Screening Method for Macroporous Resins for Purification of Ephedra Root Tannins

Abstract

The present invention discloses a method for screening macroporous resins for purifying ephedra root tannins, comprising: loading the macroporous resin to be screened into a chromatography column, passing an ephedra root extract through the pretreated chromatography column; rinsing the chromatography column with deionized water until the absorbance value of the effluent stabilizes, desorbing the resin with an ethanol solution, and collecting the desorbed liquid; measuring the wet bulk density of the resin after adsorption; inputting the amount of ephedra root tannin adsorbed per unit time during the adsorption phase, the recovery rate of the ephedra root tannins during the desorption phase, and the wet bulk density ρ into a first sub-model based on a linear regression algorithm, a second sub-model based on a support vector regression algorithm, and a third sub-model based on a random forest algorithm; summing the scores of the sub-models according to dynamic weights to calculate a comprehensive score, and selecting the macroporous resin with the highest comprehensive score. The present invention can achieve precise screening of macroporous resins.

Description

Technical Field

[0001] The present invention relates to the technical field related to the purification of tannins from Ephedrae root, and more specifically, to a method for screening macroporous resins for purifying tannins from Ephedrae root. Background Art

[0002] Macroporous resin chromatography is widely used for the separation of plant tannins due to its advantages, such as high adsorption selectivity and easy regeneration. Tannin compounds from Ephedra root possess specific pharmacological activities, and their purification requires high performance from macroporous resins. However, current macroporous resin screening methods for the purification of Ephedra root tannins leave much to be desired, resulting in difficulties in achieving a balance between adsorption efficiency, desorption stability, and process adaptability in practical applications.

[0003] Traditional macroporous resin screening methods typically focus on a single or limited number of performance indicators, such as static adsorption capacity and simple desorption recovery, while neglecting key parameters related to the resin's physical properties and the dynamic adsorption process. For example, the resin's wet bulk density directly influences fluid distribution and mass transfer efficiency within the column. Improper bulk density can lead to excessive column pressure drop or uneven flow paths, which in turn affects adsorption kinetics. While resin specific surface area, a key indicator of adsorption site abundance, lacks quantitative evaluation of its binding capacity for tannin molecules, making it difficult to accurately match the molecular characteristics of the target component during screening. Furthermore, parameters reflecting process stability, such as desorption solution concentration stability (e.g., coefficient of variation) and dynamic adsorption mass transfer coefficient, are often overlooked in traditional screening, resulting in significant fluctuations in desorption solution composition during actual production and increasing the complexity of subsequent purification steps. Existing methods often employ single linear regression models, which can only handle simple linear relationships and struggle to accurately describe the nonlinear coupling between adsorption capacity, desorption recovery, and the resin's physicochemical properties. This leads to discrepancies between screening results and actual process performance.

[0004] Therefore, it is necessary to design a technical solution that can overcome the above-mentioned defects. Summary of the Invention

[0005] An object of the present invention is to provide a method for screening macroporous resins for purifying ephedra root tannins, which can achieve accurate screening of macroporous resins.

[0006] In order to achieve these objects and other advantages of the present invention, according to one aspect of the present invention, the present invention provides S1: after loading the macroporous resin to be screened into the chromatography column, using a 60%-70% ethanol solution at a flow rate of 1.5-2.5 times the column volume per hour to flush for 2.5-3.5 hours, and then flushing with deionized water until the conductivity of the effluent is less than 15-25μS / cm; passing the ephedra root extract through the pretreated chromatography column at a flow rate of 1.0-2.0 times the column volume per hour, controlling the pH value of the sample solution to 3.0-4.0, and stopping the adsorption when the ephedra root tannin concentration in the effluent reaches 10%-20% of the initial concentration of the sample solution; S2: flushing the chromatography column with deionized water until the effluent is at a wave After the absorbance value at 280nm stabilizes at 0.03-0.08, desorption is carried out using an ethanol solution with a concentration of 70%-85% at a flow rate of 0.8-1.2 times the column volume per hour, and the volume of the desorbed liquid collected is 2-4 times the column volume; S3: Determine the wet bulk density of the resin after adsorption, transfer the resin to a measuring cylinder of known mass, let it settle for 20-40 minutes, record the resin volume and weigh the total mass, and calculate the wet bulk density using the formula (total mass - measuring cylinder mass) / resin volume; S4: Input the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, and the wet bulk density measurement value into the evaluation model, calculate the score, and select the macroporous resin with the highest score.

[0007] Furthermore, it also includes: determination of resin specific surface area, after vacuum degassing the pretreated resin, the specific surface area is determined by nitrogen adsorption method; determination of coefficient of variation of desorption liquid concentration, detecting the tannin concentration of the desorption liquid by liquid chromatography, and calculating the coefficient of variation as a percentage of the standard deviation to the average value; determination of dynamic adsorption mass transfer coefficient, obtaining the mass transfer coefficient by fitting and calculating the adsorption kinetic model based on the adsorption breakthrough curve data; inputting the adsorption amount of ephedra root tannin per unit time in the adsorption stage, the recovery rate of ephedra root tannin in the desorption stage, the measured value of wet bulk density, the resin specific surface area, the coefficient of variation of the desorption liquid concentration, and the dynamic adsorption mass transfer coefficient into the evaluation model to calculate the score.

