Real-time control method for drying degree of red yeast rice

Abstract

The invention provides a monascus drying degree real-time control method which comprises the following steps: monitoring the surface moisture content, internal metabolic activity index and pigment content change of mycelium in a monascus drying process, and generating a mycelium state distribution diagram; extracting moisture gradient characteristics and metabolic activity parameters from the mycelium state distribution diagram, analyzing a relationship between a moisture evaporation rate and mycelium activity, and determining a dynamic trend of moisture gradient and activity change; analyzing the dynamic trend of the moisture gradient and the activity change, determining the relationship between the moisture gradient and the activity change through a fitting curve, and predicting the change trend of the moisture and the activity of the mycelium in different drying stages; aiming at a moisture distribution non-uniformity index, the uniform distribution of heat in the red yeast rice is improved by optimizing an air supply mode, so that the moisture evaporation rate deviation is reduced, and a temperature and humidity control sequence is obtained.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for real-time control of the drying degree of red yeast rice. Background Art

[0002] As a traditional fermented food and medicinal ingredient, maintaining microbial activity and active ingredients during the production of red yeast rice (Red Yeast Rice) is crucial to product quality. The drying process directly impacts mycelial activity, pigment stability, and active ingredient retention, making it a key factor in determining the product's biological activity and market value. Studying the correlation between microbial activity and temperature and humidity control during red yeast rice drying not only impacts the modernization of traditional processes but also holds far-reaching significance for the sustainable development of the food and traditional Chinese medicine industries. However, existing drying technologies have significant limitations in controlling temperature and humidity. Traditional methods often employ fixed temperature and humidity parameters, which are difficult to adapt to the complex and variable mycelial state after red yeast rice fermentation. This can lead to uneven drying rates, pigment degradation, or active ingredient loss. While some dynamic control technologies have improved, they lack precise response to the uneven evaporation of water from the mycelial surface caused by temperature fluctuations and fail to fully account for the periodic changes in pigment stability and active ingredient retention. The key challenge lies in balancing drying rate with maintaining mycelial activity while addressing the uneven evaporation and pigment stability issues caused by temperature fluctuations. Temperature fluctuations can lead to differential evaporation rates on the mycelial surface, resulting in localized over-drying or over-wetting, which in turn impacts microbial metabolic activity. Pigment stability varies significantly at different drying stages, and high temperatures or uncontrolled humidity can accelerate degradation. The retention of active ingredients is affected by temperature and humidity gradients, and fluctuations are difficult to predict. These factors collectively present technical challenges in dynamically adjusting temperature and humidity control strategies. Therefore, dynamically adjusting the temperature and humidity gradient during the red yeast rice drying process to balance drying rate with mycelial activity, stabilize pigment content, and enhance active ingredient retention has become a key issue in optimizing product quality and bioactivity. Summary of the Invention

[0003] The present invention provides a method for real-time control of the drying degree of red yeast rice, which mainly comprises:

[0004] Monitor the changes in mycelium surface moisture content, internal metabolic activity indicators and pigment content during the red yeast rice drying process, and generate a mycelium status distribution map;

[0005] Extract moisture gradient characteristics and metabolic activity parameters from the mycelium state distribution map, analyze the relationship between water evaporation rate and mycelium activity, and determine the dynamic trends of moisture gradient and activity changes;

[0006] Analyze the dynamic trend of moisture gradient and activity change, determine the relationship between the two by fitting curves, and predict the changing trend of mycelium moisture and activity at different drying stages;

[0007] The impact of temperature fluctuations on uneven evaporation of red yeast rice is determined based on the dynamic trends of moisture gradient and activity changes. If the temperature fluctuation amplitude exceeds a preset threshold, the deviation of the moisture evaporation rate is predicted to obtain an index of uneven moisture distribution.

[0008] Aiming at the moisture distribution unevenness index, the air supply mode is optimized to improve the uniform distribution of heat inside the red yeast rice, thereby reducing the deviation of the moisture evaporation rate and obtaining the temperature and humidity control sequence;

[0009] Extract phase-specific variation characteristics from the changing trends of mycelium moisture and activity at different drying stages, analyze the impact of temperature and humidity fluctuations at different drying stages on pigment stability, predict the degradation risk of pigments at different drying stages of red yeast rice, and determine the sensitive range of pigment degradation;

[0010] Compare the sensitive interval for pigment degradation and the interval for decreased active ingredient retention to determine whether there is overlap between the two. If there is overlap, it indicates that temperature and humidity fluctuations in this interval may cause both pigment degradation and active ingredient loss.

[0011] In view of the overlapping sensitive areas of pigment degradation and active ingredient loss, temperature and humidity control targets are set at different stages according to the stability requirements of pigments and active ingredients, generating a temperature and humidity control sequence for pigment and ingredient protection.

[0012] A temperature and humidity control sequence was implemented on the drying equipment. The control effect was verified by continuously monitoring the status of Monascus mycelium. Indicators representing the balance between drying rate and biological activity were extracted from the verification results to generate an optimized sequence of temperature and humidity control parameters.

[0013] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0014] The present invention discloses a real-time control method for the drying degree of red yeast rice. The method obtains a mycelium state distribution diagram through multi-point sampling technology, analyzes the relationship between moisture gradient and metabolic activity, and predicts the trend of changes in mycelium moisture and activity during the drying process. According to the influence of temperature fluctuations on the uniformity of water evaporation, the air supply and heating parameters are adjusted. For the sensitive areas of pigment degradation and loss of effective ingredients, a segmented temperature and humidity control strategy is adopted to set control targets for different stages. Finally, the parameters of the drying equipment are adjusted in real time through an automated control system, and a multi-objective optimization algorithm is used to balance the drying rate and biological activity to generate an optimal control sequence. The present invention can effectively improve the retention rate of pigments and effective ingredients in the red yeast rice drying process, while ensuring the drying efficiency, thereby achieving dual optimization of red yeast rice quality and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention is a flow chart of a method for real-time control of the drying degree of red yeast rice. DETAILED DESCRIPTION

[0016] The following will describe the technical solutions in the embodiments of the present invention in detail with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of the present invention.

[0017] like Figure 1 In this embodiment, a method for real-time control of the dryness of red yeast rice may specifically include:

[0018] S101. Monitor the changes in mycelium surface moisture content, internal metabolic activity indicators and pigment content during the red yeast rice drying process, and generate a mycelium status distribution map.

[0019] An initial moisture content dataset is obtained from preset sampling points on the mycelium surface through a humidity sensor array. The initial dataset is collected by a metabolic activity sensor and a spectrophotometer to obtain a point metabolic activity dataset and a pigment content dataset. The initial dataset is converted into interval values ​​using a maximum-minimum value normalization method, and moisture distribution continuous surface data are generated based on the moisture gradient value between adjacent sampling points and an inverse distance weighted interpolation algorithm. The moisture content of the sampling points is determined based on the moisture distribution continuous surface data. If the moisture content of the sampling point is lower than a preset threshold, the metabolic activity dataset and pigment content dataset of the point with moisture content lower than the preset threshold are obtained. A Bayesian network model is constructed, and the sampling point data are calculated using a Kriging spatial interpolation algorithm to obtain a mycelium state distribution map.

