Melt viscosity stability control method for coping with great load change of polyester device

Abstract

The invention discloses a melt viscosity stability control method for coping with great load change of a polyester device, and particularly relates to the technical field of chemical process control. Constructing a reaction rate model containing a catalyst activity parameter alpha (t) based on a polyester polycondensation kinetic mechanism; operating parameters such as temperature, vacuum degree and feeding flow are collected, and alpha (t) is dynamically calculated by combining a model predicted value and actually measured viscosity difference; when the alpha (t) continuously deviates from the preset threshold value, rate model updating or sub-model switching is triggered; the updated model is used for feed-forward adjustment of process parameters, and feedback correction is carried out in combination with a viscosity real-time detection result; the combined control instruction is output to a device control system to drive equipment such as a temperature control device, a vacuum system and a metering pump to complete adjustment; by means of the method, intelligent stable control over the melt viscosity of the polyester device under the unsteady-state working condition is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chemical process control, in particular to a melt viscosity stable control method for coping with large changes in the load of a polyester device. BACKGROUND

[0002] Melt viscosity stable control refers to keeping the flow viscosity of a melt within a relatively stable range by adjusting process parameters such as temperature, pressure, and shear rate during the melting processing of high molecular materials (such as plastics and rubber). This can ensure uniform flow of the material during the forming process such as extrusion, injection molding or spinning, and avoid size errors, inconsistent mechanical properties or surface defects of the product caused by viscosity fluctuations, thereby improving the quality of the product and the stability of the production.

[0003] During the long-term operation of a polyester device, the activity of the catalyst will gradually decrease over time, or uneven distribution will occur when the load of the device fluctuates greatly. However, existing mathematical models usually assume that the activity of the catalyst remains constant, which leads to a gradual disconnection between the calculated reaction rate coefficient and the actual working condition. When the actual activity of the catalyst decreases, the predicted condensation rate of the model is often higher than the actual level, which makes the feedforward prediction result deviate from the actual trend for a long time, and the operators make process adjustments based on the wrong prediction, which can easily lead to frequent or even excessive parameter adjustments, thereby exacerbating the instability of the melt viscosity. SUMMARY

[0004] The purpose of the present application is to provide a melt viscosity stable control method for coping with large changes in the load of a polyester device to solve the problems in the background art.

[0005] In order to achieve the above-mentioned purpose, the present application provides the following technical solution: a melt viscosity stable control method for coping with large changes in the load of a polyester device, comprising:

[0006] A kinetic rate model is constructed for the polyester condensation reaction, and the kinetic rate model contains a catalyst activity parameter a(t);

[0007] Process parameters including melt viscosity, intrinsic viscosity, vacuum degree, temperature, feed flow rate, and byproduct removal rate are collected and updated according to a set period;

[0008] The catalyst activity parameter a(t) is calculated, and the corrected a(t) parameter is calculated according to the difference between the predicted value of the kinetic rate model and the measured value of the operating data;

[0009] The corrected a(t) parameter and the corresponding operating parameters are archived to form a labeled parameter set;

[0010] When the corrected α(t) parameter deviates from the preset threshold condition, the dynamic rate model parameters are updated;

[0011] The updated kinetic rate model is used for feedforward adjustment of process parameters, and feedback adjustment is performed in combination with real-time melt viscosity detection results;

[0012] Output control commands, and output the results of feedforward control and feedback control to the control system of the polyester unit to control the operation of related equipment to adjust the process parameters.

[0013] Preferably, the dynamic rate model takes the form of: r represents the polycondensation reaction rate, k0 is the standard reaction rate constant, taken from the calibration value measured during the initial operation of the unit or under optimal catalyst activity; C A C B α(t) represents the instantaneous concentrations of the two reactants participating in the polycondensation reaction; m and n are the kinetic reaction orders of the reactants; and α(t) is the catalyst activity parameter.

[0014] Preferably, the data acquisition includes process parameters such as melt viscosity, intrinsic viscosity, vacuum degree, temperature, feed flow rate, and by-product removal rate, and includes: an adjustable time interval between 10 seconds and 5 minutes for sampling period; the acquired data is assigned scene tags, which include the current unit load range, raw material batch identifier, reactor segment information, and operating status code, used to construct a multivariate data structure.

[0015] Preferably, the calculation of the catalyst activity parameter α(t) includes:

[0016] The error data between the predicted viscosity value and the measured viscosity value is obtained, wherein the predicted viscosity value is output by a polyester polycondensation kinetic model that includes a catalyst activity factor, the measured viscosity value is obtained by an online viscosity detection device installed at the outlet of the polycondensation section, and the error data is obtained by calculating the difference within a sliding time window.

[0017] A multivariate nonlinear fitting method is used to estimate the correctable catalyst activity parameter α(t) based on the current output deviation. A dynamic objective function is constructed to minimize the weighted sum of squared residuals between the predicted and measured values, which is used to obtain the value of α(t) in the current period.

