Method for preparing low high-alcohol content mead

Abstract

The application discloses a preparation method of low high-grade alcohol content honey wine, relates to the related technical field of honey wine, and aims to solve the technical problem of high high-grade alcohol content in honey wine. The method comprises the following steps: preparing a honey diluent; performing pasteurization treatment; adding potassium dihydrogen phosphate, magnesium chloride, yeast cream and an inorganic nitrogen source, uniformly mixing the inorganic nitrogen source to obtain a honey fermentation liquor, wherein the inorganic nitrogen source comprises directly added inorganic nitrogen source and inorganic nitrogen source wrapped in pH-responsive biodegradable capsules; using citric acid to adjust the pH value of the honey fermentation liquor to 3-4; activating active dry yeast in water at an addition amount of 0.2-0.4 g / L at 28-32 DEG C to obtain an activated yeast liquor; inoculating the activated yeast liquor into the honey fermentation liquor for fermentation; and performing real-time prediction on the high-grade alcohol content of the fermentation liquor through a machine learning model during the fermentation process. The application significantly reduces the high-grade alcohol content in the honey wine, and guarantees the fermentation efficiency and wine body quality.

Description

Technical Field

[0001] This invention relates to the technical field of mead. More specifically, this invention relates to a method for preparing mead with low higher alcohol content. Background Technology

[0002] Mead is an alcoholic beverage made primarily from honey and water through yeast fermentation, typically with an alcohol content between 3% vol and 18% vol (V / V). Classified by production method, mead can be categorized into fermented mead, distilled mead, and fortified mead. The main components of mead are water and ethanol, and it also contains abundant flavonoids, phenolic acids, amino acids, vitamins, minerals, and other physiologically active components, as well as volatile components such as esters, alcohols, and aldehydes. As a bee product, mead is believed to have health benefits such as improving digestion, regulating sleep, beautifying the skin, and delaying aging.

[0003] Higher alcohols, also known as higher fatty alcohols or fusel oils, are a collective term for mixtures of monohydric alcohols with three or more carbon atoms, mainly including n-propanol, n-butanol, isobutanol, active pentanol, isopentanol, and 2-phenylethanol. Higher alcohols are primary metabolites, mainly produced by yeast during the main fermentation stage of brewing. Appropriate amounts of higher alcohols can increase the harmony and fullness of the wine, impart a unique aroma to fermented wines, and enhance ester aromas, giving a pleasant feeling. However, when the concentration of higher alcohols is too high, off-flavors will be produced, making the wine bitter, harsh, and losing its original style. At the same time, excessive intake of higher alcohols can cause rapid congestion of the nervous system, and because higher alcohols are metabolized slowly and remain in the body for a long time, they can cause a rapid heartbeat, headache, and increased susceptibility to intoxication, leading to the so-called "headache" phenomenon, which is harmful to human health. Currently, there are two main ways to effectively control higher alcohols: one is to start with the fermentation strains, by selecting superior yeast strains that produce low levels of higher alcohols to fundamentally reduce the content; the other is to optimize the fermentation process by controlling raw materials, fermentation temperature, and other indicators. Summary of the Invention

[0004] One objective of this invention is to provide a method for preparing mead with low higher alcohol content, which can significantly reduce the higher alcohol content in mead while ensuring fermentation efficiency and quality.

[0005] To achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, a method for preparing mead with low higher alcohol content is provided, comprising: S1: dissolving honey in water at an addition amount of 300-350 g / L to prepare a honey dilution; S2: pasteurizing the honey dilution at 70-80 °C; S3: cooling the pasteurized honey dilution to 28-32 °C, adding potassium dihydrogen phosphate, magnesium chloride, yeast extract, and an inorganic nitrogen source, mixing evenly to obtain a honey fermentation broth, wherein the inorganic nitrogen source includes a directly added inorganic nitrogen source and an inorganic nitrogen source encapsulated in a pH-responsive biodegradable capsule; S4: adjusting the pH of the honey fermentation broth to 3-4 using citric acid; S5: adding active dry yeast at an addition amount of 0.2-0.4 g / L at 28-32 °C. S6: Activate the yeast in water at ℃ to obtain activated yeast liquid; S7: Inoculate the activated yeast liquid into honey fermentation liquid for fermentation. During the fermentation process, the higher alcohol content of the fermentation liquid is predicted in real time by a machine learning model, and the fermentation temperature and pH of the fermentation liquid are dynamically adjusted according to the prediction results.

[0006] Furthermore, in S3, the amounts of potassium dihydrogen phosphate, magnesium chloride, yeast extract, and directly added inorganic nitrogen source are 0.5 g / L, 0.2 g / L, 0.15 g / L, and 0.4-0.8 g / L, respectively. The inorganic nitrogen source is ammonium chloride or diammonium hydrogen phosphate. The inorganic nitrogen source encapsulated in the pH-responsive biodegradable capsule contains 0.2-0.4 g / L of inorganic nitrogen source.

[0007] Furthermore, the active dry yeast is Ramania D47.

[0008] Further, in S3, the inorganic nitrogen source encapsulated in the pH-responsive biodegradable capsule is prepared by the following method: S31: Sodium alginate is dissolved in deionized water to prepare a sodium alginate solution with a mass-volume concentration of 1-2%; S32: The inorganic nitrogen source powder and the sodium alginate solution are mixed at a mass-volume ratio of 1:10 (g:mL), and thoroughly stirred to obtain a mixed slurry; S33: The mixed slurry is added dropwise to a mass-volume concentration of 2-3% using a syringe. In a calcium chloride solution, capsules are formed, cross-linked and cured for 10-20 minutes, and then filtered out to obtain calcium alginate gel capsules; S34: The calcium alginate gel capsules are immersed in a chitosan acetate solution with a mass-volume concentration of 0.5-1.0%, and allowed to stand for 10-20 minutes to allow chitosan to cross-link with calcium alginate through electrostatic interaction, forming a pH-responsive chitosan-sodium alginate composite membrane on the outer layer of the capsule; S35: The capsules are filtered out, washed with deionized water, and then freeze-dried to obtain the final product.

[0009] Furthermore, in S6, a training dataset is constructed, which contains multiple feature parameters collected at multiple time points during the historical fermentation process and their corresponding measured values ​​of higher alcohol content. The dataset is trained using a random forest regression algorithm, with higher alcohol content as the prediction target, to obtain a prediction model. During the fermentation process, the feature parameters are monitored in real time by sensors and input into the prediction model, which outputs the predicted value of the current higher alcohol content. If the predicted value exceeds the higher alcohol content threshold, the fermentation temperature and / or the pH value of the fermentation broth are dynamically adjusted by regulating the temperature control device and the acid / alkali addition pump.

