Metallurgy whole-process dynamic collaborative optimization method and system based on digital twinning

Abstract

The invention provides a metallurgy full-process dynamic collaborative optimization method and system based on digital twinning, which can realize metallurgy full-process, multi-dimensional and dynamic collaborative closed-loop optimization. The method specifically aims to construct a digital twinborn base capable of reflecting the complex state of the whole metallurgical process in real time, accurate prediction and diagnosis of key indexes and potential problems in the metallurgical process are achieved based on digital twinborn and artificial intelligence technologies, and collaborative optimization decisions of time dimensions (scheduling) and space dimensions (process parameters) based on the state of the whole process are achieved. The feasibility and effect of an optimization scheme are verified through virtual simulation, the trial and error cost is reduced, a closed-loop control mechanism from the virtual world to the physical world is established, online self-learning and self-adaption of the model are achieved, and finally the metallurgical production efficiency is improved, the energy consumption is optimized, the product quality is guaranteed and improved, the operation and maintenance cost is reduced, and the flexible production capacity is enhanced.

Description

Technical Field

[0001] This invention relates to the fields of digital twin and artificial intelligence technologies, specifically to a dynamic collaborative optimization method and system for the entire metallurgical process based on digital twins. Background Technology

[0002] As a pillar industry of the national economy, the metallurgical industry's production efficiency, energy consumption, and product quality directly impact national economic development and environmental protection. Metallurgical production processes are extremely complex, highly continuous, and tightly coupled between processes (for example, ironmaking-steelmaking-continuous casting-rolling is a closely linked process chain). Simultaneously, they face harsh operating conditions such as high temperature, high pressure, and heavy loads, as well as the demand for customized production of multiple varieties in small batches. With the application of information technology and automation technology, enterprises have accumulated a large amount of production data. However, this data is scattered across different systems (ERP, MES, L1 / L2, etc.), forming data silos that are difficult to integrate and analyze systematically.

[0003] With the continuous advancement of intelligent manufacturing and Industry 4.0, the metallurgical industry, as a typical representative of high energy consumption and highly complex processes, is facing enormous challenges in synergistically improving production efficiency, energy utilization, equipment reliability, and product quality. Traditional metallurgical production processes typically consist of multiple physically distributed process segments (such as blast furnaces, converters, continuous casting, and rolling) connected in series, covering raw material pretreatment, smelting, refining, forming, and post-processing. The entire process involves hundreds of key pieces of equipment and tens of thousands of sensor nodes, exhibiting typical characteristics such as strong temporal dynamics, complex spatial coupling, significant system nonlinearity, and difficulty in coordinating multiple objectives and constraints. Although enterprises have gradually introduced information systems such as MES (Manufacturing Execution System), ERP (Enterprise Resource Planning), and SCADA (Supervisory Control and Data Acquisition) in recent years, and deployed preliminary automatic control and scheduling optimization modules, most current methods still suffer from gaps in real-time status perception and prediction due to the failure to construct high-fidelity digital twins. They often rely solely on offline models or operational experience, lacking the ability to predict sudden anomalies in advance.

[0004] Based on this, we will continue to improve the existing dynamic collaborative optimization methods and systems for the entire metallurgical process to address the technical deficiencies of the existing solutions. Summary of the Invention

[0005] The purpose of this invention is to address the core problems in the entire metallurgical production process, such as fragmented multi-source data, insufficient state awareness, delayed anomaly response, and lack of optimization decision-making. This invention provides a dynamic collaborative optimization method for the entire metallurgical process based on digital twins.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A dynamic collaborative optimization method for the entire metallurgical process based on digital twins includes the following steps: S1. By collecting metallurgical process data, automatic control data of the control system, and enterprise management system data from the upper management system of the metallurgical equipment through IoT sensors installed in the metallurgical equipment, multi-source heterogeneous data is formed. The multi-source heterogeneous data is preprocessed to form a full-process state vector. S2. Transmit the entire process state vector to the digital twin platform. The digital twin platform constructs a digital twin model that includes a geometric model, a mechanism model, a behavior model, and a data model. The digital twin platform calculates real-time state information based on the data model and imports historical data information. The artificial intelligence model located in the digital twin platform predicts and diagnoses anomalies in the production indicators of metallurgical equipment based on the real-time state information and historical data. S3. The digital twin platform generates multi-dimensional optimization strategies based on anomaly diagnosis results, production indicator prediction information, and constraints. The optimization strategies include time-dimensional strategies and spatial-dimensional strategies. Virtual simulation is performed through the optimization strategies, and the optimization strategy with the best simulation results is selected as the final optimization strategy. S4. Select the optimization strategy and output it to the system execution unit of the metallurgical equipment. Collect the execution process information of the metallurgical equipment through IoT sensors, compare the execution information with the simulation information of the final optimization strategy, and learn and iteratively correct based on the deviation driving mechanism model and artificial intelligence model.

[0007] The above technical solution produces the following technical effects: Compared with existing technologies, this invention has significant advantages. Existing technologies, such as single-process optimization and static control, lack a global perspective and struggle to address the challenges posed by the complex coupling of metallurgical processes. This invention, by constructing a full-process digital twin, breaks down data silos and achieves precise mapping of physical entities; based on artificial intelligence algorithms, it enables predictive diagnosis and proactive response to dynamic changes. More importantly, this invention innovatively coordinates optimization in the time dimension (scheduling) and the spatial dimension (process parameters), considering the coupling characteristics of the entire process, thereby generating a truly globally optimal control strategy. Virtual simulation verification reduces the trial-and-error risks in actual production. Through closed-loop control and model adaptation, the system can continuously learn and improve, constantly enhancing optimization effects and ensuring optimal operating conditions under complex and ever-changing circumstances. These are achievements difficult to attain with traditional automation systems and local optimization methods.

[0008] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, in step S1, the metallurgical process data collected by the IoT sensors includes data from the blast furnace, converter, continuous casting machine and rolling mill; the automatic control data is data from the L1 and / or L2 process control systems, including set values, control commands, alarm information and historical trend information; the enterprise management system data includes data from at least one upper-level management system from the EPR system, MES system, PLM system and EMS system.

