Humidity and temperature coordinated control method based on fermentation process of Jiuyu Hongmei

Abstract

The application relates to the technical field of tea processing process control, and discloses a temperature and humidity synergic control method based on a fermentation process of Jiugu Hongmei. The method comprises the following steps: collecting real-time temperature, humidity and tea internal humidity distribution signals in a fermentation environment; pre-processing and feature extraction are performed on the collected data to obtain a temperature and humidity sequence and a humidity distribution feature sequence; fermentation rate change characteristics are extracted based on the temperature and humidity sequence, a current fermentation stage is determined by combining humidity distribution features and identifying stage turning points; synergic control strategy parameters are generated according to the influence of temperature and humidity interaction in the current stage; a fermentation environment fine-tuning scheme is constructed, and a control instruction is generated through historical verification and secondary calibration; a control signal is outputted, and a fermentation rate trajectory is tracked; when deviating from a preset path, a correction signal is fed back. The application improves the temperature and humidity control accuracy and stability.

Description

Technical Field

[0001] This application relates to the field of tea processing control technology, and in particular to a method for coordinated temperature and humidity control based on the Jiuqu Hongmei fermentation process. Background Technology

[0002] Jiuqu Hongmei belongs to the black tea category, and its fermentation process plays a decisive role in the formation of the tea's aroma, flavor, and liquor color. During fermentation, temperature and humidity are interdependent: increased temperature accelerates moisture evaporation and alters the reaction rate, while insufficient humidity may inhibit the fermentation reaction. Furthermore, fermentation has distinct stages, with different stages exhibiting varying sensitivities to and control priorities regarding temperature and humidity conditions.

[0003] In current production processes, the fermentation environment is largely regulated by manual experience. Progress is often judged by observing leaf color, texture, or odor, and equipment such as air conditioners, humidifiers, and exhaust fans are manually controlled. This method is heavily influenced by the operator's experience level and suffers from lag, making timely and appropriate adjustments difficult at key stages. Furthermore, when the fermentation chamber is large, differences in airflow organization and stacking location can easily lead to uneven local temperature and humidity distribution. Single-point monitoring cannot reflect changes in the internal humidity distribution of the tea leaves, resulting in inconsistent fermentation levels in different areas of the same batch. External climate fluctuations, door opening and closing, and equipment start-ups and shutdowns can also cause temperature and humidity fluctuations, causing the fermentation trajectory to deviate from the target process path, leading to decreased quality stability and increased rework rates.

[0004] Therefore, there is an urgent need for a fermentation process control method that can combine internal humidity distribution information with stage identification results to achieve temperature and humidity coordination, verifiability, and feedback correction capabilities. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a temperature and humidity synergistic control method based on the Jiuqu Hongmei fermentation process. This method solves the problems in the prior art, such as response lag due to reliance on manual experience for adjustment, control imbalance under temperature and humidity coupling, and inconsistent fermentation of the same batch due to uneven temperature and humidity distribution in the fermentation room. This method improves the control precision and quality consistency of the fermentation process.

[0006] In a first aspect, this application provides a method for coordinated temperature and humidity control based on the fermentation process of Jiuqu Hongmei plum, the method comprising:

[0007] Step S1: In the fermentation environment of Jiuqu Hongmei, real-time temperature data, humidity data, and humidity distribution signals reflecting the interior of Jiuqu Hongmei are collected by sensors;

[0008] Step S2: Preprocess and extract features from the temperature data, humidity data, and humidity distribution signal to obtain temperature and humidity sequences and humidity distribution feature sequences;

[0009] Step S3: Extract the fermentation rate change features based on the temperature and humidity sequence, and identify the stage transition markers by combining the humidity distribution feature sequence to determine the current stage of the fermentation process;

[0010] Step S4: Based on the current stage, assess the interaction between temperature and humidity and environmental fluctuations, and generate temperature and humidity coordinated control strategy parameters corresponding to the current stage;

[0011] Step S5: Construct a fermentation environment fine-tuning scheme based on the control strategy parameters, and perform historical verification and secondary calibration on the fermentation environment fine-tuning scheme to generate new control instructions;

[0012] Step S6: Output temperature and humidity control signals according to the control instructions, and track the fermentation rate trajectory. If it deviates from the preset path, a correction signal will be generated.

[0013] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0014] This application provides a temperature and humidity coordinated control method based on the fermentation process of Jiuqu Hongmei. By synchronously collecting the temperature and humidity distribution signals of the fermentation environment and the humidity distribution inside the pile in the fermentation environment, and combining adaptive moving average filtering, variational mode decomposition and stability metric-driven secondary correction, the method achieves accurate preprocessing of multi-dimensional data, significantly improves data continuity and reliability, and avoids the risk of misjudgment caused by sensor noise, asynchronous sampling and humidity changes from the source.

[0015] Furthermore, by extracting fermentation rate change features through ensemble empirical mode decomposition and identifying stage transition markers by combining humidity distribution features, a dual judgment mechanism is constructed that compares the hidden Markov chain model with the historical feature database to achieve accurate and timely identification of fermentation stages, overcoming the lag and subjectivity defects of traditional empirical judgment.

[0016] Meanwhile, based on the current stage of quantitative assessment of the interaction between temperature and humidity, stage-adaptive collaborative control parameters are generated. A fine-tuning scheme is constructed by combining a dynamic correlation matrix. Control commands are optimized through historical trajectory matching verification and secondary calibration. Finally, relying on closed-loop feedback control, the fermentation rate trajectory is tracked in real time and dynamically corrected, which effectively improves the accuracy and stability of temperature and humidity control, ensures batch consistency of Jiuqu Hongmei fermentation quality, enhances the controllability and reliability of fermentation production, and provides technical support for intelligent fermentation production. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a temperature and humidity synergistic control method based on the Jiuqu Hongmei fermentation process in an embodiment of this application;

[0019] Figure 2 This is a comparison chart of the temperature and humidity sequence filtering and noise reduction effects in the embodiments of this application;

[0020] Figure 3 This is a bar chart comparing the weighted effects of temperature and humidity on fermentation rate in embodiments of this application.

[0021] Figure 4 This is a comparison chart showing the similarity between the temperature target trajectory and the historical best trajectory in this application embodiment;

[0022] Figure 5 This is a comparison chart verifying the execution effect of the control commands in this embodiment. Detailed Implementation

[0023] This application provides a method for coordinated temperature and humidity control based on the fermentation process of Jiuqu Hongmei plum. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the temperature and humidity synergistic control method based on the Jiuqu Hongmei fermentation process in this application includes:

[0025] Step S1: In the fermentation environment of Jiuqu Hongmei, real-time temperature data, humidity data, and humidity distribution signals reflecting the interior of Jiuqu Hongmei are collected by sensors.

[0026] Specifically, multiple sets of temperature and humidity sensors and a distributed humidity sensor array are deployed in the fermentation environment of Jiuqu Hongmei tea. The temperature and humidity sensors are placed in the environmental space inside the fermentation chamber to collect real-time temperature and humidity data of the fermentation environment. The distributed humidity sensor array is placed inside the fermentation pile of Jiuqu Hongmei tea and consists of multiple capacitive humidity sensor nodes. The sensor nodes are arranged in layers along the thickness direction of the fermentation pile, for example, at the surface, middle and inner layers of the pile, to collect humidity distribution signals at different locations inside the tea leaves. The internal humidity distribution signal is a signal that reflects the uniformity of humidity inside the pile, formed by the combination of humidity data from multiple locations. All collected data are timestamped to ensure time sequence consistency and are uploaded to the data terminal in real time via wired transmission.

[0027] The above-mentioned data collection method can comprehensively obtain macroscopic temperature and humidity data of the fermentation environment and microscopic humidity distribution information inside the tea leaves. It solves the problem that traditional manual control can only obtain local surface data, and provides complete data support for subsequent data processing, stage identification and precise control. It effectively avoids control imbalance caused by one-sided data, improves the pertinence and accuracy of temperature and humidity control in the fermentation process, and ensures the consistency of Jiuqu Hongmei fermentation quality.

[0028] Step S2: Preprocess and extract features from the temperature data, humidity data, and humidity distribution signal to obtain the temperature and humidity sequence and the humidity distribution feature sequence.

[0029] In step S2, the preprocessing and feature extraction of temperature data, humidity data and humidity distribution signal include: using an adaptive moving average smoothing algorithm to filter and denoise the temperature data and humidity data to generate a denoised temperature and humidity sequence.

