustment command, and the cooling valve opening command, respectively.

[0012] Optionally, it also includes: Acquire the measured temperature of the strip steel and the updated multimodal data after the actuator operates according to the composite control strategy; Based on the updated multimodal data, the process state vector is reconstructed, and combined with the measured temperature of the strip, the actual heat transfer coefficient is determined by temperature back-calculation method. The actual heat transfer coefficient is compared with the equivalent heat transfer coefficient to obtain the coefficient deviation value; If the coefficient deviation value exceeds the preset deviation threshold, then an online learning sample is formed based on the process state vector and the coefficient deviation value. The model parameters and compensation weights of the correction function are dynamically updated based on the online learning samples.

[0013] Optionally, it also includes: constructing a dynamic model of the equivalent heat transfer coefficient; The construction of the dynamic model of the equivalent heat transfer coefficient specifically includes: Acquire historical multimodal data and historical actual heat transfer coefficients during the historical hot continuous rolling production process; Based on the laws of conservation of mass, energy, and momentum, and combined with the heat transfer mechanism of the hot continuous rolling process, a basic mechanism framework covering the heat transfer relationship between strip steel and rolls, cooling medium, and environment is established. Based on the aforementioned basic mechanism framework, and combined with the correction factor, velocity influence factor, water cooling mode influence factor, and water flow density influence factor, a corresponding framework for the mechanism rules is constructed. This corresponding framework is used to determine the theoretical value of the heat transfer coefficient corresponding to the historical multimodal data based on the historical multimodal data. Heat transfer-related core features are extracted from the historical multimodal data to construct a historical state vector. The historical state vector is used as the input of the initial correction function, and the ratio of the historical actual heat transfer coefficient to the theoretical value of the heat transfer coefficient is used as the output of the initial correction function to construct a training sample set. The initial correction function is trained using a gradient boosting tree or deep neural network algorithm based on the training sample set, and the trained initial correction function is used as the correction function.

[0014] In a second aspect, the present invention provides a hot continuous rolling temperature control system, comprising: The data acquisition unit is used to acquire multimodal data of the hot strip rolling production process at the current moment; The feature extraction unit is used to extract features from the multimodal data to obtain the heat transfer features related to the strip steel during the hot continuous rolling process, and to construct a state vector based on the heat transfer features. The computing unit is used to perform collaborative calculations based on the state vector, using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient, to obtain the equivalent heat transfer coefficient at the current moment. The strategy generation unit is used to generate a composite control strategy that includes rack control and temperature control at the current moment, based on the equivalent heat transfer coefficient and a preset temperature target. The control unit is used to control the operation of the actuator according to the composite control strategy.

[0015] Thirdly, an electronic device according to the present invention includes: a processor and a memory, the memory being used to store a computer program; When the computer program is loaded by the processor, it causes the processor to execute the hot rolling temperature control method as described above.

[0016] The hot strip rolling temperature control method, system, and electronic equipment of the present invention first acquire multimodal data at the current moment, comprehensively collect operating condition information strongly related to the heat transfer process, such as rolling process, strip body, environment, and measured temperature, to avoid the inability to fully characterize complex heat transfer conditions; extract heat transfer features from the multimodal data and construct state vectors, wherein, through the integration and feature filtering of multi-source data, complex heat transfer conditions are transformed into quantifiable state vectors. Based on state vectors, a collaborative computation is performed using a dual-drive structure (mechanism rules and machine learning correction terms) of the dynamic model of the equivalent heat transfer coefficient. Mechanism rules ensure the physical rationality of the heat transfer coefficient calculation, avoiding the shortcomings of pure black-box models that lack physical support and have insufficient extrapolation capabilities. Simultaneously, the machine learning correction terms of the dynamic model compensate for the shortcomings of mechanism rules in adapting to complex and variable operating conditions. Compared to traditional steady-state empirical formula models, this approach can more accurately output the equivalent heat transfer coefficient at the current moment, while also avoiding the drawbacks of pure numerical models being time-consuming and unable to meet real-time temperature control requirements. This achieves high-precision and high-timeliness prediction of the equivalent heat transfer coefficient. A composite control strategy is generated based on the equivalent heat transfer coefficient and a preset temperature target, deeply linking the precise equivalent heat transfer coefficient with relevant parameters of rack control and temperature control. This addresses the problem of the disconnect between the heat transfer coefficient and the control strategy in traditional temperature control, ensuring the scientific nature and relevance of the control strategy. In the step of controlling the actuator according to the composite control strategy, by implementing the strategy into specific actions, precise control of the strip temperature can be achieved. This invention utilizes a dynamic model of equivalent heat transfer coefficient to achieve accurate prediction of heat transfer coefficient and precise execution of temperature control strategy, thereby comprehensively improving the temperature control effect and ensuring the stability and consistency of strip temperature under dynamic rolling conditions. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of the hot strip rolling temperature control method according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the construction process of the dynamic model of the equivalent heat transfer coefficient in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hot strip rolling temperature control system according to an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0020] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties. The collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0023] Combination Figure 1 As shown in the figure, an embodiment of the present invention provides a hot strip rolling temperature control method, comprising: Obtain multimodal data of the hot strip rolling production process at the current moment.