[0009] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, the process of preprocessing the collected multi-source heterogeneous data to form the state vector of the entire process in step S1 includes the following steps: S11. Perform data cleaning on multi-source heterogeneous data collected from IoT sensors, statistically analyze abnormal monitoring data, and process abnormal data in any way, such as removing, truncating, or replacing with predicted values. S12. Complete the missing or abnormal data, where the completed data is the predicted value; S13. The data processed in step S12 is standardized and uniformly encoded. S14. Resample the data processed in step S13 to a uniform time step according to different sampling frequencies; S15. The data processed in step S14, together with other heterogeneous data from multiple sources, constitute the full-process state vector.

[0010] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, the geometric model constructed by the digital twin platform in step S2 is a three-dimensional model of the plant where the metallurgical equipment is located and a structural model of the metallurgical equipment, which is used to achieve visualization. The mechanistic models include blast furnace reaction model, converter blowing model, continuous casting solidification and heat transfer model, rolling mechanics model, and heat treatment phase transformation model; Behavioral models describe the operating logic, process flow, and production cycle of metallurgical equipment through discrete event simulation or state machines; The data model is based on real-time mapping of the collected data to the status.

[0011] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, after the state vector of the entire process is input into the digital twin platform, the digital twin platform corrects the output of the mechanism model by using Kalman filtering or Bayesian update methods.

[0012] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, in step S2, the artificial intelligence model predicts the production indicators of metallurgical equipment based on long short-term memory neural networks or Transformer models to obtain the steel composition, steel temperature and power consumption information of metallurgical equipment in the future state. The artificial intelligence model realizes anomaly diagnosis through decision trees, SVM or convolutional neural networks, and identifies equipment failures of metallurgical equipment, abnormal quality of produced products and abnormal state information of production conditions.

[0013] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, the digital twin platform generates multi-dimensional optimization strategies through a collaborative optimization engine based on anomaly diagnosis results, production indicator prediction information, and constraints. The time-dimensional strategy is the scheduling scheme, and the spatial-dimensional strategy is the adjustment of process parameters.

[0014] As a further improvement to the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in this application, the final optimization strategy is sent to the field L1 control system and / or L2 control system in the system execution unit for execution through the closed-loop control interface module.

[0015] A dynamic collaborative optimization system for the entire metallurgical process based on digital twins is provided. The system has instructions that can be executed by at least one processor, which enables the at least one processor to execute any of the aforementioned dynamic collaborative optimization methods for the entire metallurgical process based on digital twins.

[0016] As a further improvement to the dynamic collaborative optimization system for the entire metallurgical process based on digital twins proposed in this application, it includes: The data acquisition and fusion module is used to collect metallurgical process data, automatic control data from the control system, and enterprise management system data from the upper-level management system of the metallurgical equipment from the IoT sensors in the metallurgical equipment. The metallurgical process data, automatic control data, and enterprise management system data are fused to form multi-source heterogeneous data. Digital twin platforms are used to build digital twin models that include geometric models, mechanistic models, behavioral models, and data models, including artificial intelligence models for predicting and diagnosing anomalies in metallurgical equipment production indicators. The collaborative optimization engine receives real-time status information calculated by the digital twin platform and anomaly diagnosis results output by the artificial intelligence model, and generates multi-dimensional collaborative optimization strategies. The virtual simulation verification module receives the collaborative optimization strategy and performs virtual simulation on the digital twin platform to simulate the execution of the collaborative optimization strategy. The closed-loop control interface module transforms the optimal collaborative optimization strategy verified by virtual simulation into the execution of the field L1 control system and / or L2 control system in the system execution unit. The human-computer interaction interface module is used to visually display the real-time status, key parameters, material flow, and energy flow of the digital twin model; display the information processed in the digital twin platform; support operators to confirm, modify, or manually intervene in the digital twin platform; and provide functions for historical data query, report generation, and system configuration. Attached Figure Description

[0017] Figure 1 This is a flowchart of the process of the present invention; Figure 2 This is a structural diagram of the electronic device that constitutes the system of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To facilitate a proper understanding of the solutions provided in the following embodiments of the present invention, the terms involved in the present invention will be explained as follows before describing the technical solutions provided by the present invention: Digital twin: refers to a virtual model of a physical entity or process that can reflect its state in real time and be simulated, predicted, and optimized.

[0020] IoT (Internet of Things): refers to a huge network formed by combining various information sensors, radio frequency identification technology, global positioning system, infrared sensors, laser scanners and other devices and technologies to collect information in real time about any object or process that needs to be monitored, connected and interacted with, with the Internet.

[0021] ERP (Enterprise Resource Planning): An enterprise resource planning system used to manage various internal business processes of an enterprise, such as finance, procurement, inventory, and production planning.

[0022] MES (Manufacturing Execution System): A manufacturing execution system used to monitor and manage the production process, including production scheduling, process management, and quality tracking.

[0023] PLM (Product Lifecycle Management): A product lifecycle management system used to manage information throughout the entire process of a product, from conception, design, manufacturing, service to disposal.

[0024] EMS (Energy Management System): An energy management system used to monitor and manage an enterprise's energy consumption and optimize energy scheduling.

[0025] L1 / L2 process control system: The automation control system hierarchy in the metallurgical industry. L1 usually refers to basic automation (such as PLC), and L2 usually refers to process control and monitoring (such as DCS).

[0026] AI (Artificial Intelligence): refers to the use of computer programs to simulate, extend, and expand human intelligence to achieve abilities such as perception, cognition, and decision-making.

[0027] Metallurgical process: refers to the complete production process from raw material preparation (such as sintering), ironmaking, steelmaking, continuous casting, rolling, heat treatment to finished product delivery.

[0028] Dynamic collaborative optimization: refers to adjusting production plans and process parameters in real time to achieve overall optimization across processes, equipment, and objectives in response to the constantly changing state during continuous production.

[0029] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0030] Example 1 like Figure 1 As shown, this application addresses the technical deficiencies in existing solutions by implementing a dynamic collaborative optimization method for the entire metallurgical process based on digital twins. Specifically, this application recognizes that in recent years, enterprises have gradually introduced information systems such as MES (Manufacturing Execution System), ERP (Enterprise Resource Planning), and SCADA (Supervisory Control and Data Acquisition), and deployed preliminary automatic control and scheduling optimization modules. However, most current methods still suffer from the following prominent problems: 1) Data fragmentation and modeling lag: Traditional systems generally suffer from problems such as separation of equipment data and management data and lag in model updates, making it impossible to achieve full-process modeling and real-time response for complex processes.

