Intelligent regulation and control system and method for chemical reaction process based on multi-mode online detection

Abstract

The invention relates to the technical field of chemical process control, in particular to an intelligent chemical reaction process regulation and control system and method based on multi-mode online detection, and the method comprises the steps: carrying out the Savitzky-Golay filtering and noise reduction processing of collected near infrared spectrum data, and screening characteristic wavelengths through employing a continuous projection algorithm SPA; jointly inputting the screened characteristic wavelength data and the real-time viscosity data into a pre-trained partial least squares PLS prediction model to obtain a real-time prediction value of the monomer conversion rate; the online self-adaptive mechanism is used for acquiring a latest'spectrum-granularity-viscosity-actual conversion rate 'quaternary data sample at preset time intervals, and carrying out online incremental updating on model parameters through a recursive PLS (Partial Least Square) algorithm; the control quantity is calculated by using a fuzzy PID control algorithm, and the stirring rate of the reaction kettle and the reactant feeding rate are dynamically adjusted. According to the method, comprehensive and accurate data support is provided for real-time regulation and control of the reaction process, and the prediction precision and long-term operation stability of the model are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of chemical process control technology, specifically to an intelligent control system and method for chemical reaction processes based on multimodal online detection. Background Technology

[0002] The accuracy of monitoring and the timeliness of control of chemical reactions are the core to ensuring product quality and production efficiency. In particular, polymerization reactions have extremely high technical requirements due to the complex dynamics of the system and the coupling of indicators.

[0003] Current chemical reaction monitoring and control relies on traditional technologies, which present significant bottlenecks:

[0004] 1. Offline detection has significant lag: The offline analysis and detection cycle of mainstream laboratories is 30-60 minutes, which cannot capture the reaction dynamics in real time, easily leading to reaction deviation and low product qualification rate (such as only about 86% for styrene-acrylic emulsion polymerization); frequent sampling also increases costs and damages the system's airtightness.

[0005] 2. The limitations of single online detection: Existing online detection technologies mostly rely on single-dimensional monitoring. For example, the polymerization process kinetic parameter adjustment method proposed in patent CN119989848A, although it optimizes model parameters through adaptive algorithms, relies on kinetic data and does not combine multi-source physical parameters such as spectroscopy and particle size. Single devices such as near-infrared spectroscopy cannot fully characterize the reaction. The spectrum is easily interfered with and only reflects molecular changes. Particle size analyzers and viscometers cannot capture functional group transformations. Even in some scenarios where multiple detection devices are combined, the lack of an effective data collaborative processing mechanism makes it difficult to form a comprehensive characterization of the reaction state and support precise control.

[0006] 3. Insufficient signal processing and modeling capabilities: Traditional spectral processing has poor noise reduction effects, feature selection relies on experience, and model redundancy is high; fixed offline models lack adaptive updates, with a monthly decay rate of 15-20%, and long-term prediction accuracy drops sharply (e.g., the error reaches 7.2% after 30 days in copolymerization reactions). Although some patents attempt to optimize model adaptability through algorithms, they have not been deeply integrated with multimodal detection data and are still difficult to match the dynamic control requirements of complex reactions.

[0007] 4. Lack of coordination in control strategies: Traditional PID is difficult to adapt to nonlinear reactions, and manual control is inconsistent; optimizing only a single indicator (such as conversion rate) and ignoring constraints such as particle size and energy consumption leads to poor product performance or excessive energy consumption.

[0008] In addition, insufficient sampling accuracy, transmission rate, and actuator control precision in some hardware components further amplify the error. Summary of the Invention

[0009] To address the problems existing in the prior art, the present invention provides an intelligent control system and method for chemical reaction processes based on multimodal online detection.