[0018] Preferably, the corrected α(t) parameter and the corresponding operating parameters are archived to form a tagged parameter set, including:

[0019] Obtain the α(t) parameter value for the current cycle and the operating data used in its calculation, including temperature, vacuum level, feed flow rate, load percentage, and raw material batch information;

[0020] Add a label field to the operating data. The label field includes the unit operating stage identifier, load change trend type, catalyst usage time interval, and reaction section number.

[0021] The α(t) parameter value and its corresponding tagged runtime data are packaged and stored in the historical parameter database.

[0022] Preferably, the step of updating the dynamic rate model parameters when the corrected α(t) parameter deviates from the preset threshold condition includes:

[0023] Within multiple consecutive sampling periods, the difference between α(t) and the baseline activity value is calculated. If the deviation exceeds the threshold range and continues to exceed the set number of periods, the model update judgment flag is triggered.

[0024] Once the trigger flag is set, retrain a portion of the dynamic coefficients in the rate model based on the historically archived labeled parameter set, or enable a pre-built sub-model version.

[0025] The updated dynamic rate model parameters are written into the main model module of the control system, replacing the original parameters to perform the next round of prediction and control calculations.

[0026] Preferably, the step of using the updated kinetic rate model for feedforward adjustment of process parameters and combining it with real-time melt viscosity detection results for feedback adjustment includes:

[0027] Using the temperature, vacuum level, feed flow rate, and α(t) value within the current cycle as model inputs, the predicted viscosity trend under the target operating conditions is calculated, and the optimal combination of process parameters is generated for feedforward adjustment.

[0028] The difference between the measured viscosity obtained by the online viscosity sensor deployed at the discharge port and the model predicted viscosity is analyzed to calculate the current deviation magnitude and trend, which is used to identify lag or abnormal behavior in the process response.

[0029] Based on the deviation between the predicted and measured values, the process parameters set by the feedforward output are corrected in real time, including polycondensation temperature, vacuum degree of the reaction section, metering pump speed or feed ratio, and the adjustment results are sent to the corresponding execution equipment through control commands.

[0030] Preferably, the step of outputting the feedforward control and feedback control results to the control system of the polyester unit to control the operation of relevant equipment to adjust the process parameters includes:

[0031] The feedforward adjustment results calculated by the updated kinetic rate model are integrated with the feedback correction parameters obtained by real-time viscosity deviation analysis to generate a joint control command set.

[0032] Based on the interface protocol types of different control devices, the joint control instruction set is subjected to protocol conversion and numerical format standardization processing, and then distributed to the corresponding execution device modules, including temperature control devices, vacuum regulating valve groups, feed pumps and metering pumps;

[0033] After the control command is sent, the execution status of each device is monitored. Confirmation signals or status data are collected through the feedback channel to determine whether the adjustment has been successfully implemented. If execution failure, delay or deviation occurs, the backup control logic is automatically triggered or the command is reissued.

[0034] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0035] 1. This invention introduces a custom catalyst activity parameter α(t) and combines it with dynamic rate model updates, tagged parameter archiving, feedforward-feedback closed-loop control, and command execution response mechanisms to achieve high-precision and robust control of melt viscosity in polyester plants under conditions of significant load fluctuations. Compared to traditional methods based on static models or single-path control, this invention can respond to changes in operating conditions in real time, dynamically correct model prediction biases, and improve the system's adaptability to unsteady-state reaction behavior.

[0036] 2. This invention also enhances the interpretability and scalability of the model by establishing a multi-dimensional data archiving and tagging mechanism to highly bind model parameters with operating conditions; by setting up a joint control instruction set and execution status monitoring mechanism, it ensures the closed-loop realization of process parameter adjustment at the physical equipment level, effectively reducing control deviations caused by equipment failure or response lag. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0038] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] For examples, please refer to Figure 1 As shown in this embodiment, the melt viscosity stabilization control method for dealing with large changes in the load of a polyester plant includes:

[0041] A kinetic rate model for polyester polycondensation reaction is constructed, and the kinetic rate model includes the catalyst activity parameter α(t);

[0042] The system collects process parameters including melt viscosity, intrinsic viscosity, vacuum level, temperature, feed flow rate, and by-product removal rate, and updates them according to a set cycle.

[0043] Calculate the catalyst activity parameter α(t), and calculate the corrected α(t) parameter based on the difference between the predicted value of the kinetic rate model and the measured value of the operating data.

[0044] The corrected α(t) parameters and their corresponding operating parameters are archived to form a tagged parameter set;

[0045] When the corrected α(t) parameter deviates from the preset threshold condition, the dynamic rate model parameters are updated;

[0046] The updated kinetic rate model is used for feedforward adjustment of process parameters, and feedback adjustment is performed in combination with real-time melt viscosity detection results;

[0047] Output control commands, and output the results of feedforward control and feedback control to the control system of the polyester unit to control the operation of related equipment to adjust the process parameters.