[0020] Specifically, moisture content values ​​are obtained from preset sampling points on the mycelium surface based on a humidity sensor array, metabolic activity signals inside the mycelium are collected through a metabolic activity sensor, and pigment content values ​​are obtained at the sampling points using a spectrophotometer to generate an initial moisture distribution dataset, a metabolic activity dataset, and a pigment content dataset. For the initial moisture distribution dataset, the maximum and minimum value normalization method is used to convert the moisture content values ​​to the 0-1 range, the moisture gradient values ​​between adjacent sampling points are calculated, and the inverse distance weighted interpolation algorithm is used to generate moisture distribution continuous surface data. Based on the moisture distribution continuous surface data, a moisture content monitoring threshold is set. When the moisture content of any sampling point is detected to be lower than the threshold, the metabolic activity dataset and pigment content dataset of the point are obtained. A Bayesian network model is constructed using the metabolic activity dataset and pigment content dataset. The model input parameters include moisture content, sampling time, and spatial position coordinates, and the output parameters include metabolic activity prediction value and pigment content prediction value. Based on the output of the Bayesian network model, a kriging spatial interpolation algorithm was used to interpolate the metabolic activity and pigment content between sampling points, generating spatial distribution data for metabolic activity and pigment content. A three-dimensional scatter plot was used to construct a mycelial state distribution map based on the moisture distribution continuous surface data, metabolic activity spatial distribution data, and pigment content spatial distribution data. Mycelial moisture content was monitored using a capacitive humidity sensor array. Sampling points were spaced 10 mm apart on the surface of the mycelial culture dish. The sensor probes were positioned perpendicular to the mycelial surface, with a 0.5 mm gap between the probe tips and the mycelial surface. Moisture content values ​​obtained by the sensor array ranged from 15% to 85%. Metabolic activity was detected using a fluorescence intensity sensor, measuring fluorescent metabolites produced within the mycelium. Fluorescence intensity values ​​ranged from 200 to 800 relative fluorescence units. Pigment content was detected using a spectrophotometer with a wavelength set between 460 and 660 nanometers, obtaining absorbance values ​​ranging from 0.2 to 2.0. Normalization preprocessing was performed on the initial collected data. Taking the moisture content data as an example, when the maximum value in the original data was 85% and the minimum value was 15%, the standardized data was obtained by maximum and minimum normalization calculation. The moisture gradient between adjacent sampling points was calculated using the central difference method to obtain the gradient components in the horizontal and vertical directions. The inverse distance weighted algorithm was used to interpolate the area between the sampling points. The interpolation grid density was set to 2 mm, the weight index was 2, and the search radius was 20 mm. In the drying process monitoring, when the moisture content of a sampling point was detected to be lower than the 25% threshold, the metabolic activity data and pigment content data of the point and its surrounding 8 points were extracted to establish a Bayesian network model. The Bayesian network model nodes include moisture content nodes, metabolic activity nodes, and pigment content nodes. The conditional probabilities between nodes are obtained through historical data training.Ordinary kriging was used as the spatial interpolation algorithm, and a spherical model was used as the variogram model. The sill value was determined by the variance of the sampled data, the range was set to 30 mm based on the sampling point spacing, and the nugget value was obtained experimentally. The interpolation grid was divided into a 100×100 uniform grid, and the predicted metabolic activity and pigment content values ​​were calculated at each grid point. The three-dimensional scatter plot used different colors to mark the moisture content, metabolic activity, and pigment content values, and the size of the scatter plots was positively correlated with the measured values. The coordinate axes used actual physical units: the X and Y axes represent the sampling plane coordinates in millimeters, and the Z axis represents the measured values. The map resolution was set to 1920×1080 pixels, the viewing angle was 45 degrees from above, and a color scale was added to indicate the value range. Rotation and zooming operations allowed for observation and analysis of distribution characteristics in different regions. A mycelial state distribution map was generated to visually demonstrate the spatial distribution patterns of changes in moisture content, metabolic activity, and pigment content during the drying process of Monascus mycelium.

[0021] S102. Extract moisture gradient characteristics and metabolic activity parameters from the mycelium state distribution map, analyze the relationship between water evaporation rate and mycelium activity, and determine the dynamic trend of moisture gradient and activity changes.

[0022] According to the moisture content values ​​of the sampling points in the mycelium state distribution map, the Sobel difference operator is used to calculate the moisture gradient values ​​in the horizontal and vertical directions, and the moisture gradient direction characteristic map and the moisture gradient amplitude distribution map are obtained through vector field processing; for the original metabolic activity data, the metabolic activity change data are smoothed using a third-order autoregressive filter to obtain a metabolic activity time series curve; the moisture gradient differences at adjacent time points are extracted from the moisture gradient direction characteristic map and the moisture gradient amplitude distribution map, and the water evaporation rate data are obtained by calculation; after obtaining the water evaporation rate data and the metabolic activity time series curve, the random forest regression method is used to construct a mapping relationship between water evaporation and metabolic activity, the input features of the mapping relationship include moisture gradient amplitude and water evaporation rate data, and the output features of the mapping relationship include metabolic activity change rate and metabolic activity decay curve.

[0023] Specifically, based on the moisture content value of each sampling point in the mycelium state distribution map, the Sobel difference operator is used to calculate the moisture gradient values ​​in the horizontal and vertical directions, and the moisture gradient direction characteristic map and moisture gradient amplitude distribution map are obtained through vector field processing. For the original metabolic activity data of each sampling time point, the sliding window size is set to 5 time points and the sliding step size is set to 1 time point, and the metabolic activity change data over time is calculated. The metabolic activity change data is smoothed using a third-order autoregressive filter to obtain a metabolic activity time series curve. The moisture gradient difference between adjacent time points is extracted from the moisture gradient direction characteristic map and the moisture gradient amplitude distribution map, and the water evaporation rate data per unit time is calculated. Based on the metabolic activity time series curve, the metabolic activity change rate of adjacent time points is calculated, and the metabolic activity decay curve is obtained using an exponential weighting method. The random forest regression method is used to construct a mapping relationship between water evaporation and metabolic activity. The input features include the moisture gradient amplitude and water evaporation rate data, and the output features include the metabolic activity change rate and the metabolic activity decay curve. Dynamic prediction of water evaporation rate and metabolic activity changes is performed using a Kalman filter algorithm. The state variables include water gradient values ​​and metabolic activity values, while the observed variables include measured water content and metabolic activity data. Dynamic predictions of water gradient and activity changes are generated through iterative updating. The Sobel operator is used to extract water gradient features using a differential operation. A 3×3 pixel sliding window is set on the mycelial state distribution map, and convolution calculations are performed in both the horizontal and vertical directions to obtain the rate of change of water content in both directions. As the water content on the mycelial surface decreases from the edge to the center, the gradient direction points in the direction of decreasing water content. The gradient amplitude reflects the severity of the water content change, with areas of larger gradient amplitude often corresponding to locations of rapid water loss. For metabolic activity data processing, the sliding window is set based on the characteristics of the mycelial growth cycle. A window size of five time points is selected to smooth short-term fluctuations while preserving medium-term trends. A sliding step of one time point ensures data continuity. A third-order autoregressive filter eliminates random fluctuations in metabolic activity measurements by weighted superposition of the three most recent historical data points. The filtered metabolic activity time series curve better reflects the true dynamics of mycelial activity. Water evaporation rate is calculated by comparing the difference in water gradients between adjacent time points. When the ambient temperature is maintained at 25°C and the relative humidity is 60%, the evaporation rate in the peripheral region is approximately 0.8% per hour, and in the central region is approximately 0.3% per hour. The metabolic activity decay curve uses an exponential weighting method, assigning greater weight to recent activity changes, reflecting the nonlinear nature of activity decay.A random forest regression method established a quantitative relationship between water evaporation and metabolic activity. The water gradient amplitude, an input feature, reflects the degree of local water unevenness, while the water evaporation rate characterizes the external environmental driving force. These features show a clear correlation with the rate of change and decay curve of metabolic activity. When the water gradient amplitude exceeds 2% per millimeter, the metabolic activity decay rate increases significantly, indicating that drastic water loss adversely affects mycelial activity. A Kalman filter algorithm dynamically predicts changes in water gradient and metabolic activity. The state variables include the current water gradient and metabolic activity values, while the observation variables are derived from actual measured data. The filter's prediction step is based on an established water-activity mapping, while the update step corrects the prediction based on the observed values. When the error between the predicted and observed values ​​is large, the filter increases its confidence in the observed values, thereby improving prediction accuracy. The water gradient and activity trends predicted by the filter reflect the dynamic response characteristics of mycelium at different drying stages.

