A control method for a supercritical fluid chromatography system

Abstract

The present application relates to a kind of control methods for supercritical fluid chromatography system, comprising: the real-time operating parameter of supercritical fluid chromatography system is collected and pretreated;Calculate supercritical fluid phase state perception parameter and according to phase state perception parameter dynamically divide fluid working area;Establish phase state driven dynamic coupling control matrix;The state prediction model with thermodynamic constraint is constructed, and the trend of fluid phase state change is predicted;Based on the prediction result, perform forward-looking pre-intervention control and multivariable collaborative adjustment;Based on chromatographic separation performance, perform closed-loop feedback optimization and update online self-adaptation by model.The present application can effectively inhibit the fluid property mutation in supercritical sensitive area, multi-parameter coupling disturbance and system control oscillation, improve the stability of supercritical fluid state and chromatographic separation reproducibility, realize the stable control of high precision, self-adaptive and forward-looking of supercritical fluid chromatography system.

Description

Technical Field

[0001] This invention relates to a control method for a supercritical fluid chromatography system, and belongs to the technical field of supercritical fluid chromatography. Background Technology

[0002] Supercritical fluid chromatography (SFC) utilizes the properties of supercritical fluids, which combine the high diffusivity of gases and the high solubility of liquids, to achieve efficient separation of complex components and chiral compounds. It has advantages such as high separation efficiency, low consumption of organic solvents, and environmental friendliness, and is widely used in pharmaceuticals, food, environmental monitoring, and fine chemicals.

[0003] Existing SFC systems typically maintain the supercritical stable state of the mobile phase by adjusting process parameters such as system pressure, column temperature, back pressure, and modifier ratio, using single-variable PID or fixed-parameter control methods. However, supercritical fluids exhibit significant nonlinear and multi-parameter coupling characteristics near the critical region. Traditional control methods struggle to respond promptly to changes in fluid state and effectively suppress coupled disturbances, easily leading to problems such as retention time drift, peak shape deterioration, decreased separation, and reduced batch reproducibility.

[0004] Therefore, there is an urgent need for a control method suitable for supercritical fluid chromatography systems that can effectively predict and control the trend of fluid state changes based on the nonlinear thermodynamic characteristics of supercritical fluids near the critical region, and improve the control stability and consistency of the separation process under multi-parameter coupling conditions, thereby reducing the impact of fluid state fluctuations on chromatographic separation performance and improving system operation stability and separation effect. Summary of the Invention

[0005] To address the aforementioned problems, this invention discloses a control method for a supercritical fluid chromatography system, the specific technical solution of which is as follows: A control method for a supercritical fluid chromatography system includes the following steps: Step (1) Collect and preprocess the real-time operating parameters of the supercritical fluid chromatography system; Step (2) Calculate the supercritical fluid phase state sensing parameters based on the preprocessed parameters; Step (3) Dynamically divide the fluid working area according to the phase state sensing parameters; Step (4) Construct the phase-driven dynamic coupling control matrix; Step (5) Construct a time-series state prediction model with thermodynamic constraints to predict the trend of fluid phase change; Step (6) Execute prospective pre-intervention control and multivariate collaborative adjustment based on the prediction results; Step (7) Perform closed-loop feedback optimization based on chromatographic separation performance.

[0006] Furthermore, in step (1), the preprocessing specifically includes: Step (1.1) Outlier handling: Set reasonable value ranges for each real-time operating parameter, determine data that exceeds the corresponding value range as outliers and remove them; the missing values ​​include missing data generated during the original acquisition process and missing data formed after outlier removal; for data with a number of consecutive missing points lower than the preset threshold, linear interpolation is used to complete the data, and for data with a number of consecutive missing points that reach or exceed the preset threshold, sliding window mean, historical data under the same working conditions or model prediction results are used to complete the data. Step (1.2) Normalization: The parameters are mapped to a fixed range by a preset normalization method to eliminate the impact of differences in the numerical range of different parameters on model training; By using a pre-defined normalization method, each running parameter is mapped to a fixed numerical range to eliminate the impact of differences in the units and ranges of different parameters on model training. Step (1.3) Time series reconstruction: Arrange the preprocessed running parameters into a time series according to the preset sampling time step Δt, and construct the model input samples using a sliding time window; Let the time window length be L, then the k-th input sample is represented as: , in: Let i represent the multidimensional running parameter vector at the i-th sampling time, where i∈[k,k+L-1]; L represents the length of the input time window; Adjacent samples move forward according to a preset sliding step size S to form continuous training samples; The corresponding prediction target is the running parameters at the next time step after the window ends: , Therefore, the following can be constructed: , The training dataset for the time series prediction model is composed of these datasets.

[0007] Furthermore, in step (2), the formula for calculating the phase sensitivity index is: , in, Indicates the compression factor. Represents the density gradient. This represents the rate of dynamic change of pressure. Indicates the rate of dynamic change of temperature. These are the weighting coefficients corresponding to the density gradient, pressure dynamic change rate, temperature dynamic change rate, and compressibility factor deviation, respectively, used to reflect the degree of influence of each physical quantity on the phase sensitivity index.

[0008] Furthermore, in step (3), the work area is divided as follows: Normal region: The fluid is far from the critical state, and the compressibility factor, density gradient and phase sensitivity are all below the preset threshold, with weak parameter coupling; Subcritical coupling region: The fluid approaches the critical state, the compressibility factor is in the critical range, the density gradient and phase sensitivity gradually increase, and the parameter coupling gradually strengthens. Supercritical sensitive region: Both density gradient and phase sensitivity exceed the corresponding preset critical threshold, and the nonlinear characteristics of fluid properties are significant.