[0048] In this invention, to achieve stable control of melt viscosity in a polyester plant under conditions of significant load fluctuations, a kinetic model of polyester polycondensation reaction that can dynamically reflect the actual reaction state must first be constructed. This model introduces a custom catalyst activity parameter α(t) to correct the limitations of fixing the reaction rate constant k0 in traditional models.

[0049] The kinetic model focuses on the final polycondensation reaction of polyester, a typical polycondensation process whose reaction rate is significantly affected by factors such as temperature, reactant concentration, and catalyst activity. In traditional models, catalyst activity is generally considered constant, and the reaction rate constant k depends on empirical fitting or a single temperature function relationship. However, in the actual operation of polyester industrial plants, catalyst activity gradually decreases over time, and the distribution and reaction efficiency of the catalyst may dynamically change under conditions such as sudden load changes, unstable feed, and temperature disturbances. To compensate for this deficiency, this embodiment introduces a catalyst activity function α(t), enabling the reaction rate to dynamically respond to actual operating conditions.

[0050] The mathematical expression of the dynamic model is as follows: r represents the polycondensation reaction rate, k0 is the standard reaction rate constant, taken from the calibration value measured during the initial operation of the unit or under optimal catalyst activity; C A C B , respectively, are the instantaneous concentrations of the two main reactants participating in the polycondensation reaction; m and n are the kinetic reaction orders of the reactants, which can be obtained through linear fitting or nonlinear regression; α(t) is the catalyst activity parameter, which is a function of multiple coupled factors such as operating time t, unit load, temperature, vacuum degree, and feed composition, and is used to reflect the dynamic change trend of catalyst activity.

[0051] The introduction of α(t) allows the model to dynamically adjust the original rate constant k0 using the correction factor α(t) even when faced with catalyst activity decline or operational disturbances. α(t) can be expressed in one of the following ways or a combination thereof:

[0052] The exponential decay function form, for example, α(t) = e -βt , where β is the activity decay rate constant, which can be obtained by fitting historical operating data;

[0053] In piecewise function form, the constant value of α(t) is allocated according to the operating load range of the device (such as 30%–50%, 50%–80%, 80%–100%).

[0054] Polynomial functions or interpolation forms, such as Lagrange interpolation or polynomial regression, can be used to construct the mapping relationship between α(t) and the operating conditions;

[0055] Alternatively, an inversion algorithm based on real-time monitoring data can be used to calculate the current value of α(t) by using the error between the predicted and measured values ​​of melt viscosity.

[0056] In this invention, to achieve stable control of melt viscosity under conditions of significant load changes in the polyester unit, a model-driven mechanism based on multi-source real-time data needs to be established. Therefore, it is essential to systematically collect and manage key process parameters closely related to the polycondensation reaction. These parameters include melt viscosity, intrinsic viscosity, vacuum level, temperature, feed flow rate, and by-product removal rate, covering the key kinetic and mass transfer characteristics of the polyester polycondensation reaction.

[0057] First, multi-source process sensors are deployed at typical locations within the device to collect the aforementioned operating parameters in real time. These sensors are industrial-grade equipment suitable for high-temperature, high-viscosity conditions, and are positioned at the following typical locations: First, melt viscosity and temperature sensors are installed at the outlet of the final polycondensation reactor to reflect the immediate state of the polycondensation reaction; second, an online intrinsic viscosity analyzer is installed in the middle section of the melt pipeline to detect information related to polymer molecular weight; third, a pressure and temperature composite sensor is installed at the front end of the melt metering pump to obtain the feed status and fluctuation trends; and fourth, a flow meter and vacuum pressure gauge are installed on the byproduct ethylene glycol discharge pipeline to assess the removal efficiency of polycondensation byproducts. This deployment method ensures continuous spatiotemporal acquisition of key variables in the polycondensation process.

[0058] Secondly, to ensure the collected data has a sound analytical foundation, the implementation method sets the data collection operation to a fixed sampling period, preferably an adjustable interval between 10 seconds and 5 minutes, depending on the fluctuation characteristics of the device operation and response time requirements. Data collected within the sampling period is synchronously transmitted to the central data management system and structurally categorized according to parameter type. To further enhance the analytical applicability of the data, the collected data is automatically assigned scene tags after being transmitted to the system. These scene tags include the current device load range (e.g., full load, low load, overload), raw material batch number, reactor stage identifier (e.g., first condenser, final condenser, etc.), and operating status code (e.g., start-up, stable, switching, abnormal, etc.). These tags are used to establish the correspondence between parameter data and operating scenarios, supporting the requirements for scene identification and variable fitting in subsequent adaptive model calculations.

