Special fat for plant milk and preparation method thereof based on machine learning

Abstract

The invention discloses special fat for vegetable milk and a preparation method thereof, and the special fat for vegetable milk is prepared by mixing three or four kinds of vegetable fat according to a certain proportion and carrying out transesterification. According to the fat special for the plant milk, the appearance, smell, taste, flavor, texture and other expressions of the plant milk are improved, meanwhile, the prepared plant milk has good oxidation resistance, and the SFC curve of the plant milk is close to that of cow milk, so that the plant milk can be comparable with the cow milk in the aspects of taste, flavor and nutritional characteristics, and the fat special for the plant milk has good market prospects. And the possibility of replacing cow milk with plant milk is increased. The invention also discloses a preparation method of the special fat for the plant milk based on machine learning, which comprises the following steps: screening out a plurality of groups of representative process condition parameter combinations of the grease composition and the transesterification reaction in combination with Latin hypercube sampling, and screening out an optimal grease composition and an optimal transesterification reaction condition in combination with a neural network model; the method is high in screening accuracy, and the research and development cost is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of food processing, and in particular to a special fat for plant milk and a preparation method of the special fat for plant milk based on machine learning. Background Art

[0002] As people's attention to health and environmental protection continues to increase, plant milk (plant-based dairy products) as a substitute for traditional cow's milk has a growing market demand. Plant milk is mainly based on plant raw materials such as soybeans, almonds, coconut meat, oats, walnuts, etc., and is prepared by soaking, grinding, enzymatic hydrolysis, filtration, homogenization and blending. At present, the preparation method of plant milk generally includes the following steps: first, the plant raw materials are pretreated (such as peeling and deodorizing), then the fat, protein and other components in the plant tissue are released by physical crushing or enzymatic hydrolysis technology, and then the residue is removed by centrifugation or filtration. Finally, the sensory properties of the product are improved by adding emulsifiers (such as lecithin), stabilizers (such as carrageenan) and flavor modifiers (such as sweeteners). Although this method can prepare plant milk products with good flavor and mouthfeel, due to the significant difference in the fat composition of the prepared plant milk and cow's milk fat, plant milk is difficult to compete with cow's milk in terms of taste, flavor and nutritional properties, which restricts its market penetration.

[0003] Specifically, the differences between existing plant-based milk and cow's milk are mainly reflected in two aspects. First, the fatty acid composition of cow's milk fat is unique, with a balanced ratio of saturated fatty acids to unsaturated fatty acids, and contains short-chain fatty acids and medium-chain fatty acids. Its unique fatty acid composition gives cow's milk a smooth taste and easy digestibility. The fatty acid composition contained in existing plant-based milk is relatively simple, which is quite different from the fatty acid composition in cow's milk fat. On the other hand, plant fat is easily affected by oxidation or enzymatic reactions during processing, producing unpleasant flavors such as beany and grassy flavors, which need to be masked by the addition of additional flavors. At the same time, the melting point and crystallization behavior of plant fat (reflected in the SFC curve) are different from those of cow's milk fat, resulting in poor melting characteristics of plant milk fat in the mouth, which leads to a significant difference in taste between plant-based milk and cow's milk. Although fat composition can be adjusted by physically mixing various vegetable oils or using transesterification, there are currently few studies on the use of multiple oil blends or transesterification methods to prepare fat substitutes with a flavor and taste similar to cow's milk fat. Furthermore, different vegetable oils have different properties, and the fatty acid composition of cow's milk fat is relatively complex, making it difficult to prepare a fat substitute that closely resembles cow's milk fat by simply screening oils for blending or using transesterification methods. Therefore, it is particularly important to develop a compound oil specifically for vegetable milk fat and a method for preparing vegetable milk fat. Summary of the Invention

[0004] [Technical Issues]

[0005] The fatty acid composition and SFC curve of existing plant milk and cow's milk are quite different, which leads to a significant difference in the flavor and taste of plant milk and cow's milk; there are many types of existing oils and fats, and the fatty acid composition of different oils and fats is different. There are many factors affecting the transesterification reaction, making it difficult to prepare a fat substitute that is close to cow's milk fat by simply screening oils for compounding or based on the transesterification method.

[0006] [Technical solution]

[0007] The purpose of the present invention is to provide a method for preparing special fat for plant milk, by selecting different oils for compounding, and combining the transesterification reaction to prepare special fat for plant milk with a flavor and taste and an SFC curve similar to that of cow's milk fat. The sensory score of the plant milk prepared by the compounded oil and transesterification method of the present invention is high, and its SFC curve is also close to the SFC curve of cow's milk. At the same time, the plant milk prepared by the present invention has good antioxidant properties, which helps to improve the stability and nutritional value of the plant milk, making it closer to cow's milk fat in terms of taste, flavor and health, thereby improving consumer acceptance. In addition, the present invention also proposes a method for preparing special fat for plant milk by using a neural network to screen the compounding ratio of vegetable oils and the transesterification reaction parameters. The method of the present invention can quickly screen out the plant oil composition and transesterification reaction parameters that can prepare fat with a flavor and taste close to cow's milk fat, greatly reducing the number of experiments and reducing research and development costs.

[0008] To achieve the above-mentioned purpose, the present invention first provides a fat composition for preparing special fat for plant milk, wherein the fat composition comprises ashwagandha oil, camelina oil and coconut oil, and the mass ratio of ashwagandha oil, camelina oil and coconut oil is 2.9-3.1:1.8-2.2:4.7-5.3.

[0009] The present invention also provides a second oil composition for preparing special fat for plant milk, the oil composition comprising palm oil, camelina oil and olive oil, wherein the mass ratio of palm oil, camelina oil and olive oil is 3.8-4.2:2.9-3.1:2.9-3.1.

[0010] The present invention also provides a third oil composition for preparing special fat for plant milk, the oil composition comprising ashwagandha oil, camelina oil, coconut oil and palm oil, the mass ratio of ashwagandha oil, camelina oil, coconut oil and palm oil being 3.8-4.2:1.9-2.1:2.9-3.1:0.9-1.1.

[0011] The present invention adopts a chemical analysis method to determine the composition ratio of specific fatty acids in milk fat. According to the ratio of fatty acids in milk fat obtained by analysis, the fatty acids in milk fat are selected from tung oil (saturated fatty acids SFA 7% to 10%, monounsaturated fatty acids MUFA 14% to 20%, polyunsaturated fatty acids 70% to 80%), camelina oil (saturated fatty acids SFA 8% to 10%, monounsaturated fatty acids MUFA 15% to 20%, polyunsaturated fatty acids 60% to 75%), olive oil (saturated fatty acids SFA The present invention relates to a fat composition comprising: a plant milk raw material, a vegetable milk raw material, and a vegetable milk raw material. The fat composition comprises: a plant milk raw material, a vegetable ...