[0031] 2) Lack of real-time state mapping mechanism: Due to the failure to build a high-fidelity digital twin, the existing system has gaps in real-time state perception and prediction, often relying only on offline models or operational experience, and lacks the ability to predict sudden anomalies in advance.

[0032] 3) Localized and static optimization strategies: Most optimization models are only for a single device or a local process, lacking a joint optimization framework that covers the entire process, space, and time dimensions. This results in the system operation still relying on the experience of the scheduler for adjustment, leading to slow response and poor stability.

[0033] 4) The separation between artificial intelligence and physical mechanism models: Existing artificial intelligence technologies have begun to be explored in some metallurgical scenarios, but they often exist as independent modules and lack deep integration with physical mechanism models, resulting in insufficient interpretability and poor adaptability.

[0034] In recent years, the rapid development of digital twin technology has brought new breakthroughs to the metallurgical industry. By constructing a unified metallurgical virtual entity that includes geometric models, physical mechanism models, behavioral rule models, and data models, digital twins achieve accurate mapping and prediction of real equipment and process conditions, and possess powerful real-time perception, predictive analysis, and collaborative control capabilities.

[0035] At the same time, artificial intelligence (AI) technologies, especially deep learning, reinforcement learning, and evolutionary algorithms, are showing increasingly prominent performance in industrial forecasting, fault diagnosis, and process optimization, providing technical support for the intelligent control of metallurgical processes.

[0036] However, there is currently a lack of a complete system that can deeply integrate a digital twin platform with strong physical modeling capabilities with an artificial intelligence model with strong self-learning capabilities, and can dynamically link multi-source heterogeneous data, real-time control systems, and management systems.

[0037] Specifically, this invention aims to solve core problems in the entire metallurgical production process, such as fragmented multi-source data, insufficient state awareness, delayed anomaly response, and lack of optimization decision-making. It provides a dynamic collaborative optimization method for the entire metallurgical process based on digital twins, specifically including the following steps: A dynamic collaborative optimization method for the entire metallurgical process based on digital twins includes the following steps: S1. By collecting metallurgical process data, automatic control data of the control system, and enterprise management system data from the upper management system of the metallurgical equipment through IoT sensors installed in the metallurgical equipment, multi-source heterogeneous data is formed. The multi-source heterogeneous data is preprocessed to form a full-process state vector. S2. Transmit the entire process state vector to the digital twin platform. The digital twin platform constructs a digital twin model that includes a geometric model, a mechanism model, a behavior model, and a data model. The digital twin platform calculates real-time state information based on the data model and imports historical data information. The artificial intelligence model located in the digital twin platform predicts and diagnoses anomalies in the production indicators of metallurgical equipment based on the real-time state information and historical data. S3. The digital twin platform generates multi-dimensional optimization strategies based on anomaly diagnosis results, production indicator prediction information, and constraints. The optimization strategies include time-dimensional strategies and spatial-dimensional strategies. Virtual simulation is performed through the optimization strategies, and the optimization strategy with the best simulation results is selected as the final optimization strategy. S4. Select the optimization strategy and output it to the system execution unit of the metallurgical equipment. Collect the execution process information of the metallurgical equipment through IoT sensors, compare the execution information with the simulation information of the final optimization strategy, and learn and iteratively correct based on the deviation driving mechanism model and artificial intelligence model.

[0038] Furthermore, this application further proposes that in step S1, the metallurgical process data includes data from the blast furnace, converter, continuous casting machine and rolling mill; the automatic control data is data from the L1 and / or L2 process control systems, including setpoints, control commands, alarm information and historical trend information; and the enterprise management system data includes data from at least one upper-level management system from the EPR system, MES system, PLM system and EMS system.

[0039] Furthermore, in step S1, the process of preprocessing the collected multi-source heterogeneous data to form the full-process state vector includes the following steps: S11. Perform data cleaning on multi-source heterogeneous data collected from IoT sensors, statistically analyze anomaly detection data, and process the abnormal data in any way, such as removing, truncating, or replacing with predicted values. The raw value collected by the i-th sensor of the metallurgical equipment at time point t is: ; Furthermore, outliers are defined using the sliding window mean and standard deviation, with the following calculation formula: ; in, The sliding window length represents the number of time points selected when calculating the mean and standard deviation. For example, =10 indicates that data from the past 10 sampling points is used for calculation. The moving average represents the average level of the data within the window, characterizing the baseline of the sensor's normal state during the current time period. It is the sliding standard deviation, which represents the fluctuation amplitude of the data within the window and depicts the normal fluctuation range of the sensor in the current period.

[0040] When the following conditions are met: ; where The threshold coefficient, usually taking values from 2 to 3. It defines the allowable fluctuation range. When the deviation exceeds the data point is determined to be abnormal.

[0041] Assuming that the normal data follows an approximate Gaussian distribution, most of the data should fall within μ±2σ, and the vast majority of the data should fall within μ±3σ. Data exceeding this interval is likely to be abnormal (such as sensor jitter, interference, equipment shock). For abnormal data, this application directly deletes the data point at that moment and uses step S12 for completion.

[0042] S12. Complete the missing data or abnormal data, where the completed data is the predicted value; Among them, the method for completion in this application includes: 1) Using linear interpolation, which is applicable to slowly changing physical quantities, such as temperature.

[0043] ; where t1 < t < t2, and the two ends are the nearest valid values.

[0044] 2) Sliding mean substitution: applicable to parameters with little fluctuation but requiring stable input.

[0045] <00.....​​​​​​​​​​​​​​​​​​​​​​​​

[0048] Each data entry is bound to a primary key: ; S14. Resample the data processed in step S13 to a uniform time step (e.g., 1s or 1min) according to different sampling frequencies.

[0049] S15. The data processed in step S14, along with other heterogeneous data from multiple sources, constitutes the overall process state vector. One way to express the overall process state vector is as follows: The meanings of each connected parameter are shown in Table 1 below:

[0050] Table 1 Furthermore, the final dimension d≈110 can be extended to meet the specific production line requirements. In practical applications, the number of dimensions can be improved according to the specific metallurgical production requirements.