[0059] To ensure the quality and validity of the collected data, this implementation method introduces an anomaly detection mechanism. Specifically, dynamic upper and lower limit thresholds within historical statistical intervals are set for various collected parameters, and the system continuously compares and analyzes the sampled data using a sliding window. If a parameter exceeds its corresponding threshold interval three times consecutively, the system marks the parameter as "invalid data" and writes this marking information into the data header field. Invalid data will be automatically removed or assigned a low weight in subsequent dynamic inversion calculations of the catalyst activity parameter α(t) to reduce calculation deviations caused by local abnormal fluctuations or equipment noise.

[0060] After the collected data is filtered and labeled, it is input into the α(t) calculation module. This module inverts the catalyst activity correction parameters based on the error between the melt viscosity value predicted by the current model and the measured melt viscosity value. The inversion result is used to dynamically adjust the effective value of the standard reaction rate constant in the original kinetic model to reflect the actual reaction capacity of the catalyst under the current operating conditions. The periodically updated α(t) parameters are also automatically archived into the catalyst activity parameter database for long-term learning optimization and model adaptive iteration.

[0061] In this invention, to address the model mismatch problem caused by catalyst activity decay in polyester polycondensation reactions under long-term operation and significant load fluctuations, a method for calculating the catalyst activity parameter α(t) based on model prediction bias is proposed. The α(t) parameter is used to dynamically correct the standard reaction rate constant in the original kinetic rate model, thereby enhancing the model's responsiveness to changes in actual operating conditions. Specifically, the α(t) calculation method includes:

[0062] First, within each data sampling period, the predicted melt viscosity under the current operating condition is obtained from the constructed kinetic rate model. This predicted value is calculated based on real-time input variables such as temperature, reactant concentration, vacuum level, and feed flow rate, given an initial set value for the catalyst activity parameter (e.g., 1.0). Correspondingly, the measured viscosity data is obtained through an online melt viscosity monitoring device deployed at the outlet of the polycondensation section of the unit. The system compares the predicted viscosity value with the measured viscosity value to obtain an error value, which is recorded within a set sliding time window, constructing an error dataset with a time-series structure. The length of this time window is configurable, typically 5 to 30 minutes, used to capture continuous deviation behavior in the short term.

[0063] After obtaining the error dataset, the second step involves the inversion calculation of the catalyst activity parameter α(t). This process is based on a dynamic multivariate fitting method, aiming to adjust the value of α(t) so that, under the same model structure, the generated viscosity prediction value is as close as possible to the measured value. The system first defines an objective function based on the current error data, which measures the weighted sum of squares of the prediction bias. To improve the adaptability of the inversion results, the system incorporates label data from historical operating data similar to the current operating conditions for reference, such as similarity judgment in dimensions like temperature range, load level, and vacuum depth, to dynamically adjust the error weight coefficients. Subsequently, the objective function is solved using iterative calculation methods (such as gradient descent, quasi-Newton method, or genetic algorithm) to finally obtain the α(t) parameter value that minimizes the objective function.

[0064] It should be noted that, in order to avoid calculation bias caused by occasional errors or sensor malfunctions during this inversion process, the system introduces a data validity judgment mechanism. If the error value within a certain sampling period exceeds the abnormal threshold outside the historical statistical range, or shows a significant deviation trend that does not conform to physical laws within consecutive periods, the error data will be marked as invalid and will be given low weight or discarded during the inversion process.

[0065] After the initial estimation of α(t), the third step is initiated: a multi-point collaborative update mechanism. Due to the complex thermo-mechanical coupling and material gradient characteristics of the polyester polycondensation process, feedback from a single monitoring point may not fully reflect the global state of catalyst activity. Therefore, this invention further introduces a multi-point collaborative mechanism, incorporating error data from multiple physical quantities, including melt viscosity, intrinsic viscosity, reactor temperature gradient, and byproduct removal rate, into the inversion system. Each parameter is assigned a different weight based on its correlation with the reaction rate and then fed into a unified parameter fusion module.

[0066] In the fusion module, the α(t) values ​​from multiple monitoring points are integrated through weighted averaging, data collaborative filtering, or machine learning methods (such as random forest regression and principal component analysis) to ultimately generate a fusion parameter α(t) representing the overall catalyst activity state of the current system. This fusion result is written into the model control module, replacing the original static rate factor in subsequent viscosity prediction and process adjustment calculations. The system also archives and stores the α(t) parameter values ​​generated in this cycle, their calculation basis, tagged operating status, and error evaluation indicators, forming a historical catalyst activity database that can be used for model training and trend analysis.

[0067] In this invention, to enhance the system's ability to learn from the long-term operating status of the polyester unit and the model's adaptive capability, after completing the inversion calculation of the catalyst activity parameter α(t), a structured data archiving operation is further performed. The core objective of archiving is to establish a historical parameter database with time, operating condition, regional, and process tag characteristics to support the continuous iteration of subsequent model optimization, anomaly prediction, and process adjustment strategies.

