A real-time regulation method and system of CPD and DCPD and a storage medium

Abstract

The application provides a real-time regulation method and system of CPD and DCPD and a storage medium, and relates to the technical field of chemical process control. The method comprises the following steps: obtaining C5 process stream online analysis data and process variables, outputting component concentration estimation value and prediction uncertainty through a pre-calibrated multivariate correction model; combining the process variables to output component state and prediction trajectory through a kinetic state estimator; inputting the prediction controller to generate control instructions under process constraints, and regulating C5 process variables in real time in a closed loop. The method distinguishes the credibility of concentration estimation through prediction uncertainty, fuses process variables for kinetic state estimation and model prediction control rolling optimization, regulates C5 process variables in real time in a closed loop, and guarantees that the target values of CPD product indicators and DCPD internal and external isomer are stably achieved.

Description

Technical Field

[0001] This invention belongs to the field of chemical process control technology, specifically relating to a real-time control method, system, and storage medium for CPD and DCPD. Background Technology

[0002] In the deep processing of C5 hydrocarbons and the preparation of high-value-added chemicals, when recovering cyclopentadiene (CPD) from the process stream and preparing dicyclopentadiene (DCPD), it is necessary to simultaneously monitor the concentrations of multiple components, including cyclopentadiene and endogenous and exogenous dicyclopentadiene. Currently, industry mainly relies on offline chromatography for periodic sampling analysis or online spectroscopy to monitor single components or overall trends. However, for the aforementioned isomer system, since the components have the same molecular formula and similar boiling points and spectral characteristics, existing online detection methods struggle to balance resolution and response speed. They cannot simultaneously acquire real-time concentration information of each component and isomer, nor can they provide predictive uncertainty to support state estimation. This forces process control to rely on lagging adjustments based on offline analysis results or total quantity estimates, leading to a mismatch between control actions and the actual process state. This results in fluctuations in product purity, high energy consumption, and difficulties in suppressing byproducts, making it difficult to meet the requirements for refined control of isomer proportions and purity. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a real-time control method, system, and storage medium for CPD and DCPD to solve the aforementioned problems. This method distinguishes the reliability of concentration estimation by predictive uncertainty, integrates process variables for kinetic state estimation and model predictive control rolling optimization, and performs real-time closed-loop control of C5 process variables to ensure the stable achievement of CPD product indicators and DCPD endogenous and exogenous isomer target values.

[0004] To address the aforementioned technical problems, this invention provides a real-time control method for CPD and DCPD, comprising the following steps: Online analytical measurement data and process variable data were obtained from the C5 process stream containing cyclopentadiene CPD, internal DCPD and external DCPD; The online analytical measurement data is input into a pre-calibrated multivariate correction model, so that the multivariate correction model can perform quantitative analysis based on the online analytical measurement data and output the estimated CPD concentration, the estimated internal DCPD concentration, and the estimated external DCPD concentration, as well as the corresponding CPD prediction uncertainty, the internal DCPD prediction uncertainty, and the external DCPD prediction uncertainty. The current CPD product indicators and the current DCPD ino- and outo-type isomer indicators are determined based on the estimated CPD concentration, estimated endo-DCPD concentration, and estimated exo-DCPD concentration. The estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties, along with the process variable data, are input into a preset kinetic state estimator, which outputs the current component concentration state and the future concentration prediction trajectory. The current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and endo-isomer index are input to the preset model prediction controller, so that the preset model prediction controller uses the preset CPD product index value and the preset DCPD endo- and endo-isomer target value as control targets, performs rolling optimization solution under preset process constraints, and generates real-time control commands. The process variables of the C5 process are adjusted in real time based on the real-time control commands.

[0005] In the above scheme, online analytical measurement data and process variable data are acquired in real time. A pre-calibrated multivariate correction model outputs estimated CPD concentration, internal DCPD concentration, and external DCPD concentration based on the online analytical measurement data, along with their corresponding prediction uncertainties. The current CPD product index and the current DCPD inverse / exterior isomer index are determined based on these estimated values. The prediction uncertainties enable a preset kinetic state estimator to distinguish the reliability of each concentration estimate and, by integrating process variable data, output the current component concentration state and the predicted future concentration trajectory. Based on this, a preset model predictive controller, using preset CPD product index values ​​and preset DCPD inverse / exterior isomer target values ​​as control objectives, performs rolling optimization under preset process constraints based on the current component concentration state, the predicted future concentration trajectory, the current CPD product index, and the current DCPD inverse / exterior isomer index, generating real-time control commands. The real-time control commands adjust the process variables of the C5 process. The changed process state triggers a new round of online analysis and measurement, forming a closed-loop iteration, and dynamically ensuring the stable achievement of the preset CPD product index value and the preset DCPD endogenous and endomorphic isomer target value.

[0006] It should be noted that the current CPD product index is determined based on the estimated CPD concentration and at least one of the feed composition, feed flow rate, and product flow rate related to the CPD product index. The CPD product index includes CPD product purity and / or CPD yield. The current DCPD endo-exo-isomer index is determined based on the estimated concentrations of endo-DCPD and extrinsic DCPD. The DCPD endo-exo-isomer index includes the content and / or ratio of endo-DCPD and extrinsic DCPD. The process variable data may include at least one of the following: dimer reactor temperature, pyrolysis reactor temperature, feed flow rate, recycle ratio, reaction pressure, and material residence time, and participates as input parameters of the state transition equation in the pre-set kinetic state estimator for extrapolating the component concentration evolution trajectory.

[0007] Furthermore, the calibration process of the pre-calibrated multivariate correction model includes: Obtain a C5 process logistics calibration sample set, which includes CPD, internal DCPD and external DCPD; Based on the preset reference value determination algorithm, the component values ​​of each calibration sample in the C5 process logistics calibration sample set are determined to obtain the CPD reference concentration, internal DCPD reference concentration and external DCPD reference concentration in each calibration sample; Spectral data of each calibration sample were acquired to obtain the characteristic spectral data of each calibration sample. An initial training matrix is ​​constructed based on the CPD reference concentration, internal DCPD reference concentration, external DCPD reference concentration, and characteristic spectral data of each calibration sample. Based on the initial training matrix, model construction and training are performed to obtain the initial multivariate correction model and the initial concentration loading matrix; The initial multivariate calibration model was validated to obtain the validated multivariate calibration model and the validated concentration loading matrix. The verified multivariate correction model is determined as the pre-calibrated multivariate correction model, and the verified concentration loading matrix is ​​determined as the preset concentration loading matrix.

[0008] In the above scheme, a C5 process material calibration sample set including CPD, internal DCPD, and external DCPD is obtained to ensure that the training covers the actual process component distribution range. Based on a preset reference value determination algorithm, component values ​​are assigned to each sample to obtain accurate reference concentrations, providing reliable baseline true values ​​for model establishment. Spectroscopic acquisition is performed on each sample to obtain characteristic spectral data. An initial training matrix is ​​constructed by combining the reference concentration and characteristic spectral data to establish the correspondence between spectral features and component concentrations. Model building and training are performed based on the initial training matrix to obtain an initial multivariate calibration model and an initial concentration loading matrix, endowing the model with the ability to resolve complex spectral overlap signals. The initial multivariate calibration model is validated, and a validated multivariate calibration model and validated concentration loading matrix with stable predictive performance are selected as the pre-calibration model and preset concentration loading matrix. This ensures that during online analysis, accurate output of component concentration estimates and prediction uncertainties based on spectral data is possible, providing reliable input for closed-loop control.

[0009] It should be noted that, in the above model calibration process, the step of building and training the model based on the initial training matrix can specifically employ a preset full-spectrum partial least squares algorithm to establish a quantitative correlation between the characteristic spectral data and the CPD reference concentration, the in-type DCPD reference concentration, and the out-of-type DCPD reference concentration; the step of validating the initial multivariate calibration model can specifically employ a preset leave-one-out cross-validation algorithm to evaluate the model's prediction accuracy and robustness. The selection of the above algorithms is merely illustrative, and those skilled in the art can also employ other model building and validation methods that can achieve the same function according to actual application scenarios, all of which fall within the technical concept of this invention.

[0010] Further, the online analytical measurement data is input into a pre-calibrated multivariate correction model, enabling the multivariate correction model to perform quantitative analysis based on the online analytical measurement data, and output estimated CPD concentration values, estimated internal DCPD concentration values, and estimated external DCPD concentration values, as well as the corresponding CPD prediction uncertainties, internal DCPD prediction uncertainties, and external DCPD prediction uncertainties, including: Based on a preset spectral cutoff range, the spectral data in the online analysis and measurement data is cut off to obtain the target band spectral data. The target band spectral data is baseline corrected based on a preset baseline correction algorithm to obtain baseline corrected spectral data. The baseline-corrected spectral data is vector-normalized based on a preset vector normalization algorithm to obtain normalized spectral data. The normalized spectral data is smoothed using a preset smoothing algorithm to obtain preprocessed spectral data. The preprocessed spectral data is input into a pre-calibrated multivariate correction model, which performs quantitative analysis based on online analytical measurement data and outputs CPD concentration estimates, internal DCPD concentration estimates, and external DCPD concentration estimates, as well as the corresponding CPD prediction uncertainties, internal DCPD prediction uncertainties, and external DCPD prediction uncertainties.

[0011] In the above scheme, before inputting the online analytical measurement data into the pre-calibrated multivariate correction model, band truncation is first performed based on a preset spectral cutoff range. Target bands containing characteristic information of CPD, internal DCPD, and external DCPD are retained, while irrelevant noise is removed. Baseline correction is performed based on a preset baseline correction algorithm to eliminate baseline shifts caused by light source drift and probe contamination. Vector normalization is performed based on a preset vector normalization algorithm to eliminate multiplicative interference caused by laser power fluctuations and probe distance changes, ensuring spectral comparability. Smoothing is performed based on a preset smoothing algorithm to suppress high-frequency random noise while retaining effective signal trends. After the above preprocessing, the quality and consistency of the spectral data are significantly improved. After being input into the pre-calibrated multivariate correction model, the model can more stably and accurately resolve the concentration estimates of CPD, internal DCPD, and external DCPD and their prediction uncertainties, reducing prediction bias caused by spectral quality fluctuations.

[0012] It should be noted that the execution order of baseline correction, vector normalization, and smoothing in the above spectral preprocessing steps is not unique. In practice, the order of first performing band truncation, then baseline correction, then smoothing, and finally vector normalization can also be adopted, i.e., truncation, baseline, smoothing, normalization, to obtain preprocessed spectral data. Adjusting the order of the above preprocessing steps does not change the essence of the scheme based on online spectral analysis for quantitative concentration estimation, nor will it have a substantial impact on the evaluation logic of prediction uncertainty. Those skilled in the art can reasonably choose the preprocessing order based on the actual situation such as the noise characteristics of the spectral data and the degree of baseline drift.

[0013] Further, the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index are input to a preset model prediction controller. This causes the preset model prediction controller to perform rolling optimization under preset process constraints, using preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values ​​as control objectives, to generate control commands, including: The deviation of CPD product indicators is determined based on the preset CPD product indicator value and the current CPD product indicator, and the deviation of DCPD endo-endo isomer indicators is determined based on the preset DCPD endo-endo isomer target value and the current DCPD endo-endo isomer indicator. A multi-objective optimization function is constructed based on the CPD product index deviation and the DCPD endotype isomer index deviation, and the feasible region of the manipulated variables is determined based on the preset process constraints. Based on the current component concentration state, the future concentration prediction trajectory, and the feasible region of the manipulated variables, the multi-objective optimization function is solved to obtain the sequence of manipulated variables at the current time. The real-time control command is generated based on the set value of the manipulated variable corresponding to the current control step size in the manipulated variable sequence.

