Automatic intelligent management and control method of factory and related device

Abstract

The invention discloses an automatic intelligent management and control method of a factory, an automatic intelligent management and control device of the factory, automatic intelligent management and control equipment and a computer readable storage medium, and the method comprises the steps: collecting multi-source heterogeneous data from a plurality of information systems of the factory in real time, and carrying out the data fusion of the multi-source heterogeneous data, obtaining a real-time working condition data set for describing the current operation state of the factory; based on the real-time working condition data set, performing state synchronization on a pre-constructed digital twinborn model to obtain a digital twinborn model instance for reproducing the state of the factory at the current moment; deducing a plurality of control operations in a future preset time window based on the digital twinborn model instance to obtain a deduction result, and performing optimization calculation on the deduction result based on a preset multi-objective optimization function to obtain an optimal control sequence; and analyzing the optimal control sequence into an equipment control instruction set for the bottom layer control system, and issuing the equipment control instruction set to the bottom layer control system to execute control operation.

Description

Technical Field

[0001] This invention belongs to the field of industrial control technology, and more specifically, relates to an automated intelligent management and control method for a factory, an automated intelligent management and control device for a factory, an automated intelligent management and control equipment, and a computer-readable storage medium. Background Technology

[0002] Chemical production processes are typical complex industrial systems characterized by multiple variables, strong coupling, and large time delays. Their automated control systems are the core of ensuring the safe, stable, and efficient operation of the equipment. Currently, modern large-scale chemical plants generally adopt automation schemes based on distributed control systems, and their control cores mostly rely on the classic PID (proportional-integral-derivative) feedback control algorithm.

[0003] However, existing technologies have significant drawbacks. First, PID-based feedback control inherently suffers from hysteresis. It can only make "post-event adjustments" after detecting deviations in process parameters from the setpoint. For chemical reaction processes with large inertia and long time delays, this passive response mode is prone to repeated overshooting and oscillations in key parameters, affecting not only the stability of the final product quality but also posing a challenge to the smooth operation of the plant. Second, optimization in existing technologies is usually local and static. Engineers often perform isolated parameter tuning for the control loops of individual equipment or sections, neglecting the strong coupling between process units, making it difficult to achieve overall optimization of the entire process. At the same time, these fixed control parameters cannot adaptively cope with slow changes such as fluctuations in raw material composition and catalyst activity decay, causing the system's long-term operating efficiency to gradually deviate from the optimal state.

[0004] Therefore, how to provide an intelligent control method that can proactively predict the future state of the process and achieve collaborative optimization of the entire process, so as to overcome the defects of existing technologies caused by control lag and local optimization, and thus comprehensively improve the stability, operating efficiency and safety of chemical production processes, is a key issue of concern to those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide an automated intelligent control method, an automated intelligent control device, an automated intelligent control equipment, and a computer-readable storage medium for factories, which enables proactive prediction of the future state of the process and intelligent control method for full-process collaborative optimization, in order to overcome the defects of existing technologies caused by control lag and local optimization, thereby comprehensively improving the stability, operating efficiency and safety of chemical production processes.

[0006] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an automated intelligent control method for factories, comprising: Multi-source heterogeneous data is collected in real time from multiple information systems in the factory, and the multi-source heterogeneous data is fused to obtain a real-time operating condition dataset that describes the current operating status of the factory. Based on the real-time operating condition dataset, the state of the pre-built digital twin model is synchronized to obtain a digital twin model instance that reproduces the current state of the factory; wherein, the digital twin model is a twin model constructed by combining the chemical process mechanism model and historical data model of the factory; Based on the digital twin model example, various control operations within a future preset time window are deduced to obtain the deduction results, and the deduction results are optimized based on a preset multi-objective optimization function to obtain the optimal control sequence; The optimal control sequence is parsed into a set of device control instructions for the underlying control system, and the set of device control instructions is sent to the underlying control system to execute control operations; The residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence is calculated according to a preset period. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control periods.

[0007] Optionally, the multi-source heterogeneous data includes: Real-time process data from the distributed control system, programmable logic controller, or safety instrumented system of the chemical plant; static equipment data from the equipment management database; and material attribute data from the laboratory information management system.

[0008] Optionally, the digital twin model is a model constructed by combining a chemical reaction kinetics mechanism model, a fluid dynamics mechanism model, and a neural network data model trained with historical data.