[0030] For the humidity distribution signal, variational mode decomposition (VMD) is used to decompose the signal, separating the effective component from the interference component. The effective component is then extracted as the initial humidity distribution feature sequence. Based on the sensor acquisition timestamps, the denoised temperature and humidity sequence is time-aligned with the initial humidity distribution feature sequence to generate unified time series data. The variance of the initial humidity distribution feature sequence is calculated to obtain a stability metric for the humidity distribution within the Jiuqu Hongmei plum tree. The unified time series data is then subjected to secondary correction using the stability metric to remove abnormal data points caused by abrupt changes in humidity distribution, resulting in optimized temperature and humidity sequences and humidity distribution feature sequences.

[0031] Specifically, a pre-defined adaptive moving average smoothing algorithm is used to filter and denoise the temperature and humidity data. This algorithm dynamically adjusts the window size based on data fluctuations, automatically adapting the window length to the real-time fluctuation range of the temperature and humidity data. When the fluctuation range is large, the window is reduced to preserve key trends; when the fluctuation range is small, the window is expanded to enhance the denoising effect. In the algorithm implementation, a pre-defined initial window is first calculated before and after each data point. For example, the initial window size is set to the variance of fluctuations within 5 data points. If the variance is greater than a pre-defined threshold, the window is reduced to 3 data points; if the variance is less than or equal to the pre-defined threshold, the window is expanded to 7 data points. Then, the adjusted window is used to calculate the average value of each temperature and humidity data point, replacing the original data points, thereby generating a denoised temperature and humidity sequence. Figure 2 The image shows a comparison of the denoising effects of the temperature and humidity sequences during the fermentation of Jiuqu Hongmei plums. It intuitively demonstrates the effect of using the adaptive moving average smoothing algorithm to filter and denoise the temperature and humidity data. In the image, the horizontal axis represents fermentation time, the left vertical axis represents temperature, and the right vertical axis represents humidity. The curves with larger fluctuations correspond to the original temperature and humidity sequences, while the curves with smaller fluctuations correspond to the filtered and denoised temperature and humidity sequences. It can be seen that after processing by this algorithm, the instantaneous noise interference in the temperature and humidity sequences is effectively suppressed, while the core trends of temperature and humidity changes during fermentation are preserved. For example, during the early stage of fermentation when the temperature gradually rises and the humidity slowly decreases, the denoised sequences can more accurately reflect the real dynamics of the fermentation environment, providing a high-precision data foundation for subsequent feature extraction and stage recognition.

[0032] For the collected humidity distribution signal inside tea leaves, variational mode decomposition (VMD) technology is used for signal decomposition. By constraining the energy distribution of the humidity distribution signal in the frequency domain, the signal is decomposed into several intrinsic mode functions (IMFs) with different center frequencies and bandwidths. This separates effective humidity information from environmental interference information. The changes in the humidity distribution signal inside tea leaves are mainly manifested as low-frequency components that change slowly over time or space, while interference introduced by environmental temperature and humidity fluctuations, sensor noise, and other factors is mainly manifested as high-frequency components. Therefore, VMD can distinguish between the low-frequency effective components and high-frequency interference components in the humidity distribution signal. Through the above processing, the original humidity distribution signal can be decomposed into multiple sets of IMFs. The IMFs with lower center frequencies correspond to the true changes in humidity distribution inside tea leaves, while the IMFs with higher center frequencies correspond to invalid components caused by external environmental interference or measurement noise. Furthermore, a predetermined number of IMFs with the lowest center frequencies are selected from the decomposed IMFs and superimposed to obtain an initial humidity distribution feature sequence that can characterize the essential features of humidity distribution inside tea leaves, providing a reliable data foundation for subsequent humidity state analysis and evaluation.

[0033] The denoised temperature and humidity sequence is time-aligned with the initial humidity distribution feature sequence based on sensor acquisition timestamps. A linear interpolation method is used to fill time gaps in the sequence. Using the timestamps of the temperature and humidity sequence as a reference, for missing time points in the initial humidity distribution feature sequence, linear interpolation results are calculated using the humidity distribution feature values ​​of the two adjacent valid time points to complete the data. This ensures that each time point corresponds to a complete set of temperature and humidity data and humidity distribution feature data, generating unified time series data. The variance of the initial humidity distribution feature sequence is calculated as a measure of the stability of humidity distribution within the tea leaves, and the standard deviation is also used as a stability measure. These stability measures quantitatively reflect the fluctuation range of humidity distribution over time; a larger variance indicates a more uneven humidity distribution and poorer stability. When the variance or standard deviation of the unified time series at a certain time point exceeds a pre-set threshold, it is determined that there is abnormal data at that time point caused by abrupt changes in humidity distribution. Such abnormal data usually originates from instantaneous environmental disturbances or sensor noise interference. At this point, a secondary correction process is performed on the uniform time series data using stability metrics. For example, a variance threshold of 0.25 can be preset. When the stability metric exceeds the threshold at a certain moment, the corresponding temperature and humidity data point is marked as an outlier. Then, several adjacent data points before and after the outlier, such as three unmarked data points, are selected, and a linear fit is performed using the least squares method. The original outlier data point is replaced with the value at the corresponding time position on the fitted curve, thereby achieving smooth correction of local abnormal fluctuations. The above preprocessing and feature extraction process effectively eliminates environmental fluctuations and equipment noise interference, accurately separates and retains key information reflecting the essence of the fermentation process, and lays a high-precision and high-reliability data foundation for subsequent intelligent judgment and control.

[0034] Step S3: Extract fermentation rate change features based on temperature and humidity sequences, and identify stage transition markers by combining humidity distribution feature sequences to determine the current stage of the fermentation process.

[0035] In step S3, extracting fermentation rate change features based on temperature and humidity sequences includes: using ensemble empirical mode decomposition (EMD) to decompose the temperature and humidity sequences into time series components at different time scales; calculating the amplitude change rate and peak interval of each fluctuation component; constructing a fermentation rate feature vector based on the amplitude change rate and peak interval; and mapping the fermentation rate feature vector to a preset feature space through normalization to obtain standardized fermentation rate change features.

[0036] Specifically, ensemble empirical mode decomposition (EMD) is used to decompose the temperature and humidity sequence into time series components. EMD is used to decompose this nonlinear, non-stationary time series into multiple subsequences with different time scale characteristics. The process involves first superimposing Gaussian white noise with an amplitude equal to a preset multiple of the sequence's standard deviation onto the original temperature and humidity sequence, and repeating this superposition process multiple times to construct multiple noisy temperature and humidity sequences. Then, EMD is performed on each noisy temperature and humidity sequence to obtain multiple sets of intrinsic mode functions (IMFs). Finally, the IMFs of the same order in different noisy sequences are averaged point-by-point to obtain the final EMD result, i.e., several temperature and humidity fluctuation components. Each IMF obtained through EMD corresponds to the temperature and humidity variation components at different time scales. High-frequency fluctuation components characterize rapid temperature and humidity changes over a short period, while low-frequency fluctuation components characterize the slow temperature and humidity changes over a longer time scale, reflecting the slow trend during fermentation. After obtaining the temperature and humidity fluctuation components at different time scales, the amplitude change rate and peak interval of each fluctuation component are calculated to quantify the speed and rhythm of temperature and humidity changes at each time scale. For each temperature and humidity fluctuation component, all extreme points in the sequence are first identified. The extreme points are arranged in chronological order, and the amplitude difference and time difference between adjacent extreme points are calculated. The ratio of the two is the amplitude change rate of that interval. The average amplitude change rate of all intervals is taken as the amplitude change rate of the fluctuation component. The amplitude difference refers to the ratio of the amplitude change rate of the fluctuation component to the peak interval. In a time-sequential sequence of adjacent extreme points, the absolute difference between the temperature and humidity values ​​of the later extreme point and the earlier extreme point is considered. For example, in a time series of a certain temperature and humidity fluctuation component, the extreme points obtained in chronological order are designated as the 1st extreme point, the 2nd extreme point, the 3rd extreme point, and so on. For two adjacent extreme points, such as the 1st and 2nd extreme points, the magnitude difference between them and their time interval are calculated. This magnitude difference is then divided by the corresponding time interval to obtain the rate of change of magnitude within that time interval. Similarly, the rate of change of magnitude is calculated for the intervals between adjacent extreme points, such as the 2nd and 3rd extreme points, the 3rd and 4th extreme points, etc.Finally, the amplitude change rate obtained from all adjacent extreme point intervals is averaged, and the average value is taken as the amplitude change rate of the temperature and humidity fluctuation component, which is used to characterize the speed of change of the fluctuation component over the overall time range. Then, all peak points of each fluctuation component are identified, the time difference between adjacent peak points is calculated, and the average value is taken as the peak interval of the fluctuation component. The peak interval is used to represent the average time period of the peak points in the temperature and humidity fluctuation component, which intuitively reflects the periodicity and frequency density of the fluctuation component. Based on the amplitude change rate and peak interval of each fluctuation component, they are arranged in order of decomposition order to construct a fermentation rate feature vector. This vector comprehensively covers the influence characteristics of temperature and humidity fluctuations on the fermentation rate at different time scales.

