Continuous casting machine pulling speed stability control method, system, equipment and medium

Abstract

The invention provides a continuous casting machine pulling speed stability control method, system and equipment and a medium, and the method comprises the steps: S1, obtaining real-time state data of continuous casting production of a continuous casting machine, including molten steel temperature, flow, crystallizer vibration parameters, withdrawal and straightening machine speed, casting blank size and quality detection data; s2, a virtual three-dimensional model of the continuous casting machine production process is built, the virtual three-dimensional model is used for reflecting the operation state of all equipment and the change condition of materials in the actual production process in real time, and therefore the element cosmic space of the continuous casting production process is built; s3, on the basis of the obtained real-time state data of continuous casting production of the continuous casting machine, real-time monitoring and correlation analysis are conducted on the continuous casting production process through the built element cosmic space of the continuous casting production process, and therefore abnormal judgment and casting speed change trend prediction are achieved; and S4, optimal pulling speed control is generated on the basis of abnormal judgment and pulling speed change trend prediction, and the pulling speed of the continuous casting machine is controlled and regulated through the generated optimal pulling speed control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of continuous casting machines, in particular to a continuous casting machine pulling speed stability control method, system, device and medium, and more particularly to a continuous casting machine pulling speed stability control method based on meta-universe space information synchronous mapping and disposal. BACKGROUND

[0002] In the continuous casting production process, pulling speed control is a key link to ensure the quality and production efficiency of the cast blank. The traditional continuous casting machine pulling speed control method mainly relies on empirical formula and real-time feedback control, such as adjusting the pulling speed by measuring the liquid level height, molten steel temperature and other parameters in the crystallizer. However, these methods have the following limitations: Insufficient data dimension: mainly relying on a limited number of parameters such as liquid level, temperature, etc., which cannot fully reflect the complex multi-source information in the continuous casting process, resulting in insufficient overall situational awareness of the production process and difficulty in timely discovering problems; Lack of prediction ability: existing technologies are mostly based on real-time feedback for adjustment, lacking the ability to predict future pulling speed trends. When sudden abnormalities occur in the production process, such as fluctuations in molten steel composition, equipment failure, etc., the system cannot respond in advance, resulting in pulling speed fluctuations and affecting the quality of the cast blank; Slow decision response: traditional pulling speed control methods rely on human experience and simple feedback control algorithms, with a slow decision-making process that is difficult to adapt to rapidly changing production conditions. For example, during continuous casting, when the molten steel temperature suddenly changes, the system needs a long time to adjust the pulling speed, resulting in a decrease in the quality of the cast blank; Lack of visualization and simulation functions: existing technologies lack a visual environment for the production process, making it difficult for operators to intuitively understand the production status. At the same time, there is a lack of deep review of historical production data and simulation analysis of future production scenarios, which cannot provide scientific basis for production optimization.

[0003] Patent document CN119204757A (application number: 202411710302.8) discloses a method for constructing an intelligent energy system based on a meta-universe. It includes: in the meta-universe, based on the structural characteristics and energy characteristics of the actual energy system, a virtual energy system is constructed; according to the running conditions corresponding to the actual energy system, the virtual energy system is tested for steady state, and the running parameters of the virtual energy system are obtained; according to the running parameters of the virtual energy system, the expected failure points corresponding to the actual energy system are determined; according to the expected failure points, the actual energy system is adjusted.

[0004] Patent document CN108380838B (Application Number: 201810164343.X) discloses a casting speed control method for continuous casting. The method comprises: determining the current casting speed and the target casting speed during the continuous casting process; comparing the current casting speed and the target casting speed; if the current casting speed is greater than the target casting speed, determining whether the speed variation between the target and current casting speeds is within a preset range; if the speed variation is within the preset range, adjusting the casting speed deceleration to the target deceleration corresponding to the preset range; if the current casting speed is less than the target casting speed, determining whether the current casting speed is within the preset range; if the current casting speed is within the preset range, adjusting the casting speed acceleration to the target acceleration corresponding to the preset range. However, this method lacks information dimensionality and fails to analyze and integrate the complex multi-source information in the continuous casting process, resulting in delayed detection and response to sudden anomalies. Furthermore, this method lacks predictive capabilities and optimization strategies for casting speed variations, and lacks a visualization environment for the production process, which prevents intuitive decision support. Summary of the Invention

[0005] In view of the defects in the prior art, the purpose of the present invention is to provide a method, system, equipment and medium for controlling the casting speed stability of a continuous casting machine.