[0008] Furthermore, the first sub-model is based on a linear regression algorithm, and its input parameters are the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, and the wet bulk density of the resin, and the first sub-model score is output; the second sub-model is based on a support vector regression algorithm, and its input parameters include the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the coefficient of variation of the desorption liquid concentration, and the dynamic adsorption mass transfer coefficient, and the second sub-model score is output after mapping the nonlinear relationship through the kernel function; the third sub-model is based on a random forest algorithm, and its input parameters include the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, the wet bulk density of the resin, the specific surface area of ​​the resin, and the coefficient of variation of the desorption liquid concentration, and the third sub-model score is output after parallel training of multiple decision trees; according to the historical prediction error of each sub-model on the verification data set, the weight ratio of the first sub-model, the second sub-model, and the third sub-model are dynamically allocated, and the scores of each sub-model are added according to the dynamic weight to generate the final comprehensive score.

[0009] Furthermore, the kernel function of the second sub-model is a Gaussian kernel with a kernel width σ∈[0.25,0.35], a regularization parameter C∈[1.5,2.5], and an insensitive loss parameter ε≤0.03; the number of decision trees of the third sub-model is n_trees∈[80,120], the maximum depth of a single tree is max_depth∈[12,18], and the minimum number of samples for node splitting is min_samples_split≥15.

[0010] Furthermore, when the dynamic adsorption mass transfer coefficient K≥0.45 cm / s, the weight W2 of the second submodel is increased to 50%-60%; when the coefficient of variation of the desorption liquid concentration CV≤12%, the weight W3 of the third submodel is reduced to 15%-25%; when the resin specific surface area S≥400 m² / g and the wet bulk density ρ≤1.15 g / cm³, the weight W1 of the first submodel is fixed at 20%, and the weight W2 of the second submodel and the weight W3 of the third submodel are distributed in a ratio of 2:3; the adjustment step of each submodel weight does not exceed ±5% / minute, and the total weight distribution error is controlled within ±1.5%.

[0011] Furthermore, when the column temperature T∈[25°C,35°C], the second sub-model weight W2 is modified according to the formula: , and the sum of weights is kept at 100% through normalization.

[0012] Furthermore, when the pH fluctuation ΔpH>0.3 is detected during the adsorption phase, the fault tolerance process is activated: the output value of the first sub-model is multiplied by the attenuation factor , the kernel function of the second sub-model is switched to a polynomial kernel, and the depth of the decision tree of the third sub-model is temporarily reduced to 10 layers.

[0013] Furthermore, it also includes: selecting the two macroporous resins with the highest scores for secondary verification, repeating the adsorption-desorption process three times under the same operating conditions, calculating the standard deviation of the purity of ephedra root tannins in the three experiments, and retaining the resins with a standard deviation less than 1.0%-2.0% as the final screening results.

[0014] Furthermore, each cycle must meet the following requirements: column temperature control accuracy of ±1°C, desorption liquid flow rate fluctuation range ≤±5% of the set value; the purity of the ephedra root tannins obtained in each cycle is detected, and the standard deviation σ of the three purities is calculated. If σ∈[1.0%,2.0%], the resin is judged to be qualified; when two resins are qualified, the one with the smaller σ value is selected as the final screening result; if only one is qualified, it is directly selected; if both are unqualified, the scores are re-evaluated, and the top two resins with the highest original scores are excluded.

[0015] Furthermore, the selected resins were treated under accelerated aging conditions, which were: temperature 40±2°C, relative humidity 75±5%, for 72-120 hours; after aging, the adsorption-desorption process was repeated, and the decay rate ΔQ of the tannin adsorption before and after aging was calculated. ; Where Q0 is the adsorption amount before aging, Q1 is the adsorption amount after aging; retain resin with ΔQ≤5.0%.

[0016] The present invention has at least the following beneficial effects:

[0017] The present invention incorporates wet bulk density, specific surface area, coefficient of variation of desorption liquid concentration and dynamic adsorption mass transfer coefficient into screening indicators, accurately reflecting the fluid distribution capacity, adsorption site richness, desorption uniformity and mass transfer efficiency of the resin in the actual chromatography process, avoiding the screening bias caused by a single indicator, making the resin performance deeply matched with the requirements of the tannin purification process, and greatly improving the accuracy of macroporous resin screening. Linear regression, support vector regression and random forest algorithms are used to construct a three-level sub-model, which respectively handles linear relationships, nonlinear dynamic characteristics and multivariate complex coupling effects, and responds to changes in working conditions such as temperature and pH value in real time through a dynamic weight distribution mechanism, significantly improving the model's fitting accuracy for complex adsorption and desorption processes. The secondary verification step screens resins that meet the standard deviation through repetitive experiments, and combines accelerated aging tests to control the adsorption amount decay rate to ensure that the screened resins maintain stable performance in long-term use and reduce the risk of process fluctuations in industrial production.

[0018] Other advantages, objectives and features of the present invention will be reflected in part from the following description and will be understood by those skilled in the art through study and practice of the present invention. DETAILED DESCRIPTION

[0019] The present invention is further described in detail below with reference to the embodiments so that those skilled in the art can implement the invention with reference to the description.

[0020] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are intended only to explain the relative positional relationships and movement of components in a specific posture. If the specific posture changes, the directional indications will also change accordingly. When an element is referred to as being "fixed to" or "disposed on" another element, it can be directly on the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. References to "first," "second," etc. in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features designated as "first" or "second" may explicitly or implicitly include at least one of such features.