[0009] Optionally, the multi-objective optimization function includes at least two of the following indicators: output, quality, energy consumption, and safety margin.

[0010] Optionally, when the residuals meet preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model, including: When the residual exceeds the anomaly detection threshold within a preset time, the residual pattern is analyzed for fault based on a preset fault knowledge graph or classification model to obtain a fault diagnosis report, which is then pushed to the operation and maintenance personnel. Based on the historical real-time operating condition dataset and the fault diagnosis report, the data-driven part of the digital twin model is incrementally trained or retrained to obtain the updated and optimized digital twin model.

[0011] Optionally, based on the digital twin model instance, various control operations within a future preset time window are extrapolated to obtain extrapolation results, and the extrapolation results are optimized based on a preset multi-objective optimization function to obtain the optimal control sequence, including: Within the preset set of control variable constraints, a set of candidate control sequences is generated; Each candidate control sequence is input into the digital twin model instance for forward simulation to obtain its corresponding future state prediction trajectory. The multi-objective optimization function is used to evaluate and score each future state prediction trajectory to determine the candidate control sequence corresponding to the trajectory with the highest evaluation score as the optimal control sequence.

[0012] This application also provides an automated intelligent control device for a factory, comprising: The real-time data acquisition module is used to collect multi-source heterogeneous data from multiple information systems in the factory in real time, and to fuse the multi-source heterogeneous data to obtain a real-time operating condition dataset that describes the current operating status of the factory. The twin model synchronization module is used to synchronize the state of a pre-built digital twin model based on the real-time operating condition dataset to obtain a digital twin model instance that reproduces the current state of the factory; wherein, the digital twin model is a twin model constructed by combining the chemical process mechanism model and historical data model of the factory. The control operation deduction module is used to deduce multiple control operations within a future preset time window based on the digital twin model instance to obtain deduction results, and to perform optimization calculations on the deduction results based on a preset multi-objective optimization function to obtain the optimal control sequence. The operation instruction issuing module is used to parse the optimal control sequence into a set of device control instructions for the underlying control system, and issue the set of device control instructions to the underlying control system to execute control operations; The digital twin model correction module is used to calculate the residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence according to a preset period. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control periods.

[0013] Optionally, the multi-source heterogeneous data includes at least one of the following: real-time process data from the distributed control system, programmable logic controller, or safety instrumented system of the chemical plant; equipment static data from the equipment management database; and material attribute data from the laboratory information management system.

[0014] This application also provides an automated intelligent control device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the steps of the automated intelligent control method described above.

[0015] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the automated intelligent control method described above.

[0016] This application provides an automated intelligent management and control method for a factory, comprising: real-time collection of multi-source heterogeneous data from multiple information systems in the factory, and data fusion of the multi-source heterogeneous data to obtain a real-time operating condition dataset describing the current operating state of the factory; based on the real-time operating condition dataset, state synchronization of a pre-constructed digital twin model to obtain a digital twin model instance that reproduces the current state of the factory; wherein, the digital twin model is a twin model constructed based on a hybrid model of the factory's chemical process mechanism and historical data; and based on the digital twin model instance, various control measures within a preset future time window are implemented. The operation is simulated to obtain simulation results, and the simulation results are optimized based on a preset multi-objective optimization function to obtain the optimal control sequence. The optimal control sequence is parsed into a set of equipment control instructions for the underlying control system, and the set of equipment control instructions is sent to the underlying control system to execute control operations. The residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence is calculated according to a preset period. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control cycles.

[0017] It has the following beneficial effects: By synchronizing the state of a digital twin model that blends mechanism and data, a high-fidelity virtual factory instance can be obtained. Using this instance for forward-looking extrapolation and multi-objective optimization, the resulting optimal control sequence can proactively guide the production process towards a globally optimal state, fundamentally overcoming the inherent lag and optimization locality defects of traditional feedback control. Finally, through the model's adaptive correction mechanism, this method can continuously learn and adapt to the dynamic changes in the factory, ensuring the long-term accuracy of the digital twin model and significantly enhancing the robustness and long-term operational benefits of the entire intelligent control system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an automated intelligent control method for a factory, provided as an embodiment of this application; Figure 2 A schematic diagram of the structure of an automated intelligent control device for a factory provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the automated intelligent control device provided in the embodiments of this application. Detailed Implementation

[0020] The purpose of this application is to provide an automated intelligent control method, an automated intelligent control device, an automated intelligent control equipment, and a computer-readable storage medium for factories, which enables proactive prediction of the future state of the process and intelligent control method for full-process collaborative optimization, in order to overcome the defects of existing technologies caused by control lag and local optimization, thereby comprehensively improving the stability, operating efficiency and safety of chemical production processes.