[0014] In the above scheme, during each control cycle, the preset model predictive controller, based on the updated current component concentration state and future concentration prediction trajectory, solves a multi-objective optimization function within a preset prediction time domain, with preset CPD product index values ​​and preset DCPD endo-isomeric target values ​​as control objectives, under process variable boundary constraints and oligomer generation and / or energy consumption constraints. First, the CPD product index deviation is determined based on the difference between the preset CPD product index value and the current CPD product index. Simultaneously, the DCPD endo-isomeric index deviation is determined based on the difference between the preset DCPD endo-isomeric target value and the current index, quantifying the degree of deviation of the process state from the control objectives. Based on these deviations, a multi-objective optimization function is constructed, unifying CPD purity control and DCPD isomer ratio control into a solvable optimization problem. The feasible region of the manipulated variables is determined based on preset process constraints, ensuring that the results meet physical limitations such as reactor temperature, flow rate, and residence time. Solving this optimization function based on the current component concentration state, future concentration prediction trajectory, and the feasible region of the manipulated variables yields the current time-to-time manipulated variable sequence, enabling control decisions to not only respond to current deviations but also anticipate future trends. The system executes the set value of the manipulated variable corresponding to the current control step size in the manipulated variable sequence to generate a real-time control command, precisely adjusts the C5 process variable, and makes each indicator gradually approach and stabilize at the preset target value; in the next control cycle, the system updates the current component concentration state and future concentration prediction trajectory based on the newly acquired online analysis and measurement data and process variable data, and repeats the above rolling optimization process to make each indicator gradually approach and stabilize at the preset target value.

[0015] Furthermore, the real-time adjustment of process variables in the C5 process based on the real-time control command includes: The CPD concentration estimate, internal DCPD concentration estimate, external DCPD concentration estimate, and their respective CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty are weighted and fused based on a preset weighted fusion algorithm to obtain the fused CPD concentration estimate, the fused internal DCPD concentration estimate, the fused external DCPD concentration estimate, and the corresponding global fused variance. For any component in CPD, internal DCPD and external DCPD, obtain the estimated concentration of the component and the corresponding prediction uncertainty in at least one online analysis channel; When there are concentration estimates for a component from at least two different online analysis channels, a weighted fusion is performed based on the prediction uncertainty and concentration estimates of each online analysis channel to obtain the fused concentration estimate and the variance after fusion for that component. When the concentration estimate of a component exists only in a single online analysis channel, the concentration estimate is used as the fusion concentration estimate of the component, and the fusion variance of the component is determined based on the prediction uncertainty of the online analysis channel. A fusion confidence interval is constructed based on the fusion variance of all components, and an action amplitude control strategy is generated based on the fusion confidence interval. The real-time control command is modified based on the aforementioned motion amplitude control strategy to obtain a modified real-time control command, and the process variables of the C5 process are adjusted in real time based on the modified real-time control command.

[0016] In the above scheme, for each component in CPD, internal DCPD, and external DCPD, a weighted fusion is performed based on the component concentration estimates and corresponding prediction uncertainties from at least two online analysis channels. Channels with lower uncertainties receive higher weights, while channels with higher uncertainties are suppressed, resulting in a fused concentration estimate and fused variance for the component. When a component has a concentration estimate from only a single online analysis channel, that concentration estimate is used as the fused concentration estimate for the component, and the fused variance is determined based on the prediction uncertainty of that channel. A corresponding fusion confidence interval is constructed based on the fused variance of each component, and the amplitude of the real-time control command is adjusted according to the fusion confidence interval. When the fused variance is large, i.e., the overall estimation confidence decreases, the control amplitude is automatically reduced to avoid drastic adjustments that could cause process fluctuations when measurement confidence is insufficient. This ensures that the strength of the control command matches the reliability of the measurement information, effectively preventing over-control due to single-channel failure or model drift, and improving robustness and control safety in multi-source heterogeneous data environments.

[0017] It should be noted that the online analysis channel refers to an independent analysis path based on different measurement principles or different physical installation points, including but not limited to Raman spectroscopy channel, attenuated total reflection Fourier transform infrared spectroscopy (ATR-FTIR) channel, low-field nuclear magnetic resonance (NMR) channel, and online gas chromatography (GC) channel; instruments installed at different process nodes under the same measurement principle are considered as different online analysis channels.

[0018] It should be noted that the control action amplitude in the aforementioned action amplitude control strategy can refer to the absolute value of the change in the set value of the manipulated variable obtained by the original optimization solution of the model predictive controller relative to the actual executed value of the previous control cycle, such as the absolute value of the temperature regulation amount or the absolute value of the flow regulation amount.

[0019] Furthermore, the real-time adjustment of process variables in the C5 process based on the real-time control command includes: Real-time acquisition of measured CPD content in the feed stream of the target process unit in the C5 process stream; The feedforward disturbance deviation is obtained based on the preset feed reference concentration and the measured value of the CPD content. The feedforward perturbation deviation and the current component concentration state are input into a preset reaction kinetic model so that the reaction kinetic model can perform perturbation propagation path deduction based on the feedforward perturbation deviation and the current component concentration state to obtain the expected concentration deviation trajectory. A compensation adjustment command is generated based on the expected concentration deviation trajectory. The real-time control command is optimized based on the compensation and adjustment command to obtain the optimized real-time control command, and the process variables of the C5 process are adjusted in real time based on the optimized real-time control command.

[0020] In the above scheme, the measured CPD content in the feed stream delivered to the target process unit is acquired in real time and compared with the preset feed baseline concentration to obtain the feedforward disturbance deviation. The disturbance is detected before the change in feed composition propagates downstream. The feedforward disturbance deviation and the current component concentration are input into a preset reaction kinetic model to deduce the propagation path of the disturbance in the dimerization, separation, and pyrolysis units, obtaining the expected concentration deviation trajectory and achieving advance prediction of the disturbance's impact. Compensation adjustment commands are generated based on the expected concentration deviation trajectory to make directional adjustments to process variables in advance. Real-time control commands are optimized based on the compensation adjustment commands, enabling the control action to initiate compensation before the feedback adjustment shows a deviation, shortening the system's response time to feed fluctuations, suppressing product purity fluctuations and isomer ratio drift caused by changes in feed CPD content, and improving the control stability of CPD product indicators and DCPD endo- and extrinsic isomer indicators under feed disturbance conditions.

[0021] It should be noted that the preset feed baseline concentration can be the arithmetic average of the CPD content of the feed logistics within 24 hours before the start of the current production shift, or it can be dynamically given by the upper-level production planning system based on the raw material tank analysis data.

[0022] It should be noted that the target process unit may refer to a dimerizing reactor.

[0023] Furthermore, the real-time adjustment of process variables in the C5 process based on the real-time control command further includes: Extract the characteristic band spectral vectors of oligomers from the online analysis and measurement data; Based on the pre-calibrated multivariate correction model, the oligomer characteristic band spectral vectors are quantitatively analyzed to obtain an estimated oligomer concentration. Based on the estimated oligomer concentration and the preset oligomer threshold, an oligomer exceedance determination is made. When the oligomer concentration is greater than or equal to the preset oligomer threshold, a clamping control command is generated based on the oligomer exceedance determination result. The real-time control command is forcibly modified based on the clamping and control command to obtain the modified real-time control command, and the process variables of the C5 process are controlled in real time based on the modified real-time control command.

[0024] In the above scheme, the characteristic band spectral vectors of oligomers are extracted from online analytical measurement data. Based on a pre-calibrated multivariate correction model, quantitative analysis of oligomers is performed to obtain estimated oligomer concentrations, ensuring that oligomer information is obtained synchronously with principal component analysis. The estimated oligomer concentration is compared with a preset oligomer threshold; when the oligomer concentration is greater than or equal to the preset threshold, it is determined to be an out-of-limit condition. Based on the oligomer exceedance determination result, a clamping control command is generated, bypassing conventional optimization logic to directly force correction of real-time control commands, such as forcibly reducing the reaction temperature or shortening the residence time, rapidly inhibiting further oligomer formation. By embedding oligomer monitoring and clamping control into a closed-loop circuit, a safety protection layer is added while controlling CPD product indicators and DCPD isomer indicators, preventing oligomer accumulation due to excessive pursuit of conversion rate or pyrolysis depth, thus avoiding increased downstream separation load and product quality deterioration.

[0025] Furthermore, the real-time control method for CPD and DCPD provided by the present invention further includes: NMR spectra were obtained from the low-field NMR flow cell in the C5 process stream based on a preset acquisition cycle. Characteristic peak analysis was performed based on the inner vinyl proton resonance peak and the outer vinyl proton resonance peak in the nuclear magnetic resonance spectrum to obtain the inner DCPD nuclear magnetic resonance measurement value and the outer DCPD nuclear magnetic resonance measurement value. The nuclear magnetic resonance isomer ratio was determined based on the inner DCPD nuclear magnetic resonance measurement value and the outer DCPD nuclear magnetic resonance measurement value. The proportion of spectral isomers is determined based on the estimated concentrations of the internal DCPD and the external DCPD, and a consistency check is performed based on the proportions of nuclear magnetic resonance isomers and the proportions of spectral isomers to obtain the proportion deviation value. When the proportional deviation value is greater than or equal to the preset internal correction threshold, a paired sample set is extracted from historical online analysis measurement data and nuclear magnetic resonance spectra based on the preset paired sample window; The pre-calibrated multivariate correction model is corrected and updated based on the paired sample set to obtain the updated multivariate correction model. The updated multivariate correction model is then used to estimate the multi-component concentrations of subsequent real-time online analytical measurement data.

[0026] In the above scheme, NMR spectra are acquired from a low-field NMR flow cell according to a preset acquisition cycle. Characteristic peaks are analyzed using the vinyl proton resonance peaks of both internal and external DCPDs to obtain their respective NMR measurements and determine the proportion of NMR isomers. Simultaneously, the proportion of spectral isomers is determined based on the concentration estimates output by a pre-calibrated multivariate correction model. The consistency between the two is verified to obtain the proportion deviation value, monitoring for model drift. When the proportion deviation value is greater than or equal to a preset internal correction threshold, it indicates that the model's prediction accuracy has deviated unacceptably. Based on a preset paired sample window, a paired sample set is extracted from historical online analysis measurement data and NMR spectra. The pre-calibrated model is then corrected and updated using the NMR measurements as the true benchmark, resulting in an updated multivariate correction model. This model adaptively compensates for prediction deviations caused by factors such as changes in raw materials and probe aging, ensuring that subsequent online analysis measurement data can still output accurate concentration estimates and maintaining the long-term reliability of closed-loop control.

[0027] The present invention also provides a real-time control system for CPD and DCPD, comprising: The data acquisition module is used to acquire online analytical measurement data and process variable data from the C5 process stream containing cyclopentadiene CPD, internal DCPD and external DCPD; The spectral analysis module is used to input the online analysis measurement data into a pre-calibrated multivariate correction model, so that the multivariate correction model can perform quantitative analysis based on the online analysis measurement data and output the CPD concentration estimate, the internal DCPD concentration estimate, and the external DCPD concentration estimate, as well as the corresponding CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty. The index determination module is used to determine the current CPD product index and the current DCPD endo- and endo-type isomer index based on the CPD concentration estimate, the endo-type DCPD concentration estimate, and the extrin-type DCPD concentration estimate. The state estimation module is used to input the estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties and the process variable data into a preset kinetic state estimator, and output the current component concentration state and the future concentration prediction trajectory. The optimization control module is used to input the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index to the preset model prediction controller, so that the preset model prediction controller uses the preset CPD product index value and the preset DCPD endo- and extrinsic isomer target value as control targets, performs rolling optimization solution under preset process constraints, and generates real-time control commands. The process control module is used to control the process variables of the C5 process in real time based on the real-time control command.

[0028] In the above scheme, the data acquisition module obtains online analytical measurement data and process variable data from the C5 process stream containing CPD, internal DCPD, and external DCPD, providing raw information input. The spectral analysis module inputs the online analytical measurement data into a pre-calibrated multivariate correction model, enabling the model to perform quantitative analysis and output estimated concentrations and prediction uncertainties for each component, achieving real-time conversion of spectral signals to concentration information. The index determination module determines the current CPD product index and the current DCPD internal and external isomer index based on the concentration estimates, quantifying the process state. The state estimation module inputs the concentration estimates, prediction uncertainties, and process variable data into a preset kinetic state estimator, outputting the current component concentration state and future concentration prediction trajectory, fusing measurement information with process kinetics. The optimization control module inputs the state, trajectory, and current index into a preset model predictive controller, enabling the controller to use preset CPD product index values ​​and preset DCPD internal and external isomer target values ​​as control objectives, generating real-time control commands through rolling optimization under process constraints. The process control module adjusts the C5 process variables in real time based on real-time control commands. The changed process state triggers a new round of measurement, forming a closed-loop iteration, and dynamically ensuring the stable achievement of the preset target.

