Intelligent control method and system based on digital twinning
By constructing a digital twin model and combining it with optimization algorithms and machine learning prediction models, real-time closed-loop optimization control of process-oriented production processes was achieved, solving the problem of insufficient control precision in existing technologies and improving the stability and efficiency of the production process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING HONGXIN ZHAOYANG TECH CO LTD
- Filing Date
- 2026-04-14
- Publication Date
- 2026-06-12
Smart Images
Figure CN122194925A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of industrial process control and optimization technology, and in particular to an intelligent control method and system based on digital twins. Background Technology
[0002] In the pursuit of efficient, stable, and low-consumption modern production, optimizing key control parameters of process production is one of the core issues for improving product quality, reducing energy consumption and production costs, and maximizing economic benefits. Process production (such as in the chemical, tobacco, metallurgical, pharmaceutical, and food industries) often has complex characteristics such as continuity, nonlinearity, time-varying nature, and strong coupling of multiple variables. This makes it increasingly difficult and inefficient to set process parameters solely through traditional experience-based adjustments or iterative trial and error methods, failing to guarantee that the production process is always in its optimal state.
[0003] Digital twin technology offers a new solution to these problems. Digital twins create high-fidelity virtual models in information space that correspond to physical entities, thereby enabling control over physical production processes. However, existing digital twin models have relatively poor precision in controlling physical production processes. Summary of the Invention
[0004] The purpose of this application is to provide an intelligent control method and system based on digital twins.
[0005] In a first aspect, embodiments of this application provide an intelligent control method based on digital twins, comprising: Determine the objectives that need to be optimized in the production process, the key parameters to be optimized, and the constraints that the parameters to be optimized must meet; Construct a digital twin model of the production process; wherein the digital twin model includes a feedforward and feedback composite control structure model, and the digital twin model is used to simulate the dynamic behavior of the physical production equipment and its control system; Based on the objective, the key parameters to be optimized, and the constraints, an optimization algorithm is determined, and the key parameters to be optimized are solved using the optimization algorithm to obtain the optimized key parameters. The optimized key parameters are input into the digital twin model for simulation verification; Construct a predictive model; the predictive model is used to predict the changing trend of optimized key parameters after a preset future time. The prediction model is embedded into the digital twin model to form a digital twin control system. The digital twin model generates feedforward control commands based on the prediction results of the prediction model, so that the digital twin model can simulate control behavior with prediction feedforward correction to control the actuators of the physical production device. The tested digital twin control system is deployed into the actual production control system. Real-time data from the physical production equipment, obtained through sensors and communication networks, is input into the digital twin model. The digital twin model and the optimization algorithm determine the optimized control commands, which are then sent to the actuators of the physical production equipment. This achieves real-time closed-loop optimized control of the physical production process by the digital twin control system.
[0006] In some embodiments, the process of constructing the digital twin model includes: The digital twin model is established using a hybrid modeling method that combines mechanistic modeling and digital-driven modeling. The digital twin model includes a cascade control system model of a main control loop and a secondary control loop, as well as a feedforward control channel for known disturbances. The cascade control system model and the feedforward control channel constitute the feedforward and feedback composite control structure model.
[0007] In some embodiments, the optimization algorithm includes one or more of the following algorithms: linear programming algorithm, nonlinear programming algorithm, dynamic programming algorithm or integer programming algorithm, genetic algorithm, particle swarm optimization, simulated annealing algorithm, ant colony optimization algorithm, neural network optimization algorithm, and hybrid optimization algorithm that combines traditional mathematical programming with the algorithm.
[0008] In some embodiments, the optimized key parameters are input into the digital twin model for simulation verification, including: The optimized key parameters are input into the digital twin model, and the digital twin model is tested in a virtual environment by applying different input signals and disturbances. The dynamic response and performance indicators of the system output are observed to verify the optimized key parameters through simulation. The simulation verification includes: parameter sensitivity analysis, model control parameter tuning, and robustness analysis of the optimized key parameters.
[0009] In some embodiments, constructing a prediction model includes: Based on historical operational data and process mechanisms, a prediction model is constructed using machine learning or deep learning algorithms; wherein, the prediction model includes one or more of time series prediction models, regression prediction models, and neural network prediction models.
[0010] In some embodiments, the digital twin model generates feedforward control commands based on the prediction results of the prediction model, enabling the digital twin model to simulate control behavior with predictive feedforward correction, including: The prediction results of the prediction model are input into the prediction controller in the digital twin model; Based on the prediction results, the predictive controller calculates control actions in advance according to the deviation of future production operation status and generates feedforward control instructions, so that the digital twin model can simulate control behavior with predictive feedforward correction based on the feedforward control instructions.
[0011] In some embodiments, the communication network employs industrial Internet of Things (IoT) technology, including 5G communication, industrial Ethernet, or wireless sensor networks, to achieve high-speed data transmission between the digital twin control system and the physical production equipment. The actuator includes at least one of a programmable logic controller, a distributed control system, and a robot controller, to receive instructions from the digital twin control system and act directly on the physical production process.
[0012] Secondly, embodiments of this application provide an intelligent control system based on digital twins, comprising: A digital twin simulation module is used to construct a digital twin model of a production process; wherein, the digital twin model includes a feedforward and feedback composite control structure model, and the digital twin model is used to simulate the dynamic behavior of physical production equipment and its control system; The optimization decision module, connected to the digital twin simulation module, is used to determine the optimization algorithm based on the determined production process optimization objectives, the key parameters to be optimized, and the constraints that the parameters to be optimized must satisfy. The optimization algorithm is then used to solve for the key parameters to be optimized to obtain the optimized key parameters. The prediction module, integrated with the digital twin simulation module, is used to construct a prediction model. The prediction model is used to predict the changing trend of optimized key parameters after a preset time in the future. By embedding the prediction model into the digital twin model, a digital twin control system is formed, enabling the digital twin model to generate feedforward control commands based on the prediction results of the prediction model. Thus, the digital twin model can simulate control behavior with predictive feedforward correction to control the actuators of the physical production device. The data acquisition and communication module is used to acquire real-time data of the physical production device through sensors, establish a data communication network between the physical space and the digital twin simulation module, input the real-time data to the digital twin simulation module through the data communication network, and determine the optimized control command through the digital twin model and the optimization algorithm. The execution module includes one or more physical production devices, used to execute the optimization control commands issued by the digital twin simulation module, thereby realizing real-time closed-loop optimization control of the physical production process.