[0021] It should be noted that the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0022] The embodiments of the present application provide a method for screening macroporous resins for purifying ephedra root tannins, comprising:

[0023] S1: After loading the macroporous resin to be screened into the chromatography column, flush it with 60%-70% ethanol solution at a flow rate of 1.5-2.5 column volumes per hour for 2.5-3.5 hours, and then flush it with deionized water until the conductivity of the effluent is less than 15-25μS / cm; pass the ephedra root extract through the pretreated chromatography column at a flow rate of 1.0-2.0 column volumes per hour, control the pH value of the sample solution to be 3.0-4.0, and stop adsorption when the ephedra root tannin concentration in the effluent reaches 10%-20% of the initial concentration of the sample solution; S2: flush the chromatography column with deionized water until the absorbance value of the effluent at a wavelength of 280nm is stable at 0.03 -0.08, desorb with 70%-85% ethanol solution at a flow rate of 0.8-1.2 column volumes per hour, and collect desorbed liquid with a volume of 2-4 column volumes; S3: Determine the wet bulk density of the resin after adsorption, transfer the resin to a measuring cylinder of known mass, let it settle for 20-40 minutes, record the resin volume and weigh the total mass, and calculate the wet bulk density using the formula (total mass - measuring cylinder mass) / resin volume; S4: Input the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, and the measured wet bulk density into the evaluation model, calculate the score, and select the macroporous resin with the highest score.

[0024] For example, the concentration of the ethanol solution may be selected as 60%, 65%, or 70%, the flushing flow rate may be selected as 1.5 times, 2.0 times, or 2.5 times the column volume per hour, the flushing time may be selected as 2.5 hours, 3.0 hours, or 3.5 hours, the effluent conductivity control target may be selected as 15 μS / cm, 20 μS / cm, or 25 μS / cm, the loading flow rate may be selected as 1.0 times, 1.5 times, or 2.0 times the column volume per hour, the loading solution pH value may be selected as 3.0, 3.5, or 4.0, the effluent concentration threshold for stopping adsorption may be selected as 10%, 15%, or 20%, the ethanol concentration for desorption may be selected as 70%, 75%, 80%, or 85%, the desorption flow rate may be selected as 0.8 times, 1.0 times, or 1.2 times the column volume per hour, the desorption solution collection volume may be selected as 2 times, 3 times, or 4 times the column volume, and the standing sedimentation time may be selected as 20 minutes, 30 minutes, or 40 minutes. Suitable equipment includes glass chromatography columns, conductivity meters (such as the DDS-307), UV-visible spectrophotometers (such as the UV-2550), electronic balances, and graduated cylinders. For macroporous resins, commercially available options include AB-8, D101, and X-5. To calculate wet bulk density, first weigh a graduated cylinder of known mass, then add the resin and weigh again to determine the total mass. After allowing the cylinder to settle, read the resin volume and calculate using the formula (total mass - graduated cylinder mass) / resin volume.

[0025] For example, the following data were obtained from the experimental measurements of three macroporous resins A, B, and C: Resin A: the tannin adsorption capacity per unit time in the adsorption stage was 80 mg / h, the tannin recovery rate in the desorption stage was 85%, and the wet bulk density was 1.15 g / cm³; Resin B: the tannin adsorption capacity per unit time in the adsorption stage was 75 mg / h, the tannin recovery rate in the desorption stage was 90%, and the wet bulk density was 1.20 g / cm³; if the evaluation model adopts the linear weighting method, the weights of the adsorption capacity, recovery rate, and bulk density are set to 0.4, 0.3, and 0.3 respectively, and each parameter is converted to a 0-100 score in a normalized manner (for example, the adsorption capacity is full score of 100 mg / h and 50 mg / h is 0 point, the recovery rate is full score of 100% and 70% is 0 point, and the bulk density is 1.2 g / cm³ is the central value, and the farther away from it, the lower the score). The calculation process is as follows: Resin A: adsorption capacity score 80 points, recovery score 85 points, bulk density score 90 points, total score is 80×0.4 + 85×0.3 + 90×0.3 = 84.5 points; Resin B: adsorption capacity score 75 points, recovery score 90 points, bulk density score 80 points, total score is 75×0.4 + 90×0.3 + 80×0.3 = 81 points.

[0026] During use, the macroporous resin is first loaded into a chromatography column and pretreated with a specific concentration of ethanol solution and deionized water. Then, an extract of Ephedra sinica root is passed through the column at a specified flow rate. The pH value is controlled and adsorption is stopped at the appropriate concentration. The column is then rinsed with deionized water and desorbed. The desorbed liquid is collected, and the resin's wet bulk density is measured. Finally, the relevant data is input into an evaluation model to calculate a score to screen the resin. This method can provide a quantitative basis for the screening of macroporous resins through standardized pretreatment, adsorption, desorption, and measurement steps, helping to improve the scientific nature and accuracy of the screening process.

[0027] In another embodiment, it also includes: determination of resin specific surface area, after vacuum degassing the pretreated resin, the specific surface area is determined by nitrogen adsorption method; determination of coefficient of variation of desorption liquid concentration, detecting the tannin concentration of the desorption liquid by liquid chromatography, and calculating the coefficient of variation as a percentage of the standard deviation to the average value; determination of dynamic adsorption mass transfer coefficient, obtaining the mass transfer coefficient by fitting and calculating the adsorption kinetic model based on the adsorption breakthrough curve data; inputting the adsorption amount of ephedra root tannin per unit time in the adsorption stage, the recovery rate of ephedra root tannin in the desorption stage, the measured value of wet bulk density, the resin specific surface area, the coefficient of variation of desorption liquid concentration, and the dynamic adsorption mass transfer coefficient into the evaluation model to calculate the score.

