Intelligent control method and system for polyurethane production based on digital twinning

Abstract

The invention relates to the technical field of intelligent control, in particular to an intelligent control method and system for polyurethane production based on digital twinning. The method comprises the following steps: constructing a digital twin simulation model corresponding to a polyurethane production whole process flow, and synchronously accessing a real-time sensor data flow corresponding to a production line; performing dynamic assignment based on the real-time sensor data stream to generate a digital twinborn model instance; performing prospective process simulation operation based on the digital twinborn model instance to generate multi-physics field prediction data; a prediction time sequence curve set of the key process state parameters is extracted from the multi-physical field prediction data, high-frequency dynamic comparison is carried out, and a deviation attribution analysis report is generated; and generating a real-time optimization control instruction set based on the deviation attribution analysis report, and obtaining the actual response data of the production line after execution as a new round of real-time sensor data flow execution feedback, thereby forming a closed-loop intelligent control loop. According to the invention, the production regulation efficiency of polyurethane can be comprehensively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent control method and system for polyurethane production based on digital twins. Background Technology

[0002] In recent years, with the widespread application of polyurethane materials in various fields such as construction, automobiles, and home appliances, the market has placed higher demands on the performance consistency, production efficiency, and green and low-carbon level of polyurethane products. At the same time, the polyurethane production process is characterized by strong nonlinearity, multi-parameter coupling, and complex reaction mechanisms, involving the coordinated control of multiple key parameters such as reactor temperature, pressure, material viscosity, isocyanate index, and polyol feed rate. Digital twin technology, with its unique advantages of virtual simulation, real-time mapping, forward prediction, and closed-loop optimization, is gradually replacing traditional manual operation and conventional automated control modes, becoming an important technical means for polyurethane production process control, energy consumption optimization, and product quality improvement.

[0003] Currently, several polyurethane production control methods based on digital technology have been proposed. These methods mostly involve deploying various sensors on the production line to collect real-time operating data, combining fixed parameter control with experience feedback modes to start / stop or fine-tune parameters of key equipment in the production process, and then performing preliminary optimization of the production process based on historical production data and experience models. However, existing methods lack the ability to perform global simulation and dynamic mapping of the entire production process. Affected by factors such as raw material batch fluctuations, multi-physics coupling within the reactor, and equipment performance drift, the data collected by sensors is prone to deviate from the actual reaction state, resulting in a lag in process parameter control. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a digital twin-based intelligent control method and system for polyurethane production to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a digital twin-based intelligent control method for polyurethane production includes the following steps: Step S1: Construct a digital twin simulation model corresponding to the entire polyurethane production process, and synchronously connect to the real-time sensor data stream corresponding to the production line, which includes real-time temperature, real-time pressure, material viscosity, isocyanate index and polyol feed rate in the reactor; dynamically assign values ​​to the initial state parameters of the digital twin simulation model based on the real-time sensor data stream to generate a digital twin model instance. Step S2: Based on the digital twin model instance, inject the preset production formula parameters and process set point sequence to perform forward-looking process simulation calculations, and generate multi-physics prediction data covering the entire reaction cycle; extract the set of prediction time series curves of key process state parameters from the multi-physics prediction data; Step S3: Perform high-frequency dynamic comparison between the predicted time series curve set and the actual process state parameter time series collected from the production line in real time, and calculate the set of state deviations of each parameter between the digital twin virtual space and the physical entity space at each comparison time; based on the set of state deviations, solve the potential process disturbance sources and equipment performance drift factors that cause the deviations, and generate a deviation attribution analysis report. Step S4: Based on the deviation attribution analysis report, match the current deviation attribution scenario to generate a real-time optimization control instruction set for the physical production line; send the real-time optimization control instruction set to the distributed control system of the polyurethane production line for execution, driving the physical production line equipment to make dynamic adjustments; at the same time, feed back the actual response data of the production line after execution as a new round of real-time sensor data stream to step S1 to form a closed-loop intelligent control loop.

[0006] Furthermore, step S1 includes the following steps: Step S11: Identify the core units of the entire polyurethane production process, including the raw material pretreatment unit, reaction synthesis unit, curing and modification unit, molding and cutting unit, and quality inspection unit. Collect the equipment geometric parameters, material properties, process constraints, and historical production operation data of each unit to build the basic architecture of the digital twin simulation model and clarify the material transfer path, energy exchange relationship, and timing linkage logic between each unit. Step S12: Deploy a distributed sensor network, synchronously access the real-time sensor data streams of each core unit of the production line and perform real-time noise reduction processing, remove abnormal pulse data and time-series missing data, and generate a purified real-time sensor data stream. Step S13: Based on the purified real-time sensor data stream, extract the real-time operating status parameters of each core unit, establish parameter mapping relationships, assign the real-time status parameters of the physical production line to the corresponding nodes of the digital twin simulation model, and generate an initial digital twin model instance. Step S14: Collect the running status data of the initial digital twin model instance in real time, make a preliminary comparison with the purified real-time sensor data stream, calculate the assignment deviation of the initial state of the model, and dynamically correct the initial state parameters of the initial digital twin model instance based on the assignment deviation to generate the corrected digital twin model instance.

[0007] Furthermore, step S2 includes the following steps: Step S21: Based on the preset specifications of polyurethane products, determine the production formula parameters and the sequence of time process set points, clarify the parameter control range, timing nodes and switching conditions of each process stage, and structure the production formula parameters and the sequence of time process set points to generate process control instructions. Step S22: Inject process control commands into the digital twin model instance, start forward-looking process simulation calculation, simulate the material reaction process, energy transfer process and equipment operation process of the entire reaction cycle of polyurethane production, and generate multi-physics field prediction data covering the entire reaction cycle, including time-series change data of temperature field, pressure field, material concentration field and viscosity field in the reactor. Step S23: Select key process state parameters from the multiphysics prediction data, including real-time temperature prediction, real-time pressure prediction, material viscosity prediction, isocyanate index prediction, polyol feed rate prediction, and reaction conversion rate prediction, and extract the time series data of each key process state parameter in the whole reaction cycle. Step S24: Based on the time series data of each key process state parameter, draw the corresponding predicted time series curves, mark the time series nodes, parameter change trends and key feature points of each curve, and generate a set of predicted time series curves for key process state parameters. Step S25: Perform timing consistency verification on the predicted timing curve set, calculate the timing synchronization error between each curve, and correct the predicted timing curve set based on the timing synchronization error to ensure that the predicted timing of each key process state parameter is consistent with the actual production timing.

[0008] Furthermore, step S23 includes the following steps: Based on the reaction mechanism of polyurethane production, the types of key process state parameters that affect product quality and production stability are identified, the detection standards and time sequence collection requirements for each key parameter are clarified, and key parameter screening criteria are established. The multiphysics prediction data is traversed, and the time series data corresponding to each key process state parameter is extracted according to the key parameter screening criteria. Invalid prediction data and abnormal prediction data that are outside the process control range are removed to generate the original time series dataset of key parameters. The original time series dataset of key parameters is then subjected to time series interpolation to supplement the missing time series data, so that the time series data of each key process state parameter remains continuous, and a continuous key parameter time series dataset is generated. Based on the continuous key parameter time series dataset, the rate of change and time series fluctuation amplitude of each key process state parameter in different reaction stages are calculated, the change characteristics of each parameter are extracted, and a key parameter change characteristic dataset is generated. Based on the key parameter change feature dataset, the predicted time series data of each key process state parameter are matched, and the time series data of each key process state parameter within the entire reaction cycle are extracted.

[0009] Furthermore, the step of calculating the rate of change and time-series fluctuation amplitude of each key process state parameter at different reaction stages based on a continuous key parameter time-series dataset, and extracting the change characteristics of each parameter, includes the following steps: The continuous key parameter time series dataset is divided according to the reaction stages of polyurethane production, and the time interval and time length of each reaction stage are defined to generate a stage-specific key parameter time series dataset. Based on the stage-specific key parameter time series dataset, the time series change rate of each key process state parameter in each reaction stage is calculated, and the parameter change rate value at each moment is obtained through the time series difference algorithm to generate a key parameter stage-specific change rate dataset. Based on the key parameter phase change rate dataset, calculate the rate of change fluctuation of each key process state parameter in each reaction phase, determine the stability characteristics of each parameter change, and generate a key parameter fluctuation characteristic dataset. Based on the key parameter fluctuation feature dataset, the fluctuation peak, fluctuation frequency and time series distribution of each key process state parameter in different reaction stages are extracted to generate key parameter fluctuation feature indicators. By integrating the datasets of key parameter phased change rates, key parameter fluctuation characteristics, and key parameter fluctuation characteristics indicators, a dataset of key parameter change characteristics is generated, clarifying the temporal change patterns and fluctuation characteristics of each key parameter.

[0010] Furthermore, step S3 includes the following steps: Step S31: Establish a time synchronization mapping channel between each predicted time series curve and the actual process state parameter time series collected from the production line in real time; based on the time synchronization mapping channel, perform point-by-point difference calculation on each predicted time series curve and the corresponding actual process state parameter time series, calculate the instantaneous absolute deviation and relative deviation between the predicted value and the measured value, and integrate the time series to generate a set of state deviation quantities. Step S32: Perform time-frequency analysis on the set of state deviation quantities, extract the fluctuation patterns, trend components and abrupt change characteristics of each parameter deviation signal at different time scales, and cluster deviation signals with similar fluctuation patterns and trend characteristics to generate a set of deviation pattern feature vectors; Step S33: By pre-training an inverse mapping network containing perturbation sources and drift factors, the network learns a complex nonlinear mapping relationship between typical deviation patterns and a series of potential causes; the perturbation sources include at least raw material component fluctuations, catalyst deactivation, and temperature control loop disturbances; the drift factors include at least stirring efficiency decay factors and heat transfer coefficient drift factors. Step S34: Input the feature vector set of deviation modes into the inverse mapping network to solve and output the probability of the existence of potential process disturbance sources and equipment performance drift factors corresponding to each deviation mode and the estimated value of their quantitative contribution to the current overall deviation. Integrate all probability and quantitative contribution estimates to generate a deviation attribution analysis report.