[0009] Furthermore, step (4) specifically includes the following steps: Step (4.1) Establish a dynamic coupling control matrix based on the coupling characteristics of different phase regions: , in, Let be the fluid state deviation vector, state deviation vector This includes at least one of the following: fluid density deviation, fluid viscosity deviation, pressure deviation, and temperature deviation; This refers to the system control adjustment vector. Including at least one of the following: supercritical fluid flow rate adjustment, column box temperature adjustment, system back pressure adjustment, and modifier ratio adjustment; It is a dynamically coupled control matrix that is dynamically and adaptively updated according to real-time pressure, temperature, compressibility factor, and phase sensitivity; Step (4.2) Dynamically adjust the matrix structure according to different phase regions: Regular area: The coupling matrix is ​​mainly composed of diagonal elements, while the off-diagonal cross-coupling elements have extremely small values, and lightweight control logic is used; Subcritical coupling region: Gradually increase the weights of off-diagonal cross-coupling elements in the matrix to enhance decoupling control capability; Supercritical sensitive region: Significantly improves the overall matrix gain and the weight of off-diagonal cross-coupled elements, suppressing strong coupling disturbances in the critical region; Meanwhile, linkage constraints are introduced into the actuator within the supercritical sensitive region to limit the rate of back pressure change and the slope of flow rate change, thereby preventing control overshoot from causing new phase fluctuations.

[0010] Furthermore, in step (5), the time-series state prediction model only adopts the architecture of time-series feature fitting and physical constraint correction, and is not limited to a single algorithm structure. It can be adapted to any industrial-grade time-series network such as LSTM, GRU, and Transformer.

[0011] Furthermore, in step (6), the prospective pre-intervention control process is as follows: when the prediction model determines that the system will transition from the subcritical coupling region to the supercritical sensitive region within a preset prediction time, at least one parameter among the supercritical fluid flow rate, system back pressure, column oven temperature and modifier ratio is adjusted before the predicted transition occurs to reduce the system phase sensitivity, slow down the fluid phase change process, and improve the operational stability of the supercritical fluid chromatography system.

[0012] Furthermore, in step (6), the specific process of multivariate collaborative regulation is as follows: First, based on the fluid state parameters and chromatographic separation performance parameters obtained from real-time monitoring, the deviation between the current operating state and the target state is calculated: , , , , in, Given the current fluid density, For target density; Given the current fluid viscosity, Target viscosity; To retain the time in real time, Reserve time for the target; For real-time separation, For the target separation, For fluid density error, For fluid viscosity error, This is due to chromatographic retention time error. This refers to the error in chromatographic resolution. Based on the above deviations, establish the industrial control objective function: ; in, Density control weights; For viscosity control weights; Optimize weights to retain time; Optimize weights for separation; According to the industrial control objective function To optimize the solution for the control variables, the control variables are set as follows: , in: This refers to the carbon dioxide flow rate; For modifier flow rate; This refers to the column temperature. Back pressure; Establish a sensitivity mapping relationship between the objective function and the control variables, and determine the adjustment direction of the control variables based on the changing trend of the objective function with respect to each control variable: , in, For the first Adjustment amount of control variables within each control cycle For coordinated adjustment of the gain matrix, The gradient of the objective function with respect to the control variables is expressed as follows: , Update the control parameters for the next control cycle based on the control variable adjustment amount: , in, For the control variables of the current control cycle, The control variables for the next optimized control cycle; The specific adjustment process includes: (1) When the fluid density error For the objective function When the contribution is greatest, increase the density control weight. This allows the control system to prioritize reducing density deviation; the back pressure is adjusted according to the direction of the objective function gradient change. and carbon dioxide flow By changing the density of the supercritical fluid, the actual density gradually approaches the target density; (2) When the fluid viscosity error For the objective function When the contribution is greatest, increase the viscosity control weight. Prioritize adjusting the flow rate of the modifier. With carbon dioxide flow The proportional relationship between these components can be used to optimize viscosity parameters by altering fluid polarity and solvation ability. (3) When the chromatographic retention time error For the objective function When the contribution is greatest, increase the retention time to optimize the weight. By synergistically adjusting fluid density, modifier ratio and back pressure, the distribution behavior of the target component between the stationary phase and the mobile phase is changed, so that the actual retention time converges to the target retention time. (4) When the chromatographic resolution error For the objective function When the contribution is maximized, the separation degree is increased to optimize the weights. By optimizing column temperature Back pressure In addition, the proportion of modifiers can reduce peak broadening and component interactions, thereby improving the separation performance between target components; Through the above process, the synergistic optimization of CO2 flow rate, modifier ratio, column temperature, and back pressure is achieved, making the objective function... Continue to decrease until the preset convergence condition is met: , Complete one multivariate collaborative regulation.

[0013] Furthermore, in step (7), by collecting chromatographic retention time, peak width, peak tailing factor and resolution in real time, the fluid target physical property parameters and coupling control matrix weights are dynamically corrected according to the measured chromatographic separation performance. The control parameters in steps (4) and (5) are updated using the feedback results of step (7), and the cyclic control of steps (5) to (7) continues to be executed.