[0024] Specifically, current traditional temperature control only collects a limited number of process parameters or single-point temperature data, which cannot fully reflect the complexity of the heat exchange process in strip steel. Therefore, the multimodal data collected in this invention covers rolling process parameters, strip steel parameters, environmental parameters, and measured temperature data, encompassing all the factors related to heat exchange between the strip steel and the rolls, cooling medium, and external environment. From the perspective of the technical characteristics of metallurgical process control, multi-dimensional data can completely characterize the heat exchange scenario under the current rolling conditions. Therefore, multimodal data provides a comprehensive data source that closely matches actual operating conditions, which is a prerequisite for achieving precise temperature control.

[0025] Feature extraction is performed on the multimodal data to obtain the heat transfer characteristics related to strip steel during the hot continuous rolling process, and a state vector is constructed based on the heat transfer characteristics.

[0026] Specifically, by extracting features from multimodal data, key information directly related to the heat transfer process of strip steel can be screened out, and irrelevant data interference can be eliminated. From the perspective of the heat transfer mechanism of hot continuous rolling, the extracted heat transfer features correspond to the influencing factors of core heat transfer processes such as heat transfer between strip steel and rolls, convective heat transfer with cooling medium, and radiative heat transfer with the environment. By constructing the above features into state vectors according to rules, a standardized and calculable representation of complex dynamic heat transfer conditions can be achieved, providing a suitable input carrier for the dynamic model of equivalent heat transfer coefficient.

[0027] Based on the state vector, the equivalent heat transfer coefficient at the current moment is obtained through collaborative calculation using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient.

[0028] Specifically, the dynamic model of the equivalent heat transfer coefficient, as the core of the temperature control method, essentially achieves accurate solution of the equivalent heat transfer coefficient through the dual-drive synergy of mechanistic rules and machine learning correction terms. At the mechanistic rule level, it is constructed based on fundamental laws of heat transfer such as mass conservation and energy conservation, possessing clear physical meaning and ensuring the fundamental rationality of the heat transfer coefficient calculation. This avoids the technical shortcomings of pure black-box models, which lack physical support and have insufficient extrapolation capabilities, and can stably output the theoretical heat transfer coefficient within the known operating conditions. At the machine learning correction term level, it can capture complex nonlinear heat transfer disturbances that mechanistic rules cannot cover, such as changes in contact thermal resistance due to equipment aging and differences in thermal conductivity caused by steel grade switching. This compensates for the technical shortcomings of traditional steady-state empirical formula models, which cannot adapt to varying operating conditions. Furthermore, compared to computational fluid dynamics numerical models, this collaborative operation mode has lower computational complexity, meeting the real-time technical requirements of hot continuous rolling production and achieving a technical balance between high precision and high timeliness.

[0029] Based on the equivalent heat transfer coefficient and combined with the preset temperature target, a composite control strategy including rack control and temperature control is generated for the current moment.

[0030] Specifically, the precise equivalent heat transfer coefficient is converted into executable temperature control parameters. From the perspective of hot strip rolling control technology, the equivalent heat transfer coefficient directly reflects the heat transfer efficiency under the current operating conditions. Combined with the preset temperature target, the direction and magnitude of strip temperature regulation can be clearly defined. From the perspective of control strategy design, the generated composite control strategy integrates two types of parameters: stand control and temperature control. The stand control parameters correspond to factors affecting strip deformation heat and contact time, such as rolling speed, while the temperature control parameters correspond to factors affecting convective heat transfer efficiency, such as cooling water volume. The coupling of these two types of parameters enables multi-dimensional regulation of strip temperature, changing the limitations of traditional temperature control where the heat transfer coefficient and control strategy are disconnected and only cooling parameters are adjusted. This makes the control strategy more closely aligned with actual heat transfer conditions.

[0031] The actuator is controlled to operate according to the composite control strategy.

[0032] Specifically, control through a composite control strategy requires parsing the data into control commands for specific actuators, such as commands recognizable by the main drive of the rolling mill, looper tension, and cooling valves. Finally, real-time command transmission is achieved through an industrial communication network. From the perspective of the process characteristics of hot strip rolling, the actions of the actuators directly affect the rolling and cooling processes. The actions of the stand actuators can adjust the rolling rhythm and deformation heat generation of the strip, while the actions of the cooling actuators can change the cooling intensity of the strip. The coordinated action of these two mechanisms enables precise temperature control of the strip, transforming the technical achievements of the previous steps into actual temperature control effects.

