Grain soaking process moisture content prediction method based on neural network
Through the neural network-based prediction method of moisture content in grain soaking process, the problem of cumbersome detection of grain moisture content in winemaking is solved, and rapid and accurate prediction of moisture content is achieved, which improves the quality of liquor and reduces production costs.
Patent Information
- Application Number
- CN202411315336.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2025-01-10
AI Technical Summary
During the brewing process, the prior art needs to frequently detect the moisture content of new varieties of grains, which leads to cumbersome operations and time-consuming, affecting the quality of base wine and production costs.
The moisture content prediction method during the food immersion process based on neural network is used to quickly predict the moisture content during the food immersion process by measuring the physical and chemical properties of grain, recording the moisture content at different immersion times, analyzing the correlation, and establishing a neural network prediction model.
It has achieved rapid and accurate prediction of the moisture content of the grain soaking process of different varieties of grains, reduced the exploration cycle of new varieties of grain processes, improved the quality of liquor and reduced production costs.
Smart Images
Figure CN119294657A_ABST
Abstract
Description
Technical Field
[0001] The invention mainly relates to the technical field of wine brewing, and in particular to a method for predicting moisture content in a grain soaking process based on a neural network. Background Art
[0002] In the brewing of liquor, grain moistening is a key process. The moisture content of grain after moistening directly affects the yield and flavor of base liquor. Therefore, for new varieties or grains with different physical and chemical properties, it is necessary to frequently test the moisture content of grain in the moistening process in order to determine the moistening time and meet the requirements of the moistening process. The traditional way of moistening grain is mainly to pile up and moisten the grain. High-temperature water is added to the grain pile for stirring. After several similar operations, the grain reaches a suitable moisture content. However, the pile-up and moistening process has defects such as long time consumption and uneven stirring. In addition, the operation is cumbersome, the process is difficult to control, and the production cost is high. After exploration and innovation, another way of moistening the grain process is soaking. Soaking grain is simple to operate and the corresponding parameters are easy to control, so it is more suitable for large-scale industrial production. However, the moisture content of different varieties of grain after soaking will vary due to their different physical and chemical properties. The starch composition and starch content, protein content, fat content and tannin content directly affect the moisture content of the grain after the soaking process, which in turn affects the quality of the base liquor. At the same time, the whole grain is usually soaked in constant temperature water in the soaking process, because the whole grain has a large porosity between its particles when used as a brewing raw material, which can minimize the use of bran and husk, and fundamentally reduce the introduction of the unpleasant taste of raw materials into the wine; in addition, the whole grain is fed into the mash system to slowly release nutrients, thereby ensuring the supply of nutrients during the long fermentation period. However, in the prior art, when introducing a new variety of brewing raw materials, it is usually necessary to use a simulation experiment method to determine the moisture content of the grain during the soaking process, which is cumbersome and time-consuming. Summary of the invention
[0003] Technical Problems to be Solved by the Invention
[0004] The invention provides a method for predicting the moisture content of grains in a grain soaking process based on a neural network, aiming at quickly and easily obtaining the real-time moisture content of grains in the grain soaking process.
[0005] The technical solution adopted by the present invention to solve the above technical problems
[0006] A method for predicting moisture content in a grain soaking process based on a neural network, comprising:
[0007] Step S1, obtaining a grain sample and measuring various physical and chemical properties of the grain sample;
[0008] Step S2, recording the moisture content of the grain corresponding to different soaking times in constant temperature water;
[0009] Step S3, analyzing the correlation between various physical and chemical properties of grains and the moisture content during the grain soaking process, and determining the key physical and chemical properties that affect the moisture content during the grain soaking process according to the degree of correlation;
[0010] Step S4, establishing a neural network prediction model, taking the initial moisture content, key physical and chemical properties and soaking time of the grain as inputs of the neural network prediction model, taking the moisture content corresponding to the soaking time as output of the neural network prediction model, and training the neural network prediction model;
[0011] Step S5: input the initial moisture content, key physical and chemical properties and soaking time of the grain to be tested for moisture content into the trained neural network prediction model, and the neural network prediction model outputs the moisture content corresponding to the soaking time.
[0012] Furthermore, the physical and chemical properties described in step 1 specifically include the fat content, tannin content, protein content, starch content, amylose content, amylopectin content, hardness, specific surface area and seed coat thickness of the grain.
[0013] Furthermore, step 2 specifically includes: using the peleg model The moisture content of simulated grains corresponding to different soaking times in constant temperature water, where t is the soaking time, M t is the moisture content of the grain after soaking for t time, M0 is the initial moisture content of the grain, k1 is the rate constant, and k2 is the capacity constant.