[0011] Furthermore, step S4 includes the following steps: Step S41: Establish a deviation attribution scenario matching library, covering deviation scenarios and corresponding control strategies for various process disturbances and equipment drifts. Compare the attribution results in the deviation attribution analysis report with the deviation attribution scenario matching library to determine the attribution scenario and control direction corresponding to the current deviation. Step S42: Based on the current deviation attribution scenario and control direction, and combined with the real-time operating status of the digital twin model instance, generate real-time optimized control instructions for each core unit of the physical production line, including raw material feed rate adjustment instructions, reactor temperature and pressure control instructions, and equipment operating parameter correction instructions. Integrate all control instructions to generate a real-time optimized control instruction set. Step S43: Input the real-time optimized control instruction set into the digital twin model instance for virtual simulation verification, simulate the production line response effect after the control instructions are executed, and calculate the predicted deviation after the control instructions are executed; if the predicted deviation meets the preset control requirements, the real-time optimized control instruction set is sent to the distributed control system of the polyurethane production line to drive each piece of equipment on the physical production line to make dynamic adjustments according to the control instructions; if the predicted deviation does not meet the preset control requirements, the real-time optimized control instruction set is corrected based on the predicted deviation, and the virtual simulation verification steps are repeated until the predicted deviation meets the requirements; Step S44: Collect the actual response data of the production line after the execution of the corresponding control command that meets the requirements in real time, and feed the actual response data back to step S1 as a new round of real-time sensor data stream to update the state parameters of the digital twin model instance, start the next round of simulation, comparison, attribution and control process, and form a closed-loop intelligent control loop.

[0012] Furthermore, step S43 includes the following steps: Extract the parameters of each control instruction from the real-time optimization control instruction set, clarify the execution object, execution sequence, adjustment range and execution duration of each control instruction, perform structured parsing of the control instruction parameters, and generate virtual control instructions that can be injected into digital twin model instances; Inject virtual control commands into the current digital twin model instance, keep other parameters of the model consistent with the real-time state of the physical production line, start virtual simulation verification, simulate the state changes of the digital twin model instance during the execution of control commands, and generate virtual response data; The time-series variation data of key process state parameters are extracted from the virtual response data, and a virtual response time-series curve is plotted. This curve is compared with the preset process standard time-series curve to calculate the predicted deviation after the execution of the virtual response control command. The distribution characteristics and time-series variation trend of the predicted deviation are analyzed to determine whether the predicted deviation is within the preset control requirements. If the predicted deviation meets the preset control requirements, the real-time optimized control command set is sent to the distributed control system of the polyurethane production line to drive each piece of equipment on the physical production line to dynamically adjust according to the control command. If it exceeds the range, the abnormal deviation and time-series node that exceed the control requirements and the corresponding virtual control command are marked, and the deviation node and the corresponding virtual control command are corrected based on the predicted deviation. The virtual simulation verification steps are repeated until the predicted deviation meets the requirements.

[0013] Furthermore, if the deviation exceeds the range, marking the abnormal deviation amount and timing node that exceed the control requirements and the corresponding virtual control command includes the following steps: Determine the deviation threshold range corresponding to the preset control requirements, clarify the upper and lower limits of the allowable deviation for each key process state parameter, and establish deviation judgment criteria; traverse the predicted deviation amount, compare the predicted deviation amount at each time point with the corresponding deviation threshold range, determine whether each deviation amount meets the preset control requirements, and mark the abnormal deviation amount that exceeds the threshold range and the corresponding time sequence node. Analyze the timing nodes and key parameter types corresponding to abnormal deviations, and in conjunction with the execution timing of virtual control commands, determine the target control command that caused the abnormal deviation, and clarify the correlation between the adjustment range, execution timing, and abnormal deviation of the target control command; specifically including: Based on the timing nodes corresponding to abnormal deviations, the virtual control instruction execution stage corresponding to each timing node is determined, and all control instructions being executed within this stage and their execution status are clarified. Based on the virtual response data of the digital twin model instance, the changing patterns of key process state parameters during the execution of each control instruction are analyzed, and a correlation model between control instructions and parameter changes is established. The virtual control instruction execution stage corresponding to the timing node is input into the correlation model to invert the control instruction that caused the abnormal change in the parameter, thus determining the target control instruction. The adjustment range and execution timing of the target control instruction are extracted, and the quantitative relationship between the adjustment range and execution timing of the target control instruction and the abnormal deviation is analyzed. The influence coefficient of the control instruction parameter change on the abnormal deviation is calculated. Based on the influence coefficient, the correlation strength and timing lag characteristics between the parameter adjustment of the target control instruction and the abnormal deviation are clarified, and the correlation between the adjustment range and execution timing of the target control instruction and the abnormal deviation is determined. Calculate the difference between the abnormal deviation amount and the deviation threshold to determine the extent of the deviation, generate abnormal deviation analysis data, and clarify the severity level and time propagation range of the abnormal deviation; Based on the abnormal deviation analysis data, determine whether the current real-time optimized control instruction set needs to be corrected. If it needs to be corrected, mark the control instruction that needs to be corrected and the direction of correction.

[0014] Furthermore, the present invention also provides a digital twin-based intelligent control system for polyurethane production, including a processor, a memory, and a computer program stored in the memory and executable on the processor, for performing the digital twin-based intelligent control method for polyurethane production as described above.

[0015] The beneficial effects of this invention are: The intelligent control method for polyurethane production based on digital twins proposed in this invention offers several advantages over existing technologies. Firstly, by synchronously integrating key operational data such as real-time temperature, pressure, material viscosity, isocyanate index, and polyol feed rate within the reactor into the model, it achieves comprehensive integration between the physical production line and the virtual simulation space. This overcomes the limitations of traditional control methods that rely solely on local sensor data and cannot cover the entire process. Secondly, by dynamically assigning initial parameters to the model based on real-time data streams, it can quickly generate digital twin model instances that closely match the current production line conditions. This ensures that the virtual model is always updated synchronously with the physical production line, effectively offsetting initial state differences caused by raw material batch fluctuations and environmental changes. Compared to traditional fixed-parameter modeling methods, this step significantly improves the fit between the virtual model and actual production conditions, providing a stable and reliable virtual platform for subsequent full-process simulation, deviation analysis, and optimization control. This addresses the issue of data discrepancies with actual reaction states from the source, enhancing the reliability and operational stability of the entire control process. Secondly, by injecting preset production formula parameters and process setpoint sequences, the model can completely simulate the changes in multiple physical fields such as temperature, pressure, viscosity, and component ratios throughout the entire reaction cycle in virtual space, generating predictive data covering the entire reaction process and realizing the transformation from passive monitoring to proactive prediction. Extracting the predicted time-series curves of key process state parameters from the multi-physics prediction data clearly presents the changing trends and fluctuation ranges of each indicator in future periods, providing a standardized reference for subsequent real-time comparison and anomaly identification. This effectively addresses the complex changes brought about by the coupling of multiple physical fields within the reactor, compensates for the shortcomings of traditional experience models that cannot predict the entire process in advance, reduces the blindness of process control, and lays a complete reference foundation for subsequent dynamic deviation comparison. Then, by performing high-frequency point-by-point comparisons between the predicted time-series curves and the actual process state time-series sequences, the state differences between the virtual and physical spaces can be captured in real time, forming a systematic set of state deviation quantities, and promptly detecting process deviations caused by factors such as raw material fluctuations, environmental interference, and equipment aging. Based on the set of deviations, further calculations of potential process disturbance sources and equipment performance drift factors can clearly distinguish the specific causes of deviations, generating a structured deviation attribution analysis report. This avoids the predicament of traditional control methods that can only detect parameter anomalies but cannot pinpoint the root cause, reduces the error between sensor data and the actual response state, and provides a reliable basis for subsequent development of targeted optimization control strategies, significantly improving the targeting and rationality of process control. Finally, by generating a real-time optimized control instruction set based on deviation attribution scenario matching, it can directly drive the physical production line's distributed control system to perform parameter adjustments, achieving rapid response to process disturbances and equipment drift, avoiding the problems of slow response and unreasonable adjustment range in traditional experience-based control.The actual operating data of the adjusted production line is fed back to the digital twin model, forming a complete closed loop of data acquisition, modeling, simulation, comparison, optimization, and execution. This enables the entire control system to have continuous self-optimization capabilities, continuously reducing the state deviation between the virtual and the real, thereby promoting the continuous upgrading of production control towards intelligence and precision. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps in the intelligent control method for polyurethane production based on digital twins according to the present invention. Figure 2 for Figure 1 A detailed flowchart of step S2. Detailed Implementation