[0006] According to the present invention, a method for controlling the casting speed stability of a continuous casting machine includes: Step S1: Acquire real-time status data of continuous casting production of the continuous casting machine, including: molten steel temperature, flow rate, crystallizer vibration parameters, straightening machine speed, billet size and quality inspection data; Step S2: constructing a virtual three-dimensional model of the continuous casting machine production process, using the virtual three-dimensional model to reflect in real time the operating status of each device and the changes in materials during the actual production process, thereby constructing a metaverse space of the continuous casting production process; Step S3: Based on the acquired real-time status data of the continuous casting production of the continuous casting machine, the continuous casting production process is monitored and correlated with each other using the constructed metaverse space of the continuous casting production process, thereby realizing abnormality judgment and casting speed change trend prediction; Step S4: generating an optimal casting speed control based on abnormality judgment and casting speed change trend prediction, and using the generated optimal casting speed control to regulate the casting speed of the continuous casting machine.

[0007] Preferably, step S3 includes: Step S3.1: preprocessing the acquired real-time status data to obtain preprocessed real-time status data; Step S3.2: The pre-processed real-time state data is formed into a two-dimensional single-channel feature map; wherein each row of the single-channel feature map represents a type of state data and is segmented according to the length of the same sampling point; Step S3.3: Single-channel feature map is subjected to anomaly identification and casting speed change trend prediction by the anomaly identification model; The anomaly identification model comprises a multi-parameter collaborative anomaly pattern identification sub-model, an anomaly propagation path tracing sub-model, and a hidden risk early warning correlation mining sub-model. The multi-parameter collaborative anomaly pattern identification sub-model comprises identifying single-parameter normal but combined abnormal anomaly based on the single-channel feature map. The anomaly propagation path tracing sub-model comprises locating the abnormal source and diffusion path based on the single-channel feature map through time correlation and spatial correlation analysis. The hidden risk early warning correlation mining sub-model comprises predicting the casting speed change trend based on the single-channel feature map.

[0008] Preferably, the step S4 comprises: Step S4.1: A casting speed control model is constructed according to the predicted casting speed change trend, the current continuous casting state, the abnormal situation, and the process constraint condition, and a reinforcement learning algorithm is used to generate a casting speed adjustment strategy. The action is defined as the casting speed adjustment amount, which needs to meet the process constraint:

[0009] wherein, is the upper and lower limit of single casting speed adjustment; represents single casting speed adjustment, represents the adjusted casting speed at time t+1, and v(t) represents the adjusted casting speed at time t; The reward function formula is as follows:

[0010] wherein, represents the casting speed stability reward; is the target casting speed, and δ is the penalty coefficient;

[0011] wherein, represents the production efficiency reward; is the theoretical optimal casting speed, and η is the efficiency coefficient;

[0012] wherein, represents the abnormal correction reward; is a 0-1 variable, represents that the i-th type of abnormality exists at time t, represents that the i-th type of abnormality has been eliminated or has not occurred at time t, is the weight of the i-th type of abnormality;

[0013] in, represents the constraint violation penalty; if v(t) <v_min,则 ; α, β, γ, and λ represent weight coefficients respectively, α+β+γ+λ=1; Step S4.2: Establish an evaluation model based on the quality indicators and production efficiency indicators of continuous casting production; use the constructed evaluation model to simulate and evaluate the generated casting speed adjustment strategy, update the casting speed control model parameters in real time based on the evaluation results, and obtain the current optimal casting speed adjustment strategy based on the updated casting speed control model;

[0014] Where F is the evaluation result; Q is the quality evaluation value; E is the efficiency evaluation value; θ∈[0,1] is the quality-efficiency balance coefficient; C_viol is the constraint violation cost; ρ is the constraint penalty coefficient; Preferably, the method further comprises: analyzing the cause of the abnormality using a trained traceability model based on the status data; The trained traceability model includes: Construct a traceability model based on an improved convolutional neural network model; Construct a training set and use it to train the traceability model; Wherein, the training set includes: abnormal record data and normal record data; The abnormal record data includes the real-time parameters, equipment status, quality inspection results and final confirmed issues when the abnormality occurs. When constructing the data set, a certain number of records before and after each abnormality event point are used as abnormal samples. The normal record data includes: the records of non-abnormal time are divided into a number of normal samples; Feature extraction based on abnormal samples and normal samples; The traceability model is constructed based on an improved convolutional neural network model; In which, the improved convolutional neural network model uses a multi-scale convolution kernel in the first convolution layer to perform a convolution operation on the original data; the multi-scale convolution kernel includes a large convolution kernel and a small convolution kernel; the large convolution kernel is used to increase the receptive field and extract global features; the small convolution kernel is used to extract local features.