[0029] The present invention also provides a computer-readable storage medium item, comprising: a stored computer program, wherein when the computer program is running, the device on which the computer-readable storage medium is located executes the steps of a real-time control method for CPD and DCPD as described in the present invention. Attached Figure Description

[0030] Figure 1 A schematic flowchart of a real-time control method for CPD and DCPD provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a real-time control system architecture for CPD and DCPD provided in an embodiment of the present invention. Detailed Implementation

[0031] 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, and 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.

[0032] Please see Figure 1 This embodiment provides a real-time control method for CPD and DCPD, including the following steps: Step S1: Obtain online analytical measurement data and process variable data from the C5 process stream containing cyclopentadiene CPD, internal DCPD and external DCPD; Step S2: Input the online analysis and measurement data into the pre-calibrated multivariate correction model so that the multivariate correction model can perform quantitative analysis based on the online analysis and measurement data, and output the estimated CPD concentration, the estimated internal DCPD concentration, and the estimated external DCPD concentration, as well as the corresponding CPD prediction uncertainty, the internal DCPD prediction uncertainty, and the external DCPD prediction uncertainty. Step S3: Determine the current CPD product index and the current DCPD endo- and endo-isomer index based on the estimated CPD concentration, estimated endo-DCPD concentration, and estimated extrinsic DCPD concentration. Step S4: Input the estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties and the process variable data into the preset kinetic state estimator, and output the current component concentration state and the future concentration prediction trajectory; Step S5: Input the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index into the preset model prediction controller, so that the preset model prediction controller uses the preset CPD product index value and the preset DCPD endo- and extrinsic isomer target value as control targets, performs rolling optimization solution under preset process constraints, and generates real-time control commands; Step S6: Adjust the process variables of the C5 process in real time based on the real-time control command.

[0033] In one embodiment, a method for real-time control of cyclopentadiene and dicyclopentadiene is provided. Hereinafter, cyclopentadiene is abbreviated as CPD, and dicyclopentadiene as DCPD. This method is applied to a C5 process flow comprising a dimerization reactor, a separation unit, a pyrolysis reactor, and a CPD refining unit. The main process flow is as follows: C5 feedstock is fed into the dimerization reactor via the feed unit for dimerization. The reaction product enters the separation unit, where a DCPD-containing stream is separated by distillation. This DCPD stream is then fed into the pyrolysis reactor for high-temperature pyrolysis to generate crude CPD. The crude CPD is then purified in the CPD refining unit to obtain the final CPD product. Simultaneously, a portion of the material from the CPD refining unit is returned to the dimerization reactor as a CPD recirculation stream to improve the feedstock conversion rate.

[0034] Multiple online analytical measurement points are configured in the process flow to acquire online analytical measurement data and process variable data in real time. These online analytical measurement points are sequentially arranged along the material flow direction of the C5 process stream at the dimer reactor outlet, separation unit feed or tower top, pyrolysis reactor inlet or outlet, and CPD product pipeline. Each sampling point can be equipped with at least one of a Raman spectrometer, a low-field nuclear magnetic resonance analyzer, or an online gas chromatograph to cover the component concentration monitoring needs of the entire process, including dimerization, separation, pyrolysis, and purification. The process variable data may include at least one of the following: dimer reactor temperature, pyrolysis reactor temperature, feed flow rate, recycle ratio, reaction pressure, and material residence time. These data serve as input parameters for the state transition equation and are used by the kinetic state estimator to extrapolate the component concentration evolution trajectory.

[0035] In this embodiment, a first Raman spectrometer is installed at the outlet of the dimerizing reactor to monitor the CPD conversion rate and initial endomorphic to ectomorphic ratio of the material at the reactor outlet in real time. A second Raman spectrometer is installed at the feed of the separation unit or at the top of the column to monitor the purity and composition of DCPD entering the separation unit. A third Raman spectrometer is installed at the inlet or outlet of the pyrolysis reactor to monitor the endomorphic to ectomorphic ratio before pyrolysis and the CPD yield after pyrolysis. Each Raman spectrometer directly acquires the Raman spectra of the material at the corresponding location using an online probe or a side-flow probe. In addition, a low-field nuclear magnetic resonance (NMR) analyzer is installed in the DCPD stream bypass to periodically provide highly selective endomorphic to ectomorphic DCPD ratio verification data; an online gas chromatograph is installed on the CPD product line to periodically provide authoritative reference data for model calibration. Process variable data include at least the dimerizing reactor temperature, feed flow rate, recycle ratio, and pyrolysis reactor temperature.

[0036] Each Raman spectrometer continuously acquired the Raman spectra of the material at intervals of 2 to 30 seconds, obtaining wavenumbers in the range of 200 cm⁻¹. -1 Up to 3200cm -1The original spectral data is obtained, and then spectral truncation is performed on the original spectral data based on a preset target band to obtain the characteristic band spectral vector. The preset target band covers the conjugated diene carbon-carbon double bond stretching vibration region of CPD, endo-dicyclopentadiene (endo-DCPD), and exo-dicyclopentadiene (exo-DCPD), i.e., the C=C stretching vibration region: the C=C stretching vibration peak of CPD is located at 1625 cm⁻¹. -1 Up to 1635cm -1 The C=C stretching vibration peak of norbornene in the endo-DCPD is located at approximately 1652 cm⁻¹. -1 The C=C stretching vibration peak of norbornene in the morphology-DCPD is located at approximately 1645 cm⁻¹. -1 The conjugated C=C stretching vibration of oligomers or tricyclopentadiene TCPD is located at 1580 cm⁻¹. -1 Up to 1600cm -1 The internal-DCPD and external-DCPD exhibit different Raman scattering cross-sections due to differences in molecular configuration, displaying distinct and independent characteristic responses in the aforementioned stretching vibration region. Furthermore, their spectral contributions to the CPD are clearly distinguishable. Therefore, the extracted characteristic bands cover at least 1580 cm⁻¹. -1 Up to 1700cm -1 In this embodiment, a 1000cm section is preferably selected. -1 Up to 1700cm -1 The band contains complete fingerprint area information.

[0037] The extracted spectral vectors are preprocessed, including baseline correction and vector normalization. Baseline correction employs an adaptive iterative reweighted penalized least squares method with the following parameters: polynomial order 7, maximum number of iterations 30, and penalty coefficient λ = 10. 5 Weight threshold ε=10 -4 The convergence criterion is that the L2 norm of the baseline difference between two adjacent iterations is less than 10. -3 Vector normalization employs unit vector normalization, meaning that for a spectral vector x, its L2 norm ||x||² is calculated, and then the normalized vector x' = x / ||x||² is set to eliminate multiplicative interference caused by probe distance, laser power fluctuations, or material concentration differences. The preferred preprocessing procedure is: the original spectrum is truncated to 1000 cm⁻¹ using a cosine window. -1 Up to 1700cm -1 For each band, the baseline is subtracted using the adaptive iterative reweighted penalized least squares method, followed by 5-point Savitzky-Golay smoothing using a second-order polynomial, and finally unit vector normalization to obtain the preprocessed characteristic band spectral vector.

[0038] The preprocessed characteristic band spectral vectors are input into a pre-calibrated multivariate correction model. This multivariate correction model is a Partial Least Squares (PLS) model, which uses 12 latent variables and a preset concentration loading matrix. Through regression projection, the estimated concentrations of CPD, internal-DCPD, and external-DCPD are simultaneously calculated. Based on a preset prediction uncertainty estimation algorithm, the prediction uncertainties for each component are output: CPD prediction uncertainty, internal-DCPD prediction uncertainty, and external-DCPD prediction uncertainty. The preset prediction uncertainty estimation algorithm can be constructed based on the prediction residual statistics of PLS ​​regression. In this embodiment, the prediction performance of the multivariate correction model is as follows: the root mean square error of prediction (RMSEP) for internal-DCPD is no greater than 1.5% by weight, the RMSEP for external-DCPD is no greater than 1.8% by weight, and the RMSEP for CPD is no greater than 1.2% by weight.

[0039] Based on the obtained estimated CPD concentration, estimated endo-DCPD concentration, and estimated exo-DCPD concentration, the current CPD product index and the current DCPD endo-exo-isomer index are determined. The current CPD product index is CPD purity or yield, and the current DCPD endo-exo-isomer index is the concentration ratio of endo-DCPD to exo-DCPD or their respective purities. The CPD yield in the current CPD product index is calculated based on at least one of the estimated CPD concentration at the pyrolysis reactor outlet, the measured CPD content in the feed stream, the feed flow rate, and the CPD product flow rate, rather than being directly determined by a single-point CPD concentration estimate, to reflect the actual conversion efficiency under material balance.

[0040] Subsequently, the estimated CPD concentration, internal-DCPD concentration, external-DCPD concentration, and their corresponding prediction uncertainties, along with the process variable data, are input into a preset kinetic state estimator. By introducing the prediction uncertainties, this preset kinetic state estimator can distinguish the reliability of each concentration estimate and, by integrating the process variable data, output a more reliable current component concentration state and future concentration prediction trajectory. This preset kinetic state estimator can be constructed based on extended Kalman filtering or unscented Kalman filtering, and uses the kinetic model of the C5 process to recursively estimate and predict the state.

[0041] Next, the current component concentration status, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index are input into a preset model predictive control (MPC). The MPC uses preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values ​​as control objectives, and performs rolling optimization solutions under preset process constraints to generate real-time control commands. The preset process constraints include the dimer reactor temperature range, feed flow rate range, recycle ratio range, and pyrolysis temperature range.

[0042] Finally, the process variables of the C5 process are adjusted in real time based on the aforementioned real-time control commands. Specifically, the process variable adjustments include: adjusting the temperature control valve of the dimerization reactor to maintain the dimerization reaction temperature within the range of 20°C to 120°C, thereby controlling the selectivity of endo- and extrinsic DCPDs; adjusting the feed pump to change the C5 feed flow rate, thereby controlling the reaction residence time and DCPD conversion rate; adjusting the circulation ratio adjustment valve to change the CPD circulation loop flow rate; and adjusting the temperature control valve of the pyrolysis reactor to maintain the pyrolysis temperature within the range of 140°C to 220°C, preferably 160°C to 200°C, to optimize the CPD regeneration yield. The changed process state after adjustment will trigger a new round of online analysis and measurement, thus forming a closed-loop iteration that dynamically ensures the stable achievement of preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values.

[0043] During long-term continuous operation, the low-field NMR analyzer and online GC analyzer periodically provide highly selective validation and reference data to correct the concentration loading matrix of the multivariate calibration model or update the prediction uncertainty estimation algorithm, thereby maintaining the accuracy and reliability of the model predictions. Optionally, each Raman spectrometer can also be simultaneously configured with an attenuated total reflection Fourier transform infrared (ATR-FTIR) probe to acquire data at 4000 cm⁻¹. -1 Up to 650cm -1 The mid-infrared spectrum is used to provide complementary chemical information. As a dependent embodiment, when encountering conditions with severe laser drift, the baseline correction step can be switched to an asymmetric least squares method with parameters set to λ=10. 6 Asymmetric weights p=0.001, or multivariate scattering correction instead of normalization, are used to enhance the method's environmental adaptability. Thus, the concentrations and confidence intervals of each component in the C5 process stream can be obtained in real time without relying on chromatographic separation, providing immediate and reliable feedback for process variable adjustments. This overcomes the lag of traditional offline analysis and achieves precise, targeted control of the purity of the target CPD product and the proportion of DCPD isomers.

[0044] In another embodiment, a real-time control method for CPD and DCPD is implemented through a four-layer progressive control system. The system consists of a measurement layer, a processing layer, a model layer, and a control layer from bottom to top. Each layer communicates with the other through unidirectional data transmission and adjustment commands issued via feedback loops, forming a continuous closed-loop automatic control.