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] The following example illustrates an automated intelligent control method for a factory provided in this application.

[0023] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating an automated intelligent control method for a factory, as provided in an embodiment of this application.

[0024] In this embodiment, the method may include: S101 collects multi-source heterogeneous data from multiple information systems in the factory in real time, and performs data fusion on the multi-source heterogeneous data to obtain a real-time operating condition dataset that describes the current operating status of the factory. This step aims to acquire data from the information system. A chemical plant is an extremely complex system, and its complete state cannot be described by a single data source. Dynamic parameters of the production process, static properties of equipment, chemical properties of materials, and production plans are scattered across different information silos. Therefore, it is necessary to collect this multi-source heterogeneous data, varying in structure, format, and frequency, from multiple sources and transform it into a unified, clean, and usable dataset through data fusion technology. This is the prerequisite for all subsequent digital analysis, simulation, and optimization, aiming to ensure that the digital twin model accurately reflects the true state of the physical world.

[0025] The multi-source heterogeneous data in this step includes at least one of the following: real-time process data from the distributed control system, programmable logic controller, or safety instrumented system of the chemical plant; equipment static data from the equipment management database; and material attribute data from the laboratory information management system.

[0026] The implementation process of this step can include three stages: data acquisition, data preprocessing, and data fusion.

[0027] During the data acquisition phase, the system connects to the core information systems within the factory by deploying multiple data interfaces. As a preferred implementation, the data acquisition server directly connects to the distributed control system and programmable logic controller (PLC) via an open platform communication unified architecture or standard industrial protocols such as Modbus TCP / IP. This allows for real-time acquisition of key process parameters, such as the internal temperature and pressure of reactor R-101, the top temperature of distillation column T-201, and the flow rate of feed pump P-301, among other dynamic process data, at millisecond or second-level frequencies.

[0028] Through the database connector, access the laboratory information management system to periodically retrieve offline analysis data of raw materials and intermediate products, such as material property data like the concentration of key components and impurity content.

[0029] Similarly, by accessing the database and connecting to the equipment management system or computerized maintenance management system, static information about the equipment can be obtained, such as the design heat exchange area and material of heat exchanger E-401, as well as its most recent maintenance records and health status assessment.

[0030] Connect to the manufacturing execution system to obtain production management information such as the current production batch number and planned output.

[0031] In the data preprocessing stage, the collected raw data needs to undergo a series of cleaning and normalization operations to ensure data quality. This process may include: Since there may be slight differences in the time of data generation from different systems, the system uses the Network Time Protocol (NTP) to synchronize the clocks of all data sources. After the data is aggregated, the data points are interpolated or resampled using a unified time base to ensure that all data are strictly aligned in the time dimension. To address potential outliers from sensors (such as spikes caused by signal interference) or packet loss during data transmission, statistical methods such as the 3-sigma rule and box plot analysis are used to identify and remove outliers. Missing values ​​are then filled using methods such as linear interpolation, mean imputation, or model prediction based on historical data. For sensor signals with high-frequency fluctuations, algorithms such as moving average filtering and Kalman filtering can be used for smoothing to extract the true trend of process changes.

[0032] In the data fusion phase, the preprocessed data from various sources are integrated into a real-time operating condition dataset. Specifically, the system converts all data into a standardized data structure, such as a JSON object or Pandas DataFrame containing timestamps and multiple key-value pairs. In this structure, each data point has a unique identifier, physical unit, value, and timestamp, thus forming a snapshot that provides a panoramic, high-fidelity description of the entire factory's state at a specific moment.

[0033] Furthermore, it is evident that this step, by executing it, can construct a comprehensive, accurate, and unified data foundation, which fully depicts the state of all elements of the chemical plant—human, machine, material, method, and environment—at any given moment. This not only provides high-quality input for the subsequent state synchronization of the digital twin model, ensuring the authenticity of the model's initial state, but also fundamentally solves the problem of model inaccuracy caused by a single data source or incomplete information, providing the most basic and crucial guarantee for the effectiveness of the entire intelligent control method.