[0013] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of any of the methods described in the foregoing embodiments.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of any of the methods described in the foregoing embodiments.
[0015] This application provides a digital twin-based intelligent control method that utilizes digital twin technology to offer a virtual-real integrated platform. This platform allows for repeated testing and optimization of solutions without affecting actual production, significantly improving the efficiency and safety of parameter optimization. By using an optimization algorithm to globally search for optimal parameters, the method avoids the blindness of manual parameter tuning, ensuring that the production process achieves optimization under multi-objective and multi-constraint conditions. The introduction of a machine learning prediction model enables proactive prediction and compensatory control of future disturbances, improving the response speed and stability of the control system. Finally, a real-time optimized control system with a virtual-real closed loop is constructed, continuously and adaptively adjusting the control strategy based on field feedback, ensuring that the production process always operates close to optimal conditions. The combined effect of these technologies can improve product quality consistency, reduce energy and raw material consumption, and decrease manual intervention and trial-and-error costs. It has broad application prospects and significant economic and social benefits in the digital and intelligent transformation of process industries. The digital twin-based intelligent control method provided in this application can significantly improve the accuracy and robustness of complex production process control, enhance product quality, and reduce energy consumption. Attached Figure Description
[0016] Figure 1 A flowchart illustrating an intelligent control method based on digital twins provided in an embodiment of this application; Figure 2 This is a schematic diagram of a digital twin model with an embedded prediction model provided in an embodiment of this application. Detailed Implementation
[0017] Various embodiments and features of this application are described herein with reference to the accompanying drawings.
[0018] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this application will be apparent to those skilled in the art.
[0019] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.
[0020] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.
[0021] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application.
[0022] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.
[0023] Specific embodiments of this application are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in a variety of substantially any suitable detailed structures.
[0024] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.
[0025] Digital twins create high-fidelity virtual models corresponding to physical entities in a cyberspace, ensuring real-time data synchronization and interaction between the virtual and physical environments. This allows for the simulation, analysis, and optimization of real production processes within a virtual environment. Using digital twin models, the impact of various operating conditions and disturbances on the production process can be simulated in advance in virtual space, enabling parameter sensitivity analysis and control strategy testing. The effectiveness of parameter optimization schemes can be verified through virtual simulation (simulating the real world). Simultaneously, digital twins can establish real-time connections with physical entities via IoT devices and sensors, directly guiding the operation of physical equipment after optimization decisions are made in the virtual space (virtual control of the real world), forming a closed-loop control system. Therefore, combining digital twin technology with parameter optimization and intelligent control can solve the problems of inaccurate models and untimely responses in existing technologies, achieving continuous and adaptive optimization control of the production process.
[0026] This application provides a digital twin-based intelligent control method that can be widely applied to process parameter optimization and intelligent control in process-oriented production industries such as chemical, tobacco, metallurgy, pharmaceutical, and food processing. See also... Figure 1 As shown, intelligent control methods based on digital twins may include: S10, determine the objectives that need to be optimized in the production process, the key parameters to be optimized, and the constraints that the parameters to be optimized need to meet; This step aims to determine the optimization objectives and key parameters (i.e., key process parameters). It involves identifying the optimization objectives for the production process, selecting several adjustable key process parameters as optimization decision variables, and setting the process and safety constraints that each decision variable must satisfy. Specifically, optimization objectives can be determined based on production process requirements, and these can include a single objective or multiple comprehensive objectives. A single objective might be, for example, maximizing product quality or minimizing energy consumption. A comprehensive objective might be, for example, achieving the optimal overall benefit that balances quality, efficiency, and cost.
[0027] Identify the key process parameters that are adjustable in the production process and have a significant impact on the optimization objective as the key parameters to be optimized, and clarify the constraints that each key parameter must meet, such as process specifications, physicochemical limitations, safety indicators, and equipment performance limits.
[0028] S20, Construct a digital twin model of the production process; wherein, the digital twin model includes a feedforward and feedback composite control structure model, and the digital twin model is used to simulate the dynamic behavior of physical production equipment and its control system; This step aims to establish a digital twin model. A digital twin model of the production process is constructed to simulate the dynamic behavior of the physical production equipment and its control system. This digital twin model includes virtual simulation models of feedforward and feedback control structures. Specifically, a combination of mechanistic models and data-driven models can be used to obtain high-fidelity process and controller models.
[0029] S30. Based on the objective, the key parameters to be optimized, and the constraints, determine the optimization algorithm, and solve for the key parameters to be optimized using the optimization algorithm to obtain the optimized key parameters. This step aims to select an optimization algorithm and solve for key parameters. Based on the nature of the optimization objective and the problem size, a suitable optimization algorithm can be selected to optimize the decision variables. The optimization algorithm can include traditional mathematical programming algorithms or intelligent optimization algorithms, resulting in a combination of key parameters that enables the optimization objective to reach a predetermined optimal value.
[0030] Based on the defined optimization objective, key parameters to be optimized, and constraints, a suitable optimization algorithm is selected to solve for the optimal combination of key parameters. For example, for linear or convex optimization problems, linear programming or nonlinear programming methods can be used; for complex problems with highly nonlinearity, strong coupling, or multiple local optima, intelligent optimization algorithms are preferred, such as genetic algorithms (GA), particle swarm optimization (PSO), simulated annealing, ant colony optimization, or neural network-based optimization methods. A hybrid optimization strategy combining traditional optimization methods with intelligent algorithms can also be used to improve global optimization capability and convergence efficiency. The optimal key process parameter settings or controller parameter combinations that achieve the expected optimization objective are obtained through iterative solutions on a digital twin model using optimization algorithms.