[0028] For example, determination of resin specific surface area requires vacuum degassing of the pretreated resin. Nitrogen adsorption can be performed using a specific surface area analyzer (e.g., Micromeritics ASAP 2020). Desorption solution concentration can be measured using a high-performance liquid chromatograph (e.g., Agilent 1260 Infinity), with the coefficient of variation calculated as the percentage of the standard deviation to the mean. Dynamic adsorption mass transfer coefficient determination is based on the adsorption breakthrough curve, employing either a pseudo-first-order or pseudo-second-order adsorption kinetic model. Specifically, changes in tannin concentration in the effluent are recorded during the adsorption process to generate an "adsorption breakthrough curve" (i.e., a curve showing the change in effluent concentration over time or volume). Specifically, an ephedra root extract is passed through the chromatography column at a fixed flow rate. From the start of adsorption, the tannin concentration in the effluent is measured at regular intervals until the concentration approaches the initial concentration of the sample. This results in a curve that initially rises slowly and then rapidly approaches the initial concentration. To calculate the mass transfer coefficient, an appropriate adsorption kinetic model (e.g., one that describes the relationship between the diffusion rate of a substance and the adsorption process) is used to fit the breakthrough curve. In practice, experimentally measured concentration data is input into the model using data analysis software (such as Origin or MATLAB). The software automatically adjusts the model parameters to make the model curve as close to the experimental data as possible. The key parameter directly related to the mass transfer rate obtained after adjustment is the dynamic adsorption mass transfer coefficient.

[0029] When used as a whole, the previous example adds measurements of resin specific surface area, coefficient of variation of desorption solution concentration, and dynamic adsorption mass transfer coefficient. These parameters, along with the previously described adsorption capacity, recovery rate, and wet bulk density, are input into the evaluation model to calculate the score. By incorporating more parameters characterizing resin performance, the suitability of macroporous resins for the purification of ephedra root tannins can be more comprehensively evaluated, providing a richer basis for screening and helping to select resins with superior overall performance.

[0030] In another embodiment, the first sub-model is based on a linear regression algorithm, and its input parameters are the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, and the wet bulk density of the resin, and the first sub-model score is output; the second sub-model is based on a support vector regression algorithm, and its input parameters include the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the coefficient of variation of the desorption liquid concentration, and the dynamic adsorption mass transfer coefficient, and the second sub-model score is output after mapping the nonlinear relationship through the kernel function; the third sub-model is based on a random forest algorithm, and its input parameters include the adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, the wet bulk density of the resin, the specific surface area of ​​the resin, and the coefficient of variation of the desorption liquid concentration, and the third sub-model score is output after parallel training of multiple decision trees; according to the historical prediction error of each sub-model on the verification data set, the weight ratio of the first sub-model, the second sub-model, and the third sub-model is dynamically allocated, and the scores of each sub-model are added according to the dynamic weight to generate the final comprehensive score.

[0031] Specifically, the first sub-model uses a linear regression algorithm to process the three parameters of adsorption capacity, recovery rate, and wet bulk density. The second sub-model is based on a support vector regression algorithm, and the kernel function can be selected from the Gaussian kernel, with input parameters including adsorption capacity, coefficient of variation, and mass transfer coefficient. The third sub-model is based on a random forest algorithm, with input parameters including adsorption capacity, recovery rate, wet bulk density, specific surface area, and coefficient of variation, and is trained in parallel using multiple decision trees. Data processing can be implemented on a computer using programming languages ​​such as Python, with the help of machine learning libraries such as Scikit-learn.

[0032] Three sub-models from different algorithms were constructed, each with its own parameters entered for scoring. Weights were then dynamically assigned based on historical forecast errors, and the scores of each sub-model were weighted together to produce a final composite score. This multi-model approach leverages the strengths of different algorithms, considers both linear and nonlinear relationships between various parameters, improves the accuracy and robustness of the evaluation model, and makes screening results more reliable.

[0033] In another embodiment, the kernel function of the second sub-model is a Gaussian kernel with a kernel width σ∈[0.25,0.35], a regularization parameter C∈[1.5,2.5], and an insensitive loss parameter ε≤0.03; the number of decision trees of the third sub-model is n_trees∈[80,120], the maximum depth of a single tree is max_depth∈[12,18], and the minimum number of samples for node splitting is min_samples_split≥15.

[0034] For example, the Gaussian kernel width σ of the second sub-model might be 0.25, 0.30, or 0.35, the regularization parameter C might be 1.5, 2.0, or 2.5, and the insensitive loss parameter ε might be 0.01, 0.02, or 0.03. The number of decision trees n_trees of the third sub-model might be 80, 100, or 120, the maximum depth of a single tree max_depth might be 12, 15, or 18, and the minimum number of samples for node splitting min_samples_split might be 15, 20, 25, or other values ​​≥ 15. These parameter settings can be configured in machine learning libraries, such as the SVR and RandomForestRegressor modules in Scikit-learn.

[0035] A Gaussian kernel support vector regression algorithm with specific parameters is used in the second sub-model, and appropriate random forest parameters are set in the third sub-model. Through the optimization of these parameters, the sub-model can better handle the nonlinear relationship and complex characteristics of the input data, improve the prediction accuracy of the sub-model, and thus improve the accuracy of the final comprehensive score, which helps to more accurately screen macroporous resins.

