A high-efficiency extraction and purification process for large-scale production of plant polyphenols

By optimizing process parameters through a dynamic extraction and purification framework model and a fusion optimization algorithm, the problem of multi-objective coupling in the extraction and purification of plant polyphenols was solved, enabling efficient and low-cost mass production and adapting to process adjustments for different raw material batches and production scales.

CN123120079APending Publication Date: 2026-11-17WENZHOU CELECARE MEDICAL EQUIP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611203590.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-10
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing plant polyphenol extraction and purification processes suffer from strong coupling and mutual constraints among polyphenol yield, purity, and energy consumption, resulting in long process development cycles and serious resource waste. In particular, small suboptimal processes can cause huge losses in large-scale production.

Method used

A dynamic extraction and purification framework model is adopted, and virtual process particle clusters are randomly generated. The yield, purity and energy consumption of polyphenols are calculated by combining mass transfer and adsorption kinetics models. The process parameters are optimized by using a fusion optimization algorithm with an adaptive mixed acquisition function and dual convergence criteria to achieve efficient extraction and purification of plant polyphenols.

Benefits of technology

It significantly shortens the process development cycle, reduces energy consumption costs, improves polyphenol yield and purity, and has strong robustness and flexibility, adapting to different raw material batches and production scales, thus realizing large-scale green manufacturing of plant polyphenols.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120079A_ABST
    Figure CN123120079A_ABST
Patent Text Reader

Abstract

This invention discloses a high-efficiency extraction and purification process for large-scale production of plant polyphenols. The invention relates to the field of plant active ingredient extraction and purification technology, and includes: establishing a dynamic extraction and purification framework model for plant raw materials; randomly generating several virtual process particles containing process parameters within the dynamic extraction and purification framework model; and transforming the originally complex, high-dimensional, and nonlinear extraction and purification process optimization problem into a feature matching fitness optimization problem in a feature space by establishing a dynamic virtual process framework model and randomly generating virtual process particle clusters. This avoids the local optima and inefficiency caused by traditional trial-and-error or single-factor rotation methods. Combined with the generalized distance evaluation of the target cluster centroid, it can effectively integrate multi-objective process requirements while eliminating the influence of dimensions and initial conditions, significantly improving the interpretability and practicality of the optimization results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of plant active ingredient extraction and purification technology, specifically a large-scale production process for the efficient extraction and purification of plant polyphenols. Background Technology

[0002] In the development of existing plant polyphenol extraction and purification processes, technicians typically rely on methods such as single-factor rotation, orthogonal experiments, or response surface methodology to adjust process parameters such as extraction temperature, time, liquid-to-solid ratio, ethanol concentration, and adsorbent type.

[0003] However, these traditional methods face the following problems: there is often a strong coupling and mutual constraint relationship between polyphenol yield, purity and energy consumption; the high-dimensional parameter space presents a complex response surface that is non-convex and multi-peaked; traditional methods are prone to getting trapped in local optima and cannot find the true global optimal process window; in addition, each set of process parameters requires a complete extraction and purification experiment, resulting in long process development cycles and high raw material and energy costs. Especially in large-scale production scenarios, even a small process non-optimality can cause huge waste of resources. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a high-efficiency extraction and purification process for large-scale production of plant polyphenols, solving the technical problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a large-scale production process for the efficient extraction and purification of plant polyphenols, comprising: Step S1: Establish a dynamic extraction and purification framework model for plant raw materials. In the dynamic extraction and purification framework model, randomly generate several virtual process particles containing process parameters. Use the virtual process particles as the initial operating conditions for the extraction and purification process to construct an initial virtual process particle cluster. Step S2: The mass transfer and adsorption kinetics model is used to calculate the polyphenol yield, purity and energy consumption of each virtual process particle in the virtual process particle cluster. Combined with the preset process constraints, the theoretical comprehensive performance index of the virtual extraction and purification process is calculated. Then, the theoretical comprehensive performance index is mapped to the virtual extraction and purification feature vector in the high-dimensional feature space. Step S3: Obtain the target yield, target purity, and energy consumption limit set during the actual extraction and purification process. Set the target yield and target purity as the target cluster centroid in the feature space. Calculate the generalized distance between the virtual extraction and purification feature vector and the target cluster centroid. Use the generalized distance as the feature matching fitness of virtual process particles belonging to the target process category. Step S4: Construct a non-convex optimization problem with the goal of achieving optimal feature matching fitness. Solve the non-convex optimization problem using a fusion optimization algorithm to obtain the optimal combination of process parameters, thereby achieving efficient extraction and purification of plant polyphenols. The fusion optimization algorithm uses a probabilistic surrogate model to fit the objective function. The core of the model adopts a multi-scale feature coupling kernel function. It adaptively weights and couples local non-smooth components with global isotropic components. It dynamically balances global exploration and local development through an adaptive hybrid acquisition function. Furthermore, the fusion optimization algorithm introduces a dual convergence criterion based on expected information gain and feature space compactness to dynamically evaluate the optimization state during the iteration process.