[0031] S40: Input the optimized key parameters into the digital twin model for simulation verification; This step aims to simulate and verify the optimization effect. The optimized key parameters are input into the digital twin model for simulation verification. Different input signals and disturbances are applied to the system in a virtual environment to test the system output response. The optimization scheme is then verified and evaluated, which may include analysis of parameter sensitivity, control performance, and robustness. Based on the simulation results, the key optimization parameters or the digital twin model may be adjusted if necessary.
[0032] S50, Build a predictive model; the predictive model is used to predict the changing trend of optimized key parameters after a preset time in the future; This step aims to build a predictive model. A machine learning or deep learning model can be trained based on historical operational data and mechanistic knowledge to serve as a predictive model for key process variables, used to predict the state of the production process after a predetermined time.
[0033] S60 embeds the predictive model into the digital twin model to form a digital twin control system. The digital twin model generates feedforward control commands based on the prediction results of the predictive model, enabling the digital twin model to simulate control behavior with predictive feedforward correction in order to control the actuators of the physical production device. This step aims to integrate the predictive model into the digital twin model. The predictive model can be embedded into the digital twin model, which generates feedforward control commands based on the prediction results. This allows the digital twin model to simulate control behavior with predictive feedforward correction. The system performance before and after introducing predictive control under different operating conditions can be simulated and compared, and the predictive controller parameters of the digital twin model can be adjusted to achieve the best control effect.
[0034] S70 deploys the tested digital twin control system into the actual production control system, and inputs real-time data of the physical production device obtained through sensors and communication networks into the digital twin model. The digital twin model and optimization algorithm determine the optimized control commands, and send the optimized control commands to the actuators of the physical production device, thereby realizing the real-time closed-loop optimized control of the physical production process by the digital twin control system.
[0035] This step aims to deploy a closed-loop control system. The tested digital twin control system is deployed into the actual production control system at the production site. Real-time data from the physical production equipment, acquired through sensors and communication networks, is input into the digital twin model. The digital twin model and optimization algorithms calculate optimized control commands, which are then sent to the actuators of the physical production equipment, thus achieving real-time closed-loop optimized control of the physical production process by the digital twin control system.
[0036] This application provides a virtual-physical integrated platform using digital twin technology, enabling repeated testing and optimization of solutions without affecting actual production, significantly improving the efficiency and safety of parameter optimization. By globally searching for optimal parameters through optimization algorithms, the blindness of manual parameter tuning is avoided, ensuring that the production process achieves optimization under multi-objective and multi-constraint conditions. The introduction of machine learning prediction models enables proactive prediction and compensatory control of future disturbances, improving the response speed and stability of the control system. Finally, a real-time optimization control system with a virtual-physical closed loop is constructed, continuously and adaptively adjusting the control strategy based on on-site feedback, ensuring that the production process always operates close to optimal conditions. The combined effect of these technologies can improve product quality consistency, reduce energy and raw material consumption, and reduce manual intervention and trial-and-error costs, demonstrating broad application prospects and significant economic and social benefits in the digital and intelligent transformation of process industries. The digital twin-based intelligent control method provided in this application can significantly improve the accuracy and robustness of complex production process control, enhance product quality, and reduce energy consumption.
[0037] The embodiments of this application can be summarized in the following three points regarding their systematic nature, forward-looking approach, and closed-loop intelligence: (1) A complete closed-loop system combining physical and digital elements has been constructed: The core innovation of this application lies not only in creating a static simulation model, but also in constructing a digital twin that interacts with and co-evolves with the physical entity in real time. It integrates high-precision modeling, intelligent optimization, virtual simulation, machine learning and field control into a complete, dynamic closed-loop optimization control system, realizing full-process intelligence from data perception, analysis and decision-making to precise execution.
[0038] (2) A shift in control paradigm from "passive response" to "active prediction" has been achieved: Traditional control methods are mostly "feedback control," meaning that corrections are only made after a problem occurs. This application introduces "machine learning prediction to achieve feedforward control," which is a significant improvement. The digital twin control system can predict potential fluctuations or disturbances in the production process based on real-time data and adjust parameters in advance, eliminating problems before they occur. This proactive control approach greatly enhances the robustness and control accuracy of the system.
[0039] (3) A safe, efficient, and low-cost optimization verification platform is provided: In actual production, directly trying new combinations of process parameters is risky and costly. This application uses a digital twin model to create a virtual simulation environment. All new parameters found by the optimization algorithm can be quickly and safely tested and verified in this virtual environment, and their effects can be confirmed before being deployed to physical production. This greatly shortens the R&D cycle, reduces trial and error costs, and solves the pain point of "difficulty in efficiently finding the best solution".
[0040] In some embodiments, the process of constructing a digital twin model includes: A hybrid modeling approach combining mechanistic modeling and digital-driven modeling is adopted to establish a digital twin model. The digital twin model includes a cascade control system model of the main control loop and the secondary control loop, as well as a feedforward control channel for known disturbances. The cascade control system model and the feedforward control channel constitute a feedforward and feedback composite control structure model.
[0041] Optionally, a mathematical model framework is established based on the physical mechanism of the controlled object. System identification is performed using experimental or historical data to calibrate model parameters, obtaining a model consistent with the dynamic characteristics of the actual production process. The digital twin model may include a main controller, a secondary controller, and their corresponding controlled object models, as well as measurement and execution element models, and has a feedforward control channel for known disturbances. (See [link to relevant documentation]). Figure 2 As shown.
[0042] In this embodiment, the digital twin model is used to simulate the dynamic behavior of the physical production equipment and its control system. The digital twin model includes virtual simulations of the main stages of the production process and their control structures, such as a feedforward-feedback composite control structure model containing a main control loop and a secondary control loop.