[0010] To achieve the above objectives, the technical solution of the present invention is as follows:

[0011] A smart control system for chemical reaction processes based on multimodal online detection includes a reaction vessel, a multimodal detection module, a data processing unit, a PLC controller, a stirring frequency converter, a solution pump, and an initiator storage tank. The reaction vessel is equipped with a stirring mechanism connected to the stirring frequency converter. The top of the reaction vessel has a reactant inlet and an initiator inlet, and the bottom of the reaction vessel has a discharge port. The initiator inlet of the reaction vessel is connected to the output end of the solution pump, and the input end of the solution pump is connected to the initiator storage tank. The multimodal detection module is mounted on the reaction vessel and includes a near-infrared spectral probe, a laser particle size analyzer, and a viscometer. The data processing unit includes a data acquisition card and an industrial switch. The data acquisition card is connected to the near-infrared spectral probe, the laser particle size analyzer, and the viscometer, and the industrial switch is connected to the data acquisition card. The PLC controller is connected to the industrial switch, the stirring frequency converter, and the solution pump.

[0012] Furthermore, the reactant inlet of the reactor is connected to a first pipeline, the initiator inlet of the reactor is connected to a third pipeline, and the discharge port of the reactor is connected to a second pipeline.

[0013] Furthermore, the detection wavelength range of the near-infrared spectral probe is 900-1700nm, the sampling interval is 0.5s, the measurement range of the laser particle size analyzer is 0.1-1000μm, and the measurement range of the viscometer is 0.1-10000mPa·s.

[0014] Furthermore, the power range of the stirring frequency converter is 0.75-75kW.

[0015] This invention also includes the following technical solutions:

[0016] A method for intelligent control of chemical reaction processes based on multimodal online detection using the above system includes the following steps:

[0017] (1) Near-infrared spectral data, particle size distribution data and viscosity data are collected synchronously during the chemical reaction process through a multimodal detection module;

[0018] (2) The near-infrared spectral data were subjected to Savitzky-Golay filtering for noise reduction, and the continuous projection algorithm SPA was used to screen characteristic wavelengths; at the same time, the validity of the spectral data was initially verified by combining the viscosity data collected at the same time.

[0019] (3) Input the selected characteristic wavelength data and real-time viscosity data into the pre-trained partial least squares PLS prediction model to obtain the real-time predicted value of monomer conversion rate.

[0020] (4) The PLS prediction model is maintained by an online adaptive mechanism. The online adaptive mechanism obtains the latest data samples of “spectrum-particle size-viscosity-actual conversion rate” every preset time interval, and applies a forgetting factor within a set range to update the model parameters online incrementally through the recursive PLS algorithm.

[0021] (5) Based on the real-time predicted value of the monomer conversion rate, the real-time measured particle size distribution data and the real-time viscosity data, the control quantity is calculated using the fuzzy PID control algorithm, and the stirring rate and reactant feeding rate of the reactor are dynamically adjusted so that the conversion rate and particle size distribution are maintained within the preset target range. The conversion rate, particle size distribution and energy consumption index are simultaneously optimized using a multi-objective optimization strategy.

[0022] Furthermore, in the online adaptive mechanism, the preset time interval is 5-15 minutes, and the forgetting factor ranges from 0.8 to 0.95.

[0023] Furthermore, the multi-objective optimization strategy adopts the NSGA-II algorithm, and the optimization objectives include conversion rate, particle size distribution and energy consumption. The optimal setpoint under the current operating conditions is calculated, and the optimal setpoint is used as the input target value of the fuzzy PID algorithm, wherein the conversion rate prediction error is ≤1.8%.