[0006] In some embodiments, step S2 specifically includes: using a mass transfer path algorithm based on the shortest path search approach to calculate the diffusion path length and mass transfer resistance of the extraction solvent corresponding to the virtual process particle in the plant tissue, and then calculating the theoretical dissolution amount of polyphenols; combining preset adsorption-desorption equilibrium parameters to calculate the polyphenol yield, purity and energy consumption indicators in the virtual extraction and purification process, and eliminating the influence of absolute process dimensions through centralization processing, and finally constructing a virtual extraction and purification feature vector that is independent of the initial operating conditions.

[0007] In some embodiments, in step S3, the target cluster centroid is constructed by mean-centering the actual set target yield, target purity, and energy consumption upper limit to eliminate interference caused by process baseline offset, so that the target cluster centroid represents an ideal process state in the feature space; the feature matching fitness is calculated using weighted Euclidean distance, and different feature components are assigned different confidence weight coefficients according to their measurement confidence to suppress feature distortion caused by detection noise or process fluctuations.

[0008] In some embodiments, the multi-scale feature coupling kernel function uses an adjustable multi-scale coupling factor to linearly weight the globally isotropic component describing the long-range mass transfer trend and the locally non-smooth component describing local process disturbances. The globally isotropic component is used to fit a macroscopic continuous surface showing the change in polyphenol yield with parameters such as temperature and time during the extraction and purification process. The locally non-smooth component is used to capture local fluctuations in the objective function caused by microscopic factors such as batch-to-batch differences in raw materials and adsorbent saturation. The kernel function also includes an observation noise variance term to characterize random errors in actual process measurements.

[0009] In some embodiments, the adaptive hybrid acquisition function employs a dynamic balancing weight function, which in the early stage of iteration focuses on exploring regions with unknown process parameters, and in the later stage of iteration focuses on mining known high-performance regions. The dynamic balancing weight function is an S-shaped curve, and its inflection point and switching rate can be preset or adjusted online according to the complexity of the process, thereby achieving a smooth transition from global search to local optimization.

[0010] In some embodiments, the dual convergence criteria include a feature space compactness criterion and an expected information gain criterion; the feature space compactness criterion is triggered when the minimum feature matching fitness in the historical record is lower than a preset compactness threshold; the expected information gain criterion is triggered when the moving average of the expected improvement calculated in multiple consecutive iterations is lower than a preset gain threshold; if either of the above two criteria is met, the iteration is terminated and the current optimal combination of process parameters is output.

[0011] In some embodiments, in the dynamic extraction and purification framework model, the process parameters of the virtual process particles include at least one of extraction temperature, extraction time, liquid-to-solid ratio, ethanol concentration, pH value, type and amount of adsorbent, elution flow rate, and eluent concentration; the process parameters are initially generated using a space-filling sampling method to ensure that the virtual process particle clusters are uniformly distributed throughout the entire feasible parameter domain.

[0012] In some embodiments, the probabilistic proxy model is constructed based on a Gaussian process. In each iteration, the mean function and covariance function of the Gaussian process are updated according to the feature matching fitness of all currently evaluated virtual process particles. Then, by maximizing the adaptive hybrid acquisition function, the next virtual process particle to be evaluated is selected, and its corresponding yield, purity, and energy consumption data are obtained through real experiments or high-fidelity simulations, thereby gradually narrowing the search range and approaching the global optimal process conditions.

[0013] In some embodiments, after the process completes an optimization solution, the optimal combination of process parameters obtained is used as the extraction and purification operation standard for the current batch, and the combination and its actual performance indicators are stored in the process knowledge base. When faced with a new batch of raw materials or a change in production scale, similar cases in the process knowledge base are called as the basis for generating the initial virtual process particle cluster, thereby realizing rapid adaptive migration of the batch production process.

[0014] In some embodiments, the process further includes an online feedback adjustment step: during actual mass production, key process variables in the extraction and purification process are collected in real time, and the key process variables are input into a pre-trained surrogate model to predict the feature matching fitness under the current process parameters; when the predicted fitness deviates from the optimal tolerance range, a compensatory process parameter adjustment amount is generated by solving the local optimization sub-problem, and automatically sent to the production line control system to achieve closed-loop adaptive control.