[0045] The measurement layer, acting as the on-site hardware sensing layer, acquires online analytical measurement data and process variable data from the C5 process stream containing CPDs, internal DCPDs, and external DCPDs. The measurement layer is equipped with an online Raman probe, an attenuated total reflection Fourier transform infrared spectroscopy (ATR-FTIR) probe, a low-field nuclear magnetic resonance (NMR) analyzer, an online gas chromatograph (GC) analyzer, and temperature and flow sensors. The online Raman probe is immersed and installed at the outlet of the dimer reactor, the feed or top of the separation unit, and the inlet or outlet of the pyrolysis reactor to acquire the raw Raman spectral signal of the material in real time. The ATR-FTIR probe is configured synchronously with the Raman probe, acquiring wavenumbers in the range of 4000 cm⁻¹. -1 Up to 650cm -1 The system includes mid-infrared spectroscopy; a low-field NMR analyzer performs qualitative and quantitative analysis via a bypass pipeline, providing periodic verification data on the ratio of highly selective internal and external DCPDs; an online GC analyzer performs chromatographic purity testing on CPD products or DCPD streams via a bypass pipeline, providing periodic authoritative reference data for model calibration; and temperature and flow sensors collect basic process parameters such as reactor temperature and feed flow rate as process variable data.

[0046] The processing layer receives all raw signals from the measurement layer, completes spectral preprocessing and chemometric quantitative conversion, and determines the current product specifications. Raman spectra are truncated according to a preset target band to obtain characteristic band spectral vectors. The preset target band covers the conjugated diene carbon-carbon double bond stretching vibration region of CPD, internal DCPD, and external DCPD. Internal and external DCPDs have different Raman scattering cross-sections due to differences in molecular configuration, exhibiting distinct and independent characteristic responses in this stretching vibration region, and their spectral contributions to CPD are clearly distinguishable. The truncated characteristic band covers at least 1580 cm⁻¹. -1 Up to 1700cm -1 The preferred area is 1000cm. -1 Up to 1700cm -1The spectral bands contain complete fingerprint region information. A preprocessing process is performed on the truncated spectral vectors, including adaptive iterative reweighted penalized least squares baseline correction, Savitzky-Golay (SG) smoothing, and unit vector normalization to eliminate multiplicative interference caused by probe distance, laser power fluctuations, or material concentration differences. The preprocessed characteristic band spectral vectors are input into a pre-calibrated multivariate correction model. The model uses 12 latent variables and a predefined concentration loading matrix. Through regression projection, the estimated CPD, internal DCPD, and external DCPD concentrations are simultaneously calculated. Simultaneously, based on a predefined prediction uncertainty estimation algorithm, the prediction uncertainties for each component are output, namely, CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty. The predefined prediction uncertainty estimation algorithm is constructed based on the prediction residual statistics of PLS ​​regression. The processing layer also integrates validation and reference data provided by low-field NMR and online GC to verify the spectral quantification results, remove abnormal and distorted data, and output reliable real-time component concentration data. Furthermore, the processing layer determines the current CPD product index and the current DCPD endo-endo-isomeric isomer index based on the estimated CPD concentration, the estimated endo-DCPD concentration, and the estimated extrinsic DCPD concentration, and uploads the current CPD product index, the current DCPD endo-endo-isomeric isomer index, along with the estimated concentrations of each component and their prediction uncertainties, to the model layer.

[0047] The model layer incorporates a kinetic state estimator based on CPD and DCPD reaction networks. This estimator receives estimated CPD, internal DCPD, and external DCPD concentrations, along with their corresponding prediction uncertainties, from the processing layer. It also receives the process variable data and outputs the current component concentration state and future concentration prediction trajectories. The kinetic state estimator, combined with real-time component data, calculates implicit state parameters within the reactor that cannot be directly measured. These implicit state parameters include real-time reaction rate, equilibrium conversion rate, and intermediate product inventory. It further refines the estimated values ​​for each component, distinguishing between internal DCPD, external DCPD, and oligomer impurity concentrations. Due to the introduction of prediction uncertainties, the kinetic state estimator can differentiate the reliability of each concentration estimate. The model layer also integrates spectral detection errors and kinetic estimation errors through an uncertainty quantification module, calculating the fluctuation range of each estimate to provide the control layer with state and component information with confidence intervals, preventing abnormal data from misleading control.

[0048] The control layer is equipped with a Model Predictive Control (MPC) controller. The current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer indexes output from the model layer are input to the MPC controller. The MPC controller uses preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values ​​as control objectives, and performs rolling optimization solutions under preset process constraints to generate real-time control commands. The controlled variables include CPD purity or yield, endo-DCPD content, oligomer formation, and energy consumption. The manipulated variables include dimer reactor temperature, dimer reactor feed flow rate to control residence time, recycle ratio to the dimer reactor, and pyrolysis reactor temperature. Based on the real-time control commands, the temperature control valves of the dimer reactor, feed pump, recycle ratio adjustment valve, and pyrolysis reactor are adjusted to directionally suppress oligomer side reactions, increase CPD yield, and stabilize the ratio of endo- and extrinsic DCPD isomers. The control layer also includes an alarm and diagnostic module. When component concentrations exceed limits, sensor signals are abnormal, model estimation deviations are too large, or oligomer formation exceeds thresholds, an alarm is automatically triggered and the source of the fault is diagnosed. These fault sources include probe contamination, temperature deviation, or instrument drift. Adjustment commands output from the control layer are sent to the corresponding actuators in the measurement layer. After changing the process conditions, data is re-acquired and transmitted upwards layer by layer, forming a closed-loop iteration. The temperature of the dimerization reactor is set to 20℃ to 120℃, and the temperature of the pyrolysis reactor is set to 140℃ to 220℃, preferably 160℃ to 200℃. Therefore, the concentrations and confidence intervals of each component in the C5 process stream can be obtained in real time without relying on chromatographic separation, providing immediate and reliable feedback for process variable adjustments. This overcomes the lag of traditional offline analysis and achieves precise directional control of the purity of the target CPD product and the proportion of DCPD isomers.

[0049] In another embodiment, a real-time control method for CPD and DCPD is realized through a model predictive control closed-loop system that integrates feedforward dynamic prediction and multi-analytical instrument feedback. The system includes a setpoint input module, an online spectral feedback module, a reference calibration module, a preset dynamic state estimator, a deviation comparison node, and an MPC controller, forming a closed-loop control mechanism with multiple signal links working together.

[0050] Operators or the upper-level optimization system input process setpoints into the closed-loop system. These setpoints include the target CPD purity, target DCPD conversion rate, and upper limit of oligomer impurities. Simultaneously, online analytical measurement data and process variable data are acquired from the C5 process stream containing CPD, internal DCPD, and external DCPD. The online analytical measurement data is acquired in real-time by an online Raman probe, specifically Raman spectra of materials from process nodes such as dimerizing reactors, separation units, or pyrolysis reactors. After spectral interception and preprocessing, the data is input into a pre-calibrated multivariate correction model. Quantitative analysis is performed based on the online analytical measurement data, simultaneously outputting estimated CPD concentration, estimated internal DCPD concentration, and estimated external DCPD concentration, along with their corresponding prediction uncertainties—namely, CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty—forming the online spectral feedback module. The prediction uncertainties are obtained based on the prediction residual statistics of PLS ​​regression. The reference calibration module includes a low-field nuclear magnetic resonance analyzer and an online gas chromatograph to periodically provide highly selective and accurate component concentration verification and calibration data. This data is used to correct the output of the multivariate calibration model, eliminating quantitative errors caused by Raman probe contamination, spectral drift, or model aging, ensuring the reliability of the feedback data. Based on the reference-calibrated CPD concentration estimates, endo-DCPD concentration estimates, and ecto-DCPD concentration estimates, the current CPD product index and the current DCPD endo-ecto-isomer index are determined. Subsequently, the CPD concentration estimates, endo-DCPD concentration estimates, ecto-DCPD concentration estimates, and their corresponding prediction uncertainties, along with the process variable data, are input into the preset kinetic state estimator.

[0051] The preset kinetic state estimator incorporates a kinetic model based on the reaction network mechanism of CPD and DCPD interconversion, capable of outputting the current component concentration state and the predicted future concentration trajectory. The preset kinetic state estimator receives the process setpoints and, based on the reaction network mechanism, predicts the component change trend caused by external disturbances such as feed component fluctuations, temperature disturbances, or residence time changes, thereby generating the predicted future concentration trajectory. This predicted future concentration trajectory, as a feedforward predictive signal, can reflect the evolution direction of the process state in advance, allowing the controller to initiate compensation actions before actual component deviations occur. This overcomes the inherent large hysteresis characteristic of chemical reactors and is fundamentally different from conventional proportional-integral-derivative control that relies solely on ex-post corrections.

[0052] The deviation comparison node receives the process setpoint and the measured component concentration after reference calibration, calculates the difference, and generates a real-time error signal. This real-time error signal, the current component concentration status, the predicted future concentration trajectory, the current CPD product index, and the current DCPD endo-isomeric isomer index are input to the MPC controller. The MPC controller uses preset CPD product index values ​​and preset DCPD endo-isomeric isomer target values ​​as control objectives, performs rolling optimization under preset process constraints, and generates real-time control commands. During the rolling optimization process, the MPC controller further receives the prediction uncertainty output by the multivariate correction model, incorporating it as a constraint or weighting factor into the optimization objective. This allows the controller to automatically adopt a more conservative adjustment strategy when the uncertainty of the component concentration estimation is high, avoiding malfunctions caused by abnormal spectral data or model extrapolation. Based on the real-time control commands, the process variables of the C5 process are adjusted in real time. Specifically, this includes: adjusting the temperature control valve of the dimerizing reactor to keep the dimerizing reaction temperature within the range of 20℃ to 120℃; adjusting the feed pump to change the C5 feed flow rate to control the residence time; adjusting the circulation ratio adjustment valve to change the circulation ratio to the dimerizing reactor; and adjusting the temperature control valve of the pyrolysis reactor to keep the pyrolysis temperature within the range of 140℃ to 220℃. The changed process state after adjustment is re-acquired by the online Raman probe, forming a real-time closed-loop feedback loop. This embodiment also sets up an alarm and diagnostic module in the control layer. When the component concentration exceeds the standard, the prediction uncertainty exceeds the limit, or the sensor signal is abnormal, an alarm is automatically triggered, and the source of the fault is diagnosed from dimensions such as probe contamination, temperature deviation, and instrument drift, ensuring the long-term stability and safety of the closed-loop control. Therefore, by deeply integrating real-time Raman spectroscopy quantification, kinetic feedforward prediction, NMR and gas chromatography reference calibration, and MPC rolling optimization, the system can achieve precise directional control of the purity of the target CPD product and the ratio of DCPD isomers based on rapid feedback and high-precision calibration and long-cycle feedforward prediction without relying on traditional chromatographic separation. This effectively suppresses the formation of oligomer byproducts and significantly improves the real-time performance and robustness of process control.