[0034] S102, based on the real-time operating condition dataset, the state of the pre-built digital twin model is synchronized to obtain a digital twin model instance that reproduces the current state of the factory; wherein, the digital twin model is a twin model constructed by combining the chemical process mechanism model and historical data model of the factory; Building upon S101, this step aims to synchronize the state of a pre-built digital twin model based on a real-time operating condition dataset, thereby obtaining a digital twin model instance that reproduces the current state of the factory. The digital twin model is constructed by combining the factory's chemical process mechanism model and historical data model.

[0035] The core of this step lies in activating a static, general digital twin model into a dynamic digital twin instance that is fully synchronized with the physical entity. The pre-built digital twin model is a mathematical template containing the inherent laws and operational experience of the chemical process, but it does not reflect the real-time state of the plant. The purpose of state synchronization is to use the real-time operating condition dataset obtained in the previous step to assign values ​​and calibrate this template, ensuring that all its internal state variables—whether directly measurable or not—are precisely consistent with the state of the physical plant at the current moment. Furthermore, the hybrid model construction method in this step, combining mechanism and data, aims to use the mechanistic model to ensure the model's generalization ability and interpretability, while using the data model to compensate for the simplification errors of the mechanistic model, thereby achieving higher-precision reproduction.

[0036] Furthermore, the digital twin model in this step is a hybrid model constructed from chemical reaction kinetics mechanism model, fluid dynamics mechanism model, and neural network data model trained with historical data.

[0037] First, this step is implemented based on the interaction between real-time operating condition datasets and digital twin models.

[0038] The pre-built digital twin model is a composite structure. The process of building the twin model may include: The mechanistic models are partially constructed based on the fundamental physicochemical laws of chemical processes. For example, for a continuous stirred tank reactor, the mechanistic model would include a set of differential equations based on the conservation of mass and energy, as well as the Arrhenius equation describing the relationship between chemical reaction rates and temperature and concentration. For a distillation column, the mechanistic model would include vapor-liquid balance equations and mass and energy balance equations for each tray. These models form the "skeleton" of the digital twin, ensuring that its behavior conforms to physical laws.

[0039] The historical data model is used to compensate for the shortcomings of the mechanistic model. Mechanistic models are often based on ideal assumptions, while actual operating conditions are complex and variable. Therefore, a data-driven compensation model can be trained using historical operating data, such as a deep learning model like a Long Short-Term Memory network or a gated recurrent unit. The input to this data model is the input and output of the mechanistic model, and the goal is to predict the residual between the mechanistic model output and the real factory data. During operation, the final output of the digital twin model is the "residual between the mechanistic model output and the data model prediction," thus forming a more accurate hybrid model.

[0040] Furthermore, the state synchronization process may include: Step 1: Use the measurable parameters from the "real-time operating condition dataset" obtained in the previous step as inputs or boundary conditions for the digital twin model. For example, use the measured feed flow rate, feed temperature, and component concentration from the dataset as inputs to the CSTR mechanism model.

[0041] Step 2: Under given boundary conditions, solve the driving mechanism model. For dynamic models, this means starting from a known past state and calculating up to the current time through numerical integration to obtain the values ​​of all state variables within the model. These state variables not only include measurable temperature and pressure, but more importantly, they contain a large number of key internal variables that cannot be directly measured online, such as the catalyst activity distribution at different locations within the reactor, the molecular weight distribution of the polymer, and the vapor-liquid phase composition on the trays.

[0042] Step 3: Input the parameters from the real-time operating condition dataset into the data model section. The data model will output a compensation value for the prediction error of the mechanism model based on the current operating conditions.

[0043] Step 4 combines the state variables calculated by the mechanistic model with the compensation values ​​output by the data model to obtain the final digital twin model instance, which is highly consistent with the current physical plant state. This instance not only matches reality in terms of macroscopic measurable parameters, but also obtains the most reasonable estimate of its internal microscopic and unmeasurable states.

[0044] As can be seen, by performing this step, a static model template can be transformed into a "living" digital mirror that resonates in real time with the physical plant. The value of this digital twin model instance lies in providing a virtual testing platform that is highly consistent with real-world behavior. Subsequent predictions and optimizations will be carried out in this high-fidelity virtual environment, enabling rapid and accurate extrapolation and evaluation of future control strategies without interfering with actual production or incurring any physical risks. This is the core prerequisite for achieving forward-looking predictive control.