[0017] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides an intelligent control method for polyurethane production based on digital twins. In the embodiments of this invention, please refer to... Figure 1 The diagram shown is a flowchart illustrating the steps of the intelligent control method for polyurethane production based on digital twins according to the present invention. In this example, the intelligent control method for polyurethane production based on digital twins includes the following steps: Step S1: Construct a digital twin simulation model corresponding to the entire polyurethane production process, and synchronously connect to the real-time sensor data stream corresponding to the production line, which includes real-time temperature, real-time pressure, material viscosity, isocyanate index and polyol feed rate in the reactor; dynamically assign values ​​to the initial state parameters of the digital twin simulation model based on the real-time sensor data stream to generate a digital twin model instance. In this embodiment of the invention, by analyzing the entire polyurethane production process, encompassing five core units—raw material pretreatment, reaction synthesis, curing modification, molding and cutting, and quality inspection—a digital twin simulation model corresponding to the physical production line at a 1:1 scale is constructed. The geometric parameters, material properties, and process constraints of each unit's equipment are clearly defined. For example, the raw material pretreatment unit has a 1000L feed pump made of 316 stainless steel; the reaction synthesis unit has an 8000L reactor with a 15mm wall thickness made of 316L stainless steel; the curing modification unit has a 10000L curing tank made of carbon steel; the molding and cutting unit has a cutting accuracy of ±0.5mm; and the quality inspection unit has a detection accuracy of ±0.01g. A distributed sensor network is deployed, with sensors installed at key locations in each unit, synchronously accessing the real-time sensor data stream of the production line. The data stream includes real-time temperature, pressure, material viscosity, isocyanate index, and polyol feed rate within the reactor. One set of data is collected every minute, and a moving average denoising process is used. The moving average window is set to 5 sampling points, and the denoising formula is y. n '=(y n-2 +y n-1 +y n +y n+1 +y n+2 ) / 5, abnormal pulse data with amplitudes exceeding three times the normal range and missing data with three consecutive missing sampling points are removed to generate a purified real-time sensor data stream. Based on the purified real-time sensor data stream, the initial state parameters of the digital twin simulation model are dynamically assigned. The real-time temperature (82℃), real-time pressure (0.33MPa), material viscosity (1600mPa·s), isocyanate index (1.05), and polyol feed rate (55L / h) in the reactor are assigned to the corresponding nodes of the digital twin simulation model, respectively. The initial assignment deviation is calculated using the formula Δx = |xmodel - xactual|, with a temperature deviation of 0.2℃ and a pressure deviation of 0.01MPa. Based on the deviation, the initial state parameters are corrected using the formula xcorrection = xmodel - Δx × 0.6, resulting in a corrected temperature of 81.88℃ and a pressure of 0.324MPa, generating a digital twin model instance.

[0019] Step S2: Based on the digital twin model instance, inject the preset production formula parameters and process set point sequence to perform forward-looking process simulation calculations, and generate multi-physics prediction data covering the entire reaction cycle; extract the set of prediction time series curves of key process state parameters from the multi-physics prediction data; In this embodiment of the invention, a preset production formula parameter and process setpoint sequence are injected based on a digital twin model instance. The production formula parameter is a molar ratio of isocyanate to polyol of 1.05:1, a catalyst addition of 0.3%, and a foaming agent addition of 2%. The process setpoint sequence is: raw material pretreatment 28℃ / 15 minutes → reaction synthesis 82.5℃ / 40 minutes → ripening modification 62℃ / 60 minutes → molding and cutting 2m / min → quality inspection once every 10 minutes. The parameter control range of each process stage is clearly defined. The temperature of the reaction synthesis stage is 80-85℃ and the pressure is 0.3-0.4MPa. A forward-looking process simulation was initiated, with a simulation step size of 1 minute. The simulation simulated the material reaction process, energy transfer process, and equipment operation process of the entire 130-minute reaction cycle of polyurethane production. Multiphysics prediction data covering the entire reaction cycle was generated, including time-series changes in the temperature field, pressure field, material viscosity field, isocyanate index field, and polyol feed rate field within the reactor. The temperature field gradually increased from 28℃ to 82.5℃ and then decreased to 62℃; the pressure field increased from 0.1MPa to 0.33MPa and then decreased to 0.1MPa; the material viscosity field increased from 500mPa·s to 1850mPa·s; the isocyanate index field decreased from 1.10 to 1.03; and the polyol feed rate field increased from 55L / h to 58L / h and then remained stable. The set of predicted time-series curves for key process state parameters is extracted from multiphysics prediction data. The key parameters are real-time temperature, real-time pressure, material viscosity, isocyanate index and polyol feed rate in the reactor. Each curve contains 130 time-series data points. The time-series nodes, parameter change trends and key feature points of each curve are marked to generate the set of predicted time-series curves.

[0020] Step S3: Perform high-frequency dynamic comparison between the predicted time series curve set and the actual process state parameter time series collected from the production line in real time, and calculate the set of state deviations of each parameter between the digital twin virtual space and the physical entity space at each comparison time; based on the set of state deviations, solve the potential process disturbance sources and equipment performance drift factors that cause the deviations, and generate a deviation attribution analysis report. In this embodiment of the invention, a time synchronization mapping channel is established between the set of predicted time-series curves and the time-series sequence of actual process state parameters collected in real time from the production line. Using timestamps as a reference, each data point of each predicted time-series curve is mapped one-to-one with the corresponding data point in the actual time-series sequence, with the time synchronization error controlled within 0.1 minutes. High-frequency dynamic comparisons are performed, with a full parameter comparison completed every minute. The set of state deviations of each parameter between the digital twin virtual space and the physical entity space at each comparison moment is calculated. The deviations are calculated using absolute deviations: temperature absolute deviation ΔT = |Tvirtual - Tactual|, pressure absolute deviation ΔP = |Pvirtual - Pactual|, material viscosity absolute deviation Δη = |ηvirtual - ηactual|, isocyanate index absolute deviation ΔI = |Ivirtual - Iactual|, and polyol feed rate absolute deviation Δv = |vvirtual - vactual|. At the 20th minute, the virtual temperature T is 82.3℃, the actual temperature T is 82.1℃, and the ΔT = 0.2℃; the virtual pressure P is 0.328MPa, the actual pressure P is 0.327MPa, and the ΔP = 0.001MPa; the virtual pressure η is 1700mPa·s, the actual pressure η is 1680mPa·s, and the Δη = 20mPa·s; the virtual pressure I is 1.04, the actual pressure I is 1.03, and the ΔI = 0.01; the virtual flow rate v is 57L / h, the actual flow rate v is 56.5L / h, and the Δv = 0.5L / h. The deviations at all times are calculated point by point and integrated to generate a set of state deviations. The potential process disturbance sources and equipment performance drift factors that cause deviations are calculated by using a pre-trained inverse mapping network. This network has learned the nonlinear mapping relationship between typical deviation patterns and potential causes. Disturbance sources include raw material composition fluctuations and catalyst deactivation. Drift factors include stirring efficiency decay and heat transfer coefficient drift. The calculation yields a probability of 75% for raw material composition fluctuations and a quantitative contribution of 40%, a probability of 60% for catalyst deactivation and a quantitative contribution of 30%, a probability of 55% for stirring efficiency decay and a quantitative contribution of 20%, and a probability of 50% for heat transfer coefficient drift and a quantitative contribution of 10%. The results are then integrated to generate a deviation attribution analysis report.

[0021] Step S4: Based on the deviation attribution analysis report, match the current deviation attribution scenario to generate a real-time optimization control instruction set for the physical production line; send the real-time optimization control instruction set to the distributed control system of the polyurethane production line for execution, driving the physical production line equipment to make dynamic adjustments; at the same time, feed back the actual response data of the production line after execution as a new round of real-time sensor data stream to step S1 to form a closed-loop intelligent control loop.

[0022] In this embodiment of the invention, a deviation attribution scenario matching library is established. This library covers various process disturbances and equipment drift, along with corresponding deviation scenarios and control strategies. For example, the control strategy for raw material component fluctuation deviation scenarios is to adjust the polyol feed rate; the control strategy for catalyst deactivation deviation scenarios is to replenish the catalyst; the control strategy for temperature control loop disturbances is to adjust the temperature control parameters; the control strategy for decreased stirring efficiency is to increase the stirring speed; and the control strategy for heat transfer coefficient drift is to clean the scale buildup on the reactor wall. By comparing the attribution results in the deviation attribution analysis report with the deviation attribution scenario matching library, it is determined that the current deviation is mainly attributed to raw material component fluctuations and catalyst deactivation, and the corresponding control directions are adjusting the polyol feed rate and replenishing the catalyst. Based on the current deviation attribution scenario and control direction, and combined with the real-time operating status of the digital twin model instance, a real-time optimized control instruction set for the physical production line is generated. This set includes three core instructions: adjusting the polyol feed rate of the raw material pretreatment unit from 55 L / h to 58 L / h; adding 0.05% catalyst to the reaction synthesis unit; adjusting the temperature control target of the reaction synthesis unit from 82℃ to 82.5℃; and simultaneously increasing the stirring speed from 60 r / min to 65 r / min. This real-time optimized control instruction set is then sent to the distributed control system of the polyurethane production line for execution, driving dynamic adjustments to each piece of equipment on the physical production line. Specifically, the feed rate of the raw material pretreatment unit's feed pump is adjusted, and the reaction synthesis unit adds catalyst and adjusts the temperature and stirring speed. The actual response data of the production line after the execution of the control command is collected in real time: temperature 82.4℃, pressure 0.325MPa, material viscosity 1720mPa·s, isocyanate index 1.04, and polyol feed rate 58L / h. This actual response data is fed back to step S1 as a new round of real-time sensor data stream to update the state parameters of the digital twin model instance. After correction, the temperature is 82.4℃ and the pressure is 0.325MPa. The next round of simulation, comparison, attribution and control process is started to form a closed-loop intelligent control loop.

[0023] Furthermore, step S1 includes the following steps: Step S11: Identify the core units of the entire polyurethane production process, including the raw material pretreatment unit, reaction synthesis unit, curing and modification unit, molding and cutting unit, and quality inspection unit. Collect the equipment geometric parameters, material properties, process constraints, and historical production operation data of each unit to build the basic architecture of the digital twin simulation model and clarify the material transfer path, energy exchange relationship, and timing linkage logic between each unit. Step S12: Deploy a distributed sensor network, synchronously access the real-time sensor data streams of each core unit of the production line and perform real-time noise reduction processing, remove abnormal pulse data and time-series missing data, and generate a purified real-time sensor data stream. Step S13: Based on the purified real-time sensor data stream, extract the real-time operating status parameters of each core unit, establish parameter mapping relationships, assign the real-time status parameters of the physical production line to the corresponding nodes of the digital twin simulation model, and generate an initial digital twin model instance. Step S14: Collect the running status data of the initial digital twin model instance in real time, make a preliminary comparison with the purified real-time sensor data stream, calculate the assignment deviation of the initial state of the model, and dynamically correct the initial state parameters of the initial digital twin model instance based on the assignment deviation to generate the corrected digital twin model instance.