[0053] Further, in one embodiment, the calibration process of the pre-calibrated multivariate correction model is as follows: A calibration sample set for the C5 process logistics is obtained. This sample set contains 120 calibration samples, with compositions covering a range of concentrations including cyclopentadiene (CPD) from 0 to 15% by weight, endogenous dicyclopentadiene (endo-DCPD) from 40 to 90% by weight, exogenous dicyclopentadiene (exogenous-DCPD) from 5 to 50% by weight, and oligomers such as tricyclopentadiene (TCPD) from 0 to 8% by weight. This ensures that the training domain of the multivariate correction model covers the actual process fluctuation range, fundamentally avoiding model extrapolation failure. Component values ​​were determined for each calibration sample using the following methods: offline gas chromatography with a flame ionization detector (FID) was used to determine the mass fraction of each component, with an accuracy of ±0.3% by weight. Simultaneously, 400MHz quantitative nuclear magnetic resonance (NMR) 1H NMR was used to determine the integral ratio of endogenous-DCPD vinyl protons to exogenous-DCPD vinyl protons, obtaining the CPD reference concentration, endogenous-DCPD reference concentration, and exogenous-DCPD reference concentration for each calibration sample as true values. Spectra were acquired for each calibration sample using 785nm excitation and a wavenumber range of 1000 cm⁻¹. -1 Up to 1700cm -1 Raman spectra of each sample were obtained by averaging 10 scans to acquire characteristic spectral data for each calibration sample. An initial training matrix was constructed based on the CPD reference concentration, intrinsic-DCPD reference concentration, extrinsic-DCPD reference concentration, and corresponding characteristic spectral data of each calibration sample. A full-spectrum partial least squares algorithm was used for model training to obtain the initial multivariate calibration model and initial concentration loading matrix. The full-spectrum partial least squares algorithm effectively uncovers the latent variable correlation between Raman spectral data and component concentrations, starting from the intrinsic-DCPD, which is located at approximately 1652 cm⁻¹. -1 And the shape - DCPD is located at approximately 1645cm -1 Discriminative information was extracted from subtle but stable peak position differences, enabling the multivariate calibration model to perform spectral analysis on the two isomers. Subsequently, leave-one-out cross-validation was used to validate the initial multivariate calibration model. By successively eliminating individual samples to evaluate the model's predictive performance on samples not used in training, the optimal number of latent variables was determined to be 12, resulting in the validated multivariate calibration model and the validated concentration loading matrix. This validated multivariate calibration model was designated as the pre-calibrated multivariate calibration model, and the validated concentration loading matrix was designated as the preset concentration loading matrix. Performance testing of the validated multivariate calibration model was conducted on a test set consisting of 30 independent samples. The resulting root mean square error of prediction (RMSEP) and coefficient of determination (R²) were measured. 2 The results are as follows: CPD's RMSEP is 0.9% by weight, R 2The RMSEP of the internal DCPD is 0.997; the RMSEP of the external DCPD is 1.3% by weight, and the R² is 0.995; the RMSEP of the external DCPD is 1.5% by weight, and the R² is 0.997. 2 The value is 0.993. The results show that the full-spectrum partial least squares model can achieve a wavenumber difference of approximately 7 cm⁻¹. -1 Under certain conditions, it can accurately distinguish between internal-type DCPD and external-type DCPD to meet the accuracy requirements of real-time control.

[0054] In one embodiment, to improve robustness to specific probe failure modes and reduce false positive alarm rates, the Raman probe and the ATR-FTIR probe are deployed at the same measurement point, for example, at the outlet of the dimer reactor and the inlet and outlet of the pyrolysis reactor, respectively. The Raman and ATR-FTIR spectra are stitched together, and a fusion partial least squares regression model is trained on the stitched spectral data to ensure that the root mean square error of prediction for all components is less than 1.5% by weight. An alarm is automatically triggered when the prediction deviation between the Raman channel and the ATR-FTIR channel for the same sample exceeds 3% by weight, and this data is marked to trigger gas chromatography verification, thereby effectively identifying measurement distortions caused by abnormal operating conditions such as probe contamination, laser drift, or window fouling.

[0055] Further, the online analytical measurement data is input into a pre-calibrated multivariate correction model, which outputs estimated CPD concentration, estimated internal DCPD concentration, and estimated external DCPD concentration, along with their corresponding CPD prediction uncertainties, internal DCPD prediction uncertainties, and external DCPD prediction uncertainties. Specifically, this includes: performing preprocessing on the spectral data in the online analytical measurement data. The preprocessing sequentially includes band truncation based on a preset spectral cutoff range, baseline correction based on a preset baseline correction algorithm, vector normalization based on a preset vector normalization algorithm, and smoothing based on a preset smoothing algorithm to obtain preprocessed spectral data. The preprocessed spectral data is then input into the pre-calibrated multivariate correction model, which performs quantitative analysis based on the online analytical measurement data and simultaneously outputs estimated CPD concentration, estimated internal DCPD concentration, and estimated external DCPD concentration, along with their respective prediction uncertainties.

[0056] In one embodiment, for an initial deployment scenario with limited capital expenditure and a priority on ensuring the yield of cyclopentadiene (CPD), only a single Raman probe is configured, using a 785nm excitation wavelength and installed in a side-flow manner. No nuclear magnetic resonance (NMR) or process gas chromatograph is used. Online analytical measurement data and process variable data are acquired from the C5 process stream containing CPD, endogenous dicyclopentadiene (endogenous DCPD), and exogenous dicyclopentadiene (exogenous DCPD). The process variable data includes at least the current dimer reactor temperature. The Raman probe acquires the material spectrum every 5 seconds, which is used as the spectral data in the online analytical measurement data. After band interception, baseline correction, vector normalization, and smoothing, preprocessed spectral data is obtained. The preprocessed spectral data is input into a pre-calibrated multivariate correction model. The model uses 12 latent variables and a preset concentration loading matrix. The estimated concentrations of CPD, endogenous DCPD, and exogenous DCPD are simultaneously calculated through regression projection. Simultaneously, the predicted uncertainty for each component is output based on a preset prediction uncertainty estimation algorithm. Based on the estimated CPD concentration, estimated internal DCPD concentration, and estimated external DCPD concentration, the current CPD product index and the current DCPD inverse and endo-isomer index are determined. Then, the current component concentration state vector is constructed using the estimated CPD concentration, estimated internal DCPD concentration, estimated external DCPD concentration, and their corresponding prediction uncertainties. This current component concentration state vector, the future concentration prediction trajectory, the current CPD product index, and the current DCPD inverse and endo-isomer index are input to a preset model predictive controller. The preset model predictive controller uses preset CPD product index values ​​and preset DCPD inverse and endo-isomer target values ​​as control objectives, and performs rolling optimization solutions under preset process constraints to generate real-time control commands. In this embodiment, the preset model predictive controller inputs the current component concentration state vector and the preset intrinsic DCPD target fraction into the proportional-integral-derivative (PID) controller. The PID controller directly calculates the temperature adjustment amount as the target control strategy, generating real-time control commands to regulate the dimer reactor temperature within the range of 80°C to 110°C, maintaining the intrinsic DCPD content within the target range of 80 ± 3% by weight. Based on these real-time control commands, the process variables of the C5 process are adjusted in real time. The changed process state triggers a new round of online analysis and measurement, forming a closed-loop iteration.

[0057] In another embodiment, taking the pyrolysis reactor of a C5 processing unit as an example, its feed composition is 72% by weight of internal DCPD, 22% by weight of external DCPD, and 6% by weight of oligomers, with a target CPD product purity greater than 97% by weight. Traditional operation uses a fixed pyrolysis temperature of 185°C and a residence time of 15 minutes. Although the internal DCPD is completely pyrolyzed, the conversion rate of external DCPD is only 78%, causing the CPD product purity to fluctuate between 92% and 96% by weight. Furthermore, the high temperature promotes re-dimerization of CPD, resulting in oligomer formation as high as 2.8% by weight. To overcome these shortcomings, this embodiment deploys Raman probes at the inlet and outlet of the pyrolysis reactor to monitor the material composition in real time. Online analytical measurement data and process variable data are obtained from the C5 process stream containing CPD, internal DCPD, and external DCPD. The process variable data includes at least the current pyrolysis reactor temperature and residence time. The spectral data from the online analysis and measurement data, after band interception, baseline correction, vector normalization, and smoothing, are input into a pre-calibrated multivariate correction model. This multivariate correction model is a partial least squares regression model with 12 latent variables, outputting estimated CPD concentration, estimated internal DCPD concentration, and estimated external DCPD concentration, along with their corresponding prediction uncertainties. Based on these estimated CPD concentrations, internal DCPD concentration, and external DCPD concentrations, the current CPD product index and the current DCPD internal and external isomer index are determined. The estimated CPD concentration, internal DCPD concentration, and external DCPD concentration, along with their corresponding prediction uncertainties, and the process variable data, are input into a preset kinetic state estimator. This preset kinetic state estimator uses an extended Kalman filter (EKF) to perform concentration evolution on the state vector, generating a concentration prediction sequence covering the changing trends of the internal and external DCPD ratios, and outputting the current component concentration state and future concentration prediction trajectory. The current component concentration status, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and endo-isomer index are input into the preset model prediction controller. With the preset target of CPD product purity greater than 97% by weight as the control objective, rolling optimization is performed under preset process constraints to generate real-time control commands.

[0058] In this embodiment, the preset model predictive controller predicts reaction conditions based on future concentration prediction trajectories and generates target control strategies: when the proportion of endomorphic DCPD is greater than 75% by weight, it is determined that endomorphic material is the main feed, generating a strategy of using a low temperature of 165°C to 175°C and a residence time of 12 minutes to ensure complete endomorphic pyrolysis while inhibiting CPD re-dimerization; when the proportion of exogenous DCPD is greater than 35% by weight, it is determined that exogenous material accumulates, generating a strategy of using a high temperature of 195°C to 205°C and a residence time of 8 minutes to ensure sufficient exogenous pyrolysis while controlling re-polymerization by shortening the residence time. Based on the real-time control commands, the temperature and feed flow rate of the pyrolysis reactor are dynamically adjusted, extending the control basis from the current moment to the future time period. Potential trends of purity or isomer ratio deviation from the target are identified in advance, and forward-looking control commands are generated, effectively overcoming quality fluctuations caused by large hysteresis and achieving precise directional control of the pyrolysis process.

[0059] Furthermore, in one embodiment, a method for real-time control of cyclopentadiene and dicyclopentadiene is provided. Hereinafter, cyclopentadiene is abbreviated as CPD, and dicyclopentadiene as DCPD. This method is fully equipped with a Raman spectrometer, an attenuated total reflectance Fourier transform infrared spectrometer, a low-field nuclear magnetic resonance analyzer, and an online gas chromatograph, used to obtain online analytical measurement data and process variable data from C5 process streams containing CPD, internal DCPD, and external DCPD.

[0060] The acquired online analytical measurement data are input into the pre-calibrated multivariate correction models corresponding to their respective channels. The pre-calibrated multivariate correction models for each channel all adopt partial least squares regression (PLS) models, which perform quantitative analysis based on the online analytical measurement data, and output the estimated CPD concentration, internal DCPD concentration, and external DCPD concentration for each channel, as well as their respective prediction uncertainties.

[0061] Subsequently, based on the concentration estimates and prediction uncertainties of each channel, a Gaussian Bayes weighted fusion algorithm was used for data fusion. This algorithm uses the reciprocal of the real-time variance of each channel as the weight to perform a weighted average of multiple predicted values ​​for the same component, obtaining a globally fused concentration estimate and its globally fused variance. The globally fused variance is the reciprocal of the sum of the weights. The real-time variance of each channel is dynamically estimated as follows: the variance of the Raman channel is calculated using an exponentially weighted moving average (EWMA) of the squared PLS residuals over the past 100 seconds, with a smoothing coefficient α of 0.1, and is superimposed with the variance from the spectral blank area of ​​1800 cm⁻¹. -1 Up to 2500cm -1The baseline noise variance is estimated using the standard deviation to comprehensively reflect the model residuals and baseline perturbations. The variance of the ATR-FTIR channel is estimated using the same EWMA method. The variance of the low-field NMR channel is estimated by the peak integral signal-to-noise ratio, typically 0.7% by weight squared, and a time decay factor of 300 seconds is introduced due to the low acquisition frequency to characterize the decay of its information timeliness. The variance of the GC channel is given by the repeatability of standard samples, typically 0.3% by weight squared, and a time decay factor of 1800 seconds is introduced so that its weight decays to zero after one GC cycle. Thus, the weights of outdated or anomalous data are automatically decayed, ensuring the robustness of the fusion results.

[0062] The current CPD product index and the current DCPD endo- and endo-isomer index are determined based on the globally fused concentration estimate. Simultaneously, the globally fused concentration estimate, the globally fused variance for each component, and the process variable data are input into a preset kinetic state estimator. This preset kinetic state estimator, based on the CPD and DCPD reaction network mechanism, outputs the current component concentration state and the predicted future concentration trajectory. A fusion confidence interval is constructed based on the obtained globally fused variance. When the globally fused variance exceeds a preset threshold, the current overall estimation quality is deemed unreliable. The preset threshold is, for example, 2 weight percentage squared, meaning that when the half-width of the fused double confidence interval exceeds 4 weight percentage, an action amplitude control strategy is generated: the control action amplitude in the target control strategy is halved, and an alarm is issued to the operator; otherwise, normal control action is maintained.