[0015] According to the present invention, a continuous casting machine casting speed stability control system is provided, comprising: Module M1: Acquires real-time status data of continuous casting production of the continuous casting machine, including: molten steel temperature, flow rate, crystallizer vibration parameters, straightening machine speed, billet size and quality inspection data; Module M2: Construct a virtual 3D model of the continuous casting machine production process. This model will reflect the operating status of each device and the changes in materials during the actual production process in real time, thereby constructing a metaverse space for the continuous casting production process. Module M3: Based on the acquired real-time status data of the continuous casting machine, the continuous casting process is monitored and analyzed in real time using the constructed metaverse space of the continuous casting process, thereby realizing abnormality judgment and casting speed change trend prediction; Module M4: Generate the optimal casting speed control based on abnormality judgment and casting speed change trend prediction, and use the generated optimal casting speed control to regulate the casting speed of the continuous casting machine.

[0016] Preferably, the module M3 includes: Module M3.1: pre-processing the acquired real-time status data to obtain pre-processed real-time status data; Module M3.2: composes the preprocessed real-time state data into a two-dimensional single-channel feature map; wherein each row of the single-channel feature map represents a type of state data and is segmented according to the length of the same sampling point; Module M3.3: Single-channel feature map uses anomaly recognition model to identify anomalies and predict casting speed change trends; The anomaly recognition model includes a multi-parameter collaborative anomaly pattern recognition sub-model, an anomaly propagation path tracing sub-model, and an association mining sub-model for implicit risk warning. The multi-parameter collaborative abnormal pattern recognition sub-model includes: identifying abnormal situations where a single parameter is normal but the combination is abnormal based on a single channel feature map; The abnormal propagation path tracing sub-model includes: locating the abnormal source and diffusion path through temporal correlation and spatial correlation analysis based on the single-channel feature graph; The association mining sub-model of the implicit risk warning includes: predicting the pulling speed change trend based on the single-channel characteristic graph.

[0017] Preferably, the module M4 includes: Module M4.1: Build a casting speed control model based on the predicted casting speed trend, current continuous casting status, abnormal conditions, and process constraints, and use reinforcement learning algorithms to generate casting speed adjustment strategies; The action is defined as the pulling speed adjustment, which must meet the process constraints:

[0018] Among them, Δv_min and Δv_max are the upper and lower limits of a single casting speed adjustment; Δv_t represents a single casting speed adjustment, v(t+1) represents the casting speed after adjustment at time t+1, and v(t) represents the casting speed before adjustment at time t; The reward function formula is as follows:

[0019] Among them, R_stab represents the pulling speed stability reward; v_target is the target pulling speed, and δ is the penalty coefficient; R_eff=η·min(v_opt / v(t),1) Among them, R_eff represents the production efficiency reward; v_opt is the theoretical optimal pulling speed, and η is the efficiency coefficient;

[0020] Where R_fault represents the anomaly correction reward; a_{i,t} is a 0-1 variable, a_{i,t}=1 indicates that the i-th anomaly exists at time t, a_{i,t}=0 indicates that the i-th anomaly has been eliminated or has not occurred at time t, and ω_i is the weight of the i-th anomaly; R_viol=Σmax(0,c_viol) Among them, R_viol represents the constraint violation penalty; if v(t) <v_min,则 ; α, β, γ, and λ represent weight coefficients respectively, α+β+γ+λ=1; Module M4.2: Establish an evaluation model based on the quality indicators and production efficiency indicators of continuous casting production; use the constructed evaluation model to simulate and evaluate the generated casting speed adjustment strategy, update the casting speed control model parameters in real time based on the evaluation results, and obtain the current optimal casting speed adjustment strategy based on the updated casting speed control model;

[0021] Where F is the evaluation result; Q is the quality evaluation value; E is the efficiency evaluation value; θ∈[0,1] is the quality-efficiency balance coefficient; C_viol is the constraint violation cost; ρ is the constraint penalty coefficient; Preferably, the system further comprises: analyzing the cause of the abnormality using a trained traceability model based on the status data; The trained traceability model includes: Construct a traceability model based on an improved convolutional neural network model; Construct a training set and use it to train the traceability model; Wherein, the training set includes: abnormal record data and normal record data; The abnormal record data includes the real-time parameters, equipment status, quality inspection results and final confirmed issues when the abnormality occurs. When constructing the data set, a certain number of records before and after each abnormality event point are used as abnormal samples. The normal record data includes: the records of non-abnormal time are divided into a number of normal samples; Feature extraction based on abnormal samples and normal samples; The traceability model is constructed based on an improved convolutional neural network model; In which, the improved convolutional neural network model uses a multi-scale convolution kernel in the first convolution layer to perform a convolution operation on the original data; the multi-scale convolution kernel includes a large convolution kernel and a small convolution kernel; the large convolution kernel is used to increase the receptive field and extract global features; the small convolution kernel is used to extract local features.