[0036] In another embodiment, when the dynamic adsorption mass transfer coefficient K ≥ 0.45 cm / s, the weight W2 of the second submodel is increased to 50%-60%; when the coefficient of variation of the desorption liquid concentration CV ≤ 12%, the weight W3 of the third submodel is reduced to 15%-25%; when the resin specific surface area S ≥ 400 m² / g and the wet bulk density ρ ≤ 1.15 g / cm³, the weight W1 of the first submodel is fixed at 20%, and the weight W2 of the second submodel and the weight W3 of the third submodel are distributed in a ratio of 2:3; the adjustment step of each submodel weight does not exceed ±5% / minute, and the total weight distribution error is controlled within ±1.5%.

[0037] For example, the threshold value of the dynamic adsorption mass transfer coefficient K is 0.45 cm / s. When this value is reached, the weight of the second submodel, W2, is between 50% and 60%, and may be 50%, 55%, or 60%. The threshold value of the coefficient of variation (CV) of the desorption solution concentration is 12%. When it is below this value, the weight of the third submodel, W3, is between 15% and 25%, and may be 15%, 20%, or 25%. When the resin specific surface area S is ≥ 400 m² / g and the wet bulk density ρ is ≤ 1.15 g / cm³, the weight of the first submodel, W1, is fixed at 20%, and the remaining 80% is distributed according to W2:W3 = 2:3, that is, W2 is 32%, W3 is 48%, and so on. The weight adjustment step size does not exceed ±5% / minute, and the total weight error is controlled within ±1.5%. These conditional judgments and weight adjustments can be implemented through programming in the data processing system.

[0038] Based on the actual measured values ​​of different parameters, the weights of each sub-model are dynamically adjusted according to preset rules to ensure that the weight distribution is more closely aligned with the performance characteristics of the resin. For example, when the mass transfer coefficient is high, the second sub-model is emphasized, and when the coefficient of variation is low, the third sub-model is weighted less. When the specific surface area and bulk density are specific, the weights are assigned according to a fixed ratio. This ensures that the comprehensive score can better highlight the current key performance indicators, improving the targetedness and rationality of the screening.

[0039] In another embodiment, when the column temperature T∈[25°C, 35°C], the second sub-model weight W2 is modified according to the formula: , and the sum of weights is kept at 100% through normalization.

[0040] The column temperature, T, ranges from 25°C to 35°C. The base temperature in the correction formula is 30°C. For every 1°C deviation from the base temperature, the weight of the second submodel is adjusted by a factor of 0.02. For example, when T = 25°C, W2′ = W2 × (1 - 0.1) = 0.9W2; when T = 35°C, W2′ = W2 × (1 + 0.1) = 1.1W2. After weight adjustment, normalization is performed to ensure that the sum of the weights of the three submodels is always 100%, avoiding weight imbalance. Temperature monitoring can be performed using thermocouples or temperature sensors (such as PT100), connected to the data processing system to obtain real-time temperature data and perform weight correction calculations.

[0041] When the temperature of the chromatography column changes, the weight of the second sub-model is linearly corrected according to the temperature. Taking into account that temperature may affect the adsorption and desorption processes, and thus affect related parameters such as the mass transfer coefficient, this correction makes the weight distribution more in line with actual working conditions, improves the adaptability of the evaluation model to temperature changes, and ensures the reliability of the screening results under different temperature conditions.

[0042] In another embodiment, when a pH fluctuation ΔpH>0.3 is detected during the adsorption phase, the fault tolerance process is activated: the output value of the first sub-model is multiplied by the attenuation factor , the kernel function of the second sub-model is switched to a polynomial kernel, and the depth of the decision tree of the third sub-model is temporarily reduced to 10 layers.

[0043] The threshold of pH fluctuation ΔpH is 0.3, and fault tolerance is activated when the value exceeds this value. The attenuation factor is , for example, when ΔpH=0.4, the attenuation factor is approximately The kernel function of the second sub-model is switched from a Gaussian kernel to a polynomial kernel. The degree of the polynomial kernel can remain at the default or be set according to the situation. The depth of the decision tree of the third sub-model is temporarily reduced from the original 12-18 layers to 10 layers to reduce the complexity of the model. pH value detection can use a pH meter (such as the S40 model) to monitor the pH value in real time during the adsorption stage. When the fluctuation exceeds the threshold, the data processing system triggers the corresponding fault tolerance mechanism.

[0044] When the pH value fluctuates greatly during the adsorption stage, it may affect the stability of the adsorption effect. By attenuating the output of the first sub-model, switching the kernel function of the second sub-model, and reducing the depth of the decision tree of the third sub-model, the evaluation model can maintain a certain robustness under abnormal working conditions, reduce the adverse effects of pH fluctuations on the screening results, and improve the fault tolerance and reliability of the method.

[0045] In another embodiment, the method further comprises: selecting the two macroporous resins with the highest scores for secondary verification, repeating the adsorption-desorption process three times under the same operating conditions, calculating the standard deviation of the purity of the ephedra root tannins of the three experiments, and retaining the resins with a standard deviation less than 1.0%-2.0% as the final screening results.

[0046] For secondary validation, select the two resins with the highest scores and repeat the experiment three times under the same conditions. Calculate the standard deviation of purity, with standard deviation thresholds set at less than 1.0%, 1.5%, and 2.0%, etc. Experimental conditions must remain consistent with the previous ones, including column parameters, solution concentration, flow rate, and pH. Purity testing can be performed using high-performance liquid chromatography or other appropriate analytical methods to ensure the accuracy of the test results.