[0010] Furthermore, an initial prediction model is obtained by training a subset of training data collected during the first 24 hours of fermentation, and the baseline prediction error E of the initial prediction model is calculated. The adjustment range of the number of decision trees is set based on the baseline prediction error E, with a lower limit of 100×(1+E) and an upper limit of 200×(1+E), where E is in g / L. The adjustment range of the maximum depth of the decision trees is set based on the average fluctuation range ΔpH of the pH value of the fermentation broth during the first 24 hours, with a lower limit of 10×(1+ΔpH) and an upper limit of 20×(1+ΔpH). The adjustment range of the minimum number of sample splits is set based on the average fluctuation range ΔT of the temperature of the fermentation broth during the first 24 hours, with a lower limit of 2×(1+ΔT) and an upper limit of 5×(1+ΔT), where ΔT is in °C. Every 24 hours of fermentation time, the model is retrained based on the newly collected fermentation data during that time period, and the adjustment ranges of the number of decision trees, the maximum depth of the decision trees, and the minimum number of sample splits are updated using the newly calculated E, ΔpH, and ΔT values.

[0011] Furthermore, during each model training or retraining session, the Pearson correlation coefficient between each candidate feature parameter in the current training dataset and the measured value of higher alcohol content is calculated. The candidate feature parameters include at least fermentation broth temperature, pH value, dissolved oxygen content, sugar content, yeast cell density, and fermentation time. Feature parameters with absolute Pearson correlation coefficient values ​​greater than the coefficient threshold are selected as valid input variables for this training and used to train the random forest regression model.

[0012] Furthermore, if the predicted higher alcohol content exceeds the higher alcohol content threshold but is below the first limit, the fermentation temperature is reduced; if the predicted higher alcohol content reaches or exceeds the first limit, both the fermentation temperature and the pH of the fermentation broth are reduced simultaneously.

[0013] The present invention has at least the following beneficial effects:

[0014] This invention, through the synergistic effect of multiple processes, can significantly reduce the content of higher alcohols in mead while effectively ensuring fermentation efficiency and wine quality. The pH-responsive biodegradable capsule, with its special structural design, dynamically adjusts the release rhythm of inorganic nitrogen sources according to changes in the pH of the fermentation broth. In the early stages of fermentation, the pH of the fermentation broth is relatively high, and the outer structure of the capsule is stable, releasing nitrogen sources only slowly. This avoids the problem of excessive nitrogen in the initial stage caused by adding nitrogen sources all at once in traditional processes, preventing explosive growth of yeast due to nutrient overload, and thus reducing the production of higher alcohols due to abnormal yeast metabolism. As fermentation progresses, the organic acids produced by yeast metabolism lower the pH of the fermentation broth. The outer structure of the capsule changes, accelerating the release of nitrogen sources and timely replenishing the nutrients needed for yeast growth and reproduction. This prevents metabolic disorders caused by insufficient nitrogen sources in the later stages, further reducing the formation of higher alcohols. A constant supply of nitrogen sources keeps the yeast in a stable metabolic state, effectively improving sugar conversion efficiency, ensuring the target alcohol content, and reducing residual sugar accumulation. Pasteurization can completely kill miscellaneous bacteria in honey dilution, providing a pure environment for yeast fermentation; suitable activation and fermentation temperatures can match the activity requirements of yeast and reduce abnormal metabolism; citric acid can regulate pH to both inhibit miscellaneous bacteria and create a suitable environment for yeast; machine learning models can predict the content of higher alcohols in real time and regulate temperature and pH, complementing the nitrogen source sustained-release mechanism of capsules, solving the problems of traditional processes relying on experience and lagging regulation, and ultimately achieving a unity of alcohol reduction, efficiency improvement and quality assurance.

[0015] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0016] Figure 1 A flowchart of one embodiment of this application;

[0017] Figure 2 The effects of different yeast strains on the higher alcohol content of mead were shown.

[0018] Figure 3 The effect of fermentation temperature on the higher alcohol content of mead is shown;

[0019] Figure 4 The effect of initial pH on the higher alcohol content of mead is shown;

[0020] Figure 5 The effect of honey addition on the higher alcohol content of mead was shown.

[0021] Figure 6 The effect of yeast addition on the higher alcohol content of mead is shown. Detailed Implementation

[0022] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0023] 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 only used to explain the relative positional relationship and movement of components in a specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed to" or "set on" another element, it can be directly on the other element or may have an intervening element present. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. Descriptions involving "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 indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0024] It should be noted that the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0025] The embodiments of this application provide a method for preparing mead with low higher alcohol content, comprising: S1: dissolving honey in water at an addition amount of 300-350 g / L to prepare a honey dilution; S2: pasteurizing the honey dilution at 70-80 ℃; S3: cooling the pasteurized honey dilution to 28-32 ℃, adding potassium dihydrogen phosphate, magnesium chloride, yeast extract, and an inorganic nitrogen source, mixing evenly to obtain a honey fermentation liquid, wherein the inorganic nitrogen source includes directly added inorganic nitrogen source and inorganic nitrogen source encapsulated in pH-responsive biodegradable capsules; S4: adjusting the pH value of the honey fermentation liquid to 3-4 using citric acid; S5: activating active dry yeast in water at 28-32 ℃ at an addition amount of 0.2-0.4 g / L to obtain an activated yeast liquid; S6: inoculating the activated yeast liquid into the honey fermentation liquid for fermentation, wherein during the fermentation process, the higher alcohol content of the fermentation liquid is predicted in real time using a machine learning model, and the fermentation temperature and pH of the fermentation liquid are dynamically adjusted according to the prediction results.

