Ceramic nanofiltration membrane treatment method for removing impurities from natural product macromolecules

Abstract

The invention discloses a treatment method of a ceramic nanofiltration membrane for removing impurities of natural product macromolecules, and particularly relates to the field of application of ceramic nanofiltration membrane treatment.The treatment method comprises the steps that an operation data set generated in the operation process of the ceramic nanofiltration membrane is converted into a normalized membrane behavior characteristic set through a synchronous calibration method, and the normalized membrane behavior characteristic set serves as basic input of follow-up behavior modeling; executing state sequence learning and trend prediction on the normalized membrane behavior feature set through a behavior learning method, and outputting a membrane behavior state anomaly score; and performing fusion calculation on the abnormal score of the membrane behavior state and the auxiliary signal set through an abnormal judgment method. A closed-loop processing path from real-time monitoring to intelligent early warning, dynamic adjustment to self-learning model updating is formed in a mode of combining synchronous acquisition based on multi-source operation data, behavior trend recursion, abnormal signal collaborative fusion judgment and self-adaptive process intervention; the problem that the flux mutation trend of the ceramic nanofiltration membrane cannot be recognized in advance is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of ceramic nanofiltration membrane treatment, and more specifically, to a ceramic nanofiltration membrane treatment method for removing macromolecular impurities from natural products. Background Art

[0002] In the current ceramic nanofiltration membrane treatment of macromolecular extracts of natural products, although the membrane material has good corrosion resistance and mechanical strength, the membrane flux exhibits unpredictable nonlinear fluctuations in the highly heterogeneous system of colloids, macromolecules, microbubbles, and other heterogeneous coexistence in the complex extract.

[0003] Traditional process monitoring methods mainly rely on average flow or pressure values, which can only respond passively after membrane blockage or failure occurs, and cannot effectively perceive and intervene in potential minor anomalies before mutations during membrane operation.

[0004] Especially when the batches of natural product extracts fluctuate greatly and the formation of impurity clusters is highly random, the membrane system is prone to suddenly jump from a normal state to a serious failure state without warning, causing production interruption and a sharp shortening of the membrane service life;

[0005] Therefore, the most fundamental and unresolved substantive problem in the prior art is the lack of a processing method that can identify and warn of membrane flux mutation trends in advance based on real-time data during the operation of ceramic nanofiltration membranes. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a ceramic nanofiltration membrane treatment method for removing impurities from natural product macromolecules. By combining the synchronous acquisition of multi-source operation data, behavioral trend recursion, collaborative fusion judgment of abnormal signals and adaptive process intervention, a complete closed-loop processing path is formed from real-time monitoring to intelligent early warning, dynamic adjustment and self-learning model update, which solves the problem of being unable to identify the sudden change trend of ceramic nanofiltration membrane flux in advance.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for treating natural product macromolecules with a ceramic nanofiltration membrane for impurity removal, comprising:

[0008] S1: The operating data set generated during the operation of the ceramic nanofiltration membrane is converted into a normalized membrane behavior feature set through a synchronous calibration method, which serves as the basic input for subsequent behavior modeling;

[0009] S2: Perform state sequence learning and trend prediction on the normalized membrane behavior feature set through behavioral learning methods, and output the membrane behavior state abnormality score;

[0010] S3: The membrane behavior state abnormality score is integrated with the auxiliary signal set through an abnormality judgment method to output a comprehensive abnormality risk level and determine the subsequent processing path;

[0011] S4: Execute process adjustment methods or emergency bypass switching methods according to the comprehensive abnormal risk level to obtain immediate correction of the membrane operation status;

[0012] S5: Update the abnormal case samples to the behavior prediction model through the self-learning update method to form a behavior prediction model that is adaptively optimized for the new state.

[0013] In a preferred embodiment, S1 further includes: acquiring data generated during membrane operation and defining the data as an operation data set, the operation data set including pressure data, flow data, particle concentration data, and temperature data;

[0014] The operation data set is time-aligned by a synchronous calibration method to obtain a synchronized operation data set, the synchronized operation data set is subjected to a noise removal method to remove the instantaneous fluctuation exceeding the limit value, and the purified operation data set is output, and the purified operation data set is divided into continuous sub-data sets according to a fixed time window by a data segmentation method;

[0015] The continuous sub-data set is input into the feature extraction method to extract the membrane behavior characteristics, which include the instantaneous flux change rate, the particle concentration change gradient, and the abnormal flow fluctuation value; the membrane behavior characteristics are normalized to output the normalized membrane behavior feature set.