[0037] After obtaining the fermentation rate feature vector, normalization is performed to map it to a preset feature space, eliminating dimensional differences between different feature indicators and avoiding feature weight imbalances caused by differences in numerical ranges. For example, using the min-max normalization method, the minimum and maximum values ​​of each element in the fermentation rate feature vector are first calculated. Then, the minimum value is subtracted from each element value, and the result is divided by the difference between the maximum and minimum values ​​to obtain the normalized element values. All normalized elements are arranged in their original order to form standardized fermentation rate change features. The above-described process for extracting temperature and humidity change features for inferring fermentation rate accurately captures the temperature and humidity fluctuation patterns at different time scales through adaptive decomposition technology. By combining the amplitude change rate and peak interval to construct and standardize the feature vector, it solves the problem of difficulty in comprehensively representing environmental dynamic changes strongly correlated with fermentation rate. This achieves a refined description of the change trend, enabling adaptive capture of various change patterns from instantaneous disturbances to long-term trends, providing stable and quantifiable dynamic features for accurate identification of fermentation stages.

[0038] Step S3, identifying stage transition markers and determining the current stage of the fermentation process, includes: analyzing the temperature and humidity sequence using time series decomposition technology to extract fermentation rate change features; calculating the feature difference values ​​of adjacent time nodes in the humidity distribution feature sequence as humidity distribution features, and marking the humidity distribution features as candidate stage transitions when they exceed a preset threshold; associating and matching the rate mutation points in the fermentation rate change features with the candidate stage transition markers, eliminating isolated candidate markers, and obtaining valid stage transition markers; constructing a fermentation process stage division model based on the valid stage transition markers, inputting the real-time extracted fermentation rate change features and humidity distribution features into the division model, and outputting a preliminary judgment result of the current fermentation process stage; retrieving a preset feature library of each stage of Jiuqu Hongmei fermentation, comparing the preliminary judgment result with the stage feature parameters in the feature library, and if the matching degree exceeds a preset threshold, confirming that the stage corresponding to the preliminary judgment result is the current stage of the fermentation process, and outputting identification information including the stage name and feature parameter range.

[0039] Specifically, after obtaining standardized fermentation rate change characteristics, to accurately capture the stage transition nodes of the fermentation process, it is necessary to identify stage transition markers by combining humidity distribution characteristics and rate change patterns. Specifically, firstly, time series decomposition technology is used to analyze the processed temperature and humidity series. This technology is a data analysis method that decomposes time-series data into components such as trend, period, and noise. By separating the trend component, seasonal component, and residual component in the temperature and humidity series, fermentation rate change characteristics reflecting the overall progress of fermentation are extracted. These characteristics include quantitative indicators such as rate abrupt change points, rate increase magnitude, stable duration, and rate decay, used to quantify the speed of fermentation, rate fluctuation magnitude, and abrupt changes. Subsequently, the characteristic difference values ​​of adjacent time nodes in the humidity distribution characteristic sequence are calculated. The characteristic difference value is obtained by the absolute value of the difference between the humidity distribution statistical parameters of the subsequent node and the corresponding parameters of the previous node. This difference value is directly used as the humidity distribution characteristic. Humidity distribution statistical parameters include mean, variance, and distribution uniformity. The humidity distribution characteristic reflects whether the internal humidity of the tea leaves is uniform and whether there are localized areas of excessive dryness or excessive moisture, which is a key indicator for judging whether fermentation is balanced. A preset threshold is set, which can be determined based on historical fermentation data statistics, for example, set as the 90th quantile of the historical humidity distribution characteristic difference value sequence. When the calculated characteristic difference value exceeds the preset threshold, it indicates that between two adjacent monitoring times, the humidity distribution inside the tea pile has undergone a significant abrupt change far exceeding the normal stable fluctuation range. This usually means that the dynamic balance of moisture migration, transformation, or loss within the fermentation pile has been broken, and this time point is marked as a candidate stage turning point.

[0040] Next, the rate mutation points in the fermentation rate change characteristics are associated and matched with candidate stage transition markers. The rate mutation point is the time point in which the fermentation rate changes by more than a preset threshold, such as 30%, within a continuous time node. The association and matching are achieved by calculating the time interval between the rate mutation point and the candidate stage transition marker. For example, the time interval threshold is set to 30 minutes. The timestamps of each candidate stage transition marker are compared with all rate mutation points one by one. If the time difference between a certain rate mutation point and the candidate marker is less than the time interval threshold, the two are considered to be matched successfully and the candidate stage transition marker is retained. If the time difference between a certain candidate marker and all rate mutation points exceeds the time interval threshold, or if there is no corresponding rate mutation point, it is identified as an isolated candidate marker and is eliminated. Finally, effective stage transition markers that simultaneously reflect the coordinated changes in temperature and humidity are obtained.

[0041] After obtaining effective stage transition identifiers, a fermentation process stage division model is constructed by combining historical fermentation stage label data. This model is constructed using a hidden Markov chain, which is a probability-based state transition model that includes hidden states and observation sequences. The hidden states correspond to the four different stages of Jiuqu Hongmei fermentation: preheating, acceleration, stabilization, and decay. The observation sequences are a combination of fermentation rate change characteristics and humidity distribution characteristics. The standardized fermentation rate change features and humidity distribution feature vectors extracted in real time are arranged into an observation sequence in chronological order and input into a pre-trained Hidden Markov Model. The probability distribution of each hidden state is calculated, and the hidden state with the highest probability is selected as the preliminary judgment result of the current fermentation process stage. The preliminary judgment result includes the specific stage name of the Jiuqu Hongmei fermentation, namely the preheating stage, acceleration stage, stabilization stage, and decay stage; the probability value corresponding to the stage, that is, the highest probability value among the probability distributions of the current hidden states; and the core feature description with the highest matching degree with the stage, that is, the key feature information supporting the stage judgment in the standardized fermentation rate change features and humidity distribution features extracted in real time. This provides a clear and complete basis for subsequent comparison and verification with the preset stage feature library. The detailed identification process of the preliminary judgment result will be explained in detail later.

[0042] After obtaining the preliminary judgment result of the fermentation process stage, it is necessary to retrieve the preset feature library of each stage of Jiuqu Hongmei fermentation. The feature library contains feature parameters such as the typical range of each dimension of the standardized fermentation rate change feature vector, the typical value of humidity distribution feature, and the number of historical effective stage transition markers for each fermentation stage. For example, the first dimension feature corresponding to the preheating stage, such as the typical range of the amplitude change rate of the intrinsic mode function reflecting the long-term trend, is 0.05-0.1, and the number of transition markers is less than 2. The similarity between the observed feature vector corresponding to the preliminary judgment result and the feature parameter vector of the corresponding stage in the feature library is calculated using Euclidean distance. Euclidean distance is the square root of the sum of the square differences of corresponding elements between two feature vectors. The smaller the distance, the higher the similarity. Assuming the matching degree threshold is 0.8, if the calculated similarity exceeds the matching degree threshold, the preliminary judgment result is confirmed to be accurate, and the corresponding stage is the current stage of the fermentation process. The output includes the stage name and the range of feature parameters.

[0043] By integrating the dual perspectives of humidity distribution changes and fermentation rate abrupt changes, and combining them with an intelligent partitioning model based on hidden Markov chains, this method significantly improves the accuracy and robustness of identifying inflection points in the fermentation process. This approach not only effectively eliminates the interference of isolated noise but also achieves self-correction through comparison with a historical feature database, providing reliable state input for subsequent precise control and greatly reducing the risk of misjudgment.