[0026] For example, in S1, the honey can be acacia honey or jujube honey with a Baume degree of 42°, and the water is deionized water treated by reverse osmosis. The amount added can be 320g / L or 340g / L. In a stainless steel mixing tank equipped with a paddle stirrer, the solution is stirred at 300r / min for 20 minutes until completely dissolved. A temperature sensor is installed inside the tank to monitor the solution temperature in real time. In S2, pasteurization is performed using an electrically heated jacketed kettle. The diluted solution is heated to 75°C or 78°C and maintained at this temperature for 30 minutes. During this period, the temperature is maintained stable by a temperature control system in the kettle jacket to avoid local overheating that could damage the nutritional components of the honey. In S3, cooling is achieved through stainless steel cooling coils installed on the inner wall of the mixing tank. Cooling water at 20°C is circulated to lower the solution temperature to 30°C or 31°C. Analytical grade potassium dihydrogen phosphate is used, magnesium chloride is in anhydrous crystalline form, and yeast extract is food-grade powder. Ammonium chloride can be added directly as an inorganic nitrogen source at a rate of 0.5 g / L or 0.7 g / L. Encapsulated nitrogen sources are added at a ratio equivalent to 0.3 g / L of ammonium chloride. Before adding, the stirrer is turned on at 200 rpm for pre-stirring, then the nutrients are added sequentially, and stirring continues for 10 minutes until completely dissolved. In S4, food-grade anhydrous citric acid is prepared as a 10% aqueous solution and slowly added dropwise along the tank wall using a separatory funnel. The addition is paused after every 50 mL, and the pH is monitored using a pH meter inserted into the middle of the solution until the pH stabilizes at 3.5 or 3.8. In S5, brewing-specific active dry yeast is used at an addition rate of 0.3 g / L or 0.35 g / L. It is activated in a constant-temperature water bath with deionized water at 29°C or 31°C, with gentle stirring every 5 minutes using a glass rod. After 30 minutes, microscopic observation is performed to ensure a yeast cell survival rate of over 90%. In S6, fermentation takes place in a fermenter with an insulation layer. The fermenter is equipped with a platinum resistance temperature sensor, a glass electrode pH sensor, and an online refractometer, collecting data every 15 minutes. A machine learning model is deployed in an industrial control computer, with prediction frequency synchronized with data acquisition. The higher alcohol content threshold is set at 250 mg / L. If the predicted value exceeds this threshold, the temperature is lowered by 2-3°C via the fermenter's electric heating / cooling system, or citric acid is added via an acid metering pump to adjust the pH, with a response time not exceeding 5 minutes.

[0027] In existing technologies, honey wine preparation often involves directly dissolving honey in water and adding a nitrogen source all at once, with fermentation relying solely on manual temperature and pH adjustments. This method frequently results in excessive initial nitrogen supply leading to yeast overgrowth and the production of large amounts of higher alcohols, while insufficient nitrogen in the later stages disrupts yeast metabolism. Incomplete sterilization also easily breeds other microorganisms, and delayed temperature and pH adjustments often lead to large fluctuations in higher alcohol content. This embodiment addresses this issue by supplying nitrogen in stages, with the encapsulated nitrogen source being slowly released throughout fermentation to prevent nitrogen metabolic imbalances. Pasteurization strictly controls temperature and time to ensure a pure fermentation environment. Real-time monitoring combined with model prediction allows for timely adjustment of temperature and pH, keeping fermentation conditions consistently within a suitable range, thus solving the problem of controlling higher alcohols in traditional processes.

[0028] In another embodiment, in S3, the amounts of potassium dihydrogen phosphate, magnesium chloride, yeast extract, and directly added inorganic nitrogen source are 0.5 g / L, 0.2 g / L, 0.15 g / L, and 0.4-0.8 g / L, respectively. The inorganic nitrogen source is ammonium chloride or diammonium hydrogen phosphate. The inorganic nitrogen source encapsulated in the pH-responsive biodegradable capsule contains 0.2-0.4 g / L of directly added inorganic nitrogen source.

[0029] For example, in step S3, potassium dihydrogen phosphate, a white crystalline powder with a purity ≥99%, serves as a phosphorus source to promote yeast nucleic acid synthesis. The added amount is 0.5 g / L, which can be accurately weighed using an electronic balance, dissolved in a beaker containing a small amount of deionized water, and then slowly poured in along the wall of the mixing tank. The inorganic nitrogen source encapsulated in pH-responsive biodegradable capsules contains 0.3 g / L of directly added inorganic nitrogen source. Anhydrous magnesium chloride with a purity ≥98% is selected to provide magnesium ions to the yeast to activate metabolic enzyme activity. The added amount is 0.2 g / L, weighed, and dissolved separately from the potassium dihydrogen phosphate solution to avoid precipitation. Yeast extract, an extract from yeast after cell wall disruption, contains amino acids, B vitamins, etc. A food-grade product with a protein content ≥40% is selected. The added amount is 0.15 g / L. Due to its hygroscopic nature, it needs to be stored in a sealed container and taken with a dry spatula when using. The amount of inorganic nitrogen source added directly can be 0.5 g / L or 0.7 g / L. If ammonium chloride is selected, it must be agricultural grade white crystals with a purity ≥96%. If diammonium hydrogen phosphate is selected, it must be food grade powder with a purity ≥98%. Both nitrogen sources must be filtered through a 100-mesh stainless steel sieve to remove impurities. When adding, first slowly sprinkle the nitrogen source powder onto the surface of the fermentation liquid, then turn on the stirrer and stir at 150 r / min for 5 minutes to ensure complete dissolution before proceeding with subsequent operations.

[0030] In existing technologies, the addition of nutrients during mead fermentation lacks a fixed ratio and is often estimated based on experience. This leads to low yeast activity when phosphorus and magnesium are insufficient, while excessive amounts increase the metabolic burden. The nitrogen source is also limited in type and dosage, easily causing abnormal metabolism in yeast due to nitrogen deficiency or excess. This embodiment specifies the precise dosage of each nutrient, with a phosphorus-magnesium ratio tailored to the yeast's metabolic needs. Two inorganic nitrogen sources can be selected based on the characteristics of the honey matrix. The direct addition amount and the capsule slow-release amount work synergistically to ensure that the yeast receives balanced nutrition at each growth stage, solving the problem of blind nutrient regulation in traditional methods. ### Embodiment of Claim 3

[0031] In another embodiment, the active dry yeast is Ramania D47.

[0032] For example, the active dry yeast used in this embodiment is specifically Ramania D47, which has the characteristics of low production of higher alcohols and high fermentation efficiency, and is purchased from a regular microbial reagent supplier. During activation, yeast is weighed at an addition amount of 0.3 g / L and added to deionized water preheated to 30°C. The mixture is gently stirred with a sterile glass rod until it is uniformly suspended, and then placed in a 30°C constant temperature water bath for 30 minutes for activation. During this period, samples are taken every 10 minutes, and the morphology of yeast cells is observed through an optical microscope to ensure that the cells are plump and undamaged and that the survival rate reaches more than 95% before being inoculated into the fermentation broth.

[0033] In existing technologies, mead fermentation often uses common baker's yeast or brewer's yeast. These strains are prone to over-proliferation in high-sugar environments, producing large amounts of higher alcohols and having low utilization rates of the unique sugars in honey.