[0016] In a preferred embodiment, S2 further includes: inputting the normalized membrane behavior feature set into a behavior learning method, wherein the behavior learning method performs behavior state sequence learning based on a cyclic recursive prediction algorithm, wherein the cyclic recursive prediction algorithm includes cyclic recursive regression prediction, abnormal mutation trajectory fitting prediction, and state vector trend recursive prediction;

[0017] The behavior state sequence learning result is output as a historical behavior state trajectory; the current membrane behavior feature is input into the state similarity calculation method, and the state similarity calculation method performs the optimal similarity matching between the historical behavior state trajectory and the current membrane behavior feature through the optimal path dynamic time warping algorithm, and outputs the behavior state matching result;

[0018] Determine whether the behavior state matching result is lower than the set behavior deviation threshold. If so, return to the first step to continue monitoring; if not, enter the behavior mutation trend prediction step;

[0019] The behavior state matching result is input into the behavior mutation trend prediction method. The behavior mutation trend prediction method solves the future change trajectory of the membrane behavior state through a multi-order trend difference fitting algorithm and outputs a set of membrane behavior state prediction trajectories.

[0020] The membrane behavior state prediction trajectory set is input into the anomaly scoring calculation method. The anomaly scoring calculation method uses a composite anomaly scoring mechanism to calculate the degree of anomaly of each membrane behavior feature and outputs a membrane behavior state anomaly score.

[0021] In a preferred embodiment, S3 further includes: inputting the membrane behavior state abnormality score into the abnormality judgment method, the abnormality judgment method performing a first abnormality judgment based on whether the abnormality score exceeds a first threshold, and if not, returning to the first part to continue monitoring; if it exceeds the first threshold, inputting the membrane behavior state abnormality score and an auxiliary signal set into a multi-feature fusion judgment method, the auxiliary signal set including a pressure fluctuation abnormality signal, a flow velocity sharp drop signal, and a microbubble concentration abnormal peak signal;

[0022] The multi-feature fusion judgment method is based on the hierarchical decision tree fusion algorithm to perform comprehensive correlation judgment between the auxiliary signal set and the membrane behavior state abnormality score, and output the comprehensive abnormality risk level; judge whether the comprehensive abnormality risk level belongs to the low risk level, if so, return to S1 to continue monitoring; if the comprehensive abnormality risk level belongs to the medium risk level, enter the process adjustment method; if it belongs to the high risk level, enter the emergency bypass switching method.

[0023] In a preferred embodiment, S4 further includes: inputting the medium risk level into a process adjustment method, the process adjustment method solving an optimal process parameter adjustment path based on the behavior trend and the historical abnormal case knowledge base, the optimal process parameter adjustment path including reducing the feed flow rate, reducing the membrane pressure difference, and increasing the short-time flushing frequency;

[0024] After executing the optimal process parameter adjustment path, return to S1 to continue forming a closed-loop monitoring process;

[0025] The high risk level is input into the emergency bypass switching method. The emergency bypass switching method switches the current feed flow to the bypass buffer flow path by setting the bypass trigger condition; after executing the emergency bypass switching, it returns to S1 to continue to form a closed-loop monitoring process.

[0026] In a preferred embodiment, S5 further includes: inputting the anomaly score, the comprehensive anomaly risk level, and the final process result as new case samples into the self-learning update method, wherein the self-learning update method evaluates the contribution of the new case sample to the behavior prediction model based on a sample weight optimization algorithm and outputs a sample weight update result;

[0027] The sample weight update results are input into the behavior learning method to perform behavior model retraining to form a new behavior prediction model.