[0051] Data cleaning refers to identifying and processing abnormal data using algorithms, specifically employing methods such as the Three Sigma rule based on statistical thresholds or the Isolation Forest algorithm, to eliminate outliers caused by noise interference or equipment malfunctions during sensor acquisition. Data completion involves filling in missing or invalid data, typically using linear interpolation or prediction models based on Long Short-Term Memory (LSTM) neural networks to ensure the integrity of the data sequence. Standardization eliminates the impact of different units of measurement on data fusion, using Z-score standardization or maximum-minimum normalization methods to unify the numerical range of different process parameters. Unified encoding converts unstructured data into structured features, using independent encoding or embedding techniques to address compatibility issues between different system data formats. Resampling unifies the temporal frequency of multi-source data, using linear interpolation or spline interpolation to downsample high-frequency data and upsample low-frequency data to a unified time base, achieving temporal alignment of data across processes.

[0052] Compared to existing technologies, traditional methods typically employ only simple data cleaning or interpolation, lacking a systematic processing flow for multi-source heterogeneous data in metallurgy. Existing technologies often discard low-frequency data or perform simple averaging when processing frequently sampled data, leading to distortion of time-series information. This solution constructs a tiered data processing flow, preserving the physical meaning of the original data while effectively integrating heterogeneous data through multi-dimensional data reduction, thus addressing four core issues: data anomalies, missing data, dimensional differences, and time-series misalignment.

[0053] Specifically, in the continuous casting process, IoT sensors collect data on crystallizer water temperature and casting speed vibration; the L2 system provides the setpoint for secondary cooling zone water distribution; and the MES system acquires production order priority data. After data preprocessing, a 32-dimensional state vector is generated. This vector is then input into a digital twin platform. The mechanistic model calculates the solidification front position of the billet, and the artificial intelligence model predicts the probability of surface cracks on the billet. When the predicted crack probability exceeds a threshold, the collaborative optimization engine generates a secondary cooling zone water volume adjustment strategy and a production rhythm optimization scheme. Virtual simulation verifies that the adjusted system extends the production cycle by 15 minutes while reducing the scrap rate. During execution, the optimized secondary cooling zone water volume parameters are sent to the PLC controller, while simultaneously monitoring actual billet quality data. When there is a deviation between the ultrasonic flaw detection results and the simulation prediction, the mechanistic model is driven to correct the heat transfer coefficient.

[0054] Compared to existing technologies, traditional methods can only formulate fixed production plans based on historical data and cannot dynamically respond to changes in operating conditions. For example, a company previously used an offline optimization model to adjust rolling parameters. When the thickness of the incoming material fluctuated, the parameters needed to be recalculated manually, resulting in a 30-minute production line downtime. This solution uses real-time data to drive a digital twin model, which can complete parameter optimization and simulation verification within 5 minutes, avoiding production interruptions.

[0055] Preferably, this application further proposes a technical solution for constructing four types of complementary models in a digital twin platform. The geometric model is defined as the three-dimensional model of the plant where the metallurgical equipment is located and the equipment structure model. The mechanism model includes the blast furnace reaction model, converter blowing model, continuous casting solidification and heat transfer model, rolling mechanics model, and heat treatment phase transformation model. The behavioral model describes the equipment operation logic through discrete event simulation or state machine. The data model maps the physical entity state based on real-time acquired data.

[0056] The geometric model refers to the spatial layout and equipment structure model of the plant constructed using 3D modeling technology. This can be achieved using CAD software or BIM tools, and an interactive 3D visualization interface can be generated by importing equipment drawings to visually display the spatial relationships of the equipment. The mechanistic model refers to a set of mathematical equations based on the physicochemical principles of metallurgical processes. Specifically, a finite element method can be used to establish a model of the iron ore reduction reaction in a blast furnace, and thermodynamic equations can be used to construct a solidification heat transfer model of the continuously cast billet to accurately simulate the material transformation process. The behavioral model refers to a temporal model describing the equipment's operating logic. Specifically, Petri nets can be used to construct a continuous casting machine casting rhythm model, and state transition diagrams can be used to express the rolling mill's working mode switching logic to characterize process connection relationships. The data model refers to the mapping mechanism between real-time data and the virtual model. Specifically, the OPC UA protocol can be used to achieve synchronous updates of sensor data and digital twins to maintain consistency between the virtual and real systems.

[0057] Specifically, the four types of models form a complete digital twin system through a collaborative working mechanism. The geometric model's 3D visualization interface provides operators with spatial monitoring support; for example, real-time observation of molten steel level changes can be seen in a 3D view of the converter workshop. The mechanistic model calculates process parameters based on physical equations; for example, predicting the final carbon content through a converter blowing model provides a theoretical benchmark for anomaly diagnosis. The behavioral model uses discrete event simulation technology to simulate production rhythms; for example, constructing a linkage model between the continuous casting machine's billet pulling speed and the rolling mill's bite sequence to ensure material balance between processes. The data model corrects the virtual model's state using real-time sensor data; for example, when a blast furnace temperature sensor detects abnormal fluctuations, the temperature field distribution of the digital twin is updated synchronously. The integrated application of these four types of models achieves a unity of interpretability of physical laws and adaptability to real-time data, establishing a multi-dimensional model foundation for anomaly prediction.

[0058] Compared to existing technologies, traditional methods typically employ only a single mechanistic model or a data-driven model, such as constructing a static model of a blast furnace solely through thermodynamic equations, lacking visualization support for the spatial layout of equipment. Existing systems, when diagnosing anomalies in the continuous casting process, cannot identify billet quality defects caused by production rhythm disruptions due to the lack of a behavioral model of the casting rhythm. This solution, by combining a three-dimensional geometric model with a discrete event behavioral model, can simultaneously monitor changes in equipment spatial location and process flow timing anomalies. For example, in the rolling process, it can visualize the location of roll wear and detect strip thickness fluctuations caused by abnormal rolling rhythms.

[0059] Through the above technical solutions, this application addresses the problems of insufficient visualization of equipment status and weak anomaly prediction capabilities in metallurgical processes. The three-dimensional geometric model enables visualized monitoring of equipment spatial distribution, the discrete event behavior model can identify production cycle anomalies, and the collaborative work of the mechanism model and data model improves the accuracy of status prediction. For example, in the continuous casting process, by predicting the surface temperature of the billet using a solidification heat transfer mechanism model and comparing it with real-time infrared temperature measurement data, a warning of surface crack defects can be issued 30 minutes in advance, allowing operators to adjust the water distribution in the secondary cooling zone in a timely manner.

