Search

CN-121986895-A - AI-driven sweet tea glycoside syrup preparation and milk tea flavor adaptation method

CN121986895ACN 121986895 ACN121986895 ACN 121986895ACN-121986895-A

Abstract

The invention relates to an AI-driven sweet tea glycoside syrup preparation and milk tea flavor adaptation method, belonging to the field of intersection of artificial intelligence and food processing, wherein the method adopts an AI double engine driving mode, wherein a first engine classifies sweet tea raw materials through a random forest algorithm, optimizes extraction parameters based on a BP neural network and an NSGA-II algorithm, and utilizes AI dynamic regulation and control membrane filtration, double resin chromatography and double effect concentration processes to realize stable preparation of high-concentration syrup with the sweet tea glycoside extraction rate of more than or equal to 96%, the purity of more than or equal to 99% and the phenolic content of less than or equal to 0.05%; and the second engine trains a CNN sensory prediction model by constructing a flavor database of milk tea and syrup, inputs the flavor characteristics of the milk tea, rapidly outputs optimal adaptation parameters, and supports AI simulation evaluation and real-time fine adjustment. The invention realizes the standardization and high yield of syrup quality, shortens the flavor adaptation period of the milk tea from the traditional 15-30 days to 24 hours, ensures the adaptation success rate to reach 100 percent, and obviously improves the taste coordination and the production efficiency of the sugar-free milk tea.

Inventors

  • HUANG YUYUAN
  • WEI YULIN

Assignees

  • 黑马当先(金秀)绿色农业科技有限公司

Dates

Publication Date
20260508
Application Date
20260129

Claims (10)

  1. 1. An AI-driven sweet tea glycoside syrup preparation and milk tea flavor adaptation method is characterized by comprising the following steps: Preparation of AI-driven Jin Xiutian tea glycoside highly concentrated syrup: S11, classifying and preprocessing raw materials, namely collecting Jin Xiutian tea leaf raw material characteristics, constructing a database containing altitude, picking period, initial content C0 of rubusoside and phenolic content P0, training a classification model based on a random forest algorithm, outputting raw material grades, and automatically optimizing crushing parameters based on raw material moisture content by AI; S12, AI collaborative extraction parameter optimization, namely, constructing a BP neural network extraction prediction model by taking extraction temperature, solid-liquid ratio and extraction time as input variables and the extraction rate of rubusoside and the phenolic dissolution rate as output variables, performing multi-objective optimization by using an NSGA-II algorithm, outputting optimal extraction parameters, and monitoring and feeding back and adjusting in real time by using a sensor in the extraction process; s13, AI multistage purification and debitterizing, namely, based on the characteristics of raw materials and extracting solution, AI dynamically optimizes the membrane filtration operation pressure and temperature, the resin chromatography flow rate and the addition amount of active carbon, and finally ensures that the phenolic content in syrup is less than or equal to 0.05%; S14, AI stabilization concentration control, namely predicting double-effect concentration process parameters by adopting an LSTM time sequence model, detecting the concentration of syrup in real time by using a refractometer, performing closed-loop adjustment, controlling the final concentration of syrup to be 68+/-0.5 Brix, and regulating and controlling the addition amount of the stabilizer based on the viscosity AI; AI-driven linked milk tea flavor adaptation: s21, constructing a flavor characteristic database of milk tea and syrup, wherein the flavor characteristic database comprises aroma components and taste characteristics; S22, constructing a sensory prediction model based on a Convolutional Neural Network (CNN), inputting milk tea flavor characteristics, syrup flavor characteristics and syrup addition amount, and outputting milk tea taste scores and adaptation parameters; S23, AI simulates sensory evaluation and carries out real-time parameter fine adjustment, and the adaptation success rate is 100%; And S3, supporting an AI full-flow data center, namely constructing a distributed data platform, and realizing real-time synchronization and AI model iteration update of data of each link of extraction, purification and adaptation.
  2. 2. The method of claim 1, wherein the feedstock classification model input features in step S11 include 12 features of elevation 500-1200m, picking period 9-11 months, rubusoside content C0 10-14%, phenolic content P0 0.8-1.5%, class a feedstock requirement C0 > 12.5%, P0 > 1.2%.
  3. 3. The method of claim 1, wherein the training set samples of the BP neural network model in the step S12 are not less than 500 groups, the fitting degree R2 is not less than 0.95, the learning rate is 0.001-0.003, and the optimization range of the extraction process parameters is that the temperature is 65-70 ℃, the solid-liquid ratio is 1:12-1:13, and the time is 55-60min.
  4. 4. The method according to claim 1, wherein the membrane filtration in step S13 is performed by using an explosion-proof ceramic ultrafiltration membrane, the molecular weight cut-off is 5000Da, the operating pressure is 0.25-0.27MPa, the temperature is 42-44 ℃, and the AI dynamically adjusts the backwash period to 8-12 h/time according to the membrane flux to 15-18L/(m 2. H).
  5. 5. The method according to claim 1, wherein the double effect concentration process parameters in the step S14 are that the first effect temperature is 75-80 ℃, the vacuum degree is-0.06-0.07 MPa, the second effect temperature is 55-60 ℃, the vacuum degree is-0.08-0.09 MPa, the purity of the concentrated syrup is more than or equal to 99%, and the moisture is less than or equal to 13.5%.
  6. 6. The method of claim 1, wherein the flavor database in step S21 contains more than 50 pieces of milk tea product data from more than 20 linked milk tea enterprises, aroma characteristics are detected by GC-MS, and taste characteristics are detected by electronic tongue.
  7. 7. The method according to claim 1, wherein the input feature dimension of the CNN sensory predictive model in step S22 is 256 dimensions, the output is a mouth feel score of 1-5 minutes, and the adaptation parameter generation time is not more than 10 minutes.
  8. 8. The method of claim 1, wherein the AI parameter tuning logic in step S23 includes decreasing the syrup addition temperature by 2-3 ℃ if the aroma retention rate is less than 90% and increasing the syrup addition amount by 0.1-0.2 mL/500mL if the sweetness is insufficient.
  9. 9. A system for implementing the method of any one of claims 1-8, comprising: the AI raw material analysis unit comprises a near infrared spectrometer and a random forest classification model; The AI extraction control unit comprises a sensor array, a BP neural network extraction model and a real-time regulation and control module; the AI flavor adaptation unit comprises a GC-MS, an electronic tongue, a CNN sensory prediction model and a parameter optimization module; the distributed data center supports full-flow data synchronization and model iteration based on MySQL and Redis architecture.
  10. 10. The system of claim 9, wherein the AI flavor adaptation unit supports 24 hours online invocation, the system outputs adaptation parameters within 10 minutes after the cascade of enterprises inputs the milk tea type through the client, and the parameter library is updated in synchronization with the new milk tea product.