[0068] Specifically, after completing the α(t) inversion in each data cycle, the system automatically extracts the original set of operating parameters associated with the α(t) calculation process. These operating parameters include, but are not limited to: the temperature inside the reactor, the vacuum level of the corresponding reaction section, the flow rate data of the polymer feed system, the percentage representation of the current operating load, and the identification number of the raw material batch. All of the data comes from a multi-source sensor system coupled to the α(t) model and has undergone data validity screening and outlier removal.

[0069] Next, based on the real-time status information of the process operation, the system automatically adds a set of predefined label fields to the above operating data. These label fields are used to identify the operating condition context of the current α(t) value, facilitating model recognition and subsequent retrieval. The label fields include the following:

[0070] Unit operation stage identifier: This field is used to identify which stage the current operation is in, such as start-up, stable operation, load switching, or shutdown. The system automatically assigns a value by judging the unit's operating time, output trend, and parameter fluctuation range.

[0071] Load change trend type: This field is used to characterize the current load change trend of the unit, which may include load increase, load decrease, load change, or load stability. The system calculates the trend type based on the relative change rate of feed flow and total load over multiple cycles.

[0072] Catalyst usage time interval: This field records the running time of the catalyst since its addition, and is used for subsequent analysis of the functional relationship between catalyst activity and usage time. The system calculates this by combining the catalyst addition timestamp and the current cycle timestamp.

[0073] Reaction segment number: This field indicates which reactor segment (e.g., first condenser, second condenser, or final condenser) corresponds to the α(t) value. It is used for region division when modeling multiple segments. The label information can be automatically labeled by the system according to the sensor installation location.

[0074] After automatically assigning values ​​to the label fields, the system packages and stores the generated α(t) value for that period, along with its associated operating parameters and all label data, into the historical parameter database. The database is organized in a structured table format and indexed by the period timestamp as the primary key to ensure temporal consistency of the data. Once stored, the data can be used for the following purposes: including but not limited to sample selection during model reconstruction, automatic identification of abnormal operating conditions, catalyst lifetime assessment, prediction deviation trend analysis, and training machine learning models for α(t) prediction.

[0075] In addition, to improve the scalability and long-term maintenance capabilities of the database, the system also supports periodic compression and tiered storage of archived data. High-frequency data can be retained for nearly three months in its entirety, while data exceeding this time limit is clustered by tags to retain statistical summary information and migrated to a cold data storage area.

[0076] In this invention, to further improve the adaptability of the kinetic rate model under long-term operation and drastic load fluctuations in polyester plants, a model update mechanism based on the deviation trend of the catalyst activity parameter α(t) is set up. When the system continuously detects that the value of α(t) deviates significantly from the originally set activity range, and this deviation has a certain persistence, the parameter update operation of the polyester polycondensation reaction kinetic rate model is triggered.

[0077] The process begins with a dynamic assessment of the α(t) deviation. The system acquires the current α(t) parameter value within each sampling period and compares it with the baseline setting (i.e., the catalyst activity value assumed in the initial calibration model, typically set to 1.00), calculating the absolute value of the deviation. To avoid unnecessary model updates due to occasional disturbances or sensor errors, the system employs a sliding window strategy to analyze the α(t) deviation over multiple consecutive sampling periods. If, within a set number of consecutive periods, such as 10 consecutive sampling periods (configurable), the deviation between α(t) and the baseline value consistently exceeds a set threshold range (e.g., ±10%), the system marks this state as "significant shift in catalyst activity" and automatically sets a model update judgment flag.

[0078] Once the determination flag is met, the system immediately enters the model parameter update phase. This phase includes two implementation paths: one is to retrain the existing rate model based on historical data; the other is to switch to a pre-built sub-model version.

[0079] In the first implementation path, the system calls upon the archived tagged parameter set to filter out historical operating samples with similar trends to the current α(t) value. These samples include information such as the operating temperature, vacuum level, reactor stage, feed batch, catalyst usage time, load change trend, and the recorded α(t) parameter value at that time. Based on these samples, the system recalculates some parameters in the kinetic rate model, such as the reaction order, activation term weight, or temperature correction factor, using nonlinear fitting algorithms (e.g., least squares method, gradient optimization method). The retrained model retains the original model framework, but its key coefficients have been corrected for catalyst activity decline or unsteady-state operation, making its output more consistent with the current actual operating conditions.

[0080] In the second implementation path, if the system has already established multiple sub-model versions applicable to different α(t) intervals (e.g., corresponding to typical states such as "initial catalytic activity," "mid-term activity decline," and "final activity failure"), then the matching sub-model can be automatically selected directly based on the current α(t) interval range, and this sub-model can be activated as the primary model for the control system. Compared to retraining, sub-model switching is faster in terms of response speed and is suitable for operational scenarios requiring real-time response. The sub-model can be established based on fitting long-term production data or constructed through expert experience.