[0015] This invention provides a high-efficiency extraction and purification process for large-scale production of plant polyphenols, which has the following advantages: By establishing a dynamic virtual process framework model and randomly generating virtual process particle clusters, the originally complex, high-dimensional, and nonlinear extraction and purification process optimization problem is transformed into a feature matching fitness optimization problem in the feature space, avoiding local optima and low efficiency caused by traditional experience trial and error or single-factor rotation methods; Secondly, by using a mass transfer and adsorption kinetics model to calculate comprehensive performance indicators such as yield, purity, and energy consumption and mapping them to virtual extraction and purification feature vectors, combined with the generalized distance evaluation of the target cluster centroid, the invention can effectively integrate multi-objective process requirements while eliminating the influence of dimensions and initial conditions, significantly improving the interpretability and practicality of the optimization results. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process optimization of the present invention.

[0017] Figure 2 This is a schematic diagram of the iterative process of the fusion optimization algorithm of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 and Figure 2As shown, this invention provides a high-efficiency extraction and purification process technology for large-scale production of plant polyphenols, comprising: Step S1, establishing a dynamic extraction and purification framework model of plant raw materials, in which several virtual process particles containing process parameters are randomly generated, and the virtual process particles are used as the initial operating conditions for the extraction and purification process, thereby constructing an initial virtual process particle cluster; Step S2, using a mass transfer and adsorption kinetics model to calculate the polyphenol yield, purity, and energy consumption index corresponding to each virtual process particle in the virtual process particle cluster, and combining the preset process constraints to calculate the theoretical comprehensive performance index of the virtual extraction and purification process, and then mapping the theoretical comprehensive performance index to a virtual extraction and purification feature vector in a high-dimensional feature space; Step S3, obtaining the target yield, target purity, and energy consumption upper limit set in the actual extraction and purification process, and comparing the target yield with the target... Purity is set as the centroid of the target cluster in the feature space. The generalized distance between the virtual extraction and purification feature vector and the centroid of the target cluster is calculated. The generalized distance is used as the feature matching fitness to quantitatively characterize the virtual process particles belonging to the target process category. In step S4, a non-convex optimization problem is constructed with the optimal feature matching fitness as the objective. The problem is solved by a fusion optimization algorithm to obtain the optimal combination of process parameters to achieve efficient extraction and purification of plant polyphenols. The fusion optimization algorithm uses a probabilistic surrogate model to fit the objective function. The core of the model adopts a multi-scale feature coupling kernel function. It uses adaptive weighted coupling of local non-smooth components and global isotropic components. It uses an adaptive hybrid acquisition function to dynamically balance global exploration and local development. The fusion optimization algorithm introduces a dual convergence criterion based on expected information gain and feature space compactness to dynamically evaluate the optimization state during the iteration process.

[0020] Figure 2 This includes constructing a probabilistic surrogate model for a Gaussian process, fitting the response surface of the objective function using a multi-scale feature coupling kernel function, dynamically balancing global exploration and local development through an adaptive hybrid acquisition function, selecting the next virtual process particle for evaluation, updating the model and historical data, and controlling the iteration termination through a dual convergence criterion composed of a feature space density criterion and an expected information gain criterion, ultimately outputting the optimal combination of process parameters.

[0021] In the iterative process of the fusion optimization algorithm, a Gaussian process surrogate model is first constructed using initial evaluation data. The response surface of the objective function is then accurately fitted using a multi-scale feature coupling kernel function, simultaneously capturing macroscopic mass transfer trends and local process disturbances. Subsequently, an adaptive hybrid acquisition function dynamically balances global exploration and local development, guiding the selection of the next virtual process particle for real experiments or simulation evaluation, and updating the model and historical data. When the algorithm reaches the "dual convergence criterion satisfied?" check: if the criterion is not satisfied (no branch), the algorithm returns to the adaptive hybrid acquisition function, reselects sampling points for iteration, and continuously explores under-explored process parameter regions or further refines known optimal regions, ensuring that no global optimal solution is missed in the complex search space; if the criterion is satisfied (yes branch), the iteration immediately terminates and outputs the current optimal combination of process parameters, effectively avoiding the waste of computational resources caused by over-search and achieving an optimal balance between solution accuracy and computational cost. This dual convergence criterion, through the synergistic evaluation of feature space compactness and expected information gain, enables the algorithm to automatically stop when a satisfactory solution is reached or the model tends to converge, significantly improving optimization efficiency and engineering practicality.