Description

AI-driven sweet tea glycoside syrup preparation and milk tea flavor adaptation method Technical Field The invention relates to the technical field of intersection of artificial intelligence and food processing, in particular to an AI-driven sweet tea glycoside syrup preparation and milk tea flavor adaptation method. Background Along with popularization of health consumption concepts, the sugar-free milk tea becomes a core product category and a key growing point of a chain drink enterprise. However, the large-scale development and quality upgrading of the products are facing a series of technical bottlenecks to be solved. First, at the level of sugar substitute raw materials, the existing mainstream natural sugar substitutes have remarkable sensory defects. For example, steviol glycosides have a high sweetness, but generally have a bitter aftertaste that is difficult to mask, and erythritol has an insufficient sweetness and is prone to cool at high concentrations. These inherent defects lead to difficulty in realizing the synergy and balance of flavors with different tea bases (such as mellow black tea, fresh green tea and flower and fruit fragrance of oolong tea) and diversified milk bases (creamy fresh milk and rich non-dairy creamer), and result in unnatural mouthfeel and poor coordination of the final beverage, namely the industry pain point of difficult adaptation and poor mouthfeel. Secondly, in the syrup preparation process, the traditional rubusoside extraction and purification process is highly dependent on manual experience. Due to lack of dynamic response mechanism to raw material characteristics (such as fluctuation of rubusoside content and phenolic impurity content caused by different altitudes and picking periods), key technological parameters (such as temperature, time and solid-to-liquid ratio) are solidified, so that the fluctuation of syrup core indexes among production batches is large, for example, the concentration fluctuation can reach +/-5 Brix, and the removal of impurities with bitter and astringent taste such as phenols is incomplete. This makes the syrup quality unstable, and fails to meet the strict requirements of the interlocking enterprises on raw material standardization and consistency, namely the pain point of 'batch instability'. Although the artificial intelligence technology is primarily applied to agriculture optimization and food flavor analysis, two key technical blanks exist in solving the problem of the whole chain of the natural sugar industry. First, there is a lack of dynamic process optimization AI models for the fusion of feedstock properties for specific natural products (e.g., jin Xiutian theaglycosides). The existing model is mostly based on fixed raw material assumption, and cannot predict and optimize extraction and purification process parameters in real time according to a raw material database (such as altitude, picking period, initial content C0 and phenols P0), so that the process robustness is poor, and natural fluctuation of agricultural raw materials cannot be adapted. Secondly, an AI prediction and adaptation system which runs through the syrup physicochemical index-beverage sensory flavor is not established yet. Current sugar substitute applications remain in the manual trial-and-error stage and lack a systematic method for correlating big data, modeling and accurately predicting formulas of objective flavor fingerprint data (e.g., electronic tongue, GC-MS data) of syrup and sensory evaluation of target milk tea. This directly results in lengthy adaptation cycles (typically 15-30 days) for new products or syrups, and low success rates (about 60%), severely limiting product innovation efficiency and market response speed. In conclusion, the prior art system cannot realize stable, efficient and high-purity production of rubusoside syrup at the preparation end and cannot realize rapid and accurate flavor adaptation with a multi-component milk tea system at the application end. Therefore, an integrated method for deeply integrating AI into the whole process of raw material treatment, extraction and purification and terminal adaptation is urgently needed. Disclosure of Invention The invention provides a preparation method of AI-driven rubusoside syrup and a milk tea flavor adaptation method for solving the problems in the prior art. In order to solve the technical problems, the invention is realized by the following technical scheme that in the first aspect, the AI-driven sweet tea glycoside syrup preparation and milk tea flavor adaptation method comprises the following steps: Preparation of AI-driven Jin Xiutian tea glycoside highly concentrated syrup: S11, classifying and preprocessing raw materials, namely collecting Jin Xiutian tea leaf raw material characteristics, constructing a database containing altitude, picking period, initial content C0 of rubusoside and phenolic content P0, training a classification model based on a random forest algorit