[0047] After initially selecting the two highest-scoring resins, secondary validation experiments were repeated, purity standard deviations were calculated, and the resins with good stability (smallest standard deviations) were retained. This step further verifies the repeatability and stability of the resins in actual operation, avoids screening errors caused by accidental factors, ensures the final selected resins have reliable performance in actual applications, and improves the practicality of the screening results.

[0048] In another embodiment, each cycle must meet the following conditions: the temperature control accuracy of the chromatography column is ±1°C, and the fluctuation range of the desorption liquid flow rate is ≤±5% of the set value; the purity of the ephedra root tannin obtained in each cycle is detected, and the standard deviation σ of the three purities is calculated. If σ∈[1.0%, 2.0%], the resin is determined to be qualified; when two resins are qualified, the one with the smaller σ value is selected as the final screening result; if only one is qualified, it is directly selected; if both are unqualified, the scores are re-evaluated, and the top two resins with the highest scores are excluded.

[0049] The column temperature should be controlled with an accuracy of ±1°C. A column equipped with a temperature control device (e.g., a constant-temperature column) should be used. The desorption solution flow rate should fluctuate within ±5% of the set value and be controlled by a peristaltic pump (e.g., BT100-2J). After purity testing, calculate the triplicate standard deviation (σ), with an acceptable range of 1.0% to 2.0%. If both resins pass the test, select the one with the smaller σ. If only one passes, select that one. If both fail, re-evaluate and exclude the first two.

[0050] During secondary validation, experimental conditions are strictly controlled to ensure temperature and flow rate stability, and the resin's eligibility and quality are determined by calculating the purity standard deviation. This rigorous validation and screening process further ensures that the final selected resin exhibits stable purification performance under the same conditions, improving the reliability and effectiveness of screening results and providing a more stable resin selection for actual production applications.

[0051] In another embodiment, the selected resins were treated under accelerated aging conditions, wherein the conditions were: temperature 40±2° C., relative humidity 75±5%, for 72-120 hours; after the aging treatment, the adsorption-desorption process was repeated, and the tannin adsorption decay rate ΔQ before and after aging was calculated. ; Where Q0 is the adsorption amount before aging, Q1 is the adsorption amount after aging; retain resin with ΔQ≤5.0%.

[0052] Accelerated aging conditions are 40±2°C, 75±5% relative humidity, and durations of 72, 96, or 120 hours. Aging can be performed in a constant temperature and humidity chamber (such as the LHS-100CH). After aging, the adsorption-desorption process is repeated, and the decay rate ΔQ is calculated, with a threshold of ≤5.0%. To calculate this, first measure the adsorption amount before aging, Q0, and the adsorption amount after aging, Q1, and then use the formula to calculate the decay rate.

[0053] Screened resins are subjected to accelerated aging to simulate actual aging conditions. The adsorption capacity decay rate before and after aging is measured, and resins with the lowest decay rates are retained. This step assesses the durability and stability of the resins, ensuring that the selected resins maintain optimal performance over long-term use, minimizing the degradation of purification efficiency due to resin aging, and improving the durability and reliability of the screening results in real-world applications.

[0054] In another embodiment, the screening method comprises:

[0055] Commercially available macroporous resins, types X, Y, and Z, were loaded into glass chromatography columns with an inner diameter of 2.5 cm and a height of 50 cm (column volume approximately 245 mL). The columns were first rinsed with 65% ethanol at a flow rate of 2.0 column volumes per hour (490 mL / h) for 3.0 hours, followed by deionized water flushing until the effluent conductivity reached 18 μS / cm. An extract of Ephedra sinica root at pH 3.5 was loaded at a flow rate of 1.5 column volumes per hour (367.5 mL / h). Adsorption was terminated when the tannin concentration in the effluent reached 15% of the initial concentration in the loading solution.

[0056] After rinsing with deionized water until the absorbance at 280 nm reached a stable level of 0.05, desorption was performed with 80% ethanol solution at a flow rate of 1.0 column volume / hour (245 mL / h), and 3 column volumes (735 mL) of the desorbate were collected.

[0057] Transfer the adsorbed X-type resin to a 100 mL graduated cylinder of known mass (50 g), let it stand for 30 minutes, record the resin volume as 80 mL, weigh the total mass as 135 g, and calculate the density as (135-50) / 80=1.06 g / cm³.

[0058] The pretreated X-type resin was vacuum degassed and measured using a Micromeritics ASAP 2020 surface area analyzer. The specific surface area was 450 m 2 / g.

[0059] The tannin concentration of the X-type resin desorption solution was detected by Agilent 1260 high performance liquid chromatography. The values ​​of three repeated measurements were 25.1 mg / mL, 24.5 mg / mL, and 25.8 mg / mL, with an average of 25.13 mg / mL, a standard deviation of 0.65, and a coefficient of variation of 0.65 / 25.13×100%=2.59%.

[0060] The liquid film diffusion model was fitted using Origin software based on the X-type resin adsorption breakthrough curve data, and the mass transfer coefficient K = 0.48 cm / s was obtained.

[0061] First submodel (linear regression): Input the adsorption capacity (85 mg / h), recovery rate (88%), and bulk density (1.06 g / cm³) of resin X. The first submodel can assume the following model form: Score = a × adsorption capacity + b × recovery rate + c × bulk density + d, where a, b, c, and d are coefficients obtained through least squares fitting. After entering specific values, the first submodel score is calculated according to the fitted equation (this requires normalization or standardization, for example, converting the parameters to a 0-100 scale and then weighting them).