[0060] Example 2 Unlike Example 1, in this further step, after the entire process state vector is input into the digital twin platform, the platform corrects the output of the mechanistic model using Kalman filtering or Bayesian update methods. In step S2, the artificial intelligence model predicts the production indicators of the metallurgical equipment based on a long short-term memory neural network or a Transformer model to obtain the steel composition, steel temperature, and power consumption information of the metallurgical equipment in the future state. The artificial intelligence model uses decision trees, SVM, or convolutional neural networks to perform anomaly diagnosis, identifying equipment faults, abnormal product quality, and abnormal production conditions.

[0061] The specific working method of the artificial intelligence model includes the following steps: S21. Calculate the real-time state vector using the data model. Based on the material flow model, energy flow model, information flow and quality evolution path model maintained by the data model in the digital twin platform, and combined with various types of data provided by the acquisition module at the current moment, calculate the current system state vector: ; in, The value of the nth state variable at the current time t (such as molten steel temperature, secondary cooling zone flow rate, oxygen lance position, furnace pressure, etc.); State vectors are dynamic snapshots of digital twin models and can be used to model the time-varying evolution of a system.

[0062] S22. Query the state trajectory over a historical period through the data model interface. Used to form the temporal input for modeling: ; in, This represents the sequence of state trajectories over the past k steps. Data sources include historical data buffers in the data model of the twin platform or offline data lakes (i.e., historical data). S23. Predict production indicators (such as steel temperature, composition, casting speed, etc. as examples, but not limited to these), using typical time series prediction models, such as LSTM (Long Short-Term Memory Network) or Transformer: ; in, For the prediction model, the parameters θ are obtained through training on historical data; The predicted values ​​of key indicators for the next H steps (e.g., molten steel temperature in the next 10 minutes); the feature input is... The label is the actual historical measurement value y; Specific, predictable indicators include, but are not limited to: Physical quantities in the metallurgical process (molten steel temperature, inclusion probability, composition, liquid core length); Remaining life of the equipment (based on load and vibration characteristics); Energy consumption trends (unit gas consumption / electricity consumption); Quality output indicators (such as first-grade product rate, defect rate); S24. Perform anomaly detection and diagnosis. The anomaly diagnosis process uses classification algorithms, such as Support Vector Machine (SVM), Random Forest, Convolutional Neural Network (CNN), Graph Neural Network (GNN), etc. ; in, This refers to the output exception label or fault category; This is the current state; It represents historical reference status / diagnostic templates (which can be modeled using clustering). From this, it can also output alarm levels (low / medium / high), root cause paths of faults, anomaly confidence scores, etc.

[0063] Common diagnostic tasks include: Continuous casting nozzle blockage warning (abnormal flow rate and pressure); Furnace temperature overshoot prediction; Abnormal fluctuations in rolling thickness; Diagnosis of water supply fluctuations in the secondary cooling zone system; In the specific implementation process, the model training data comes from historical production process data (annotated alarm / defect / parameter over-limit records) and simulation data generated by the digital twin platform (simulating various working conditions); the training method uses cross-validation to evaluate the model's generalization ability, and the anomaly diagnosis can use oversampling techniques such as SMOTE to process imbalanced data, and the model can be automatically updated according to real-time deviations.

[0064] Furthermore, the model's predicted output (such as the trend of molten steel temperature over the next 10 minutes) will be fed into the collaborative optimization engine to participate in scheduling and process parameter optimization.

[0065] This application further proposes that in a digital twin platform, an artificial intelligence model based on a long short-term memory neural network or a Transformer model predicts the production indicators of metallurgical equipment to obtain information on the composition, temperature, and power consumption of molten steel in the future state. The artificial intelligence model uses decision trees, SVMs, or convolutional neural networks to achieve anomaly diagnosis, identify equipment failures of metallurgical equipment, abnormal product quality, and abnormal production conditions.

[0066] Among them, Long Short-Term Memory Neural Network refers to a time series processing model with memory units and gating mechanisms. Specifically, it can be implemented using forget gate, input gate and output gate structures to capture the long-term dependence of steel composition and temperature changes in the metallurgical process.

[0067] The Transformer model refers to a deep learning model based on a self-attention mechanism. Specifically, it can be implemented using multi-head attention modules and positional encoding to extract global temporal features from multivariate time series data.

[0068] Among them, decision trees refer to classification models based on feature splitting rules. Specifically, information gain or Gini impurity can be used as the splitting criteria to determine product quality anomalies based on process parameter thresholds.

[0069] SVM refers to the Support Vector Machine classification model, which can be implemented by using kernel function mapping to process nonlinearly separable high-dimensional working condition data, and is used to identify abnormal patterns in production conditions.

[0070] Among them, convolutional neural networks refer to deep learning models with convolutional layers and pooling layers. Specifically, they can be implemented by using two-dimensional convolutional kernels to extract spatial features of equipment vibration signals or temperature field images, and are used to detect mechanical faults.

[0071] Specifically, multi-source heterogeneous data from the metallurgical process are preprocessed to form time series data, which are then input into the artificial intelligence model. A Long Short-Term Memory (LSTM) neural network uses a gating mechanism to filter key time-series information and predict the evolution trend of steel composition. The Transformer model uses self-attention weights to calculate the correlation between different time steps, predicting temperature gradients and equipment energy consumption. In the anomaly diagnosis module, a decision tree classifies excessive sulfur and phosphorus content in the molten steel according to preset rules; a SVM uses kernel functions to map high-dimensional operating data to a separable space, identifying abnormal furnace pressure or deviations in cooling water flow; and a convolutional neural network extracts local features from the equipment vibration spectrum to determine bearing wear or motor failure. The prediction and diagnosis modules share real-time data from the digital twin platform. When an anomaly is detected, parameter updates between the models are triggered, forming a closed-loop optimization mechanism.

[0072] Compared to existing technologies, traditional methods typically employ single statistical models or empirical rules for prediction and diagnosis, failing to effectively handle the temporal dynamics and spatial complexity of metallurgical data. For example, the existing ARIMA model can only handle linear time-series relationships and struggles to predict multivariate coupled changes in molten steel composition; threshold-based alarm anomaly detection methods cannot identify early, latent characteristics of equipment failures. This solution integrates time-series models and spatial feature extraction models, simultaneously covering the dynamic evolution of production indicators and the multi-dimensional representation of anomalies, thus addressing the problems of prediction lag and high false alarm rates inherent in traditional methods.