[0045] S103, based on a digital twin model example, extrapolates various control operations within a future preset time window to obtain extrapolation results, and performs optimization calculations on the extrapolation results based on a preset multi-objective optimization function to obtain the optimal control sequence; Building upon S102, this step aims to use a digital twin model instance to extrapolate various control operations within a future preset time window, obtain extrapolation results, and perform optimization calculations on the extrapolation results based on a preset multi-objective optimization function to obtain the optimal control sequence.

[0046] This step represents a shift from passive response to proactive prediction. The principle is to use the high-fidelity digital twin model instance created in the previous step as a virtual model to simulate the potential consequences of implementing different control strategies within a future timeframe (i.e., a preset time window, such as the next 30 minutes). Then, a multi-objective optimization function is introduced to evaluate all simulation results, and an optimization algorithm is used to find the operational scheme that maximizes the overall future benefits—the optimal control sequence. This fundamentally solves the lag problem of traditional control methods by anticipating the future and selecting the optimal path, thus enabling proactive guidance of the production process.

[0047] The multi-objective optimization function includes at least two of the following: output, quality, energy consumption, and safety margin.

[0048] Furthermore, this step may include: First, define the operable control variables (such as valve opening, pump speed, heater power, etc.) and their constraints (such as maximum / minimum opening, maximum rate of change).

[0049] Then, candidate control sequences are generated. This can be achieved by the system generating a series of candidate schemes for operating on the control variables within a preset future time window. These sequences can be generated in various ways. A simpler approach is to use a parameterized control strategy; for example, defining the future change in reaction temperature as a slope and then sampling the slope. A more advanced approach is to use population initialization with intelligent optimization algorithms to generate a diverse and complex set of control sequences. Each sequence represents a complete future operational plan, such as: "Increase the opening of valve A from 50% to 55% at minute 5, and decrease the power of heater B by 3% at minute 15."

[0050] Next, each candidate control sequence is used as input and loaded into the "digital twin model instance". Starting from the current state, the model performs rapid forward simulation calculations according to the operations set by the control sequence, step by step deducing the dynamic change trajectory of all key process parameters of the factory (such as product purity, energy consumption, and reaction conversion rate) over the next 30 minutes. The set of this series of trajectories is the deduction result.

[0051] Finally, in the optimization phase, the system evaluates and filters the simulation results.

[0052] The optimization process may include: First, based on production targets, a comprehensive evaluation function J is predefined. This function is a weighted sum or more complex form that includes multiple, even conflicting, targets. For example: J = w1*(output) + w2*(product purity) - w3*(energy consumption per unit product) - w4*(safety risk index). Here, w1, w2, w3, and w4 are weighting coefficients, reflecting the priority of different targets. The safety risk index can be calculated based on the degree to which key parameters (such as pressure and temperature) deviate from safety boundaries.

[0053] Then, for each candidate control sequence, the system substitutes the predicted trajectory into the multi-objective optimization function J for calculation, obtaining a comprehensive score. The optimization algorithm iterates continuously, eliminating sequences with low scores, retaining and evolving sequences with high scores, and continuously generating new and better candidate control sequences for a new round of deduction and evaluation. This iterative process continues until a control sequence that maximizes (or minimizes) the value of the optimization function J is found.

[0054] Ultimately, the control sequence with the highest score found when the algorithm converges is the optimal control sequence. It represents a set of operating instructions that, under the current operating conditions, comprehensively considers multiple aspects such as output, quality, energy consumption, and safety, and can guide the factory to its best state in the future.

[0055] As can be seen, this method can elevate control decision-making from a feedback loop based on history and the present to a future-based feedforward-feedback composite loop. It can select a path to the global optimum from numerous possibilities. This not only effectively avoids overshoot and oscillations in traditional control, greatly improving process stability and product quality consistency, but also significantly enhances the overall economic benefits of the plant through synergistic optimization of targets such as energy consumption and output.

[0056] S104, the optimal control sequence is parsed into a set of device control instructions for the underlying control system, and the set of device control instructions is sent to the underlying control system to execute control operations; Building upon S103, this step aims to parse the optimal control sequence into a set of device control instructions for the underlying control system, and then send the set of device control instructions to the underlying control system to execute control operations.

[0057] Furthermore, the optimal control sequence is typically a time-series data structure, such as an array whose elements are tuples of time points, control variable labels, and setpoints. The parsing module iterates through this sequence. For example, the sequence might contain an entry (t=300s, TIC-101.SP, 150.5), where TIC-101.SP is the setpoint label for the reactor temperature controller.