[0081] Regardless of the update path used, the updated model parameters will be written into the main model module of the control system in the next stage. This module is the core operating unit of the kinetic control model, containing all mathematical expressions and coefficient values ​​used for melt viscosity prediction, reaction rate estimation, and feedforward control calculations. After writing the updated parameters, the system stops using the old parameters and starts the prediction logic based on the new parameters in the next sampling period. Simultaneously, it records the timestamp of this model switching event, the update method (retraining or sub-model switching), the original α(t) deviation value, and the trigger label. This information is also archived in the model version control database for subsequent performance backtracking, version management, or bug correction.

[0082] To avoid instability in the control system caused by frequent model switching, a model switching lockout period is also set. This means that within a lockout period after each parameter update (e.g., 30 minutes), the system will not respond to new α(t) offset trigger events unless an extreme operational anomaly is detected. This mechanism ensures the stability of model parameter updates and the consistency of the control logic.

[0083] In this invention, to achieve precise control of melt viscosity in a polyester unit under conditions of significant load fluctuations, the system employs a closed-loop control method that uses an updated kinetic rate model for feedforward adjustment of process parameters and combines this with real-time viscosity monitoring results for feedback correction. This method relies on a fusion mechanism of model prediction and online monitoring to establish a bidirectional adjustment pathway for melt viscosity stability.

[0084] After calculating the catalyst activity parameter α(t) and dynamically updating the rate model, the system first uses the updated kinetic rate model to perform feedforward prediction of melt viscosity. This model uses key operating parameters collected during the current sampling period as input variables, specifically: the reaction temperature of the polyester polycondensation section, the vacuum level of the reaction section, the actual feed flow rate of the polymer raw material, and the value of the α(t) parameter. By inputting these variables into the corrected rate model, the system can deduce the viscosity evolution trend under the current operating conditions based on the kinetic equation of the polyester polycondensation reaction. The predicted viscosity trend includes the target viscosity value, the direction of change, and the rate of change.

[0085] Based on the viscosity prediction results, the system further invokes optimization algorithms to combine and calculate adjustable process parameters such as temperature, vacuum level, pump speed, and feed ratio to generate the optimal process control strategy under the current target operating conditions. This strategy outputs recommended values ​​for the parameter set, which are used to correct the process setpoint in the feedforward control path. Without relying on real-time error information, the system can adjust process parameters in advance based solely on model prediction results to guide the melt viscosity towards the set target range.

[0086] Subsequently, the system initiates a feedback adjustment path to dynamically correct deviations in the feedforward output. The feedback path uses a high-temperature online viscosity sensor deployed at the outlet of the polycondensation section as its core data source, periodically collecting the actual viscosity value of the melt and comparing its real-time value with the model's predicted value. The system calculates the deviation between the predicted and measured viscosity and further analyzes the trend of this deviation over several sampling periods to determine whether the current process response exhibits lag, overshoot, or abnormal behavior. For example, if the model predicts a target viscosity, and the measured viscosity remains consistently lower than the predicted value and shows a downward trend, it may indicate insufficient polycondensation reaction intensity or an excessively low temperature setting; conversely, if the measured viscosity is significantly higher than the predicted value and rises rapidly, there may be a risk of excessive polycondensation or heat buildup.

[0087] Based on the above deviation analysis results, the system will perform dynamic correction operations on the current combination of process parameters. The correction logic includes the following aspects: when the measured viscosity is significantly lower than the predicted value, the system can appropriately increase the polycondensation temperature or increase the vacuum level to enhance the polycondensation reaction rate; when the viscosity is too high, the system can lower the temperature, slow down the reaction, or increase the ethylene glycol removal efficiency to regulate the reaction equilibrium state. Furthermore, the system can adjust the metering pump speed according to the degree of deviation to affect the material's reaction residence time, or fine-tune the feed ratio to control the reactant concentration ratio. These adjustments form a feedback correction pathway for the predicted deviation.

[0088] All the aforementioned adjustment operations are output to the interfaces of each target device through the control system's execution command module, including the reactor temperature control unit, vacuum control valve group, feed pump drive unit, and metering pump frequency conversion system. Control commands can be analog signals, electrical switching quantities, or communication protocol data streams, automatically adapting to the device type. In actual operation, the feedforward adjustment path and the feedback adjustment path operate in parallel. The system formulates a main strategy based on predicted trends and then corrects for feedback errors, achieving a stable and rapid response.

[0089] In addition, the system also sets up a steady-state identification mechanism and tolerance band parameters for the closed-loop control logic to prevent frequent adjustment behavior caused by small deviations. When the viscosity deviation is within the set tolerance range and the trend of change is stable, the system suspends feedback correction and only maintains feedforward control output; when a trend deviation is detected or the deviation continues to increase, the system automatically restores the feedback path to enhance the correction force.