[0012] The present invention also discloses a method for preparing special fat for plant milk, comprising the following steps:

[0013] (1) mixing the oil composition for preparing the plant milk fat according to a certain proportion to obtain a mixed oil;

[0014] (2) subjecting the mixed oil to transesterification reaction in the presence of a catalyst;

[0015] (3) separating and purifying the product after the transesterification reaction in step (2), removing the catalyst, and drying to obtain the plant milk-specific fat.

[0016] In one embodiment of the present invention, the temperature of the transesterification reaction in step (2) is 40-60° C., and the reaction time is 0.25-1.25 h.

[0017] In one embodiment of the present invention, the pressure of the transesterification reaction in step (2) is 0.1 to 0.15 MPa.

[0018] In one embodiment of the present invention, the catalyst in step (2) comprises at least one of citric acid, Novozymes Lipozyme lipase, and tartaric acid, preferably citric acid.

[0019] In one embodiment of the present invention, in step (3), the specific steps of separation are: selecting n-hexane as the solvent, extracting at room temperature (25°C) in a ratio of 1:2 between the volume of the solvent and the volume of the transesterified product, extracting three times, each extraction time being 30 minutes, and centrifuging (3000 rpm, for 10 minutes) to obtain the upper organic phase.

[0020] In one embodiment of the present invention, in step (3), column chromatography is used to purify the product after extraction and centrifugation.

[0021] In one embodiment of the present invention, in step (3), the specific steps of the purification are: eluting in a silica gel column using a mixed solvent of n-hexane and ethyl acetate (the solvent ratio is 80:20), and the elution rate is controlled at 0.5 mL / min.

[0022] In one embodiment of the present invention, in step (3), thin layer chromatography (TLC) is further used for separation and purification. When performing thin layer chromatography, the stationary phase selected is silica gel, the solvent system is n-hexane / ethyl acetate = 85:15 (v / v), the stationary phase is an ordinary silica gel plate, and the development distance is 7 to 10 cm.

[0023] In one embodiment of the present invention, in step (3), for solid catalysts, the catalyst is removed by filtering using filter paper (glass fiber filter paper with a pore size of 0.45 μm); for liquid catalysts, the catalyst is removed by inactivating the enzyme by adjusting the pH value to 5.0-6.0 or increasing the reaction temperature to 70°C.

[0024] In one embodiment of the present invention, in step (3), the drying is one of vacuum drying, spray drying, freeze drying or nitrogen drying.

[0025] In one embodiment of the present invention, the vacuum degree during vacuum drying is -0.08 MPa, the drying temperature is 45-50° C., and the drying time is 24 h.

[0026] In one embodiment of the present invention, the temperature during spray drying is 120° C., the feed flow rate is 5 to 10 mL / min, and the air flow velocity is controlled at 30 m / s.

[0027] In one embodiment of the present invention, the freeze-drying temperature is -40°C, the pressure is 0.03 MPa, and the time is 48 hours.

[0028] In one embodiment of the present invention, the nitrogen drying refers to drying the product under a nitrogen flow at a nitrogen flow rate of 0.5 L / min at a temperature of 50° C. for 2 h.

[0029] The invention also discloses a special fat for plant milk prepared according to the method.

[0030] The present invention also discloses an application of the above-mentioned special fat for plant milk in plant milk, wherein the application is to add the special fat for plant milk to plant milk without adding plant oil, and the addition amount of the special fat for plant milk accounts for 0.5-1.0% of the total mass of the finally prepared plant milk. The plant milk without adding plant oil is prepared according to a conventional method.

[0031] Among them, one implementation method of preparing vegetable milk without adding vegetable oil by conventional method includes the following steps:

[0032] S1. The plant raw material is crushed into 50 μm using a jet mill, and then spray-dried. The inlet air temperature is set to 170°C, the outlet air temperature is 80°C, and the spray pressure is controlled at 150 bar to obtain plant dry powder;

[0033] S2. The plant dry powder obtained in step S1 was mixed with water in a ratio of 1:8 (w / v), and stirred at a speed of 2000 rpm for 20 minutes at 50° C. using an ultrasonic stirrer to fully disperse it, and then subjected to high-pressure homogenization treatment at a pressure of 150 MPa, and homogenized twice to improve the stability of the emulsion;

[0034] S3. The emulsion homogenized in step S2 needs to be pasteurized at 80°C for 25 seconds to ensure microbial safety;

[0035] S4. Add 0.2% xanthan gum, 3% sucrose and 0.3% lecithin, accounting for the total mass of the plant milk, to the emulsion sterilized in step S3, and stir to obtain the plant milk without adding vegetable oil.

[0036] In one embodiment of the present invention, after adding the special fat for plant milk, it is necessary to stir to obtain the plant milk, with a stirring speed of 1500 rpm and a stirring time of 10 minutes.

[0037] The invention also discloses the plant milk prepared according to the method.

[0038] The present invention also discloses a method for preparing plant milk-specific fat based on machine learning, comprising the following steps:

[0039] (1) Select several groups of different oils and fats and use Latin hypercube sampling to screen oil compositions with different proportions;

[0040] (2) mixing the oil composition screened out in step (1), performing an ester exchange reaction, separating, purifying and drying to obtain a plant-based fat, and adding the obtained plant-based fat to the raw materials of the plant milk to prepare the plant milk;

[0041] (3) performing sensory evaluation on the plant milk obtained in step (2) to obtain sensory scores of the plant milk based on different oil compositions;

[0042] (4) using a neural network model to predict the optimal ratio of the oil composition, wherein the input feature of the neural network model is the ratio of each oil in the oil composition in step (2), and the output feature is the sensory score of the plant milk corresponding to the oil composition;

[0043] (5) using Latin hypercube sampling to screen several sets of process condition combinations for transesterification reactions, wherein the process parameters include temperature, pressure, catalyst, and reaction time;

[0044] (6) using the oil composition with the optimal ratio obtained in step (4) as a raw material, performing an ester exchange reaction according to the process condition combination screened in step (5) and preparing the plant milk according to the method of step (2);

[0045] (7) performing a sensory evaluation on the plant milk obtained in step (6), measuring the antioxidant properties and SFC curve of the plant milk-specific fat, and obtaining the sensory scores, antioxidant properties, and SFC curves of the plant milk based on different process condition combinations;

[0046] (8) using a neural network model to predict the optimal process condition combination for the transesterification reaction, wherein the input features of the neural network model are the process condition combination based on Latin hypercube screening in step (5), and the output features are the sensory scores of the plant milk, the antioxidant properties and SFC curves of the fat, and the SFC curves of the milk fat corresponding to different process condition combinations;

[0047] (9) performing an ester exchange reaction based on the optimal ratio of the oil and fat composition selected in step (4) in combination with the optimal temperature, pressure, catalyst and reaction time obtained in step (8), and separating, purifying and drying to obtain the plant milk-specific fat.

[0048] In one embodiment of the present invention, the oil in step (1) comprises three or four of tung oil, camelina oil, olive oil, coconut oil, and palm oil.