[0022] According to an electronic device provided by the present invention, the electronic device includes a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the various steps of the continuous casting machine casting speed stability control method as described above.

[0023] According to a computer-readable storage medium provided by the present invention, instructions are stored on the computer-readable storage medium, and when the instructions are executed by a processor, the various steps of the continuous casting machine casting speed stability control method as described above are implemented.

[0024] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention deploys sensors at key locations on the continuous casting machine to collect multi-source data, including molten steel temperature, flow rate, mold vibration parameters, straightener speed, ingot dimensions, and quality inspection data. This data is then mapped to a virtual 3D model in real time using an information space mapping module. This system can accurately reflect changes in the operating status of each device during the actual production process with millisecond-level accuracy, providing operators and algorithms with comprehensive and intuitive situational awareness. This system can more comprehensively reflect the complexities of the production process, significantly improving monitoring accuracy and response speed. 2. The present invention uses a deep learning model to analyze historical data and real-time operating conditions to predict casting speed trends in the short term, and feeds the predicted results back to the decision-making optimization and control module. This module further uses a reinforcement learning algorithm to comprehensively consider the predicted casting speed trends, the current continuous casting status, abnormal conditions, and process constraints to generate an optimal casting speed adjustment strategy. This technical feature enables the system to not only predict casting speed changes in advance but also formulate precise control strategies based on the predicted results, thereby achieving stable casting speed control. 3. The anomaly sensing module of the present invention can monitor abnormal patterns in the continuous casting process in real time and highlight warnings in the metaverse, buying time for rapid response. At the same time, the decision optimization and control module feeds back the adjusted difference between the expected and actual casting speeds to the anomaly sensing and casting speed prediction module, adjusting anomaly monitoring and correcting the prediction model to achieve closed-loop optimization. This technical feature enables the system to quickly respond to sudden anomalies and continuously improve the accuracy and stability of casting speed control by continuously optimizing the prediction model and control strategy. 4. Relying on the powerful simulation and backtracking functions of the metaverse space, the present invention can, on the one hand, conduct in-depth mining and review of historical production data, reproduce past production scenarios, speed changes, and problems that occurred, and assist in analyzing the root causes of problems; on the other hand, it can carry out forward-looking simulation analysis based on preset scenarios, discover potential problems in advance, and optimize and adjust production plans, process parameters, etc.; this technical feature provides operators with an intuitive production process visualization environment, and also provides a scientific basis for production optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings: Figure 1 Schematic diagram of the continuous casting machine casting speed stability control system.

[0026] Figure 2 A diagram showing the data flow.

[0027] Figure 3 A schematic diagram of the layered architecture.

[0028] Figure 4 This is the closed-loop control logic diagram.

[0029] Figure 5 It is a two-dimensional single-channel feature map. DETAILED DESCRIPTION

[0030] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0031] Example 1 The present invention provides a method and system for controlling the casting speed stability of a continuous casting machine, comprising: constructing a metaverse space of the continuous casting production process, integrating multi-source real-time data through information space mapping, using deep learning and intelligent optimization algorithms to predict and optimize casting speeds, and feeding back to physical equipment to achieve stable control of the casting speed of the continuous casting machine; at the same time, supporting production history review and future simulation to improve the control accuracy, analysis capabilities and production efficiency of continuous casting production.

[0032] The continuous casting machine casting speed stability control system is as follows: Figures 1 to 4 Shown, including: Data Acquisition Module: Sensors deployed at key locations on the continuous casting machine collect data on molten steel temperature, flow rate, mold vibration parameters, straightener speed, ingot dimensions, and quality inspection. This data comprehensively reflects the real-time status of continuous casting production and is transmitted to the Information Space Mapping Module. It also provides basic data support for the anomaly perception and prediction, decision optimization, and control modules.