[0028] In a preferred embodiment, S2 further includes: establishing a ceramic nanofiltration membrane behavior evolution model, performing state recursion and behavior trend tracking on the normalized membrane behavior feature set through a behavior learning method, and outputting a membrane behavior prediction feature set; the ceramic nanofiltration membrane behavior evolution model is expressed as:

[0029]

[0030] in:

[0031]

[0032] Among them, M t+1 is the membrane behavior prediction feature set, which represents the predicted behavior state at time t+1; F t is the current normalized membrane behavior feature set, which includes the instantaneous flux change rate Particle concentration gradient Abnormal flow fluctuation value ΔF t =F t -F t-1 Represents the current behavior feature change rate set; D t is the membrane external disturbance set, which includes the feed flow impact energy Transient shear wave intensity α k is the behavioral recursion coefficient of the impact of historical behavioral changes on current behavior. In practical applications, the behavioral recursion coefficient of the impact of historical behavioral changes on current behavior is a fixed empirical parameter; β k is the recursive coefficient of the impact of the historical behavior state difference on the current behavior. In practical applications, the recursive coefficient of the impact of the historical behavior state difference on the current behavior is a fixed empirical parameter; n is the time window length for reviewing the historical behavior state, and k is the index; It is a behavior state asymmetric difference tracking function, which is used to detect abnormal trends of mutations in historical behaviors; is the disturbance response coupling function, which is used to analyze the impact of the current external disturbance on the behavior prediction results; sign(x) is the sign function, which is defined as returning +1 when x>0, returning 0 when x=0, and returning -1 when x<0. It is used to determine the direction of the change between two behavior state values.

[0033] In a preferred embodiment, S3 further includes: based on the multi-feature comprehensive judgment and abnormality confirmation, fusing and calculating the membrane behavior prediction feature set and the auxiliary abnormal signal set through the abnormality judgment method to output a comprehensive abnormality risk level;

[0034]

[0035] where R final The final comprehensive abnormal risk level is used to judge the subsequent process path; M score is the comprehensive anomaly score of the membrane behavior prediction feature set, and the comprehensive anomaly score of the membrane behavior prediction feature set is calculated by M t+1 Calculated according to preset rules; A j is the jth auxiliary abnormal signal value, which includes the pressure mutation value A1, the abnormal peak value of microbubble concentration A2, and the drastic change value of feed flow rate A3; m is the number of signals in the auxiliary abnormal signal set; γ j The auxiliary abnormal signal sensitivity amplification factor is used to control the contribution strength of each signal to the abnormal level; is the predicted value of the pth dimension of the membrane behavior feature set prediction value; is the real-time observation value of the pth dimension of the current value of the membrane behavior feature set; q is the number of dimensions of the membrane behavior feature set (same as F in S2 t consistent); δ p is the membrane behavior prediction error amplification factor, which is used to enhance the contribution of the prediction error to the comprehensive score;

[0036] R final The model formula forms a strong ability to detect extreme abnormal states through dual-channel coupling of multi-source signal anomalies and behavior prediction errors.

[0037] In a preferred embodiment, S5 further includes: adaptive model updating, feeding back abnormal case samples to the behavior prediction logic through a self-learning update method to optimize the model parameters for the next cycle prediction;

[0038]

[0039] in The contribution weight of the abnormal case sample in the behavior prediction logic after updating; The original contribution weight of the abnormal case sample; is the state value of the membrane behavior feature set predicted by the abnormal case sample at time t; is the state value of the membrane behavior feature set actually observed by the abnormal case sample at time t; T is the length of the behavior state data sequence recorded by the abnormal case sample;

[0040] The above formula forms a dynamic smooth weight attenuation mechanism through the accumulation of cubic error and the square accumulation of actual observations, thus achieving long-term adaptive error suppression.

[0041] Technical effects and advantages of the present invention:

[0042] 1. Through multi-source data acquisition and behavioral trend prediction, a continuous monitoring and early warning mechanism has been established that can identify sudden changes in ceramic nanofiltration membrane flux in advance, solving the problem that existing technologies cannot achieve real-time early warning;

[0043] 2. Through simultaneous calibration and feature extraction, data such as membrane operating pressure, flow rate, and particle concentration are acquired and standardized to form a set of membrane behavior features that can be used for behavior prediction, thereby improving the accuracy of model input data;

[0044] 3. Through behavioral learning and dynamic recursive methods, a membrane behavior state sequence model was established, which enables continuous tracking and trend prediction of minor abnormal behaviors during membrane operation, enhancing the ability to perceive abnormalities;

[0045] 4. Through the fusion judgment of anomaly scoring and auxiliary anomaly signals, the anomaly risk level is calculated by multi-signal collaboration, which reduces the error caused by single anomaly judgment and improves the accuracy of anomaly detection;

[0046] 5. Through adaptive process parameter optimization and bypass switching driven by comprehensive abnormality risk levels, graded intervention and fault prevention of membrane system operation status are achieved according to different abnormality levels;