[0058] The parsing module queries a pre-defined device address mapping table. This table establishes a correspondence between high-level tags and specific addresses in the underlying control system. Subsequently, the parsing module converts the high-level commands into a format conforming to the underlying system's communication protocol. For example, the aforementioned command would be converted into an OPC UA write service request, including: service type, node ID, value to be written, data type, etc. For commands with continuous changes over a time period, the parsing module breaks them down into a series of step setpoint commands issued consecutively within a short period.

[0059] After the above processing, the originally abstract optimal control sequence is transformed into one or more structured device control instruction sets containing explicit target addresses and values. This instruction set can be directly read and executed by the underlying system.

[0060] During the instruction issuance phase, the system sends the set of equipment control instructions to the target device through the industrial network.

[0061] The instruction generation server establishes a communication connection with the underlying devices via industrial Ethernet using protocols such as OPC UA and Modbus TCP / IP.

[0062] Then, the server sends out the instructions in the instruction set one by one or in batches. After receiving the new setpoint, the underlying control system immediately starts working on its internal control loops and other basic control functions, adjusting the corresponding actuators to drive the process parameters toward the new setpoint. The upper-level intelligent management and control system continuously monitors whether the actual value has successfully followed the setpoint through the data acquisition link in step 1, forming a higher-level monitoring closed loop.

[0063] S105: Calculate the residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence according to the preset cycle; when the residual meets the preset correction conditions, adaptively correct the digital twin model to obtain an updated and optimized digital twin model for state synchronization in subsequent control cycles.

[0064] Based on S104, this step aims to calculate the residual between the actual operating data of the plant and the predicted state trajectory generated based on the optimal control sequence according to a preset cycle; when the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control cycles.

[0065] This step introduces learning and evolution capabilities into the entire intelligent control system, with its core principle being continuous improvement based on feedback. No model can achieve 100% accuracy, and the characteristics of a physical plant change slowly over time (e.g., equipment wear, pipe scaling, catalyst activity degradation). This step quantifies the model's accuracy by periodically comparing the differences between model predictions and plant reality. Once this difference accumulates to a certain level, it triggers the model's adaptive correction mechanism. This ensures that the digital twin model does not become outdated over time, thereby guaranteeing that predictions and optimization decisions based on the model remain accurate and effective.

[0066] Furthermore, this step can be divided into three stages: residual calculation, condition judgment, and model correction.

[0067] In the residual calculation stage.

[0068] First, acquire the data: one is the real-time operating condition dataset from step S101, which represents the actual operating data of the factory in the past cycle (e.g., the past 24 hours); the other is the predicted state trajectory from the deduction stage in step S103, which corresponds to the optimal control sequence actually executed.

[0069] The system aligns the two time series data sets and calculates the difference between the actual and predicted values ​​of each key parameter at each time point. These differences constitute the residual sequence. The statistical properties of the residuals, such as root mean square error and mean absolute error, can be further calculated as metrics to measure the overall performance of the model.

[0070] In the conditional judgment phase, the system will decide whether to trigger the correction procedure based on preset correction conditions. These conditions may include: Threshold condition: For example, if the RMSE of key parameter A exceeds the preset threshold Y (e.g., exceeds the normal fluctuation range by 2%) in the past 24 hours, then the condition is met.

[0071] Trend condition: For example, if the model’s prediction error on the key parameter B shows a monotonically increasing trend for 7 consecutive days, this indicates that the deviation of the model from reality is systematic rather than random noise, and the correction condition is met.

[0072] Event triggering conditions: For example, manual correction is triggered after a major overhaul of the plant or replacement of critical equipment (such as catalyst replacement).

[0073] In the model adaptive correction phase, once the correction conditions are met, the system will initiate the update process for the digital twin model. Since the mechanistic model portion is usually fixed, the focus of the correction is on the historical data model portion of the hybrid model. The correction process may include: The system uses real-time operating condition datasets accumulated over a recent period (e.g., the past month) as new training samples. This data contains the input-output relationships of the factory under the latest characteristics.

[0074] For the neural network used as the compensation module, the system performs incremental learning. This means that based on the original model parameters, it uses new data to train for a few cycles to fine-tune the model to adapt to new operating conditions. Alternatively, when the model has a large deviation, it combines new data with some old data to perform a complete retraining of the data model.