[0049] In one embodiment of the present invention, a constraint condition is imposed when performing Latin hypercube sampling so that the sum of the proportions of the selected oils and fats is 100%.

[0050] In one embodiment of the present invention, the temperature of the transesterification reaction in step (2) is 40-60° C., the reaction time is 0.25-1.25 h, the pressure is 0.1-0.15 MPa, and the catalyst comprises at least one of citric acid, Novozymes Lipozyme lipase, and tartaric acid, preferably citric acid.

[0051] In one embodiment of the present invention, the neural network model in step (4) is constructed by minimizing the loss function L θ For training, the loss function is:

[0052]

[0053] Among them, y i For real sensory rating, is the sensory score predicted by the model, and θ is the model parameter.

[0054] Latin hypercube (LHS) is an efficient statistical method for uniformly covering multidimensional variable space to ensure that the training data of the neural network model is representative. In step (2), LHS is used to generate the combination ratio of different oils and fats to ensure the breadth and rationality of different ratio schemes. Specifically, the input variables include the content of saturated fatty acids (SFA), monounsaturated fatty acids (MUFA) and polyunsaturated fatty acids (PUFA) of each oil and fat, and constraints are imposed to ensure that the total ratio of the three oils is always 100%. The sampling process first sets the number of samples and performs uniformly distributed random sampling within the range of each variable to ensure that the ratio of different plant fats is not concentrated in a specific area, and is normalized to meet the sum constraint.

[0055] The data sampled in step (2) is used to train a neural network, where the input is the ratio of each oil in the oil mixture and the output is the sensory evaluation score of the plant milk corresponding to the oil mixture. The neural network predicts the optimal oil combination by learning the relationship between the ratio and the sensory evaluation, so that the flavor of the final product is closest to that of cow's milk. LHS sampling ensures that the neural network can fully learn the impact of different plant oil ratios, improve prediction accuracy, and ultimately optimize the plant oil ratio to maximize the simulation of cow's milk flavor.

[0056] In one embodiment of the present invention, in step (5), when Latin hypercube sampling is performed, the temperature is selected in the range of 40°C, 50°C, 60°C, 70°C, and 80°C; the time is selected in the range of 0.25h, 0.5h, 0.75h, 1h, and 1.25h; the pressure is selected in the range of 0.09MPa, 0.1MPa, 0.11MPa, 0.12MPa, and 0.13MPa; and the catalyst is selected in the range of Novozymes Lipozyme lipase, aluminate catalyst, citric acid, calcium hydroxide, and tartaric acid.

[0057] In step (5), LHS is used to generate the combination ratios of different processing conditions (temperature, pressure, reaction time, catalyst) to ensure the breadth and rationality of different ratio schemes. Specifically, the input variables include each processing condition such as temperature, pressure, reaction time, and catalyst. At the same time, constraints are imposed to ensure that the four processing conditions are always within the preset range, namely temperature (40°C, 50°C, 60°C, 70°C, 80°C), time (0.25h, 0.5h, 0.75h, 1h, 1.25h), pressure (0.09MPa, 0.1MPa, 0.11MPa, 0.12MPa, 0.13MPa) and catalyst (Novozymes Lipozyme lipase, aluminate catalyst, citric acid, calcium hydroxide, tartaric acid). The sampling process first sets the number of samples, and performs uniformly distributed random sampling within the range of each variable, which is normalized to meet the total constraint.

[0058] The data sampled in step (5) is used to train a neural network, wherein the input is the process condition combination and the output is the comprehensive score of the sensory evaluation, oxidative stability and SFC curve of the plant milk. The neural network predicts the optimal combination of process conditions by learning the relationship between the process condition combination and the sensory evaluation, so that the flavor and SFC curve of the final product are close to those of cow's milk and the oxidative stability is improved. Sampling ensures that the neural network can fully learn the effects of different process condition combinations, improve prediction accuracy, and ultimately optimize the process conditions to maximize the simulation of cow's milk flavor.

[0059] Traditional neural network methods often only address a single output task. However, the present invention requires simultaneous consideration of the flavor score, oxidative stability, and SFC curve of the test sample. Therefore, to optimize these conditions simultaneously, the present invention incorporates a Pareto optimization approach. The goal is to find an optimal solution between sensory scores, antioxidant capacity, and SFC curve, avoiding the potential bias associated with simple weighted sum methods.

[0060] In one embodiment of the present invention, in step (8), the specific objective function of the neural network model is:

[0061] Maximizey=Maximize{y1,y2,y3}

[0062] Where y1 is the sensory evaluation score, y2 is the oxidative stability, and y3 is the similarity of the SFC curves. y3 is calculated by subtracting the SFC value of the mixed vegetable oil at the corresponding temperature from the SFC value of the milk fat at each temperature.

[0063] Using the Pareto optimization algorithm, given reaction conditions, a set of solutions can be found, each optimal for at least one objective and not dominated by any other solution for another objective. For example, when selecting reaction conditions, if improving sensory scores results in a slight decrease in antioxidant and SFC curves, Pareto optimization can help select a solution that strikes a balance between the three objectives, avoiding an extreme bias toward any one objective. The Pareto optimization algorithm can be used for multi-objective optimization. Neural network optimization results will provide a set of Pareto frontier solutions, allowing the selection of the combination of reaction conditions that best meets the requirements without relying on manually set weights.

[0064] In one embodiment of the present invention, in step (8), the optimization objective function of the neural network is as follows:

[0065] Maximizey=β0+β1*T+β2*P+β3*t+β4*X

[0066] Among them, β0~β4 are the weight parameters to be optimized, and T, P, t and X are the temperature, pressure, time and catalyst in the transesterification reaction process, respectively.

[0067] The objective function is optimized by the gradient descent method. The formula for gradient descent is:

[0068]

[0069] in, is the value of the i-th parameter in the k-th iteration, ψ is the learning rate, and J(β) is the objective function (comprehensive score) on the parameter β k The partial derivative of , which represents the influence of the parameter on the objective function.

[0070] Beneficial effects:

[0071] (1) The present invention selects different combinations of plant oils and fats, and then carries out an ester exchange reaction on the oil composition to prepare a special fat for plant milk. The special fat for plant milk of the present invention can be used in the preparation of plant milk, thereby improving the appearance, smell, taste, flavor and texture of the plant milk. At the same time, the prepared plant milk has good antioxidant properties, and the SFC curve of the plant milk is close to that of cow's milk, which makes the plant milk comparable to cow's milk in taste, flavor and nutritional characteristics, and increases the possibility of plant milk replacing cow's milk.

[0072] (2) The vegetable oils used in the present invention are widely available and relatively low in cost. The preparation method of the plant milk-specific fat is simple and easy to operate, and is suitable for large-scale production.

[0073] (3) The plant milk fat of the present invention does not contain cholesterol and is suitable for consumption by a wider range of people. It also provides new technical support for the development of the plant milk industry.