[0024] S103. Analyze the dynamic trends of moisture gradient and activity changes, determine the relationship between the two by fitting curves, and predict the changing trends of mycelium moisture and activity at different drying stages.

[0025] A cubic polynomial function is used to fit the moisture gradient data to obtain a moisture gradient trend function, and a negative exponential function is used to fit the activity change data to obtain an activity trend function; the correlation coefficient within a fixed time interval is calculated for the moisture gradient trend function and the activity trend function to obtain a water activity correlation curve; the K-means clustering method is used to group the correlation coefficients in the water activity correlation curve to obtain a drying stage time interval; for the data within the drying stage time interval, a support vector regression method with a radial basis kernel function is used to establish a water activity mapping function, wherein the input variable of the mapping function is the moisture gradient value, and the output variable is the activity change value; based on the water activity mapping function, an iterative prediction is performed with a fixed step size within each drying stage interval to obtain the dynamic development trend of the moisture gradient and activity change.

[0026] Specifically, based on the time series data of moisture gradient and activity change, a cubic polynomial function was fitted to the moisture gradient data to obtain a moisture gradient trend function, and a negative exponential function was fitted to the activity change data to obtain an activity trend function. For the moisture gradient trend function and the activity trend function, fixed time intervals were set to integer multiples of the sampling period. The correlation coefficients of the two trend functions within each time interval were calculated to generate a water activity correlation curve. Based on the water activity correlation curve, the correlation coefficients were grouped using the K-means clustering method. The clustering results were used to divide the drying stage time intervals, and the start and end time points of the different drying stages were determined. For the data within each drying stage time interval, a water activity mapping function was constructed using support vector regression with a radial basis kernel function. The input variable of this mapping function was the moisture gradient value, and the output variable was the activity change value. Based on the water activity mapping function, an error threshold range was set. When the prediction error exceeded the threshold range, the mapping function parameters were updated using the gradient descent method to obtain a new water activity mapping function. Using this updated water activity mapping function, iterative predictions were performed with a fixed step size within each drying stage interval to determine the dynamic development trends of moisture gradient and activity change. The moisture gradient during mycelial drying exhibits nonlinear characteristics. Fitting a cubic polynomial function accurately describes the spatial distribution of moisture. Typical moisture gradient values ​​decrease from the edge to the center, with values ​​of approximately 0.8 per millimeter at the edge and 0.2 per millimeter at the center. Activity changes exhibit exponential decay, with a negative exponential function reflecting a continuous decline in activity over time. Initial activity was set at 1.0, and decreased to approximately 0.6 when the moisture content dropped to 25%. Correlation analysis used a fixed 15-minute time interval, calculating the correlation coefficient between the moisture gradient trend function and the activity trend function. The correlation coefficient ranged from -1 to 1, with positive values ​​indicating an increase in activity with increasing moisture gradients and negative values ​​indicating a decrease in activity with increasing moisture gradients. Initially, the correlation coefficient remained above 0.8, indicating a strong correlation between moisture gradient and activity. As drying progressed, the correlation coefficient gradually decreased to approximately 0.3. K-means clustering was used to group the correlation coefficients, with a set number of clusters of three, corresponding to the rapid drying, transitional drying, and slow drying stages. The cluster center for the rapid drying phase was approximately 0.85, lasting from 0 to 2 hours; the cluster center for the transitional drying phase was approximately 0.6, lasting from 2 to 4 hours; and the cluster center for the slow drying phase was approximately 0.3, lasting from 4 hours onwards. Support vector regression used a radial basis kernel function to establish the water activity mapping relationship. The kernel function parameters were selected based on the data distribution characteristics, and mapping functions were constructed for different drying stages.When the prediction error exceeds the set threshold of 0.1, the kernel function parameters and weight coefficients are adjusted by the gradient descent method, the learning rate is set to 0.01, and the number of iterations does not exceed 100. During the parameter update process, the prediction error gradually decreases from the initial 0.15 to below 0.08. Based on the updated mapping function, 5 minutes is used as the prediction step size for iterative calculation in each drying stage. In the rapid drying stage, the moisture gradient prediction value drops rapidly from 0.8 per millimeter to 0.5 per millimeter, and the activity prediction value drops from 1.0 to 0.8; in the transition drying stage, the moisture gradient drops slowly to 0.3 per millimeter, and the activity drops to 0.6; in the slow drying stage, the moisture gradient remains at around 0.2 per millimeter, and the activity continues to decrease slowly. The prediction results reflect the changing characteristics of moisture and activity in different drying stages, and the goodness of fit between the prediction curve and the measured data reaches 0.92.

[0027] S104: Determine the impact of temperature fluctuations on uneven evaporation of red yeast rice based on the dynamic trends of moisture gradient and activity changes. If the temperature fluctuation exceeds a preset threshold, predict the deviation in the evaporation rate to obtain a moisture distribution unevenness index.

[0028] According to the moment when the temperature fluctuation amplitude exceeds the preset temperature threshold, the actual value of the water evaporation rate is obtained, and the functional relationship between the temperature fluctuation amplitude and the evaporation rate deviation is established using a quadratic polynomial regression equation; based on the functional relationship between the temperature fluctuation amplitude and the evaporation rate deviation, a time series prediction function is established using a long short-term memory network, and the input layer nodes of the time series prediction function include the temperature fluctuation amplitude, the fluctuation frequency, and the moisture gradient value, and the output layer nodes are the evaporation rate prediction values; based on the evaporation rate prediction values ​​output by the time series prediction function, the evaporation rate difference between adjacent measuring points on the red curved surface is calculated, and the moisture distribution variation coefficient is obtained by using the ratio of the standard deviation to the mean value, and the moisture distribution variation coefficient is used as an indicator of moisture distribution unevenness.

[0029] Specifically, a Fourier transform method is used to perform spectrum analysis on time series data collected by the temperature sensor. The fluctuation amplitude and frequency of the temperature data are calculated to determine whether the fluctuation amplitude exceeds a preset temperature threshold. When the temperature fluctuation amplitude exceeds the preset temperature threshold, the measured water evaporation rate at the corresponding moment is extracted, and a functional relationship between the temperature fluctuation amplitude and the evaporation rate deviation is established based on a quadratic polynomial regression equation. Using this functional relationship, the theoretical water evaporation rate at each measuring point on the red yeast rice surface is calculated, and the evaporation rate deviation distribution caused by temperature fluctuations is obtained. Based on this evaporation rate deviation distribution, a time series prediction function is established using a long-short-term memory network. The input layer nodes contain the temperature fluctuation amplitude, fluctuation frequency, and moisture gradient value. The number of hidden layer nodes is set to twice that of the input layer. The output layer nodes contain the predicted evaporation rate values. Based on the predicted evaporation rate values, the evaporation rate differences between adjacent measuring points on the red yeast rice surface are calculated. The coefficient of variation of the moisture distribution is calculated as the ratio of the standard deviation to the mean. The coefficient of variation of the moisture distribution is used to establish a quantitative indicator for the uneven distribution of moisture on the red yeast rice surface. The coefficient of variation value is a quantitative representation of the degree of moisture unevenness. The temperature sensor monitors the red yeast rice fermentation environment with a 10-second sampling period. The acquired temperature time-series data is decomposed into frequency fluctuation components using a Fourier transform, with the primary fluctuation frequency concentrated in the 0.01 Hz to 0.1 Hz range. A preset temperature threshold of 2 degrees Celsius is set for determining fluctuation amplitude. Temperature fluctuations exceeding this threshold indicate significant fluctuations in the fermentation environment. When the temperature fluctuation amplitude reaches 2.5 degrees Celsius, the measured water evaporation rate exhibits significant fluctuations. A quadratic polynomial regression fit to the relationship between temperature fluctuation and evaporation rate reveals an upward-opening parabolic curve. When the temperature fluctuation amplitude reaches 3 degrees Celsius, the evaporation rate deviation reaches 0.15 grams per square centimeter per hour, a 25% increase compared to stable temperature conditions. Sixteen measurement points are placed on the red yeast rice surface, arranged in a 4×4 matrix with 20 mm spacing between points. The theoretical evaporation rate is calculated at each measurement point based on the distribution of evaporation rate deviations caused by temperature fluctuations. The theoretical evaporation rate is 0.8 grams per square centimeter per hour in the center, rising to 1.2 grams per square centimeter per hour at the edges. The spatial distribution of evaporation rate shows an increasing trend from the center to the edge. The long-short-term memory network adopts a three-layer structure. The three nodes in the input layer receive the temperature fluctuation amplitude, fluctuation frequency, and moisture gradient value respectively. The hidden layer has six nodes, and the output layer has one node that outputs the predicted evaporation rate. The network training uses 30 minutes of historical data to predict the evaporation rate change in the next 5 minutes. The average error of the trained network in evaporation rate prediction is controlled within 0.05 grams per square centimeter per hour. The difference in evaporation rate between adjacent measuring points reflects the unevenness of moisture distribution. The coefficient of variation of moisture distribution is obtained by calculating the standard deviation of the evaporation rate of 16 measuring points and dividing it by the average evaporation rate.When the ambient temperature is stable, the coefficient of variation remains below 0.1, indicating relatively uniform moisture distribution. When temperature fluctuations occur, the coefficient of variation increases to 0.2 to 0.3, indicating increased moisture inequality. The coefficient of variation is used as an indicator of moisture inequality, which comprehensively reflects the degree of dispersion of moisture distribution on the red yeast rice surface. Experimental data shows that for every 0.1 increase in the coefficient of variation, the uniformity of the pigment content of the red yeast rice product decreases by 5% to 8%. By real-time monitoring of the moisture inequality indicator, a quantitative evaluation standard for the impact of temperature fluctuations on red yeast rice quality has been established. This indicator is significantly correlated with product quality parameters such as pigment content and moisture content, with a correlation coefficient exceeding 0.85.