[0034] In another embodiment, in S3, the inorganic nitrogen source encapsulated in the pH-responsive biodegradable capsule is prepared by the following method: S31: Sodium alginate is dissolved in deionized water to prepare a sodium alginate solution with a mass-volume concentration of 1-2%; S32: The inorganic nitrogen source powder and the sodium alginate solution are mixed at a mass-volume ratio of 1:10 (g:mL), and thoroughly stirred to obtain a mixed slurry; S33: The mixed slurry is added dropwise using a syringe until the mass-volume concentration reaches 2-3%. In a 0.5% calcium chloride solution, capsules are formed, cross-linked and cured for 10-20 minutes, and then filtered out to obtain calcium alginate gel capsules; S34: The calcium alginate gel capsules are immersed in a 0.5-1.0% (w / v) chitosan acetate solution and allowed to stand for 10-20 minutes to allow chitosan to cross-link with calcium alginate through electrostatic interaction, forming a pH-responsive chitosan-sodium alginate composite membrane on the outer layer of the capsule; S35: The capsules are filtered out, washed with deionized water, and then freeze-dried to obtain the final product.

[0035] For example, in S31, sodium alginate is selected as a low molecular weight food-grade product with a viscosity of 200-300 mPa·s, soluble in deionized water, and the concentration can be selected as 1.5% or 1.8%. It is stirred in a 60℃ constant temperature water bath at 400 r / min for 1 hour until completely dissolved, and undissolved impurities are removed by filtration through a glass funnel. In S32, the inorganic nitrogen source powder is ammonium chloride with a purity ≥96%. It is mixed at a ratio of 1g powder to 10mL sodium alginate solution, poured into a high-speed disperser, and stirred at 2000 r / min for 20 minutes to form a uniform, particle-free slurry, which is then poured into a 50mL beaker for later use. In S33, a 10mL glass syringe with a 0.8mm needle was used. The slurry was placed in the syringe and added dropwise from 10cm above the surface of the calcium chloride solution while stirring with a magnetic stirrer (300r / min). The calcium chloride solution concentration was 2.5% or 2.8%. This formed capsules with a diameter of 2-3mm. After cross-linking and curing at room temperature for 15 minutes, the capsules were separated by filtration using a Buchner funnel and the gel capsules were collected. In S34, chitosan, a food-grade product with a degree of deacetylation ≥90%, was dissolved in a 1% acetic acid aqueous solution to prepare a 0.7% or 0.9% solution. The calcium alginate capsules were poured into the solution, with the liquid level 5cm above the capsules. The mixture was allowed to stand at room temperature for 18 minutes, during which the container was gently shaken to ensure uniform reaction. The positively charged groups of chitosan combined with the negatively charged groups of sodium alginate to form a composite film. In S35, the capsules were washed three times with deionized water, each time soaking for 5 minutes to remove residual solution on the surface. Then, they were placed in a freeze dryer and dried at -40℃ and 0.1MPa for 24 hours to obtain dried pH-responsive capsules, which were then sealed and stored in a desiccator for later use.

[0036] To test the pH response, three 1g capsules were added to 100mL of phosphate buffer solutions at pH 3.5, 4.5, and 5.5, respectively. The solutions were shaken at 150 rpm in a 30℃ constant-temperature shaker. 5mL samples were taken at 2 hours and 24 hours, and the nitrogen content in the solutions was measured using a UV spectrophotometer to calculate the nitrogen source release rate. The results showed that at pH 3.5, the release rate was 45% at 2 hours and 85% at 24 hours; at pH 4.5, the release rate was 30% at 2 hours and 65% at 24 hours; and at pH 5.5, the release rate was 15% at 2 hours and 30% at 24 hours. This indicates that the capsules' release rate significantly increases as the pH decreases, demonstrating a clear pH-responsive characteristic.

[0037] In existing technologies, nitrogen source encapsulation often uses a single sodium alginate capsule, which lacks pH response functionality, resulting in a fixed nitrogen source release rate that cannot adapt to the needs of different fermentation stages. Furthermore, the capsule membrane is prone to premature or delayed rupture, leading to uncontrolled nitrogen source supply. This embodiment constructs a pH-responsive structure using a chitosan-sodium alginate composite membrane, automatically adjusting the release rate when the pH of the fermentation broth changes, thus solving the problem of inaccurate release from traditional capsules.

[0038] In another embodiment, in S6, a training dataset is constructed, which includes multiple feature parameters collected at multiple time points during the historical fermentation process and their corresponding measured values ​​of higher alcohol content; the dataset is trained using a random forest regression algorithm, with higher alcohol content as the prediction target, to obtain a prediction model; during the fermentation process, the feature parameters are monitored in real time by sensors and input into the prediction model, and the predicted value of the current higher alcohol content is output; if the predicted value exceeds the higher alcohol content threshold, the fermentation temperature and / or the pH value of the fermentation broth are dynamically adjusted by adjusting the temperature control device and the acid / alkali addition pump.

[0039] For example, in S6, the training dataset comes from the fermentation records of the past 100 batches. For each batch, characteristic parameters are collected every 2 hours from the start to the end of fermentation, including fermentation broth temperature (measured with a platinum resistance sensor), pH value (glass electrode sensor), dissolved oxygen content (fluorescence dissolved oxygen meter), sugar content (online refractometer), yeast cell density (counted with a hemocytometer), and fermentation time, totaling 2000 sets of data. Each set of data corresponds to the measured value of the total content of higher alcohols (isoamyl alcohol, isobutanol, etc.) measured by gas chromatography. A random forest regression model is constructed using the Scikit-learn library of Python, with 200 decision trees, a maximum depth of 15, and a minimum number of sample splits of 5. The model is trained with higher alcohol content as the target variable, and the model parameters are optimized through 5-fold cross-validation. During fermentation, each sensor is installed at the sampling port in the middle of the fermenter. Data is transmitted to the industrial computer every 10 minutes via the PLC control system, and the model outputs predicted values ​​in real time. The threshold for higher alcohol content is set to 250 mg / L. If the predicted value reaches 260 mg / L, the temperature control device will adjust the cooling water flow rate of the fermenter jacket to reduce the temperature from 30℃ to 28℃; if it reaches 280 mg / L, the acid addition pump will be started at the same time to pump 10% citric acid solution into the tank at a rate of 5 mL / min until the pH drops from 3.8 to 3.5.