[0047] 6. Through self-learning updates and dynamic adjustment of model parameters based on abnormal case error feedback, the adaptability and stability of the behavior prediction model in the processing of various natural product extracts are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flowchart of the method steps of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] Refer to the instruction manual Figure 1 A ceramic nanofiltration membrane treatment method for removing impurities from natural product macromolecules according to one embodiment of the present invention comprises:

[0051] S1: The operating data set generated during the operation of the ceramic nanofiltration membrane is converted into a normalized membrane behavior feature set through a synchronous calibration method, which serves as the basic input for subsequent behavior modeling;

[0052] S2: Perform state sequence learning and trend prediction on the normalized membrane behavior feature set through behavioral learning methods, and output the membrane behavior state abnormality score;

[0053] S3: The membrane behavior state abnormality score is integrated with the auxiliary signal set through an abnormality judgment method to output a comprehensive abnormality risk level and determine the subsequent processing path;

[0054] S4: Execute process adjustment methods or emergency bypass switching methods according to the comprehensive abnormal risk level to obtain immediate correction of the membrane operation status;

[0055] S5: Update the abnormal case samples to the behavior prediction model through the self-learning update method to form a behavior prediction model that is adaptively optimized for the new state.

[0056] S1 also includes: acquiring data generated during membrane operation and defining it as an operation data set, the operation data set including pressure data, flow data, particle concentration data, and temperature data;

[0057] The operation data set is time-aligned by a synchronous calibration method to obtain a synchronized operation data set, the synchronized operation data set is subjected to a noise removal method to remove the instantaneous fluctuation exceeding the limit value, and the purified operation data set is output, and the purified operation data set is divided into continuous sub-data sets according to a fixed time window by a data segmentation method;

[0058] The continuous sub-data set is input into the feature extraction method to extract the membrane behavior characteristics, which include the instantaneous flux change rate, the particle concentration change gradient, and the abnormal flow fluctuation value; the membrane behavior characteristics are normalized to output the normalized membrane behavior feature set.

[0059] S2 also includes: inputting the normalized membrane behavior feature set into a behavior learning method, the behavior learning method performing behavior state sequence learning based on a cyclic recursive prediction algorithm, the cyclic recursive prediction algorithm including cyclic recursive regression prediction, abnormal mutation trajectory fitting prediction, and state vector trend recursive prediction;

[0060] The behavior state sequence learning result is output as a historical behavior state trajectory; the current membrane behavior feature is input into the state similarity calculation method, and the state similarity calculation method performs the optimal similarity matching between the historical behavior state trajectory and the current membrane behavior feature through the optimal path dynamic time warping algorithm, and outputs the behavior state matching result;

[0061] Determine whether the behavior state matching result is lower than the set behavior deviation threshold. If so, return to the first step to continue monitoring; if not, enter the behavior mutation trend prediction step;

[0062] The behavior state matching result is input into the behavior mutation trend prediction method. The behavior mutation trend prediction method solves the future change trajectory of the membrane behavior state through a multi-order trend difference fitting algorithm and outputs a set of membrane behavior state prediction trajectories.

[0063] The membrane behavior state prediction trajectory set is input into the anomaly scoring calculation method. The anomaly scoring calculation method uses a composite anomaly scoring mechanism to calculate the degree of anomaly of each membrane behavior feature and outputs a membrane behavior state anomaly score.

[0064] S3 further includes: inputting the membrane behavior state abnormality score into the abnormality judgment method, the abnormality judgment method performing a first abnormality judgment based on whether the abnormality score exceeds a first threshold, and if not, returning to the first part to continue monitoring; if it exceeds the first threshold, inputting the membrane behavior state abnormality score and an auxiliary signal set into a multi-feature fusion judgment method, the auxiliary signal set including a pressure fluctuation abnormality signal, a flow velocity sharp drop signal, and a microbubble concentration abnormal peak signal;

[0065] The multi-feature fusion judgment method is based on the hierarchical decision tree fusion algorithm to perform comprehensive correlation judgment between the auxiliary signal set and the membrane behavior state abnormality score, and output the comprehensive abnormality risk level; judge whether the comprehensive abnormality risk level belongs to the low risk level, if so, return to S1 to continue monitoring; if the comprehensive abnormality risk level belongs to the medium risk level, enter the process adjustment method; if it belongs to the high risk level, enter the emergency bypass switching method.