[0090] In this invention, to ensure that the control strategy generated based on the updated kinetic rate model can be accurately executed to each execution unit of the polyester plant, the system sets up an integrated control command output mechanism to fuse the results of feedforward control and feedback control to generate joint control commands and complete the entire process control closed loop, including command distribution, protocol adaptation, execution confirmation and abnormal response.

[0091] First, based on the completion of feedforward prediction and feedback correction, the system logically integrates the data output from the two control paths. The feedforward adjustment results are derived from the optimized combination of process parameters based on the model's predicted viscosity trend, mainly including the temperature setpoint, target vacuum value, recommended metering pump speed, and feed ratio adjustment suggestions for the polycondensation reaction. The feedback control parameters are derived from the error analysis between the measured viscosity and the predicted viscosity in the current cycle; their correction values ​​reflect the degree of lag in the actual process response and the direction of deviation of the current adjustment effect. The system performs numerical weighting, priority judgment, and conflict resolution on both to generate the final joint control instruction set.

[0092] This instruction set is a structured data format, containing multiple control instruction fields. Each field includes a parameter name, target setpoint, priority level, expiration date, instruction source tag (feedforward or feedback), execution tolerance range, and redundancy checksum. The priority level is used to prioritize the execution of core parameters that significantly affect melt viscosity changes, such as instructions where the reaction temperature is higher than the metering pump speed setting, in cases of control conflicts or system resource constraints. The expiration date prevents expired instructions from being executed incorrectly; the system automatically cleans up unresponsive instructions that have exceeded their set lifespan.

[0093] Secondly, to ensure that the joint control commands can be correctly recognized and responded to by various actuators within the polyester plant, the system includes a protocol conversion and standardization adaptation module. Since there are various types of control devices in the polyester plant, such as heaters with analog temperature control, vacuum valves with digital control, and variable frequency pumps controlled by PLCs, their communication interface protocols may be Modbus, PROFIBUS, EtherCAT, CAN bus, or custom industrial communication protocols. Therefore, the system performs protocol parsing, data format conversion, and parameter value standardization on the control command set according to the communication interface type of each target device. For example, setting the temperature may need to be converted to a 4–20mA analog signal, and the metering pump speed may need to be converted to the frequency value received by the inverter. All conversion processes are completed automatically, ensuring consistency of commands at both the semantic and physical layers.

[0094] After protocol adaptation is complete, the system sends control commands to the control interfaces of each target execution device. Simultaneously, the system activates a command response monitoring mechanism to collect real-time execution status feedback information from the devices. This feedback information may include the device's current status code, actual response parameter values, signal confirmation responses (such as ACK / NAK), and operational alarm information. The system uses this feedback information to determine whether each control command has been successfully executed. If a control command does not receive a confirmation response within a specified time window, or if the execution result deviates from the set target by more than the allowable error range, the system considers the execution to have failed.

[0095] Upon detecting execution failures, response delays, or abnormal feedback data, the system will automatically trigger backup control logic. This logic includes one or a combination of the following strategies: 1) reissue control commands for the current cycle; 2) switch to a suboptimal parameter combination for redundant adjustment; 3) issue a manual intervention alert, requiring the operator to confirm on-site; 4) enter control protection mode according to the system safety policy, temporarily freezing some adjustment paths to prevent cascading fluctuations. All abnormal response events will be recorded in the operation log for subsequent system evaluation and continuous improvement.

[0096] To further improve execution efficiency and system reliability, this invention also supports adaptive optimization of instruction execution strategies. For example, if a certain type of device responds slowly for multiple cycles, the system will automatically reduce the frequency of adjustment for that device and extend the adjustment interval; if a device has had high execution accuracy in past operations, the system can increase the weighting coefficient of its parameter adjustment in future cycles to enhance the adjustment effect.

[0097] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for stabilizing melt viscosity in response to significant load changes in polyester plants, characterized by: include: A kinetic rate model for polyester polycondensation reaction is constructed, and the kinetic rate model includes the catalyst activity parameter α(t); The system collects process parameters including melt viscosity, intrinsic viscosity, vacuum level, temperature, feed flow rate, and by-product removal rate, and updates them according to a set cycle. Calculate the catalyst activity parameter α(t), and calculate the corrected α(t) parameter based on the difference between the predicted value of the kinetic rate model and the measured value of the operating data. The corrected α(t) parameters and their corresponding operating parameters are archived to form a tagged parameter set; When the corrected α(t) parameter deviates from the preset threshold condition, the dynamic rate model parameters are updated; The updated kinetic rate model is used for feedforward adjustment of process parameters, and feedback adjustment is performed in combination with real-time melt viscosity detection results; Output control commands, and output the results of feedforward control and feedback control to the control system of the polyester unit to control the operation of related equipment to adjust the process parameters.