[0044] The preliminary judgment result of the current fermentation process stage includes: fusing the real-time extracted fermentation rate change features and humidity distribution features to generate an observation feature vector; inputting the observation feature vector into the constructed fermentation process stage division model, calculating the state probability corresponding to each fermentation stage through the preset parameters in the model; using the Viterbi algorithm to traverse the state probabilities of all time points to obtain a state sequence whose corresponding state probabilities meet the preset requirements, and taking the state corresponding to the current time point in the state sequence as the preliminary judgment result of the current fermentation process stage.

[0045] Specifically, after constructing the fermentation process stage segmentation model, preliminary stage judgment results need to be obtained through feature integration and model computation to provide basic support for subsequent comparison and verification with the feature library. Specifically, the real-time extracted fermentation rate change features and humidity distribution features are fused to generate observation feature vectors. Feature fusion is a process of concatenating and integrating the two types of features in a preset order, with the elements of the fermentation rate change features first, followed by the elements of the humidity distribution features, ensuring that the vector structure is consistent with the observation sequence format during model training. This allows the generated observation feature vectors to fully represent the comprehensive state of the current fermentation environment and rate. The observation feature vectors are then input into the constructed fermentation process stage segmentation model, which is based on a hidden Markov chain and is a probability-based state transition model. Its hidden states correspond to the four stages of Jiuqu Hongmei fermentation: preheating, acceleration, stabilization, and decay. The state probabilities corresponding to each fermentation stage are calculated using preset parameters of the model. The model's preset parameters include the initial state probability, transition probability, and emission probability defined by a multivariate Gaussian distribution. The initial state probability is set to a uniform distribution, meaning that the probability of being in all four stages at the initial moment of fermentation is equal, ensuring the objectivity of the initial state judgment. The transition probability is calculated from a large amount of historical fermentation data using the maximum likelihood estimation method. Specifically, it is calculated by statistically analyzing the frequency ratio of transitions between different stages in the historical data. For example, the probability of transitioning from the acceleration stage to the stable stage is statistically estimated to be 0.7, ensuring that the transition probability matches the actual transition patterns of fermentation stages. The emission probability is assumed to be a multivariate Gaussian distribution, with its mean set to the average rate of change over time in the historical fermentation data, such as 0.15, and its variance set to a value suitable for humidity fluctuation scenarios, such as 0.02. This distribution describes the corresponding probability relationship between the observed feature vector and each hidden state.

[0046] In the probability calculation process, the initial state probability is used as the starting point, and the transition probability is combined to obtain the prior probability of each stage at the current time point. This prior probability reflects the possibility of the current stage based on the historical stage transition pattern. Then, the prior probability is corrected by combining the emission probability, that is, the probability value is adjusted according to the matching degree between the current observed feature vector and the features of each stage. Finally, the posterior probability corresponding to each fermentation stage is obtained. This posterior probability is the state probability of the fermentation process in each stage under the current observed feature vector. For example, when the observed feature vector corresponds to the features of a steady increase in temperature and a slow decrease in humidity, the model calculates that the state probability of the preheating stage is 0.1, the state probability of the acceleration stage is 0.65, the state probability of the stabilization stage is 0.2, and the state probability of the decay stage is 0.05, which intuitively reflects the probability of each stage.

[0047] Finally, the Viterbi algorithm is used to iterate through the state probabilities of each fermentation stage at all time points. Specifically, the state probabilities of each time point corresponding to the real-time generated observation feature vector are used as input. Starting from the fermentation start time point, the state probability of each time point is recursively calculated one by one. During the calculation process, the maximum cumulative probability and forward associated state corresponding to each state at each time point are recorded simultaneously. The forward associated state refers to the associated fermentation stage state of the previous adjacent time point when a certain state at the current time point achieves the maximum cumulative probability, forming complete path tracing data. For example: In the first hour of fermentation, the probabilities of each stage are: preheating 0.4, acceleration 0.3, stabilization 0.2, and decay 0.1. The maximum cumulative probability of the preheating stage is 0.4. Since there is no preceding time point, the forward associated state is marked as none, and it is recorded as "t1: state - preheating, maximum cumulative probability -0.4, forward associated state - none". In the second hour, the maximum cumulative probabilities of each stage are calculated recursively as follows: preheating 0.35, acceleration 0.5, stabilization 0.25, and decay 0.15, and it is recorded as "t2: state - acceleration, maximum cumulative probability -0.5, forward associated state - t1 preheating". This process is repeated to complete the traversal of all time points, and finally a complete path tracing data containing the state, maximum cumulative probability and corresponding forward associated state of each time point is formed. During the traversal, the preset state probability confidence threshold is 0.5. For the state probability of each fermentation stage at each time point, only the state with a probability value higher than this threshold is retained as a valid candidate state, and low-probability invalid states are removed to eliminate interference. After traversing all collected data time points, an algorithm is used to select a continuous optimal state sequence where the state at each time point meets the preset confidence level requirements. This sequence completely corresponds to the stage change trajectory of the fermentation process from the start to the current moment, such as "t1-preheating (0.4, none) → t2-acceleration (0.5, t1 preheating) → t3-acceleration (0.6, t2 acceleration) → ...". After obtaining the above continuous optimal state sequence where the state at each time point meets the preset confidence level requirements, the state information in the sequence that precisely corresponds to the current fermentation monitoring time point is extracted, and this state is directly determined as the preliminary judgment result of the current stage of the fermentation process. Specifically, if the current fermentation monitoring time point is the 6th hour, and the corresponding node in the optimal state sequence is "t6: state - stable, maximum cumulative probability -0.7, forward correlation state - t5 acceleration", then the stable stage is directly taken as the preliminary judgment result of the current stage of the fermentation process. This method can effectively avoid misjudgments caused by fluctuations in the probability of a single point in time, improve the stability and reliability of the stage judgment results, and provide an accurate basis for subsequent comparison and verification.

[0048] Step S4: Based on the current stage, assess the interaction between temperature and humidity and environmental fluctuations, and generate temperature and humidity coordinated control strategy parameters corresponding to the current stage.

[0049] In step S4, generating the temperature and humidity synergistic control strategy parameters corresponding to the current stage includes: calculating the proportion of the influence of temperature on the fermentation rate at the current stage using preset temperature and humidity analysis data; evaluating the influence of humidity fluctuations on the rate fluctuation amplitude based on the humidity distribution characteristic sequence, and generating a quantitative influence factor of humidity fluctuations on the fermentation rate fluctuation amplitude; generating environmental adaptability assessment results based on the influence proportion and the quantitative influence factor; determining the stage-specific control strategy parameters based on the assessment results; comparing the control strategy parameters with historical control data; adjusting the applicable range of the parameters; and outputting the adjusted control strategy parameters.