[0066] S4 also includes: inputting the medium risk level into the process adjustment method. The process adjustment method is based on the behavioral trends and historical abnormal case knowledge base to solve the optimal process parameter adjustment path. The optimal process parameter adjustment path includes reducing the feed flow rate, reducing the membrane pressure difference, and increasing the short-time flushing frequency;

[0067] After executing the optimal process parameter adjustment path, return to S1 to continue forming a closed-loop monitoring process;

[0068] The high risk level is input into the emergency bypass switching method. The emergency bypass switching method switches the current feed flow to the bypass buffer flow path by setting the bypass trigger condition; after executing the emergency bypass switching, it returns to S1 to continue to form a closed-loop monitoring process.

[0069] S5 also includes: inputting the anomaly score, the comprehensive anomaly risk level, and the final process result as new case samples into the self-learning update method, the self-learning update method evaluating the contribution of the new case sample to the behavior prediction model based on the sample weight optimization algorithm, and outputting the sample weight update result;

[0070] The sample weight update results are input into the behavior learning method to perform behavior model retraining to form a new behavior prediction model.

[0071] It should be noted that in the formula structure involved in this solution, dimensionless terms can serve as proportionality or structural adjustment factors. When combined with quantities with units, they only play a numerical scaling role and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system of expression. This combination of "dimensionless terms and units" can be understood as a composite structural expression commonly used in mathematical and physical modeling, conforming to the principle of dimensional consistency and having a clear physical interpretation basis.

[0072] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can be formed into a unified structure through function mapping, ratio combination or normalization adjustment. The units and meanings are clear, and the overall expression conforms to the principle of dimensional consistency and the common formula of engineering modeling.

[0073] In this solution, any design constants, weights, adjustment factors, threshold parameters, and proportional coefficients are adjustable control parameters for different application environments. Their values depend on the target device configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are set within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have unique preset values, they have clear adjustment logic and calculation paths, and are part of the deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution is both universally adaptable, reproducible, and operable, without affecting its technical clarity and feasibility.

[0074] S2 also includes: establishing a ceramic nanofiltration membrane behavior evolution model, performing state recursion and behavior trend tracking on the normalized membrane behavior feature set through a behavior learning method, and outputting a membrane behavior prediction feature set; the ceramic nanofiltration membrane behavior evolution model is expressed as:

[0075]

[0076] in:

[0077]

[0078] Among them, M t+1 is the membrane behavior prediction feature set, which represents the predicted behavior state at time t+1; F t is the current normalized membrane behavior feature set, which includes the instantaneous flux change rate Particle concentration gradient Abnormal flow fluctuation value ΔF t =F t -F t-1 Represents the current behavior feature change rate set; D t is the membrane external disturbance set, which includes the feed flow impact energy Transient shear wave intensity α k is the behavioral recursion coefficient of the impact of historical behavioral changes on current behavior. In practical applications, the behavioral recursion coefficient of the impact of historical behavioral changes on current behavior is a fixed empirical parameter; β k is the recursive coefficient of the impact of the historical behavior state difference on the current behavior. In practical applications, the recursive coefficient of the impact of the historical behavior state difference on the current behavior is a fixed empirical parameter; n is the time window length for reviewing the historical behavior state, and k is the index; It is a behavior state asymmetric difference tracking function, which is used to detect abnormal trends of mutations in historical behaviors; is the disturbance response coupling function, which is used to analyze the impact of the current external disturbance on the behavior prediction results; sign(x) is the sign function, which is defined as returning +1 when x>0, returning 0 when x=0, and returning -1 when x<0. It is used to determine the direction of the change between two behavior state values.