[0014] Furthermore, the analysis of the correlation between various physicochemical properties of grain and the moisture content of the grain soaking process described in step 3 specifically includes: performing correlation analysis on the physicochemical properties of the grain with the rate constant k1 and the capacity constant k2 respectively; if the degree of correlation between a certain physicochemical property and k1 or k2 exceeds a set threshold, then the physicochemical property is determined as a key physicochemical property that affects the moisture content of the grain soaking process.
[0015] Furthermore, the various physicochemical properties of grain were subjected to bivariate Pearson correlation analysis with the rate constant k1 and the capacity constant k2, and a two-sided significance test was selected. The physicochemical properties with a significant p-value less than the set value were identified as the key physicochemical properties affecting the moisture content of the grain soaking process.
[0016] Furthermore, step S4 specifically includes: converting the key physical and chemical properties, soaking time and the moisture content corresponding to the soaking time into values in the interval [0,1] and inputting them into the established neural network prediction model. The specific conversion method is: X' represents the value after preprocessing, X i represents the value before preprocessing, X min Represents the minimum value in the data, X maxIndicates the maximum value in the data.
[0017] Beneficial effects of the present invention
[0018] The present invention provides a method for predicting the moisture content in a grain soaking process based on a neural network. Firstly, the key physical and chemical properties that can affect the moisture content of the grain after soaking are screened out through correlation analysis between the physical and chemical properties of the grain and the moisture content of the grain in the grain soaking process, and a neural network prediction model is constructed. The key physical and chemical properties of the grain, the soaking time and the initial moisture content are used as inputs of the neural network prediction model, and the moisture content in the grain soaking process is used as the output of the neural network prediction model to train the neural network prediction model. The neural network prediction model after training can quickly and accurately predict the moisture content of different varieties of brewing grains in the grain soaking process. The prediction result is helpful for screening brewing raw material varieties and can also be used for adjusting process parameters, thereby greatly shortening the exploration cycle of the new variety of grain soaking process, which is of great significance to improving the quality of liquor and reducing production costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 This is a flow chart of a method for predicting moisture content in a grain soaking process based on a neural network according to the present invention;
[0020] Figure 2 It is the regression result of the training set BP neural network prediction model and the actual value in the embodiment;
[0021] Figure 3 It is the regression result of the BP neural network prediction model and the actual value of the verification set in the embodiment;
[0022] Figure 4 It is the regression result of the BP neural network prediction model and the actual value of the test set in the embodiment;
[0023] Figure 5 The comprehensive regression results of the BP neural network prediction model and the actual value are summarized for the embodiment. DETAILED DESCRIPTION
[0024] like Figure 1 As shown, the method for predicting moisture content in a grain soaking process based on a neural network described in the present invention specifically includes:
[0025] Step S1: Obtain grain samples, and measure and record the fat content, tannin content, protein content, amylopectin content, amylose content, hardness, specific surface area, seed coat thickness, initial moisture content, etc. of the grain samples.
[0026] Step S2: Soak the grain sample in constant temperature water, and measure the moisture content of the grain sample at different soaking time points, wherein the temperature of the constant temperature water is set between 30-80°C.
[0027] As a preferred embodiment, the peleg model is selected in this embodiment. Simulate the hydration dynamics of grain during soaking to obtain the moisture content of grain at different soaking times, where M t It represents the moisture content of grain after soaking for t time; M0 represents the initial moisture content of grain before soaking; k1 is the rate constant; k2 is the capacity constant.
[0028] Step S3: Perform Pearson correlation analysis on the physicochemical properties of the grain sample, such as fat content, tannin content, protein content, starch content, amylose content, hardness, specific surface area, seed coat thickness, initial moisture content, etc., with the rate constant k1 and the equilibrium constant k2, and select a two-sided significance test. If the correlation degree of a certain physicochemical property with k1 or k2 exceeds the set threshold, the physicochemical property is determined as the key physicochemical property affecting the moisture content of the grain soaking process.
[0029] Step S4: The key physical and chemical properties, soaking time and the moisture content corresponding to the soaking time are converted into values in the interval [0,1], and the conversion formula is as follows: X' represents the value after preprocessing, X i represents the value before preprocessing, X min Represents the minimum value in the data, X max Represents the maximum value in the data. The neural network prediction model is trained with the preprocessed parameters. A BP neural network prediction model is constructed, wherein the BP neural network prediction model includes an input layer, an output layer, and a hidden layer. The preprocessed key physical and chemical properties, the soaking time, and the initial moisture content of the grain are used as data inputs of the input layer, and the moisture content corresponding to the preprocessed soaking time is used as data input of the output layer to train the established BP neural network prediction model.