[0074] (4) The present invention cleverly combines Latin hypercube sampling to screen out multiple groups of representative oil compositions and process parameter combinations for transesterification reactions, and combines the neural network model to screen out the optimal oil composition and optimal transesterification reaction conditions. Under the optimal oil composition and optimal transesterification reaction conditions of the present invention, the best-performing plant milk fat was prepared. After the screened plant milk fat was applied to the preparation of plant milk, the resulting plant milk had a high sensory score and an SFC curve close to that of cow's milk, indicating that the screening accuracy of the present invention method is high. In addition, if several groups of oil compositions and transesterification conditions are manually randomly screened and then combined with neural network prediction, the prediction effect is poor, and it is difficult to obtain a plant milk fat with excellent performance within a limited number of experiments. However, the present invention, combined with Latin hypercube sampling, can omit a large number of experiments and screening processes, greatly reducing R&D costs.

[0075] (5) Conventional neural network model predictions are difficult to take into account the sensory scores, antioxidant properties, and SFC curves of plant milk. Instead, only one set of indicators is considered to screen the process condition combinations. The screened process conditions are difficult to achieve a balance among the sensory scores, antioxidant properties, and SFC curves. After multiple experiments, it is also difficult to screen out the transesterification conditions that can produce plant milk fat with excellent performance in all aspects. The present invention accurately predicts the transesterification conditions that are conducive to obtaining plant milk fat with excellent performance through Latin hypercube sampling and Pareto neural network model. The oil composition and transesterification parameters obtained by the present invention are used to carry out transesterification reactions, and plant milk fat with the same SFC curve as cow milk fat is successfully prepared. The application of plant milk can significantly improve the flavor and taste of plant milk, making it closer to cow milk, and greatly reducing the number of experiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 The SFC curves of the plant milk fat and cow milk fat prepared in Examples 1 to 2 and Comparative Examples 1 to 4;

[0077] Figure 2 The SFC curves of the plant milk fat and cow milk fat prepared in Examples 3 to 4 and Comparative Examples 5 to 7;

[0078] Figure 3 The SFC curves of the plant milk fat and cow milk fat prepared in Examples 5-6 and Comparative Examples 8-9;

[0079] Figure 4 The SFC curves of plant milk fat and cow milk fat prepared according to the optimal oil composition and optimal process parameter combination screened in Example 7;

[0080] Figure 5The SFC curves of plant milk fat and cow milk fat prepared according to the optimal oil composition and optimal process parameter combination screened in Example 8;

[0081] Figure 6 The SFC curves of plant milk fat and cow milk fat prepared according to the optimal oil composition and optimal process parameter combination screened in Example 9;

[0082] Figure 7 The present invention provides a flow chart for preparing plant milk-specific fat based on machine learning. DETAILED DESCRIPTION

[0083] The present invention is further described in detail below with reference to specific embodiments.

[0084] Source of raw materials:

[0085] Novozymes Lipozyme Lipase is derived from Rhizomucor miehei and has an enzyme activity of 500 LU / g. Enzyme activity is defined as the amount of substrate decomposition catalyzed by the enzyme per minute under standard reaction conditions.

[0086] Test method:

[0087] 1. Fatty Acid Composition of Vegetable Oils: Fatty acid methyl esters (FAMs) were determined by gas chromatography. A 0.5 g vegetable oil sample was added to 1 mL of methanol and 0.2 mL of sodium hydroxide, and the mixture was reacted at 65°C for 30 minutes. After the reaction, 3 mL of chloroform and 1 mL of water were added for separation and extraction. The chloroform solvent was removed after drying to obtain the FAME sample. Analysis was performed using a gas chromatograph equipped with a flame ionization detector (FID) using a DB-23 column (30 m × 0.25 mm × 0.25 μm) and helium as the carrier gas at a flow rate of 1.0 mL / min. The injection volume was 1 μL. The temperature program was as follows: an initial temperature of 140°C for 2 minutes, followed by a temperature increase of 3°C / min to 250°C and a temperature hold of 20 minutes. The detector and injector temperatures were set at 260°C and 250°C, respectively. Qualitative analysis of the FAMEs was performed based on the retention time of the FAMEs compared with a standard. Quantitative analysis was performed using the peak area method to calculate the relative content of each fatty acid.

[0088] 2. Sensory scoring method for plant milk:

[0089] The overall sensory evaluation of plant-based milk adopts a 10-point scale, using cow's milk as a reference, to evaluate the comprehensive performance of plant-based milk, including appearance, smell, taste, flavor and texture. The evaluation environment should be kept clean, free of interfering odors, and the temperature and humidity should be appropriate. Evaluators must undergo basic sensory evaluation training and conduct blind tests to avoid interference from external factors. The scoring process includes appearance observation, olfactory evaluation, taste and texture testing, comprehensive flavor evaluation, and ultimately an overall score. Each evaluator must score independently to avoid mutual influence, and the final score is the arithmetic average of multiple evaluators.

[0090] The scoring criteria are shown in the table below:

[0091] Fraction Rating Performance 9.0-10.0 Excellent Perfect or near-perfect performance in all aspects 8.0-8.9 excellent Excellent overall performance, with only minor shortcomings 7.0-7.9 good Overall good but with minor flaws 5.0-6.9 generally Acceptable but with obvious shortcomings that affect the experience 3.0-4.9 Poor There are obvious defects that affect edibility 1.0-2.9 Extremely poor There are serious problems with the food and it is difficult to accept 0.0-1.0 Unacceptable Food is completely spoiled or inedible

[0092] 3. Antioxidant test method:

[0093] Fat peroxide value (POV) test: Weigh 3.0g of fat sample, add 30mL of glacial acetic acid-chloroform mixture (the volume ratio of glacial acetic acid and chloroform is 3:2), dissolve, add 1mL of saturated potassium iodide solution, shake for 0.5min, and let it stand in the dark for 3min. After the reaction is completed, add 100mL of distilled water and mix well, add 1mL of starch indicator, and titrate with 0.01mol / L sodium thiosulfate solution until the blue color disappears completely. Record the volume of sodium thiosulfate consumed. Calculate the peroxide value according to the peroxide value calculation formula:

[0094]

[0095] Where: POV is the peroxide value, in g / 100 g, V is the volume of sodium thiosulfate consumed by the sample (mL), V0 is the volume of sodium thiosulfate consumed by the blank (mL), C is the concentration of the sodium thiosulfate standard solution (mol / L), m is the sample mass (g), 0.1269 is the mass of iodine equivalent to 1 mL of sodium thiosulfate standard titration solution, in g / mmol, and 100 is the conversion factor for 100 g of sample.