[0043] Optionally, a hybrid modeling approach combining mechanistic modeling and data-driven modeling can be used to establish the digital twin model. Specifically, firstly, a basic mathematical model of the physical process (differential equations or transfer functions, etc.) is constructed based on first principles and domain mechanistic knowledge. Then, historical or experimental data are used to identify and correct unknown parameters in the basic mathematical model through system identification methods, or data-driven modeling is used to compensate for deviations between the mechanistic model and the actual production process, thereby obtaining a high-fidelity digital twin model. The system's transfer function is determined using the above methods, including the control laws of the main and secondary controllers, the dynamic characteristics of the primary and secondary controlled objects, the dynamic response models of measurement sensors and actuators (regulating valves, etc.), and the model of the feedforward control channel, ensuring that the digital twin model accurately reflects the dynamic performance of the physical system.
[0044] In a cascade control system, the master controller and slave controller play distinct roles and follow specific control laws to achieve precise control of the final controlled variable. The master controller typically employs a proportional-integral-derivative (PID) control law, or, depending on specific process requirements, proportional-integral (PI) control. PID control comprehensively considers current deviations, historical deviation accumulation, and future deviation trends, thus enabling comprehensive and precise adjustment. The output signal of the master controller does not directly act on the actuator (such as a valve) but serves as the setpoint for the slave controller.
[0045] The slave controller typically employs proportional (P) or proportional-integral (PI) control. Because the slave loop requires a fast response, derivative (D) action is rarely used to avoid excessive sensitivity to measurement noise. In many cases, proportional control alone is sufficient to meet the requirements for rapid adjustment. The output signal of the slave controller directly drives the final actuator (such as the opening of a control valve), directly intervening in the controlled object.
[0046] In some embodiments, the optimization algorithm includes one or more of the following: linear programming algorithm, nonlinear programming algorithm, dynamic programming algorithm or integer programming algorithm, genetic algorithm, particle swarm optimization, simulated annealing algorithm, ant colony optimization algorithm, neural network optimization algorithm, and hybrid optimization algorithm that combines traditional mathematical programming with algorithms.
[0047] Selecting an optimization algorithm and solving for key parameters involves choosing and applying an optimization algorithm to optimize parameters.
[0048] In optimal control, parameter optimization refers to finding the optimal control parameters (such as control inputs, state variables, or system parameters) through numerical or analytical methods to minimize or maximize the objective function (such as cost function or performance index) while satisfying dynamic system constraints (such as state equations). Since optimal control problems are often nonlinear, nonconvex, and involve high-dimensional parameter spaces, directly solving analytical solutions is usually not feasible. Therefore, efficient numerical optimization algorithms are needed to accelerate convergence, reduce computational complexity, and handle uncertainties or constraints. The core of efficient optimization lies in: (1) using gradient information or surrogate models to reduce the number of iterations; (2) reducing dimensionality through parallel computing or low-dimensional parameterization; (3) combining global and local searches to avoid local optima; and (4) using hot start or surrogate acceleration in real-time applications (such as Model Predictive Control, MPC).
[0049] Alternatively, gradient descent: parameter optimization is performed by iteratively updating the parameters along the negative gradient direction; Alternatively, the Levenberg-Marquardt algorithm combines gradient descent and the Gauss-Newton method, adaptively adjusting the step size through damping parameters to minimize the squared error for parameter optimization. Alternatively, Newton's method and its variants (such as the Conjugate Gradient) can accelerate convergence by using the second derivative (Hessian matrix) for parameter optimization. Alternatively, Bayesian optimization: uses surrogate models such as Gaussian processes to select evaluation points through acquisition functions (as expected) to optimize parameters.
[0050] In some embodiments, the optimized key parameters are input into the digital twin model for simulation verification, including: The optimized key parameters are input into the digital twin model, and the digital twin model is tested in a virtual environment by applying different input signals and disturbances. The dynamic response and performance indicators of the system output are observed to verify the optimized key parameters through simulation. The simulation verification includes: parameter sensitivity analysis, model control parameter tuning, and robustness analysis of the optimized key parameters.
[0051] Specifically, sensitivity analysis is performed on the digital twin model to identify parameters and disturbances that significantly affect system performance; different controller parameter settings are tried in a virtual environment and the system dynamic response is compared to tune and optimize the controller parameters; and process parameter deviations or external disturbances are simulated in the digital twin model to evaluate the robustness of the optimized control scheme.
[0052] Specifically, the optimized key parameters are input into the digital twin model for simulation to verify its effectiveness. Various typical input signals and possible disturbances are applied in a virtual environment, and the system's dynamic response and performance indicators are observed to evaluate the effectiveness of the optimization scheme. Simulation verification includes, but is not limited to: (1) Parameter sensitivity analysis: By changing different input variables or disturbance conditions, the system response is simulated on the digital twin model. The influence of each key parameter on the optimization objective and system performance is analyzed. The selected combination of key optimization parameters can maintain good performance under various working conditions. The factors that have the greatest impact on the system are identified and the digital twin model is robust enough to them.
[0053] (2) Model control parameter tuning: Try different control parameter settings (e.g., proportional, integral, and derivative parameter combinations for PID control) on the digital twin model, compare the corresponding system responses (e.g., overshoot, steady-state error, response time, etc.), and select the optimal parameter combination that optimizes system performance. This effectively uses digital twins for virtual tuning, reducing the workload of actual main and auxiliary control loop parameter tuning.
[0054] (3) Robustness analysis: Simulate whether the control system of the digital twin model can still maintain stable control and keep key production indicators within the allowable range when process parameters or environmental conditions change within a certain range (e.g., fluctuations in raw material characteristics, parameter deviations caused by equipment aging).