[0063] Next, the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo-isomeric index are input to a preset model predictive controller. This preset model predictive controller uses preset CPD product index values ​​and preset DCPD endo-isomeric target values ​​as control objectives, and performs rolling optimization under preset process constraints to generate real-time control commands. Specifically, the preset model predictive controller determines the CPD product index deviation based on the preset CPD product index values ​​and the current CPD product index, and determines the DCPD endo-isomeric index deviation based on the preset DCPD endo-isomeric target values ​​and the current DCPD endo-isomeric index; it constructs a multi-objective optimization function based on the CPD product index deviation and the DCPD endo-isomeric index deviation, and determines the feasible region of manipulated variables based on the preset process constraints. The manipulated variables include at least the dimer reactor temperature, feed flow rate, and recycle ratio; it solves the multi-objective optimization function based on the current component concentration state, future concentration prediction trajectory, and feasible region of manipulated variables to obtain the current time-to-time sequence of manipulated variables. The preset model predictive controller then receives the fused global concentration estimate, the fusion confidence interval, and the predicted future concentration trajectory. It performs rolling optimization within a 10-step prediction time domain, calculating the optimal control trajectory for the dimerizing reactor temperature, feed flow rate, and recycle ratio. Based on the action amplitude control strategy, it corrects the setpoints of the manipulated variables and generates the real-time control command based on the corrected setpoints. The real-time control command is used to regulate the process variables of the C5 process in real time, making it suitable for large-scale C5 processing plants requiring comprehensive process digitization and refined control.

[0064] Further, in one embodiment, when the process variables of the C5 process are adjusted in real time based on the real-time control command, a preset weighted fusion algorithm is used to weight and fuse the estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties to obtain the fused estimated CPD concentration, the fused estimated internal DCPD concentration, the fused estimated external DCPD concentration, and the corresponding global fused variance. For any component among CPD, internal DCPD, and external DCPD, when the component has at least two different online analysis channels with estimated concentration values, a weighted fusion is performed based on the prediction uncertainty and the estimated concentration value of each online analysis channel to obtain the fused estimated concentration value and the fused variance of the component; when the component has only a single online analysis channel with estimated concentration value, the estimated concentration value is directly used as the fused estimated concentration value, and the fused variance is determined based on the prediction uncertainty of that channel. Subsequently, a fusion confidence interval is constructed based on the fusion variance of all components, and an action amplitude control strategy is generated. The real-time control command is then modified using this action amplitude control strategy to obtain the modified real-time control command, and the process variables of the C5 process are adjusted in real time accordingly.

[0065] In one embodiment, an online process gas chromatograph is used to monitor the cyclopentadiene (CPD) content in the C5 process stream in real time with an analysis cycle of 5 minutes, obtaining the measured CPD content as part of the process variable data. Simultaneously, Raman spectra of the material, acquired in real time by a Raman probe, are obtained from the C5 process stream containing CPD, endogenous dicyclopentadiene (endogenous DCPD), and exogenous dicyclopentadiene (exogenous DCPD). After spectral interception, baseline correction, vector normalization, and smoothing, the spectra are input into a pre-calibrated multivariate correction model. Quantitative analysis is performed based on the online analytical measurement data, outputting estimated CPD concentration, estimated endogenous DCPD concentration, estimated exogenous DCPD concentration, and their corresponding prediction uncertainties. Based on the estimated CPD concentration, estimated endogenous DCPD concentration, and estimated exogenous DCPD concentration, the current CPD product index and the current DCPD endogenous and exogenous isomer index are determined. Subsequently, the estimated CPD concentration, estimated internal DCPD concentration, estimated external DCPD concentration, and their respective prediction uncertainties, along with the process variable data, are input into a preset kinetic state estimator. This preset kinetic state estimator incorporates a reaction kinetic model based on the dimerization, cracking, and isomerization reaction network of CPD and DCPD, and can deduce the propagation path of feed disturbances in the reactor and separation unit based on material balance and reaction mechanisms. The state estimator also receives the feedforward disturbance deviation obtained based on the difference between the preset feed reference concentration and the measured CPD content, and combines this feedforward disturbance deviation with the current component concentration state vector to perform state deduction and output the expected concentration deviation trajectory. This expected concentration deviation trajectory, as a form of future concentration prediction trajectory, also has a feedforward prediction function. Then, the current component concentration state, the future concentration prediction trajectory, the current CPD product index, and the current DCPD internal and external isomer index are input into a preset model prediction controller. The preset model predictive controller uses preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values ​​as control objectives. Under preset process constraints, it performs rolling optimization to solve the problem. Based on the expected concentration deviation trajectory, it generates compensation adjustment commands to optimize the target control strategy, thereby generating real-time control commands. Based on these real-time control commands, it adjusts the temperature, feed flow rate, and recycle ratio of the dimer reactor. Thus, changes in feed CPD content are introduced as feedforward perturbation variables into the preset model predictive controller, enabling the reactor to make predictive adjustments before compositional changes reach the reactor.

[0066] In a certain application scenario, the CPD content in the feed C5 of a factory abruptly decreased from 16% by weight to 11% by weight over approximately 2 hours. Traditional pure temperature feedback control only responds after gas chromatography detects the change in CPD product composition. Due to a 30-minute chromatographic analysis cycle and a 10-minute reactor thermal hysteresis, the feedback adjustment is delayed by approximately 40 minutes. During this period, the CPD product purity drops to 93.5% by weight, resulting in approximately 2.1 tons of defective products. However, using the method described in this embodiment, after the online process GC analyzer detects a decrease in the feed CPD content, it immediately transmits the disturbance information to the MPC controller in a feedforward manner: acquiring the measured CPD content value, calculating the feedforward disturbance deviation based on the baseline concentration of 16% by weight, and combining it with the current component concentration state vector output by the pre-calibrated multivariate correction model. A preset kinetic state estimator then deduces the expected concentration deviation trajectory where product purity will decrease, generating compensation adjustment instructions accordingly. The MPC controller then generates real-time control instructions, specifically including increasing the dimer reactor temperature by 5°C, increasing the residence time by 10%, and increasing the recycle ratio from 35% to 42%. After implementation, the purity of CPD products remained stable at over 98.2% by weight, with no defective products. This embodiment actively counteracts the propagation effect of fluctuations in the feed CPD content, forming a synergy with feedback regulation based on concentration prediction sequences. The feedforward path compensates for disturbances in advance, and the feedback path corrects residual deviations, thereby shortening the response delay and effectively suppressing the impact of feed disturbances on the purity of CPD products and the ratio of DCPD isomers.

[0067] It should be noted that the recycle ratio can be specifically expressed as the ratio of the mass flow rate of the recycle stream returning to the dimerizing reactor to the mass flow rate of the C5 feedstock entering the device.

[0068] In another embodiment, an adaptive update mechanism for a Raman model corrected by nuclear magnetic resonance is provided. After 30 days of continuous operation, due to the change in raw material source, the Raman PLS model systematically under-predicts the external DCPD, with an average deviation of +4.2 wt%. The low-field nuclear magnetic resonance (NMR) analyzer performs a verification measurement every 5 minutes. When the deviation between the low-field NMR measurement and the Raman prediction exceeds a preset trigger threshold, the correction process is automatically triggered. This preset trigger threshold is, for example, 3 wt%. The correction process first temporarily uses the measured low-field NMR value as the input to the MPC controller to ensure control continuity; simultaneously, 150 sets of paired Raman and low-field NMR spectra samples from the past 24 hours are extracted, and the model is incrementally updated using a recursive least squares method with a forgetting factor of 0.95. After the update, the root mean square error (RMSEP) of the model's prediction recovers to 1.2 wt%, meeting the requirement of not exceeding 1.5 wt%, thereby automatically restoring the Raman-dominated control mode. This embodiment utilizes highly selective verification data provided by low-field NMR to periodically calibrate the spectral quantitative model, which can effectively overcome model mismatch when the raw material composition changes, and keep the model accuracy stable during long-term operation.

[0069] Further, in one embodiment, when the process variables of the C5 process are adjusted in real time based on the real-time control command, the preset kinetic state estimator is specifically configured as a preset reaction kinetic model with an embedded extended Kalman filter, used to connect the spectral measurement results and perform disturbance propagation path deduction to generate the expected concentration deviation trajectory. The measured value of cyclopentadiene (CPD) content in the feed stream of the target process unit in the C5 process stream is acquired in real time, and the feedforward disturbance deviation is obtained based on the difference between the preset feed reference concentration and the measured value of CPD content. Simultaneously, the CPD concentration state, the internal dicyclopentadiene concentration state, and the external dicyclopentadiene concentration state are determined based on the current component concentration state vector, and the oligomer concentration state is determined according to the preset material balance relationship, wherein the oligomer includes at least tricyclopentadiene TCPD. An augmented state vector is constructed, which is a six-dimensional vector. The first four dimensions represent the component concentration state variables, and the last two dimensions represent online adaptive rate constants. These are the dimerization rate constants for CPD dimerizing to dicyclopentadiene (DCPD) via the Diels-Alder reaction, and the fragmentation rate constants for DCPD fragmenting to CPD via the reverse Diels-Alder reaction. Incorporating the dimerization and fragmentation rate constants into the augmented state vector for online adaptive updating allows the state estimation to simultaneously encompass component distribution and key kinetic parameters. The augmented state vector is time-evolved using a preset process noise matrix Q and a feedforward perturbation bias to obtain a predicted state vector. Then, the predicted state vector is corrected for bias using a preset measurement noise matrix R and the current component concentration state vector to obtain the updated state vector. The preset process noise matrix Q represents the uncertainty of the model prediction and is a diagonal matrix. The preset measurement noise matrix R represents the uncertainty of Raman spectroscopy measurements of each component concentration and is also a diagonal matrix, corresponding to the root mean square error of the Raman partial least squares regression model's predictions for each component. Subsequently, concentration trajectory was extrapolated based on the isomerization rate constant, the polymerization rate constant of irreversible polymerization of endogenous dicyclopentadiene to oligomers, the polymerization rate constant of irreversible polymerization of exogenous dicyclopentadiene to oligomers, and the updated dimerization rate constant and pyrolysis rate constant. During the extrapolation process, the extrapolation results were convolved by the residence time distribution function to reflect the influence of non-ideal mixing. Finally, the updated state estimate and the predicted future concentration trajectory with a prediction step size of 10 steps were output, forming the expected concentration deviation trajectory.

[0070] It should be noted that the dimerization rate constant and the cracking rate constant were chosen as online adaptive update variables in this embodiment because they are extremely sensitive to operating conditions such as catalyst residue, feedstock moisture, and oxygen content, and drift significantly. They need to be identified online with EKF to maintain prediction accuracy. The isomerization rate constant, the polymerization rate constant of endogenous dicyclopentadiene to oligomer, and the polymerization rate constant of exogenous dicyclopentadiene to oligomer are intrinsic rates at the molecular level. If all five rate constants are identified at the same time, the observability of EKF will be insufficient and there is a risk of divergence. Therefore, these three are treated as fixed parameters.

[0071] The isomerization rate constant characterizes the rate of reversible isomerization between endogenous and exogenous dicyclopentadiene, with its Arrhenius parameter activation energy Ea taken as 93 kJ / mol and pre-exponential factor A as 3.2 × 10⁻⁶. 10 The process model is embedded per second as a function of temperature, without being updated individually; the initial value of the dimerization rate constant at 100℃ is 5.0 × 10⁻⁶. -4 The initial value of the pyrolysis rate constant at 180 °C is 2.8 × 10⁻⁶ mol / s. -3 Every second. The above reaction network fully describes the molecular transformation mechanism of the entire CPD process: two CPD molecules reversibly dimerize to form endogenous and exogenous dicyclopentadiene. These two DCPDs can isomerize at high temperatures and irreversibly polymerize into oligomeric impurities. Temperature and material residence time directly determine product distribution. At low temperatures and short residence times, the dimerization rate constant is much greater than the cracking rate constant, kinetically controlling the endogenous dicyclopentadiene as the main product. At high temperatures and long residence times, the isomerization pathway is favorable, thermodynamically controlling the exogenous dicyclopentadiene as the main product. The polymerization pathway is intensified under high temperature and high DCPD concentration conditions, and is a key side reaction to be suppressed. The process maximizes CPD monomer yield by segmented temperature control and directional regulation of the reaction equilibrium. Based on the obtained expected concentration deviation trajectory, a compensation adjustment command is generated, and the real-time control command is optimized based on this compensation adjustment command to obtain the optimized real-time control command, thereby enabling real-time regulation of process variables in the C5 process.