[0050] Specifically, a systematic assessment of the interaction between temperature and humidity and environmental fluctuations at the current fermentation stage is conducted to generate temperature and humidity synergistic control strategy parameters that are precisely matched to the current stage, providing a quantitative basis for real-time control of the fermentation environment. First, a pre-set temperature and humidity analysis database is retrieved. This database pre-stores multiple sets of parallel temperature gradient experimental data, synchronous monitoring data of fermentation rate, and characteristic parameters of temperature and humidity interaction corresponding to each fermentation stage of Jiuqu Hongmei. Based on this database, the controlled variable method was used to calculate the proportion of the influence of temperature on the fermentation rate at the current stage. The specific operation is as follows: The baseline humidity and other fermentation environmental parameters at the current stage were kept constant. Five equidistant temperature gradient points within the typical temperature range of the current stage were selected, such as 25℃, 27℃, 29℃, 31℃, and 33℃. The mean fermentation rate corresponding to each temperature gradient was extracted, and the difference in fermentation rate between adjacent temperature gradients was calculated. The average value of each difference was taken as the change in fermentation rate corresponding to a unit temperature change. The ratio of this change to the baseline fermentation rate at the current stage was then used to determine the proportion of the influence of temperature on the fermentation rate. This proportion directly represents the weight of temperature regulation on the fermentation process at the current stage; the higher the proportion, the higher the priority of temperature regulation. Subsequently, based on the extracted humidity distribution feature sequence, the impact of humidity fluctuations on the amplitude of fermentation rate fluctuations was evaluated, generating a quantitative influence factor of humidity fluctuations on the amplitude of fermentation rate fluctuations. Specifically, humidity distribution difference values ​​at consecutive time points are extracted from the humidity distribution characteristic sequence, and the standard deviation of these differences is calculated to characterize the humidity fluctuation amplitude. At the same time, the fluctuation data of fermentation rate within the corresponding time period are extracted, and the standard deviation of fermentation rate is calculated as an indicator of rate fluctuation amplitude. A linear correlation model between humidity fluctuation amplitude and rate fluctuation amplitude is established using the Pearson correlation analysis method. The correlation coefficient is obtained through model fitting, and the correlation coefficient is normalized and used as a quantitative influencing factor. The quantitative influencing factor ranges from 0 to 1. The larger the value, the stronger the interference of humidity fluctuation on the stability of fermentation rate at the current stage, and the higher the humidity uniformity control level needs to be.Subsequently, based on the calculated influence ratio of temperature on fermentation rate and the quantified influence factor of humidity fluctuation, combined with the preset fermentation rate target range for the current stage (e.g., 0.2-0.35 for the accelerated stage), an environmental adaptability assessment result is generated. This assessment result includes three core components: First, the priority of temperature control, determined according to the influence ratio of temperature: >40% is set as Level 1 priority, 20%-40% as Level 2 priority, and <20% as Level 3 priority; Second, the level of humidity uniformity control requirement, determined according to the quantified influence factor: >0.6 is set as high requirement level, 0.3-0.6 as medium requirement level, and <0.3 as low requirement level; Third, the environmental fluctuation tolerance threshold, combined with the allowable range of rate fluctuations for the current stage, the maximum allowable value for a single temperature fluctuation (e.g., ±0.5℃ for Level 1 priority, ±1℃ for Level 2) and the maximum allowable threshold for humidity distribution difference (e.g., <3% for High requirement level) are derived in reverse. Figure 3 The bar chart shown is a comparison of the weights of temperature and humidity on the fermentation rate, illustrating the distribution of the weights of temperature and humidity on the fermentation rate under different dimensions. The horizontal axis represents the influencing dimensions, the vertical axis represents the influencing weights, and the values ​​in parentheses represent the contribution rate of each dimension's weight. Dark bars correspond to the temperature influence dimension, and light bars correspond to the humidity influence dimension. As can be seen from the chart, temperature has the highest weight (0.42) and contribution rate (31.4%), followed by humidity, which has the second highest weight (0.28) and contribution rate (20.7%). The weights of the remaining dimensions decrease sequentially. This distribution clearly indicates the control priority of temperature and humidity dimensions at the current fermentation stage. The direct influence of temperature is the core control direction, while the direct influence of humidity is the secondary control direction, providing a quantitative basis for generating parameters for the subsequent temperature and humidity synergistic control strategy.

[0051] Based on the above environmental adaptability assessment results, phased control strategy parameters were determined. The parameter system covers core temperature control parameters, such as target temperature range, single adjustment range, and adjustment interval; core humidity control parameters, such as target humidity range, humidity distribution uniformity control standards, and humidity compensation rate; and temperature and humidity synergy rules, such as the synchronous compensation coefficient for humidity after temperature adjustment. In addition, it includes weighted parameters for the influence of temperature and humidity on the fermentation rate. The temperature influence weight corresponds to the previously calculated proportion of temperature's influence on the fermentation rate, and the humidity influence weight corresponds to the quantitative influence factor of humidity fluctuations on the amplitude of fermentation rate fluctuations. Subsequently, the generated phased control strategy parameters were compared and verified with a pre-set historical control database. This historical database stores the optimal control parameter ranges for each stage of multiple standard fermentation batches, the corresponding fermentation rate change curves, and the final fermentation quality test data. By calculating the Euclidean distance between the current control parameters and the historical best parameters, one or more sets of historical parameter cases with the smallest distance are selected. Referring to the correspondence between parameters and fermentation quality in these cases, the applicable range of the current control parameters is adjusted. For example, if historical cases show that the fermented product has a higher content of tea polyphenols within a certain temperature range, the current temperature control range is narrowed by 0.5-1℃ towards that range. If historical data indicates that fermentation batch stability is higher under a certain humidity uniformity standard, the current humidity uniformity control standard is further tightened. Finally, the adjusted control strategy parameters are validated for rationality, ensuring that the parameters meet the operating parameter limitations of the current fermentation equipment and process safety requirements. After successful validation, the final temperature and humidity coordinated control strategy parameters are output. The parameters are presented in the form of quantified numerical ranges and linkage rule tables, which can be directly imported into the fermentation environment control system to achieve precise coordinated control of temperature and humidity, ensuring stable progress in the current fermentation stage and improving the consistency of fermentation quality.

[0052] Step S5: Construct a fermentation environment fine-tuning scheme based on the control strategy parameters, and perform historical verification and secondary calibration on the fermentation environment fine-tuning scheme to generate new control instructions.

[0053] Step S5, which involves constructing a fermentation environment fine-tuning scheme based on the control strategy parameters, includes: determining the priority order of temperature and humidity control at the current stage based on the influence weights of temperature and humidity on the fermentation rate in the control strategy parameters and the rate change trends at each stage on the rate change time axis; calculating the dynamic correlation matrix between temperature and humidity for the priority, and generating a temporary control scheme containing target temperature and humidity values ​​and adjustment timing based on the correlation matrix; determining the trajectory balance benchmark value that conforms to the preset process path based on the temporary control scheme and historical fermentation trajectory data; comparing the trajectory balance benchmark value with the real-time collected temperature and humidity data to obtain the deviation sequence, and generating a trajectory plan for adjustment at each time period based on the deviation sequence, and outputting the trajectory plan as the basis for the fine-tuning scheme.

[0054] Specifically, the influence weights of temperature and humidity on the fermentation rate are extracted from the control strategy parameters. Combined with the rate change trends at each stage on the rate change time axis, the priority order of temperature and humidity control at the current stage is determined. The influence weights are the quantified proportions of the effects of temperature and humidity on the fermentation rate in the control strategy parameters, and the rate change trends are the increasing and decreasing patterns shown on the rate change time axis. When the influence weight of temperature is higher than that of humidity, temperature is prioritized to quickly stabilize the fermentation rate; conversely, humidity is prioritized. If the two weights are close, for example, when the interpolation value of temperature and humidity weights is less than 0.1, it means that the influence of temperature and humidity on the fermentation rate is relatively small. When the weight differences are minimal, the priority is dynamically adjusted according to the rate change trend. The rate change trend is obtained by calculating the modulus change rate of the recent fermentation rate change feature vector. When the rate of change exceeds a preset threshold, such as an increase of 0.05 per minute, it indicates that the fermentation system has accumulated too much heat, and the rate is judged to be rising too fast. In this case, the temperature is adjusted first. When the rate of change is lower than another threshold, such as a decrease of 0.03 per minute, it indicates that the humidity of the fermentation system is insufficient. Low humidity leads to a rapid decline in microbial activity, and the rate is judged to be falling too fast. In this case, the humidity is adjusted first to ensure microbial activity and avoid incomplete fermentation.

[0055] Next, based on the priority, a dynamic correlation matrix between temperature and humidity is calculated. This dynamic correlation matrix is ​​a two-dimensional matrix that characterizes the degree of mutual influence between temperature and humidity under different control priorities. Its matrix dimensions match the monitoring dimensions of the temperature and humidity control parameters. For example, a 3×3 matrix is ​​constructed with the macroscopic temperature, macroscopic humidity, and internal humidity distribution characteristics of the fermentation environment as the core dimensions. The rows and columns of the matrix correspond to the three parameter dimensions of macroscopic temperature, macroscopic humidity, and internal humidity distribution characteristics, respectively. The elements on the diagonal of the matrix are the autocorrelation coefficients of each parameter, reflecting the continuity and stability of the fluctuation of a single parameter during the control process. The elements off-diagonal are the cross-correlation coefficients between any two different parameters. Specifically, this is obtained by calculating the covariance of the change series of two target parameters under the same control priority scenario in historical fermentation data and dividing it by the product of the standard deviations of the change series of the two parameters, thereby quantifying the intensity of the interaction between different parameters.