[0014] Furthermore, the specific process of dynamically correcting the fluid target physical property parameters and coupling control matrix weights based on the measured chromatographic separation performance is as follows: 1) Establish a chromatographic separation performance evaluation model, and use the retention time acquired in real time. Peak width Peak tailing factor and resolution The separation performance deviation is calculated by comparing it with the preset target value. , in, , , , These are the target retention time, target peak width, target peak tailing factor, and target separation, respectively. for , , , Corresponding evaluation weights; 2) Based on the separation performance deviation, feedback correction is performed on the target fluid properties: When the chromatographic retention time deviates from the target value, the density and viscosity parameters of the target fluid are adjusted according to the direction of the deviation. Specifically, when the actual retention time is greater than the target retention time, the density of the target fluid is increased to enhance the solubilization ability and reduce the retention time of the target substance; when the actual retention time is less than the target retention time, the density of the target fluid is decreased to allow the target substance to maintain a longer retention time. When an increase in peak width or a decrease in separation is detected, the stability constraint of the target fluid density is increased, and the control weights corresponding to the rate of pressure and temperature change are reduced, so that the fluid state changes more smoothly. When an increase in peak tailing factor is detected, the target parameter of the corrector ratio is adjusted according to the peak shape change trend, and the mass transfer process of the target component is improved by changing the fluid polarity and solubility selectivity. The specific correction process is represented as follows: , in: These are the target physical properties of the fluid, including target density, target viscosity, target compressibility factor, and target modifier ratio; This is the performance deviation feedback gain matrix; Indicates the current optimization cycle; 3) Update the weights of the coupling control matrix online based on the chromatographic separation performance feedback results: The coupling control matrix is ​​used to describe the mapping relationship between changes in target physical property parameters and control variables: , in, This includes carbon dioxide flow rate regulation, corrective agent flow rate regulation, column temperature regulation, and back pressure regulation; Including density, viscosity, pressure, and temperature deviation; The matrix weights are adaptively adjusted based on the real-time separation performance deviation: , in, For online learning speed; When the system detects a decrease in separation, the control weights for density changes and corrector ratio changes are increased; when an increase in peak width is detected, the control weights related to temperature stability and pressure stability are increased; when an increase in peak tailing is detected, the control weight for corrector ratio is increased.

[0015] The beneficial effects of this invention are: (1) By acquiring supercritical fluid phase state sensing parameters, this invention realizes the quantitative characterization of fluid operating state and critical sensitivity, which is beneficial to improving the system's ability to identify complex working conditions and its control adaptability.

[0016] (2) By establishing a phase-driven dynamic coupling control model, the present invention can adaptively adjust the control strategy according to different working areas, thereby reducing the coupling influence between parameters such as pressure, temperature, flow rate and modifier ratio, and improving the stability of multivariable control.

[0017] (3) This invention predicts the trend of fluid phase change by constructing a state prediction model with thermodynamic constraints, and combines it with forward-looking pre-intervention control, which helps to reduce the hysteresis effect of traditional feedback control and improve the dynamic response capability of the system.

[0018] (4) By introducing a multi-variable collaborative control mechanism, this invention realizes the linkage adjustment of multiple actuators, improves the control accuracy of supercritical fluid operation status, and reduces the impact of process disturbances on system operation.

[0019] (5) By establishing a closed-loop feedback optimization mechanism based on chromatographic separation performance, this invention combines fluid state control with chromatographic separation effect, which can improve retention time stability, resolution and batch reproducibility. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the method described in this invention; Figure 2 Comparison of phase sensitivity change curves; Figure 3 Comparison of chromatographic UV response curves; Figure 4 Comparison of retention time drift curves. Detailed Implementation

[0021] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0022] A control method for a supercritical fluid chromatography system includes the following steps: Step (1) Collect and preprocess the real-time operating parameters of the supercritical fluid chromatography system: Preprocessing specifically includes: Step (1.1) Outlier handling: Set reasonable value ranges for each real-time operating parameter, determine data that exceeds the corresponding value range as outliers and remove them; the missing values ​​include missing data generated during the original acquisition process and missing data formed after outlier removal; for data with a number of consecutive missing points lower than the preset threshold, linear interpolation is used to complete the data, and for data with a number of consecutive missing points that reach or exceed the preset threshold, sliding window mean, historical data under the same working conditions or model prediction results are used to complete the data. Step (1.2) Normalization: The running parameters are mapped to a fixed numerical range by a preset normalization method to eliminate the impact of differences in the units and numerical ranges of different parameters on model training.

[0023] 1) Preferably, the mapping is performed using a minimum-maximum normalization method, and its calculation formula is as follows: , in, This is the original data; The minimum value in the historical or training data for the corresponding parameter; The maximum value of the corresponding parameter in historical or training data; The data is normalized and its value range is [0,1].

[0024] Mapping process: For each running parameter, iterate through all time series samples to find the global maximum and minimum values ​​of the feature; perform linear scaling point by point according to the above formula, and uniformly constrain the parameter value range to [0,1]; retain the independent extreme value mapping relationship of each parameter, and use the same extreme value to perform inverse normalization to restore the true physical dimensions in the subsequent model prediction stage.

[0025] 2) Alternatively, Z-score standardization can be used: , in, The mean, The standard deviation is denoted as .

[0026] Step (1.3) Time series reconstruction: Arrange the preprocessed running parameters into a time series according to the preset sampling time step Δt, and construct the model input samples using a sliding time window; Let the time window length be L, then the k-th input sample is represented as: , in: Let i represent the multidimensional running parameter vector at the i-th sampling time, where i∈[k,k+L-1]; L represents the length of the input time window; Adjacent samples move forward according to a preset sliding step size S to form continuous training samples; The corresponding prediction target is the running parameters at the next time step after the window ends: , Therefore, the following can be constructed: , The training dataset for the time series prediction model is composed of these datasets.

[0027] Step (2) Calculate the supercritical fluid phase state sensing parameters based on the preprocessed parameters: The formula for calculating the phase sensitivity index is: , in, Indicates the compression factor. Represents the density gradient. This represents the rate of dynamic change of pressure. Indicates the rate of dynamic change of temperature. These are the weighting coefficients corresponding to the density gradient, pressure dynamic change rate, temperature dynamic change rate, and compressibility factor deviation, respectively, used to reflect the degree of influence of each physical quantity on the phase sensitivity index.

[0028] Step (3) Dynamically divide the fluid working region according to the phase state sensing parameters: The work area is divided as follows: Normal region: The fluid is far from the critical state, and the compressibility factor, density gradient and phase sensitivity are all below the preset threshold, with weak parameter coupling; Subcritical coupling region: The fluid approaches the critical state, the compressibility factor is in the critical range, the density gradient and phase sensitivity gradually increase, and the parameter coupling gradually strengthens. Supercritical sensitive region: Both density gradient and phase sensitivity exceed the corresponding preset critical threshold, and the nonlinear characteristics of fluid properties are significant.