[0033] Information space mapping module: Utilizes modeling technology to construct a virtual three-dimensional model of the continuous casting machine production process, and maps the production data collected in real time from different sensors into the virtual three-dimensional model, so that the virtual model can reflect the operating status of each device and the changes in materials in the actual production process in real time, thereby constructing a metaverse space of the continuous casting production process, providing operators and algorithms with intuitive and comprehensive situational awareness, while ensuring that the status of the virtual model is synchronized with the actual operating status of the physical continuous casting machine at the millisecond level, realizing the linkage between virtual and real objects, and providing visualization scenarios for abnormal perception and prediction, decision optimization and control, and historical data and simulation analysis modules.

[0034] The continuous casting machine production process is a dynamically changing system, involving multiple data sources such as molten steel temperature, flow rate, mold vibration, and straightener speed. These data sources vary widely in sampling frequency, data format, and communication protocols. For example, a molten steel temperature sensor may collect data at a frequency of seconds, while mold vibration parameters require high-frequency sampling in milliseconds. Data from these different sources is uniformly converted into a standardized data structure in JSON format, using the encoding rule: heat number + billet number + process code + timestamp (ms) to ensure data consistency across time. Normalization is then performed to ensure accurate data synchronization into the Metaverse virtual model.

[0035] The Anomaly Perception and Prediction Module utilizes fused real-time data and, based on a virtual model, conducts real-time monitoring and correlation analysis of the continuous casting process. Upon detecting abnormal patterns or potential risks that deviate from normal ranges, an alert is highlighted in the Metaverse, buying time for a rapid response. Furthermore, based on historical data and real-time operating conditions, a deep learning model is used to predict casting speed trends over the short term. These predictions are fed back to the Decision Optimization and Control Module to inform control strategy formulation. Simultaneously, the prediction model is continuously updated and optimized online based on actual casting speed data. Working together, these two modules not only enhance the system's ability to rapidly respond to emergencies but also lay the foundation for more stable and optimized casting speed control.

[0036] Specifically, firstly, the collected real-time status data of the continuous casting production of the continuous casting machine is preprocessed, including: removing noise and outlier processing to obtain preprocessed real-time status data; The pre-processed state data, such as molten steel temperature, flow rate, crystallizer vibration parameters, straightening machine speed, billet size and quality inspection data, are composed into a two-dimensional single-channel feature map. Each row of the feature map is the data collected by a type of sensor, which is divided according to the length of the same sampling point, such as Figure 5 As shown; The single-channel feature map is used to identify anomalies and predict the trend of casting speed changes through anomaly recognition models; The anomaly recognition model includes a multi-parameter collaborative anomaly pattern recognition sub-model, an anomaly propagation path tracing sub-model, and an association mining sub-model for implicit risk warning. The multi-parameter collaborative abnormal pattern recognition sub-model identifies hidden problems in which a single parameter is normal but a combination of parameters is abnormal based on a single-channel characteristic diagram; for example, the molten steel temperature is within the normal range, but the crystallizer liquid level fluctuates greatly and the drawing speed oscillates synchronously. Combined with the change in the sudden drop in cooling water volume, the chain abnormal pattern of "insufficient cooling intensity → uneven shell shrinkage → liquid level fluctuation → passive adjustment of drawing speed" can be identified, rather than a single parameter failure.

[0037] The anomaly propagation path tracing sub-model uses temporal and spatial correlation analysis based on single-channel feature graphs to locate the anomaly's source and diffusion path. For example, if a continuous decrease in casting speed is detected over a 10-minute period, retrospective data reveals that the molten steel temperature from the ladle to the tundish had already dropped 30 minutes prior, leading to a gradual decrease in tundish superheat, accelerated mold shell growth, increased load on the straightening machine, and ultimately triggering a reduction in casting speed. Temporal correlation can be used to trace the anomaly's source to "upstream molten steel temperature control deviation."

[0038] The implicit risk early warning association mining sub-model identifies potential risks that have not yet caused obvious abnormalities based on single-channel feature maps; for example, when the copper plate wear exceeds 5% of the threshold, and the vibration amplitude deviation is large, even if the current pulling speed is still stable, but the protective slag consumption has increased, it can be warned that "copper plate wear → increased friction of crystallizer → possible pulling speed fluctuation or sticking and breakout risk within 24 hours in the future".

[0039] The decision optimization and control module generates optimal pulling speed control instructions by integrating various types of information, accurately regulates the pulling speed of the continuous casting machine, and ensures the quality of the cast slab and production efficiency.