[0040] In existing technologies, the control of higher alcohols in mead fermentation relies on manual, periodic sampling and testing. The results are delayed by 2-4 hours, and by the time excessive levels are detected, it is often too late to reverse the process. Furthermore, temperature adjustments are based solely on experience, lacking a systematic parameter linkage mechanism, leading to large fluctuations in higher alcohol content. This embodiment combines real-time sensing with a machine learning model to achieve early prediction and automatic control, solving the problems of delayed detection and blind control inherent in traditional methods.

[0041] In another embodiment, an initial prediction model is trained based on a subset of training data collected during the first 24 hours of fermentation, and the baseline prediction error E of the initial prediction model is calculated. The adjustment range of the number of decision trees is set based on the baseline prediction error E, with a lower limit of 100×(1+E) and an upper limit of 200×(1+E), where E is in g / L. The adjustment range of the maximum depth of the decision trees is set based on the average fluctuation range ΔpH of the pH value of the fermentation broth during the first 24 hours, with a lower limit of 10×(1+ΔpH) and an upper limit of 20×(1+ΔpH). The adjustment range of the minimum number of sample splits is set based on the average fluctuation range ΔT of the temperature of the fermentation broth during the first 24 hours, with a lower limit of 2×(1+ΔT) and an upper limit of 5×(1+ΔT), where ΔT is in °C. Every 24 hours of fermentation time, the model is retrained based on the newly collected fermentation data during that time period, and the adjustment ranges of the number of decision trees, the maximum depth of the decision trees, and the minimum number of sample splits are updated using the newly calculated E, ΔpH, and ΔT values.

[0042] For example, the initial model was trained using 300 sets of data from the first 24 hours. By calculating the difference between the predicted values ​​and the gas chromatographic measured values, the baseline prediction error E was found to be 0.02 g / L. The lower limit for the number of decision trees was 100 × (1 + 0.02) = 102, and the upper limit was 200 × (1 + 0.02) = 204. In actual training, 150 trees were selected. The pH value was recorded hourly during the first 24 hours. The difference between the maximum and minimum values ​​was calculated to be 0.6, and the average fluctuation amplitude ΔpH was 0.3. The lower limit for the maximum depth of the decision tree was 10 × (1 + 0.3) = 13, and the upper limit was 20 × (1 + 0.3) = 26. In practice, it was set to 20. The temperature was recorded hourly, with a maximum fluctuation of 0.8℃ and an average fluctuation amplitude ΔT of 0.5℃. The lower limit for the minimum number of sample splits was 2 × (1 + 0.5) = 3, and the upper limit was 5 × (1 + 0.5) = 7.5. 7 was selected, and in practice, it was set to 5. When fermentation reaches 48 hours, retrain using new data from 24-48 hours. Calculate E as 0.015 g / L, ΔpH as 0.2, and ΔT as 0.4℃. Then update the number of decision trees to 101.5-203, the maximum depth to 12-24, and the minimum number of sample splits to 2.8-6. After rounding to the nearest integer, adjust the model parameters and retrain.

[0043] In existing technologies, prediction model parameters are fixed once set and remain so throughout the fermentation process. However, yeast activity and substrate composition change continuously during fermentation, leading to a decline in prediction accuracy in later stages and an inability to adapt to the dynamic fermentation environment. This embodiment dynamically adjusts the model's structural parameters based on real-time data, ensuring the model always matches the fermentation state. This solves the problem of poor adaptability in traditional fixed models and improves prediction stability.

[0044] In another embodiment, during each model training or retraining, the Pearson correlation coefficient between each candidate feature parameter in the current training dataset and the measured value of higher alcohol content is calculated; the candidate feature parameters include at least fermentation broth temperature, pH value, dissolved oxygen content, sugar content, yeast cell density, and fermentation time; feature parameters with absolute Pearson correlation coefficient values ​​greater than the coefficient threshold are selected as valid input variables for this training and used to train the random forest regression model.

[0045] For example, during each training iteration, all candidate feature parameters were extracted from the dataset. These parameters included a fermentation broth temperature range of 25-35℃, pH value of 3-5, dissolved oxygen content of 0.5-5 mg / L, sugar content of 5-25 Brix, yeast cell density of 10^6-10^8 cells / mL, fermentation time of 0-120 hours, and the higher alcohol content measured by gas chromatography for each parameter. Pearson correlation coefficients between each parameter and the higher alcohol content were calculated using SPSS statistical software, with a threshold of 0.6. The calculated correlation coefficients were 0.72 (positive correlation), -0.65 (negative correlation) for temperature, -0.55 for pH, -0.68 for dissolved oxygen, -0.68 for sugar content, 0.63 for yeast cell density, and 0.59 for fermentation time. Temperature, pH, sugar content, and yeast cell density were selected as valid input variables because their absolute values ​​were all greater than 0.6; dissolved oxygen and fermentation time were excluded because their coefficients did not meet the threshold. The selected variables are used to construct the input matrix, reducing interference from irrelevant parameters and improving model training efficiency.

[0046] In existing technologies, model input variables often include all collectable parameters without distinguishing the strength of correlation, leading to excessive model complexity, long training times, and the potential introduction of noise by irrelevant parameters, thus reducing prediction accuracy. This embodiment uses the Pearson coefficient to select core parameters, focusing on factors that significantly influence higher alcohols, thus solving the problem of input redundancy in traditional models and improving training efficiency and prediction accuracy.

[0047] In another embodiment, if the predicted higher alcohol content exceeds the higher alcohol content threshold but is below the first limit, the fermentation temperature is reduced; if the predicted higher alcohol content reaches or exceeds the first limit, the fermentation temperature and the pH of the fermentation broth are reduced simultaneously.

[0048] For example, the higher alcohol content threshold is set to 250 mg / L, and the first limit is set to 300 mg / L. When the model predicts a value of 270 mg / L (exceeding the threshold but below the limit), the fermenter's PLC control system sends a command to the temperature control module to adjust the opening of the jacket cooling water valve, reducing the fermentation broth temperature from the current 30°C to 28°C at a rate of 0.5°C / hour. The temperature is recorded every 30 minutes during the cooling process to ensure stable adjustment. When the predicted value is 310 mg / L (reaching the first limit), the system simultaneously activates the temperature control and pH adjustment modules, reducing the temperature to 27°C. At the same time, the acid addition pump pumps 10% citric acid solution into the fermentation broth at a rate of 8 mL / min. Through real-time feedback from the pH sensor, the pH value is reduced from the current 3.7 to 3.4 at a rate of 0.1 units / 10 minutes. After adjustment, monitoring continues for 2 hours, and control is stopped only after confirming that the predicted higher alcohol value has fallen below the threshold.