[0055] Multi-scenario simulations were used to verify the robustness of the optimized control scheme against uncertainties in the digital twin model and external disturbances, ensuring its disturbance resistance capability in practical applications.
[0056] In some embodiments, constructing a prediction model includes: Based on historical operational data and process mechanisms, a prediction model is constructed using machine learning or deep learning algorithms; the prediction model includes one or more of time series prediction models, regression prediction models, and neural network prediction models.
[0057] The predictive model is built using machine learning or deep learning algorithms. The input to the predictive model is real-time and historical process data of the production process, and the output is the predicted value of one or more key variables at a predetermined future time.
[0058] Specifically, a predictive model for key variable prediction is trained based on historical operational data and process mechanisms. Preferably, machine learning or deep learning algorithms (such as support vector machines, time series models, or neural networks like LSTM and GRU) can be used to learn the dynamic patterns of the production process from historical data. This predictive model takes current and past process parameters and states as input and can predict the changing trends of key control variables (such as product quality indicators and key process parameter values) at future time t (or several future time steps).
[0059] In some embodiments, the digital twin model generates feedforward control commands based on the prediction results of the prediction model, enabling the digital twin model to simulate control behavior with predictive feedforward correction, including: The prediction results of the prediction model are input into the prediction controller in the digital twin model; Based on the prediction results, the predictive controller calculates control actions in advance according to the deviation of future production operation status and generates feedforward control commands, so that the digital twin model can simulate control behavior with predictive feedforward correction based on the feedforward control commands.
[0060] Specifically, the prediction results output by the predictive model are used as auxiliary control signals and input to the predictive controller in the digital twin model, so that the predictive controller can calculate the control action in advance based on the future state deviation and generate feedforward control commands. The response performance of the system under disturbance is compared with and without predictive control through digital twin model simulation, so as to adjust the parameters or model structure of the predictive controller.
[0061] Figure 2 A schematic diagram of a digital twin model that incorporates a prediction model. Figure 2 The digital twin model includes a schematic diagram of a feedforward and feedback composite control structure model, and shows a dual-loop structure in which the main controller and the sub-controller control the main object and the sub-object respectively, as well as a control block diagram in which the feedforward controller pre-compensates the main loop using known disturbances.
[0062] like Figure 2 As shown, the predictive model is integrated into the digital twin model to form a future-oriented intelligent control strategy. Specifically, a predictive controller is added to the digital twin simulation platform to collect key parameter data of the current physical production process in real time (such as...). Figure 2 The process operating parameters are input into the equipment, process, and other operating parameter database and fed into the prediction model. The prediction model then obtains the prediction result of the future system behavior at time t. The prediction result is compared with the given optimization objective (such as...). Figure 2A comparison operation is performed on a given value R1(s) to obtain the comparison result, which is then output to the predictive controller. The predictive controller generates feedforward control commands or auxiliary control commands based on the comparison result. These feedforward or auxiliary control commands, generated by the predictive controller, serve as additional inputs to the secondary controller in the digital twin simulation. This allows the digital twin model to simulate closed-loop behavior with "predictive control" in the simulation, i.e., to correct future deviations in advance using virtual space. The feedforward controller utilizes known disturbances (such as...) Figure 2 The input parameters (such as disturbances caused by input parameter fluctuations) are pre-compensated for in the main loop. In this way, the digital twin control system is upgraded to an intelligent control system, which can adjust the control input in advance based on predicted future deviations, thereby improving the response speed and accuracy of the entire control system and reducing lag and overshoot.
[0063] Input parameters refer to the input parameters of the process, such as the moisture content and flow rate of the incoming material, while process operating parameters refer to the relevant process and equipment parameters for equipment operation.
[0064] When the feedforward controller receives input parameters, it calculates and sets the relevant operating parameters of the equipment, usually based on a mathematical model.
[0065] Predictive controllers adjust based on predicted key parameter states, similar to master controllers. The difference lies in the objects they control. Master controllers adjust based on current key parameter states, while predictive controllers adjust based on future states.
[0066] Optionally, the predictive model is embedded into the digital twin model to form a digital twin control system. The digital twin model generates feedforward control commands based on the prediction results of the predictive model, enabling the digital twin model to simulate control behavior with predictive feedforward correction to control the actuators of the physical production device. After that, the process includes testing the digital twin control system.
[0067] This step aims to test the predictive control performance of the digital twin control system. Specifically, in a digital twin simulation environment, a comprehensive test is conducted on the digital twin control system integrating the predictive model. Various input parameter fluctuations, process disturbances, and model uncertainties are simulated, comparing the system performance before and after the introduction of predictive control, and verifying the effectiveness of the predictive model in improving system stability, reducing overshoot and oscillations, and handling delays and constraints. Evaluation metrics may include tracking error, fluctuation amplitude, and energy consumption changes. Through repeated simulation experiments, the predictive model or control strategy is continuously adjusted and improved to ensure the accuracy and reliability of the predictive model.
[0068] A tested and performance-compliant digital twin control system is deployed into the actual production control system to achieve optimized control of the physical space from the virtual space. To this end, a two-way communication link is established between the physical production unit and the digital twin model: real-time data (including process parameters, equipment status, environmental parameters, etc.) from the physical production unit is collected in real time via industrial IoT devices and sensor networks and transmitted to the digital twin model; the digital twin model analyzes the real-time data in its information space, compares it with its internal virtual model, and calculates optimized control commands in real time based on the optimization objectives and control strategies, such as optimized control parameters, process commands, or operation sequences. Subsequently, the optimized control commands are sent to the actuators of the physical production unit, such as programmable logic controllers (PLCs), robot controllers, distributed control systems (DCS), and supervisory control systems (SCADA), via industrial communication networks (such as 5G communication and industrial Ethernet). Upon receiving the optimized control commands, the physical production unit automatically adjusts its operating status or process setpoint, ensuring that the actual production process operates according to the optimized control commands. Meanwhile, any state changes resulting from the execution of optimized control commands are again collected by sensors and fed back to the digital twin model, thus forming a closed-loop control between the information space and the physical space. Through continuous real-time feedback, the digital twin control system can continuously correct and optimize the control strategy, ensuring that the physical entity is always in an optimal or near-optimal operating state.