[0024] Furthermore, the proportional coefficient Kp of the fuzzy PID control algorithm is 0.1-10, the integral coefficient Ki is 0.01-1, and the derivative coefficient Kd is 0.001-0.1.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] This invention provides an intelligent control system and method for chemical reaction processes based on multimodal online detection. By integrating a near-infrared spectroscopy probe, a laser particle size analyzer, and a viscometer to construct a multimodal detection module, it simultaneously acquires multi-dimensional data on molecular structure, microparticle size, and macroscopic viscosity during the reaction process. Combined with a high-precision data acquisition card and an industrial switch, it achieves high-speed synchronous data transmission, completely solving the limitations of traditional single-detection technologies and the lag issues of offline detection. This provides comprehensive and accurate data support for real-time control of the reaction process. The system optimizes the signal processing flow through Savitzky-Golay filtering for noise reduction and SPA feature selection, and incorporates a training method based on sufficient offline samples with an online adaptive update mechanism. The PLS prediction model incorporates a forgetting factor ranging from 0.8 to 0.95 and a recursive PLS algorithm, effectively avoiding the accuracy decay defects of traditional fixed models and significantly improving the model's prediction accuracy and long-term operational stability. A control strategy combining the NSGA-II multi-objective optimization algorithm and fuzzy PID control algorithm is employed to simultaneously optimize key production indicators such as conversion rate, particle size distribution, and energy consumption. Combined with high-precision actuators such as stirring frequency converters and solution pumps, dynamic and precise adjustment of stirring and feeding rates is achieved, improving product qualification rates and shortening the polymerization reaction production cycle. Simultaneously, the multi-modal online detection mode significantly reduces the frequency of offline detection, comprehensively improving production efficiency and reducing production costs. Attached Figure Description

[0027] The embodiments of the present invention will be further described below with reference to the accompanying drawings, wherein:

[0028] Figure 1 A schematic diagram of an embodiment of an intelligent control system for chemical reaction processes based on multimodal online detection is shown.

[0029] Figure labels: 1-Reaction vessel, 2-Near-infrared spectroscopy probe, 3-Laser particle size analyzer, 4-Viscometer, 5-Data acquisition card, 6-Industrial switch, 7-PLC controller, 8-Stirring frequency converter, 9-Solution pump, 10-Initiator storage tank, G1-First pipeline, G2-Second pipeline, G3-Third pipeline. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0031] Reference Appendix Figure 1A smart control system for chemical reaction processes based on multimodal online detection includes a reaction vessel 1, a multimodal detection module, a data processing unit, a PLC controller 7, a stirring frequency converter 8, a solution pump 9, and an initiator storage tank 10. The reaction vessel 1 has an internal stirring mechanism connected to the stirring frequency converter 8. The top of the reaction vessel 1 has a reactant inlet and an initiator inlet, and the bottom of the reaction vessel 1 has a discharge port. The initiator inlet of the reaction vessel 1 is connected to the output end of the solution pump 9, and the input end of the solution pump 9 is connected to the initiator storage tank 10. The multimodal detection module is mounted on the reaction vessel 1 and includes a near-infrared spectral probe 2, a laser particle size analyzer 3, and a viscometer 4. The data processing unit includes a data acquisition card 5 and an industrial switch 6. The data acquisition card 5 is connected to the near-infrared spectral probe 2, the laser particle size analyzer 3, and the viscometer 4, and the industrial switch 6 is connected to the data acquisition card 5. The PLC controller 7 is connected to the industrial switch 6, the stirring frequency converter 8, and the solution pump 9.

[0032] In one embodiment of the present invention, the reactant inlet of the reactor 1 is connected to the first pipeline G1, the initiator inlet of the reactor 1 is connected to the third pipeline G3, and the discharge port of the reactor 1 is connected to the second pipeline G2.

[0033] In one embodiment of the present invention, the near-infrared spectral probe 2 has a detection wavelength range of 900-1700 nm and a sampling interval of 0.5 s, the laser particle size analyzer 3 has a measurement range of 0.1-1000 μm, and the viscometer 4 has a measurement range of 0.1-10000 mPa·s.

[0034] In one embodiment of the present invention, the power range of the stirring frequency converter 8 is 0.75-75kW.