[0073] Through the above technical solutions, this application achieves high-precision prediction of metallurgical equipment production indicators and real-time accurate diagnosis of anomaly types. The prediction error of molten steel composition is reduced to within the allowable range of the process; equipment fault identification time is shortened to the millisecond level; quality anomaly detection covers various types, from excessive composition to surface defects; and the accuracy of production condition anomaly diagnosis meets online control requirements. Prediction results and diagnostic information interact in real time through a digital twin platform, forming a dynamic collaborative optimization basis, effectively avoiding chain production accidents caused by prediction deviations or diagnostic delays.

[0074] Furthermore, this application proposes that the digital twin platform, through a collaborative optimization engine, generates multi-dimensional optimization strategies based on anomaly diagnosis results, production indicator prediction information, and constraints. The time-dimensional strategy is a scheduling scheme, and the spatial-dimensional strategy is an adjustment of process parameters.

[0075] The collaborative optimization engine refers to a computational module based on multi-objective optimization algorithms and constraint satisfaction techniques. Specifically, it can be implemented using mixed integer programming or genetic algorithms to integrate anomaly diagnosis results, prediction information, and constraints to generate a global optimization strategy.

[0076] Among them, the time-dimensional strategy refers to a dynamic scheduling scheme that optimizes the rhythm of process connections and the start-up and shutdown sequence of equipment. Specifically, it can be implemented by using the rolling time-domain optimization method, which resolves resource allocation conflicts by adjusting the production cycle.

[0077] Among them, the spatial dimension strategy refers to the local optimization scheme for the operating parameters of metallurgical equipment. Specifically, it can be implemented by gradient descent or particle swarm optimization, and physical space coordination can be achieved by adjusting parameters such as blast furnace temperature and rolling mill pressure.

[0078] Specifically, the collaborative optimization engine first receives anomaly diagnosis results to identify current production bottlenecks, such as abnormal cooling water flow in the continuous casting machine. Then, it combines these with future steel composition trends predicted by an AI model, overlaid with constraints such as energy consumption limits and equipment load thresholds, to perform multi-dimensional coupled analysis. In the time dimension, it generates dynamic scheduling schemes, such as delaying converter tapping time to match the continuous casting machine anomaly repair cycle; in the spatial dimension, it generates process parameter adjustment instructions, such as reducing the mill roll gap setpoint to compensate for plate thickness deviations caused by temperature fluctuations. Both strategies are verified through virtual simulation and then simultaneously sent to the execution unit to ensure collaborative optimization of temporal dynamics and spatial coupling.

[0079] Compared to existing technologies, traditional methods typically employ single-dimensional optimization strategies, such as adjusting only production scheduling or optimizing only equipment parameters, leading to conflicts between local optimization and global objectives. For example, simply increasing the production cycle time may cause equipment to overload and shut down, while optimizing only the blast furnace temperature may disrupt the overall energy efficiency balance. This solution, through collaborative optimization in both time and space dimensions, avoids the equipment risks associated with scheduling optimization and eliminates the global imbalances caused by adjusting a single parameter.

[0080] Through the above technical solution, this application resolves the optimization conflict caused by the temporal dynamism and spatial coupling in the entire metallurgical process, achieving bidirectional coordination between production scheduling and process parameters. Specific effects include: dynamically adjusting the rhythm of upstream and downstream processes when the continuous casting machine malfunctions, avoiding a complete shutdown; and simultaneously optimizing the heating furnace temperature and rolling mill pressure when rolling process parameters fluctuate, reducing plate quality defects.

[0081] Specifically, this application further proposes that the final optimization strategy be sent to the field L1 control system and / or L2 control system in the system execution unit through the closed-loop control interface module.

[0082] The closed-loop control interface module refers to the physical channel used to connect the digital twin platform and the industrial field control system. Specifically, it can be implemented using OPC UA or Profinet protocol conversion interfaces. Its function is to convert the optimization strategy generated by virtual simulation into control commands that can be recognized by industrial equipment.

[0083] The L1 control system refers to the basic automation system that directly controls the actuators of metallurgical equipment. Specifically, it can be implemented using a PLC or DCS controller, and its function is to control the equipment's actions in real time.

[0084] The L2 control system refers to a process control system that sets process parameters. Specifically, it can be implemented using a process optimization computer, and its function is to dynamically adjust process parameters such as temperature and pressure.

[0085] Specifically, the closed-loop control interface module analyzes the optimization strategy output by the digital twin platform, decomposing it into equipment action commands required by the L1 system and process parameter setpoints required by the L2 system. For example, in the rolling process, when the optimization strategy requires adjusting the roll pressure, the module converts the pressure setpoint into a Modbus TCP protocol command and sends it to the L2 system; if the continuous casting machine speed needs to be changed, a Profibus DP protocol signal is generated and transmitted to the PLC controller of the L1 system. For cross-process collaboration scenarios, the module can simultaneously send blast temperature parameters to the blast furnace L2 system and oxygen lance height commands to the converter L1 system, achieving joint control in a spatial dimension. The protocol conversion function is completed through the built-in industrial communication middleware, ensuring compatibility of control commands from different manufacturers' equipment.

[0086] Compared to existing technologies, traditional methods rely on manual distribution of optimization strategies layer by layer to the workshop execution system, resulting in command distortion and transmission delays. In existing technologies, data interaction between the MES system and the underlying control system is typically limited to the transmission of production commands and cannot directly drive equipment parameter adjustments. This solution, through a closed-loop control interface module, penetrates the information hierarchy and directly connects to the L1 / L2 control system, eliminating the data packet loss problem caused by multi-system interface conversion in traditional architectures.

[0087] Through the above technical solution, this application achieves real-time comparison between virtual simulation results and actual production execution data. When there is a deviation between the actual rolling force fed back by the sensor and the set value of the optimization strategy, the model can be immediately triggered for iterative correction. At the same time, this solution supports two modes: independent optimization of a single process and collaborative control of multiple processes. For example, under abnormal operating conditions of the continuous casting machine, only the crystallizer vibration parameters of the L1 system can be adjusted, while during full-process energy efficiency optimization, the converter oxygen supply and mill temperature set values ​​of the L2 system can be issued simultaneously.