[0069] In some embodiments, the communication network employs industrial Internet of Things (IoT) technology, including 5G communication, industrial Ethernet, or wireless sensor networks, to enable high-speed data transmission between the digital twin control system and the physical production equipment. The actuator includes at least one of a programmable logic controller (PLC), a distributed control system (DCS), and a robot controller, to receive instructions from the digital twin control system and act directly on the physical production process. Of course, the actuator may also include other industrial automation devices.
[0070] The method provided in this application combines the high-fidelity simulation capability of digital twin models with the optimization capability of intelligent optimization algorithms to achieve a closed-loop optimization control throughout the entire process, from offline parameter optimization and virtual simulation verification to the introduction of predictive control and real-time online optimization.
[0071] In a specific application scenario, this embodiment addresses the temperature and pressure control of a process industry production device (such as a continuous reactor system). It applies the intelligent control method based on digital twins provided in this application embodiment to achieve simultaneous improvement in product quality and energy efficiency.
[0072] Key Parameter Selection and Digital Twin Model Construction: First, the optimization objective of this reactor process is determined to be to increase product yield and reduce energy consumption while meeting safety and process constraints. Therefore, the key parameters to be optimized are the reactor's temperature setpoint and feed flow rate (these two parameters directly affect the reaction rate and energy consumption). Constraints include ensuring the reaction temperature does not exceed the safety threshold, maintaining the pressure within the equipment's pressure tolerance range, and meeting product quality standards. Next, a digital twin model of the reactor is established using a mechanistic model combined with data identification. Based on chemical reaction kinetics and heat transfer mechanisms, a differential equation model of the dynamic changes in reactor temperature and pressure is established, while also considering the effects of stirring, heat exchange, and other factors. After the basic framework of the model is established, the model parameters are corrected using historical process data. For example, the least squares method is used to identify unknown parameters such as the heat transfer coefficient and reaction rate constant, so that the model output matches the actual historical process records, ultimately resulting in a digital twin model. This digital twin model includes a main control loop (temperature control, where the main controller drives the heater power and adjusts the temperature inside the vessel) and a secondary control loop (pressure control, where the secondary controller adjusts the opening of the exhaust valve and controls the vessel pressure), and includes a feedforward control module to adjust the feeding rate in advance based on fluctuations in the raw material composition. The transfer function and response characteristics of the digital twin model are highly consistent with the actual equipment within the normal operating range.
[0073] Optimization Algorithm: Based on the digital twin model, the objective function is defined as a weighted comprehensive score of product yield and unit energy consumption, with the goal of maximizing this score. Temperature setpoint and feed flow rate are used as decision variables, with constraints including temperature, pressure, and safety indicators that do not violate limits. Since this optimization problem is nonlinear and involves multiple objective trade-offs, a genetic algorithm (GA) is used as the primary optimization algorithm. The digital twin model is embedded into the GA's fitness evaluation function: GA generates a set of candidate solutions for temperature and flow rate parameters, and the corresponding yield and energy consumption performance are obtained through simulation using the digital twin model to calculate the fitness value. The algorithm iteratively evolves the population, eventually converging to obtain a set of approximately optimal temperature and feed settings. To improve optimization efficiency, this embodiment also incorporates the global search advantage of the particle swarm optimization (PSO) algorithm. Specifically, when GA stagnates at local convergence, PSO is introduced to mutate a portion of the population, expanding the search range and avoiding getting trapped in local optima. After a certain number of iterations, the optimal solution (the optimized key parameters) is as follows: the reactor temperature setting is increased by 2℃ (within the safe range), the feed flow rate is increased by 5%, and the yield is expected to increase by about 3% and the unit energy consumption is reduced by about 4%.
[0074] Simulation Verification and Adjustment: The optimal key parameters (optimized key parameters) were input into the digital twin model for simulation verification. Scenarios were set in the simulation, including raw material composition fluctuations of ±5% and ambient temperature changes of ±3℃, and the response of the control system was observed. Results showed that after adopting the optimized parameters, the reactor temperature could quickly stabilize near the new set value, with the steady-state error approaching zero; the product yield increased to approximately 103% of the original under various disturbances, and energy consumption was reduced. Sensitivity analysis revealed that the feed flow rate had a significant impact on the yield, while temperature had a more pronounced impact on energy consumption. Subsequently, the PID parameters of the temperature controller were further tuned and optimized on the digital twin platform: by scanning different proportional gains and integral times, a set of PID parameters that could achieve no overshoot and fast response under disturbances was selected, improving control quality. Robustness testing showed that even when the reaction rate constant decreased by 10% due to catalyst activity decay, the new control strategy could still control the temperature and pressure within the target range, with the yield decrease not exceeding 2%, demonstrating good robustness. This verifies that the optimized key parameter combination is effective and reliable within the expected range.
[0075] Predictive control was introduced: To further improve control performance, a deep learning predictive model was trained based on the historical operating data of the unit over the past few months for key parameter prediction. A Long Short-Term Memory (LSTM) neural network model was used to predict the temperature and pressure trends of the reactor over the next 5 minutes. The input to the predictive model included time-series data such as current and past temperatures, pressures, feed flow rates, and cooling water flow rates, and the output was the predicted temperature and pressure increments after 5 minutes. After training, the predictive model achieved an average temperature error of ±0.2℃ and a pressure error within ±0.01MPa on the validation data, with accuracy within the predetermined range.