[0033] The hot strip rolling temperature control method of the present invention firstly acquires multimodal data at the current moment, comprehensively collects operating condition information strongly related to the heat transfer process, such as rolling process, strip body, environment and measured temperature, to avoid the inability to fully characterize complex heat transfer conditions; extracts heat transfer features from the multimodal data and constructs a state vector, wherein, through the integration and feature screening of multi-source data, the complex heat transfer conditions are transformed into a quantifiable state vector. Based on state vectors, a collaborative computation is performed using a dual-drive structure (mechanism rules and machine learning correction terms) of the dynamic model of the equivalent heat transfer coefficient. Mechanism rules ensure the physical rationality of the heat transfer coefficient calculation, avoiding the shortcomings of pure black-box models that lack physical support and have insufficient extrapolation capabilities. Simultaneously, the machine learning correction terms of the dynamic model compensate for the shortcomings of mechanism rules in adapting to complex and variable operating conditions. Compared to traditional steady-state empirical formula models, this approach can more accurately output the equivalent heat transfer coefficient at the current moment, while also avoiding the drawbacks of pure numerical models being time-consuming and unable to meet real-time temperature control requirements. This achieves high-precision and high-timeliness prediction of the equivalent heat transfer coefficient. A composite control strategy is generated based on the equivalent heat transfer coefficient and a preset temperature target, deeply linking the precise equivalent heat transfer coefficient with relevant parameters of rack control and temperature control. This addresses the problem of the disconnect between the heat transfer coefficient and the control strategy in traditional temperature control, ensuring the scientific nature and relevance of the control strategy. In the step of controlling the actuator according to the composite control strategy, by implementing the strategy into specific actions, precise control of the strip temperature can be achieved. This invention utilizes a dynamic model of equivalent heat transfer coefficient to achieve accurate prediction of heat transfer coefficient and precise execution of temperature control strategy, thereby comprehensively improving the temperature control effect and ensuring the stability and consistency of strip temperature under dynamic rolling conditions.

[0034] Optionally, acquiring the multimodal data of the hot rolling mill production process at the current moment includes: Real-time acquisition of rolling process parameters, strip body parameters, environmental parameters, and measured temperature data during the hot continuous rolling production process; The multimodal data at the current moment are obtained by combining the rolling process parameters, strip body parameters, environmental parameters, and measured temperature data.

[0035] Specifically, at the data acquisition level, for rolling process parameters, real-time acquisition of parameters such as rolling speed, rolling force, roll gap, cooling manifold opening mode, water flow rate, and water pressure is achieved through an industrial sensor network. Rolling speed is acquired via the mill's main drive encoder; water flow rate and water pressure rely on electromagnetic flowmeters and pressure transmitters in the cooling pipes; and the cooling manifold opening mode is captured via PLC switching signals. For strip body parameters, core data such as steel grade, thickness, width, and inlet temperature are acquired through the production line's steel grade identification system, thickness and width gauges, and inlet temperature measurement devices. Information can be obtained through work order data linkage from the production planning system, while thickness and width are measured precisely and non-contactly using laser thickness and width gauges. For environmental parameters, temperature and humidity sensors are deployed in key areas of the production line to collect real-time ambient temperature and humidity data, eliminating the implicit interference of environmental factors on the heat exchange process. For measured temperature data, infrared thermometers deployed at the inlet, between stands, and outlet of the finishing mill, as well as the coiler inlet, enable multi-point, continuous acquisition of strip surface temperature. Infrared thermometry technology offers advantages such as fast response and non-contact operation, avoiding equipment damage and measurement errors caused by direct contact with high-temperature strip. At the data integration level, a data aggregation platform is built using a Supervisory Control and Data Acquisition (SCADA) system. Parameters from different acquisition terminals and of different types are aligned by timestamp, and the current rolling process, strip body, environment, and measured temperature data are structurally integrated to form a unified format multimodal dataset. This integration process relies on industrial Ethernet for data transmission and uses an edge computing gateway for initial data cleaning and format conversion, ensuring data timeliness and consistency. This provides a standardized and highly available data source for subsequent heat exchange feature extraction and state vector construction.

[0036] In this embodiment of the invention, through multi-dimensional and high-precision data acquisition and standardized integration, a comprehensive characterization of the hot continuous rolling production conditions is achieved. This not only overcomes the limitations of traditional temperature control that only collects single temperature or process data, but also provides complete operating condition data support for the accurate calculation of the equivalent heat transfer coefficient.

[0037] Optionally, the step of extracting features from the multimodal data to obtain heat transfer features related to strip steel during the hot continuous rolling process, and constructing a state vector based on the heat transfer features, includes: The multimodal data is subjected to data standardization processing to obtain the effective data in the multimodal data; Key features directly related to the strip heat exchange process are extracted from the effective data. These key features include process features, physical features, and environmental features that affect the heat transfer effect between the strip and the rolls, cooling medium, and environment. The process features, the ontology features, and the environmental features are dimensionally optimized to obtain the core effective features corresponding to the process features, the ontology features, and the environmental features, respectively. The core effective features are then arranged and combined in a preset order to obtain the state vector.

[0038] Specifically, for data standardization, the Z-score standardization algorithm can be used to normalize multimodal data. For continuous process parameters such as rolling speed and water flow rate, the mean and standard deviation of the data are calculated to eliminate dimensional differences. At the same time, the Laida criterion is used to remove outliers in the measured temperature, ambient temperature and humidity data. In this embodiment, industrial data cleaning tools can also be used to interpolate and complete missing data, ultimately obtaining effective data without redundancy or interference, thus ensuring the weight balance of different types of data in subsequent feature extraction. Then, the Pearson correlation coefficient analysis method is used to calculate the correlation between each effective data and the heat transfer efficiency of the strip. From this, core correlation features such as process features (rolling speed, cooling manifold opening mode and water flow rate), body features (steel grade, strip thickness and inlet temperature), and environmental features (ambient temperature and humidity) are selected. Among them, process features correspond to the core influencing factors of heat transfer between the strip and the roll and convective heat transfer with the cooling medium, body features determine the heat conduction characteristics of the strip itself, and environmental features are related to the radiative heat transfer effect between the strip and the outside world. Subsequently, the core effective features are arranged and combined in a structured manner according to a preset order to form a high-dimensional state vector that can be directly input into the dynamic model of the equivalent heat transfer coefficient, for example, in the order of process features - ontological features - environmental features. Feature purification is achieved by relying on machine learning dimensionality reduction algorithms, while the fixed arrangement order ensures the adaptability of the vector to the operating conditions, thus fully realizing feature extraction and state vector construction.