[0096] 4. SFC curve testing method: Weigh 3g of a fat sample into a dedicated NMR test tube and maintain the sample at 90°C for at least 30 minutes to eliminate the crystallization memory effect of the fat. Then, quickly transfer the sample to a 0°C constant temperature bath and maintain it for 2 hours. Use pulsed NMR to measure the solid fat content of the sample at 0-40°C, with a temperature gradient of 5°C. The sample is stabilized at each temperature for 30 minutes before testing to ensure measurement accuracy. By recording the solid fat content at different temperatures, an SFC curve can be drawn for the fat, reflecting the distribution of solid and liquid fat in the fat at different temperatures.

[0097] Example 1

[0098] A fat composition for special fat for plant milk comprises tung oil, camelina oil and coconut oil, wherein the mass ratio of tung oil, camelina oil and coconut oil is 3:2:5.

[0099] A method for preparing fat specifically for plant milk comprises the following steps:

[0100] (1) mixing the above-mentioned oil composition for preparing the plant milk fat according to proportion to obtain a mixed oil;

[0101] (2) The mixed oil was subjected to transesterification reaction at 60°C, 0.10 MPa and in the presence of citric acid as a catalyst for 0.75 h;

[0102] (3) separating and purifying the product after the transesterification reaction in step (2), removing the catalyst, and drying the product under a nitrogen flow at 50° C. for 2 h at a nitrogen flow rate of 0.5 L / min to obtain a fat for plant milk;

[0103] The specific steps of separation are as follows: n-hexane is selected as the solvent, and extraction is carried out at room temperature (25°C) in a ratio of 1:2 between the volume of the solvent and the volume of the transesterified product, and the extraction is performed three times, each extraction time being 30 minutes, and the fatty acids and other impurities in the solution are separated by centrifugation (3000 rpm, for 10 minutes).

[0104] The specific purification steps are as follows: elution on a silica gel column using a mixed solvent of n-hexane and ethyl acetate (solvent ratio of 80:20) at a controlled elution rate of 0.5 mL / min; after elution, separation and purification are performed using thin layer chromatography (TLC). When performing TLC, the stationary phase selected is silica gel, and the solvent system is n-hexane / ethyl acetate = 85:15 (v / v).

[0105] Removal of catalyst: The catalyst can be removed by adjusting the pH to 5.0-6.0 or increasing the reaction temperature to 70°C to inactivate the enzyme.

[0106] A method for preparing plant milk comprises the following steps:

[0107] First, the soybeans were pulverized to 50 μm using a jet mill and then spray-dried. The inlet air temperature was set at 170°C, the outlet air temperature was 80°C, and the spray pressure was controlled at 150 bar. The entire drying process was completed within 3 seconds. The resulting dry powder was then mixed with water in a 1:8 (w / v) ratio and stirred at 2000 rpm using an ultrasonic agitator for 20 minutes to fully disperse it. The temperature was maintained at 50°C during this process. Next, high-pressure homogenization was performed at a pressure of 150 MPa and homogenized twice to improve the stability of the emulsion. The homogenized emulsion was pasteurized at 80°C for 25 seconds to ensure microbial safety. Next, 0.2% xanthan gum, 3% sucrose, 0.3% lecithin, and 0.5% plant-based milk fat were added to the emulsion. The addition process was performed with stirring at 1500 rpm for 10 minutes to ensure that the oils were fully and evenly dispersed in the emulsion, resulting in the plant-based milk.

[0108] Example 2

[0109] The difference between Example 2 and Example 1 is that in the method for preparing the special fat for plant milk, the mixed oil is subjected to an ester exchange reaction at 50° C., 0.11 MPa and in the presence of a catalyst, citric acid, and the reaction time is 0.5 h.

[0110] Comparative Example 1

[0111] The difference between Comparative Example 1 and Example 1 is that the proportions of the oils in the oil composition are different, and the mass ratio of tung oil, camelina oil and coconut oil is 3:3:4.

[0112] Comparative Example 2

[0113] The difference between Comparative Example 2 and Example 1 is that the types and proportions of the oils in the oil composition are different. The oil composition contains camelina oil and coconut oil, and the mass ratio of the two is 1:1.

[0114] Comparative Example 3

[0115] The difference between Comparative Example 3 and Example 1 is that camelina oil is replaced by sesame oil.

[0116] Comparative Example 4

[0117] The difference between Comparative Example 4 and Example 1 is that in the preparation method of special fat for plant milk, the mixed oil is subjected to an ester exchange reaction at 50°C, 0.09MPa and in the presence of a catalyst calcium hydroxide, and the reaction time is 0.25h; after the reaction, filter paper (glass fiber filter paper with a pore size of 0.45μm) is used for filtration to remove the catalyst.

[0118] The sensory score results and antioxidant data of the plant milk obtained in Examples 1 to 2 and Comparative Examples 1 to 4 are shown in Table 1. It can be seen from Table 1 that only when the mass ratio of tung oil, camelina oil and coconut oil and the ester exchange reaction conditions are within the scope of protection claimed in the present invention, the prepared plant milk fat has a higher sensory score and better antioxidant property. Figure 1 It can also be seen that the SFC curves of the plant-based fats prepared in Examples 1 and 2 are relatively close to those of bovine milk fat (AMF). Comparative Example 1 varied the ratios of the various oils based on Example 1, while Comparative Examples 2 and 4 replaced one of the oils with another oil with a similar fatty acid content. This ultimately resulted in significantly reduced sensory scores and antioxidant activity for the plant-based fats, and the SFC curves of these fats also differed significantly from those of bovine milk fat (AMF). The experimental results of Comparative Example 4 demonstrate that transesterification reaction conditions can also significantly impact the properties of the prepared fats.

[0119] Table 1 Sensory evaluation results of the plant milk obtained in Examples 1 to 2 and Comparative Examples 1 to 4

[0120] example Sensory score Antioxidant properties Example 1 9.5 0.078 Example 2 9.2 0.085 Comparative Example 1 5.5 0.105 Comparative Example 2 4.6 0.112 Comparative Example 3 5.2 0.098 Comparative Example 4 6.4 0.101

[0121] Example 3

[0122] A fat composition for special fat for plant milk comprises palm oil, camelina oil and olive oil, wherein the mass ratio of palm oil, camelina oil and olive oil is 4:3:3.

[0123] A method for preparing fat specifically for plant milk comprises the following steps:

[0124] (1) mixing the above-mentioned oil composition for preparing the plant milk fat according to proportion to obtain a mixed oil;

[0125] (2) The mixed oil was subjected to transesterification reaction at 50°C, 0.11 MPa and in the presence of citric acid as a catalyst for 0.5 h;

[0126] (3) The product after the transesterification reaction in step (2) is separated and purified, the catalyst is removed, and the fat for plant milk is obtained after drying; the separation, purification, catalyst removal and drying methods are the same as those in Example 1.

[0127] The preparation method of the plant milk is the same as that in Example 1.

[0128] Example 4

[0129] The difference between Example 4 and Example 3 is that in the method for preparing the special fat for plant milk, the mixed oil is subjected to an ester exchange reaction at 60° C., 0.10 MPa and in the presence of a catalyst citric acid, and the reaction time is 0.75 h.