[0030] Step S5: key physical and chemical properties, soaking time and initial moisture content of the grain to be tested are input into the trained BP neural network prediction model, and the BP neural network prediction model outputs the moisture content of the grain to be tested.
[0031] Embodiment: This embodiment takes sorghum as an example to further explain the method for predicting moisture content in a grain soaking process based on a neural network described in the present invention.
[0032] Step S1: In this embodiment, 23 varieties of sorghum were selected to measure fat content, tannin content, protein content, starch content, amylose content, hardness, specific surface area, seed coat thickness, and initial moisture content; wherein, the initial moisture content refers to GB 5009.3-2016; the starch content is measured in accordance with GB 5009.9-2016; the fat content is measured in accordance with GB5009.6-2016; the amylose content is measured in accordance with GB / T 15683-2008; the amylopectin content is calculated by subtracting the amylose content from the starch content; the tannin content is measured in accordance with GB / T 15686-2008; and the protein content is measured in accordance with GB5009.5-2016. The specific surface area Sv of sorghum is measured as follows: 30 grains of sorghum of each variety are selected, the measurement error is set to ±0.02 mm, and the length L, width W and thickness T of each sorghum are measured using a vernier caliper; the surface area of sorghum is estimated respectively. and volume Then calculate the specific surface area based on the surface area S and volume V The seed coat thickness measurement method is as follows: the seed coat thickness (pericarp plus seed coat) of sorghum grains is measured using a scanning electron microscope. The grain sample is first cut longitudinally with a scalpel, and then the middle part of the grain is taken, and the two ends are cut into planes. The sample is attached to an aluminum stake with silver conductive paint and metalized in a gold evaporator using sputtering technology. The image is obtained using a magnification at 15KV. Each sample is fixed and extracted from the side of the middle part of the sorghum for measurement; each variety of sorghum sample uses a different grain for measurement, and the average result is repeated twice. The hardness of dry sorghum grains is measured using a texture analyzer. The diameter of the texture analyzer probe is 10mm, the test speed is 0.5mm / s, and the grain is compressed to 60% of the height. The force of grain rupture is used as the hardness, and the average value of 10 grains of sorghum for each variety is used as the measurement result.
[0033] Step S2: The 23 selected sorghum varieties were soaked at a constant temperature of 40°C with a grain-water ratio of 1:5 (m:m) and the peleg model was used. The hydration dynamics of sorghum was simulated and the moisture content of each variety of sorghum was measured at 0 min, 20 min, 40 min, 60 min, 120 min, 180 min, 240 min, 300 min, 360 min and 420 min, respectively. t represents the moisture content at soaking time t; M0 represents the initial moisture content of unsoaked sorghum; k1 is the rate constant; k2 is the capacity constant.
[0034] Step S3: The physicochemical properties of sorghum and the Plege model parameters k1 and k2 were subjected to bivariate Pearson correlation analysis (Pearson correlation analysis) by software SPSS20.0, and a bilateral significance test was selected. When p < 0.05, * was used to indicate significant correlation, and when p < 0.01, ** was used to indicate extremely significant correlation. One or more physicochemical properties significantly correlated or extremely significantly correlated with k1 or k2 were set as key physicochemical properties affecting the moisture content of sorghum. As shown in the correlation coefficient in the correlation analysis results of the physicochemical properties of sorghum and the Plege model constants k1 and k2 in Table 1, the key physicochemical properties affecting the moisture content of sorghum in this embodiment include fat, tannin, amylose, amylopectin content, hardness and specific surface area; at the same time, since the contents of amylose and amylopectin are negatively correlated, the correlation coefficient is -0.875, and there is a strong linear correlation between the two, so one of amylose and amylopectin is selected as one of the key physicochemical properties, and amylose is selected in this embodiment.