[0040] Specifically, the decision optimization and control module includes: Sub-module N1: comprehensively considers the predicted pulling speed change trend, current continuous casting state, abnormal situation, and process constraint conditions, and uses a reinforcement learning algorithm to generate an optimal pulling speed adjustment strategy. According to the quality indicators and production efficiency indicators of continuous casting production, an evaluation model is established to simulate and evaluate the generated pulling speed adjustment strategy, and the model parameters are updated in real time to continuously improve the accuracy and adaptability of the pulling speed adjustment strategy.

[0041] Sub-module N2: sends the optimized pulling speed control instructions, visually presents them in the virtual space, and confirms them through human-computer interaction (optional), finally generates accurate control instructions, sends them to physical devices, and records all decision-making and disposal processes for subsequent review.

[0042] The historical data and simulation analysis module: on the one hand, based on the models and historical data constructed in the metaverse space, it reviews the past continuous casting production process, reproduces the production scene, pulling speed change, and problems at the time, and assists in analyzing the root cause of the problem; on the other hand, using prediction models and optimization algorithms, it simulates future continuous casting production, simulates the pulling speed control effect under different working conditions, identifies potential problems in advance, and optimizes production plans, process parameters, etc., to improve production efficiency and benefits.

[0043] Specifically, first, based on a large amount of historical data, a neural network is built, and the traditional convolutional neural network model structure is improved. In the first convolutional layer, a multi-scale convolution kernel is used to convolve the original data to mine the rich and diverse features in the original signal. The large convolution kernel mainly functions to increase the receptive field and extract global features, and the small convolution is mainly used to extract local features, thereby constructing a traceability model. A training set was then constructed, consisting of both abnormal and normal data. The abnormality records included real-time parameters (such as casting speed and temperature), equipment status (such as current and vibration), quality inspection results, and the final confirmed issue. When constructing the training set, a certain number of records before and after each abnormality occurred were used as abnormal samples, while records from non-abnormal times were divided into several normal samples. 70% of these samples were selected as training samples, and 30% as test samples.

[0044] Extract features from the data set, then build a network model, configure model parameters and perform network training. Once the requirements are met, save the network parameters and output the trained model.

[0045] When an anomaly is detected, the current status data is obtained and the root cause of the problem is analyzed using the trained traceability model based on the status data.

[0046] The present invention solves the problems of insufficient information dimension, slow decision response, and lack of in-depth review and foresight of the production process in traditional continuous casting machine speed control by adopting a method based on synchronous mapping and processing of metaverse space information, combined with abnormal perception, speed prediction and decision optimization technology. By building a virtual-real linkage bridge, integrating multi-source data in the production process, and a real-time data-driven intelligent model, high-precision and stable control of the continuous casting machine speed is achieved. In addition, relying on the powerful simulation and backtracking functions of the metaverse space, not only can historical production data be deeply mined and reviewed, but also forward-looking simulation analysis can be carried out based on preset scenarios to provide a scientific basis for decision-making. The present invention effectively improves the quality and efficiency of steel production, reduces production costs, and expands new paths for production process analysis and optimization.

[0047] Those skilled in the art will appreciate that, in addition to implementing the system, device, and various modules provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, and the like by logically programming the method steps. Therefore, the system, device, and various modules provided by the present invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; the modules for implementing various functions can also be considered both software programs for implementing the method and structures within the hardware component.

[0048] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A method for controlling the casting speed stability of a continuous casting machine, characterized in that: include: Step S1: Acquire real-time status data of continuous casting production of the continuous casting machine, including: molten steel temperature, flow rate, crystallizer vibration parameters, straightening machine speed, billet size and quality inspection data; Step S2: constructing a virtual three-dimensional model of the continuous casting machine production process, using the virtual three-dimensional model to reflect in real time the operating status of each device and the changes in materials during the actual production process, thereby constructing a metaverse space of the continuous casting production process; Step S3: Based on the acquired real-time status data of the continuous casting production of the continuous casting machine, the continuous casting production process is monitored and correlated with each other using the constructed metaverse space of the continuous casting production process, thereby realizing abnormality judgment and casting speed change trend prediction; Step S4: generating an optimal casting speed control based on abnormality judgment and casting speed change trend prediction, and using the generated optimal casting speed control to regulate the casting speed of the continuous casting machine.