[0072] Furthermore, the preset kinetic state estimator also has an oligomer formation prediction function, which predicts the oligomer formation trend under high temperature and high DCPD concentration conditions by tracking the polymerization rate constant and polymerization rate constant path. When the predicted oligomer concentration exceeds the preset threshold, an early warning is issued to the control layer and corresponding process adjustment actions are triggered, thereby effectively suppressing side reactions while ensuring product purity.

[0073] Furthermore, in one embodiment, the preset kinetic state estimator also has an oligomer formation prediction function. By tracking the reaction paths corresponding to the polymerization rate constants of the irreversible polymerization of endogenous dicyclopentadiene into oligomers and the polymerization rate constants of the irreversible polymerization of exogenous dicyclopentadiene into oligomers, it quantifies and predicts the oligomer formation trend under high temperature and high dicyclopentadiene (DCPD) concentration conditions. When the predicted oligomer concentration exceeds a preset threshold, it issues an early warning to the control layer and triggers corresponding process adjustment actions, effectively suppressing side reactions while ensuring product purity.

[0074] Furthermore, when real-time controlling the process variables of the C5 process based on the aforementioned real-time control commands, the characteristic band spectral vectors of oligomers are extracted from the online analytical measurement data. Based on the pre-calibrated multivariate correction model, the characteristic band spectral vectors of oligomers are quantitatively analyzed to obtain an estimated oligomer concentration. The oligomer concentration at 1580 cm⁻¹... -1 Up to 1600cm -1 The region exhibits a characteristic Raman response to the stretching vibration of conjugated carbon-carbon double bonds, which is distinguishable from the spectral contributions of monomers and dimers. Therefore, it can be quantified simultaneously using a partial least squares regression model. The oligomers mainly include tricyclopentadiene and higher polymers. The estimated oligomer concentration is compared with a preset oligomer threshold. When the oligomer concentration is greater than or equal to the preset threshold, it is determined to be an out-of-limit condition. Based on this out-of-limit determination, a clamping control command is generated. This clamping control command bypasses conventional optimization logic and directly forces a correction to the target control strategy generated based on the estimated concentrations and prediction uncertainties of cyclopentadiene (CPD), internal DCPD, and external DCPD. For example, it forcibly reduces the temperature of the pyrolysis reactor or shortens the residence time, rapidly inhibiting further oligomer formation. This adds a safety protection layer while controlling CPD product indicators and DCPD isomer indicators, preventing oligomer accumulation due to excessive pursuit of conversion rate or pyrolysis depth, and avoiding increased downstream separation load and product quality deterioration.

[0075] In another embodiment, the real-time control method for CPD and DCPD further integrates a model adaptive update mechanism based on NMR verification. A low-field NMR flow cell is set up in the sidestream of the DCPD product stream, and NMR spectra are acquired from this flow cell based on a preset acquisition cycle. Characteristic peak analysis is performed on the inner-type vinyl proton resonance peak and the outer-type vinyl proton resonance peak in the NMR spectrum to obtain the NMR measurements of the inner-type DCPD and the outer-type DCPD, and to determine the NMR isomer ratio; wherein, the chemical shift of the inner-type vinyl proton resonance peak is 6.50 to 6.55 ppm, and the chemical shift of the outer-type vinyl proton resonance peak is 6.42 to 6.48 ppm, and the chemical shift represents the frequency shift of the proton resonance peak relative to the reference material. Simultaneously, the spectral isomer ratio is determined based on the estimated concentrations of the inner-type DCPD and the outer-type DCPD, and the consistency between the NMR isomer ratio and the spectral isomer ratio is checked to obtain the ratio deviation value to monitor whether there is drift in the spectral model. When the proportional deviation reaches or exceeds a preset internal calibration threshold, the model is deemed to have a statistically significant systematic deviation, and the calibration process is automatically triggered. If the deviation further expands to an external alarm threshold, an alarm requiring manual intervention is issued. The preset internal calibration threshold is, for example, 3 weight percentages, approximately twice the confidence width of the root mean square error of the partial least squares regression model prediction; the external alarm threshold is, for example, 5 weight percentages. After triggering internal calibration, the NMR measurement value is temporarily used as the control input to maintain control continuity. Simultaneously, a set of paired samples of historical Raman spectra and NMR data is extracted from a preset paired sample window. The preset paired sample window covers the past 24 hours, corresponding to approximately 150 pairs of paired samples. Based on this paired sample set, the pre-calibrated multivariate calibration model is updated using a recursive partial least squares algorithm with a forgetting factor, resulting in an updated multivariate calibration model, which is then used for multi-component concentration estimation of subsequently acquired online analytical measurement data.

[0076] During the update process, for new samples, the covariance matrix is ​​recursively calculated based on the forgetting factor and the current covariance matrix, followed by the calculation of the gain vector and the updating of the regression coefficient matrix. The number of latent variables in the model is fixed at 12. Before the update, outlier samples are removed within the 95% confidence ellipsoid based on the Q statistic and the Hotelling T² statistic. The Q statistic measures the deviation of the new samples from the model residual space, and the Hotelling T² statistic measures the deviation of the new samples from the principal component space. The update convergence criterion is that the root mean square error of the leave-one-out cross-validation prediction changes by less than 0.2 weight percentages compared to the previous round. After the update, the root mean square error of the validation prediction recovers to 1.2 weight percentages, meeting the requirement of not exceeding 1.5 weight percentages, and the Raman dominant control mode is automatically restored. The forgetting factor ranges from 0.92 to 0.98. Therefore, the multivariate calibration model can achieve self-calibration without stopping the machine, effectively overcome model mismatch when raw material fluctuates or operating conditions change, and ensure that the concentration estimate on which the closed-loop control is based maintains reliable accuracy in the long term. It is suitable for downstream applications with strict limitations on the ratio of internal to external shape of polymer-grade DCPD products, such as the manufacture of polydicyclopentadiene resin.

[0077] Please see Figure 2 This embodiment also provides a real-time control system for CPD and DCPD, including: The data acquisition module is used to acquire online analytical measurement data and process variable data from the C5 process stream containing cyclopentadiene CPD, internal DCPD and external DCPD; The spectral analysis module is used to input the online analysis measurement data into a pre-calibrated multivariate correction model, so that the multivariate correction model can perform quantitative analysis based on the online analysis measurement data and output the CPD concentration estimate, the internal DCPD concentration estimate, and the external DCPD concentration estimate, as well as the corresponding CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty. The index determination module is used to determine the current CPD product index and the current DCPD endo- and endo-type isomer index based on the CPD concentration estimate, the endo-type DCPD concentration estimate, and the extrin-type DCPD concentration estimate. The state estimation module is used to input the estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties and the process variable data into a preset kinetic state estimator, and output the current component concentration state and the future concentration prediction trajectory. The optimization control module is used to input the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index to the preset model prediction controller, so that the preset model prediction controller uses the preset CPD product index value and the preset DCPD endo- and extrinsic isomer target value as control targets, performs rolling optimization solution under preset process constraints, and generates real-time control commands. The process control module is used to control the process variables of the C5 process in real time based on the real-time control command.

[0078] In one embodiment, a real-time control system for cyclopentadiene and dicyclopentadiene is provided. Hereinafter, cyclopentadiene is referred to as CPD and dicyclopentadiene as DCPD. The system includes a data acquisition module, a spectral analysis module, an index determination module, a state estimation module, an optimization control module, and a process control module.

[0079] The data acquisition module is used to acquire online analytical measurement data and process variable data from the C5 process stream containing CPD, internal DCPD, and external DCPD. The data acquisition module is configured with online analytical measurement points along the material flow direction of the C5 process stream at the dimer reactor outlet, separation unit feed or tower top, pyrolysis reactor inlet or outlet, and CPD product pipeline. At each measurement point, at least one of a Raman spectrometer, a low-field nuclear magnetic resonance analyzer, or an online gas chromatograph is installed to form a multi-point sampling network covering the entire process chain. The process variable data includes at least one of the following: dimer reactor temperature, pyrolysis reactor temperature, feed flow rate, recycle ratio, reaction pressure, and material residence time. These data variables are used as input parameters in the state transition equation to participate in the deduction of the component concentration evolution trajectory by the preset kinetic state estimator. This data acquisition module is implemented using an integrated online spectrometer combining Raman and attenuated total reflectance Fourier transform infrared spectroscopy, directly deployed at each measurement point. The integrated online spectrometer is internally equipped with, along the optical path, a laser source driver and control unit, a Raman laser, a flow cell Raman probe, an optical signal switching coupler, an ATR-FTIR immersion probe, a spectroscopic detection host, and a spectroscopic data acquisition and industrial control processing unit. The laser source drive and control unit provides a stable excitation source for the Raman laser; the Raman laser has a built-in filter and collimating lens group to output a parallel excitation beam; the flow cell Raman probe is inserted into the process pipeline in an online or bypass manner to introduce the laser into the material and collect the scattered Raman signal; the optical signal switching coupler realizes the switching and beam combining of Raman and infrared light signals, allowing both to share the back-end detection channel; the ATR-FTIR immersion probe uses the attenuated total internal reflection principle to collect the absorption signal of infrared light by the material, and has a built-in spectroscopic crystal and optical path collimation component; the spectroscopic detection host separates the light signal by wavelength through a grating or interferometer and converts it into an electrical signal by a photodetector; the spectral data acquisition and industrial control processing unit completes analog-to-digital conversion and has a built-in chemometrics algorithm library to analyze the spectral output component concentration in real time, and at the same time interfaces with the plant's distributed control system to realize over-limit alarms and closed-loop process control. Process variable data include at least the temperature of the dimerizing reactor, feed flow rate, circulation ratio, and the temperature of the pyrolysis reactor, which are collected by temperature and flow sensors.

[0080] The spectral analysis module is used to input the online analytical measurement data into a pre-calibrated multivariate correction model. This model performs quantitative analysis based on the online data, outputting estimated CPD concentrations, estimated internal DCPD concentrations, estimated external DCPD concentrations, and corresponding prediction uncertainties. Specifically, the spectral analysis module intercepts Raman spectra based on preset target bands to obtain characteristic band spectral vectors. It then performs preprocessing on these vectors, including baseline correction, smoothing, and vector normalization. The preprocessed spectral vectors are then input into the pre-calibrated multivariate correction model. This pre-calibrated multivariate correction model is a partial least squares regression model, which simultaneously calculates the estimated concentrations and prediction uncertainties of each component through regression projection.

[0081] The index determination module is used to determine the current CPD product index and the current DCPD endo-DCPD isomer index based on the estimated CPD concentration, the estimated endo-DCPD concentration, and the estimated extrinsic DCPD concentration.

[0082] The state estimation module is used to input the estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, their respective prediction uncertainties, and the process variable data into a preset kinetic state estimator, and output the current component concentration state and the predicted future concentration trajectory. The preset kinetic state estimator embeds an extended Kalman filter, which is used to perform concentration evolution on the state vector to generate a concentration prediction sequence covering the changing trends of the ratio of internal and external DCPD.

[0083] The optimization control module inputs the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index to the preset model predictive controller. The preset model predictive controller uses preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values ​​as control objectives, and performs rolling optimization under preset process constraints to generate real-time control commands. The preset model predictive controller employs a model predictive control (MPC) strategy, with a defined time-domain structure: a control step size of 10 seconds, configurable within the range of 5 to 30 seconds; a prediction time domain of 10 steps, corresponding to a 100-second look-ahead window; and a control time domain of 3 steps, corresponding to a 30-second control command sequence. The timing matching relationship between sampling and control is as follows: Raman spectroscopy is acquired every 5 seconds, ATR-FTIR spectroscopy is acquired every 5 seconds, and process parameters are acquired every 1 second from temperature and flow sensors; every 10 seconds, all sensor data are aggregated to form a set of MPC input vectors, ensuring that the spectral sampling interval is naturally nested within the control step size.