[0130] Comparative Example 5

[0131] The difference between Comparative Example 5 and Example 3 is that the proportions of the oils in the oil composition are different, and the mass ratio of tung oil, camelina oil and coconut oil is 3:3:4.

[0132] Comparative Example 6

[0133] The difference between Comparative Example 6 and Example 3 is that camelina oil is replaced by sesame oil.

[0134] Comparative Example 7

[0135] The difference between Comparative Example 7 and Example 3 is that, in the method for preparing the plant milk-specific fat, the mixed oil is subjected to transesterification reaction at 80°C, 0.11 MPa and in the presence of aluminate catalyst, and the reaction time is 0.5 h;

[0136] The sensory score results and antioxidant data of the plant milks obtained in Examples 3 to 4 and Comparative Examples 5 to 7 are shown in Table 2. It can be seen from Table 2 that only when the mass ratio of palm oil, camelina oil and olive oil and the ester exchange reaction conditions are within the scope of the present invention, the prepared plant milk fat has a higher sensory score and better antioxidant property. Figure 2 It can also be seen that the SFC curves of the plant-based fats prepared in Examples 3 and 4 are relatively close to those of bovine milk fat. Comparative Example 5 varied the ratios of the various oils based on Example 1, and Comparative Example 6 replaced one of the oils with another with a similar fatty acid content. This ultimately resulted in significantly lower sensory scores and antioxidant activity for the plant-based fats, and the SFC curves of these fats also differed significantly from those of bovine milk fat. The experimental results of Comparative Example 7 demonstrate that transesterification reaction conditions can also significantly influence the properties of the prepared fats.

[0137] Table 2 Sensory evaluation results of the plant milk obtained in Examples 1 to 2 and Comparative Examples 1 to 4

[0138]

[0139]

[0140] Example 5

[0141] A fat composition for special fat for plant milk comprises tung oil, camelina oil, coconut oil and palm oil, wherein the mass ratio of tung oil, camelina oil, coconut oil and palm oil is 4:2:3:1.

[0142] A method for preparing fat specifically for plant milk comprises the following steps:

[0143] (1) mixing the above-mentioned oil composition for preparing the plant milk fat according to proportion to obtain a mixed oil;

[0144] (2) The mixed oil was subjected to transesterification reaction at 60°C, 0.10 MPa and in the presence of citric acid as a catalyst for 0.75 h;

[0145] (3) The product after the transesterification reaction in step (2) is separated and purified, the catalyst is removed, and the fat for plant milk is obtained after drying; the separation, purification, catalyst removal and drying methods are the same as those in Example 1.

[0146] The preparation method of the plant milk is the same as that in Example 1.

[0147] Example 6

[0148] The difference between Example 6 and Example 5 is that in the method for preparing the special fat for plant milk, the mixed oil is subjected to an ester exchange reaction at 50° C., 0.13 MPa and in the presence of a catalyst citric acid, and the reaction time is 1.25 h.

[0149] Comparative Example 8

[0150] The difference between Comparative Example 8 and Example 5 is that the proportions of the oils in the oil composition are different, and the mass ratio of tung oil, camelina oil and coconut oil is 2:2:3:3.

[0151] Comparative Example 9

[0152] The difference between Comparative Example 9 and Example 5 is that camelina oil is replaced by sesame oil.

[0153] Comparative Example 10

[0154] The difference between Comparative Example 10 and Example 5 is that, in the method for preparing the plant milk-specific fat, the mixed oil is subjected to transesterification reaction at 80° C., 0.12 MPa and in the presence of tartaric acid as a catalyst, and the reaction time is 0.75 h;

[0155] The sensory evaluation results and antioxidant data of the plant milks obtained in Examples 5-6 and Comparative Examples 8-10 are shown in Table 3. It can be seen from Table 3 that only when the mass ratio of tung oil, camelina oil, coconut oil and palm oil and the ester exchange reaction conditions are within the scope of protection claimed in the present invention, the prepared plant milk fat has a higher sensory score and better antioxidant property. Figure 3It can also be seen that the SFC curves of the plant-based fats prepared in Examples 5 and 6 are relatively similar to those of bovine milk fat. Comparative Example 8 varied the ratios of the various oils based on Example 1, and Comparative Example 9 replaced one of the oils with another with a similar fatty acid content. This ultimately resulted in significantly lower sensory scores and antioxidant activity for the plant-based fats, and the SFC curves of these fats also differed significantly from those of bovine milk fat. The experimental results of Comparative Example 10 demonstrate that transesterification reaction conditions can also significantly influence the properties of the prepared fats.

[0156] Table 3 Sensory evaluation results of the plant milks obtained in Examples 5 to 6 and Comparative Examples 8 to 10

[0157] example Sensory score Antioxidant properties Example 5 9.8 0.076 Example 6 8.8 0.096 Comparative Example 8 6.8 0.093 Comparative Example 9 7.5 0.096 Comparative Example 10 5.6 0.081

[0158] Example 7

[0159] A method for preparing plant milk-specific fat based on machine learning, comprising the following steps:

[0160] (1) Castanopsis oil, camelina oil, and coconut oil were selected and Latin hypercube sampling was used to screen oil compositions with different proportions. The input variables during sampling were the proportions of saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), and polyunsaturated fatty acids (PUFA) of the three oils. Constraints were imposed to ensure that the total proportion of the three oils was always 100%, and the number of samples was set to 9.

[0161] (2) mixing the oil and fat composition screened out in step (1), performing an ester exchange reaction, and preparing the plant milk-specific fat and plant milk according to the method of Example 1;

[0162] (3) performing sensory evaluation on the plant milk obtained in step (2) to obtain sensory scores of the plant milk based on different oil compositions;

[0163] (4) Using a neural network model to predict the optimal ratio of the oil composition, the input feature of the neural network model is the ratio of each oil in the oil composition in step (2), and the output feature is the sensory score of the plant milk corresponding to the oil composition; the neural network model minimizes the loss function L θ For training, the loss function is:

[0164]

[0165] Among them, y i For real sensory rating, is the sensory score predicted by the model, and θ is the model parameter.

[0166] (5) Latin hypercube sampling was used to screen several sets of process condition combinations for transesterification reactions, wherein the process parameters included temperature, pressure, catalyst, and reaction time. The input variables were each processing condition, such as temperature, pressure, reaction time, and catalyst. Constraints were imposed to ensure that the four processing conditions were always within the preset range, namely, temperature (40°C, 50°C, 60°C, 70°C, and 80°C), time (0.25h, 0.5h, 0.75h, 1h, and 1.25h), pressure (0.09MPa, 0.1MPa, 0.11MPa, 0.12MPa, and 0.13MPa), and catalyst (Novozymes Lipozyme lipase, aluminate catalyst, citric acid, calcium hydroxide, and tartaric acid). The number of samples was set to 9.