[0075] The newly trained model will first be tested on an independent validation dataset to ensure that its performance is significantly improved compared to the old model. After successful validation, the updated and optimized digital twin model will be deployed into the system and will replace the old model for state synchronization in step S102 at the beginning of the next control cycle.

[0076] As can be seen, this step ensures that the method can continuously provide accurate forecasts and optimal control for the factory over a period of several years or even longer, maintaining its production operations in the best possible state.

[0077] In summary, this embodiment achieves a high-fidelity virtual factory instance by synchronizing the state of a digital twin model that integrates mechanism and data. Using this instance for forward-looking extrapolation and multi-objective optimization, the resulting optimal control sequence proactively guides the production process towards a globally optimal state, fundamentally overcoming the inherent lag and optimization locality defects of traditional feedback control. Finally, through the model's adaptive correction mechanism, this method can continuously learn and adapt to the dynamic changes in the factory, ensuring the long-term accuracy of the digital twin model and significantly enhancing the robustness and long-term operational benefits of the entire intelligent control system.

[0078] The following describes an automated intelligent control device for a factory provided in an embodiment of this application. The automated intelligent control device and the automated intelligent control method for a factory described below can be referred to each other.

[0079] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of an automated intelligent control device for a factory, provided as an embodiment of this application.

[0080] In this embodiment, the device may include: The real-time data acquisition module 100 is used to collect multi-source heterogeneous data from multiple information systems in the factory in real time, and to fuse the multi-source heterogeneous data to obtain a real-time operating condition dataset that describes the current operating status of the factory. The twin model synchronization module 200 is used to synchronize the state of a pre-built digital twin model based on a real-time operating condition dataset, so as to obtain a digital twin model instance that reproduces the current state of the factory. The digital twin model is a twin model constructed by combining the chemical process mechanism model and historical data model of the factory. The control operation simulation module 300 is used to simulate multiple control operations within a preset time window based on a digital twin model instance to obtain simulation results, and to perform optimization calculations on the simulation results based on a preset multi-objective optimization function to obtain the optimal control sequence. The operation instruction issuing module 400 is used to parse the optimal control sequence into a set of equipment control instructions for the underlying control system, and issue the set of equipment control instructions to the underlying control system to execute control operations; The twin model correction module 500 is used to calculate the residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence according to a preset cycle. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control cycles.

[0081] This application also provides automated intelligent control equipment; please refer to it. Figure 3 , Figure 3 This is a schematic diagram of the structure of the automated intelligent control device provided in the embodiments of this application. The automated intelligent control device may include: Memory, used to store computer programs; A processor is used to execute computer programs to implement the steps of any of the above-described automated intelligent control methods for factories.

[0082] like Figure 3 The diagram shows the structural composition of an automated intelligent control device, which may include a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, memory 11, and communication interface 12 all communicate with each other through the communication bus 13.

[0083] In this embodiment, the processor 10 may be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0084] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiment of the abnormal IP identification method.

[0085] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions: Multi-source heterogeneous data is collected in real time from multiple information systems in the factory, and the multi-source heterogeneous data is fused to obtain a real-time operating condition dataset that describes the current operating status of the factory. Based on real-time operating condition datasets, the state of a pre-built digital twin model is synchronized to obtain a digital twin model instance that reproduces the current state of the factory. The digital twin model is constructed by combining the chemical process mechanism model and historical data model of the factory. Based on a digital twin model example, various control operations within a future preset time window are simulated to obtain simulation results. The simulation results are then optimized based on a preset multi-objective optimization function to obtain the optimal control sequence. The optimal control sequence is parsed into a set of device control instructions for the underlying control system, and the set of device control instructions is sent to the underlying control system to execute control operations. The residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence is calculated according to a preset cycle. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control cycles.

[0086] In one possible implementation, the memory 11 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; and the data storage area may store data created during use.

[0087] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0088] Communication interface 12 can be an interface for the communication module, used to connect with other devices or systems.

[0089] Of course, it should be noted that, Figure 3 The structure shown does not constitute a limitation on the automated intelligent control device in the embodiments of this application. In practical applications, the automated intelligent control device may include more than Figure 3 More or fewer components as shown, or combinations of certain components.

[0090] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps of any of the above-described automated intelligent control methods for factories.

[0091] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0094] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0095] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0096] The above provides a detailed description of an automated intelligent control method, device, equipment, and computer-readable storage medium for a factory. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of these embodiments are merely illustrative of the method and its core concepts. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the scope of protection of the claims.