[0022] By constructing a dynamic virtual process framework model and generating virtual process particle clusters, the traditional empirical trial-and-error approach is transformed into a fitness optimization problem in the feature space, effectively solving the technical problems of existing technologies that are prone to getting trapped in local optima, have slow convergence, and are difficult to balance multiple objectives. This scheme uses a mass transfer and adsorption kinetics model to calculate yield, purity, and energy consumption, and combines the target cluster centroids to construct a weighted generalized distance as the feature matching fitness, achieving multi-objective fusion and noise suppression. Based on this, a fusion optimization algorithm is used to solve the non-convex optimization problem: a multi-scale feature coupling kernel function is used to simultaneously fit the macroscopic mass transfer trend and local process perturbations, combined with an adaptive hybrid acquisition function to dynamically balance global exploration and local development, and a dual convergence criterion is introduced to automatically terminate the iteration. Therefore, this invention can approach the globally optimal process parameters with far fewer experiments than traditional methods, significantly shortening the process development cycle, reducing raw material and energy costs, and flexibly balancing the constraints of yield, purity, and energy consumption. The resulting comprehensive performance is significantly better than orthogonal experiments or response surface methodology, and it has strong robustness to batch-to-batch differences in raw materials, realizing intelligent and efficient large-scale green manufacturing of plant polyphenols.

[0023] In some embodiments, step S2 specifically includes: using a mass transfer path algorithm based on the shortest path search approach to calculate the diffusion path length and mass transfer resistance of the extraction solvent corresponding to the virtual process particles in the plant tissue, and then calculating the theoretical dissolution amount of polyphenols; combining preset adsorption-desorption equilibrium parameters to calculate the polyphenol yield, purity, and energy consumption indicators in the virtual extraction and purification process, and eliminating the influence of absolute process dimensions through centralization processing, and finally constructing a virtual extraction and purification feature vector that is independent of the initial operating conditions.

[0024] This method can more realistically reflect the physicochemical nature of extraction and purification, improve the accuracy of virtual evaluation, and enhance the universality and comparability of feature vectors through centralized processing.

[0025] In some embodiments, in step S3, the target cluster centroid is constructed by mean centering the actual set target yield, target purity and energy consumption upper limit to eliminate the interference caused by process baseline offset, so that the target cluster centroid represents an ideal process state in the feature space; the feature matching fitness is calculated using weighted Euclidean distance, and different feature components are assigned different confidence weight coefficients according to their measurement confidence to suppress feature distortion caused by detection noise or process fluctuation.

[0026] This mechanism effectively enhances the anti-interference capability of fitness assessment and the ability to flexibly adjust the priority of multiple objectives, making the optimization results more in line with actual production needs.

[0027] In some embodiments, the multi-scale feature coupling kernel function uses an adjustable multi-scale coupling factor to linearly weight the globally isotropic component describing the long-range mass transfer trend and the locally non-smooth component describing local process disturbances. The globally isotropic component is used to fit the macroscopic continuous surface of polyphenol yield as a function of parameters such as temperature and time during the extraction and purification process, while the locally non-smooth component is used to capture local fluctuations in the objective function caused by microscopic factors such as batch-to-batch differences in raw materials and adsorbent saturation. The kernel function also includes an observation noise variance term to characterize random errors in actual process measurements.

[0028] This kernel function can simultaneously fit the macroscopic trend and microscopic perturbation of the process response, significantly improving the surrogate model's ability to express complex non-convex, multi-peak objective functions and enhancing the algorithm's robustness.

[0029] In some embodiments, the adaptive hybrid acquisition function employs a dynamic balancing weight function, which in the early stage of iteration focuses on exploring regions with unknown process parameters, and in the later stage of iteration focuses on mining known high-performance regions. The dynamic balancing weight function is an S-shaped curve, and its inflection point and switching rate can be preset or adjusted online according to the complexity of the process, thereby achieving a smooth transition from global search to local optimization.

[0030] This strategy effectively avoids getting stuck in local optima in the early stages and the problem of low search efficiency in the later stages, while ensuring global convergence and accelerating the convergence speed.

[0031] In some embodiments, the dual convergence criteria include a feature space compactness criterion and an expected information gain criterion; the feature space compactness criterion is triggered when the minimum feature matching fitness in the historical record is lower than a preset compactness threshold; the expected information gain criterion is triggered when the moving average of the expected improvement calculated in multiple consecutive iterations is lower than a preset gain threshold; if either of the above two criteria is met, the iteration is terminated and the current optimal combination of process parameters is output.

[0032] This dual criterion can automatically terminate the iteration based on the optimization process, avoiding over-search and significantly reducing the consumption of computing resources, making it particularly suitable for rapid process iteration in mass production.

[0033] In some embodiments, in the dynamic extraction and purification framework model, the process parameters of the virtual process particles include at least one of the following: extraction temperature, extraction time, liquid-to-solid ratio, ethanol concentration, pH value, type and amount of adsorbent, elution flow rate, and eluent concentration; the process parameters are initially generated using a space-filling sampling method to ensure that the virtual process particle clusters are uniformly distributed throughout the entire parameter feasible domain.

[0034] This design ensures that the initial samples fully cover the parameter space, providing high-quality initial training data for the surrogate model and preventing optimization blind spots caused by sampling bias.