[0049] In existing technologies, addressing excessive levels of higher alcohols relies solely on cooling measures, resulting in a fixed control intensity that cannot be flexibly adjusted according to the degree of exceedance. Mild exceedances can lead to over-control, prolonging the fermentation cycle, while severe exceedances are difficult to control effectively due to insufficient intensity. This embodiment employs single-factor or dual-factor control based on the degree of exceedance, precisely matching the control intensity and overcoming the rigidity of traditional control methods, thus improving control effectiveness.

[0050] The following is a description of a specific embodiment.

[0051] Example 1: Complete preparation process of low-high alcohol content mead

[0052] Preparation of honey dilution: Select nectar and dissolve it in deionized water at an addition rate of 320 g / L. Stir at 300 r / min for 20 minutes in a stainless steel mixing tank with a paddle stirrer until the honey is completely dissolved.

[0053] Pasteurization: Transfer the diluted honey solution into an electrically heated jacketed kettle, heat it to 75°C and maintain it for 30 minutes. During this time, the temperature is kept stable by the jacket temperature control system to avoid local overheating.

[0054] Preparation of honey fermentation liquid: After sterilization, 20°C cooling water is introduced through the cooling coils on the tank wall to cool the diluted liquid to 30°C;

[0055] Add the following nutrients in sequence: potassium dihydrogen phosphate (0.5 g / L, analytical grade), magnesium chloride (0.2 g / L, anhydrous crystals), yeast extract (0.15 g / L, food grade, protein content ≥40%), and directly added ammonium chloride (0.5 g / L, agricultural grade, filtered through a 100-mesh sieve). Turn on the stirrer and stir at 200 r / min for 10 minutes.

[0056] Preparation method of pH-responsive capsule encapsulating ammonium chloride (equivalent to 0.3 g / L ammonium chloride): 1.5% sodium alginate solution and ammonium chloride are mixed at a ratio of 1:10, crosslinked with 2.5% calcium chloride for 15 minutes, reacted with 0.7% chitosan acetate solution for 18 minutes, freeze-dried at -40℃ for 24 hours, and stirred for another 5 minutes until uniformly mixed.

[0057] Adjusting pH: Slowly add a 10% food-grade citric acid aqueous solution to the fermentation broth through a separatory funnel, pausing after every 50 mL of addition and monitoring with a pH meter inserted into the broth until the pH stabilizes at 3.5.

[0058] Activating yeast: Take Raman yeast D47 (vacuum packaged in aluminum foil, refrigerated at 4℃), add it to 30℃ deionized water at a rate of 0.3g / L, and let it stand in a constant temperature water bath for 30 minutes to activate it, stirring once every 5 minutes during the process. Observe the cell viability under a microscope and the cell viability is ≥95%.

[0059] Fermentation and Control: Activated yeast broth was inoculated into the fermentation broth and fermented in a fermenter with an insulation layer. A platinum resistance temperature sensor, a glass electrode pH sensor, an online refractometer, and a fluorescence dissolved oxygen meter were installed in the tank, and data were collected every 15 minutes. A machine learning model was deployed on an industrial control computer and trained on a random forest regression model (initial decision trees: 150, maximum depth: 20, minimum number of sample splits: 5) using 100 historical batches of data (including temperature, pH, dissolved oxygen, sugar content, yeast density, and fermentation time). The higher alcohol threshold was set at 100 mg / L, and the first limit was 120 mg / L.

[0060] Before each training session, the Pearson coefficient (threshold 0.6) was calculated, and temperature, pH, sugar content, and yeast density were selected as valid inputs. The model was retrained every 24 hours with new data, and the model parameters were adjusted based on the baseline error E (0.02 g / L), pH fluctuation ΔpH (0.3), and temperature fluctuation ΔT (0.5℃).

[0061] If the predicted concentration of higher alcohols is 110 mg / L (exceeding the threshold but not reaching the limit), the temperature is lowered from 30°C to 28°C using jacket cooling water; if it reaches 130 mg / L (exceeding the limit), the temperature is simultaneously lowered to 27°C and citric acid is added to adjust the pH to 3.4, with a control response time of ≤5 minutes.

[0062] The higher alcohol content, alcohol content, total acid content, residual sugar content, total flavonoid content, and transmittance of the mead obtained in Example 1 were determined. Alcohol content: determined according to GB 5009.225-2023 (alcohol meter method); Total acid (calculated as citric acid): determined according to GB12456-2021 (acid-base indicator titration method); Residual sugar: determined using a saccharimeter method; Total flavonoids: determined according to GB / T 20574-2006 (spectrophotometric colorimetric method); Transmittance: after centrifugation at 8000 r / min for 5 min, 2 mL of the mead was measured at 680 nm. Higher alcohol content: determined by gas chromatography.

[0063] Testing revealed that the mead in this embodiment is clear and transparent in color, has a harmonious and pleasantly sweet and sour taste, a rich honey aroma, and no off-odors, exhibiting typical mead characteristics. Specific results are shown in Tables 1 and 2 below (the physicochemical indicators of different meads were obtained by replacing honey from different sources).

[0064] Example 2: The ammonium chloride in Example 1 was replaced with diammonium hydrogen phosphate.

[0065] Table 1 Physicochemical properties of the honey wine obtained in Example 1

[0066]

[0067] Table 2 Physicochemical properties of the honey wine obtained in Example 2

[0068]

[0069] Higher alcohols, as potentially harmful byproducts in fermented beverages, are crucial to consumer safety and health, and controlling their concentration within a reasonable range is extremely important. Factors influencing the formation of higher alcohols include yeast strain, assimilable nitrogen, and fermentation process conditions. Therefore, this application systematically explores the appropriate concentrations of each influencing factor from aspects such as raw material selection, strain breeding, and fermentation control, aiming to reasonably control the higher alcohol content in mead at a low level to reduce its risk.

[0070] One influencing factor is brewing yeast. Higher alcohols are typically produced by yeast via the Ehrlich and Harris pathways. Different yeasts have varying abilities to adapt to the fermentation environment and utilize nutrients, resulting in differences in the production of related metabolites. This invention investigated the effects of six commercially available yeasts (A, B, D, M, and V) on the higher alcohol content of mead. The results showed that yeast D produced the lowest total higher alcohol content, at 218.34 ± 1.23 mg / L. Therefore, yeast D was selected as the yeast used in this invention. (See [link to relevant documentation]). Figure 2 , Figure 2 The letters (a, b, c, d) are significance markers, used to indicate the statistical significance of differences in higher alcohol content at different fermentation temperatures. Figures 3-6 Similarly). Among them, yeast A is Aroma White, B is Angel Wine Active Dry Yeast BV818, D is Raman Yeast D47, E is Raman Yeast EC1118, M is Mangrove Jack Yeast M05, and V is Vintage White.