[0029] Step (4) Constructing the phase-driven dynamic coupling control matrix, specifically including the following steps: Step (4.1) Establish a dynamic coupling control matrix based on the coupling characteristics of different phase regions: , in, Let be the fluid state deviation vector, state deviation vector This includes at least one of the following: fluid density deviation, fluid viscosity deviation, pressure deviation, and temperature deviation; This refers to the system control adjustment vector. Including at least one of the following: supercritical fluid flow rate adjustment, column box temperature adjustment, system back pressure adjustment, and modifier ratio adjustment; It is a dynamically coupled control matrix that is dynamically and adaptively updated according to real-time pressure, temperature, compressibility factor, and phase sensitivity; Step (4.2) Dynamically couple the control matrix structure and control strategy according to different phase state regions: Regular area: The coupling matrix is ​​mainly composed of diagonal elements, while the off-diagonal cross-coupling elements have extremely small values, and lightweight control logic is used; Subcritical coupling region: Gradually increase the weights of off-diagonal cross-coupling elements in the matrix to enhance decoupling control capability; Supercritical sensitive region: Significantly improves the overall matrix gain and the weight of off-diagonal cross-coupled elements, suppressing strong coupling disturbances in the critical region; Meanwhile, linkage constraints are introduced into the actuator within the supercritical sensitive region to limit the rate of back pressure change and the slope of flow rate change, thereby preventing control overshoot from causing new phase fluctuations.

[0030] The matrix structure of each of the above-mentioned different phase regions is determined according to the following method: Based on phase sensitivity parameters The current supercritical fluid operating state is determined by region, and the correlation weights between the control variables in the coupling control matrix are adjusted according to different phase regions.

[0031] The coupling control matrix is ​​represented as: , in: , To control the adjustment amount of the variable; , This refers to the fluid state deviation. This is the dynamic coupling control matrix that varies with the phase state region.

[0032] The specific strategies and processes for each of the above work areas are explained in detail below: (1) Conventional area control strategy When the system is in the normal range, the fluid density, viscosity and thermodynamic state change relatively slowly, and the coupling effect between the control variables is weak.

[0033] At this point, the coupling control matrix is ​​set to a weakly coupled form: , in, To adjust the gain according to the corresponding control variable itself, Let be the cross-coupling compensation coefficient, and: , That is, adjustments are mainly made based on univariate deviations.

[0034] During execution, the back pressure is adjusted first based on the density deviation, and the pump output is adjusted based on the flow deviation, with only a small amount of cross compensation to reduce the amount of control calculation and improve the system response speed.

[0035] (2) Subcritical coupling region control strategy When the system enters the subcritical coupling region, the effects of pressure and temperature changes on fluid density and viscosity increase, and the mutual influence between different control variables increases.

[0036] At this point, the cross-coupling elements in the enhancement matrix are based on real-time state changes: , Among them, the coupling compensation terms between back pressure and density change are increased; the coupling compensation terms between CO2 flow rate and density change are increased; and the compensation terms between temperature change and viscosity change are increased.

[0037] Right now: , During the control process, instead of adjusting a single actuator, the back pressure, CO2 flow rate and column temperature are simultaneously calculated based on the target density deviation to achieve multivariable decoupled control.

[0038] (3) Control strategy for supercritical sensitive area When the system enters the supercritical sensitive region, the fluid state parameters are highly sensitive to pressure and temperature disturbances, and small changes in control can lead to rapid changes in density. Therefore, it is necessary to increase the overall gain of the coupled control matrix and enhance off-diagonal coupling compensation. , in, Based on the control gain matrix; Enhancement compensation matrix for critical regions.

[0039] Specifically, the following aspects are improved: the compensation weight for pressure changes on density changes; the compensation weight for temperature changes on density and viscosity changes; and the compensation weight for changes in the proportion of modifiers on mass transfer performance.

[0040] This enables the control system to predict in advance the impact of a change in a certain variable on other state parameters and adjust the relevant actuators in advance to reduce the risk of abrupt phase changes.

[0041] Simultaneously, actuator linkage constraints are introduced in the supercritical sensitive region: , , in, To limit the rate of change of back pressure; Limit the rate of change of flow.

[0042] When a pressure or flow rate change is detected to exceed the limit threshold, the control gain is automatically reduced to limit the adjustment range of the actuator, so as to avoid the fluid state from crossing the sensitive area due to excessively rapid adjustment, which could cause new pressure fluctuations or phase oscillations.

[0043] Step (5) Construct a time-series state prediction model with thermodynamic constraints to predict the trend of fluid phase change: The time-series state prediction model only adopts the architecture of time-series feature fitting and physical constraint correction. It is not limited to a single algorithm structure and can be adapted to any industrial-grade time-series network such as LSTM, GRU, and Transformer.

[0044] Step (6) Execute prospective pre-intervention control and multivariate collaborative adjustment based on the prediction results: The prospective pre-intervention control process is as follows: When the prediction model determines that the system will transition from the subcritical coupling region to the supercritical sensitive region within a preset prediction time, at least one parameter among the supercritical fluid flow rate, system back pressure, column oven temperature, and modifier ratio is adjusted before the predicted transition occurs to reduce the system's phase sensitivity, slow down the fluid phase transition process, and improve the operational stability of the supercritical fluid chromatography system.