[0030] S105. Aiming at the moisture distribution unevenness index, the air supply mode is optimized to improve the uniform distribution of heat inside the red yeast rice, so as to reduce the deviation of the moisture evaporation rate and obtain the temperature and humidity control sequence.

[0031] According to the moisture distribution unevenness index, a fuzzy control method is used to obtain the wind speed adjustment amount and the heating power adjustment amount, and the adjustment amount is obtained by mapping the membership function of the temperature deviation and the humidity deviation; for the wind speed adjustment amount and the heating power adjustment amount, the temperature field distribution map of the red yeast material layer is solved by the computational fluid dynamics equation, and the temperature field distribution map is used as the input data for the heat transfer calculation; a finite element meshing method is used to establish a heat transfer calculation model, and the heat distribution uniformity value is calculated by the temperature field distribution map, and the grid unit size is set according to the thickness of the red yeast material layer; according to the heat distribution uniformity value, a recursive neural network structure is used to calculate the water evaporation rate prediction value, and the adjustment sequence of the air supply speed and the heating power is obtained by the dynamic programming state transfer equation, and the temperature setting value and the humidity setting value of the fixed sampling period are generated to obtain the temperature and humidity control sequence.

[0032] Specifically, according to the moisture distribution unevenness index, a fuzzy control method is used to calculate the air supply speed adjustment value. The control rule is established based on the temperature deviation and humidity deviation, and the wind speed adjustment value and the heating power adjustment value are generated by membership function mapping. Using the wind speed adjustment value and the heating power adjustment value, the boundary conditions of the hot air flow field are set, and the temperature field distribution diagram inside the red koji material layer is solved by the computational fluid dynamics equation. For the temperature field distribution diagram, a heat transfer calculation model is established by the finite element meshing method. The grid unit size is set to one tenth of the thickness of the red koji material layer, and the heat distribution uniformity value is calculated. A recursive neural network structure is used to establish an air supply mode optimization function. The input layer contains three nodes: the air supply speed value, the heating power value, and the heat distribution uniformity value. The hidden layer uses a long short-term memory unit, and the output layer is the water evaporation rate prediction value. According to the water evaporation rate prediction value, a dynamic programming state transition equation is constructed. With the moisture distribution uniformity as the objective function, the optimal adjustment sequence of the air supply speed and heating power is calculated. For the optimal adjustment sequence, the piecewise linear interpolation method is used to generate the temperature setting value and humidity setting value of the fixed sampling period to obtain the temperature and humidity control sequence data. The temperature deviation in the fuzzy control method is divided into five levels: negative large, negative small, zero, positive small, and positive large. The deviation values ​​correspond to -2 degrees Celsius, -1 degrees Celsius, 0 degrees Celsius, 1 degree Celsius, and 2 degrees Celsius respectively. The humidity deviation is also divided into five levels, with a numerical range of -10% to 10% relative humidity. The wind speed adjustment value and the heating power adjustment value are obtained by mapping the trapezoidal membership function. The wind speed adjustment range is -0.5 meters per second to 0.5 meters per second, and the heating power adjustment range is -200 watts to 200 watts. The hot air flow field boundary conditions of the red koji material layer include air inlet temperature, air inlet wind speed and air outlet pressure. The air inlet temperature is set between 25 degrees Celsius and 35 degrees Celsius, the air inlet wind speed range is 1 meter per second to 2 meters per second, and the air outlet is set to atmospheric pressure. The temperature field distribution obtained by solving computational fluid dynamics equations reflects the temperature gradient within the material layer, with a spatial resolution of 5 mm. Finite element meshing uses hexahedral elements with a cell size of 10 mm. For a red yeast rice layer thickness of 100 mm, a total of 10 mesh layers are created. Heat transfer calculations consider three heat transfer modes: conduction, convection, and radiation. Material thermophysical properties vary with moisture content, with the thermal conductivity coefficient varying between 0.2 and 0.4 watts per meter Kelvin. Heat distribution uniformity is determined by calculating the standard deviation of the temperature field. A recursive neural network employs a three-layer long-short-term memory (LSTM) structure. The input layer has three nodes corresponding to air velocity, heating power, and heat uniformity. The hidden layer has 20 memory cells, and the output layer has one node outputting the water evaporation rate. The network is trained using four hours of historical data, with the validation set using the most recent hour of data. The prediction error of the trained network is kept within 0.05 grams per square centimeter per hour.The dynamic programming optimization process divides the air supply speed and heating power control sequence into 12 time periods. The state variables for each period include wind speed levels (five wind speed levels) and power levels (eight power levels). Using moisture distribution uniformity as the objective function, the optimal control path is calculated using the state transition equation. The resulting control sequence exhibits a step-like characteristic. The sampling period for the temperature and humidity control sequence is set to 1 minute, and the step-like control sequence is converted into a continuously changing control curve using piecewise linear interpolation. The temperature setpoints vary between 28°C and 32°C, and the relative humidity setpoints vary between 55% and 65%. The control sequence data records the temperature and humidity setpoints over a 4-hour drying cycle, with 240 data points. By implementing this control sequence, the moisture distribution uniformity of the red yeast rice layer was significantly improved, with the coefficient of variation reduced from 0.25 to 0.12.

[0033] S106. Extract the phased change characteristics from the changing trends of mycelium moisture and activity at different drying stages, analyze the impact of temperature and humidity fluctuations at different drying stages on pigment stability, predict the degradation risk of pigments at different drying stages of red yeast rice, and determine the sensitive range of pigment degradation.

[0034] The Dobesi wavelet basis function is used to perform wavelet transform decomposition on the time series data of mycelium moisture content and activity, and the drying stage division result is obtained by the peak and trough recognition method; the characteristic parameters of temperature fluctuation and humidity fluctuation are extracted based on the drying stage division result, and the significant characteristic indicators of temperature and humidity fluctuation are obtained by the variance analysis method; a multiple regression equation is established for the significant characteristic indicators of temperature and humidity fluctuation, and the influence function of temperature and humidity fluctuation on pigment stability is obtained from the mycelium moisture content and activity data; a forward neural network predictor is constructed based on the influence function, and the pigment degradation rate is obtained from the temperature and humidity fluctuation amplitude, moisture content and activity value. The pigment degradation rate is divided into sensitive intervals using the K-means clustering method to obtain the pigment degradation sensitive time interval.