[0084] In this embodiment, at time zero, Raman and infrared spectra are processed by a multivariate calibration model to output component concentrations, which the EKF uses to update its state estimate. At the 10th second, spectral and temperature / flow sensor data are reassembled, and the MPC solves a quadratic programming problem, outputting a sequence of three control actions and adjusting the actuator accordingly. The above process is repeated at the 20th second, with the sliding window moving forward one step. During this period, the low-field NMR analyzer performs cross-validation every 300 seconds. If the deviation exceeds the 3% weight percentage internal calibration threshold, a recursive partial least squares incremental update is triggered, without interrupting normal control. The online gas chromatograph provides a reference measurement every 600 seconds for data archiving and long-term model calibration, which also does not affect the main control loop. The MPC's control cycle of 10 seconds is much shorter than the typical residence time of 5 to 60 minutes in a dimer reactor and 5 to 30 minutes in a pyrolysis reactor, ensuring a reasonable match between the control timescale and process kinetics, fully responding to reaction dynamics while avoiding frequent actuator actions.

[0085] The process control module is used to regulate the process variables of the C5 process in real time based on the real-time control commands. Specifically, it includes adjusting the temperature control valve of the dimerizing reactor, the feed pump, the circulation ratio adjustment valve, and the temperature control valve of the pyrolysis reactor. Thus, the above modules are connected sequentially according to the data flow, forming a complete information flow from light signal sensing to control action output, from the acquisition and preprocessing of Raman and infrared spectra, quantitative analysis of multiple components to MPC closed-loop control with feedforward and feedback. This enables real-time, precise, and fully automated control of CPD recovery and DCPD isomer distribution without manual intervention.

[0086] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute a real-time control method for CPD and DCPD as described in any of the above-described method embodiments of the present invention.

[0087] The integrated modules of the system, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0088] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A real-time control method for CPD and DCPD, characterized in that, Includes the following steps: Online analytical measurement data and process variable data were obtained from the C5 process stream containing cyclopentadiene CPD, internal DCPD and external DCPD; The online analytical measurement data is input into a pre-calibrated multivariate correction model, so that the multivariate correction model can perform quantitative analysis based on the online analytical measurement data and output the estimated CPD concentration, the estimated internal DCPD concentration, and the estimated external DCPD concentration, as well as the corresponding CPD prediction uncertainty, the internal DCPD prediction uncertainty, and the external DCPD prediction uncertainty. The current CPD product indicators and the current DCPD intermolecular and exotype isomer indicators are determined based on the estimated CPD concentration, estimated endotype DCPD concentration, and estimated exotype DCPD concentration. The estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties, along with the process variable data, are input into a preset kinetic state estimator, which outputs the current component concentration state and the future concentration prediction trajectory. The current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and endo-isomer index are input to the preset model prediction controller, so that the preset model prediction controller uses the preset CPD product index value and the preset DCPD endo- and endo-isomer target value as control targets, performs rolling optimization solution under preset process constraints, and generates real-time control commands. The process variables of the C5 process are adjusted in real time based on the real-time control commands.

2. The real-time control method for CPD and DCPD according to claim 1, characterized in that, The calibration process of the pre-calibrated multivariate correction model includes: Obtain a C5 process material calibration sample set, which includes CPDs, internal DCPDs, and external DCPDs; Based on the preset reference value determination algorithm, the component values ​​of each calibration sample in the C5 process logistics calibration sample set are determined to obtain the CPD reference concentration, internal DCPD reference concentration and external DCPD reference concentration in each calibration sample; Spectral data of each calibration sample were acquired to obtain the characteristic spectral data of each calibration sample. An initial training matrix is ​​constructed based on the CPD reference concentration, internal DCPD reference concentration, external DCPD reference concentration, and characteristic spectral data of each calibration sample. Based on the initial training matrix, model construction and training are performed to obtain the initial multivariate correction model and the initial concentration loading matrix; The initial multivariate calibration model was validated to obtain the validated multivariate calibration model and the validated concentration loading matrix. The verified multivariate correction model is determined as the pre-calibrated multivariate correction model, and the verified concentration loading matrix is ​​determined as the preset concentration loading matrix.

3. The real-time control method for CPD and DCPD according to claim 1, characterized in that, The online analytical measurement data is input into a pre-calibrated multivariate correction model, enabling the model to perform quantitative analysis based on the online analytical measurement data. The model outputs estimated CPD concentration, estimated internal DCPD concentration, and estimated external DCPD concentration, along with the corresponding CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty, including: Based on a preset spectral cutoff range, the spectral data in the online analysis and measurement data is cut off to obtain the target band spectral data. The target band spectral data is baseline corrected based on a preset baseline correction algorithm to obtain baseline corrected spectral data. The baseline-corrected spectral data is vector-normalized based on a preset vector normalization algorithm to obtain normalized spectral data. The normalized spectral data is smoothed using a preset smoothing algorithm to obtain preprocessed spectral data. The preprocessed spectral data is input into a pre-calibrated multivariate correction model, which performs quantitative analysis based on online analytical measurement data and outputs CPD concentration estimates, internal DCPD concentration estimates, and external DCPD concentration estimates, as well as the corresponding CPD prediction uncertainties, internal DCPD prediction uncertainties, and external DCPD prediction uncertainties.

4. The real-time control method for CPD and DCPD according to claim 1, characterized in that, The current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index are input to a preset model prediction controller. This causes the preset model prediction controller to perform rolling optimization under preset process constraints, using preset CPD product index values ​​and preset DCPD endo- and extrinsic isomer target values ​​as control objectives, and generate control commands, including: The deviation of CPD product indicators is determined based on the preset CPD product indicator value and the current CPD product indicator, and the deviation of DCPD endo-endo isomer indicators is determined based on the preset DCPD endo-endo isomer target value and the current DCPD endo-endo isomer indicator. A multi-objective optimization function is constructed based on the CPD product index deviation and the DCPD endotype isomer index deviation, and the feasible region of the manipulated variables is determined based on the preset process constraints. Based on the current component concentration state, the future concentration prediction trajectory, and the feasible region of the manipulated variables, the multi-objective optimization function is solved to obtain the sequence of manipulated variables at the current time. The real-time control command is generated based on the set value of the manipulated variable corresponding to the current control step size in the manipulated variable sequence.

5. The real-time control method for CPD and DCPD according to claim 1, characterized in that, The real-time adjustment of process variables in the C5 process based on the real-time control command includes: The CPD concentration estimate, internal DCPD concentration estimate, external DCPD concentration estimate, and their respective CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty are weighted and fused based on a preset weighted fusion algorithm to obtain the fused CPD concentration estimate, the fused internal DCPD concentration estimate, the fused external DCPD concentration estimate, and the corresponding global fused variance. For any component in CPD, internal DCPD and external DCPD, obtain the estimated concentration of the component and the corresponding prediction uncertainty in at least one online analysis channel; When there are concentration estimates for a component from at least two different online analysis channels, a weighted fusion is performed based on the prediction uncertainty and concentration estimates of each online analysis channel to obtain the fused concentration estimate and the variance after fusion for that component. When the concentration estimate of a component exists only in a single online analysis channel, the concentration estimate is used as the fusion concentration estimate of the component, and the fusion variance of the component is determined based on the prediction uncertainty of the online analysis channel. A fusion confidence interval is constructed based on the fusion variance of all components, and an action amplitude control strategy is generated based on the fusion confidence interval. The real-time control command is modified based on the aforementioned motion amplitude control strategy to obtain a modified real-time control command, and the process variables of the C5 process are adjusted in real time based on the modified real-time control command.

6. The real-time control method for CPD and DCPD according to claim 1, characterized in that, The real-time adjustment of process variables in the C5 process based on the real-time control command includes: Real-time acquisition of measured CPD content in the feed stream of the target process unit in the C5 process stream; The feedforward disturbance deviation is obtained based on the preset feed reference concentration and the measured value of the CPD content. The feedforward perturbation deviation and the current component concentration state are input into a preset reaction kinetic model so that the reaction kinetic model can perform perturbation propagation path deduction based on the feedforward perturbation deviation and the current component concentration state to obtain the expected concentration deviation trajectory. A compensation adjustment command is generated based on the expected concentration deviation trajectory. The real-time control command is optimized based on the compensation and adjustment command to obtain the optimized real-time control command, and the process variables of the C5 process are adjusted in real time based on the optimized real-time control command.

7. The real-time control method for CPD and DCPD according to claim 1, characterized in that, The real-time adjustment of process variables in the C5 process based on the real-time control command further includes: Extract the characteristic band spectral vectors of oligomers from the online analysis and measurement data; Based on the pre-calibrated multivariate correction model, the oligomer characteristic band spectral vectors are quantitatively analyzed to obtain an estimated oligomer concentration. Based on the estimated oligomer concentration and the preset oligomer threshold, an oligomer exceedance determination is made. When the oligomer concentration is greater than or equal to the preset oligomer threshold, a clamping control command is generated based on the oligomer exceedance determination result. The real-time control command is forcibly modified based on the clamping and control command to obtain the modified real-time control command, and the process variables of the C5 process are controlled in real time based on the modified real-time control command.

8. The real-time control method for CPD and DCPD according to claim 1, characterized in that, Also includes: NMR spectra were obtained from the low-field NMR flow cell in the C5 process stream based on a preset acquisition cycle. Characteristic peak analysis was performed based on the inner vinyl proton resonance peak and the outer vinyl proton resonance peak in the nuclear magnetic resonance spectrum to obtain the inner DCPD nuclear magnetic resonance measurement value and the outer DCPD nuclear magnetic resonance measurement value. The nuclear magnetic resonance isomer ratio was determined based on the inner DCPD nuclear magnetic resonance measurement value and the outer DCPD nuclear magnetic resonance measurement value. The proportion of spectral isomers is determined based on the estimated concentrations of the internal DCPD and the external DCPD, and a consistency check is performed based on the proportions of nuclear magnetic resonance isomers and the proportions of spectral isomers to obtain the proportion deviation value. When the proportional deviation value is greater than or equal to the preset internal correction threshold, a paired sample set is extracted from historical online analysis measurement data and nuclear magnetic resonance spectra based on the preset paired sample window; The pre-calibrated multivariate correction model is corrected and updated based on the paired sample set to obtain the updated multivariate correction model. The updated multivariate correction model is then used to estimate the multi-component concentrations of subsequent real-time online analytical measurement data.

9. A real-time control system for CPD and DCPD, characterized in that, include: The data acquisition module is used to acquire online analytical measurement data and process variable data from the C5 process stream containing cyclopentadiene CPD, internal DCPD and external DCPD; The spectral analysis module is used to input the online analysis measurement data into a pre-calibrated multivariate correction model, so that the multivariate correction model can perform quantitative analysis based on the online analysis measurement data and output the CPD concentration estimate, the internal DCPD concentration estimate, and the external DCPD concentration estimate, as well as the corresponding CPD prediction uncertainty, internal DCPD prediction uncertainty, and external DCPD prediction uncertainty. The index determination module is used to determine the current CPD product index and the current DCPD endo- and endo-type isomer index based on the CPD concentration estimate, the endo-type DCPD concentration estimate, and the extrin-type DCPD concentration estimate. The state estimation module is used to input the estimated CPD concentration, the estimated internal DCPD concentration, the estimated external DCPD concentration, and their respective prediction uncertainties and the process variable data into a preset kinetic state estimator, and output the current component concentration state and the future concentration prediction trajectory. The optimization control module is used to input the current component concentration state, future concentration prediction trajectory, current CPD product index, and current DCPD endo- and extrinsic isomer index to the preset model prediction controller, so that the preset model prediction controller uses the preset CPD product index value and the preset DCPD endo- and extrinsic isomer target value as control targets, performs rolling optimization solution under preset process constraints, and generates real-time control commands. The process control module is used to control the process variables of the C5 process in real time based on the real-time control command.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a real-time control method for CPD and DCPD as described in any one of claims 1-8.

Images