[0079] S3 also includes: based on multi-feature comprehensive judgment and abnormality confirmation, the membrane behavior prediction feature set and the auxiliary abnormal signal set are fused and calculated through the abnormality judgment method to output the comprehensive abnormality risk level;

[0080]

[0081] where R final The final comprehensive abnormal risk level is used to judge the subsequent process path; M score is the comprehensive anomaly score of the membrane behavior prediction feature set, and the comprehensive anomaly score of the membrane behavior prediction feature set is calculated by M t+1 Calculated according to preset rules; A j is the jth auxiliary abnormal signal value, which includes the pressure mutation value A1, the abnormal peak value of microbubble concentration A2, and the drastic change value of feed flow rate A3; m is the number of signals in the auxiliary abnormal signal set; γ j The auxiliary abnormal signal sensitivity amplification factor is used to control the contribution strength of each signal to the abnormal level; is the predicted value of the pth dimension of the membrane behavior feature set prediction value; is the real-time observation value of the pth dimension of the current value of the membrane behavior feature set; q is the number of dimensions of the membrane behavior feature set (same as F in S2 t consistent); δ p is the membrane behavior prediction error amplification factor, which is used to enhance the contribution of the prediction error to the comprehensive score;

[0082] R final The model formula forms a strong ability to detect extreme abnormal states through dual-channel coupling of multi-source signal anomalies and behavior prediction errors.

[0083] In order to realize the classification of comprehensive abnormal risk levels in this method, it includes but is not limited to proposing to judge based on the comprehensive abnormality score results, combining the membrane behavior prediction error and the auxiliary abnormal signal with the normalized normalized score of the abnormal value; the specific classification logic includes: setting the comprehensive abnormality score as a continuous interval R final ∈[0,1], where a scoring interval threshold model is established through a large number of historical case samples; when R final <0.3 is defined as low risk, indicating that the membrane operation state is stable and only real-time monitoring is required; when 0.3≤R final When R<0.6, it is defined as medium risk, indicating that the membrane state deviates from the normal trajectory and the process needs to be activated for adaptive optimization and adjustment. final A value ≥0.6 is defined as high risk, indicating that the membrane operating state has entered a high failure probability range and bypass switching must be triggered immediately to ensure system safety. The above interval division is determined through cluster analysis of historical abnormal case data and behavioral state prediction error distribution analysis, ensuring that the algorithm has high adaptability and robustness in the processing of different batches of natural product extracts.

[0084] S5 also includes: adaptive model updates, which use a self-learning update method to feed abnormal case samples back to the behavior prediction logic to optimize the model parameters for the next cycle prediction;

[0085]

[0086] in The contribution weight of the abnormal case sample in the behavior prediction logic after updating; The original contribution weight of the abnormal case sample; is the state value of the membrane behavior feature set predicted by the abnormal case sample at time t; is the state value of the membrane behavior feature set actually observed by the abnormal case sample at time t; T is the length of the behavior state data sequence recorded by the abnormal case sample;

[0087] The above formula forms a dynamic smooth weight attenuation mechanism through the accumulation of cubic error and the square accumulation of actual observations, thus achieving long-term adaptive error suppression.

[0088] It should be generally explained that the ceramic nanofiltration membrane treatment method for removing impurities from natural product macromolecules provided by the present invention is intended to solve the problems of unpredictable flux fluctuations and difficulty in early warning of membrane blockage mutations in the current ceramic nanofiltration membrane treatment process of natural product extracts;

[0089] This method is based on high-frequency operational data. It first generates a standardized set of membrane behavior characteristics through simultaneous calibration and data preprocessing. These characteristics include key features such as the instantaneous flux change rate, particle concentration gradient, and abnormal flow fluctuations, providing complete input for subsequent prediction models.

[0090] The design concept is to establish a recursive membrane behavior evolution model that not only comprehensively considers the changing trends of historical behavior states, but also introduces a set of external perturbations (such as feed impact energy and shear strength) to achieve dynamic coupling prediction. This can capture subtle behavioral deviations and perceive mutation risks in advance.

[0091] Through the anomaly judgment method, the membrane behavior anomaly score and auxiliary anomaly signals (such as abnormal pressure fluctuation, abnormal microbubble concentration, and sudden flow rate) are integrated in multiple dimensions to form a comprehensive anomaly risk level assessment with multi-feature collaboration, which improves the accuracy and robustness of anomaly detection. When the comprehensive anomaly risk level reaches a medium or high level, a dual-path intervention mechanism is designed: for medium risk, the adaptive process adjustment logic is activated to automatically optimize the feed flow rate, membrane pressure difference, and flushing frequency to achieve membrane status recovery. For high risk, bypass switching is immediately executed to ensure system safety. To avoid model failure, the solution further introduces a self-learning update method, which feeds back the prediction error and actual behavior status of each abnormal case into the behavior prediction model, and dynamically optimizes the sample weight based on the error contribution, thereby continuously improving the model's generalization ability and prediction accuracy under variable working conditions. The overall process forms a complete closed-loop mechanism from real-time data monitoring, dynamic prediction, multi-source fusion judgment of anomalies, to intelligent intervention and model self-learning, which is engineering feasible.