[0035] Specifically, the wavelet transform method was used to decompose the time series data of mycelial moisture content and activity. Dobesi wavelet basis functions were selected, and the decomposition scale was set to four levels. Peak and trough identification was used to identify key time nodes and divide the drying stages. Based on the drying stage division results, characteristic parameters of temperature and humidity fluctuations were extracted for each stage. The significance level of the fluctuation amplitude was determined using analysis of variance to obtain significant characteristic indicators of temperature and humidity fluctuations. A multiple regression equation was established based on the mycelial moisture content and activity data to calculate the rate of change of pigment content and obtain the impact function of temperature and humidity fluctuations on pigment stability. A forward neural network was used to construct a pigment degradation predictor. The input layer had four nodes corresponding to temperature fluctuation amplitude, humidity fluctuation amplitude, moisture content, and activity value, the hidden layer had eight nodes, and the output layer had one node outputting the pigment degradation rate. Using this pigment degradation rate, a threshold segmentation method based on maximum inter-class variance was established to divide the drying process into stable and sensitive regions, thereby obtaining a risk indicator for pigment degradation. Based on the pigment degradation risk indicator, the K-means clustering method was used to finely delineate sensitive areas, determine the start and end points of the sensitive intervals, and generate pigment degradation-sensitive time intervals. The Dobesi wavelet basis function performed a four-level decomposition of the time series data of mycelial moisture content and activity, generating detail coefficients and approximation coefficients at each level. By analyzing the peaks and troughs of the detail coefficients, key time points of moisture content and activity changes were identified, and the drying process was divided into rapid dehydration, slow dehydration, and balanced dehydration stages. The typical rapid dehydration stage lasts 0 to 2 hours, with the moisture content decreasing from 85% to 40%; the slow dehydration stage lasts 2 to 4 hours, with the moisture content decreasing to 25%; and the balanced dehydration stage lasts for more than 4 hours, with the moisture content maintained at around 20%. The characteristic parameters of temperature and humidity fluctuations include the temperature fluctuation amplitude and the humidity fluctuation frequency, and the significance level was calculated using analysis of variance. When the temperature fluctuation exceeded 2 degrees Celsius, the significance level was less than 0.05, indicating that temperature fluctuation had a significant impact on pigment stability. When the humidity fluctuation exceeded 5%, the significance level was also less than 0.05. The impact of temperature fluctuation was more pronounced during the rapid dehydration phase, while the influence of humidity fluctuation gradually increased during the slow dehydration phase. A multiple regression equation established a quantitative relationship between temperature and humidity fluctuations and changes in pigment content, with the regression coefficients reflecting the degree of influence of different factors. For every 1 degree Celsius increase in temperature fluctuation, the rate of pigment content decline increased by 0.2% per hour; for every 1% increase in relative humidity fluctuation, the rate of pigment content decline increased by 0.1% per hour. The goodness-of-fit of the regression equation reached 0.85, indicating that the model has good explanatory power. The feed-forward neural network adopts a 4-8-1 structure. The input layer receives four parameters: temperature fluctuation, humidity fluctuation, moisture content, and activity value. These parameters are standardized before input into the network.The hidden layer uses a hyperbolic tangent activation function, with eight neurons performing feature extraction and nonlinear mapping. The output layer uses a linear activation function, directly outputting the predicted pigment degradation rate. Network training uses a four-hour sample from historical data, while the validation set uses the most recent one-hour data. A threshold segmentation method based on the maximum inter-class variance is used to classify pigment degradation rates, with a threshold of 0.3% per hour. When the degradation rate exceeds this threshold, it is identified as a sensitive region; when it falls below this threshold, it is identified as a stable region. Segmentation results show that the latter half of the rapid dehydration phase and the early stages of the slow dehydration phase are typical sensitive regions. K-means clustering is used to fine-tune the sensitive regions, with the number of clusters set to three, corresponding to high, medium, and low risk levels. Cluster analysis results indicate that the time period with a moisture content of 35% to 45% and a temperature fluctuation exceeding 2°C is classified as a high-risk sensitive region; the time period with a moisture content of 25% to 35% and a temperature fluctuation between 1 and 2°C is classified as a medium-risk sensitive region; and the remaining time periods are classified as low-risk sensitive regions. The starting and ending boundary points of the sensitive interval are determined by the cluster center, providing an accurate risk control basis for the drying process.

[0036] S107. Compare the sensitive interval of pigment degradation and the interval of decreased effective ingredient retention rate to determine whether there is overlap between the two. If there is overlap, it indicates that the temperature and humidity fluctuations in this interval may cause pigment degradation and effective ingredient loss at the same time.

[0037] Fast Fourier transform is used to perform spectral analysis on the time series data of the pigment degradation sensitive interval, and a correlation curve between the pigment content and the effective ingredient content is obtained according to the temperature fluctuation frequency distribution and the humidity fluctuation frequency distribution; the boundary of the sensitive interval is determined according to the correlation curve using the Otsu method adaptive threshold method, and the overlapping time period of the pigment degradation interval and the effective ingredient loss interval is obtained through intersection operation; the root mean square value of the temperature fluctuation amplitude and the root mean square value of the humidity fluctuation amplitude are calculated for the overlapping time period, and a mapping relationship is established using a radial basis kernel function support vector machine, the input variables include the temperature fluctuation root mean square value and the humidity fluctuation root mean square value, and the output variables include the pigment degradation rate and the effective ingredient loss rate; the degradation rate threshold and the loss rate threshold are set according to the output results of the mapping relationship, the temperature and humidity fluctuation data are grouped using the K-means clustering method, and the position of the sensitive interval boundary point is determined by inter-class distance calculation.