[0056] A temporary control scheme containing target values ​​for temperature and humidity and adjustment timing is generated based on the dynamic correlation matrix. Combining core parameters such as the target temperature range and target humidity range in the control strategy parameters, and using the interaction strength of parameters in the correlation matrix as constraints, specific target values ​​for each priority control parameter are determined. For high-priority parameters, target values ​​are set according to the accuracy requirements of the control strategy parameters. For low-priority parameters, collaborative target values ​​are set based on the cross-correlation coefficient with high-priority parameters. For example, if the current temperature is determined to be a high-priority control parameter, and the temperature accuracy requirement in the control strategy parameters is ±0.3℃ and the target temperature range is 28-30℃, then the target temperature value is directly set to 29℃. The system strictly adheres to an accuracy requirement of ±0.3℃. If humidity is a low-priority parameter, the cross-correlation coefficient between humidity and temperature is calculated to be 0.85 using the dynamic correlation matrix. Based on the target temperature of 29℃ and the cross-correlation coefficient, the target humidity is set to 68%. This target value is determined based on the average humidity value corresponding to the stable fermentation rate at 29℃ in historical data, while also adapting to the temperature-humidity synergy represented by the cross-correlation coefficient. At the same time, according to the adjustment interval and compensation rate requirements in the control strategy parameters, the adjustment sequence of each parameter is planned to ensure that high-priority parameters are adjusted first, and low-priority parameters are adjusted synchronously and collaboratively, avoiding fermentation fluctuations caused by interaction interference between parameters.

[0057] Based on this, a pre-set historical fermentation trajectory database is retrieved. This database stores temperature and humidity change trajectories, fermentation rate change trajectories, and corresponding fermentation quality data for various stages of multiple standard fermentation batches. The temperature and humidity control trajectory corresponding to the temporary control scheme is matched with the historical fermentation trajectory data. Historical trajectory segments consistent with the current fermentation stage, rate change trend, and control priority are selected. The temperature and humidity balance parameters when the fermentation rate is stable within the target range in the historical trajectory are extracted and determined as the trajectory balance benchmark value that conforms to the pre-set process path. Then, the trajectory balance benchmark value is compared with the real-time collected temperature and humidity data point by point. The deviation value between the real-time data and the benchmark value at each time point is calculated and arranged in chronological order to form a deviation sequence. Based on the deviation sequence, a proportional-integral-derivative (PID) algorithm is used to generate a trajectory plan that is adjusted for each time period. Specifically, the parameter adjustment amount for each time period is calculated through the proportional, integral, and derivative terms of the deviation sequence to correct the temperature and humidity target values ​​and adjustment sequence in the temporary control scheme, so that the trajectory plan can dynamically adapt to the fluctuations of the real-time fermentation environment. Finally, this trajectory plan is output as the basis for the fine-tuning scheme, providing a core scheme carrier for subsequent historical verification and secondary calibration.

[0058] By comparing the fine-tuning scheme with historical fermentation control data for historical verification, if there is a risk of abnormal rate deviation, the temperature and humidity interaction parameters are calibrated a second time, and the scheme is adjusted based on the calibration results to generate new control instructions. This solves the problems of insufficient targeting and easy deviation from the process path of traditional fine-tuning schemes, ensuring that the fine-tuning scheme accurately matches the needs of the current fermentation stage, providing reliable support for the generation of subsequent control instructions, and effectively improving the accuracy of fermentation environment control.

[0059] In step S5, the fermentation environment fine-tuning scheme undergoes historical verification and secondary calibration to generate new control instructions. This includes: aligning the temperature and humidity target value sequence corresponding to the trajectory planning in the current fine-tuning scheme with the historical fermentation trajectory data of the same stage in terms of time sequence and similarity; calculating the probability that the implementation of the scheme will cause abnormal deviation in the fermentation rate based on the matching deviation; if the probability exceeds a preset safety threshold, performing secondary calibration on the temperature and humidity control parameters in the fine-tuning scheme to address the interaction between temperature and humidity; updating the fine-tuning scheme using the calibrated parameters to generate fermentation environment fine-tuning instructions containing specific temperature and humidity target values ​​and execution time sequences; comparing and verifying the fine-tuning instructions with the effective instruction set in the historical control records; and outputting the final executable control instructions after confirming that they meet the process constraints and stability requirements.

[0060] Specifically, after constructing the fermentation environment fine-tuning scheme, the reliability of the scheme needs to be evaluated through historical data verification. Schemes with potential risks require parameter calibration. Once verified, executable control commands are generated to ensure precise control actions that meet process requirements. Specifically, the temperature and humidity target value sequence corresponding to the trajectory planning in the current fine-tuning scheme is first time-aligned and similarity-matched with historical fermentation trajectory data from the same stage. The temperature and humidity target value sequence corresponding to the trajectory planning is the core content of the fine-tuning scheme, including temperature and humidity target values ​​for each control period. These target values ​​are arranged in chronological order to form a continuous sequence. The historical fermentation trajectory data from the same stage consists of temperature and humidity control parameter sequences and corresponding fermentation rate change sequences from multiple batches of qualified Jiuqu Hongmei fermentation. Time alignment uses a dynamic time warping algorithm to adjust the time axes of the two sequences, ensuring precise correspondence at key time nodes. Similarity matching quantifies the similarity by calculating the cumulative distance between the aligned sequences; a smaller cumulative distance indicates higher similarity. Figure 4As shown in the figure, this is a similarity matching comparison chart between the target temperature trajectory and the historical best trajectory. It intuitively shows the matching status between the current target temperature trajectory and the historical best temperature trajectory. In the figure, the light-colored curve represents the historical best temperature trajectory, the dark-colored curve represents the current target temperature trajectory, and the gray area between the two is the temperature matching deviation. As can be seen from the figure, the changing trends of the current target temperature trajectory and the historical best temperature trajectory are highly consistent, and the temperature matching deviation is always kept within a small range. This indicates that the method of determining the trajectory balance benchmark value through the dynamic time warping algorithm in this embodiment can make the currently generated target temperature trajectory closely match the historical best fermentation trajectory. This verifies the effectiveness of this method in ensuring that the fermentation trajectory conforms to the preset process path. It further illustrates that the method of this embodiment can make the temperature and humidity control trajectory of the fermentation environment consistent with the historical trajectory of high-quality fermentation, effectively reducing the risk of the fermentation process deviating from the optimal process path and ensuring the fermentation quality of Jiuqu Hongmei.

[0061] The probability of abnormal fermentation rate deviation after the implementation of this scheme is calculated based on the matching deviation. The matching deviation is the set of differences between the corresponding nodes of the current trajectory planning temperature and humidity target value sequence and the historical optimal trajectory parameter sequence after time-series alignment. By statistically analyzing the frequency of abnormal rate deviations corresponding to different matching deviations in historical data, an abnormal rate deviation is determined when the fermentation rate exceeds a preset fluctuation range during historical fermentation processes. A mapping relationship between deviation and deviation probability is established. Substituting the current matching deviation yields the corresponding probability of abnormal rate deviation. The safety threshold is set based on historical experience, for example, 0.25, which means that a scheme with a probability exceeding 25% is considered to have significant risk.

[0062] If the probability exceeds the preset safety threshold, a secondary calibration of the temperature and humidity control parameters in the fine-tuning scheme is performed to address the interaction between temperature and humidity. This interaction refers to the effect of temperature changes on humidity distribution and the inverse effect of humidity fluctuations on temperature control effectiveness. Specifically, the temperature and humidity distribution characteristics in the dynamic correlation matrix are first extracted to clarify the coupling relationship between temperature and humidity at the current stage. For example, if the cross-correlation coefficient between temperature and macro-humidity in the matrix is ​​-0.7, it indicates that for every 1°C increase in temperature, macro-humidity is likely to decrease by 0.7 units, serving as the quantitative basis for calibration. The source of risk is then located by combining the matching deviation, determining whether the risk is caused by unreasonable temperature target values ​​leading to humidity fluctuations, or by deviations in humidity target values ​​causing an imbalance in temperature control effectiveness. For instance, if the matching deviation is concentrated in periods with high temperature target values, and the predicted humidity value for the corresponding period is lower than the humidity value of the historical optimal trajectory, then compensation calibration is needed to address the insufficient humidity caused by the high temperature. The specific calibration compensation amount is calculated based on the cross-correlation coefficient of the dynamic correlation matrix, combined with the current temperature... The deviation between the target humidity value and the historical best trajectory is used to derive the compensation value. The formula logic is: Compensation value = Temperature (humidity) target deviation × Corresponding cross-correlation coefficient. For example, if the target temperature value is 2℃ higher than the historical best value, and the corresponding cross-correlation coefficient is -0.7, then 1.4 units of compensation (2℃ × 0.7) need to be added to the target humidity value for the corresponding period to offset the impact of the decrease in humidity caused by the higher temperature. Finally, the calibration effect is iteratively verified. The calculated compensation value is substituted into the temperature and humidity target value sequence, the fine-tuning scheme is updated, and the matching deviation and the corresponding rate abnormal deviation probability are recalculated. If the probability is still higher than the safety threshold, the correction coefficient of the compensation value is adjusted. The correction coefficient is set based on historical calibration experience, and the calculation is iterated again until the rate abnormal deviation probability drops below the safety threshold, completing the second calibration.