[0092] This design takes into account the industry characteristics of natural product extracts, such as complex impurities, drastic fluctuations in operating conditions, and extremely high membrane loss costs. Combined with the rigid characteristics of the ceramic nanofiltration membrane process, it cleverly implements data-driven predictive maintenance and process control, extending the service life of the membrane, reducing operation and maintenance costs, and improving the continuity and quality stability of natural product extraction.

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane, characterized in that: include: S1: The operating data set generated during the operation of the ceramic nanofiltration membrane is converted into a normalized membrane behavior feature set through a synchronous calibration method, which serves as the basic input for subsequent behavior modeling; S2: Perform state sequence learning and trend prediction on the normalized membrane behavior feature set through behavioral learning methods, and output the membrane behavior state abnormality score; S3: The membrane behavior state abnormality score is integrated with the auxiliary signal set through an abnormality judgment method to output a comprehensive abnormality risk level and determine the subsequent processing path; S4: Execute process adjustment methods or emergency bypass switching methods according to the comprehensive abnormal risk level to obtain immediate correction of the membrane operation status; S5: Update the abnormal case samples to the behavior prediction model through the self-learning update method to form a behavior prediction model that is adaptively optimized for the new state.

2. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 1, wherein: S1 also includes: acquiring data generated during membrane operation and defining it as an operation data set, the operation data set including pressure data, flow data, particle concentration data, and temperature data; The operation data set is time-aligned by a synchronous calibration method to obtain a synchronized operation data set, the synchronized operation data set is subjected to a noise removal method to remove the instantaneous fluctuation exceeding the limit value, and the purified operation data set is output, and the purified operation data set is divided into continuous sub-data sets according to a fixed time window by a data segmentation method; The continuous sub-data set is input into the feature extraction method to extract the membrane behavior characteristics, which include the instantaneous flux change rate, the particle concentration change gradient, and the abnormal flow fluctuation value; the membrane behavior characteristics are normalized to output the normalized membrane behavior feature set.

3. The method for treating natural product macromolecules with a ceramic nanofiltration membrane according to claim 2, wherein: S2 also includes: inputting the normalized membrane behavior feature set into a behavior learning method, the behavior learning method performing behavior state sequence learning based on a cyclic recursive prediction algorithm, the cyclic recursive prediction algorithm including cyclic recursive regression prediction, abnormal mutation trajectory fitting prediction, and state vector trend recursive prediction; The behavior state sequence learning result is output as a historical behavior state trajectory; the current membrane behavior feature is input into the state similarity calculation method, and the state similarity calculation method performs similarity matching between the historical behavior state trajectory and the current membrane behavior feature through the path dynamic time warping algorithm, and outputs the behavior state matching result; Determine whether the behavior state matching result is lower than the set behavior deviation threshold. If so, return to the first step to continue monitoring; if not, enter the behavior mutation trend prediction step; The behavior state matching result is input into the behavior mutation trend prediction method. The behavior mutation trend prediction method solves the future change trajectory of the membrane behavior state through a multi-order trend difference fitting algorithm and outputs a set of membrane behavior state prediction trajectories. The membrane behavior state prediction trajectory set is input into the anomaly scoring calculation method. The anomaly scoring calculation method uses a composite anomaly scoring mechanism to calculate the degree of anomaly of each membrane behavior feature and outputs a membrane behavior state anomaly score.

4. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 3, wherein: S3 further includes: inputting the membrane behavior state abnormality score into the abnormality judgment method, the abnormality judgment method performing a first abnormality judgment based on whether the abnormality score exceeds a first threshold, and if not, returning to the first part to continue monitoring; if it exceeds the first threshold, inputting the membrane behavior state abnormality score and an auxiliary signal set into a multi-feature fusion judgment method, the auxiliary signal set including a pressure fluctuation abnormality signal, a flow velocity sharp drop signal, and a microbubble concentration abnormal peak signal; The multi-feature fusion judgment method is based on the hierarchical decision tree fusion algorithm to perform comprehensive correlation judgment between the auxiliary signal set and the membrane behavior state abnormality score, and output the comprehensive abnormality risk level; judge whether the comprehensive abnormality risk level belongs to the low risk level, if so, return to S1 to continue monitoring; if the comprehensive abnormality risk level belongs to the medium risk level, enter the process adjustment method; if it belongs to the high risk level, enter the emergency bypass switching method.

5. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 4, characterized in that: S4 also includes: inputting the medium risk level into the process adjustment method. The process adjustment method is based on the behavioral trend and historical abnormal case knowledge base to solve the process parameter adjustment path. The process parameter adjustment path includes reducing the feed flow rate, reducing the membrane pressure difference, and increasing the short-time flushing frequency; After executing the process parameter adjustment path, return to S1 to continue forming a closed-loop monitoring process; The high risk level is input into the emergency bypass switching method. The emergency bypass switching method switches the current feed flow to the bypass buffer flow path by setting the bypass trigger condition; after executing the emergency bypass switching, it returns to S1 to continue to form a closed-loop monitoring process.

6. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 5, characterized in that: S5 also includes: inputting the anomaly score, the comprehensive anomaly risk level, and the final process result as new case samples into the self-learning update method, the self-learning update method evaluating the contribution of the new case sample to the behavior prediction model based on the sample weight optimization algorithm, and outputting the sample weight update result; The sample weight update results are input into the behavior learning method to perform behavior model retraining to form a new behavior prediction model.

7. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 6, characterized in that: S2 also includes: establishing a ceramic nanofiltration membrane behavior evolution model, performing state recursion and behavior trend tracking on the normalized membrane behavior feature set through a behavior learning method, and outputting a membrane behavior prediction feature set; the ceramic nanofiltration membrane behavior evolution model is expressed as: in: Among them, M t+1 is the membrane behavior prediction feature set, which represents the predicted behavior state at time t+1; F t is the current normalized membrane behavior feature set, which includes the instantaneous flux change rate Particle concentration gradient Abnormal flow fluctuation value ΔF t =F t -F t-1 Represents the current behavior feature change rate set; D t is the membrane external disturbance set, which includes the feed flow impact energy Transient shear wave intensity α k is the behavioral recursion coefficient of the impact of historical behavioral changes on current behavior; β k is the recursive coefficient of the impact of the historical behavior state difference on the current behavior. The recursive coefficient of the impact of the historical behavior state difference on the current behavior is a fixed empirical parameter; n is the time window length for reviewing the historical behavior state, and k is the index; It is a behavior state asymmetric difference tracking function, which is used to detect abnormal trends of mutations in historical behaviors; is the disturbance response coupling function, which is used to analyze the impact of the current external disturbance on the behavior prediction results; sign(x) is the sign function, which is defined as returning +1 when x>0, returning 0 when x=0, and returning -1 when x<0. It is used to determine the direction of the change between two behavior state values.

8. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 7, characterized in that: S3 also includes: based on multi-feature comprehensive judgment and abnormality confirmation, the membrane behavior prediction feature set and the auxiliary abnormal signal set are fused and calculated through the abnormality judgment method to output the comprehensive abnormality risk level; where R final is the final comprehensive abnormal risk level; M score A is the comprehensive anomaly score of the membrane behavior prediction feature set; j is the jth auxiliary abnormal signal value, which includes the pressure mutation value A1, the abnormal peak value of microbubble concentration A2, and the drastic change value of feed flow rate A3; m is the number of signals in the auxiliary abnormal signal set; γ j It is the auxiliary abnormal signal sensitivity amplification factor; is the predicted value of the pth dimension of the membrane behavior feature set prediction value; is the real-time observation value of the pth dimension of the current value of the membrane behavior feature set; q is the number of dimensions of the membrane behavior feature set; δ p is the error amplification factor for membrane behavior prediction.

9. The method for removing impurities from natural product macromolecules using a ceramic nanofiltration membrane according to claim 8, characterized in that: S5 also includes: adaptive model updates, which use a self-learning update method to feed abnormal case samples back to the behavior prediction logic to optimize the model parameters for the next cycle prediction; in The contribution weight of the abnormal case sample in the behavior prediction logic after updating; The original contribution weight of the abnormal case sample; is the state value of the membrane behavior feature set predicted by the abnormal case sample at time t; is the state value of the membrane behavior feature set actually observed by the abnormal case sample at time t; T is the length of the behavior state data sequence recorded by the abnormal case sample.

Images