[0071] Factor Two – Fermentation Temperature. Both excessively high and low fermentation temperatures can cause temperature stress, affecting yeast growth and reproduction, as well as their alcohol production capacity and efficiency. This invention investigated the effect of fermentation temperature on the higher alcohol content of mead. The results showed that as the fermentation temperature increased, the higher alcohol content initially increased and then gradually decreased. Furthermore, to achieve the target alcohol content, this invention selected 28℃ as the initial fermentation temperature for the mead. (See [link to relevant documentation]). Figure 3 .

[0072] Factor Three – Initial pH Value. The pH value of the honey fermentation broth is related to its buffering capacity and microbial stability, and also affects the sweet and sour balance of mead. This invention uses citric acid as an acidity regulator to investigate the effect of initial pH value on the higher alcohol content of mead. The results show that maintaining the initial pH value of the fermentation broth in a slightly acidic environment (pH=3.0-4.0) is beneficial for the formation of fewer higher alcohols, and also for the fermentation stability and taste of the mead. See [link to relevant documentation]. Figure 4 .

[0073] Factor Four – Honey Addition Amount. The amount of honey added represents the initial sugar content of the fermentation broth. Too high an initial sugar content results in a high osmotic pressure, which is detrimental to yeast reproduction and growth. Conversely, too low an initial sugar content hinders fermentation and can lead to a bland taste. This invention uses vitex honey as the research object to investigate the effect of honey addition amount on the higher alcohol content of mead. The results show that with increasing honey addition amount, the higher alcohol content initially increases and then gradually decreases. Furthermore, to achieve the target alcohol content and avoid excessive sweetness in the mead, a honey addition amount of 300-350 g / L is preferable. However, mead in this concentration range has a relatively high higher alcohol content, requiring consideration of other factors to reduce its concentration. See [link to relevant documentation]. Figure 5 .

[0074] Factor Five – Yeast Addition Amount. The amount of yeast added typically affects the fermentation time and rate of fermented wines. Insufficient yeast addition makes it difficult for a dominant population to form, leading to problems such as incomplete fermentation or premature termination of fermentation. This invention investigated the effect of yeast addition amount on the content of mead. The results showed that, under sufficient fermentation time, the content of higher alcohols produced by yeast at different mass concentrations remained relatively stable. Considering production costs and scale of production, a yeast addition amount of 0.2–0.4 g / L is recommended. (See [link to relevant documentation]). Figure 6 .

[0075] Comparative Example 1: All inorganic nitrogen sources were added directly (without capsules), and the other steps were the same as in Example 1 (including machine learning regulation).

[0076] Comparative Example 2: A portion of the inorganic nitrogen source was encapsulated using ordinary calcium alginate capsules (without pH-responsive chitosan-sodium alginate composite membrane), and the other steps were the same as in Example 1 (including machine learning regulation).

[0077] Comparative Example 3: pH-responsive biodegradable capsules were used, but the fermentation process was not regulated by machine learning models (only the initial temperature was fixed at 28-32℃ and pH at 3-4), and the other steps were the same as in Example 1.

[0078] Control group: Traditional process, inorganic nitrogen source is added directly at once, without encapsulation, fermentation process is not dynamically controlled (temperature and pH are measured manually once a day, without real-time adjustment), other basic steps are operated in accordance with conventional procedures.

[0079] Table 3

[0080]

[0081] Taking acacia honey wine as an example, the honey wines obtained from Example 1, Comparative Example 1, Comparative Example 2, Comparative Example 3 and the control group were tested, and the results are shown in Table 3.

[0082] Total higher alcohol content:

[0083] Example 1 (134.81 mg / L) was significantly lower than all comparative and control examples. The core reason for this was the synergistic effect of pH-responsive biodegradable capsules and a machine learning model, combined with pasteurization and appropriate temperature and pH regulation to form a multi-stage alcohol reduction system. The pH-responsive capsules achieve dynamic nitrogen supply through their special structure: in the early stage of fermentation, the pH of the fermentation broth is relatively high (alkaline environment), the outer layer of the capsule is stable, and nitrogen is released slowly, avoiding the initial nitrogen excess caused by the one-time nitrogen addition in traditional processes, preventing the yeast from growing explosively due to nutrient overload, and reducing the production of higher alcohols by abnormal metabolism; as fermentation progresses, the yeast metabolism produces organic acids, causing the pH to drop (acidic environment), the outer layer of the capsule changes structure, accelerates nitrogen release, and timely replenishes nutrients, avoiding the yeast metabolic disorder caused by insufficient nitrogen source in the later stage, and further controlling alcohol production. The comparative example 1 (189.56 mg / L) without capsules and the control group (225.47 mg / L) had significantly higher levels of higher alcohols due to nitrogen source imbalance, which confirms the importance of dynamic nitrogen release. The comparative example 3 (156.72 mg / L) without machine learning regulation lacked real-time temperature and pH intervention and could not correct metabolic deviations in time, so its level was higher than that of example 1, demonstrating the alcohol-inhibiting effect of real-time regulation.

[0084] Alcohol content:

[0085] Example 1 showed the highest alcohol content (9.00% v / v), while the control group showed the lowest (8.20% v / v), indicating that the technical solution in Example 1 effectively ensures yeast fermentation efficiency. The constant nitrogen supply from the pH-responsive capsules keeps the yeast in a stable metabolic state, improving the conversion efficiency of sugar to alcohol. Pasteurization (70℃, 20 min) thoroughly kills unwanted bacteria in the diluted honey solution, preventing them from competing for nutrients and interfering with fermentation. The activation and fermentation temperature of 28-32℃ matches the yeast's activity requirements (the optimal fermentation temperature range for yeast), reducing metabolic blockage. The control group, lacking dynamic nitrogen supply and real-time environmental control, was prone to nitrogen imbalance or interference from unwanted bacteria, resulting in incomplete yeast fermentation and a lower alcohol content.

[0086] Total acid and residual sugar:

[0087] The total acidity differences among the groups were small, but Example 1 had the lowest residual sugar (15.25 Brix%), while the control group had the highest (17.30 Brix%), further validating the promoting effect of precise regulation on yeast sugar metabolism. The pH-responsive capsules dynamically released nitrogen, providing stable nutritional support to the yeast throughout the process, maintaining its strong sugar-decomposing ability. The machine learning model controlled temperature and pH in real time, avoiding environmental fluctuations that inhibited yeast activity (yeast's optimal pH is 4-6), thus reducing sugar residue. In the control group, the one-time addition of nitrogen caused a supply-demand imbalance, and the lack of real-time environmental intervention resulted in low yeast sugar metabolism efficiency and ultimately, a higher accumulation of residual sugar.