[0167] (6) using the oil composition with the optimal ratio obtained in step (4) as a raw material, performing an ester exchange reaction according to the process condition combination screened in step (5) and preparing the plant milk according to the method of step (2);

[0168] (7) performing a sensory evaluation on the plant milk obtained in step (6), measuring the antioxidant properties and SFC curve of the plant milk-specific fat, and obtaining the sensory evaluation, antioxidant properties, and SFC curve of the plant milk based on different process condition combinations;

[0169] (8) A neural network model is used to predict the optimal process condition combination for the transesterification reaction. The input features of the neural network model are the process condition combination based on Latin hypercube screening in step (5), and the output features are the sensory scores of plant milk, fat antioxidant properties and SFC curves, and milk fat SFC curves corresponding to different process condition combinations. The objective function of the neural network model is:

[0170] Maximizey=Maximize{y1,y2,y3}

[0171] Where y1 is the sensory evaluation score, y2 is the oxidative stability, and y3 is the similarity of the SFC curves. y3 is calculated by subtracting the SFC value of the mixed vegetable oil at the corresponding temperature from the SFC value of the milk at each temperature.

[0172] (9) performing an ester exchange reaction based on the optimal ratio of the oil and fat composition selected in step (4) in combination with the optimal temperature, pressure, catalyst and reaction time obtained in step (8), and separating, purifying and drying to obtain the plant-based fat.

[0173] Table 4 Optimization of the addition ratio of tung oil, camelina oil and coconut oil and sensory scores

[0174] Tung oil camelina oil coconut oil Sensory evaluation scores 10 0 0 6.7 0 10 0 5.3 0 0 10 4.0 3 3 4 5.5 2 4 4 4.7 4 3 3 5.3 5 2 3 7.5 6 2 2 5.7 6 3 1 5.5 3 2 5 9.2

[0175] The ratio of the oil composition screened according to step (1) and the scoring results obtained in step (3) are shown in Table 4. The data in Table 4 were subjected to neural network training according to step (4), and the optimal ratio of tung oil, camelina oil, and coconut oil was obtained to be 3:2:5. The oil composition was used to prepare a plant milk, and the sensory score of the plant milk was 9.5 points.

[0176] Table 5 Optimization of processing conditions, sensory scores and peroxide values of tung oil, camelina oil and coconut oil

[0177]

[0178]

[0179] The process condition combinations selected according to step (5) and the scoring results obtained in step (7) are shown in Table 5. The data in Table 5 were subjected to neural network training according to step (8), and the optimal process condition combination was obtained as follows: temperature 60°C, pressure 0.10 MPa, catalyst citric acid, and reaction time 0.75 h (i.e., A-10 in Table 5). The oil composition was used to prepare a plant milk, and the sensory score of the plant milk was 9.5 points, and the peroxide value was 0.078.

[0180] Figure 4 Table 5 is the SFC curve results of the plant milk prepared with the mass ratio of tung oil, camelina oil and coconut oil of 3:2:5 combined with the process conditions in Table 5. Figure 4 It can be seen that the SFC curve of A-10 (SFC curve of plant milk fat prepared with the optimal oil ratio and optimal process conditions) is close to the SFC curve of cow milk fat. Combined with the sensory score and antioxidant results, it can be concluded that the flavor and taste of the plant milk prepared using the oil ratio and process conditions screened by the method of the present invention are close to that of cow milk. The method of the present invention is simple and has a high accuracy rate.

[0181] Example 8

[0182] The difference between Example 8 and Example 7 is that the oils used in step (1) are palm oil, camelina oil and olive oil, and the transesterification reaction conditions in step (2) are the same as those in Example 3.

[0183] Table 6 Optimization of palm oil, camelina oil and olive oil addition ratio and sensory scores

[0184] palm oil camelina oil olive oil Sensory evaluation scores 10 0 0 6.8 0 10 0 4.3 0 0 10 4.7 3 3 4 5.2 2 4 4 3.8 4 3 3 8.2 5 2 3 8.7 6 2 2 8.4 8 1 1 7.9 4 3 3 9.3

[0185] The ratio of the oil composition screened according to step (1) and the scoring results obtained in step (3) are shown in Table 6. The data in Table 6 were subjected to neural network training according to step (4), and the optimal ratio of palm oil, camelina oil, and olive oil was obtained to be 4:3:3. The oil composition was used to prepare a plant milk, and the sensory score of the plant milk was 9.3 points.

[0186] Table 7 Optimization of processing conditions, sensory scores and peroxide values of palm oil, camelina oil and olive oil

[0187]

[0188] The process condition combinations selected according to step (5) and the scoring results obtained in step (7) are shown in Table 7. The data in Table 7 were subjected to neural network training according to step (8), and the optimal process condition combination was obtained as follows: temperature 50°C, pressure 0.10 MPa, catalyst citric acid, and reaction time 0.75 h (i.e., A-10 in Table 5). The oil composition was used to prepare a plant milk, and the sensory score of the plant milk was 9.4 points, and the peroxide value was 0.068.

[0189] Figure 5 Table 7 is the SFC curve results of the plant milk prepared with the mass ratio of palm oil, camelina oil and olive oil 4:3:3 combined with the process conditions in Table 7. Figure 5 It can be seen that the SFC curve of B-10 (SFC curve of plant milk prepared with the optimal oil ratio combined with the optimal process conditions screened out) is close to that of cow's milk. Combined with the sensory score and antioxidant results, it can be concluded that the flavor and taste of the plant milk prepared using the oil ratio and process conditions screened out by the method of the present invention are close to that of cow's milk. The method of the present invention is simple and has a high accuracy rate.

[0190] Example 9

[0191] The difference between Example 9 and Example 7 is that the oils used in step (1) are tung oil, camelina oil, coconut oil and palm oil.

[0192] Table 8 Optimization of addition ratios of tung oil, camelina oil, coconut oil and palm oil and sensory scores

[0193]

[0194]

[0195] The ratio of the oil composition screened according to step (1) and the scoring results obtained in step (3) are shown in Table 8. The data in Table 8 were subjected to neural network training according to step (4), and the optimal ratio of tung oil, camelina oil, coconut oil, and palm oil was obtained to be 4:2:3:1. The oil composition was used to prepare a plant milk, and the sensory score of the plant milk was 9.8 points.