[0039] In this embodiment of the invention, the availability of multimodal data is improved through standardization, the strong correlation between features and heat transfer process is ensured through feature screening based on mechanism association, and redundancy is eliminated through dimensional optimization. The final constructed state vector can accurately quantify the heat transfer characteristics of the current rolling condition, providing a highly adaptable input carrier for the dynamic model of equivalent heat transfer coefficient, greatly reducing the complexity of model calculation, and improving the accuracy of subsequent coefficient prediction.

[0040] Optionally, the step of obtaining the equivalent heat transfer coefficient at the current moment by performing a joint operation based on the state vector using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient includes: The state vector is input into the dynamic model of the equivalent heat transfer coefficient, and a dynamic correction factor that changes with time is determined based on the current moment corresponding to the state vector. Based on the aforementioned mechanism rules, determine the velocity influence factor, water cooling mode influence factor, and water flow density influence factor; Based on the velocity influence factor, the water cooling mode influence factor, and the water flow density influence factor, combined with the dynamic correction factor, the theoretical value of the heat transfer coefficient at the current moment is obtained. The machine learning correction term corresponding to the state vector is generated using the correction function based on the state vector. The theoretical value of the heat transfer coefficient is corrected according to the machine learning correction term to obtain the equivalent heat transfer coefficient at the current moment.

[0041] Specifically, in the dynamic correction factor determination stage, time series analysis technology is used. Based on a pre-set time-varying law model of equipment aging and operating condition decay, combined with the current production time sequence information (rolling batch and equipment running time) in the state vector, the dynamic correction factor τ, which changes with time, is calculated through an exponential decay function. The dynamic correction factor can quantify the implicit interference of time-varying factors such as equipment wear and pipeline scaling on heat transfer. In the stage of determining the three types of mechanism influencing factors, for the speed influencing factor Kv, a power function fitting technique is used. Based on the correlation data of historical rolling speed and heat transfer efficiency, the calculation model is obtained through least squares regression. K v =a•v b ; Where a is the proportionality coefficient and b is the speed influence index, the mechanism of rolling speed V on heat transfer coefficient is quantified; for the water cooling mode influence factor K m A weighted cumulative algorithm is used to calculate the ratio of the sum of the weights of the open cooling manifolds to the sum of the weights of all open manifolds, based on the basic weights of the number and location of the open cooling manifolds (e.g., 1.0 for the core cooling section and 0.7 for the auxiliary section). This completes the mechanism characterization of the water cooling mode. The influence factor K of the water flow density is also considered. w The piecewise function modeling technique is adopted, and the low / medium / high water flow density range is divided based on the boiling heat transfer theory. The heat transfer correlation function of each range is calibrated by measured data to realize the mechanism mapping of water flow density.

[0042] In the derivation of the theoretical value of the heat transfer coefficient, the dynamic correction factor and three types of mechanism factors are substituted into the preset mechanism formula: ; in, As the reference heat transfer coefficient, The theoretical value of the heat transfer coefficient is obtained through numerical calculation to ensure the physical rationality of the calculation results. In the machine learning correction term generation stage, the gradient boosting tree algorithm is used to construct the correction function. , with state vector Using the measured temperature as input and the deviation between the actual heat transfer coefficient and the theoretical value as the output target, the model parameters are continuously trained through an online sequence learning algorithm. This technique can capture complex nonlinear disturbances not covered by the mechanistic rules. In the final coefficient correction stage, the theoretical value of the heat transfer coefficient is coupled with the machine learning correction term through multiplication. ; in, is the theoretical value of the heat transfer coefficient, and heq is the final output equivalent heat transfer coefficient.

[0043] In this embodiment of the invention, the physical rationality of the equivalent heat transfer coefficient calculation is ensured by the mechanism rules, avoiding the shortcomings of the insufficient extrapolation capability of the pure data model. At the same time, the complex nonlinear interference not covered by the mechanism model is compensated by the machine learning correction term. This not only takes into account the real-time performance of the calculation, but also improves the accuracy of coefficient prediction under varying operating conditions, providing reliable data support for the subsequent generation of temperature control strategies.