[0076] A predictive model is embedded into a digital twin model to form a digital twin control system. Specifically, the predictive model is deployed in a digital twin simulation environment as part of predictive control, combined with... Figure 2As shown, when the current state of the physical reactor (process operating parameters) is transmitted to the predictive model via sensors, the predictive model calculates that the reactor temperature may rise by 1°C and the pressure may increase by 0.05 MPa within the next 5 minutes (e.g., due to the accumulation of exothermic reaction heat). Based on this, the digital twin control system simulates the future trend in advance in the virtual environment and generates preventative adjustment actions (feedforward control commands) through control strategy calculations: slightly reducing the heater power and opening the exhaust valve in advance to relieve pressure. These feedforward control commands are applied in the digital twin model, and simulations show that they can effectively avoid potential temperature and pressure exceedances after 5 minutes. Through multiple simulation tests, it was found that after introducing predictive control, in the event of sudden increases in raw material concentration or temporary failure of cooling water, the peak reactor temperature was reduced by approximately 1.5°C, pressure fluctuations were reduced by approximately 10%, and the system recovery time was shortened by more than 30% compared to traditional control. This demonstrates that predictive control improves the system's ability to cope with future disturbances and enhances its dynamic performance.
[0077] Actual Deployment and Operation: After ensuring the safety and effectiveness of the digital twin control system's control strategy, it is integrated with the actual reactor control system. The reactor's PLC and DCS systems are connected to the digital twin platform via a field industrial Ethernet connection. Real-time data collected by temperature and pressure sensors is continuously transmitted to the digital twin model, which operates in real-time, comparing and analyzing the real-time data with the results calculated by the digital twin model. When an upward temperature trend is detected, the digital twin control system calculates an adjustment scheme based on optimization and predictive control algorithms and generates optimized control commands, such as reducing heating power and introducing cooling water earlier. These optimized control commands are then sent to the field PLC for execution via the network. Similarly, when pressure is predicted to exceed limits, the optimized control commands generated by the digital twin control system control the premature opening of the exhaust valve. Under the action of these optimized control commands, the physical reactor maintains a stable reaction state. Throughout the process, the data from the digital twin model and the actual physical production unit are updated synchronously at the millisecond level. The interaction between the virtual and reality enters a closed loop; any fluctuations in the physical production process are promptly perceived and responded to by the digital twin model, and virtual optimization decisions are immediately fed back to the physical production process for correction. After a period of operation and observation, the application of the digital twin-based intelligent control method provided in this application increased the product qualification rate of the device from 96% to 99%, reduced energy consumption by approximately 5%, and made the operation process more stable, greatly reducing the frequency of manual intervention. Therefore, it is evident that the digital twin-based intelligent control method provided in this application has achieved significant results in actual production.
[0078] This application also provides an intelligent control system based on digital twins, including: The digital twin simulation module is used to build a digital twin model of the production process; the digital twin model includes a feedforward and feedback composite control structure model, and is used to simulate the dynamic behavior of physical production equipment and its control system. The optimization decision module, connected to the digital twin simulation module, is used to determine the optimization algorithm based on the determined production process optimization objectives, the key parameters to be optimized, and the constraints that the parameters to be optimized must satisfy. The optimization algorithm is then used to solve for the key parameters to be optimized, and the optimized key parameters are obtained. The prediction module, integrated with the digital twin simulation module, is used to build a prediction model. The prediction model is used to predict the changing trend of optimized key parameters after a preset time in the future. By embedding the prediction model into the digital twin model, a digital twin control system is formed, which enables the digital twin model to generate feedforward control commands based on the prediction results of the prediction model. Thus, the digital twin model can simulate control behavior with predictive feedforward correction to control the actuators of the physical production device. The data acquisition and communication module is used to acquire real-time data of the physical production device through sensors, establish a data communication network between the physical space and the digital twin simulation module, input the real-time data to the digital twin simulation module through the data communication network, and determine the optimized control command through the digital twin model and optimization algorithm. The execution module, including one or more physical production devices, is used to execute the optimization control instructions issued by the digital twin simulation module to achieve real-time closed-loop optimization control of the physical production process.
[0079] Specifically, the digital twin simulation module is used to build and run a digital twin model of the production process, performing real-time simulation and deduction of the behavior of the physical system. This digital twin simulation module includes a mechanistic model unit and a data-driven model unit, ensuring high accuracy of the simulation model in representing the actual process and reproducing the dynamic characteristics of the main control loop, auxiliary control loop, feedforward control channel, and various key equipment.
[0080] Specifically, the optimization decision module includes an optimization algorithm unit, which, based on the set optimization objective, the key parameters to be optimized, and the constraints that the parameters to be optimized must satisfy, calls the corresponding optimization algorithm to perform optimization calculations on the key parameters to be optimized, and obtains the optimal control parameters (optimized key parameters) or control commands. Multiple optimization algorithms can be switched or mixed, including linear and nonlinear programming algorithms and intelligent optimization algorithm libraries, to adapt to problems of different natures.
[0081] Specifically, the data acquisition and communication module includes industrial sensors, data acquisition devices, and communication interfaces. It is used to collect real-time data such as process parameters and status data from the physical production unit and transmit this data to the digital twin simulation module. Simultaneously, it sends optimized control commands determined by the digital twin model and optimization algorithms to the field actuators via wired or wireless networks. This data acquisition and communication module ensures the timeliness and reliability of data interaction between the virtual and physical spaces.
[0082] Specifically, the prediction module contains a pre-trained machine learning or deep learning model for predicting key process variables. It calculates the process state at a specified future time based on real-time collected data. This prediction module works in conjunction with the optimization decision-making module, enabling the digital twin control system to possess forward-looking control capabilities.
[0083] Specifically, the control execution module consists of execution units such as field programmable controllers (PLCs), valves, motor drivers, and robot controllers. It receives optimized control commands from the digital twin control system and directly acts on the physical production process (such as adjusting valve opening, changing feeding speed, and modifying set temperature) to achieve automatic control of the production equipment operation.