[0045] The specific process of multivariate collaborative regulation is as follows: First, based on the fluid state parameters and chromatographic separation performance parameters obtained from real-time monitoring, the deviation between the current operating state and the target state is calculated: , , , , in, Given the current fluid density, For target density; Given the current fluid viscosity, Target viscosity; To retain the time in real time, Reserve time for the target; For real-time separation, For the target separation, For fluid density error, For fluid viscosity error, This is due to chromatographic retention time error. This refers to the error in chromatographic resolution. Based on the above deviations, establish the industrial control objective function: , in, Density control weights; For viscosity control weights; Optimize weights to retain time; Optimize weights for separation; According to the industrial control objective function To optimize the solution for the control variables, the control variables are set as follows: , in: This refers to the carbon dioxide flow rate; For modifier flow rate; This refers to the column temperature. Back pressure; Establish a sensitivity mapping relationship between the objective function and the control variables, and determine the adjustment direction of the control variables based on the changing trend of the objective function with respect to each control variable: , in, For the first Adjustment amount of control variables within each control cycle For coordinated adjustment of the gain matrix, The gradient of the objective function with respect to the control variables is expressed as follows: , Update the control parameters for the next control cycle based on the control variable adjustment amount: , in, For the control variables of the current control cycle, The control variables for the next optimized control cycle. The specific adjustment process includes: (1) When the fluid density error For the objective function When the contribution is greatest, increase the density control weight. This allows the control system to prioritize reducing density deviation; the back pressure is adjusted according to the direction of the objective function gradient change. and carbon dioxide flow By changing the density of the supercritical fluid, the actual density gradually approaches the target density; (2) When the fluid viscosity error For the objective function When the contribution is greatest, increase the viscosity control weight. Prioritize adjusting the flow rate of the modifier. With carbon dioxide flow The proportional relationship between these components can be used to optimize viscosity parameters by altering fluid polarity and solvation ability. (3) When the chromatographic retention time error For the objective function When the contribution is greatest, increase the retention time to optimize the weight. By synergistically adjusting fluid density, modifier ratio and back pressure, the distribution behavior of the target component between the stationary phase and the mobile phase is changed, so that the actual retention time converges to the target retention time. (4) When the chromatographic resolution error For the objective function When the contribution is maximized, the separation degree is increased to optimize the weights. By optimizing column temperature Back pressure In addition, the proportion of modifiers can reduce peak broadening and component interactions, thereby improving the separation performance between target components; Through the above process, the synergistic optimization of CO2 flow rate, modifier ratio, column temperature, and back pressure is achieved, making the objective function... Continue to decrease until the preset convergence condition is met: , Complete one multivariate collaborative regulation.

[0046] Step (7) Perform closed-loop feedback optimization based on chromatographic separation performance: By collecting chromatographic retention time, peak width, peak tailing factor and resolution in real time, the fluid target physical property parameters and coupling control matrix weights are dynamically corrected based on the measured chromatographic separation performance. The control parameters in steps (4) and (5) are updated using the feedback results of step (7), and the cyclic control of steps (5) to (7) continues.

[0047] The specific process of dynamically correcting the fluid target physical property parameters and coupling control matrix weights based on measured chromatographic separation performance is as follows: 1) Establish a chromatographic separation performance evaluation model, and use the retention time acquired in real time. Peak width Peak tailing factor and resolution The separation performance deviation is calculated by comparing it with the preset target value. , in, , , , These are the target retention time, target peak width, target peak tailing factor, and target separation, respectively. for , , , Corresponding evaluation weights; 2) Based on the separation performance deviation, feedback correction is performed on the target fluid properties: When the chromatographic retention time deviates from the target value, the density and viscosity parameters of the target fluid are adjusted according to the direction of the deviation. Specifically, when the actual retention time is greater than the target retention time, the density of the target fluid is increased to enhance the solubilization ability and reduce the retention time of the target substance; when the actual retention time is less than the target retention time, the density of the target fluid is decreased to allow the target substance to maintain a longer retention time. When an increase in peak width or a decrease in separation is detected, the stability constraint of the target fluid density is increased, and the control weights corresponding to the rate of pressure and temperature change are reduced, so that the fluid state changes more smoothly. When an increase in peak tailing factor is detected, the target parameter of the corrector ratio is adjusted according to the peak shape change trend, and the mass transfer process of the target component is improved by changing the fluid polarity and solubility selectivity. The specific correction process is represented as follows: , in: These are the target physical properties of the fluid, including target density, target viscosity, target compressibility factor, and target modifier ratio; This is the performance deviation feedback gain matrix; Indicates the current optimization cycle; 3) Update the weights of the coupling control matrix online based on the chromatographic separation performance feedback results: The coupling control matrix is ​​used to describe the mapping relationship between changes in target physical property parameters and control variables: , in, This includes carbon dioxide flow rate regulation, corrective agent flow rate regulation, column temperature regulation, and back pressure regulation; Including density, viscosity, pressure, and temperature deviation; The matrix weights are adaptively adjusted based on the real-time separation performance deviation: , in, This refers to the online learning speed.

[0048] When the system detects a decrease in separation, the control weights for density changes and corrector ratio changes are increased; when an increase in peak width is detected, the control weights related to temperature stability and pressure stability are increased; when an increase in peak tailing is detected, the control weight for corrector ratio is increased.

[0049] To more clearly demonstrate the objectives, technical solutions, and advantages of this invention, the technical solutions in the embodiments of this application are now clearly and completely described in conjunction with the accompanying drawings and specific examples. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this application.

[0050] This embodiment verifies the control method of the present invention using a supercritical fluid chromatography (SFC) experimental platform. A 10 mm × 250 mm semi-preparative column was used. The mobile phase consisted of supercritical CO2 and methanol entrainer, with a constant total flow rate of 8 mL / min, of which the CO2 flow rate was 6.4 mL / min and the methanol entrainer flow rate was 1.6 mL / min (20% by volume). The system back pressure was set to 10 MPa, and the column temperature was set to 35 °C.

[0051] The control process of this invention is as follows: Figure 1 As shown, by real-time sensing of the system's phase state information, region division is completed according to operating conditions, and control parameters are pre-adjusted based on system state prediction results. Then, feedback information is used for closed-loop correction, achieving stable control of the supercritical fluid chromatography system. The specific implementation steps are as follows: (1) During the continuous operation of the equipment, the key operating parameters of the SFC system are continuously collected with a sampling period of 1 second, including system pressure, fluid temperature, CO2 mobile phase flow rate, modifier flow rate and chromatographic ultraviolet detection signal, to obtain continuous time-series operating data and provide the original data basis for subsequent status analysis and control.