[0035] Table 1 Correlation analysis results between sorghum physicochemical properties and Plege model parameters k1 and k2
[0036]
[0037]
[0038] Step S4: Preprocess the key physicochemical properties and soaking time and moisture content data. Construct a BP neural network, use the sorghum fat, tannin, amylose content, hardness and specific surface area, soaking time, and initial moisture content after data preprocessing as the input layer, and the sorghum moisture content during the soaking process as the output layer, and import the BP neural network, specifically: import the key physicochemical properties of sorghum, soaking time, and moisture content of the corresponding time in the matalab R2021a software, use the trainlm function to perform Levenberg-Marquardt training, create a network through newff, first select 23 different varieties of brewing sorghum data as training samples, use the dividerand function to randomly divide the training sample data into three groups, 70% of the data as the training set, 15% of the data as the validation set, and 15% of the data as the test set. The final number of hidden layer nodes is determined to be 10 by trial and error, and the model is established using the optimal parameters. According to the correlation coefficient R, the training is repeated until the model is optimal, and the final prediction model for the moisture content of the whole grain sorghum soaking process based on the BP neural network is output. The regression results of the training set based on the BP neural network prediction model and the actual value are as follows Figure 2 As shown; the regression results of the validation set based on the BP neural network prediction model and the actual value are as follows Figure 3As shown, the regression results of the test set based on the BP neural network prediction model and the actual value are as follows Figure 4 As shown; the regression results of the training set, validation set and test set based on the BP neural network prediction model and the actual value are shown in Figure 5 shown.
[0039] Step S5: Select 6 sorghum samples belonging to the above 23 varieties but not involved in the construction of the BP neural network model as the samples to be tested, input their key physical and chemical properties, soaking time and initial moisture content into the trained BP neural network prediction model, and the BP neural network prediction model outputs the predicted moisture content. Compared with the actual measurement results, the average error is 4.2403%. After calculation, the correlation coefficient between the test value and the predicted value is 0.990, and the mean square error (MSE) is 2.147, which can realize the prediction of the moisture content of different varieties of brewing sorghum during the soaking process.
Claims
1. A method for predicting moisture content in grain soaking process based on neural network, characterized in that: include: Step S1, obtaining a grain sample and measuring various physical and chemical properties of the grain sample; Step S2, recording the moisture content of the grain corresponding to different soaking times in constant temperature water; Step S3, analyzing the correlation between various physical and chemical properties of grains and the moisture content during the grain soaking process, and determining the key physical and chemical properties that affect the moisture content during the grain soaking process according to the degree of correlation; Step S4, establishing a neural network prediction model, taking the initial moisture content, key physical and chemical properties and soaking time of the grain as inputs of the neural network prediction model, taking the moisture content corresponding to the soaking time as output of the neural network prediction model, and training the neural network prediction model; Step S5: input the initial moisture content, key physical and chemical properties and soaking time of the grain to be tested for moisture content into the trained neural network prediction model, and the neural network prediction model outputs the moisture content corresponding to the soaking time.
2. The method for predicting moisture content in grain soaking process based on neural network according to claim 1, characterized in that: The physicochemical properties described in step 1 specifically include the fat content, tannin content, protein content, starch content, amylose content, amylopectin content, hardness, specific surface area and seed coat thickness of the grain.
3. The method for predicting moisture content in grain soaking process based on neural network according to claim 1, characterized in that: Step 2 specifically includes: using the peleg model The moisture content of simulated grains corresponding to different soaking times in constant temperature water, where t is the soaking time, M t is the moisture content of the grain after soaking for t time, M0 is the initial moisture content of the grain, k1 is the rate constant, and k2 is the capacity constant.
4. The method for predicting moisture content in grain soaking process based on neural network according to claim 3, characterized in that: The analysis of the correlation between the various physicochemical properties of the grain and the moisture content of the grain soaking process described in step 3 specifically includes: performing correlation analysis on the physicochemical properties of the grain with the rate constant k1 and the capacity constant k2 respectively; if the degree of correlation between a certain physicochemical property and k1 or k2 exceeds a set threshold, then the physicochemical property is determined as a key physicochemical property that affects the moisture content of the grain soaking process.
5. The method for predicting moisture content in grain soaking process based on neural network according to claim 4, characterized in that: The physicochemical properties of grain were subjected to bivariate Pearson correlation analysis with the rate constant k1 and the capacity constant k2, and a two-sided significance test was selected. The physicochemical properties with a significant p-value less than the set value were identified as the key physicochemical properties affecting the moisture content in the grain soaking process.
6. The method for predicting moisture content in grain soaking process based on neural network according to claim 1, characterized in that: In step S4, the key physical and chemical properties, soaking time and the moisture content corresponding to the soaking time are converted into values in the interval [0,1] and then input into the established neural network prediction model. The specific conversion method is: X' represents the value after preprocessing, X i is the value before preprocessing, X min is the minimum value in the data, X max is the maximum value in the data.