[0024] In this embodiment of the invention, by sorting out the core units of the entire polyurethane production process, it is clearly covered that include raw material pretreatment unit, reaction synthesis unit, curing and modification unit, molding and cutting unit and quality inspection unit. The equipment geometric parameters, material properties, process constraints, and historical production operation data of each unit were collected. The raw material pretreatment unit's reactor has a volume of 5000L, a wall thickness of 15mm, and is made of 316 stainless steel. The process constraints are: raw material mixing temperature 25-30℃, mixing time 15 minutes. The reaction synthesis unit's reactor has a volume of 8000L, a stirring speed of 60r / min, and is made of 316L stainless steel. The process constraints are: reaction temperature 80-85℃, reaction pressure 0.3-0.4MPa. The curing and modification unit's curing tank has a volume of 10000L, is made of carbon steel, and the process constraints are: curing temperature 60-65℃, curing time 60 minutes. The forming and cutting unit's cutting machine has a cutting accuracy of ±0.5mm, is made of alloy steel, and the process constraint is: cutting speed 2m / min. The quality inspection unit has an inspection accuracy of ±0.01g, and the process constraint is: inspection frequency once every 10 minutes. Historical production operation data, totaling 180 sets over nearly 6 months, was collected, covering parameters such as temperature, pressure, and rotational speed for each unit. The basic architecture for a digital twin simulation model was constructed, clearly defining the material transfer path between units as: raw material pretreatment unit, reaction synthesis unit, ripening and modification unit, forming and cutting unit, and quality inspection unit. The energy exchange relationship is the transfer of waste heat from the reaction synthesis unit to the ripening and modification unit. The timing linkage logic is that the reaction synthesis unit is triggered to start after raw material pretreatment, with an interval of 5 minutes. A distributed sensor network was deployed, with sensors installed at key locations in each unit, synchronously accessing real-time sensor data streams from each core unit of the production line. Moving average denoising was used, with a moving window of 5 sampling points. The denoising formula is y... n '=(y n-2 +y n-1 +y n +y n+1 +y n+2) / 5, abnormal pulse data with amplitudes exceeding three times the normal range and missing data with three consecutive missing sampling points are removed to generate a purified real-time sensor data stream. Based on the purified real-time sensor data stream, the real-time operating status parameters of each core unit are extracted, and a parameter mapping relationship is established. The real-time temperature of 82℃ and pressure of 0.35MPa of the physical production line reaction synthesis unit are assigned to the reaction synthesis unit node of the digital twin simulation model. The assignment of parameters for all units is completed in sequence to generate an initial digital twin model instance. The operating status data of the initial digital twin model instance is collected in real time and preliminarily compared with the purified real-time sensor data stream to calculate the assignment deviation of the initial state of the model. The deviation calculation formula is Δx=|xmodel-xactual|. The temperature deviation of the reaction synthesis unit is 0.8℃ and the pressure deviation is 0.02MPa. Based on this assignment deviation, the initial state parameters of the initial digital twin model instance are dynamically corrected. The correction formula is xcorrection=xmodel-Δx×0.6. After correction, the temperature is 81.5℃ and the pressure is 0.34MPa, generating a corrected digital twin model instance.

[0025] Furthermore, as an embodiment of the present invention, reference is made to... Figure 2 As shown, Figure 1 A detailed flowchart of step S2 is shown below. In this embodiment, step S2 includes the following steps: Step S21: Based on the preset specifications of polyurethane products, determine the production formula parameters and the sequence of time process set points, clarify the parameter control range, timing nodes and switching conditions of each process stage, and structure the production formula parameters and the sequence of time process set points to generate process control instructions. Step S22: Inject process control commands into the digital twin model instance, start forward-looking process simulation calculation, simulate the material reaction process, energy transfer process and equipment operation process of the entire reaction cycle of polyurethane production, and generate multi-physics field prediction data covering the entire reaction cycle, including time-series change data of temperature field, pressure field, material concentration field and viscosity field in the reactor. Step S23: Select key process state parameters from the multiphysics prediction data, including real-time temperature prediction, real-time pressure prediction, material viscosity prediction, isocyanate index prediction, polyol feed rate prediction, and reaction conversion rate prediction, and extract the time series data of each key process state parameter in the whole reaction cycle. Step S24: Based on the time series data of each key process state parameter, draw the corresponding predicted time series curves, mark the time series nodes, parameter change trends and key feature points of each curve, and generate a set of predicted time series curves for key process state parameters. Step S25: Perform timing consistency verification on the predicted timing curve set, calculate the timing synchronization error between each curve, and correct the predicted timing curve set based on the timing synchronization error to ensure that the predicted timing of each key process state parameter is consistent with the actual production timing.

[0026] In this embodiment of the invention, based on the preset specifications of the polyurethane product, the product density is 1.2 g / cm³. 3The hardness is Shore A85 and the tensile strength is 35MPa. The production formula parameters are determined as follows: isocyanate to polyol molar ratio 1.05:1, catalyst addition 0.3%, foaming agent addition 2%. The time sequence of process set points is: raw material pretreatment 28℃ / 15min → reaction synthesis 83℃ / 40min → ripening modification 62℃ / 60min → molding and cutting 2m / min → quality inspection once / 10min. The parameter control range of each process stage is defined. The temperature control range of the reaction synthesis stage is 80-85℃ and the pressure control range is 0.3-0.4MPa. The time nodes are: start raw material pretreatment at 0 minutes, start reaction synthesis at 15 minutes, start ripening modification at 55 minutes, start molding and cutting at 115 minutes, and start quality inspection at 125 minutes. The switching condition is that the mixing uniformity of the raw material pretreatment reaches 95% and then switch to the reaction synthesis stage. The production formula parameters and the sequence of time-series process setpoints are structured and coded, categorized by unit. The raw material pretreatment unit is coded as YC-01, and the reaction synthesis unit is coded as FY-01, generating process control instructions. These instructions are then injected into a digital twin model instance, initiating a forward-looking process simulation with a simulation step size of 1 minute. This simulates the material reaction process, energy transfer process, and equipment operation process throughout the 130-minute polyurethane production reaction cycle, generating multiphysics prediction data covering the entire reaction cycle. The temperature field within the reactor gradually increases from 28℃ to 83℃ and then decreases to 62℃; the pressure field increases from 0.1MPa to 0.35MPa and then decreases to 0.1MPa; the isocyanate concentration decreases from 25% to 5%; and the viscosity increases from 500mPa·s to 2000mPa·s. Key process state parameters were selected from multiphysics prediction data, including real-time predicted temperatures, pressures, material viscosity, isocyanate index, polyol feed rate, and reaction conversion rate within the reactor. Time-series data for each key process state parameter were extracted over the entire reaction cycle, recording one data point every minute, for a total of 130 data points. Based on the time-series data of each key process state parameter, corresponding predicted time-series curves were plotted. The temperature curves were labeled with time nodes of 28℃ at 15 minutes, 83℃ at 55 minutes, and 62℃ at 115 minutes. The marked parameter change trends were early-stage heating, mid-stage isothermal, and late-stage cooling. The key characteristic point was the peak temperature of 83℃. All curves were labeled sequentially, generating a set of predicted time-series curves for the key process state parameters. The timing consistency of the predicted timing curve set is checked, and the timing synchronization error between each curve is calculated. The error calculation formula is Δt=|t1-t2|. The maximum timing synchronization error between the temperature curve and the pressure curve is 0.3 minutes. Based on the timing synchronization error, the predicted timing curve set is corrected, and the timing node corresponding to the pressure curve is advanced by 0.3 minutes to ensure that the predicted timing of each key process state parameter is consistent with the actual production timing.

[0027] Furthermore, step S23 includes the following steps: Based on the reaction mechanism of polyurethane production, the types of key process state parameters that affect product quality and production stability are identified, the detection standards and time sequence collection requirements for each key parameter are clarified, and key parameter screening criteria are established. The multiphysics prediction data is traversed, and the time series data corresponding to each key process state parameter is extracted according to the key parameter screening criteria. Invalid prediction data and abnormal prediction data that are outside the process control range are removed to generate the original time series dataset of key parameters. The original time series dataset of key parameters is then subjected to time series interpolation to supplement the missing time series data, so that the time series data of each key process state parameter remains continuous, and a continuous key parameter time series dataset is generated. Based on the continuous key parameter time series dataset, the rate of change and time series fluctuation amplitude of each key process state parameter in different reaction stages are calculated, the change characteristics of each parameter are extracted, and a key parameter change characteristic dataset is generated. Based on the key parameter change feature dataset, the predicted time series data of each key process state parameter are matched, and the time series data of each key process state parameter within the entire reaction cycle are extracted.