2. The melt viscosity stabilization control method for dealing with large load changes in polyester plants according to claim 1, characterized in that: The dynamic rate model is in the form of: r represents the polycondensation reaction rate, k0 is the standard reaction rate constant, taken from the calibration value measured during the initial operation of the unit or under optimal catalyst activity; C A C B α(t) represents the instantaneous concentrations of the two reactants participating in the polycondensation reaction; m and n are the kinetic reaction orders of the reactants; and α(t) is the catalyst activity parameter.

3. The method for stabilizing melt viscosity in response to large load changes in polyester plants according to claim 1, characterized in that: The data collection includes process parameters such as melt viscosity, intrinsic viscosity, vacuum degree, temperature, feed flow rate, and by-product removal rate. The data collection period is an adjustable time interval between 10 seconds and 5 minutes. The collected data is assigned scene tags, which include the current unit load range, raw material batch identifier, reactor segment information, and operating status code, and are used to construct a multivariate data structure.

4. The melt viscosity stabilization control method for dealing with large load changes in polyester plants according to claim 3, characterized in that: The calculation of the catalyst activity parameter α(t) includes: The error data between the predicted viscosity value and the measured viscosity value is obtained, wherein the predicted viscosity value is output by a polyester polycondensation kinetic model that includes a catalyst activity factor, the measured viscosity value is obtained by an online viscosity detection device installed at the outlet of the polycondensation section, and the error data is obtained by calculating the difference within a sliding time window. A multivariate nonlinear fitting method is used to estimate the correctable catalyst activity parameter α(t) based on the current output deviation. A dynamic objective function is constructed to minimize the weighted sum of squared residuals between the predicted and measured values, which is used to obtain the value of α(t) in the current period.

5. The melt viscosity stabilization control method for dealing with large load changes in polyester plants according to claim 4, characterized in that: The corrected α(t) parameters and their corresponding operating parameters are archived to form a tagged parameter set, including: Obtain the α(t) parameter value for the current cycle and the operating data used in its calculation, including temperature, vacuum level, feed flow rate, load percentage, and raw material batch information; Add a label field to the operating data. The label field includes the unit operating stage identifier, load change trend type, catalyst usage time interval, and reaction section number. The α(t) parameter value and its corresponding tagged runtime data are packaged and stored in the historical parameter database.

6. The melt viscosity stabilization control method for dealing with large load changes in polyester plants according to claim 5, characterized in that: The step of updating the dynamic rate model parameters when the corrected α(t) parameter deviates from the preset threshold condition includes: Within multiple consecutive sampling periods, the difference between α(t) and the baseline activity value is calculated. If the deviation exceeds the threshold range and continues to exceed the set number of periods, the model update judgment flag is triggered. Once the trigger flag is set, retrain a portion of the dynamic coefficients in the rate model based on the historically archived labeled parameter set, or enable a pre-built sub-model version. The updated dynamic rate model parameters are written into the main model module of the control system, replacing the original parameters to perform the next round of prediction and control calculations.

7. The melt viscosity stabilization control method for dealing with large load changes in polyester plants according to claim 6, characterized in that: The step of using the updated kinetic rate model for feedforward adjustment of process parameters and combining it with real-time melt viscosity detection results for feedback adjustment includes: Using the temperature, vacuum level, feed flow rate, and α(t) value within the current cycle as model inputs, the predicted viscosity trend under the target operating conditions is calculated, and the optimal combination of process parameters is generated for feedforward adjustment. The difference between the measured viscosity obtained by the online viscosity sensor deployed at the discharge port and the model predicted viscosity is analyzed to calculate the current deviation magnitude and trend, which is used to identify lag or abnormal behavior in the process response. Based on the deviation between the predicted and measured values, the process parameters set by the feedforward output are corrected in real time, including polycondensation temperature, vacuum degree of the reaction section, metering pump speed or feed ratio, and the adjustment results are sent to the corresponding execution equipment through control commands.

8. The melt viscosity stabilization control method for dealing with large load changes in polyester plants according to claim 7, characterized in that: The step of outputting the feedforward and feedback control results to the control system of the polyester unit to control the operation of relevant equipment to adjust the process parameters includes: The feedforward adjustment results calculated by the updated kinetic rate model are integrated with the feedback correction parameters obtained by real-time viscosity deviation analysis to generate a joint control command set. Based on the interface protocol types of different control devices, the joint control instruction set is subjected to protocol conversion and numerical format standardization processing, and then distributed to the corresponding execution device modules, including temperature control devices, vacuum regulating valve groups, feed pumps and metering pumps; After the control command is sent, the execution status of each device is monitored. Confirmation signals or status data are collected through the feedback channel to determine whether the adjustment has been successfully implemented. If execution failure, delay or deviation occurs, the backup control logic is automatically triggered or the command is reissued.

Images