[0196] Table 9 Optimization of processing conditions for tung oil, camelina oil, coconut oil and palm oil

[0197] Sample number Temperature (℃) Reaction pressure (MPa) Catalyst type Reaction time (h) Sensory evaluation scores Peroxide value C-1 50 0.11 Citric acid 0.50 7.8 0.086 C-2 70 0.10 calcium hydroxide 0.25 5.4 0.084 C-3 80 0.12 NovozymesLipozyme lipase 1.25 6.2 0.091 C-4 80 0.12 tartaric acid 0.75 5.6 0.081 C-5 40 0.13 NovozymesLipozyme lipase 0.25 7.4 0.094 C-6 50 0.11 tartaric acid 0.75 6.2 0.105 C-7 50 0.09 calcium hydroxide 0.25 4.7 0.112 C-8 80 0.11 Aluminate catalyst 0.50 5.2 0.094 C-9 50 0.13 Citric acid 1.25 8.8 0.096 C-10 60 0.10 Citric acid 0.75 9.8 0.076

[0198] The process condition combinations selected according to step (5) and the scoring results obtained in step (7) are shown in Table 9. The data in Table 9 were subjected to neural network training according to step (8), and the optimal process condition combination was obtained as follows: temperature 60°C, pressure 0.10 MPa, catalyst citric acid, and reaction time 0.75 h (i.e., C-10 in Table 9). The oil composition was used to prepare plant milk, and the sensory score of the plant milk was 9.8 points, and the peroxide value was 0.076.

[0199] Figure 6 Table 9 is the SFC curve results of the plant milk prepared with the mass ratio of tung oil, camelina oil, coconut oil and palm oil of 4:2:3:1 combined with the process conditions in Table 7. Figure 6 It can be seen that the SFC curve of C-10 (SFC curve of plant milk prepared with the optimal oil ratio and optimal process conditions) is close to that of cow's milk. Combined with the sensory score and antioxidant results, it can be concluded that the flavor and taste of the plant milk prepared using the oil ratio and process conditions screened by the method of the present invention are close to that of cow's milk. The method of the present invention is simple and has a high accuracy rate.

[0200] The embodiments provided above are not intended to limit the scope of the present invention, nor are the steps described to limit their execution order. Any obvious improvements to the present invention made by those skilled in the art in combination with existing common knowledge shall fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for preparing fat for vegetable milk, characterized in that: The following steps are involved: (1) mixing a fat composition for preparing a fat for vegetable milk to obtain a mixed fat; (2) subjecting the mixed oil to transesterification reaction in the presence of a catalyst; (3) separating and purifying the product after the transesterification reaction in step (2), removing the catalyst, and drying to obtain a fat specifically for plant milk; the oil composition comprises tung oil, camelina oil and coconut oil, and the mass ratio of the tung oil, camelina oil and coconut oil is 2.9-3.1:1.8-2.2:4.7-5.

3.

2. The preparation method according to claim 1, characterized in that The oil composition comprises palm oil, camelina oil and olive oil, and the mass ratio of the palm oil, camelina oil and olive oil is 3.8-4.2:2.9-3.1:2.9-3.

1.

3. The preparation method according to claim 1, characterized in that The oil composition comprises tung oil, camelina oil, coconut oil and palm oil, and the mass ratio of the tung oil, camelina oil, coconut oil and palm oil is 3.8-4.2:1.9-2.1:2.9-3.1:0.9-1.

1.

4. The preparation method according to any one of claims 1 to 3, characterized in that: The temperature of the transesterification reaction in step (2) is 40-60° C., the reaction time is 0.25-1.25 h, the pressure is 0.1-0.15 MPa, and the catalyst includes at least one of citric acid, Novozymes Lipozyme lipase, and tartaric acid, preferably citric acid.

5. The preparation method according to claim 4, characterized in that The drying is one of vacuum drying, spray drying, freeze drying or nitrogen drying.

6. The fat for vegetable milk prepared according to the method according to any one of claims 1 to 3.

7. Use of the fat for plant milk according to claim 6 in preparing plant milk, characterized in that: The application is to add special fat for plant milk to plant milk without adding plant oil, and the addition amount of the special fat for plant milk accounts for 0.5-1.0% of the total mass of the finally prepared plant milk. The plant milk without adding plant oil is prepared according to a conventional method.

8. A method for preparing plant milk-specific fat based on machine learning, characterized in that: The following steps are involved: (1) Select several groups of different oils and fats and use Latin hypercube sampling to screen oil compositions with different proportions; (2) mixing the oil composition screened out in step (1), performing an ester exchange reaction according to the parameters of claim 4, separating, purifying and drying to obtain a plant-based fat, and adding the obtained plant-based fat to a plant milk to which no plant oil is added at a ratio of 0.5 to 1.0% of the total mass of the plant milk to prepare the plant milk; (3) performing sensory evaluation on the plant milk obtained in step (2) to obtain sensory scores of the plant milk based on different proportions of the oil and fat composition; (4) using a neural network model to predict the optimal ratio of the oil composition, wherein the input feature of the neural network model is the ratio of each oil in the oil composition in step (2), and the output feature is the sensory score of the plant milk corresponding to the oil composition; (5) using Latin hypercube sampling to screen several sets of process condition combinations for transesterification reactions, wherein the process parameters include temperature, pressure, catalyst, and reaction time; (6) using the oil composition with the optimal ratio obtained in step (4) as a raw material, performing an ester exchange reaction according to the process condition combination screened in step (5) and preparing the plant milk according to the method of step (2); (7) performing a sensory evaluation on the plant milk obtained in step (6), measuring the antioxidant properties and SFC curve of the plant milk-specific fat, and obtaining the sensory scores, antioxidant properties, and SFC curves of the plant milk based on different process condition combinations; (8) using a neural network model to predict the optimal process condition combination for the transesterification reaction, wherein the input features of the neural network model are the process condition combination based on Latin hypercube screening in step (5), and the output features are the sensory scores of the plant milk, the antioxidant properties and SFC curves of the fat, and the SFC curves of the milk fat corresponding to different process condition combinations; (9) performing an ester exchange reaction based on the optimal ratio of the oil and fat composition selected in step (4) in combination with the optimal temperature, pressure, catalyst and reaction time obtained in step (8), and separating, purifying and drying to obtain the plant milk-specific fat.

9. The preparation method according to claim 8, wherein the Latin hypercube sampling in step (2) is performed with a constraint condition such that the sum of the proportions of the selected oils and fats is 100%, and the neural network model in step (4) is constructed by minimizing the loss function L θ For training, the loss function is: in, y i For real sensory rating, is the sensory score predicted by the model, and θ is the model parameter.

10. The method according to claim 8, characterized in that In step (5), when performing Latin hypercube sampling, the temperature selection range is 40°C, 50°C, 60°C, 70°C, and 80°C; the time selection range is 0.25h, 0.5h, 0.75h, 1h, and 1.25h; the pressure selection range is 0.09MPa, 0.1MPa, 0.11MPa, 0.12MPa, and 0.13MPa; and the catalyst selection range is Novozymes Lipozyme lipase, aluminate catalyst, citric acid, calcium hydroxide, and tartaric acid; the specific objective function of the neural network model in step (8) is: Maximizey=Maximize{y1,y2,y3} Where y1 is the sensory evaluation score, y2 is the oxidative stability, and y3 is the similarity of the SFC curves. y3 is calculated by subtracting the SFC value of the mixed vegetable oil at the corresponding temperature from the SFC value of the milk fat at each temperature.

Images