[0028] In this embodiment of the invention, based on the reaction mechanism of polyurethane production, the types of key process state parameters affecting product quality and production stability are determined. These parameters are identified as real-time temperature and pressure inside the reactor, material viscosity, isocyanate index, polyol feed rate, and reaction conversion rate. The detection standards and time-series acquisition requirements for each key parameter are also defined. The temperature detection standard is 80-85℃, the pressure detection standard is 0.3-0.4MPa, the material viscosity detection standard is 500-2000mPa·s, the isocyanate index detection standard is 0.95-1.15, the polyol feed rate detection standard is 50-60L / h, and the reaction conversion rate detection standard is 70%-95%. The time-series acquisition requirement is to collect data once every 1 minute. Key parameter screening criteria are established, which require that the parameter values ​​be within the detection standard range and that the time-series data be continuous and without jumps. The multiphysics prediction data was traversed, and the time series data corresponding to each key process state parameter was extracted according to the key parameter screening criteria. Abnormal prediction data that exceeded the process control range, such as temperature 86℃ and pressure 0.42MPa, were removed. A total of 6 abnormal data were removed, and the original time series dataset of key parameters was generated. Temporal interpolation processing is performed on the original time-series dataset of key parameters. For two missing time-series data points, linear interpolation is used, with the interpolation formula being y=y1+(x-x1)×(y2-y1) / (x2-x1), where x is the timestamp of the missing point, x1 and x2 are the timestamps of the adjacent valid sampling points before and after the missing point, and y1 and y2 are the parameter values ​​of the corresponding valid sampling points. For example, if the missing point timestamp is 20 minutes, x1=19 minutes, y1=75℃, x2=21 minutes, y2=78℃, the temperature of the missing point is calculated as y=75+(20-19)×(78-75) / (21-19)=76.5℃. By supplementing the missing time-series data, the time-series data of each key process state parameter are kept continuous, generating a continuous key parameter time-series dataset. Based on a continuous time-series dataset of key parameters, the rate of change and the amplitude of fluctuation for each key process state parameter at different reaction stages are calculated. The rate of change is calculated using the formula v = (x2 - x1) / (t2 - t1), and the amplitude of fluctuation is calculated using the formula A = |maximum x - minimum x|. During the reaction synthesis stage, the temperature rises from 80℃ to 83℃ in 40 minutes, with a rate of change v = (83 - 80) / 40 = 0.075℃ / minute and an amplitude of fluctuation A = 83 - 80 = 3℃. The rate of change and amplitude of change for all parameters at each stage are calculated sequentially, and the change characteristics of each parameter are extracted to generate a dataset of key parameter change characteristics. Based on this dataset, the predicted time-series data of each key process state parameter are matched, and 130 time-series data points for each key process state parameter throughout the entire reaction cycle are extracted to ensure complete data correspondence.

[0029] Furthermore, the step of calculating the rate of change and time-series fluctuation amplitude of each key process state parameter at different reaction stages based on a continuous key parameter time-series dataset, and extracting the change characteristics of each parameter, includes the following steps: The continuous key parameter time series dataset is divided according to the reaction stages of polyurethane production, and the time interval and time length of each reaction stage are defined to generate a stage-specific key parameter time series dataset. Based on the stage-specific key parameter time series dataset, the time series change rate of each key process state parameter in each reaction stage is calculated, and the parameter change rate value at each moment is obtained through the time series difference algorithm to generate a key parameter stage-specific change rate dataset. Based on the key parameter phase change rate dataset, calculate the rate of change fluctuation of each key process state parameter in each reaction phase, determine the stability characteristics of each parameter change, and generate a key parameter fluctuation characteristic dataset. Based on the key parameter fluctuation feature dataset, the fluctuation peak, fluctuation frequency and time series distribution of each key process state parameter in different reaction stages are extracted to generate key parameter fluctuation feature indicators. By integrating the datasets of key parameter phased change rates, key parameter fluctuation characteristics, and key parameter fluctuation characteristics indicators, a dataset of key parameter change characteristics is generated, clarifying the temporal change patterns and fluctuation characteristics of each key parameter.

[0030] In this embodiment of the invention, the continuous key parameter time-series dataset is divided according to the reaction stages of polyurethane production, clearly defining the time interval and length of each reaction stage. The raw material pretreatment stage has a time interval of 0-15 minutes and a length of 15 minutes; the reaction synthesis stage has a time interval of 15-55 minutes and a length of 40 minutes; the curing and modification stage has a time interval of 55-115 minutes and a length of 60 minutes; and the molding, cutting, and quality inspection stage has a time interval of 115-130 minutes and a length of 15 minutes. This generates a phased key parameter time-series dataset. Based on this dataset, the temporal change rate of each key process state parameter within each reaction stage is calculated. The parameter change rate value at each moment is obtained using a temporal difference algorithm. The difference calculation formula is v. n =(x n -x n-1 ) / (t n -t n-1 ), t n -t n-1=1 minute. During the reaction synthesis stage, the temperature is 80.5℃ at the 16th minute and 80℃ at the 15th minute. The rate of change at these moments is calculated as v = (80.5-80) / 1 = 0.5℃ / minute. The rate values ​​at each moment are calculated sequentially to generate a dataset of the phased change rates of key parameters. Based on this dataset, the fluctuation amplitude of the rate of change of each key process state parameter in each reaction stage is calculated. The fluctuation amplitude is calculated using the formula A = v_maximum - v_min. During the reaction synthesis stage, the maximum temperature change rate is 0.8℃ / minute and the minimum is 0.05℃ / minute, with a fluctuation amplitude A = 0.8 - 0.05 = 0.75℃ / minute. The stability characteristics of each parameter's changes are determined; smaller temperature fluctuation amplitudes indicate better stability. A dataset of key parameter fluctuation characteristics is then generated. Based on the key parameter fluctuation characteristic dataset, the peak value, frequency, and temporal distribution of fluctuations of each key process state parameter at different reaction stages were extracted. During the reaction synthesis stage, the peak temperature change rate fluctuation was 0.8℃ / min, with a frequency of once every 5 minutes. The temporal distribution pattern showed a high rate in the early stage, a stable rate in the middle stage, and a low rate in the later stage, generating key parameter fluctuation characteristic indices. Integrating the key parameter stage change rate dataset, the key parameter fluctuation characteristic dataset, and the key parameter fluctuation characteristic indices, it was determined that the temperature change rate during the reaction synthesis stage was 0.075-0.8℃ / min with a fluctuation amplitude of 0.75℃ / min, and the pressure change rate was 0.006-0.01MPa / min with a fluctuation amplitude of 0.004MPa, generating a key parameter variation characteristic dataset and clarifying the temporal variation patterns and fluctuation characteristics of each key parameter.

[0031] Furthermore, step S3 includes the following steps: Step S31: Establish a time synchronization mapping channel between each predicted time series curve and the actual process state parameter time series collected from the production line in real time; based on the time synchronization mapping channel, perform point-by-point difference calculation on each predicted time series curve and the corresponding actual process state parameter time series, calculate the instantaneous absolute deviation and relative deviation between the predicted value and the measured value, and integrate the time series to generate a set of state deviation quantities. Step S32: Perform time-frequency analysis on the set of state deviation quantities, extract the fluctuation patterns, trend components and abrupt change characteristics of each parameter deviation signal at different time scales, and cluster deviation signals with similar fluctuation patterns and trend characteristics to generate a set of deviation pattern feature vectors; Step S33: By pre-training an inverse mapping network containing perturbation sources and drift factors, the network learns a complex nonlinear mapping relationship between typical deviation patterns and a series of potential causes; the perturbation sources include at least raw material component fluctuations, catalyst deactivation, and temperature control loop disturbances; the drift factors include at least stirring efficiency decay factors and heat transfer coefficient drift factors. Step S34: Input the feature vector set of deviation modes into the inverse mapping network to solve and output the probability of the existence of potential process disturbance sources and equipment performance drift factors corresponding to each deviation mode and the estimated value of their quantitative contribution to the current overall deviation. Integrate all probability and quantitative contribution estimates to generate a deviation attribution analysis report.

[0032] In this embodiment of the invention, a time synchronization mapping channel is established between each predicted time-series curve and the actual process state parameter time-series sequence collected in real time from the production line. Using timestamps as a reference, each data point of the predicted time-series curve is mapped one-to-one with the corresponding actual data point in the actual time-series sequence, with the time synchronization error controlled within 0.1 minutes. Based on the time synchronization mapping channel, point-by-point difference calculations are performed on each predicted time-series curve and the corresponding actual process state parameter time-series sequence to calculate the instantaneous absolute deviation and relative deviation between the predicted and measured values. The absolute deviation is calculated using the formula Δx = |xpredicted - xactual|, and the relative deviation is calculated using the formula δ = Δx / xactual × 100%. At the 20th minute of the reaction synthesis stage, the predicted temperature was 81℃, and the actual temperature was 80.5℃, with an absolute deviation Δx = |81-80.5| = 0.5℃ and a relative deviation δ = 0.5 / 80.5×100%≈0.62%. At the 30th minute, the predicted pressure was 0.34MPa, and the actual pressure was 0.33MPa, with an absolute deviation Δx = 0.01MPa and a relative deviation δ = 0.01 / 0.33×100%≈3.03%. The deviation values ​​of all 130 data points were calculated point by point, and the time series was integrated to generate a set of state deviation values ​​containing timestamps, parameter types, absolute deviations, and relative deviations. Time-frequency analysis was performed on the set of state deviations. A time-frequency transformation method was used to decompose the deviation signals of each parameter, extracting fluctuation patterns, trend components, and abrupt change characteristics at different time scales. The temperature deviation signal exhibited periodic fluctuations on a 10-20 minute time scale, with a fluctuation period of 5 minutes. The trend component showed a slow increase, with an abrupt change at the 45th minute, where the absolute deviation changed abruptly from 0.4℃ to 0.8℃. The pressure deviation signal fluctuated on a 5-10 minute time scale, with no obvious trend component and no abrupt change characteristics. Deviation signals with similar fluctuation patterns and trend characteristics were clustered: temperature and material viscosity deviation signals were clustered into one class, and pressure and isocyanate index deviation signals were clustered into another class, generating a set of deviation pattern feature vectors. A pre-trained inverse mapping network was used, which learned a nonlinear mapping relationship from typical deviation patterns to potential causes. Disturbance sources included raw material component fluctuations, catalyst deactivation, and temperature control loop disturbances; drift factors included stirring efficiency attenuation factors and heat transfer coefficient drift factors. The feature vector set of deviation modes is input into the inverse mapping network for solution, and the output is the probability of the existence of potential process disturbance sources and equipment performance drift factors corresponding to each deviation mode and the estimated quantitative contribution value to the current overall deviation. The probability of raw material component fluctuation is 75%, and the quantitative contribution is 40%; the probability of catalyst deactivation is 60%, and the quantitative contribution is 30%; the probability of stirring efficiency decay is 55%, and the quantitative contribution is 20%; the probability of heat transfer coefficient drift is 50%, and the quantitative contribution is 10%. All probability and quantitative contribution estimates are integrated to generate a deviation attribution analysis report.