[0063] The fine-tuning scheme is updated using calibrated parameters, generating fermentation environment fine-tuning instructions containing specific temperature and humidity target values ​​and execution sequences. The execution sequence refers to the order and time interval of each temperature and humidity adjustment action, ensuring orderly progress of the control actions. The fine-tuning instructions are compared and verified with the effective instruction set in historical control records. The effective instruction set consists of control instructions that have been verified in historical fermentation processes to ensure smooth fermentation. Verification is achieved by calculating the matching degree between the current fine-tuning instructions and each instruction in the effective instruction set. Simultaneously, it verifies whether the temperature and humidity target values ​​in the instructions conform to the preset constraints of the Jiuqu Hongmei fermentation process, such as a temperature range of 22-28℃ and a relative humidity range of 75-90%, and meets stability requirements. After confirming that the fine-tuning instructions meet the process constraints and control effectiveness requirements, the final executable control instructions are output. Through historical verification, secondary calibration, and verification closed-loop, potential risks of the fine-tuning scheme are accurately identified and avoided, solving the problems of traditional control instructions easily deviating from process requirements and lacking stability. This ensures that the output control instructions are scientifically reliable and guarantee the smooth progress of the Jiuqu Hongmei fermentation process according to the preset process path.

[0064] Step S6: Output temperature and humidity control signals according to the control instructions, and track the fermentation rate trajectory. If it deviates from the preset path, a correction signal will be generated.

[0065] Step S6, which involves outputting temperature and humidity control signals according to control instructions and tracking the fermentation rate trajectory, includes: outputting specific temperature and humidity control signals to the control system according to control instructions; adjusting the real-time state of the fermentation environment and tracking the fermentation rate trajectory in real time through the control signals; updating the prediction range of the fermentation rate trajectory by combining historical fermentation process parameter records, wherein the prediction range is the allowable range of fermentation rate fluctuations; comparing the real-time tracked fermentation rate trajectory with the updated prediction range to determine whether it conforms to the preset path; if it does not conform, calculating the correction amount based on the degree of deviation and generating a deviation correction signal, and feeding the deviation correction signal back to the control system for adjustment.

[0066] Specifically, according to the control command, specific temperature and humidity control signals are output to the control system. The temperature and humidity target values ​​contained in the control command need to be converted into a digital signal format that the control system can recognize. The conversion process is realized through a preset signal conversion module. This module converts the analog quantity requirements of the target temperature and humidity into the pulse width modulation signal or digital control signal of the corresponding device, and then transmits the converted signal to the main controller of the control system through the communication interface to ensure the accuracy and real-time performance of the signal transmission. For example, for the temperature-dominant control command in the early stage of Jiuqu Hongmei fermentation, if the target temperature is 26℃, the signal conversion module converts the temperature value into the digital control signal of the corresponding heating device and transmits it to the main controller to activate the heating device.

[0067] Next, the real-time state of the fermentation environment is adjusted by controlling signals. After receiving the signal, the control system activates the corresponding heating, humidification, and ventilation equipment in the fermentation chamber according to the signal command, changing the temperature and humidity parameters of the fermentation environment. At the same time, sensors continuously collect the adjusted environmental temperature and humidity data to monitor whether the environmental state reaches the temperature and humidity level specified by the command. If it does not reach the level, the operating parameters of the equipment are continuously adjusted until the environmental state meets the requirements. During this process, the key process characteristic trajectory is calculated in real time. Through the continuously acquired optimized temperature and humidity sequence and humidity distribution characteristic sequence, the standardized fermentation rate change feature vector is calculated in real time. Principal component analysis is performed on the standardized fermentation rate change feature vector. Specifically, firstly, a feature matrix is ​​constructed for the standardized fermentation rate change feature vector at all times, where the rows of the matrix represent the time dimension and the columns represent the feature dimension. The covariance matrix of this feature matrix is ​​calculated to quantify the linear correlation between different feature dimensions. The covariance matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvalues ​​and corresponding eigenvectors. The eigenvector with the largest eigenvalue is selected as the projection direction of the primary principal component. The standardized fermentation rate change feature vector is projected onto the projection direction of the primary principal component, and the resulting projection value is the primary principal component score. The primary principal component score is used as the key process characteristic value. The primary principal component scores calculated at different time points are arranged chronologically to obtain a key process characteristic value sequence. After smoothing this sequence, a continuous key process characteristic trajectory is generated, which intuitively reflects the dynamic trend of the overall rise and fall and fluctuation of the fermentation rate during the fermentation process. Figure 5 As shown in the figure, this is a comparison chart verifying the effect of the control command execution. It shows the trend of the fermentation environment temperature before and after the control command execution. The solid line in the figure represents the temperature before control, the dashed line represents the temperature target value, and the dotted line represents the execution time of the control command. As can be seen from the figure, in the first 24 hours before the control command was executed, the fermentation environment temperature fluctuated greatly and was generally higher than the target value. However, after the control command was executed 24 hours later, the temperature quickly converged to the target value and remained near the target value with slight fluctuations. This shows that the control command generated by the temperature and humidity coordinated control method provided in this application embodiment can effectively stabilize the fermentation environment temperature within the preset target range, verifying the effectiveness of this method in the precise control of temperature and humidity during the fermentation process. It shows that it can stabilize the temperature and humidity parameters of the fermentation environment within the target range, providing support for ensuring that the fermentation process proceeds according to the preset path.

[0068] Subsequently, based on historical fermentation process parameter records, the prediction interval for the fermentation rate trajectory is updated, and the prediction interval is the allowable range of process characteristic fluctuations. Historical fermentation process parameter records include temperature and humidity sequences, humidity distribution characteristics, and corresponding standardized fermentation rate change feature vector sequences from multiple batches of fermentation. When updating the prediction interval, the historical key process characteristic value sequence is first seasonally decomposed. This method uses the least squares method to fit a trend line and separate the trend component. The trend component refers to the continuous upward, downward, or stable change pattern of the sequence over a time scale, reflecting the core change trend of the Jiuqu Hongmei fermentation rate over time. The seasonal index is calculated to quantify periodic fluctuations, decomposing the time series into three parts: trend, seasonality, and residual. Then, based on the decomposition results and current control instructions, an autoregressive integral moving average model is used to predict its characteristic trend components. Specifically, firstly, based on the trend components obtained from the decomposition, combined with the temperature and humidity target value range in the current control instructions, a prediction baseline for the trend components is determined. For example, the current control instructions require... If the temperature is maintained at 25-27℃, the trend component corresponding to the same temperature range in historical data is selected as the benchmark. The stationarity of the trend component sequence is tested. If the sequence is non-stationary, the trend is eliminated through differencing to obtain a stationary differencing sequence. The autoregressive order and moving average order of the autoregressive integral moving average model are determined based on the Akaike Information Criterion, a statistical criterion used for model selection and order determination to screen for optimal model parameters. The model parameters are fitted using historical differencing sequences. The time dimension corresponding to the current control command is substituted into the fitted model to predict the trend component prediction value for future periods. Combining the seasonal component obtained from seasonal decomposition with the fluctuation range of the residual, the upper and lower fluctuation boundaries of the trend component prediction value are calculated. Finally, the fluctuation range of the key process characteristic value for future periods is generated. This range is directly related to the fermentation rate and is used to define the allowable fluctuation range of the fermentation rate.