[0052] (2) For the original time series data collected, outlier removal, missing data compensation, parameter normalization and time series window reconstruction are performed in sequence to filter out equipment sampling noise and random disturbances, unify the dimensions of each parameter and data scale, and construct a stable, reliable, and standardized time series dataset that can be used for model calculation.

[0053] (3) Calculate the compressibility factor of supercritical fluid in real time based on the preprocessed time series data. density gradient Dynamic change rate of pressure and the rate of change of temperature A phase sensitivity index is constructed through a weighted fusion method. : , in: The calculation results are shown in Table 1.

[0054] Table 1 Calculation results of phase sensing parameters

[0055] As shown in Table 1, as the system operates, the compressibility factor gradually decreases, the density gradient gradually increases, and the phase sensitivity S gradually increases from 0.296 to 0.842, indicating that the fluid thermodynamic sensitivity is continuously enhanced.

[0056] (4) Based on the phase sensitivity index S obtained from real-time calculation, the fluid operating state is dynamically partitioned, wherein: This is a regular area; This is the subcritical coupling region; This is the supercritical sensitive region. Therefore, the system is in the normal region at 12s; it enters the subcritical coupling region at 18s; and it enters the supercritical sensitive region at 30s. The controller automatically adjusts the dynamic coupling control matrix parameters according to the operating region.

[0057] (5) Based on the thermodynamic coupling characteristics corresponding to different working regions, establish the phase-driven dynamic coupling control matrix: , in, This is the fluid state deviation vector; This is the system control adjustment vector; It is a dynamically coupled control matrix that is dynamically and adaptively updated according to real-time pressure, temperature, compressibility factor, and phase sensitivity.

[0058] When the fluid enters the normal region, a low-coupling control strategy is adopted to improve the system response speed; when the fluid enters the subcritical coupling region, the multi-parameter coupling compensation capability is enhanced to improve the system's adaptability to phase changes; when the fluid enters the supercritical sensitive region, the weight of multi-variable collaborative control is increased to strengthen the collaborative adjustment capability between pressure, flow rate, temperature and modifier ratio, suppress strong coupling disturbances, and maintain fluid state stability.

[0059] (6) Based on historical operating data and thermodynamic constraints, a time-series state prediction model was constructed to predict the phase change trend of supercritical fluid within future time windows. An LSTM network was used to establish the prediction model. The network structure consists of an input layer, two hidden layers, and an output layer. A Dropout layer was set between the hidden and output layers to effectively reduce the risk of model overfitting. The input layer has 5 neurons, corresponding to CO2 flow rate, modifier flow rate, system back pressure, column oven temperature, and phase sensitivity S, respectively. Each hidden layer has 64 neurons, and ReLU is selected as the activation function. The output layer has 2 neurons, which output the predicted value of phase sensitivity and the state of the fluid working area at future time.

[0060] The mean squared error (MSE) is used as the loss function, and the formula is as follows: , in, For the sample size, These are the model's predicted values. The values ​​are actual values; the optimizer is Adam, the learning rate is set to 0.001, and the batch size is set to 32; the preprocessed dataset is divided into training, validation, and test sets in a ratio of 7:2:1; the number of training iterations is set to 100-200 rounds, and training is stopped when the validation set loss function does not decrease for 10 consecutive rounds, and the optimal model is saved; when the system runs for 24 seconds, the time series state prediction model predicts that the phase sensitivity will exceed 0.8 within the next 10 seconds based on historical running data, corresponding to the system about to enter the supercritical sensitive region.

[0061] (7) Based on the phase change trend predicted by the model, pre-intervention regulation is performed in advance before the fluid fully enters the supercritical sensitive region. The back pressure valve opening, CO2 flow rate and entrainer ratio are synergistically adjusted. The specific adjustment process is as follows: the system back pressure is increased from 10.0 MPa to 10.2 MPa, the entrainer ratio is reduced from 20% to 18%, and the CO2 flow rate is increased from 6.4 mL / min to 6.56 mL / min, so as to reduce the growth rate of phase sensitivity and suppress the trend of sudden change in fluid properties.

[0062] Construct a multi-objective optimization control objective function: , in, For fluid density error; This is for fluid viscosity error; This is due to chromatographic retention time error; This refers to the error in chromatographic resolution. These represent the optimization weights corresponding to fluid state stability, mass transfer performance, and chromatographic separation performance, respectively.

[0063] Using fluid property deviation and chromatographic separation deviation as optimization indicators, the controller obtains the optimal multivariate coordinated adjustment amount according to the objective function to achieve decoupled coordinated control of pressure, flow rate, and temperature. At the same time, the adjustment rates of the back pressure regulating valve, flow pump, and column temperature module are dynamically constrained to avoid control overshoot and secondary transition of fluid phase caused by excessively rapid linkage of multiple mechanisms, thus ensuring smooth transition of operating conditions.

[0064] (8) By collecting chromatographic separation indicators such as chromatographic retention time, peak width, peak tailing factor and resolution in real time, the fluid target physical property parameters and the weight of the coupling control matrix are dynamically corrected according to the measured chromatographic separation performance to achieve closed-loop feedback optimization.

[0065] This comparative example employs a traditional single-variable PID control strategy, using only the system back pressure as the sole control objective, and achieving pressure regulation by adjusting the opening of the back pressure regulating valve. All other experimental conditions in this comparative example are completely consistent with those in the embodiment of this invention; the PID controller parameters are set as follows: proportional coefficient... Integral coefficient Differential coefficients After the system has been running continuously, various performance indicators are collected and compared with those of the embodiments of the present invention. The comparison results are as follows: Figures 2-4 As shown in Table 2, the statistical data for each indicator are presented.