[0044] Optionally, the step of generating a composite control strategy that includes rack control and temperature control at the current moment based on the equivalent heat transfer coefficient and a preset temperature target includes: Based on the equivalent heat transfer coefficient at the current moment, combined with the heat conduction law of strip steel and the heat transfer model of the rolling process, the predicted temperature value of the strip steel at the current moment is determined. The predicted temperature value is compared with the preset temperature target to determine the temperature deviation of the strip at the current moment; The rack-level control quantity and the cooling-level control quantity are simultaneously solved based on the temperature deviation using a composite control algorithm. The frame-level control quantities include mill speed setting values ​​and acceleration compensation values, and the cooling-level control quantities include water volume setting values ​​for each cooling section and valve opening correction values. The rack-level control variables and cooling-level control variables are coupled and optimized according to preset priorities and safety constraints to generate the composite control strategy.

[0045] Specifically, in the temperature forecast determination stage, firstly, a coupled model of strip heat conduction and rolling heat transfer is constructed using the finite element analysis method. The equivalent heat transfer coefficient at the current moment is used as the core boundary condition of the model. Combining the strip body parameters (thermal conductivity and specific heat capacity) and rolling process parameters (deformation heat and friction heat generation rate), the temperature field distribution of the strip can be solved through the ANSYS thermal analysis module, and then the temperature forecast values ​​of key points such as the finishing mill exit and coiling entrance are output. For temperature deviation determination, the predicted temperature value is compared point-by-point with the preset temperature target (e.g., 850±5℃ for the finishing mill exit and 680±5℃ for the coiling inlet), calculating the absolute deviation value and deviation trend. Simultaneously, threshold judgment logic distinguishes between normal and abnormal deviations, providing a clear direction for subsequent control variable calculation. In the synchronous solution stage of the two-level control variables, a feedforward-feedback composite control algorithm can be used. Based on the heat transfer characteristic changes predicted by the equivalent heat transfer coefficient, the algorithm outputs the pre-adjustment of the stand speed and the preset value of the cooling water volume. Then, the control variable is dynamically corrected using a PID algorithm, and the setpoints for the cooling water volume and valve opening correction values ​​for each cooling section are determined using a piecewise linear interpolation algorithm. In the control variable coupling optimization stage, a multi-objective optimization algorithm is used, with the optimization objectives being minimizing temperature deviation and balancing equipment operating load. The preset stand control priority is higher than cooling control, and safety constraints such as speed adjustment boundaries and cooling water volume adjustment thresholds are set. A genetic algorithm is used to collaboratively optimize the stand-level and cooling-level control variables, ultimately integrating them into a composite control strategy.

[0046] In this embodiment of the invention, the direction of regulation is determined by accurate temperature forecast, and the rack and cooling control quantities are solved collaboratively by relying on the composite control algorithm. Then, the safety and adaptability of the control strategy are ensured by coupling optimization. This not only changes the limitation of traditional temperature control that only adjusts cooling parameters, but also realizes the linkage between heat transfer coefficient and control strategy, which greatly improves the regulation accuracy and stability of strip temperature under dynamic working conditions.

[0047] Optionally, controlling the operation of the actuator according to the composite control strategy includes: The composite control strategy is parsed into low-level executable instructions, which include mill main drive speed instructions, looper tension adjustment instructions, and cooling valve opening instructions. The executable commands are sent in real time through the industrial communication network to the actuators corresponding to the main drive speed command of the rolling mill, the looper tension adjustment command, and the cooling valve opening command, respectively.

[0048] Specifically, firstly, the generated composite control strategy parameters related to rack control and temperature control are converted into standardized instructions recognizable by the underlying actuators. Specifically, for the mill main drive speed instruction, a PID control parameter conversion algorithm transforms the speed setpoint and acceleration compensation value into a frequency adjustment instruction receivable by the main drive frequency converter. For the looper tension adjustment instruction, the tension control quantity is converted into a torque adjustment instruction for the looper motor. For the cooling valve opening instruction, the cooling section water flow setpoint and valve opening correction value are converted into a stroke control instruction for the pneumatic valve. In the instruction issuance stage, this embodiment uses an industrial Ethernet (Profinet) network to build a real-time communication network and deploys the OPC UA protocol to achieve data interaction between the control layer and the execution layer. Instruction addressing technology is used to accurately match the communication address of the corresponding device to avoid instruction issuance errors. Furthermore, the issued instructions are verified using CRC cyclic redundancy check technology to detect data loss or errors during instruction transmission, ensuring the integrity and accuracy of the instructions.

[0049] In this embodiment of the invention, the upper-layer strategy and the lower-layer execution are connected by parsing the instructions. The high real-time industrial communication network ensures the rapid and error-free issuance of instructions, and completes the terminal landing of the temperature control closed loop. This not only improves the execution efficiency of the control strategy, but also ensures the accuracy of the execution action, and ultimately achieves precise control of the strip temperature.

[0050] Optionally, it also includes: Acquire the measured temperature of the strip steel and the updated multimodal data after the actuator operates according to the composite control strategy; Based on the updated multimodal data, the process state vector is reconstructed, and combined with the measured temperature of the strip, the actual heat transfer coefficient is determined by temperature back-calculation method. The actual heat transfer coefficient is compared with the equivalent heat transfer coefficient to obtain the coefficient deviation value; If the coefficient deviation value exceeds the preset deviation threshold, then an online learning sample is formed based on the process state vector and the coefficient deviation value. The model parameters and compensation weights of the correction function are dynamically updated based on the online learning samples.