[0033] Furthermore, step S4 includes the following steps: Step S41: Establish a deviation attribution scenario matching library, covering deviation scenarios and corresponding control strategies for various process disturbances and equipment drifts. Compare the attribution results in the deviation attribution analysis report with the deviation attribution scenario matching library to determine the attribution scenario and control direction corresponding to the current deviation. Step S42: Based on the current deviation attribution scenario and control direction, and combined with the real-time operating status of the digital twin model instance, generate real-time optimized control instructions for each core unit of the physical production line, including raw material feed rate adjustment instructions, reactor temperature and pressure control instructions, and equipment operating parameter correction instructions. Integrate all control instructions to generate a real-time optimized control instruction set. Step S43: Input the real-time optimized control instruction set into the digital twin model instance for virtual simulation verification, simulate the production line response effect after the control instructions are executed, and calculate the predicted deviation after the control instructions are executed; if the predicted deviation meets the preset control requirements, the real-time optimized control instruction set is sent to the distributed control system of the polyurethane production line to drive each piece of equipment on the physical production line to make dynamic adjustments according to the control instructions; if the predicted deviation does not meet the preset control requirements, the real-time optimized control instruction set is corrected based on the predicted deviation, and the virtual simulation verification steps are repeated until the predicted deviation meets the requirements; Step S44: Collect the actual response data of the production line after the execution of the corresponding control command that meets the requirements in real time, and feed the actual response data back to step S1 as a new round of real-time sensor data stream to update the state parameters of the digital twin model instance, start the next round of simulation, comparison, attribution and control process, and form a closed-loop intelligent control loop.

[0034] In this embodiment of the invention, a deviation attribution scenario matching library is established. This library covers various process disturbances and equipment drift, along with corresponding deviation scenarios and control strategies. For example, the control strategy for raw material component fluctuation deviation scenarios is to adjust the polyol feed rate; the control strategy for catalyst deactivation deviation scenarios is to replenish the catalyst; the control strategy for temperature control loop disturbances is to adjust the temperature control parameters; the control strategy for decreased stirring efficiency is to increase the stirring speed; and the control strategy for heat transfer coefficient drift is to clean the scale buildup on the reactor wall. By comparing the attribution results in the deviation attribution analysis report with the deviation attribution scenario matching library, it is determined that the current deviation is mainly attributed to raw material component fluctuations and catalyst deactivation, and the corresponding control directions are adjusting the polyol feed rate and replenishing the catalyst. Based on the current deviation attribution scenario and control direction, and combined with the real-time operating status of the digital twin model instance, the reaction synthesis stage temperature is 82℃, pressure is 0.33MPa, and material viscosity is 1600mPa·s. Real-time optimized control instructions are generated for each core unit of the physical production line. The raw material pretreatment unit feed rate adjustment instruction is to adjust the polyol feed rate from 55L / h to 58L / h. The reaction synthesis unit control instruction is to supplement the catalyst by 0.05% and adjust the temperature control target to 82.5℃. The equipment operating parameter correction instruction is to increase the stirring speed from 60r / min to 65r / min. All control instructions are integrated to generate a real-time optimized control instruction set. The real-time optimized control command set is input into the digital twin model instance for virtual simulation verification. The simulation step size is 1 minute, simulating the production line response effect after the control command is executed. The predicted deviation after the control command is executed is calculated. The absolute deviation of temperature is 0.3℃ and the absolute deviation of pressure is 0.005MPa, which meets the preset control requirements (absolute deviation ≤0.5℃, ≤0.01MPa). The real-time optimized control command set is sent to the distributed control system of the polyurethane production line, driving each piece of equipment on the physical production line to dynamically adjust according to the control command. The actual response data of the production line after the control command is executed is collected in real time. The temperature is 82.4℃ and the pressure is 0.325MPa. This actual response data is used as a new round of real-time sensor data stream feedback to step S1 to update the state parameters of the digital twin model instance. After correction, the temperature is 82.4℃ and the pressure is 0.325MPa. The next round of simulation, comparison, attribution and control process is started to form a closed-loop intelligent control loop.

[0035] Furthermore, step S43 includes the following steps: Extract the parameters of each control instruction from the real-time optimization control instruction set, clarify the execution object, execution sequence, adjustment range and execution duration of each control instruction, perform structured parsing of the control instruction parameters, and generate virtual control instructions that can be injected into digital twin model instances; Inject virtual control commands into the current digital twin model instance, keep other parameters of the model consistent with the real-time state of the physical production line, start virtual simulation verification, simulate the state changes of the digital twin model instance during the execution of control commands, and generate virtual response data; The time-series variation data of key process state parameters are extracted from the virtual response data, and a virtual response time-series curve is plotted. This curve is compared with the preset process standard time-series curve to calculate the predicted deviation after the execution of the virtual response control command. The distribution characteristics and time-series variation trend of the predicted deviation are analyzed to determine whether the predicted deviation is within the preset control requirements. If the predicted deviation meets the preset control requirements, the real-time optimized control command set is sent to the distributed control system of the polyurethane production line to drive each piece of equipment on the physical production line to dynamically adjust according to the control command. If it exceeds the range, the abnormal deviation and time-series node that exceed the control requirements and the corresponding virtual control command are marked, and the deviation node and the corresponding virtual control command are corrected based on the predicted deviation. The virtual simulation verification steps are repeated until the predicted deviation meets the requirements.

[0036] In this embodiment of the invention, for example, the real-time optimization control instruction set includes four core control instructions. The first instruction is a polyol feed rate adjustment instruction for the raw material pretreatment unit, which is executed on the feed pump of the raw material pretreatment unit. The execution time is from the 15th minute to the 55th minute of the reaction synthesis stage, and the adjustment range is to adjust the polyol feed rate from 55 L / h to 58 L / h. The execution duration is 40 minutes. The second instruction is a catalyst replenishment instruction for the reaction synthesis unit, which is executed on the catalyst addition device of the reaction synthesis unit. The execution time is from the 20th minute to the 21st minute of the reaction synthesis stage, and the adjustment range is to replenish 0 catalyst. The first command is a 0.05% adjustment, with an execution time of 1 minute. The second command is a temperature control command for the reaction synthesis unit, targeting the temperature control module of the reaction synthesis unit. The execution sequence is from the 20th to the 60th minute of the reaction synthesis stage, adjusting the temperature target from 82℃ to 82.5℃, with an execution time of 40 minutes. The third command is a stirring speed adjustment command for the reaction synthesis unit, targeting the stirring device of the reaction synthesis unit. The execution sequence is from the 10th to the 70th minute of the reaction synthesis stage, adjusting the stirring speed from 60 r / min to 65 r / min, with an execution time of 60 minutes. These four control commands are structured and analyzed, extracting the execution target code, execution start and end timestamps, adjustment range value, and execution duration for each command. These parameters are then encoded according to the preset format of the digital twin model, generating four injectable virtual control commands. Each virtual control command contains three core fields: object code, timing parameters, and adjustment parameters. Four virtual control commands are simultaneously injected into the current digital twin model instance. Parameters in the model, such as reactor temperature (82℃), pressure (0.33MPa), and material viscosity (1600mPa·s), are kept consistent with the real-time state of the physical production line. Virtual simulation verification is initiated with a simulation step size of 1 minute. The state changes of the digital twin model instance during the execution of the control commands are simulated, generating virtual response data covering 70 minutes. A set of key process state parameters, including reactor temperature, pressure, and material viscosity, is recorded every minute. The time-series changes in reactor temperature and pressure are extracted from the virtual response data, and virtual response time-series curves are plotted. The temperature response curve gradually increases from 82℃, reaching 82.3℃ at the 20-minute mark, and then 82.5℃ at the 60-minute mark, where it remains stable. The pressure response curve gradually decreases from 0.33MPa, dropping to 0.328MPa at the 20-minute mark and then to 0.325MPa at the 60-minute mark. The material viscosity response curve gradually increases from 1600mPa·s, reaching 1850mPa·s at the 70-minute mark.In the preset process standard time sequence curve, the standard temperature value is 82.5℃ and the standard pressure value is 0.33MPa. The predicted deviation between the virtual response time sequence curve and the process standard time sequence curve is calculated. At the 20th minute, the absolute temperature deviation ΔT = |82.3-82.5| = 0.2℃ and the absolute pressure deviation ΔP = |0.328-0.33| = 0.002MPa; at the 60th minute, the absolute temperature deviation ΔT = |82.5-82.5| = 0℃ and the absolute pressure deviation ΔP = |0.325-0.33| = 0.005MPa; at the 70th minute, the absolute temperature deviation ΔT = |82.5-82.5| = 0℃ and the absolute pressure deviation ΔP = |0.325-0.33| = 0.005MPa. The preset control requirements are absolute temperature deviation ≤ 0.5℃ and absolute pressure deviation ≤ 0.01MPa. The distribution characteristics and time-series change trends of the deviations were analyzed and predicted. The temperature deviation decreased from 0.2℃ to 0℃ and the pressure deviation increased from 0.002MPa to 0.005MPa, both of which were within the preset control requirements.