2. The method for controlling the casting speed stability of a continuous casting machine according to claim 1, wherein: The step S3 comprises: Step S3.1: preprocessing the acquired real-time status data to obtain preprocessed real-time status data; Step S3.2: The pre-processed real-time state data is formed into a two-dimensional single-channel feature map; wherein each row of the single-channel feature map represents a type of state data and is segmented according to the length of the same sampling point; Step S3.3: The single-channel feature map is used to identify anomalies and predict the casting speed change trend through an anomaly recognition model; The anomaly recognition model includes a multi-parameter collaborative anomaly pattern recognition sub-model, an anomaly propagation path tracing sub-model, and an association mining sub-model for implicit risk warning. The multi-parameter collaborative abnormal pattern recognition sub-model includes: identifying abnormal situations where a single parameter is normal but the combination is abnormal based on a single channel feature map; The abnormal propagation path tracing sub-model includes: locating the abnormal source and diffusion path through temporal correlation and spatial correlation analysis based on the single-channel feature graph; The association mining sub-model of the implicit risk warning includes: predicting the pulling speed change trend based on the single-channel characteristic graph.

3. The method for controlling the casting speed stability of a continuous casting machine according to claim 1, wherein: The step S4 comprises: Step S4.1: Build a casting speed control model based on the predicted casting speed change trend, current continuous casting status, abnormal conditions, and process constraints, and generate a casting speed adjustment strategy using a reinforcement learning algorithm; The action is defined as the pulling speed adjustment, which must meet the process constraints: Among them, Δv_min and Δv_max are the upper and lower limits of a single casting speed adjustment; Δv_t represents a single casting speed adjustment, v(t+1) represents the casting speed after adjustment at time t+1, and v(t) represents the casting speed before adjustment at time t; The reward function formula is as follows: Among them, R_stab represents the pulling speed stability reward; v_target is the target pulling speed, and δ is the penalty coefficient; R_eff=η·min(v_opt / v(t),1) Among them, R_eff represents the production efficiency reward; v_opt is the theoretical optimal pulling speed, and η is the efficiency coefficient; Where R_fault represents the anomaly correction reward; a_{i,t} is a 0-1 variable, a_{i,t}=1 indicates that the i-th anomaly exists at time t, a_{i,t}=0 indicates that the i-th anomaly has been eliminated or has not occurred at time t, and ω_i is the weight of the i-th anomaly; R_viol=Σmax(0,c_viol) where \(R_{viol}\) represents the constraint violation penalty; if \(v(t)<v_{min}\), then ; α, β, γ, and λ represent weight coefficients respectively, α+β+γ+λ=1; Step S4.2: Establish an evaluation model based on the quality indicators and production efficiency indicators of continuous casting production; use the constructed evaluation model to simulate and evaluate the generated casting speed adjustment strategy, update the casting speed control model parameters in real time based on the evaluation results, and obtain the current optimal casting speed adjustment strategy based on the updated casting speed control model; Among them, F is the evaluation result; Q is the quality evaluation value; E is the efficiency evaluation value; θ∈[0,1] is the quality-efficiency balance coefficient; C_viol is the constraint violation cost; ρ is the constraint penalty coefficient.

4. The method for controlling the casting speed stability of a continuous casting machine according to claim 1, wherein: The method further includes: analyzing the cause of the anomaly using the trained traceability model based on the status data; The trained traceability model includes: Construct a traceability model based on an improved convolutional neural network model; Construct a training set and use it to train the traceability model; Wherein, the training set includes: abnormal record data and normal record data; The abnormal record data includes the real-time parameters, equipment status, quality inspection results and final confirmed issues when the abnormality occurs. When constructing the data set, a certain number of records before and after each abnormality event point are used as abnormal samples. The normal record data includes: the records of non-abnormal time are divided into a number of normal samples; Feature extraction based on abnormal samples and normal samples; The traceability model is constructed based on an improved convolutional neural network model; In which, the improved convolutional neural network model uses a multi-scale convolution kernel in the first convolution layer to perform a convolution operation on the original data; the multi-scale convolution kernel includes a large convolution kernel and a small convolution kernel; the large convolution kernel is used to increase the receptive field and extract global features; the small convolution kernel is used to extract local features.

5. A continuous casting machine casting speed stability control system, characterized in that: include: Module M1: Acquires real-time status data of continuous casting production of the continuous casting machine, including: molten steel temperature, flow rate, crystallizer vibration parameters, straightening machine speed, billet size and quality inspection data; Module M2: Construct a virtual 3D model of the continuous casting machine production process. This model will reflect the operating status of each device and the changes in materials during the actual production process in real time, thereby constructing a metaverse space for the continuous casting production process. Module M3: Based on the acquired real-time status data of the continuous casting machine, the continuous casting process is monitored and analyzed in real time using the constructed metaverse space of the continuous casting process, thereby realizing abnormality judgment and casting speed change trend prediction; Module M4: Generate the optimal casting speed control based on abnormality judgment and casting speed change trend prediction, and use the generated optimal casting speed control to regulate the casting speed of the continuous casting machine.