[0051] Specifically, in the strip steel measured temperature and updated multimodal data acquisition stage, infrared thermometers deployed at the finishing mill exit and coiling entrance collect the strip steel measured temperature in real time after the control strategy is executed. Simultaneously, the existing sensor network on the production line is used to synchronously collect updated rolling process parameters, strip steel parameters, and environmental parameters. All data is aggregated and uploaded at the edge via a 5G industrial gateway, and timestamp synchronization technology ensures the temporal consistency of the data, providing a real-time data source for subsequent state vector reconstruction. In the actual heat transfer coefficient determination stage, a temperature back-calculation method is used, taking the measured strip steel temperature as a known condition and substituting it into the strip steel heat conduction and rolling heat transfer coupling model. The equivalent heat transfer coefficient in the model is solved in reverse using the Newton-Raphson iteration method. In the coefficient deviation acquisition stage, deviation quantification calculation technology is used to compare the back-calculated actual heat transfer coefficient with the model... The predicted equivalent heat transfer coefficient is interpolated to obtain the absolute deviation value and relative deviation rate. A preset deviation threshold (e.g., a relative deviation rate of 5%) is used as the trigger condition for model updates. When the deviation value exceeds the preset threshold, the reconstructed process state vector is used as the input feature, and the coefficient deviation value is used as the output label. These are encapsulated as online learning samples in the form of input-output key-value pairs and stored in the edge sample database. In the parameter and weight update stage of the correction function, an online sequential extreme learning machine algorithm is used. The newly generated online learning samples are input into the correction function model, and the connection weights and thresholds of the model are dynamically adjusted through incremental learning. A sliding window mechanism is used to retain recently valid samples and remove outdated samples to avoid model overfitting. This algorithm ensures the model's real-time update capability in industrial settings, allowing parameter iteration to be completed without interrupting production.

[0052] In this embodiment of the invention, by constructing an online learning closed loop for the model, dynamic adaptive optimization of the correction function is achieved. This not only makes up for the shortcomings of traditional models with fixed parameters that cannot adapt to long-term operating conditions, but also continuously improves the prediction accuracy of the equivalent heat transfer coefficient by utilizing real-time production data. This ensures the long-term effectiveness and stability of the temperature control strategy and reduces the frequency of manual intervention.

[0053] Optionally, it also includes: constructing a dynamic model of the equivalent heat transfer coefficient; The construction of the dynamic model of the equivalent heat transfer coefficient specifically includes: Acquire historical multimodal data and historical actual heat transfer coefficients during the historical hot continuous rolling production process; Based on the laws of conservation of mass, energy, and momentum, and combined with the heat transfer mechanism of the hot continuous rolling process, a basic mechanism framework covering the heat transfer relationship between strip steel and rolls, cooling medium, and environment is established. Based on the aforementioned basic mechanism framework, and combined with the correction factor, velocity influence factor, water cooling mode influence factor, and water flow density influence factor, a corresponding framework for the mechanism rules is constructed. This corresponding framework is used to determine the theoretical value of the heat transfer coefficient corresponding to the historical multimodal data based on the historical multimodal data. Heat transfer-related core features are extracted from the historical multimodal data to construct a historical state vector. The historical state vector is used as the input of the initial correction function, and the ratio of the historical actual heat transfer coefficient to the theoretical value of the heat transfer coefficient is used as the output of the initial correction function to construct a training sample set. The initial correction function is trained using a gradient boosting tree or deep neural network algorithm based on the training sample set, and the trained initial correction function is used as the correction function.

[0054] Specifically, in the historical data acquisition stage, industrial big data storage and retrieval technology is employed. Through historical databases deployed on the production line (such as the PI database), historical multimodal data from past production batches are acquired in batches. This data covers process, physical, environmental, and temperature data for different steel grades, specifications, rolling speeds, and cooling parameters. Simultaneously, a temperature back-calculation method is used to retrospectively calculate historical temperature data to obtain the actual historical heat transfer coefficients under corresponding operating conditions. Data mining tools are used to filter and clean the data, eliminating invalid and abnormal batch data to ensure the reliability of the data source. In the basic mechanism framework establishment stage, multiphysics coupling modeling technology is employed, based on the three laws of mass conservation, energy conservation, and momentum conservation, combined with the contact heat transfer between the strip and rolls, the convective boiling heat transfer with the cooling medium, and the heat transfer with the environment during hot continuous rolling. The core heat transfer mechanisms, such as radiative heat transfer, are established using the ANSYS multiphysics simulation platform to build a basic heat transfer mechanism framework, clarifying the energy transfer equations and boundary conditions for each heat transfer stage. In the framework construction stage corresponding to the mechanism rules, factor quantification modeling technology is used to embed dynamic correction factors, velocity influence factors, water cooling mode influence factors, and water flow density influence factors into the basic mechanism framework. Among them, the dynamic correction factor establishes a correlation model with production time and equipment status through time series fitting technology; the velocity influence factor constructs a quantitative relationship with rolling speed based on power function fitting (least square regression); the water cooling mode influence factor is determined by weight assignment method combined with manifold functional zoning; and the water flow density influence factor fits the heat transfer efficiency of different flow ranges through piecewise function fitting, ultimately forming a mechanism rule framework with calculable theoretical values ​​of heat transfer coefficients. In the training sample set construction stage, this embodiment employs feature engineering techniques to extract core heat transfer-related features from historical multimodal data. After dimensionality reduction through principal component analysis, a historical state vector is constructed. This vector is then used as input, and the ratio of the historical actual heat transfer coefficient to the theoretical value is used as output. The data is divided into training, validation, and test sets according to a preset ratio to form a standardized training sample set. In the correction function training stage, gradient boosting trees or deep neural network algorithms are used. An initial correction function model is built using Python's sklearn or TensorFlow framework. The training set is used to iteratively train the model parameters, and the validation set is used to adjust the model hyperparameters (such as the learning rate and number of decision trees in the gradient boosting tree, and the number of hidden layers and neurons in the deep neural network). The test set is used to verify the model accuracy. Finally, the model that meets the training criteria (temperature and roller shape prediction errors are less than preset thresholds) is determined as the correction function and integrated with the mechanistic rule framework to form a complete dynamic model of the equivalent heat transfer coefficient.