[0037] Step S24, the step of marking abnormal deviations and timing nodes exceeding control requirements and corresponding virtual control commands if the deviation exceeds the range, includes the following steps: Determine the deviation threshold range corresponding to the preset control requirements, clarify the upper and lower limits of the allowable deviation for each key process state parameter, and establish deviation judgment criteria; traverse the predicted deviation amount, compare the predicted deviation amount at each time point with the corresponding deviation threshold range, determine whether each deviation amount meets the preset control requirements, and mark the abnormal deviation amount that exceeds the threshold range and the corresponding time sequence node. Analyze the timing nodes and key parameter types corresponding to abnormal deviations, and in conjunction with the execution timing of virtual control commands, determine the target control command that caused the abnormal deviation, and clarify the correlation between the adjustment range, execution timing, and abnormal deviation of the target control command; specifically including: Based on the timing nodes corresponding to abnormal deviations, the virtual control instruction execution stage corresponding to each timing node is determined, and all control instructions being executed within this stage and their execution status are clarified. Based on the virtual response data of the digital twin model instance, the changing patterns of key process state parameters during the execution of each control instruction are analyzed, and a correlation model between control instructions and parameter changes is established. The virtual control instruction execution stage corresponding to the timing node is input into the correlation model to invert the control instruction that caused the abnormal change in the parameter, thus determining the target control instruction. The adjustment range and execution timing of the target control instruction are extracted, and the quantitative relationship between the adjustment range and execution timing of the target control instruction and the abnormal deviation is analyzed. The influence coefficient of the control instruction parameter change on the abnormal deviation is calculated. Based on the influence coefficient, the correlation strength and timing lag characteristics between the parameter adjustment of the target control instruction and the abnormal deviation are clarified, and the correlation between the adjustment range and execution timing of the target control instruction and the abnormal deviation is determined. Calculate the difference between the abnormal deviation amount and the deviation threshold to determine the extent of the deviation, generate abnormal deviation analysis data, and clarify the severity level and time propagation range of the abnormal deviation; Based on the abnormal deviation analysis data, determine whether the current real-time optimized control instruction set needs to be corrected. If it needs to be corrected, mark the control instruction that needs to be corrected and the direction of correction.

[0038] In this embodiment of the invention, by determining the deviation threshold range corresponding to the preset control requirements, the upper and lower limits of the allowable deviation of each key process state parameter are clarified, and a deviation judgment standard is established. The upper limit of the allowable temperature deviation is 0.5℃ and the lower limit is -0.5℃, and the upper limit of the allowable pressure deviation is 0.01MPa and the lower limit is -0.01MPa. The deviation judgment standard is that the deviation is within the upper and lower limits and meets the control requirements, while the deviation is abnormal if it exceeds the upper and lower limits. Iterate through all predicted deviations, comparing the predicted deviation at each time point with the corresponding deviation threshold range. For example, in the virtual simulation verification, at minute 50, the virtual temperature value is 83.1℃, and the standard temperature value is 82.5℃. The calculated absolute temperature deviation ΔT = |83.1 - 82.5| = 0.6℃, exceeding the allowable deviation limit of 0.5℃. At minute 50, the virtual pressure value is 0.342 MPa, and the standard pressure value is 0.33 MPa. The calculated absolute pressure deviation ΔP = |0.342 - 0.33| = 0.012 MPa, exceeding the allowable deviation limit of 0.01 MPa. Mark the abnormal deviations of temperature (0.6℃) and pressure (0.012 MPa) and the corresponding time point (minute 50). Analyze the time point (minute 50) and the key parameter types (temperature and pressure) corresponding to this abnormal deviation, and combine this with the execution timing of the virtual control commands to determine the target control command that caused the abnormal deviation. Based on the 50th minute of the time sequence, the virtual control command execution phase corresponding to this node is determined to be from the 20th to the 60th minute of the reaction synthesis stage. All control commands being executed during this phase are identified as the reaction synthesis unit temperature control command and the stirring speed adjustment command, with both continuously executing. Based on the virtual response data of the digital twin model instance, the temperature and pressure changes during the execution of the two control commands are analyzed. When the temperature control command adjusts the temperature from 82℃ to 82.5℃, the temperature deviation gradually increases from 0℃ to 0.2℃; when the stirring speed increases from 60 r / min to 65 r / min, the temperature deviation increases from 0.2℃ to 0.6℃, and the pressure deviation increases from 0.002 MPa to 0.012 MPa. A correlation model between the control commands and parameter changes is established. Inputting the execution phase corresponding to the 50th minute of the time sequence into the correlation model, the abnormal temperature and pressure deviations are found to be related to the reaction synthesis unit temperature control command and the stirring speed adjustment command, thus identifying the target control commands as the temperature control command and the stirring speed adjustment command.The adjustment range and execution sequence of the target control commands were extracted. The temperature control command's adjustment range was from 82℃ to 82.5℃, and the execution sequence was from the 20th to the 60th minute. The stirring speed adjustment command's adjustment range was from 60 r / min to 65 r / min, and the execution sequence was from the 10th to the 70th minute. The quantitative relationship between these two commands and the abnormal deviation was analyzed, and the influence coefficient of the control command parameter changes on the abnormal deviation was calculated. The formula for calculating the influence coefficient of the temperature control command is k1=ΔT / Δt, where ΔT is the difference between the abnormal deviation and the upper limit of the threshold (0.1℃), and Δt is the temperature adjustment amount (0.5℃). The calculated k1=0.1 / 0.5=0.2℃ / ℃. The formula for calculating the influence coefficient of the stirring speed adjustment command is k2=ΔP / Δn, where ΔP is the difference between the abnormal deviation and the upper limit of the threshold (0.002MPa), and Δn is the speed adjustment amount (5r / min). The calculated k2=0.002 / 5=0.0004MPa·min / r. Based on the influence coefficient, it is clear that the larger the adjustment range of the temperature control command, the larger the temperature deviation, with a correlation strength of 0.2℃ / ℃ and a time lag characteristic that the temperature deviation reaches its maximum within 8 minutes after the temperature adjustment; the larger the adjustment range of the stirring speed, the larger the pressure deviation, with a correlation strength of 0.0004MPa·min / r and a time lag characteristic that the pressure deviation reaches its maximum within 10 minutes after the speed adjustment, thus determining the correlation between the two. The difference between the abnormal deviation and the deviation threshold is calculated. The temperature difference is 0.6-0.5=0.1℃, and the pressure difference is 0.012-0.01=0.002MPa. The deviation is determined to exceed the thresholds of 0.1℃ for temperature and 0.002MPa for pressure. Abnormal deviation analysis data is generated, and deviations exceeding the thresholds of 0-0.1℃ and 0-0.002MPa are defined as mild anomalies. This abnormal deviation is classified as a mild anomaly. The time propagation range is from the 42nd minute to the 50th minute. During this period, the temperature deviation increases from 0.3℃ to 0.6℃, and the pressure deviation increases from 0.005MPa to 0.012MPa, showing a continuous upward trend. Based on the abnormal deviation analysis data, it is determined that the current real-time optimized control command set needs to be corrected. The control commands that need correction are the temperature control command for the reaction synthesis unit and the stirring speed adjustment command. The correction direction is to reduce the temperature control target and the stirring speed adjustment range to ensure that the temperature and pressure deviations are reduced to within the preset control requirements after correction.

[0039] Furthermore, the present invention also provides a digital twin-based intelligent control system for polyurethane production, including a processor, a memory, and a computer program stored in the memory and executable on the processor, for performing the digital twin-based intelligent control method for polyurethane production as described above.

[0040] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent control of polyurethane production based on digital twins, characterized in that, Includes the following steps: Step S1: Construct a digital twin simulation model corresponding to the entire polyurethane production process, and synchronously connect to the real-time sensor data stream corresponding to the production line, which includes real-time temperature, real-time pressure, material viscosity, isocyanate index and polyol feed rate in the reactor; dynamically assign values ​​to the initial state parameters of the digital twin simulation model based on the real-time sensor data stream to generate a digital twin model instance. Step S2: Based on the digital twin model instance, inject the preset production formula parameters and process setpoint sequence to perform forward-looking process simulation calculations and generate multi-physics prediction data covering the entire reaction cycle; A set of prediction time series curves for key process state parameters extracted from multiphysics prediction data; Step S3: Perform high-frequency dynamic comparison between the predicted time series curve set and the actual process state parameter time series collected from the production line in real time, and calculate the set of state deviations of each parameter between the digital twin virtual space and the physical entity space at each comparison time; based on the set of state deviations, solve the potential process disturbance sources and equipment performance drift factors that cause the deviations, and generate a deviation attribution analysis report. Step S4: Based on the deviation attribution analysis report, match the current deviation attribution scenario to generate a real-time optimization control instruction set for the physical production line; send the real-time optimization control instruction set to the distributed control system of the polyurethane production line for execution, driving the physical production line equipment to make dynamic adjustments; at the same time, feed back the actual response data of the production line after execution as a new round of real-time sensor data stream to step S1 to form a closed-loop intelligent control loop.

2. The intelligent control method for polyurethane production based on digital twins according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Identify the core units of the entire polyurethane production process, including the raw material pretreatment unit, reaction synthesis unit, curing and modification unit, molding and cutting unit, and quality inspection unit. Collect the equipment geometric parameters, material properties, process constraints, and historical production operation data of each unit to build the basic architecture of the digital twin simulation model and clarify the material transfer path, energy exchange relationship, and timing linkage logic between each unit. Step S12: Deploy a distributed sensor network, synchronously access the real-time sensor data streams of each core unit of the production line and perform real-time noise reduction processing, remove abnormal pulse data and time-series missing data, and generate a purified real-time sensor data stream. Step S13: Based on the purified real-time sensor data stream, extract the real-time operating status parameters of each core unit, establish parameter mapping relationships, assign the real-time status parameters of the physical production line to the corresponding nodes of the digital twin simulation model, and generate an initial digital twin model instance. Step S14: Collect the running status data of the initial digital twin model instance in real time, make a preliminary comparison with the purified real-time sensor data stream, calculate the assignment deviation of the initial state of the model, and dynamically correct the initial state parameters of the initial digital twin model instance based on the assignment deviation to generate the corrected digital twin model instance.