6. The continuous casting machine casting speed stability control system according to claim 5, characterized in that: The module M3 includes: Module M3.1: pre-processing the acquired real-time status data to obtain pre-processed real-time status data; Module M3.2: composes the preprocessed real-time state data into a two-dimensional single-channel feature map; wherein each row of the single-channel feature map represents a type of state data and is segmented according to the length of the same sampling point; Module M3.3: Single-channel feature map uses anomaly recognition model to identify anomalies and predict casting speed change trends; The anomaly recognition model includes a multi-parameter collaborative anomaly pattern recognition sub-model, an anomaly propagation path tracing sub-model, and an association mining sub-model for implicit risk warning. The multi-parameter collaborative abnormal pattern recognition sub-model includes: identifying abnormal situations where a single parameter is normal but the combination is abnormal based on a single channel feature map; The abnormal propagation path tracing sub-model includes: locating the abnormal source and diffusion path through temporal correlation and spatial correlation analysis based on the single-channel feature graph; The association mining sub-model of the implicit risk warning includes: predicting the pulling speed change trend based on the single-channel characteristic graph.

7. The continuous casting machine casting speed stability control system according to claim 5, characterized in that: The module M4 includes: Module M4.1: Build a casting speed control model based on the predicted casting speed trend, current continuous casting status, abnormal conditions, and process constraints, and use reinforcement learning algorithms to generate casting speed adjustment strategies; The action is defined as the pulling speed adjustment, which must meet the process constraints: Among them, Δv_min and Δv_max are the upper and lower limits of a single casting speed adjustment; Δv_t represents a single casting speed adjustment, v(t+1) represents the casting speed after adjustment at time t+1, and v(t) represents the casting speed before adjustment at time t; The reward function formula is as follows: Among them, R_stab represents the pulling speed stability reward; v_target is the target pulling speed, and δ is the penalty coefficient; R_eff=η·min(v_opt / v(t),1) Among them, R_eff represents the production efficiency reward; v_opt is the theoretical optimal pulling speed, and η is the efficiency coefficient; Where R_fault represents the anomaly correction reward; a_{i,t} is a 0-1 variable, a_{i,t}=1 indicates that the i-th anomaly exists at time t, a_{i,t}=0 indicates that the i-th anomaly has been eliminated or has not occurred at time t, and ω_i is the weight of the i-th anomaly; R_viol=Σmax(0,c_viol) where \(R_{viol}\) represents the constraint violation penalty; if \(v(t)<v_{min}\), then ; α, β, γ, and λ represent weight coefficients respectively, α+β+γ+λ=1; Module M4.2: Establish an evaluation model based on the quality indicators and production efficiency indicators of continuous casting production; use the constructed evaluation model to simulate and evaluate the generated casting speed adjustment strategy, update the casting speed control model parameters in real time based on the evaluation results, and obtain the current optimal casting speed adjustment strategy based on the updated casting speed control model; Among them, F is the evaluation result; Q is the quality evaluation value; E is the efficiency evaluation value; θ∈[0,1] is the quality-efficiency balance coefficient; C_viol is the constraint violation cost; ρ is the constraint penalty coefficient.

8. The continuous casting machine casting speed stability control system according to claim 5, characterized in that: The system further includes: analyzing the cause of the anomaly using the trained traceability model based on the status data; The trained traceability model includes: Construct a traceability model based on an improved convolutional neural network model; Construct a training set and use it to train the traceability model; Wherein, the training set includes: abnormal record data and normal record data; The abnormal record data includes the real-time parameters, equipment status, quality inspection results and final confirmed issues when the abnormality occurs. When constructing the data set, a certain number of records before and after each abnormality event point are used as abnormal samples. The normal record data includes: the records of non-abnormal time are divided into a number of normal samples; Feature extraction based on abnormal samples and normal samples; The traceability model is constructed based on an improved convolutional neural network model; In which, the improved convolutional neural network model uses a multi-scale convolution kernel in the first convolution layer to perform a convolution operation on the original data; the multi-scale convolution kernel includes a large convolution kernel and a small convolution kernel; the large convolution kernel is used to increase the receptive field and extract global features; the small convolution kernel is used to extract local features.

9. An electronic device comprising a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the electronic device to execute each step of the continuous casting machine casting speed stability control method according to any one of claims 1 to 4.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the method for controlling the casting speed stability of a continuous casting machine as claimed in any one of claims 1 to 4 are implemented.

Images