[0038] Specifically, a fast Fourier transform (FFT) was used to perform spectral analysis on the time series data of the pigment degradation sensitive interval. The frequency distribution of temperature and humidity fluctuations was calculated, and the Pearson correlation coefficient was used to calculate the correlation curve between pigment content and active ingredient content. Based on this correlation curve, the Otsu method adaptive threshold method was used to determine the boundaries of the sensitive interval. An intersection operation was used to determine the overlapping time periods of the pigment degradation interval and the active ingredient loss interval. For these overlapping time periods, the root mean square (RMS) values ​​of the temperature fluctuation amplitude and the RMS values ​​of the humidity fluctuation amplitude were calculated to generate a characteristic vector of the temperature and humidity fluctuations. Using this temperature and humidity fluctuation characteristic vector, a radial basis kernel support vector machine (RBFK) was used to establish a mapping relationship. The input variables included the RMS values ​​of the temperature fluctuation and the RMS values ​​of the humidity fluctuation, and the output variables included the pigment degradation rate and the active ingredient loss rate. Based on the output of this mapping relationship, thresholds for the pigment degradation rate and the active ingredient loss rate were set, and dual sensitive intervals were determined through threshold judgment. For these dual sensitive intervals, the temperature and humidity fluctuation data were grouped using the K-means clustering method, and the boundary points of the sensitive intervals were determined based on inter-cluster distance calculations. Fast Fourier transform (FFT) analysis of temperature and humidity data during the pigment degradation-sensitive period was performed, with a sampling frequency set to once per minute. A 4-hour data series was transformed. The dominant frequency component of temperature fluctuations was concentrated between 0.005 Hz and 0.02 Hz, indicating a temperature variation period of 50 to 200 minutes. The dominant frequency component of humidity fluctuations was distributed between 0.01 Hz and 0.03 Hz, corresponding to a variation period of 33 to 100 minutes. Pearson correlation coefficient calculations showed that the correlation between pigment content and active ingredient content remained above 0.9 during the initial drying phase, gradually decreasing to around 0.7 as the drying process progressed. The correlation curve was segmented using the Otsu method using an adaptive threshold. The point with the largest inter-class variance was selected as the threshold, dividing the drying process into sensitive and non-sensitive periods. The pigment degradation-sensitive period occurred primarily between 35% and 45% moisture content, while the active ingredient loss-sensitive period occurred between 30% and 40%. The overlapping period between the two periods, corresponding to the 35% to 40% moisture content, lasted approximately 45 minutes. During the overlapping time period, the RMS value of the temperature fluctuation amplitude reached 1.8 degrees Celsius, and the RMS value of the humidity fluctuation amplitude reached 4.2%. These two eigenvalues ​​constitute the temperature and humidity fluctuation eigenvector, reflecting the comprehensive performance of the fluctuation intensity. The radial basis kernel support vector machine uses a Gaussian kernel function, with the kernel parameter determined to be 0.5 through cross-validation, and the penalty factor set to 100. The training data contains 200 sample points, and the validation set accounts for 20% of the total sample. The support vector machine prediction results show that when the RMS value of the temperature fluctuation exceeds 1.5 degrees Celsius and the RMS value of the humidity fluctuation exceeds 3.5%, the pigment degradation rate reaches 0.3% per hour and the active ingredient loss rate reaches 0.2% per hour. These two rate values ​​are set as the dual sensitivity judgment thresholds to determine the range of the dual sensitivity interval.This range is characterized by large fluctuations in temperature and humidity, which simultaneously cause significant changes in pigments and active ingredients. The K-means clustering method grouped the temperature and humidity data within the dual-sensitivity range into three categories: high-risk, medium-risk, and low-risk areas. Cluster analysis results showed that areas with RMS temperature fluctuations greater than 2 degrees Celsius and RMS humidity fluctuations greater than 5% were classified as high-risk; areas with RMS temperature fluctuations between 1.5 and 2 degrees Celsius and RMS humidity fluctuations between 3.5% and 5% were classified as medium-risk; and the remaining areas were classified as low-risk. The location of boundary points was determined by calculating the Euclidean distance between clusters, providing a quantitative basis for the precise delineation of the dual-sensitivity range.

[0039] S108. For the overlapping sensitive areas of pigment degradation and active ingredient loss, set temperature and humidity control targets for different stages according to the stability requirements of pigments and active ingredients, and generate a temperature and humidity control sequence for pigment and ingredient protection.

[0040] Temperature constraint equations and humidity constraint equations are established based on the pigment degradation rate and the effective ingredient loss rate, and the temperature and humidity target values ​​are calculated using piecewise cubic spline interpolation. A convolutional neural network controller is constructed for the temperature and humidity target values, and the temperature and humidity fluctuation characteristics are extracted through the hidden layer of the convolutional neural network controller, and the fluctuation suppression parameters are obtained from the output layer of the convolutional neural network controller. A fuzzy control rule base is established using the fluctuation suppression parameters, and the temperature and humidity adjustment increments are calculated through the membership function of the fuzzy control rule base. Multi-scale wavelet decomposition is performed on the temperature and humidity signals according to the temperature and humidity adjustment increments, and soft threshold denoising is performed on the detail components obtained by the wavelet decomposition, and a smooth temperature and humidity control sequence is obtained through wavelet reconstruction.

[0041] Specifically, based on the pigment degradation rate and active ingredient loss rate in overlapping sensitive intervals, temperature and humidity constraint equations are established. Piecewise cubic spline interpolation is used to calculate the target temperature and humidity values ​​for each stage, generating a segmented temperature and humidity control benchmark sequence. A convolutional neural network controller is constructed for this segmented temperature and humidity control benchmark sequence. The input layer receives the temperature and humidity values, along with the fluctuation amplitude; the hidden layer extracts the temperature and humidity fluctuation characteristics; and the output layer generates fluctuation suppression parameters. Based on these fluctuation suppression parameters, a fuzzy control rule base is established. Temperature and humidity deviations are classified into five levels, and temperature and humidity adjustment increments are calculated using membership functions. These temperature and humidity adjustment increments are used to modify the segmented control benchmark sequence to generate a temperature and humidity compensation control sequence. A Butterworth low-pass filter is used to eliminate high-frequency fluctuations in this temperature and humidity compensation control sequence, with a set cutoff frequency and filter order. Based on this filtered sequence, multi-scale wavelet decomposition is used to extract the approximate and detail components of the temperature and humidity signals. The detail components are then denoised using a soft thresholding method. The denoising results are then used to perform wavelet reconstruction on the temperature and humidity signals to generate a smoothed temperature and humidity control sequence. Segmented temperature and humidity control targets overlapping sensitive regions of pigment degradation and active ingredient loss. A temperature and humidity baseline sequence is established using piecewise cubic spline interpolation. Within the sensitive range of 35% to 40% moisture content, the temperature baseline is set between 28 and 32 degrees Celsius, and the relative humidity baseline is controlled between 55% and 65%. The interpolation node interval is set to 15 minutes. The smoothness of the baseline sequence is ensured by the continuity of the second-order derivative of the spline function. The convolutional neural network controller utilizes a five-layer structure. The input layer has three nodes receiving temperature, humidity, and fluctuation amplitude signals. The first convolutional layer has 16 feature maps with a convolution kernel size of 3. The second convolutional layer has 32 feature maps and a fully connected layer with 64 neurons. The output layer has two nodes outputting temperature and humidity fluctuation suppression parameters. The network is trained using four hours of historical data with a learning rate of 0.001, and the training error converges to within 0.05. The fuzzy control rule base classifies temperature deviation into five levels: large negative, small negative, zero, small positive, and large positive, corresponding to a temperature deviation range of -2°C to 2°C. Humidity deviation is similarly classified into five levels, ranging from -5% to 5%. A triangular membership function is used to calculate membership values, and the center of gravity method is used to defuzzify the temperature and humidity adjustment increments. The temperature adjustment increment ranges from -1°C to 1°C, and the humidity adjustment increment ranges from -2% to 2%. A 4th-order Butterworth low-pass filter with a cutoff frequency of 0.02 Hz is used to filter out high-frequency fluctuations with a period of less than 50 seconds. The filtered temperature and humidity series retains slowly changing trends and eliminates rapidly fluctuating components. Within the sensitive range, the temperature fluctuation amplitude is reduced from 1.5°C to 0.5°C, and the humidity fluctuation amplitude is reduced from 3.5% to 1.5%. Wavelet decomposition uses the db4 wavelet basis function to perform a three-level decomposition of the temperature and humidity signals, yielding three detail components and one approximate component.The threshold selection for soft threshold denoising uses the maximum likelihood estimation method, setting thresholds for detail components of different scales. In the detail coefficients after denoising, the noise component is effectively suppressed, and the mutation characteristics of the signal are preserved. Wavelet reconstruction synthesizes the denoised detail components and approximate components to obtain a smooth temperature and humidity control sequence. The fluctuation amplitude of the reconstructed sequence is significantly reduced, with temperature fluctuations controlled within 0.3 degrees Celsius and humidity fluctuations within 1%. The smoothness of the sequence is significantly improved while maintaining the ability to respond quickly to environmental changes. The optimized temperature and humidity control sequence achieves stable control within the sensitive range, reducing the pigment degradation rate by 40% and the loss rate of active ingredients by 35%.

[0042] S109. Execute a temperature and humidity control sequence on the drying equipment, verify the control effect by continuously monitoring the state of the Monascus mycelium, extract indicators representing the balance between drying rate and biological activity from the verification results, and generate an optimized sequence of temperature and humidity control parameters.

[0043] The temperature set value and the humidity set value are binary-encoded using a standard communication protocol to obtain a control instruction sequence, which is transmitted to a drying equipment controller via a serial communication interface. Real-time data is obtained from a temperature sensor, a humidity sensor, and a photoelectric sensor according to the control instruction sequence, and the real-time data is processed via a long short-term memory network to generate a state prediction value. Normalized values ​​of a drying rate index, a mycelium activity index, a pigment content index, and an effective ingredient index are calculated based on the state prediction value to obtain an evaluation parameter set. A non-dominated sorting genetic algorithm optimizer is constructed based on the evaluation parameter set, and weight coefficients are calculated using an entropy weight method and linear weighted combination is performed to obtain an optimized sequence of temperature and humidity control parameters.