[0088] Total flavonoids and light transmittance:

[0089] Example 1 showed the highest total flavonoids (34.33 mg / L) and transmittance (83.93%) compared to the control group, demonstrating the synergistic protection of wine quality through multi-stage regulation. pH-responsive capsules combined with pasteurization thoroughly eliminated unwanted microorganisms, preventing them from consuming total flavonoids and other active ingredients or producing turbidity. A machine learning model stabilized fermentation temperature and pH in real time, reducing abnormal yeast metabolism caused by environmental fluctuations and lowering the risk of total flavonoid degradation and wine turbidity. Appropriate process parameters further protected the active ingredients and clarity of the wine. In contrast, the control group, lacking thorough microbial control and environmental stabilization measures, experienced easier consumption of total flavonoids by unwanted microorganisms, and the fermentation process was prone to producing turbidity due to abnormal metabolism, resulting in significantly lower quality than Example 1.

[0090] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A method for preparing mead with low higher alcohol content, characterized in that, include: S1: Dissolve honey in water at an addition rate of 300-350 g / L to prepare a diluted honey solution; S2: Pasteurize the diluted honey solution by heating it at 70-80℃; S3: Cool the sterilized honey dilution to 28-32 ℃, add potassium dihydrogen phosphate, magnesium chloride, yeast extract and inorganic nitrogen source, mix well to obtain honey fermentation liquid. Inorganic nitrogen source includes directly added inorganic nitrogen source and inorganic nitrogen source encapsulated in pH-responsive biodegradable capsules. S4: Use citric acid to adjust the pH of the honey fermentation liquid to 3-4; S5: Activate the active dry yeast in water at 28-32 ℃ at an addition rate of 0.2-0.4 g / L to obtain activated yeast liquid; S6: Inoculate the activated yeast liquid into the honey fermentation liquid for fermentation. During the fermentation process, the higher alcohol content of the fermentation liquid is predicted in real time using a machine learning model, and the fermentation temperature and pH of the fermentation liquid are dynamically adjusted based on the prediction results. The active dry yeast is Ramania D47; In S3, the inorganic nitrogen source encapsulated in a pH-responsive biodegradable capsule is prepared by the following method: S31: Dissolve sodium alginate in deionized water to prepare a sodium alginate solution with a mass-volume concentration of 1-2%; S32: Mix the inorganic nitrogen source powder with the sodium alginate solution at a mass-volume ratio of 1:10 (g:mL), stir and disperse thoroughly to obtain a mixed slurry; S33: Using a syringe, the mixed slurry is added dropwise to a calcium chloride solution with a mass-volume concentration of 2-3% to form capsules. After cross-linking and curing for 10-20 minutes, the capsules are filtered out to obtain calcium alginate gel capsules. S34: Immerse calcium alginate gel capsules in a chitosan acetate solution with a mass-volume concentration of 0.5-1.0% and let them stand for 10-20 minutes to allow chitosan to crosslink with calcium alginate through electrostatic interaction, forming a pH-responsive chitosan-sodium alginate composite film on the outer layer of the capsule. S35: Filter out the capsules, wash with deionized water, and freeze-dry to obtain the final product.

2. The method for preparing low-higher alcohol content mead as described in claim 1, characterized in that, In S3, the amounts of potassium dihydrogen phosphate, magnesium chloride, yeast extract, and directly added inorganic nitrogen source are 0.5 g / L, 0.2 g / L, 0.15 g / L, and 0.4-0.8 g / L, respectively. The inorganic nitrogen source is ammonium chloride or diammonium hydrogen phosphate. The inorganic nitrogen source encapsulated in the pH-responsive biodegradable capsule contains 0.2-0.4 g / L of inorganic nitrogen source.

3. The method for preparing mead with low higher alcohol content as described in claim 1, characterized in that, In S6, a training dataset is constructed, which contains multiple feature parameters collected at multiple time points during the historical fermentation process and their corresponding measured values ​​of higher alcohol content. The dataset was trained using a random forest regression algorithm, with higher alcohol content as the prediction target, to obtain a prediction model. During fermentation, sensors monitor the process in real time and input characteristic parameters into the prediction model, outputting a predicted value of the current higher alcohol content. If the predicted value exceeds the higher alcohol content threshold, the fermentation temperature and / or the pH of the fermentation broth are dynamically adjusted by regulating the temperature control device and the acid / alkali addition pump.

4. The method for preparing low-higher alcohol content mead as described in claim 3, characterized in that, An initial prediction model was obtained by training a subset of training data collected during the first 24 hours of fermentation, and the baseline prediction error E of the initial prediction model was calculated. The adjustment range of the number of decision trees is set based on the baseline prediction error E, with a lower limit of 100×(1+E) and an upper limit of 200×(1+E), where the unit of E is g / L; The adjustment range of the maximum depth of the decision tree is set based on the average fluctuation range ΔpH of the pH value of the fermentation broth in the previous 24 hours, with a lower limit of 10×(1+ΔpH) and an upper limit of 20×(1+ΔpH). The adjustment range of the minimum sample division number is set based on the average fluctuation range ΔT of the fermentation broth temperature in the previous 24 hours, with a lower limit of 2×(1+ΔT) and an upper limit of 5×(1+ΔT), where the unit of ΔT is ℃; Every 24 hours of fermentation time, the model is retrained based on the newly collected fermentation data during that period, and the number of decision trees, the maximum depth of decision trees, and the minimum number of sample splits are updated using the newly calculated E, ΔpH, and ΔT values.

5. The method for preparing mead with low higher alcohol content as described in claim 4, characterized in that, During each model training or retraining session, the Pearson correlation coefficient between each candidate feature parameter and the measured value of higher alcohol content in the current training dataset is calculated. Candidate characteristic parameters include at least fermentation broth temperature, pH value, dissolved oxygen content, sugar content, yeast cell density, and fermentation time; Feature parameters whose absolute Pearson correlation coefficient is greater than the coefficient threshold are selected as valid input variables for this training and used to train the random forest regression model.

6. The method for preparing mead with low higher alcohol content as described in claim 1, characterized in that, If the predicted higher alcohol content exceeds the higher alcohol content threshold but is below the first limit, then reduce the fermentation temperature. If the predicted higher alcohol content reaches or exceeds the first limit, then the fermentation temperature and the pH of the fermentation broth should be reduced simultaneously.

Images