[0035] A method for intelligent control of chemical reaction processes based on multimodal online detection using the above system includes the following steps:

[0036] (1) Near-infrared spectral data, particle size distribution data and viscosity data are collected synchronously through the multi-modal detection module during the chemical reaction process; among them, viscosity data is the core characterization parameter of the macroscopic physical state of the reaction system, which reflects the changes in the intermolecular forces of the reaction liquid in real time and provides a basic physical basis for the subsequent reaction process judgment.

[0037] (2) The collected near-infrared spectral data were subjected to Savitzky-Golay filtering for noise reduction, and the continuous projection algorithm SPA was used to screen characteristic wavelengths. At the same time, the validity of the spectral data was initially verified by combining the viscosity data collected at the same time. If the viscosity data showed abnormal changes (such as an increase of more than 50% in a short period of time), the spectral data of the corresponding time period was marked as suspicious data to improve the quality of subsequent modeling data.

[0038] (3) Input the selected characteristic wavelength data and real-time viscosity data into the pre-trained partial least squares PLS prediction model to obtain the real-time predicted value of monomer conversion rate; the introduction of viscosity data can make up for the deficiency of spectral data in reflecting the macroscopic polymerization degree of the system. Through the fusion of the two-dimensional features of "molecular structure signal-macroscopic physical state", the conversion rate prediction error is reduced to below 1.8%.

[0039] (4) The PLS prediction model is maintained through an online adaptive mechanism. The online adaptive mechanism obtains the latest data samples of “spectrum-particle size-viscosity-actual conversion rate” at preset time intervals, and applies a forgetting factor within a set range to update the model parameters online incrementally through the recursive PLS algorithm. Among them, viscosity data is a key indicator for verifying model stability. When the conversion rate predicted by the model deviates from the viscosity change trend, the model emergency update process is triggered to ensure the long-term accuracy of the model.

[0040] (5) Based on the real-time predicted value of monomer conversion rate, real-time measured particle size distribution data and real-time viscosity data, the control quantity is calculated using the fuzzy PID control algorithm, and the stirring rate and reactant feed rate of the reactor are dynamically adjusted to keep the conversion rate and particle size distribution within the preset target range. The conversion rate, particle size distribution and energy consumption index are optimized simultaneously using a multi-objective optimization strategy. In the control logic, viscosity data is used as the core constraint index. When the viscosity is lower than the lower limit of the preset range (indicating that the reaction process is lagging), the stirring rate is increased first and the initiator feed rate is increased. When the viscosity is higher than the upper limit of the preset range (indicating local overpolymerization or particle agglomeration), the stirring rate is appropriately reduced and the monomer feed rate is slowed down, so that the conversion rate, particle size distribution and viscosity are all kept within the preset target range.

[0041] In one embodiment of the present invention, in the online adaptive mechanism, the preset time interval is 5-15 minutes, and the forgetting factor ranges from 0.8 to 0.95.

[0042] In one embodiment of the present invention, the multi-objective optimization strategy adopts the NSGA-II algorithm, and the optimization objectives include conversion rate, particle size distribution and energy consumption. The optimal setpoint under the current operating conditions is calculated, and the optimal setpoint is used as the input target value of the fuzzy PID algorithm, wherein the conversion rate prediction error is ≤1.8%.

[0043] In one embodiment of the present invention, the proportional coefficient Kp of the fuzzy PID control algorithm is 0.1-10, the integral coefficient Ki is 0.01-1, and the derivative coefficient Kd is 0.001-0.1.

[0044] Example 1: Intelligent control of styrene-acrylic emulsion polymerization reaction

[0045] This embodiment uses styrene-acrylic emulsion polymerization as the control object, and the target product is a styrene-acrylic emulsion with a solid content of 45%. The core control indicators are: monomer conversion rate ≥95% (prediction error ≤1.8%), latex particle size distribution D50=50±5μm, and stirring energy consumption ≤1.2kWh / t.