[0066] Table 2 Comparison of Results Pressure fluctuation (MPa) ±0.12 ±0.08 Retention time fluctuation ±1.84% ±1.08% Separation fluctuation ±8.5% ±5.6% Peak tailing factor 1.42 1.22 Number of times entering sensitive areas 6 times 4 times Control overshoot 11.2% 7.3% The results are shown in Table 2. This invention, by constructing phase-sensing parameters, a dynamic coupling control model, and a thermodynamic constraint prediction model, can identify the phase change trend of supercritical fluids in advance and implement proactive pre-intervention control before entering the supercritical sensitive region. Compared with traditional PID control, this invention can reduce pressure fluctuations by approximately 30%–40%, retention time fluctuations by approximately 40%, resolution fluctuations by approximately 30%, and control overshoot by approximately 30%, effectively reducing the number of times the system enters the supercritical sensitive region and improving system stability and chromatographic separation consistency.

[0067] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0068] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A control method for a supercritical fluid chromatography system, characterized in that, Includes the following steps: Step (1) Collect and preprocess the real-time operating parameters of the supercritical fluid chromatography system; Step (2) Calculate the supercritical fluid phase state sensing parameters based on the preprocessed parameters; Step (3) Dynamically divide the fluid working area according to the phase state sensing parameters; Step (4) Construct the phase-driven dynamic coupling control matrix; Step (5) Construct a time-series state prediction model with thermodynamic constraints to predict the trend of fluid phase change; Step (6) Execute prospective pre-intervention control and multivariate collaborative adjustment based on the prediction results; Step (7) Perform closed-loop feedback optimization based on chromatographic separation performance.

2. The control method for a supercritical fluid chromatography system according to claim 1, characterized in that, In step (1), the preprocessing specifically includes: Step (1.1) Outlier handling: Set reasonable value ranges for each real-time operating parameter, determine data that exceeds the corresponding value range as outliers and remove them; the missing values ​​include missing data generated during the original acquisition process and missing data formed after outlier removal; for data with a number of consecutive missing points lower than the preset threshold, linear interpolation is used to complete the data, and for data with a number of consecutive missing points that reach or exceed the preset threshold, sliding window mean, historical data under the same working conditions or model prediction results are used to complete the data. Step (1.2) Normalization: The parameters are mapped to a fixed range by a preset normalization method to eliminate the impact of differences in the numerical range of different parameters on model training; By using a pre-defined normalization method, each running parameter is mapped to a fixed numerical range to eliminate the impact of differences in the units and ranges of different parameters on model training. Step (1.3) Time series reconstruction: Arrange the preprocessed running parameters into a time series according to the preset sampling time step Δt, and construct the model input samples using a sliding time window; Let the time window length be L, then the k-th input sample is represented as: ， in: Let i represent the multidimensional running parameter vector at the i-th sampling time, where i∈[k,k+L-1]; L represents the length of the input time window; Adjacent samples move forward according to a preset sliding step size S to form continuous training samples; The corresponding prediction target is the running parameters at the next time step after the window ends: ， Therefore, the following can be constructed: ， The training dataset for the time series prediction model is composed of these datasets.

3. The control method for a supercritical fluid chromatography system according to claim 1, characterized in that, In step (2), the formula for calculating the phase sensitivity index is: ， in, Indicates the compression factor. Represents the density gradient. This represents the rate of dynamic change of pressure. Indicates the rate of dynamic change of temperature. These are the weighting coefficients corresponding to the density gradient, pressure dynamic change rate, temperature dynamic change rate, and compressibility factor deviation, respectively, used to reflect the degree of influence of each physical quantity on the phase sensitivity index.

4. The control method for a supercritical fluid chromatography system according to claim 3, characterized in that, In step (3), the work area is divided as follows: Normal region: The fluid is far from the critical state, and the compressibility factor, density gradient and phase sensitivity are all below the preset threshold, with weak parameter coupling; Subcritical coupling region: The fluid approaches the critical state, the compressibility factor is in the critical range, the density gradient and phase sensitivity gradually increase, and the parameter coupling gradually strengthens. Supercritical sensitive region: Both density gradient and phase sensitivity exceed the corresponding preset critical threshold, and the nonlinear characteristics of fluid properties are significant.

5. The control method for a supercritical fluid chromatography system according to claim 4, characterized in that, Step (4) specifically includes the following steps: Step (4.1) Establish a dynamic coupling control matrix based on the coupling characteristics of different phase regions: ， in, Let be the fluid state deviation vector, state deviation vector This includes at least one of the following: fluid density deviation, fluid viscosity deviation, pressure deviation, and temperature deviation; This refers to the system control adjustment vector. Including at least one of the following: supercritical fluid flow rate adjustment, column box temperature adjustment, system back pressure adjustment, and modifier ratio adjustment; It is a dynamically coupled control matrix that is dynamically and adaptively updated according to real-time pressure, temperature, compressibility factor, and phase sensitivity; Step (4.2) Dynamically adjust the matrix structure according to different phase regions: Regular area: The coupling matrix is ​​mainly composed of diagonal elements, while the off-diagonal cross-coupling elements have extremely small values, and lightweight control logic is used; Subcritical coupling region: Gradually increase the weights of off-diagonal cross-coupling elements in the matrix to enhance decoupling control capability; Supercritical sensitive region: Significantly improves the overall matrix gain and the weight of off-diagonal cross-coupled elements, suppressing strong coupling disturbances in the critical region; Meanwhile, linkage constraints are introduced into the actuator within the supercritical sensitive region to limit the rate of back pressure change and the slope of flow rate change, thereby preventing control overshoot from causing new phase fluctuations.

6. The control method for a supercritical fluid chromatography system according to claim 5, characterized in that, In step (5), the time-series state prediction model only adopts the architecture of time-series feature fitting and physical constraint correction. It is not limited to a single algorithm structure and can be adapted to any industrial-grade time-series network such as LSTM, GRU, or Transformer.