[0062] The second sub-model (Support Vector Regression, Gaussian kernel σ = 0.3, C = 2.0, ε = 0.02) was fed with inputs of adsorption capacity, coefficient of variation (2.59%), and mass transfer coefficient (0.48 cm / s). Kernel function and parameters: A Gaussian kernel (RBF kernel) was used with the following parameters: kernel width (sigma = 0.3), regularization parameter (C = 2.0), and insensitivity loss parameter (varepsilon = 0.02). The kernel function was used to map the input data into a high-dimensional space, constructing a nonlinear regression model. After kernel mapping of the input parameters, the trained SVR model was used to output the second sub-model score.

[0063] The third sub-model (Random Forest, n_trees=100, max_depth=15, min_samples_split=20): All six parameters listed above (adsorption capacity, recovery rate, bulk density, specific surface area, coefficient of variation, and mass transfer coefficient) are input. The algorithm is initialized to generate a specified number of decision trees (controlled by n_trees). For example, setting it to 100 generates 100 different trees. Training data for each tree is extracted from the original dataset using bootstrap sampling to form differentiated subsets. Samples not selected are called out-of-bag (OOB) data and can be used for subsequent validation. Next, for each tree, the algorithm recursively splits the tree starting from the root node. At each node, the algorithm checks whether the number of samples in the current node meets the min_samples_split requirement (for example, a value of 20 means a node must have at least 20 samples before a split is permitted). If not, the split is terminated, and a leaf node is generated directly, generating a prediction. If so, a random subset of features (feature randomness) is selected from all features. These features and their possible thresholds are iterated, and the feature and threshold that minimizes impurity (such as the Gini index or information gain) are selected for splitting. This process is repeated for the split child node until the maximum depth set by max_depth is reached (for example, a depth of 5 means a maximum of 5 splits from the root to a leaf node) or no further splits are possible. This splitting mechanism ensures that the complexity of each tree is both limited by its depth (max_depth prevents overfitting) and by the sample size threshold (min_samples_split) to avoid overfitting to extremely small populations. Once all trees are constructed, the random forest outputs a final result using an ensemble strategy: the predictions from all trees are averaged. After inputting 6 parameters, the third sub-model score is output through the random forest model.

[0064] The calculation process of W1, W2, and W3 of X-type resin is as follows:

[0065] 1. Initial weight distribution conditions

[0066] When the dynamic adsorption mass transfer coefficient (K ≥ 0.45 cm / s), the weight of the second sub-model W2 is increased to 50%-60%. For X-type resin, K = 0.48 is greater than 0.45, so the initial W2 = 55% (within the range of 50%-60%).

[0067] When the resin specific surface area (S≥400 m² / g) and the wet bulk density ρ≤1.15 g / cm³, the weight of the first sub-model W1 is fixed at 20%, and the remaining weights are distributed according to W2:W3= 2:3.

[0068] For X-type resin, S=450m² / g, ρ=1.06 g / cm³, which meets the requirements, W1=20%. The remaining 80% originally needed to be distributed in a 2:3 ratio, W2=32% and W3=48%. However, due to the priority trigger K≥0.45cm / s, W2 was increased to 55%.

[0069] 2. Temperature correction formula

[0070] When the column temperature T∈[25℃,35℃], the weight of the second sub-model is modified to ;

[0071] In type X resin, T=28℃, and substituting it into W2′=55%×[1+0.02×(28-30)]=52.8%.

[0072] 3. Weight normalization

[0073] Since W1+ W2' + W3= 100%

[0074] Combined with the above, W1=20%, W2'=52.8%, W3 =27.2%. The original W1=22% and W3=25.2% are fine-tuned by taking into account historical forecast errors.

[0075] The top two scoring resins, X and Y, were selected and adsorption-desorption was repeated three times under the same conditions (temperature 30 ± 1°C, flow rate fluctuation ≤ 5%) to determine the tannin purity:

[0076] The three purities of X-type resin were 92.1%, 91.8%, and 92.4%, with a standard deviation of 0.26%;

[0077] The purity of Y-type resin was 89.5%, 90.2% and 88.9% at three times, with a standard deviation of 0.65%.

[0078] Since the standard deviations of both are ≤2.0%, the type X resin with the smaller standard deviation is selected to proceed to the next step.

[0079] The X-type resin was placed in a constant temperature and humidity chamber at 40±2°C and a relative humidity of 75±5% and aged for 96 hours. After aging, the screening process was repeated. The adsorption capacity before aging was Q0=90 mg / g, the adsorption capacity after aging was Q1=86 mg / g, and the decay rate ΔQ=(90-86) / 90≈4.4%≤5.0%. Finally, the X-type resin was determined to be the screening result.

[0080] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A method for screening macroporous resins for purifying tannins from Ephedra root, characterized in that: include: S1: The macroporous resin to be screened is loaded into a chromatography column and flushed with a 60%-70% ethanol solution at a flow rate of 1.5-2.5 column volumes per hour for 2.5-3.5 hours, followed by flushing with deionized water until the conductivity of the effluent is less than 15-25 μS / cm; the ephedra root extract is passed through the chromatography column at a flow rate of 1.0-2.0 column volumes per hour, and the pH value of the loading solution is controlled at 3.0-4.