[0046] 1. System Setup and Parameter Initialization

[0047] Reactor 1: Made of 316L stainless steel, with a volume of 1000L. The internal stirring mechanism is a three-layer anchor-type stirring paddle, connected to a stirring frequency converter 8 with a power range of 0.75-75kW, and the initial stirring speed is set to 500rpm. The top reactant inlet of reactor 1 is connected to the first pipeline G1, the initiator inlet is connected to the third pipeline G3, and the bottom outlet is connected to the second pipeline G2.

[0048] Multimodal detection module: Near-infrared spectroscopy probe 2 is installed on the top of reactor 1, with a detection wavelength range of 900-1700nm and a sampling interval of 0.5s; laser particle size analyzer 3 is connected to the sampling port on the side of reactor 1, with a measurement range of 0.1-1000μm; viscometer 4 is inserted and installed in the lower middle part of the side of reactor 1, with a range of 0.1-10000mPa·s, to monitor the viscosity of the reaction liquid in real time.

[0049] Data processing unit: Data acquisition card 5 synchronously connects to near-infrared spectroscopy probe 2, laser particle size analyzer 3, and viscometer 4, converting analog signals into digital signals and transmitting them to PLC controller 7 via industrial switch 6; PLC controller 7 preloads a partial least squares (PLS) prediction model (based on 600 sets of offline styrene-acrylic emulsion polymerization samples, R... 2 =0.986) and fuzzy PID algorithm (preset proportional coefficient Kp=5.0, integral coefficient Ki=0.5, derivative coefficient Kd=0.05).

[0050] Feeding system: Solution pump 9 is connected to initiator storage tank 10 and third pipeline G3, with an initial feed rate of 0.8L / min; first pipeline G1 is connected to monomer mixing tank, with an initial feed rate of 2.0L / min.

[0051] 2. Reaction process control and execution

[0052] Multi-source data synchronous acquisition: After the system is started, the multi-modal detection module synchronously acquires three sets of core data per second: near-infrared spectral probe 2 acquires spectral data of 2000 wavelength points, laser particle size analyzer 3 outputs latex particle D50 parameters, and viscometer 4 provides feedback on the viscosity value of the reaction liquid; after the data is converted by data acquisition card 5 and transmitted by industrial switch 6, it is uploaded to PLC controller 7 in real time, with a data transmission delay of ≤0.3s.

[0053] Spectral data preprocessing and validity verification: PLC controller 7 performs Savitzky-Golay filtering and noise reduction on the spectral data (window width 11 points, polynomial order 3), and then uses the continuous projection algorithm (SPA) to select 12 characteristic wavelengths (mainly corresponding to the absorption peaks of benzene ring and ester functional groups); at the same time, it combines the viscosity data collected by the viscometer 4 for verification - when the viscosity changes by more than 30% within 10 seconds, the spectral data collected by the corresponding near-infrared spectral probe 2 is marked as suspicious data and a backup sample is activated to ensure the validity of the input data.

[0054] Monomer conversion rate prediction: The characteristic wavelength data of the screened near-infrared spectroscopy probe 2 and the real-time viscosity data (initial viscosity of approximately 8 mPa·s) collected by the viscometer 4 were input into the PLS prediction model. The model outputs a real-time predicted value of monomer conversion rate through a two-dimensional correlation between "molecular structure signal and macroscopic physical state". Ten minutes after the reaction started, the predicted conversion rate was 65%, and the viscosity detected by the viscometer 4 rose to 25 mPa·s, which is consistent with the regular changes in the early stage of styrene-acrylic emulsion polymerization.

[0055] Online adaptive model update: As set according to claim 6, the preset time interval is 10 minutes, and the forgetting factor is 0.9. Every 10 minutes, the PLC controller 7 automatically extracts the latest 50 sets of quaternary samples: "near-infrared spectroscopy probe 2 data - laser particle size analyzer 3 data - viscometer 4 data - actual conversion rate" (actual conversion rate is calibrated through offline sampling), and updates the model parameters through the recursive PLS algorithm. After 60 minutes of reaction, the model prediction error stabilizes at 1.2% from the initial 0.9%, which is lower than the requirement of ≤1.8%.