Claims

1. A method for automated intelligent control of a factory, characterized in that, include: Multi-source heterogeneous data is collected in real time from multiple information systems in the factory, and the multi-source heterogeneous data is fused to obtain a real-time operating condition dataset that describes the current operating status of the factory. Based on the real-time operating condition dataset, the state of the pre-built digital twin model is synchronized to obtain a digital twin model instance that reproduces the current state of the factory; wherein, the digital twin model is a twin model constructed by combining the chemical process mechanism model and historical data model of the factory; Based on the digital twin model example, various control operations within a future preset time window are deduced to obtain the deduction results, and the deduction results are optimized based on a preset multi-objective optimization function to obtain the optimal control sequence; The optimal control sequence is parsed into a set of device control instructions for the underlying control system, and the set of device control instructions is sent to the underlying control system to execute control operations; The residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence is calculated according to a preset period. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control periods.

2. The automated intelligent control method according to claim 1, characterized in that, The multi-source heterogeneous data includes: Real-time process data from the distributed control system, programmable logic controller, or safety instrumented system of the chemical plant; static equipment data from the equipment management database; and material attribute data from the laboratory information management system.

3. The automated intelligent control method according to claim 2, characterized in that, The digital twin model is a hybrid model constructed from chemical reaction kinetics mechanism model, fluid dynamics mechanism model, and neural network data model trained with historical data.

4. The automated intelligent control method according to claim 3, characterized in that, The multi-objective optimization function includes at least two of the following: output, quality, energy consumption, and safety margin.

5. The automated intelligent control method according to claim 4, characterized in that, When the residuals meet preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model, including: When the residual exceeds the anomaly detection threshold within a preset time, the residual pattern is analyzed for fault based on a preset fault knowledge graph or classification model to obtain a fault diagnosis report, which is then pushed to the operation and maintenance personnel. Based on the historical real-time operating condition dataset and the fault diagnosis report, the data-driven part of the digital twin model is incrementally trained or retrained to obtain the updated and optimized digital twin model.

6. The automated intelligent control method according to claim 5, characterized in that, Based on the aforementioned digital twin model example, various control operations within a future preset time window are extrapolated to obtain extrapolation results. Then, based on a preset multi-objective optimization function, the extrapolation results are optimized to obtain the optimal control sequence, including: Within the preset set of control variable constraints, a set of candidate control sequences is generated; Each candidate control sequence is input into the digital twin model instance for forward simulation to obtain its corresponding future state prediction trajectory. The multi-objective optimization function is used to evaluate and score each future state prediction trajectory to determine the candidate control sequence corresponding to the trajectory with the highest evaluation score as the optimal control sequence.

7. An automated intelligent control device for a factory, characterized in that, include: The real-time data acquisition module is used to collect multi-source heterogeneous data from multiple information systems in the factory in real time, and to fuse the multi-source heterogeneous data to obtain a real-time operating condition dataset that describes the current operating status of the factory. The twin model synchronization module is used to synchronize the state of a pre-built digital twin model based on the real-time operating condition dataset to obtain a digital twin model instance that reproduces the current state of the factory; wherein, the digital twin model is a twin model constructed by combining the chemical process mechanism model and historical data model of the factory. The control operation deduction module is used to deduce multiple control operations within a future preset time window based on the digital twin model instance to obtain deduction results, and to perform optimization calculations on the deduction results based on a preset multi-objective optimization function to obtain the optimal control sequence. The operation instruction issuing module is used to parse the optimal control sequence into a set of device control instructions for the underlying control system, and issue the set of device control instructions to the underlying control system to execute control operations; The digital twin model correction module is used to calculate the residual between the actual operating data of the factory and the predicted state trajectory generated based on the optimal control sequence according to a preset period. When the residual meets the preset correction conditions, the digital twin model is adaptively corrected to obtain an updated and optimized digital twin model for state synchronization in subsequent control periods.

8. The automated intelligent control device according to claim 7, characterized in that, The multi-source heterogeneous data includes at least one of the following: real-time process data from the distributed control system, programmable logic controller, or safety instrumented system of the chemical plant; static equipment data from the equipment management database; and material attribute data from the laboratory information management system.

9. An automated intelligent control device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the automated intelligent control method as described in any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the automated intelligent control method as described in any one of claims 1 to 6.

Images