0. Adsorption is stopped when the ephedra root tannin concentration in the effluent reaches 10%-20% of the initial concentration of the loading solution; S2: Rinse the column with deionized water until the absorbance of the effluent at a wavelength of 280 nm stabilizes at 0.03-0.08, then desorb with 70%-85% ethanol at a flow rate of 0.8-1.2 column volumes per hour, collecting a desorbed solution with a volume of 2-4 column volumes; S3: Determine the wet bulk density ρ of the adsorbed macroporous resin. Transfer the macroporous resin to a graduated cylinder and allow it to settle for 20-40 minutes. Record the resin volume and weigh the total mass. The wet bulk density ρ is calculated as (total mass - graduated cylinder mass) / resin volume. S4: input the adsorption amount of ephedra root tannin per unit time in the adsorption stage, the recovery rate of ephedra root tannin in the desorption stage, and the wet bulk density ρ into the first sub-model, and output the first sub-model score; The adsorption amount of tannin from Ephedra sinica root per unit time, the coefficient of variation CV of desorption solution concentration and the dynamic adsorption mass transfer coefficient K were input into the second sub-model, and the nonlinear relationship was mapped by the kernel function to output the score of the second sub-model. The adsorption amount of ephedra root tannins per unit time in the adsorption stage, the recovery rate of ephedra root tannins in the desorption stage, the wet bulk density ρ, the specific surface area S of the macroporous resin, and the coefficient of variation CV of the desorption solution concentration are input into the third sub-model, and the third sub-model score is output after parallel training of multiple decision trees. Based on the historical prediction errors of each sub-model on the validation data set, the weight ratios of the first, second, and third sub-models are dynamically allocated. The scores of each sub-model are added together according to the dynamic weights to generate a final comprehensive score. The macroporous resin with the highest comprehensive score is selected. Among them, the specific surface area S of the macroporous resin was determined by vacuum degassing the pretreated resin and then using nitrogen adsorption method to determine the specific surface area S of the macroporous resin; the coefficient of variation of the desorption liquid concentration CV was determined by detecting the tannin concentration of the desorption liquid by liquid chromatography, and the coefficient of variation of the desorption liquid concentration CV was calculated as the percentage of the standard deviation to the average value; the dynamic adsorption mass transfer coefficient K was determined by fitting and calculating the adsorption kinetic model based on the adsorption breakthrough curve data.

2. The method for screening macroporous resin for purifying tannins from Ephedra root according to claim 1, wherein: The kernel function of the second sub-model is a Gaussian kernel with a kernel width σ∈[0.25,0.35], a regularization parameter C∈[1.5,2.5], and an insensitive loss parameter ε≤0.03; The number of decision trees of the third sub-model is n_trees∈[80,120], the maximum depth of a single tree is max_depth∈[12,18], and the minimum number of samples for node splitting is min_samples_split≥15.

3. The method for screening macroporous resin for purifying tannins from Ephedra root according to claim 2, wherein: When the dynamic adsorption mass transfer coefficient K ≥ 0.45 cm / s, the weight of the second sub-model W2 is increased to 50%-60%; When the coefficient of variation of the desorption solution concentration CV≤12%, the weight of the third sub-model W3 is reduced to 15%-25%; When the specific surface area S of the macroporous resin is ≥ 400 m² / g and the wet bulk density ρ is ≤ 1.15 g / cm³, the weight W1 of the first sub-model is fixed at 20%, and the weight W2 of the second sub-model and the weight W3 of the third sub-model are distributed with the remaining weight in a ratio of 2:3; The adjustment step of each sub-model weight does not exceed ±5% / minute, and the total weight distribution error is controlled within ±1.5%.

4. The method for screening macroporous resin for purifying tannins from Ephedra root according to claim 3, wherein: When the column temperature T∈[25℃,35℃], the second sub-model weight W2 is modified according to the formula: , and the sum of weights is kept at 100% through normalization.

5. The method for screening macroporous resin for purifying tannins from Ephedra root according to claim 3, wherein: When the pH fluctuation ΔpH>0.3 is detected during the adsorption phase, the fault tolerance process is activated: The output value of the first sub-model is multiplied by the attenuation factor , the kernel function of the second sub-model is switched to a polynomial kernel, and the depth of the decision tree of the third sub-model is temporarily reduced to 10 layers.

6. The method for screening macroporous resin for purifying tannins from Ephedra root according to claim 1, wherein: Also includes: The two macroporous resins with the highest scores were selected for secondary verification. The adsorption-desorption process was repeated three times under the same operating conditions. The standard deviation of the purity of ephedra root tannins in the three experiments was calculated, and the macroporous resins with a standard deviation less than 1.0%-2.0% were retained as the final screening results.

7. The method for screening macroporous resin for purifying tannins from Ephedrae root according to claim 6, wherein: Each cycle must meet the following requirements: column temperature T control accuracy ±1°C, desorption liquid flow rate fluctuation range ≤±5% of the set value; The purity of the Ephedra root tannin obtained in each cycle was tested, and the standard deviation σ of the three purities was calculated. If σ∈[1.0%, 2.0%], the macroporous resin was determined to be qualified. When both macroporous resins are qualified, the one with the smaller σ value is selected as the final screening result; if only one is qualified, it is directly selected; if both are unqualified, they are re-scored and the first two macroporous resins with the highest original scores are excluded.

8. The method for screening macroporous resin for purifying tannins from Ephedra root according to claim 7, wherein: The selected macroporous resin was treated under accelerated aging conditions, wherein the conditions were: temperature 40±2°C, relative humidity 75±5%, for 72-120 hours; After the aging treatment, the adsorption-desorption process of claim 1 is repeated, and the attenuation rate ΔQ of the adsorption amount of the ephedra root tannin before and after aging is calculated; , where Q0 is the adsorption amount before aging, and Q1 is the adsorption amount after aging; Retain macroporous resin with ΔQ≤5.0%.