[0056] Dynamic control and multi-objective optimization: The NSGA-II algorithm described in claim 7 is used for multi-objective optimization. The predicted conversion rate, the real-time D50 value of the laser particle size analyzer 3 (initial 42μm), and the real-time viscosity data of the viscometer 4 are input to calculate the optimal target setpoint. The fuzzy PID algorithm uses this setpoint as input to dynamically adjust the stirring rate and feed rate of the stirring frequency converter 8.

[0057] After 30 minutes of reaction, the predicted conversion rate was 82% (lower than the target of 95%). The viscosity measured by viscometer 4 was 48 mPa·s, and the D50 measured by laser particle size analyzer 3 was 45 μm. PLC controller 7 instructed stirring frequency converter 8 to increase the stirring speed to 600 rpm, solution pump 9 to increase the initiator feed rate to 1.0 L / min, and the monomer feed rate of the first pipeline G1 to 2.5 L / min.

[0058] After 50 minutes of reaction, the predicted conversion rate was 96%. Viscometer 4 detected that the viscosity had increased to 120 mPa·s, and laser particle size analyzer 3 detected D50 = 58 μm (out of ±5 μm range). After the NSGA-II algorithm calculated the optimal solution, the fuzzy PID command of the stirring frequency converter 8 reduced the stirring speed to 550 rpm, and the monomer feed rate of the first pipeline G1 was reduced to 1.8 L / min. After 3 minutes, laser particle size analyzer 3 detected that D50 had dropped back to 52 μm, and the conversion rate remained at 95.5%.

[0059] The reaction reached its endpoint after 60 minutes. Final testing showed a monomer conversion rate of 95.8%, a D50 of 51 μm as measured by laser particle size analyzer 3, a viscosity of 135 mPa·s as measured by viscometer 4, and a stirring energy consumption of 1.1 kWh / t corresponding to the stirring frequency converter 8. All indicators met the requirements. Compared with traditional manual control, the product qualification rate increased from 86% to 99.2%, and the production cycle was shortened by 15%.

[0060] Example 2: Intelligent Regulation of the Copolymerization Reaction of Methyl Methacrylate-Butyl Acrylate

[0061] This embodiment focuses on the copolymerization reaction of methyl methacrylate (MMA) and butyl acrylate (BA), with a monomer ratio of MMA:BA = 3:1. The target product is a transparent copolymer resin, and the control indicators are: conversion rate ≥ 94%, D50 = 40 ± 5 μm, and viscosity stable at 80-150 mPa·s.

[0062] 1. System Initialization and Model Preparation

[0063] The system configuration is the same as in Example 1. The PLC controller 7 loads the PLS model (R) trained based on 500 sets of MMA-BA co-aggregated offline samples. 2 =0.983), the initial settings of the fuzzy PID parameters are Kp=4.5, Ki=0.4, Kd=0.04, the model update interval is 15 minutes, and the forgetting factor is 0.85.

[0064] 2. Core Implementation Process

[0065] In the initial stage of the reaction, reactor 1 is heated to 85℃, the stirring frequency converter 8 is set to a stirring speed of 450 rpm, the monomer feed rate in the first pipeline G1 is 1.8 L / min, and the initiator feed rate in the solution pump 9 is set to 0.6 L / min. Key control points:

[0066] After 25 minutes of reaction: the predicted conversion rate was 78%, the viscosity measured by viscometer 4 was 55 mPa·s, and the D50 measured by laser particle size analyzer 3 was 36 μm. After optimization by the NSGA-II algorithm, the stirring frequency converter 8 increased the stirring speed to 550 rpm, and the solution pump 9 increased the initiator feed rate to 0.8 L / min, resulting in a rapid increase in conversion rate.