[0035] In some embodiments, the probabilistic proxy model is constructed based on a Gaussian process. In each iteration, the mean function and covariance function of the Gaussian process are updated according to the feature matching fitness of all currently evaluated virtual process particles. Then, by maximizing the adaptive hybrid acquisition function, the next virtual process particle to be evaluated is selected, and its corresponding yield, purity and energy consumption data are obtained through real experiments or high-fidelity simulations, thereby gradually narrowing the search range and approaching the global optimal process conditions.

[0036] This Bayesian optimization framework can obtain near-global optimal solutions with a very small number of real evaluations, which greatly reduces the number of experiments and raw material consumption, and lowers the process development cost.

[0037] In some embodiments, after the process completes an optimization solution, the optimal combination of process parameters obtained is used as the extraction and purification operation standard for the current batch, and the combination and its actual performance indicators are stored in the process knowledge base. When faced with a new batch of raw materials or a change in production scale, similar cases in the process knowledge base are called as the basis for generating the initial virtual process particle cluster, thereby realizing the rapid adaptive migration of the batch production process.

[0038] This knowledge base reuse mechanism significantly shortens the process re-optimization time for new batches or new scales, and improves the response speed and flexibility of the production line.

[0039] In some embodiments, the process further includes an online feedback adjustment step: during actual mass production, key process variables in the extraction and purification process are collected in real time, and the key process variables are input into a pre-trained surrogate model to predict the feature matching fitness under the current process parameters; when the predicted fitness deviates from the optimal tolerance range, a compensatory process parameter adjustment amount is generated by solving the local optimization sub-problem, and automatically sent to the production line control system to achieve closed-loop adaptive control.

[0040] This online feedback mechanism can respond to disturbances such as raw material fluctuations and equipment drift in real time, keeping the process running at its optimal state and significantly improving the stability and product consistency of mass production.

[0041] It is worth noting that all standard parts used in this invention can be purchased from the market, and irregularly shaped parts can be customized according to the description and drawings. The specific connection methods of each part all adopt conventional methods such as bolts, rivets, and welding that are mature in the prior art. The models of electrical structure equipment involved can be selected according to the user's needs, as long as they meet the requirements of this application. In addition, the circuit connection adopts conventional connection methods in the prior art. The supporting electrical structures such as the control, current detection, position feedback, predicted voltage synchronization and parameter adjustment of the electrical equipment are all existing technologies, such as PLC controllers and module structures, so they will not be described in detail here.

[0042] A specific embodiment of this application is described below with reference to the accompanying drawings: Example 1: Large-scale extraction and purification process of polyphenols from grape seeds This embodiment uses grape seeds as raw material and employs the high-efficiency extraction and purification process for plant polyphenols described in this invention for large-scale production. The specific implementation steps are as follows: Step S1: Establish a dynamic extraction and purification framework model First, based on the physical characteristics of grape seeds (particle size, cell wall structure) and the actual production line layout of the factory, a dynamic extraction and purification framework model was established, including equipment such as extraction tanks, centrifuges, macroporous adsorption resin columns, and vacuum concentrators. In this model, 500 virtual process particles were randomly generated. Each particle contained the following process parameters: extraction temperature (30℃~70℃), extraction time (30min~120min), liquid-to-solid ratio (6:1~20:1, unit L / kg), ethanol concentration (30%~80%), pH value (2.0~6.0), adsorbent type (one of AB-8, D101, XAD-7 and its application amount), elution flow rate (1~5 BV / h), and eluent concentration (50%~95%). The initial virtual process particle clusters were generated using Latin hypercube sampling to ensure uniform coverage across the entire feasible parameter domain.

[0043] Step S2: Calculate the virtual extraction and purification feature vector Using a mass transfer and adsorption kinetics model, the polyphenol yield (in gallic acid equivalent, mg / g raw material), purity (mass percentage of polyphenols in the extract), and energy consumption (including total electrical energy consumption for heating, stirring, pumping, and refrigeration, kWh / kg raw material) were calculated for each virtual process particle. Specifically, Fick's second law was used to simulate the diffusion process of ethanol solution within grape seed particles, and the dynamic adsorption-desorption behavior of the resin column was predicted using adsorption isotherms (Langmuir model). Subsequently, the yield, purity, and energy consumption of each particle were weighted and summed to obtain a comprehensive performance index, which was then mean-centered to construct a 12-dimensional virtual extraction and purification feature vector.

[0044] Step S3: Calculate feature matching fitness The actual production targets are: polyphenol yield ≥ 45 mg / g, purity ≥ 75%, and energy consumption ≤ 8 kWh / kg. These target values ​​are then subjected to the same mean-centering process to obtain the target cluster centroids in the feature space. The weighted Euclidean distance between the feature vector of each virtual particle and this centroid is the feature matching fitness. The confidence weight coefficients for the yield, purity, and energy consumption components are set to 0.4, 0.4, and 0.2, respectively, to reflect the priority given to yield and purity.