[0088] Other aspects that are the same as in Example 1 will not be repeated in this example.

[0089] Example 3 Unlike Example 1, this application also relates to a dynamic collaborative optimization system for the entire metallurgical process based on digital twins, specifically including: The data acquisition and fusion module is used to collect metallurgical process data, automatic control data from the control system, and enterprise management system data from the upper-level management system of the metallurgical equipment from the IoT sensors in the metallurgical equipment. The metallurgical process data, automatic control data, and enterprise management system data are fused to form multi-source heterogeneous data. Digital twin platforms are used to build digital twin models that include geometric models, mechanistic models, behavioral models, and data models, including artificial intelligence models for predicting and diagnosing anomalies in metallurgical equipment production indicators. The collaborative optimization engine receives real-time status information calculated by the digital twin platform and anomaly diagnosis results output by the artificial intelligence model, and generates multi-dimensional collaborative optimization strategies. The virtual simulation verification module receives the collaborative optimization strategy and performs virtual simulation on the digital twin platform to simulate the execution of the collaborative optimization strategy. The closed-loop control interface module transforms the optimal collaborative optimization strategy verified by virtual simulation into the execution of the field L1 control system and / or L2 control system in the system execution unit. The human-computer interaction interface module is used to visually display the real-time status, key parameters, material flow, and energy flow of the digital twin model; display the information processed in the digital twin platform; support operators to confirm, modify, or manually intervene in the digital twin platform; and provide functions for historical data query, report generation, and system configuration.

[0090] Specifically, the data acquisition and fusion module is responsible for real-time acquisition of various data from the metallurgical site, including but not limited to: (1) IoT data. Sensor data from key equipment such as blast furnaces, converters, continuous casting machines, and rolling mills (furnace temperature, pressure, flow rate, composition, vibration, current, rolling force, speed, cooling rate, etc.). (2) Process control data (L1 / L2). Data from basic automation and process control systems (set values, control commands, alarm information, historical trends, etc.). (3) Enterprise management data (ERP / MES / PLM / EMS, etc.). Data from upper-level management systems (production plans, material information, energy consumption, product quality data, equipment ledgers, maintenance records, etc.). This module also includes data cleaning, standardization, alignment, and preliminary fusion functions to build a unified data storage format.

[0091] Digital Twin Platform Module: This is the core of the system, responsible for building, maintaining, and operating a digital twin model of the entire metallurgical process. This model should include: (1) Geometric Model: Three-dimensional models of equipment, workshops, etc., for visualization. (2) Physical / Mechanism Model: Process models based on metallurgical principles (such as blast furnace internal reaction models, converter blowing models, continuous casting solidification heat transfer models, rolling force models, heat treatment phase transformation models, etc.), used to simulate key physicochemical processes. (3) Behavioral Model: Models describing equipment operating logic, process flow, production cycle time, etc.

[0092] The data model is used to organize and correlate real-time and historical data from the data acquisition module, mapping them to the various components and attributes of the digital twin model to form a data structure that can reflect the real-time state of the physical entity. This includes material flow models, energy flow models, information flow models, and quality evolution path models. This module can update the state of the digital twin model based on real-time data.

[0093] Artificial intelligence models utilize real-time status data and historical data from digital twin platforms, and employ AI algorithms such as machine learning and deep learning for prediction and diagnosis. (1) Prediction function: Predict the trend of key process parameters (such as changes in molten steel temperature and composition prediction), equipment remaining life, energy consumption prediction, and product quality prediction (such as potential defects). (2) Diagnosis function: Diagnose the causes of abnormalities in the production process, the root causes of equipment failures, and trace the source of quality defects.

[0094] The Collaborative Optimization Engine Module is the core intelligent decision-making unit for achieving dynamic collaborative optimization. It receives real-time status and predictive diagnostic results from the digital twin platform. Based on optimization objectives (such as maximizing total output, minimizing energy consumption, minimizing unit cost, and maximizing first-pass yield) and constraints (equipment capacity, process window, material supply, order delivery time, etc.), it generates multi-dimensional collaborative optimization strategies using optimization algorithms (such as mathematical programming, heuristic algorithms, and reinforcement learning). (1) Time dimension: Dynamically adjust production plans and scheduling, optimize furnace sequence, iron matching, continuous casting speed and steelmaking tapping rhythm matching, continuous casting billet entry time, etc., to achieve optimal production cycle time throughout the entire process.

[0095] (2) Spatial dimension: Optimize the process parameter settings of each equipment (such as converter oxygen lance position / flow rate, auxiliary material ratio, continuous casting secondary cooling water supply, rolling mill reduction procedure, heating furnace temperature curve, heat treatment heat preservation and cooling curve, etc.), and consider the parameter linkage of upstream and downstream processes.

[0096] The Virtual Simulation Validation Module receives optimization strategies generated by the collaborative optimization engine and runs them in a digital twin environment. Based on the digital twin model (including mechanistic and behavioral models), it simulates the execution of the optimization strategies and observes their impact on the overall process status, key indicators, and benefit objectives. It evaluates the feasibility, robustness, and potential risks of different optimization strategies. It outputs simulation result reports to assist or automatically select the optimal optimization strategy.

[0097] The Closed-loop Control Interface Module (CLI) converts the optimal optimization strategy (including scheduling instructions and process parameter setpoints) verified through virtual simulation into an instruction format recognizable by the field L1 / L2 control system and issues it for execution. It is responsible for reliable, real-time two-way communication with the field control system. Simultaneously, it receives feedback data from the field control system after execution and transmits it back to the data acquisition and fusion module, forming a closed loop.

[0098] The Human-Machine Interface Module (HMI) provides a user-friendly interface for visually displaying the real-time status, key parameters, and material and energy flows of the digital twin model; showcasing predictive diagnostic results and early warning information; displaying optimization strategy suggestions and simulation results; and supporting operator confirmation, modification, or manual intervention of optimization suggestions. It also provides historical data query, report generation, and system configuration functions.

[0099] Furthermore, embodiments of this application also provide a specific implementation of an electronic device capable of implementing all steps in the digital twin-based dynamic collaborative optimization method for the entire metallurgical process described in the above embodiments. See [link to implementation details]. Figure 2 The electronic devices specifically include the following: Processor 1201, memory 1202, communications interface 1203, and bus 1204; The processor 1201, memory 1202, and communication interface 1203 communicate with each other via bus 1204; the communication interface 1203 is used to realize information transmission between server-side devices, metering devices, and user-side devices and other related devices.