[0067] After 40 minutes of reaction: the predicted conversion rate was 95%, the viscosity measured by viscometer 4 was 155 mPa·s (exceeding the upper limit), and the D50 measured by laser particle size analyzer 3 was 46 μm. PLC controller 7 instructed the monomer feed rate of the first pipeline G1 to be reduced to 1.2 L / min, and the stirring frequency converter 8 reduced the stirring speed to 500 rpm. At the same time, a small amount of the reaction liquid was discharged through the second pipeline G2 to cool it down. After 5 minutes, the viscosity measured by viscometer 4 dropped to 140 mPa·s.

[0068] Model update: The model is updated every 15 minutes, and the prediction error after 60 minutes is 0.8%, which is far below the requirement of ≤1.8%; the error of the traditional fixed model during the same period has reached 7.2%.

[0069] At the 70-minute endpoint of the reaction, the conversion rate was 94.5%, the D50 of the laser particle size analyzer was 41 μm, the viscosity of the viscometer was 138 mPa·s, and the product transmittance was ≥92%. Compared with traditional offline detection and control, the detection frequency was reduced by 80%, the annual maintenance cost was reduced by 62%, and the material loss rate was reduced from 7.5% to 2.1%.

[0070] This invention provides an intelligent control system and method for chemical reaction processes based on multimodal online detection. By integrating a near-infrared spectroscopy probe, a laser particle size analyzer, and a viscometer to construct a multimodal detection module, it simultaneously acquires multi-dimensional data on molecular structure, microparticle size, and macroscopic viscosity during the reaction process. Combined with a high-precision data acquisition card and an industrial switch, it achieves high-speed synchronous data transmission, completely solving the limitations of traditional single-detection technologies and the lag issues of offline detection. This provides comprehensive and accurate data support for real-time control of the reaction process. The system optimizes the signal processing flow through Savitzky-Golay filtering for noise reduction and SPA feature selection, and incorporates a training method based on sufficient offline samples with an online adaptive update mechanism. The PLS prediction model incorporates a forgetting factor ranging from 0.8 to 0.95 and a recursive PLS algorithm, effectively avoiding the accuracy decay defects of traditional fixed models and significantly improving the model's prediction accuracy and long-term operational stability. A control strategy combining the NSGA-II multi-objective optimization algorithm and fuzzy PID control algorithm is employed to simultaneously optimize key production indicators such as conversion rate, particle size distribution, and energy consumption. Combined with high-precision actuators such as stirring frequency converters and solution pumps, dynamic and precise adjustment of stirring and feeding rates is achieved, improving product qualification rates and shortening the polymerization reaction production cycle. Simultaneously, the multi-modal online detection mode significantly reduces the frequency of offline detection, comprehensively improving production efficiency and reducing production costs.

[0071] The foregoing descriptions have outlined some exemplary embodiments of the present invention. It is understood that these embodiments are merely illustrative and do not constitute a limitation on the scope of protection of the present invention. Features in these embodiments can be rearranged in suitable ways, and the resulting solutions remain within the scope of protection claimed by the present invention. All other embodiments obtained by those skilled in the art based on the foregoing embodiments without inventive effort, i.e., all modifications, equivalent substitutions, and improvements made within the spirit and principles of this application, fall within the scope of protection claimed by the present invention.