[0045] Step S4: Solve using the fusion optimization algorithm The fusion optimization algorithm described in claim 1 is used for iterative solution. The probabilistic surrogate model employs a multi-scale feature coupling kernel function. The feature scale parameter of the globally isotropic component is set to 15°, the scale parameter of the locally non-smooth component is set to 3°, and the multi-scale coupling factor is initially set to 0.6, subsequently adaptively adjusted according to the roughness of the response surface. In the adaptive hybrid acquisition function, the strategy switching inflection point is set to the 30th iteration, and the switching rate factor is set to 0.3. In the dual convergence criterion, the compactness threshold is set to 0.01, the expected information gain threshold is set to 0.001, and the sliding window length is set to 5 iterations.

[0046] After 47 iterations, the algorithm triggered the convergence criterion and output the optimal combination of process parameters: extraction temperature 58℃, extraction time 75 min, liquid-to-solid ratio 14:1, ethanol concentration 72%, pH 3.5, adsorbent AB-8 resin (1.2 times the raw material mass), elution flow rate 2.5 BV / h, and eluent concentration 85%. Under these conditions, three batches of large-scale production were carried out (each batch processing 500 kg of raw material), with an average yield of 46.8 mg / g, a purity of 77.2%, and a unit energy consumption of 7.6 kWh / kg, fully meeting the target requirements. Compared with the process optimized by traditional orthogonal experiments, the yield increased by 9%, the purity increased by 12%, and the energy consumption decreased by 18%.

[0047] Example 2: Continuous mass production process for high-purity polyphenol products using green tea as raw material This embodiment uses green tea as raw material, with the goal of preparing high-purity (≥90%) epigallocatechin gallate (EGCG) products, and the production scale is 2000 kg of dried tea leaves per batch.

[0048] Step S1: Dynamic Virtual Fault Framework Model (Adapted to Dynamic Virtual Process Framework Model) Based on the characteristics of thin green tea leaves and easily broken cell walls, an ultrasonic-assisted extraction device and a membrane separation module were added to the model. The parameter space of the virtual process particles was expanded to include: ultrasonic power (200W–800W), ultrasonic frequency (20kHz–40kHz), extraction temperature (room temperature–50℃), extraction time (20min–60min), liquid-to-solid ratio (8:1–15:1), macroporous adsorption resin type (HP20, ADS-17, HPD600), gradient elution program (three-step concentration and time combinations), and membrane molecular weight cutoff (1kDa–10kDa). The initial particle cluster size was set to 800, generated using a pseudo-Monte Carlo method.

[0049] Step S2: Eigenvector Calculation Due to the high purity requirements, the weight of the purity component is significantly increased. The comprehensive performance index is calculated as follows: Q = 0.2 × yield + 0.7 × purity - 0.1 × energy consumption (normalized). After calculating the yield, purity, and energy consumption of each virtual particle, a 10-dimensional feature vector is obtained through centering. In particular, to characterize the isomerization loss of EGCG, a thermal degradation kinetic model is introduced, and the purity decay factor extracted at temperatures exceeding 45℃ is embedded as a penalty term in the feature vector.

[0050] Step S3: Fitness Calculation The objectives were set as follows: EGCG yield ≥ 40 mg / g, purity ≥ 90%, and energy consumption ≤ 10 kWh / kg. The centroids of the target clusters were constructed in the same way as in Example 1. In the weighted Euclidean distance of the feature matching fitness, the reliability weight coefficient of the purity component was set to 0.7, the yield to 0.25, and the energy consumption to 0.05, in order to focus on ensuring the purity index.

[0051] Step S4: Fusion Optimization and Results Due to the extremely steep objective function (purity is highly sensitive to temperature), the algorithm repeatedly fell into local optima in the early stages. However, thanks to the local non-smooth components in the multi-scale feature coupling kernel function (smoothness parameter ν set to 1.5, which effectively fits discontinuous response surfaces) and the exploration mechanism in the adaptive hybrid acquisition function, the algorithm successfully escaped the local trap on the 62nd iteration and converged on the 89th iteration. The optimal process parameters were: ultrasonic power 550W, ultrasonic frequency 28kHz, extraction temperature 42℃, extraction time 35min, liquid-to-solid ratio 12:1, first using HPD600 resin for dynamic adsorption, then eluting impurities with 30% ethanol, then eluting EGGC with 70% ethanol, and finally filtering through a 5kDa membrane. Actual production verification showed an average yield of 41.2 mg / g, purity of 92.5%, and unit energy consumption of 9.2 kWh / kg. This process achieved stable large-scale production of high-purity EGCG, with a caffeine content of less than 0.3% in the finished product, meeting the requirements for high-end health product raw materials.