[0055] In a preferred embodiment of the present invention, combined with Figure 2As shown, firstly, historical process parameters and measured values ​​of temperature, plate shape, and crown are retrieved as multimodal data. The corresponding actual heat transfer coefficient is obtained through temperature back-calculation. Then, a coupled temperature field model is built based on the laws of conservation of mass, energy, and momentum, as well as the heat transfer mechanism. A mechanism rule framework is constructed by combining temperature and roller shape threshold conditions, and the theoretical value of the heat transfer coefficient is calculated. Next, the core heat transfer features are extracted from historical data to construct a historical state vector. This vector is used as input, and the ratio of the actual heat transfer coefficient to the theoretical value is used as output to construct a training sample set. Then, a gradient boosting tree or deep neural network algorithm is used to train the initial correction function. The trained correction function and the mechanism rule framework are integrated into a mechanism data dual-driven model. After that, the current process parameters are input, and the model outputs the predicted value of the coupled temperature field model. The predicted value is compared with the temperature and roller shape threshold conditions. If they are not satisfied, the correction function is retrained using new data until the predicted value meets the threshold requirements. Finally, the model is put into process control.

[0056] In this embodiment of the invention, the physical interpretability of the model is ensured by the mechanistic framework, avoiding the shortcomings of the pure data model in terms of extrapolation capability. At the same time, the machine learning correction function is used to make up for the shortcomings of the mechanistic model in adapting to complex nonlinear conditions. The constructed equivalent heat transfer coefficient dynamic model has both high accuracy and strong adaptability.

[0057] Combination Figure 3 As shown, a hot strip rolling temperature control system of the present invention includes: The data acquisition unit is used to acquire multimodal data of the hot strip rolling production process at the current moment; The feature extraction unit is used to extract features from the multimodal data to obtain the heat transfer features related to the strip steel during the hot continuous rolling process, and to construct a state vector based on the heat transfer features. The computing unit is used to perform collaborative calculations based on the state vector, using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient, to obtain the equivalent heat transfer coefficient at the current moment. The strategy generation unit is used to generate a composite control strategy that includes rack control and temperature control at the current moment, based on the equivalent heat transfer coefficient and a preset temperature target. The control unit is used to control the operation of the actuator according to the composite control strategy.

[0058] The advantages of the hot rolling temperature control system of the present invention compared with the prior art are the same as those of the above-mentioned hot rolling temperature control method compared with the prior art, and will not be repeated here.

[0059] An electronic device according to the present invention includes: a processor and a memory, wherein the memory is used to store a computer program; When the computer program is loaded by the processor, it causes the processor to execute the hot rolling temperature control method as described above.

[0060] The electronic device of the present invention has the same advantages over the prior art as the hot rolling temperature control method described above, and will not be repeated here.