[0100] The processor 1201 is used to call the computer program stored in the memory 1202. When the processor executes the computer program, it implements all the steps in the dynamic collaborative optimization method for the entire metallurgical process based on digital twins in the above embodiments. In addition, the processor also supports dynamic adjustment of the simulation process according to the user's real-time instructions to meet the simulation requirements under different scenarios.

[0101] It is noteworthy that those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0103] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0104] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic collaborative optimization method for the entire metallurgical process based on digital twins, characterized in that, Includes the following steps: S1. The metallurgical process data, automatic control data of the control system, and enterprise management system data from the upper management system of the metallurgical equipment are collected by the IoT sensors installed in the metallurgical equipment to form multi-source heterogeneous data. The multi-source heterogeneous data is preprocessed to form a full-process state vector. S2. The entire process state vector is transmitted to the digital twin platform, which constructs a digital twin model including a geometric model, a mechanism model, a behavior model, and a data model. The digital twin platform calculates real-time state information based on the data model and imports historical data information. The artificial intelligence model located in the digital twin platform predicts and diagnoses anomalies in the production indicators of the metallurgical equipment based on the real-time state information and the historical data. S3. The digital twin platform generates multi-dimensional optimization strategies based on the anomaly diagnosis results, production indicator prediction information, and constraints. The optimization strategies include time-dimensional strategies and spatial-dimensional strategies. Virtual simulation is performed using the optimization strategies, and the optimization strategy with the best simulation results is selected as the final optimization strategy. S4. Select an optimization strategy and output it to the system execution unit of the metallurgical equipment. Collect the execution process information of the metallurgical equipment through the IoT sensor. Compare the execution information with the simulation information of the final optimization strategy. Based on the deviation, drive the mechanism model and the artificial intelligence model to learn and iteratively correct.

2. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 1, characterized in that, In step S1, the metallurgical process data collected by the IoT sensor includes data from blast furnace, converter, continuous casting machine and rolling mill; the automatic control data is data from L1 and / or L2 process control systems, including set values, control commands, alarm information and historical trend information; the enterprise management system data includes data from at least one upper-level management system from EPR system, MES system, PLM system and EMS system.

3. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 1, characterized in that, In step S1, the process of preprocessing the collected multi-source heterogeneous data to form the full-process state vector includes the following steps: S11. Perform data cleaning on the multi-source heterogeneous data collected from the IoT sensor, statistically analyze the abnormal detection data, and process the abnormal data in any way, such as removing, truncating, or replacing with predicted values. S12. Complete the missing or abnormal data, where the completed data is the predicted value; S13. The data processed in step S12 is standardized and uniformly encoded. S14. The data processed in step S13 is resampled to a uniform time step according to different sampling frequencies. S15. The data processed in step S14, together with other heterogeneous data from multiple sources, constitute the full-process state vector.

4. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 1, characterized in that, In step S2, the geometric model constructed by the digital twin platform is a three-dimensional model of the plant where the metallurgical equipment is located and a structural model of the metallurgical equipment, which is used to achieve visualization. The mechanism model includes a blast furnace reaction model, a converter blowing model, a continuous casting solidification and heat transfer model, a rolling mechanics model, and a heat treatment phase transformation model. The behavioral model describes the operating logic, process flow, and production cycle of the metallurgical equipment through discrete event simulation or state machine. The data model is based on the real-time mapping status of the collected data.

5. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 4, characterized in that, After the entire process state vector is input into the digital twin platform, the digital twin platform corrects the output of the mechanism model using Kalman filtering or Bayesian update methods.

6. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 1, characterized in that, In step S2, the artificial intelligence model predicts the production indicators of the metallurgical equipment based on a long short-term memory neural network or a Transformer model to obtain the steel composition, steel temperature and power consumption information of the metallurgical equipment in the future state. The artificial intelligence model realizes anomaly diagnosis through decision tree, SVM or convolutional neural network to identify equipment failures of the metallurgical equipment, abnormal quality of the produced products and abnormal state information of the production conditions.

7. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 1, characterized in that, The digital twin platform generates multi-dimensional optimization strategies based on the anomaly diagnosis results, production indicator prediction information, and constraints through a collaborative optimization engine. The time-dimensional strategy is a scheduling scheme, and the spatial-dimensional strategy is the adjustment of process parameters.

8. The dynamic collaborative optimization method for the entire metallurgical process based on digital twins according to claim 1, characterized in that, Finally, the optimization strategy is sent to the field L1 control system and / or L2 control system in the system execution unit through the closed-loop control interface module for execution.

9. A dynamic collaborative optimization system for the entire metallurgical process based on digital twins, characterized in that, The metallurgical whole-process dynamic collaborative optimization system has instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the metallurgical whole-process dynamic collaborative optimization method based on digital twin according to any one of claims 1 to 8.

10. A dynamic collaborative optimization system for the entire metallurgical process based on digital twins as described in claim 9, characterized in that, include: The data acquisition and fusion module is used to collect metallurgical process data, automatic control data from the control system, and enterprise management system data from the upper-level management system of the metallurgical equipment from the IoT sensors in the metallurgical equipment, and to fuse the metallurgical process data, automatic control data, and enterprise management system data to form multi-source heterogeneous data. A digital twin platform is used to construct a digital twin model that includes a geometric model, a mechanism model, a behavior model, and a data model, including an artificial intelligence model for predicting and diagnosing anomalies in the production indicators of the metallurgical equipment. The collaborative optimization engine receives real-time status information calculated by the digital twin platform and anomaly diagnosis results output by the artificial intelligence model, and generates multi-dimensional collaborative optimization strategies. The virtual simulation verification module receives the collaborative optimization strategy and performs virtual simulation on the digital twin platform to simulate the execution of the collaborative optimization strategy. The closed-loop control interface module transforms the optimal collaborative optimization strategy verified by virtual simulation into the execution of the field L1 control system and / or L2 control system in the system execution unit. The human-computer interaction interface module is used to visually display the real-time status, key parameters, material flow, and energy flow of the digital twin model; display the information processed in the digital twin platform; support operators to confirm, modify, or manually intervene in the digital twin platform; and provide functions for historical data query, report generation, and system configuration.

Images