[0052] Example 3: Adaptive process migration for mass production processes to address batch-to-batch variations in raw materials This embodiment uses apple pomace as raw material, highlighting the adaptive optimization capability of this invention when there are significant batch variations in raw materials. The apple pomace comes from different producing areas and different harvesting seasons, and its moisture content, polyphenol composition, and cellulose content fluctuate significantly.

[0053] Step S1: Construct a transferable virtual process framework model In the dynamic extraction and purification framework model, raw material characteristic parameters are introduced as adjustable input variables, including: moisture content (10%–30%), total phenolic content (5–15 mg / g dry residue), and cellulose mass fraction (20%–40%). The generation of the initial virtual process particle cluster is no longer fixed. Instead, based on the measured characteristics of the current batch of raw materials, the optimal solution of similar raw material cases in the process knowledge base is used as the initial particle center, and then a small-scale random perturbation is performed in its neighborhood to generate 200 initial particles.

[0054] Step S2: Fast Feature Vector Calculation For apple pomace polyphenols (mainly composed of phlorizin and chlorogenic acid), a simplified mass transfer model is employed to reduce computational load. In this embodiment, the overall performance indicators focus on yield and cost, with purity being a secondary objective (≥60%). The feature vector has eight dimensions, including yield, purity, energy consumption, and raw material utilization rate.

[0055] Step S3: Fitness Calculation and Online Feedback Actual target settings: theoretical maximum extraction yield ≥ 85%, unit energy consumption ≤ 5 kWh / kg, purity ≥ 60%. In addition to offline optimization, this embodiment also performs the online feedback adjustment step described in claim 10: during production, the extract is sampled every 30 minutes, and the current polyphenol concentration and impurity composition are rapidly predicted using near-infrared spectroscopy. The data is then input into a pre-trained Gaussian process surrogate model. If the model's predicted feature matching fitness deviates from the optimal tolerance range (deviation exceeding 15%), a compensation adjustment is automatically generated, such as increasing the extraction temperature by 2°C, extending the extraction time by 5 minutes, or increasing the eluent concentration by 5%, and this is automatically adjusted via a PLC system.

[0056] Step S4: Optimize Results The method of this embodiment was applied to three different batches of apple pomace (with significant differences in characteristics). The first batch (high moisture content 28%, low cellulose 22%): the algorithm converged on the 18th iteration. The optimal process was an extraction temperature of 52℃, an extraction time of 45 min, a liquid-to-solid ratio of 10:1, and an ethanol concentration of 55%, achieving a yield of 92% and a purity of 67%. The second batch (low moisture content 12%, high cellulose 38%): due to cellulose hindering mass transfer, the initial particle clusters needed to be regenerated. After 23 iterations, the optimal process was obtained: the extraction temperature was increased to 65℃, the extraction time was extended to 70 min, and cellulase pretreatment (0.2% enzyme dosage) was added, achieving a yield of 88% and a purity of 63%. The third batch (medium characteristics): directly utilizing the knowledge from the first two batches, a satisfactory process was obtained with only 5 iterations of fine-tuning. Statistics show that after applying the method of this invention, the process switching time between different batches was shortened from an average of 4 hours to 35 minutes, and the standard deviation of yield between batches decreased from 8% to 2.5%, achieving adaptive and rapid migration in mass production.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-efficiency extraction and purification process for large-scale production of plant polyphenols, characterized in that, include: Step S1: Establish a dynamic extraction and purification framework model for plant raw materials. In the dynamic extraction and purification framework model, randomly generate several virtual process particles containing process parameters. Use the virtual process particles as the initial operating conditions for the extraction and purification process to construct an initial virtual process particle cluster. Step S2: The mass transfer and adsorption kinetics model is used to calculate the polyphenol yield, purity and energy consumption of each virtual process particle in the virtual process particle cluster. Combined with the preset process constraints, the theoretical comprehensive performance index of the virtual extraction and purification process is calculated. Then, the theoretical comprehensive performance index is mapped to the virtual extraction and purification feature vector in the high-dimensional feature space. Step S3: Obtain the target yield, target purity, and energy consumption limit set during the actual extraction and purification process. Set the target yield and target purity as the target cluster centroid in the feature space. Calculate the generalized distance between the virtual extraction and purification feature vector and the target cluster centroid. Use the generalized distance as the feature matching fitness of virtual process particles belonging to the target process category. Step S4: Construct a non-convex optimization problem with the goal of achieving optimal feature matching fitness, and use a fusion optimization algorithm to solve the non-convex optimization problem to obtain the optimal combination of process parameters, so as to achieve efficient extraction and purification of plant polyphenols. The fusion optimization algorithm uses a probabilistic surrogate model to fit the objective function. The core of the model adopts a multi-scale feature coupling kernel function, which adaptively weights and couples local non-smooth components with global isotropic components. It also dynamically balances global exploration and local development through an adaptive hybrid acquisition function. Furthermore, the fusion optimization algorithm introduces a dual convergence criterion based on expected information gain and feature space density to dynamically evaluate the optimization state during the iteration process.