Claims

1. A smart control system for chemical reaction processes based on multimodal online detection, characterized in that, The system includes a reaction vessel (1), a multimodal detection module, a data processing unit, a PLC controller (7), a stirring frequency converter (8), a solution pump (9), and an initiator storage tank (10). The reaction vessel (1) is equipped with a stirring mechanism connected to the stirring frequency converter (8). The top of the reaction vessel (1) has a reactant inlet and an initiator inlet, and the bottom of the reaction vessel (1) has a discharge port. The initiator inlet of the reaction vessel (1) is connected to the output end of the solution pump (9), and the input end of the solution pump (9) is connected to the initiator storage tank (10). The multimodal detection module is installed on the reactor (1) and includes a near-infrared spectral probe (2), a laser particle size analyzer (3), and a viscometer (4); the data processing unit includes a data acquisition card (5) and an industrial switch (6). The data acquisition card (5) is connected to the near-infrared spectral probe (2), the laser particle size analyzer (3), and the viscometer (4). The industrial switch (6) is connected to the data acquisition card (5); the PLC controller (7) is connected to the industrial switch (6), the stirring frequency converter (8), and the solution pump (9).

2. The intelligent control system for chemical reaction processes based on multimodal online detection according to claim 1, characterized in that, The reactant inlet of the reactor (1) is connected to the first pipeline (G1), the initiator inlet of the reactor (1) is connected to the third pipeline (G3), and the outlet of the reactor (1) is connected to the second pipeline (G2).

3. The intelligent control system for chemical reaction processes based on multimodal online detection according to claim 1, characterized in that, The detection wavelength range of the near-infrared spectral probe (2) is 900-1700nm, the sampling interval is 0.5s, the measurement range of the laser particle size analyzer (3) is 0.1-1000μm, and the range of the viscometer (4) is 0.1-10000mPa·s.

4. The intelligent control system for chemical reaction processes based on multimodal online detection according to claim 1, characterized in that, The power range of the stirring frequency converter (8) is 0.75-75kW.

5. A method for intelligent control of a chemical reaction process based on multimodal online detection using the system described in any one of claims 1-4, characterized in that, Includes the following steps: (1) Near-infrared spectral data, particle size distribution data and viscosity data are collected synchronously during the chemical reaction process through a multimodal detection module; (2) The near-infrared spectral data were subjected to Savitzky-Golay filtering for noise reduction, and the continuous projection algorithm SPA was used to screen characteristic wavelengths; at the same time, the validity of the spectral data was initially verified by combining the viscosity data collected at the same time. (3) Input the selected characteristic wavelength data and real-time viscosity data into the pre-trained partial least squares PLS prediction model to obtain the real-time predicted value of monomer conversion rate. (4) The PLS prediction model is maintained by an online adaptive mechanism. The online adaptive mechanism obtains the latest data samples of "spectrum-particle size-viscosity-actual conversion rate" every preset time interval, and applies a forgetting factor within a set range to update the model parameters online incrementally through the recursive PLS algorithm. (5) Based on the real-time predicted value of the monomer conversion rate, the real-time measured particle size distribution data and the real-time viscosity data, the control quantity is calculated using the fuzzy PID control algorithm, and the stirring rate and reactant feeding rate of the reactor are dynamically adjusted so that the conversion rate and particle size distribution are maintained within the preset target range. The conversion rate, particle size distribution and energy consumption index are simultaneously optimized using a multi-objective optimization strategy.

6. The intelligent control method for chemical reaction processes based on multimodal online detection according to claim 5, characterized in that, In the online adaptive mechanism, the preset time interval is 5-15 minutes, and the forgetting factor ranges from 0.8 to 0.

95.

7. The intelligent control method for chemical reaction processes based on multimodal online detection according to claim 5, characterized in that, The multi-objective optimization strategy adopts the NSGA-II algorithm. The optimization objectives include conversion rate, particle size distribution, and energy consumption. The optimal setpoint under the current operating conditions is calculated and used as the input target value of the fuzzy PID algorithm, wherein the conversion rate prediction error is ≤1.8%.

8. The intelligent control method for chemical reaction processes based on multimodal online detection according to claim 5, characterized in that, The proportional coefficient Kp of the fuzzy PID control algorithm is 0.1-10, the integral coefficient Ki is 0.01-1, and the derivative coefficient Kd is 0.001-0.1.

Images