3. The intelligent control method for polyurethane production based on digital twins according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Based on the preset specifications of polyurethane products, determine the production formula parameters and the sequence of time process set points, clarify the parameter control range, timing nodes and switching conditions of each process stage, and structure the production formula parameters and the sequence of time process set points to generate process control instructions. Step S22: Inject process control commands into the digital twin model instance, start forward-looking process simulation calculation, simulate the material reaction process, energy transfer process and equipment operation process of the entire reaction cycle of polyurethane production, and generate multi-physics field prediction data covering the entire reaction cycle, including time-series change data of temperature field, pressure field, material concentration field and viscosity field in the reactor. Step S23: Select key process state parameters from the multiphysics prediction data, including real-time temperature prediction, real-time pressure prediction, material viscosity prediction, isocyanate index prediction, polyol feed rate prediction, and reaction conversion rate prediction, and extract the time series data of each key process state parameter in the whole reaction cycle. Step S24: Based on the time series data of each key process state parameter, draw the corresponding predicted time series curves, mark the time series nodes, parameter change trends and key feature points of each curve, and generate a set of predicted time series curves for key process state parameters. Step S25: Perform timing consistency verification on the predicted timing curve set, calculate the timing synchronization error between each curve, and correct the predicted timing curve set based on the timing synchronization error to ensure that the predicted timing of each key process state parameter is consistent with the actual production timing.

4. The intelligent control method for polyurethane production based on digital twins according to claim 3, characterized in that, Step S23 includes the following steps: Based on the reaction mechanism of polyurethane production, the types of key process state parameters that affect product quality and production stability are identified, the detection standards and time sequence collection requirements for each key parameter are clarified, and key parameter screening criteria are established. The multiphysics prediction data is traversed, and the time series data corresponding to each key process state parameter is extracted according to the key parameter screening criteria. Invalid prediction data and abnormal prediction data that are outside the process control range are removed to generate the original time series dataset of key parameters. The original time series dataset of key parameters is then subjected to time series interpolation to supplement the missing time series data, so that the time series data of each key process state parameter remains continuous, and a continuous key parameter time series dataset is generated. Based on the continuous key parameter time series dataset, the rate of change and time series fluctuation amplitude of each key process state parameter in different reaction stages are calculated, the change characteristics of each parameter are extracted, and a key parameter change characteristic dataset is generated. Based on the key parameter change feature dataset, the predicted time series data of each key process state parameter are matched, and the time series data of each key process state parameter within the entire reaction cycle are extracted.

5. The intelligent control method for polyurethane production based on digital twins according to claim 4, characterized in that, The process of calculating the rate of change and time-series fluctuation amplitude of each key process state parameter at different reaction stages based on a continuous key parameter time-series dataset, and extracting the change characteristics of each parameter includes the following steps: The continuous key parameter time series dataset is divided according to the reaction stages of polyurethane production, and the time interval and time length of each reaction stage are defined to generate a stage-specific key parameter time series dataset. Based on the stage-specific key parameter time series dataset, the time series change rate of each key process state parameter in each reaction stage is calculated, and the parameter change rate value at each moment is obtained through the time series difference algorithm to generate a key parameter stage-specific change rate dataset. Based on the key parameter phase change rate dataset, calculate the rate of change fluctuation of each key process state parameter in each reaction phase, determine the stability characteristics of each parameter change, and generate a key parameter fluctuation characteristic dataset. Based on the key parameter fluctuation feature dataset, the fluctuation peak, fluctuation frequency and time series distribution of each key process state parameter in different reaction stages are extracted to generate key parameter fluctuation feature indicators. By integrating the datasets of key parameter phased change rates, key parameter fluctuation characteristics, and key parameter fluctuation characteristics indicators, a dataset of key parameter change characteristics is generated, clarifying the temporal change patterns and fluctuation characteristics of each key parameter.

6. The intelligent control method for polyurethane production based on digital twins according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Establish a time synchronization mapping channel between each predicted time series curve and the actual process state parameter time series collected from the production line in real time; based on the time synchronization mapping channel, perform point-by-point difference calculation on each predicted time series curve and the corresponding actual process state parameter time series, calculate the instantaneous absolute deviation and relative deviation between the predicted value and the measured value, and integrate the time series to generate a set of state deviation quantities. Step S32: Perform time-frequency analysis on the set of state deviation quantities, extract the fluctuation patterns, trend components and abrupt change characteristics of each parameter deviation signal at different time scales, and cluster deviation signals with similar fluctuation patterns and trend characteristics to generate a set of deviation pattern feature vectors; Step S33: By pre-training an inverse mapping network containing perturbation sources and drift factors, the network learns a complex nonlinear mapping relationship between typical deviation patterns and a series of potential causes; the perturbation sources include at least raw material component fluctuations, catalyst deactivation, and temperature control loop disturbances; the drift factors include at least stirring efficiency decay factors and heat transfer coefficient drift factors. Step S34: Input the feature vector set of deviation modes into the inverse mapping network to solve and output the probability of the existence of potential process disturbance sources and equipment performance drift factors corresponding to each deviation mode and the estimated value of their quantitative contribution to the current overall deviation. Integrate all probability and quantitative contribution estimates to generate a deviation attribution analysis report.

7. The intelligent control method for polyurethane production based on digital twins according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Establish a deviation attribution scenario matching library, covering deviation scenarios and corresponding control strategies for various process disturbances and equipment drifts. Compare the attribution results in the deviation attribution analysis report with the deviation attribution scenario matching library to determine the attribution scenario and control direction corresponding to the current deviation. Step S42: Based on the current deviation attribution scenario and control direction, and combined with the real-time operating status of the digital twin model instance, generate real-time optimized control instructions for each core unit of the physical production line, including raw material feed rate adjustment instructions, reactor temperature and pressure control instructions, and equipment operating parameter correction instructions. Integrate all control instructions to generate a real-time optimized control instruction set. Step S43: Input the real-time optimized control instruction set into the digital twin model instance for virtual simulation verification, simulate the production line response effect after the control instructions are executed, and calculate the predicted deviation after the control instructions are executed; if the predicted deviation meets the preset control requirements, the real-time optimized control instruction set is sent to the distributed control system of the polyurethane production line to drive each piece of equipment on the physical production line to make dynamic adjustments according to the control instructions; if the predicted deviation does not meet the preset control requirements, the real-time optimized control instruction set is corrected based on the predicted deviation, and the virtual simulation verification steps are repeated until the predicted deviation meets the requirements; Step S44: Collect the actual response data of the production line after the execution of the corresponding control command that meets the requirements in real time, and feed the actual response data back to step S1 as a new round of real-time sensor data stream to update the state parameters of the digital twin model instance, start the next round of simulation, comparison, attribution and control process, and form a closed-loop intelligent control loop.

8. The intelligent control method for polyurethane production based on digital twins according to claim 7, characterized in that, Step S43 includes the following steps: Extract the parameters of each control instruction from the real-time optimization control instruction set, clarify the execution object, execution sequence, adjustment range and execution duration of each control instruction, perform structured parsing of the control instruction parameters, and generate virtual control instructions that can be injected into digital twin model instances; Inject virtual control commands into the current digital twin model instance, keep other parameters of the model consistent with the real-time state of the physical production line, start virtual simulation verification, simulate the state changes of the digital twin model instance during the execution of control commands, and generate virtual response data; The time-series variation data of key process state parameters are extracted from the virtual response data, and a virtual response time-series curve is plotted. This curve is compared with the preset process standard time-series curve to calculate the predicted deviation after the execution of the virtual response control command. The distribution characteristics and time-series variation trend of the predicted deviation are analyzed to determine whether the predicted deviation is within the preset control requirements. If the predicted deviation meets the preset control requirements, the real-time optimized control command set is sent to the distributed control system of the polyurethane production line to drive each piece of equipment on the physical production line to dynamically adjust according to the control command. If it exceeds the range, the abnormal deviation and time-series node that exceed the control requirements and the corresponding virtual control command are marked, and the deviation node and the corresponding virtual control command are corrected based on the predicted deviation. The virtual simulation verification steps are repeated until the predicted deviation meets the requirements.

9. The intelligent control method for polyurethane production based on digital twins according to claim 8, characterized in that, If the deviation exceeds the range, marking the abnormal deviation amount and timing node that exceed the control requirements and the corresponding virtual control command includes the following steps: Determine the deviation threshold range corresponding to the preset control requirements, clarify the upper and lower limits of the allowable deviation for each key process state parameter, and establish deviation judgment criteria; traverse the predicted deviation amount, compare the predicted deviation amount at each time point with the corresponding deviation threshold range, determine whether each deviation amount meets the preset control requirements, and mark the abnormal deviation amount that exceeds the threshold range and the corresponding time sequence node. Analyze the timing nodes and key parameter types corresponding to abnormal deviations, and in conjunction with the execution timing of virtual control commands, determine the target control command that caused the abnormal deviation, and clarify the correlation between the adjustment range, execution timing, and abnormal deviation of the target control command; specifically including: Based on the timing nodes corresponding to abnormal deviations, the virtual control instruction execution stage corresponding to each timing node is determined, and all control instructions being executed within this stage and their execution status are clarified. Based on the virtual response data of the digital twin model instance, the changing patterns of key process state parameters during the execution of each control instruction are analyzed, and a correlation model between control instructions and parameter changes is established. The virtual control instruction execution stage corresponding to the timing node is input into the correlation model to invert the control instruction that caused the abnormal change in the parameter, thus determining the target control instruction. The adjustment range and execution timing of the target control instruction are extracted, and the quantitative relationship between the adjustment range and execution timing of the target control instruction and the abnormal deviation is analyzed. The influence coefficient of the control instruction parameter change on the abnormal deviation is calculated. Based on the influence coefficient, the correlation strength and timing lag characteristics between the parameter adjustment of the target control instruction and the abnormal deviation are clarified, and the correlation between the adjustment range and execution timing of the target control instruction and the abnormal deviation is determined. Calculate the difference between the abnormal deviation amount and the deviation threshold to determine the extent of the deviation, generate abnormal deviation analysis data, and clarify the severity level and time propagation range of the abnormal deviation; Based on the abnormal deviation analysis data, determine whether the current real-time optimized control instruction set needs to be corrected. If it needs to be corrected, mark the control instruction that needs to be corrected and the direction of correction.

10. A digital twin-based intelligent control system for polyurethane production, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, for performing the intelligent control method for polyurethane production based on digital twins as described in any one of claims 1-9.

Images