[0061] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for temperature control in hot continuous rolling, characterized in that, include: Acquiring multimodal data of the hot strip rolling production process at the current moment specifically includes: real-time acquisition of rolling process parameters, strip body parameters, environmental parameters, and measured temperature data during the hot strip rolling production process; integrating the rolling process parameters, strip body parameters, environmental parameters, and measured temperature data at the current moment to obtain the multimodal data at the current moment; wherein, the rolling process parameters include rolling speed, rolling force, roll gap, cooling manifold opening mode, water flow rate, and water pressure; the strip body parameters include steel grade, thickness, width, and inlet temperature; the environmental parameters include ambient temperature and humidity data; and the measured temperature data includes strip surface temperature. Feature extraction is performed on the multimodal data to obtain the heat transfer characteristics related to strip steel during the hot continuous rolling process, and a state vector is constructed based on the heat transfer characteristics. Based on the state vector, the equivalent heat transfer coefficient at the current moment is obtained through collaborative computation using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient. Specifically, this includes: inputting the state vector into the dynamic model of the equivalent heat transfer coefficient; determining a dynamic correction factor that changes over time based on the current moment corresponding to the state vector; determining a velocity influence factor, a water cooling mode influence factor, and a water flow density influence factor based on the mechanism rules; obtaining the theoretical value of the heat transfer coefficient at the current moment based on the velocity influence factor, the water cooling mode influence factor, and the water flow density influence factor, combined with the dynamic correction factors; generating a machine learning correction term corresponding to the state vector using a correction function; and correcting the theoretical value of the heat transfer coefficient based on the machine learning correction term to obtain the equivalent heat transfer coefficient at the current moment. The construction of the equivalent heat transfer coefficient dynamic model specifically includes: acquiring historical multimodal data and historical actual heat transfer coefficients during the hot continuous rolling production process; establishing a basic mechanism framework covering the heat transfer relationship between strip steel and rolls, cooling medium, and environment, based on the laws of conservation of mass, energy, and momentum, and combined with the heat transfer mechanism of the hot continuous rolling production process; constructing a corresponding framework for the mechanism rules based on the basic mechanism framework, combined with correction factors, velocity influence factors, water cooling mode influence factors, and water flow density influence factors, which is used to determine the theoretical value of the heat transfer coefficient corresponding to the historical multimodal data; extracting heat transfer-related core features from the historical multimodal data to construct a historical state vector, using the historical state vector as the input of the initial correction function and the ratio of the historical actual heat transfer coefficient to the theoretical value of the heat transfer coefficient as the output of the initial correction function, and constructing a training sample set; training the initial correction function based on the training sample set using a gradient boosting tree or deep neural network algorithm, and using the trained initial correction function as the correction function. The process involves acquiring the measured temperature of the strip steel and updated multimodal data after the actuator operates according to the composite control strategy; reconstructing the process state vector based on the updated multimodal data; determining the actual heat transfer coefficient using a temperature back-calculation method in conjunction with the measured temperature of the strip steel; comparing the actual heat transfer coefficient with the equivalent heat transfer coefficient to obtain a coefficient deviation value; if the coefficient deviation value exceeds a preset deviation threshold, forming an online learning sample based on the process state vector and the coefficient deviation value; and dynamically updating the model parameters and compensation weights of the correction function based on the online learning sample. Based on the equivalent heat transfer coefficient and combined with the preset temperature target, a composite control strategy including rack control and temperature control is generated at the current moment. The actuator is controlled to operate according to the composite control strategy.

2. The hot continuous rolling temperature control method according to claim 1, characterized in that, The step of extracting features from the multimodal data to obtain heat transfer features related to strip steel during the hot continuous rolling process, and constructing a state vector based on the heat transfer features, includes: The multimodal data is subjected to data standardization processing to obtain the effective data in the multimodal data; Key features directly related to the strip heat exchange process are extracted from the effective data. These key features include process features, physical features, and environmental features that affect the heat transfer effect between the strip and the rolls, cooling medium, and environment. The process features, the ontology features, and the environmental features are dimensionally optimized to obtain the core effective features corresponding to the process features, the ontology features, and the environmental features, respectively. The core effective features are then arranged and combined in a preset order to obtain the state vector.

3. The hot continuous rolling temperature control method according to claim 1, characterized in that, The process of generating a composite control strategy that includes rack control and temperature control at the current moment, based on the equivalent heat transfer coefficient and a preset temperature target, includes: Based on the equivalent heat transfer coefficient at the current moment, combined with the heat conduction law of strip steel and the heat transfer model of the rolling process, the predicted temperature value of the strip steel at the current moment is determined. The predicted temperature value is compared with the preset temperature target to determine the temperature deviation of the strip at the current moment; The rack-level control quantity and the cooling-level control quantity are simultaneously solved based on the temperature deviation using a composite control algorithm. The frame-level control quantities include mill speed setting values ​​and acceleration compensation values, and the cooling-level control quantities include water volume setting values ​​for each cooling section and valve opening correction values. The rack-level control variables and cooling-level control variables are coupled and optimized according to preset priorities and safety constraints to generate the composite control strategy.

4. The hot continuous rolling temperature control method according to claim 1, characterized in that, The step of controlling the operation of the actuator according to the composite control strategy includes: The composite control strategy is parsed into low-level executable instructions, which include mill main drive speed instructions, looper tension adjustment instructions, and cooling valve opening instructions. The executable commands are sent in real time through the industrial communication network to the actuators corresponding to the main drive speed command of the rolling mill, the looper tension adjustment command, and the cooling valve opening command, respectively.

5. A hot continuous rolling temperature control system applied to the hot continuous rolling temperature control method of claim 1, characterized in that, include: The data acquisition unit is used to acquire multimodal data of the hot strip rolling production process at the current moment; The feature extraction unit is used to extract features from the multimodal data to obtain the heat transfer features related to the strip steel during the hot continuous rolling process, and to construct a state vector based on the heat transfer features. The computing unit is used to perform collaborative calculations based on the state vector, using the mechanism rules and machine learning correction terms in the dynamic model of the equivalent heat transfer coefficient, to obtain the equivalent heat transfer coefficient at the current moment. The strategy generation unit is used to generate a composite control strategy that includes rack control and temperature control at the current moment, based on the equivalent heat transfer coefficient and a preset temperature target. The control unit is used to control the operation of the actuator according to the composite control strategy.

6. An electronic device, characterized in that, include: Processor and memory, the memory being used to store computer programs; When the computer program is loaded by the processor, it causes the processor to execute the hot rolling temperature control method as described in any one of claims 1-4.

Images