7. The control method for a supercritical fluid chromatography system according to claim 5, characterized in that, In step (6), the prospective pre-intervention control process is as follows: when the prediction model determines that the system will transition from the subcritical coupling region to the supercritical sensitive region within a preset prediction time, at least one parameter among the supercritical fluid flow rate, system back pressure, column oven temperature and modifier ratio is adjusted before the predicted transition occurs to reduce the system phase sensitivity, slow down the fluid phase change process, and improve the operating stability of the supercritical fluid chromatography system.

8. The control method for a supercritical fluid chromatography system according to claim 1, characterized in that, In step (6), the specific process of multivariate collaborative regulation is as follows: First, based on the fluid state parameters and chromatographic separation performance parameters obtained from real-time monitoring, the deviation between the current operating state and the target state is calculated: ， ， ， ， in, Given the current fluid density, For target density; Given the current fluid viscosity, Target viscosity; To retain the time in real time, Reserve time for the target; For real-time separation, For the target separation, For fluid density error, For fluid viscosity error, This is due to chromatographic retention time error. This refers to the error in chromatographic resolution. Based on the above deviations, establish the industrial control objective function: ； in, Density control weights; For viscosity control weights; Optimize weights to retain time; Optimize weights for separation; According to the industrial control objective function To optimize the solution for the control variables, the control variables are set as follows: ， in: This refers to the carbon dioxide flow rate; For modifier flow rate; This refers to the column temperature. Back pressure; Establish a sensitivity mapping relationship between the objective function and the control variables, and determine the adjustment direction of the control variables based on the changing trend of the objective function with respect to each control variable: ， in, For the first Adjustment amount of control variables within each control cycle For coordinated adjustment of the gain matrix, The gradient of the objective function with respect to the control variables is expressed as follows: ， Update the control parameters for the next control cycle based on the control variable adjustment amount: ， in, For the control variables of the current control cycle, The control variables for the next optimized control cycle; The specific adjustment process includes: (1) When the fluid density error For the objective function When the contribution is greatest, increase the density control weight. This allows the control system to prioritize reducing density deviation; the back pressure is adjusted according to the direction of the objective function gradient change. and carbon dioxide flow By changing the density of the supercritical fluid, the actual density gradually approaches the target density; (2) When the fluid viscosity error For the objective function When the contribution is greatest, increase the viscosity control weight. Prioritize adjusting the flow rate of the modifier. With carbon dioxide flow The proportional relationship between these components can be used to optimize viscosity parameters by altering fluid polarity and solvation ability. (3) When the chromatographic retention time error For the objective function When the contribution is greatest, increase the retention time to optimize the weight. By synergistically adjusting fluid density, modifier ratio and back pressure, the distribution behavior of the target component between the stationary phase and the mobile phase is changed, so that the actual retention time converges to the target retention time. (4) When the chromatographic resolution error For the objective function When the contribution is maximized, the separation degree is increased to optimize the weights. By optimizing column temperature Back pressure In addition, the proportion of modifiers can reduce peak broadening and component interactions, thereby improving the separation performance between target components; Through the above process, the synergistic optimization of CO2 flow rate, modifier ratio, column temperature, and back pressure is achieved, making the objective function... Continue to decrease until the preset convergence condition is met: ， Complete one multivariate collaborative regulation.

9. The control method for a supercritical fluid chromatography system according to claim 1, characterized in that, In step (7), by collecting chromatographic retention time, peak width, peak tailing factor and resolution in real time, the fluid target physical property parameters and coupling control matrix weights are dynamically corrected according to the measured chromatographic separation performance. The control parameters in steps (4) and (5) are updated using the feedback results of step (7), and the cyclic control of steps (5) to (7) continues.

10. The control method for a supercritical fluid chromatography system according to claim 9, characterized in that, The specific process of dynamically correcting the fluid target physical property parameters and coupling control matrix weights based on measured chromatographic separation performance is as follows: 1) Establish a chromatographic separation performance evaluation model, and use the retention time acquired in real time. Peak width Peak tailing factor and resolution The separation performance deviation is calculated by comparing it with the preset target value. ， in, , , , These are the target retention time, target peak width, target peak tailing factor, and target separation, respectively. for , , , Corresponding evaluation weights; 2) Based on the separation performance deviation, feedback correction is performed on the target fluid properties: When the chromatographic retention time deviates from the target value, the density and viscosity parameters of the target fluid are adjusted according to the direction of the deviation. Specifically, when the actual retention time is greater than the target retention time, the density of the target fluid is increased to enhance the solubilization ability and reduce the retention time of the target substance; when the actual retention time is less than the target retention time, the density of the target fluid is decreased to allow the target substance to maintain a longer retention time. When an increase in peak width or a decrease in separation is detected, the stability constraint of the target fluid density is increased, and the control weights corresponding to the rate of pressure and temperature change are reduced, so that the fluid state changes more smoothly. When an increase in peak tailing factor is detected, the target parameter of the corrector ratio is adjusted according to the peak shape change trend, and the mass transfer process of the target component is improved by changing the fluid polarity and solubility selectivity. The specific correction process is represented as follows: ， in: These are the target physical properties of the fluid, including target density, target viscosity, target compressibility factor, and target modifier ratio; This is the performance deviation feedback gain matrix; Indicates the current optimization cycle; 3) Update the weights of the coupling control matrix online based on the chromatographic separation performance feedback results: The coupling control matrix is ​​used to describe the mapping relationship between changes in target physical property parameters and control variables: ， in, This includes carbon dioxide flow rate regulation, corrective agent flow rate regulation, column temperature regulation, and back pressure regulation; Including density, viscosity, pressure, and temperature deviation; The matrix weights are adaptively adjusted based on the real-time separation performance deviation: ， in, For online learning speed; When the system detects a decrease in separation, the control weights for density changes and corrector ratio changes are increased; when an increase in peak width is detected, the control weights related to temperature stability and pressure stability are increased; when an increase in peak tailing is detected, the control weight for corrector ratio is increased.

Images