[0084] The modules of the aforementioned digital twin-based intelligent control system are assembled into a distributed control system architecture via industrial Ethernet, 5G, and other networks. The digital twin simulation module, optimization decision-making module, and prediction module are typically deployed on industrial cloud or edge computing servers, working collaboratively with the field control execution module. This system configuration supports global perception, intelligent analysis, and decision optimization of the physical space from the information space, upgrading traditional post-production process adjustments to real-time predictive control, and significantly improving the intelligence level of the production process.
[0085] This application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of any of the methods described in the foregoing embodiments.
[0086] The computer device provided in this application has the same inventive concept and the same beneficial effects as the previous embodiments. For the contents of the computer device not shown in detail, please refer to the previous embodiments, and will not be repeated here.
[0087] This application also provides a computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of any of the methods described in the foregoing embodiments.
[0088] The computer device provided in this application has the same inventive concept and the same beneficial effects as the previous embodiments. For the contents of the computer device not shown in detail, please refer to the previous embodiments, and will not be repeated here.
[0089] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.
Claims
1. A digital twin-based intelligent control method, characterized in that, include: Determine the objectives that need to be optimized in the production process, the key parameters to be optimized, and the constraints that the parameters to be optimized must meet; Construct a digital twin model of the production process; wherein the digital twin model includes a feedforward and feedback composite control structure model, and the digital twin model is used to simulate the dynamic behavior of the physical production equipment and its control system; Based on the objective, the key parameters to be optimized, and the constraints, an optimization algorithm is determined, and the key parameters to be optimized are solved using the optimization algorithm to obtain the optimized key parameters. The optimized key parameters are input into the digital twin model for simulation verification; Construct a predictive model; the predictive model is used to predict the changing trend of optimized key parameters after a preset future time. The prediction model is embedded into the digital twin model to form a digital twin control system. The digital twin model generates feedforward control commands based on the prediction results of the prediction model, so that the digital twin model can simulate control behavior with prediction feedforward correction to control the actuators of the physical production device. The tested digital twin control system is deployed into the actual production control system. Real-time data from the physical production equipment, obtained through sensors and communication networks, is input into the digital twin model. The digital twin model and the optimization algorithm determine the optimized control commands, which are then sent to the actuators of the physical production equipment. This achieves real-time closed-loop optimized control of the physical production process by the digital twin control system.
2. The method according to claim 1, characterized in that, The construction process of the digital twin model includes: The digital twin model is established using a hybrid modeling method that combines mechanistic modeling and digital-driven modeling. The digital twin model includes a cascade control system model of a main control loop and a secondary control loop, as well as a feedforward control channel for known disturbances. The cascade control system model and the feedforward control channel constitute the feedforward and feedback composite control structure model.
3. The method according to claim 1, characterized in that, The optimization algorithms include one or more of the following: linear programming, nonlinear programming, dynamic programming or integer programming, genetic algorithm, particle swarm optimization, simulated annealing, ant colony optimization, neural network optimization, and hybrid optimization algorithms that combine traditional mathematical programming with the aforementioned algorithms.
4. The method according to claim 1, characterized in that, The optimized key parameters are input into the digital twin model for simulation verification, including: The optimized key parameters are input into the digital twin model, and the digital twin model is tested in a virtual environment by applying different input signals and disturbances. The dynamic response and performance indicators of the system output are observed to verify the optimized key parameters through simulation. The simulation verification includes: parameter sensitivity analysis, model control parameter tuning, and robustness analysis of the optimized key parameters.
5. The method according to claim 1, characterized in that, Building a predictive model includes: Based on historical operational data and process mechanisms, a prediction model is constructed using machine learning or deep learning algorithms; wherein, the prediction model includes one or more of time series prediction models, regression prediction models, and neural network prediction models.
6. The method according to claim 1, characterized in that, The digital twin model generates feedforward control commands based on the prediction results of the prediction model, enabling the digital twin model to simulate control behavior with predictive feedforward correction, including: The prediction results of the prediction model are input into the prediction controller in the digital twin model; Based on the prediction results, the predictive controller calculates control actions in advance according to the deviation of future production operation status and generates feedforward control instructions, so that the digital twin model can simulate control behavior with predictive feedforward correction based on the feedforward control instructions.
7. The method according to claim 1, characterized in that, The communication network adopts industrial Internet of Things (IoT) technology, including 5G communication, industrial Ethernet, or wireless sensor networks, to achieve high-speed data transmission between the digital twin control system and the physical production equipment. The actuator includes at least one of a programmable logic controller, a distributed control system, and a robot controller, to receive instructions from the digital twin control system and act directly on the physical production process.
8. An intelligent control system based on digital twins, characterized in that, include: A digital twin simulation module is used to construct a digital twin model of a production process; wherein, the digital twin model includes a feedforward and feedback composite control structure model, and the digital twin model is used to simulate the dynamic behavior of physical production equipment and its control system; The optimization decision module, connected to the digital twin simulation module, is used to determine the optimization algorithm based on the determined production process optimization objectives, the key parameters to be optimized, and the constraints that the parameters to be optimized must satisfy. The optimization algorithm is then used to solve for the key parameters to be optimized to obtain the optimized key parameters. The prediction module, integrated with the digital twin simulation module, is used to construct a prediction model. The prediction model is used to predict the changing trend of optimized key parameters after a preset time in the future. By embedding the prediction model into the digital twin model, a digital twin control system is formed, enabling the digital twin model to generate feedforward control commands based on the prediction results of the prediction model. Thus, the digital twin model can simulate control behavior with predictive feedforward correction to control the actuators of the physical production device. The data acquisition and communication module is used to acquire real-time data of the physical production device through sensors, establish a data communication network between the physical space and the digital twin simulation module, input the real-time data to the digital twin simulation module through the data communication network, and determine the optimized control command through the digital twin model and the optimization algorithm. The execution module includes one or more physical production devices, used to execute the optimization control commands issued by the digital twin simulation module, thereby realizing real-time closed-loop optimization control of the physical production process.
9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.