[0044] Specifically, based on the temperature and humidity adjustment scheme, the temperature and humidity setpoints are binary-encoded using a standard communication protocol to generate a sequence of equipment control instructions, which are then transmitted to the drying equipment controller via a serial communication interface. Temperature sensors, humidity sensors, and photoelectric sensors are deployed to collect real-time data on the status of the Monascus mycelium. A long-short-term memory network is established based on this sensor data, with input layer nodes corresponding to the sensor data, hidden layer memory units, and an output layer generating state predictions. Based on these state predictions, normalized values ​​for the drying rate, mycelium activity, pigment content, and active ingredient indices are calculated to generate a set of evaluation parameters for multi-objective optimization. A non-dominated sorting genetic algorithm optimizer is constructed for this set of evaluation parameters, with crossover and mutation operator probabilities set. A tournament selection approach is used to generate an optimal solution set. Using this optimal solution set, a comprehensive evaluation function based on the entropy weight method is established, and drying rate and biological activity weight coefficients are calculated to obtain a weight vector for the optimization objectives. Based on this weight vector, a linear weighted combination of the optimal solution set is performed to generate an optimized sequence of temperature and humidity control parameters. The binary encoding of temperature and humidity control parameters uses the standard Modbus communication protocol. Temperature setpoints use 16-bit precision, with the range of 25 to 35 degrees Celsius mapped to integers between 0 and 65535. Humidity setpoints also use 16-bit precision, with the relative humidity range of 45% to 75% mapped to corresponding integers. Control instruction sequences are organized in a format consisting of function code, data address, data length, and data value, and are transmitted to the device controller via an RS485 serial interface at a baud rate of 9600. The multi-sensor assembly includes eight temperature sensors, eight humidity sensors, and four photoelectric sensors. The temperature and humidity sensors are evenly arranged in a matrix on the surface of the red yeast rice layer, spaced 200 mm apart. The photoelectric sensors measure pigment absorbance. The long-short-term memory network utilizes a three-layer structure: 20 nodes in the input layer correspond to sensor data, 40 memory cells in the hidden layer, and four nodes in the output layer correspond to drying rate, mycelium activity, pigment content, and active ingredient content, respectively. State prediction values ​​are normalized to their maximum and minimum values ​​and converted to the range of 0 to 1. The drying rate reflects the amount of water lost per unit time, with values ​​ranging from 0.2 to 0.8. Mycelial activity is measured by fluorescence intensity, with normalized values ​​ranging from 0.4 to 0.9. Pigment content is calculated based on absorbance and ranges from 0.3 to 0.8. The active ingredient index is expressed as a relative retention rate, ranging from 0.5 to 0.95. The non-dominated sorting genetic algorithm was set with a population size of 100, 200 generations, a crossover probability of 0.8, and a mutation probability of 0.1. A binary tournament selection method was used, with two individuals randomly selected from the population each time for a non-dominated comparison. The winners advanced to the next generation. After 200 generations of evolution, an optimal solution set consisting of 50 non-dominated solutions was obtained, all of which were Pareto optimal.The entropy weight method calculates weights based on the degree of data variation and calculates information entropy for each optimization objective. Lower information entropy indicates higher discrimination and a greater weight for the metric. The resulting weight vector includes a weight of 0.3 for drying rate, 0.25 for mycelium activity, 0.25 for pigment content, and 0.2 for active ingredient. A linear weighted combination method was used to comprehensively score the 50 solutions in the optimal solution set, and the solution with the highest score was selected as the final optimization scheme. The optimized temperature and humidity control sequence achieved a balance between drying rate and biological activity. While ensuring the water loss rate, mycelium activity remained above 0.8, pigment content loss was controlled within 15%, and active ingredient retention reached 90%. The control sequence had a time resolution of 1 minute and contained 240 control nodes, forming a complete 4-hour drying process control solution.

[0045] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for real-time control of the drying degree of red yeast rice, characterized in that: The method comprises: Monitor the changes in mycelium surface moisture content, internal metabolic activity indicators and pigment content during the red yeast rice drying process, and generate a mycelium status distribution map; Extract moisture gradient characteristics and metabolic activity parameters from the mycelium state distribution map, analyze the relationship between water evaporation rate and mycelium activity, and determine the dynamic trends of moisture gradient and activity changes; Analyze the dynamic trend of moisture gradient and activity change, determine the relationship between the two by fitting curves, and predict the changing trend of mycelium moisture and activity at different drying stages; The impact of temperature fluctuations on uneven evaporation of red yeast rice is determined based on the dynamic trends of moisture gradient and activity changes. If the temperature fluctuation amplitude exceeds a preset threshold, the deviation of the moisture evaporation rate is predicted to obtain an index of uneven moisture distribution. Aiming at the moisture distribution unevenness index, the air supply mode is optimized to improve the uniform distribution of heat inside the red yeast rice, thereby reducing the deviation of the moisture evaporation rate and obtaining the temperature and humidity control sequence; Extract phase-specific variation characteristics from the changing trends of mycelium moisture and activity at different drying stages, analyze the impact of temperature and humidity fluctuations at different drying stages on pigment stability, predict the degradation risk of pigments at different drying stages of red yeast rice, and determine the sensitive range of pigment degradation; Compare the sensitive interval for pigment degradation and the interval for decreased active ingredient retention to determine whether there is overlap between the two. If there is overlap, it indicates that temperature and humidity fluctuations in this interval may cause both pigment degradation and active ingredient loss. In view of the overlapping sensitive areas of pigment degradation and active ingredient loss, temperature and humidity control targets are set for different stages according to the stability requirements of pigments and active ingredients, and a temperature and humidity control sequence is generated for the protection of pigments and ingredients.

2. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The method of monitoring the changes in surface moisture content, internal metabolic activity index and pigment content of mycelium during the red yeast rice drying process and generating a mycelium state distribution map includes: An initial moisture content dataset is obtained from preset sampling points on the mycelium surface using a humidity sensor array. The initial dataset is collected using a metabolic activity sensor and a spectrophotometer to obtain a point metabolic activity dataset and a pigment content dataset. The initial data set is converted into interval values ​​using the maximum and minimum value normalization method, and the continuous surface data of moisture distribution is generated according to the moisture gradient value between adjacent sampling points and the inverse distance weighted interpolation algorithm; The moisture content of the sampling point is determined based on the moisture distribution continuous surface data. If the moisture content of the sampling point is lower than a preset threshold, the metabolic activity data set and the pigment content data set of the point with moisture content lower than the preset threshold are obtained, a Bayesian network model is constructed, and the Kriging spatial interpolation algorithm is used to calculate the sampling point data to obtain a mycelium state distribution map.

3. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The method of extracting moisture gradient characteristics and metabolic activity parameters from the mycelium state distribution map, analyzing the relationship between water evaporation rate and mycelium activity, and determining the dynamic trend of moisture gradient and activity changes includes: According to the moisture content values ​​of the sampling points in the mycelium state distribution map, the Sobel difference operator is used to calculate the moisture gradient values ​​in the horizontal and vertical directions. The moisture gradient direction characteristic map and moisture gradient amplitude distribution map are obtained through vector field processing. For the original data of metabolic activity, a third-order autoregressive filter was used to smooth the metabolic activity change data to obtain the metabolic activity time series curve; Extracting the moisture gradient difference at adjacent time points from the moisture gradient direction characteristic map and the moisture gradient amplitude distribution map, and obtaining the moisture evaporation rate data by calculation; After obtaining the water evaporation rate data and the metabolic activity time series curve, the random forest regression method is used to construct a mapping relationship between water evaporation and metabolic activity. The input features of the mapping relationship include the water gradient amplitude and the water evaporation rate data, and the output features of the mapping relationship include the metabolic activity change rate and the metabolic activity attenuation curve.

4. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The analysis of the dynamic trend of moisture gradient and activity change, determining the relationship between the two by fitting curves, and predicting the changing trends of mycelium moisture and activity at different drying stages include: The moisture gradient trend function was obtained by fitting a cubic polynomial function based on the moisture gradient data, and the activity trend function was obtained by fitting a negative exponential function based on the activity change data. Calculating a correlation coefficient within a fixed time interval for the moisture gradient trend function and the activity trend function to obtain a water activity correlation curve; The K-means clustering method is used to group the correlation coefficients in the water activity correlation curve to obtain the drying stage time interval; For the data in the drying stage time interval, a water activity mapping function is established using a support vector regression method of a radial basis kernel function, wherein the input variable of the mapping function is the water gradient value and the output variable is the activity change value; According to the water activity mapping function, iterative prediction is performed with a fixed step size within each drying stage interval to obtain the dynamic development trend of the water gradient and activity change.

5. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The effect of temperature fluctuation on uneven evaporation of red yeast rice water is determined based on the dynamic trend of moisture gradient and activity change; If the temperature fluctuation amplitude exceeds the preset threshold, the deviation of the water evaporation rate is predicted to obtain the water distribution unevenness index, including: Obtaining a measured value of the water evaporation rate according to the moment when the temperature fluctuation amplitude exceeds a preset temperature threshold, and establishing a functional relationship between the temperature fluctuation amplitude and the evaporation rate deviation using a quadratic polynomial regression equation; Based on the functional relationship between the temperature fluctuation amplitude and the evaporation rate deviation, a time series prediction function is established using a long short-term memory network. The input layer nodes of the time series prediction function include the temperature fluctuation amplitude, the fluctuation frequency and the moisture gradient value, and the output layer node is the evaporation rate prediction value. According to the evaporation rate prediction value output by the time series prediction function, the evaporation rate difference between adjacent measuring points on the red yeast rice surface is calculated, and the moisture distribution variation coefficient is obtained by using the ratio of the standard deviation to the average value. The moisture distribution variation coefficient is used as an indicator of moisture distribution unevenness.

6. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The temperature and humidity control sequence is obtained by optimizing the air supply mode to improve the uniform distribution of heat inside the red yeast rice and reduce the deviation of the water evaporation rate, including: According to the moisture distribution unevenness index, a fuzzy control method is used to obtain wind speed adjustment and heating power adjustment, wherein the adjustment is obtained by mapping the membership function of temperature deviation and humidity deviation; The temperature field distribution diagram of the red yeast rice layer is solved by computational fluid dynamics equations for the wind speed adjustment amount and the heating power adjustment amount, and the temperature field distribution diagram is used as input data for heat transfer calculation; A heat transfer calculation model is established using a finite element meshing method, and a heat distribution uniformity value is calculated using the temperature field distribution diagram. The mesh unit size is set according to the thickness of the red yeast rice layer. According to the heat distribution uniformity value, a recursive neural network structure is used to calculate the predicted value of the water evaporation rate. The adjustment sequence of the air supply speed and heating power is obtained through the dynamic programming state transfer equation, and the temperature set value and humidity set value of the fixed sampling period are generated to obtain the temperature and humidity control sequence.

7. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The method extracts the phased change characteristics from the changing trends of mycelium moisture and activity at different drying stages, analyzes the effects of temperature and humidity fluctuations at different drying stages on pigment stability, predicts the degradation risk of pigments at different drying stages of red yeast rice, and determines the sensitive range of pigment degradation, including: The Dobesi wavelet basis function was used to perform wavelet transform decomposition on the time series data of mycelium moisture content and activity, and the drying stage division results were obtained by the peak and valley recognition method. Extracting characteristic parameters of temperature fluctuation and humidity fluctuation according to the drying stage division results, and obtaining significant characteristic indicators of temperature and humidity fluctuation by variance analysis method; A multiple regression equation was established for the significant characteristic indicators of the temperature and humidity fluctuations, and the influence function of the temperature and humidity fluctuations on the pigment stability was obtained from the mycelium moisture content and activity data; A forward neural network predictor is constructed based on the influence function, and the pigment degradation rate is obtained from the temperature and humidity fluctuation amplitude, moisture content and activity value. The pigment degradation rate is divided into sensitive intervals using the K-means clustering method to obtain the pigment degradation sensitive time interval.

8. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The comparison of the sensitive interval of pigment degradation and the interval of decreased effective ingredient retention rate to determine whether there is overlap between the two, if overlap, indicates that the temperature and humidity fluctuations in the interval may cause pigment degradation and effective ingredient loss at the same time, including: Fast Fourier transform was used to perform spectrum analysis on the time series data of the pigment degradation sensitive range, and the correlation curve between pigment content and active ingredient content was obtained based on the temperature fluctuation frequency distribution and humidity fluctuation frequency distribution. The Otsu method adaptive threshold method is used to determine the sensitive interval boundary according to the correlation curve, and the overlapping time period of the pigment degradation interval and the effective ingredient loss interval is obtained by intersection operation; Calculating the root mean square value of the temperature fluctuation amplitude and the root mean square value of the humidity fluctuation amplitude for the overlapping time period, and establishing a mapping relationship using a radial basis kernel function support vector machine, where the input variables include the root mean square value of the temperature fluctuation and the root mean square value of the humidity fluctuation, and the output variables include the pigment degradation rate and the effective ingredient loss rate; The degradation rate threshold and loss rate threshold are set according to the output results of the mapping relationship. The temperature and humidity fluctuation data are grouped using the K-means clustering method, and the location of the boundary points of the sensitive interval is determined by calculating the inter-class distance.

9. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The overlapping sensitive intervals for pigment degradation and active ingredient loss are set according to the stability requirements of the pigment and active ingredient, respectively, to set temperature and humidity control targets at different stages, and generate a temperature and humidity control sequence for pigment and ingredient protection, including: Temperature constraint equations and humidity constraint equations were established based on the pigment degradation rate and the effective component loss rate, and the temperature and humidity target values ​​were obtained using piecewise cubic spline interpolation. Constructing a convolutional neural network controller for the temperature and humidity target values, extracting temperature and humidity fluctuation characteristics through a hidden layer of the convolutional neural network controller, and obtaining a fluctuation suppression parameter from an output layer of the convolutional neural network controller; A fuzzy control rule base is established by using the fluctuation suppression parameter, and a temperature and humidity adjustment increment is obtained by calculating the membership function of the fuzzy control rule base; The temperature and humidity signals are subjected to multi-scale wavelet decomposition according to the temperature and humidity adjustment increments, the detail components obtained by the wavelet decomposition are subjected to soft threshold noise reduction processing, and a smooth temperature and humidity control sequence is obtained through wavelet reconstruction.

10. The method for real-time control of red yeast rice dryness according to claim 1, wherein: The method includes: executing a temperature and humidity control sequence on a drying device, verifying the control effect by continuously monitoring the state of Monascus mycelium, extracting an indicator representing the balance between drying rate and biological activity from the verification results, and generating an optimized sequence of temperature and humidity control parameters, specifically including: The temperature setting value and the humidity setting value are binary-encoded using a standard communication protocol to obtain a control instruction sequence, which is transmitted to the drying equipment controller via a serial communication interface; Acquire real-time data from a temperature sensor, a humidity sensor, and a photoelectric sensor according to the control instruction sequence, and process the real-time data via a long short-term memory network to generate a state prediction value; Calculating normalized values ​​of a drying rate index, a mycelium activity index, a pigment content index, and an effective component index based on the state prediction value to obtain an evaluation parameter set; A non-dominated sorting genetic algorithm optimizer is constructed according to the evaluation parameter set, and the weight coefficients are calculated by the entropy weight method and linear weighted combination is performed to obtain the optimized sequence of temperature and humidity control parameters.

Images