2. The process for high-efficiency extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, Step S2 specifically includes: using a mass transfer path algorithm based on the shortest path search approach, calculating the diffusion path length and mass transfer resistance of the extraction solvent corresponding to the virtual process particles in the plant tissue, and then calculating the theoretical dissolution amount of polyphenols; combining preset adsorption-desorption equilibrium parameters, calculating the polyphenol yield, purity, and energy consumption indicators in the virtual extraction and purification process, and eliminating the influence of absolute process dimensions through centralization processing, finally constructing a virtual extraction and purification feature vector that is independent of the initial operating conditions.

3. The process for high-efficiency extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, In step S3, the centroid of the target cluster is constructed by mean centering the actual target yield, target purity and energy consumption limit. The feature matching fitness is calculated using weighted Euclidean distance. Different feature components are assigned different confidence weight coefficients according to their measurement confidence to suppress feature distortion caused by detection noise or process fluctuations.

4. The process for high-efficiency extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, The multi-scale feature coupling kernel function uses an adjustable multi-scale coupling factor to linearly weight the globally isotropic component describing the long-range mass transfer trend and the locally non-smooth component describing local process disturbances. The globally isotropic component is used to fit the macroscopic continuous surface of polyphenol yield as a function of parameters such as temperature and time during the extraction and purification process. The locally non-smooth component is used to capture local fluctuations in the objective function caused by microscopic factors such as batch-to-batch differences in raw materials and adsorbent saturation. The kernel function also includes an observation noise variance term to characterize random errors in actual process measurements.

5. The large-scale production process for efficient extraction and purification of plant polyphenols according to claim 1, characterized in that, The adaptive hybrid acquisition function adopts a dynamic balancing weight function, which in the early stage of iteration focuses on exploring regions with unknown process parameters, and in the later stage of iteration focuses on mining known high-performance regions. The dynamic balancing weight function is an S-shaped curve, and its inflection point and switching rate can be preset or adjusted online according to the complexity of the process, thereby achieving a smooth transition from global search to local optimization.

6. The process for efficient extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, The dual convergence criteria include a feature space density criterion and an expected information gain criterion; the feature space density criterion is triggered when the minimum feature matching fitness in the historical record is lower than a preset density threshold. When the moving average of the expected improvement calculated in multiple consecutive iterations is lower than the preset gain threshold, the expected information gain criterion is triggered; if either of the above two criteria is met, the iteration is terminated and the current optimal combination of process parameters is output.

7. The process for high-efficiency extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, In the dynamic extraction and purification framework model, the process parameters of the virtual process particles include at least one of the following: extraction temperature, extraction time, liquid-to-solid ratio, ethanol concentration, pH value, type and amount of adsorbent, elution flow rate, and eluent concentration. The process parameters are initially generated using a space-filling sampling method to ensure that the virtual process particle clusters are uniformly distributed throughout the entire feasible parameter domain.

8. The process for efficient extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, The probabilistic proxy model is constructed based on a Gaussian process. In each iteration, the mean function and covariance function of the Gaussian process are updated according to the feature matching fitness of all currently evaluated virtual process particles. Then, by maximizing the adaptive hybrid acquisition function, the next virtual process particle to be evaluated is selected, and its corresponding yield, purity and energy consumption data are obtained through real experiments or high-fidelity simulations, thereby gradually narrowing the search range and approaching the global optimal process conditions.

9. The process for efficient extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, After completing one optimization solution, the process obtains the optimal combination of process parameters as the extraction and purification operation standard for the current batch, and stores the combination and its actual performance indicators in the process knowledge base. When faced with a new batch of raw materials or a change in production scale, similar cases in the process knowledge base are called as the basis for generating the initial virtual process particle cluster, thereby realizing rapid adaptive migration of the batch production process.

10. The process for efficient extraction and purification of plant polyphenols for large-scale production according to claim 1, characterized in that, The process also includes an online feedback adjustment step: during actual mass production, key process variables in the extraction and purification process are collected in real time, and the key process variables are input into a pre-trained surrogate model to predict the feature matching fitness under the current process parameters; when the predicted fitness deviates from the optimal tolerance range, a compensatory process parameter adjustment amount is generated by solving the local optimization sub-problem, and automatically sent to the production line control system to achieve closed-loop adaptive control.