[0069] The real-time tracked fermentation rate trajectory is compared with the updated prediction interval to determine if it conforms to the preset path. If the real-time trajectory point falls within the prediction interval, it is determined to conform to the preset path; if the trajectory point exceeds the prediction interval, it is determined not to conform. If it is determined not to conform to the preset path, a correction amount is calculated based on the degree of deviation, and a deviation correction signal is generated. The degree of deviation is determined by calculating the deviation between the real-time characteristic value and the median of the prediction interval. The correction amount is calculated using proportional control rules, for example: temperature correction amount = characteristic value deviation × temperature proportional coefficient; humidity correction amount = characteristic value deviation × humidity proportional coefficient. The proportional coefficient is obtained through regression analysis of historical data. After the correction amount is converted into a deviation correction signal that the control system can recognize, it is fed back to the control system for adjustment. The control system fine-tunes the operating parameters of heating, humidification, and other equipment according to the signal until the key process characteristic trajectory returns to the prediction interval and conforms to the preset path requirements. By accurately converting control commands into equipment control signals and tracking the fermentation rate in real time, and dynamically updating the prediction interval based on historical data, closed-loop feedback control of the fermentation process is achieved, significantly improving the dynamic response capability and process stability of the control, ensuring that the fermentation process always proceeds along the optimal process path.

[0070] In summary, this application provides a temperature and humidity coordinated control method based on the fermentation process of Jiuqu Hongmei (a type of plum). It utilizes multiple sets of temperature and humidity sensors and a distributed humidity sensor array to collect real-time temperature and humidity data of the fermentation environment and humidity distribution signals within the fermentation pile. This method innovatively employs adaptive moving average filtering and variational mode decomposition techniques for data preprocessing and feature extraction, combining ensemble empirical mode decomposition and hidden Markov chain models to accurately identify fermentation stages. Coordinated control parameters are generated based on the evaluation of the interaction between temperature and humidity. A fine-tuning scheme is constructed through historical trajectory matching verification and secondary calibration. Finally, a closed-loop feedback control outputs the control signal and dynamically tracks the fermentation rate trajectory. This method effectively solves the problems of incomplete data, inaccurate stage identification, and insufficient strategy targeting in traditional control methods, significantly improving the accuracy and stability of fermentation temperature and humidity control, and providing reliable technical support for ensuring the consistency of Jiuqu Hongmei fermentation quality and intelligent production.

[0071] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for coordinated temperature and humidity control based on the fermentation process of Jiuqu Hongmei plum, characterized in that, The method includes: Step S1: In the fermentation environment of Jiuqu Hongmei, real-time temperature data, humidity data, and humidity distribution signals reflecting the interior of Jiuqu Hongmei are collected by sensors; Step S2: Preprocess and extract features from the temperature data, humidity data, and humidity distribution signal to obtain temperature and humidity sequences and humidity distribution feature sequences; Step S3: Extract the fermentation rate change features based on the temperature and humidity sequence, and identify the stage transition markers by combining the humidity distribution feature sequence to determine the current stage of the fermentation process; The process of determining the current stage of fermentation includes: analyzing the temperature and humidity sequence using time series decomposition technology to extract fermentation rate change characteristics; calculating the feature difference value of adjacent time nodes in the humidity distribution feature sequence as a humidity distribution feature; when the humidity distribution feature exceeds a preset threshold, marking the corresponding time node as a candidate stage transition identifier; associating and matching the rate mutation points in the fermentation rate change characteristics with the candidate stage transition identifiers. The association and matching is achieved by calculating the time interval between the rate mutation point and the candidate stage transition identifier. Each candidate stage transition identifier is compared with the timestamps of all rate mutation points. If the time difference between a rate mutation point and a candidate identifier is less than the time interval threshold, the candidate stage transition identifier is retained; otherwise, isolated candidate identifiers are removed, and finally, effective stage transition identifiers are obtained. A fermentation process stage segmentation model is constructed based on the effective stage transition identifier. The fermentation rate change features and humidity distribution features extracted in real time are input into the segmentation model, and a preliminary judgment result of the current fermentation process stage is output. The preliminary judgment result includes the specific stage name of the current fermentation of Jiuqu Hongmei and the standardized fermentation rate change features and humidity distribution features extracted in real time. A preset feature library of each stage of Jiuqu Hongmei fermentation is retrieved, and the preliminary judgment result is compared with the stage feature parameters in the feature library. The similarity between the observed feature vector in the preliminary judgment result and the corresponding stage feature parameter vector in the feature library is calculated using Euclidean distance as the matching degree. If the matching degree exceeds a preset threshold, the stage corresponding to the preliminary judgment result is confirmed as the current stage of the fermentation process, and an identifier information including the stage name and feature parameter range is output. Step S4: Based on the current stage, assess the interaction between temperature and humidity and environmental fluctuations, and generate temperature and humidity coordinated control strategy parameters corresponding to the current stage; Step S5: Construct a fermentation environment fine-tuning scheme based on the control strategy parameters, and perform historical verification and secondary calibration on the fermentation environment fine-tuning scheme to generate new control instructions; Step S6: Output temperature and humidity control signals according to the control instructions, and track the fermentation rate trajectory. If it deviates from the preset path, a correction signal will be generated.

2. The method according to claim 1, characterized in that, Step S2 involves preprocessing and feature extraction of the temperature data, humidity data, and humidity distribution signal, including: An adaptive moving average smoothing algorithm is used to filter and denoise the temperature and humidity data to generate a denoised temperature and humidity sequence. For the humidity distribution signal, variational mode decomposition technology is used to decompose the signal, and after separating the effective component and the interference component, the effective component is extracted as the initial humidity distribution feature sequence. Based on the sensor acquisition timestamps, the denoised temperature and humidity sequence is time-aligned with the initial humidity distribution feature sequence to generate unified time series data; The variance of the initial humidity distribution characteristic sequence is calculated to obtain a stability measure of the humidity distribution within the Jiuquhongmei plum tree. The uniform time series data is then subjected to secondary correction using the stability metric to remove abnormal data points caused by abrupt changes in humidity distribution, resulting in an optimized temperature and humidity sequence and humidity distribution feature sequence.

3. The method according to claim 1, characterized in that, Step S3, which involves extracting fermentation rate change features based on the temperature and humidity sequence, includes: The temperature and humidity sequence was decomposed into time series components at different time scales using ensemble empirical mode decomposition technology. Calculate the amplitude change rate and peak interval of each fluctuation component, and construct a fermentation rate feature vector based on the amplitude change rate and peak interval; The fermentation rate feature vector is mapped to a preset feature space by normalization to obtain standardized fermentation rate change features.

4. The method according to claim 1, characterized in that, The preliminary assessment of the current fermentation stage includes: The fermentation rate change features and humidity distribution features extracted in real time are fused to generate an observation feature vector; The observed feature vector is input into the fermentation process stage division model, and the state probability corresponding to each fermentation stage is calculated through the preset parameters in the model. The Viterbi algorithm is used to traverse the state probabilities at all time points to obtain a state sequence whose corresponding state probabilities meet the preset requirements. The state corresponding to the current time point in the state sequence is used as the preliminary judgment result of the current fermentation process stage.

5. The method according to claim 1, characterized in that, Step S4 generates the temperature and humidity coordinated control strategy parameters corresponding to the current stage, including: The influence ratio of temperature on the fermentation rate at the current stage is calculated using preset temperature and humidity analysis data. For the humidity distribution characteristic sequence, the impact of humidity fluctuation on the rate fluctuation amplitude is evaluated, and a quantitative impact factor of humidity fluctuation on the fermentation rate fluctuation amplitude is generated. Based on the stated impact ratio and the stated quantified impact factors, an environmental adaptability assessment result is generated; Based on the evaluation results, phased control strategy parameters are determined, the control strategy parameters are compared with historical control data, the applicable range of the parameters is adjusted, and the adjusted control strategy parameters are output.

6. The method according to claim 1, characterized in that, Step S5, which involves constructing a fermentation environment fine-tuning scheme based on the aforementioned regulation strategy parameters, includes: Based on the influence weights of temperature and humidity on fermentation rate in the control strategy parameters, and combined with the rate change trend of each stage on the rate change time axis, the priority order of temperature and humidity control in the current stage is determined. For the priorities mentioned above, a dynamic correlation matrix between temperature and humidity is calculated, and based on the correlation matrix, a temporary control scheme containing target values ​​for temperature and humidity and adjustment timing is generated. Based on the temporary control scheme and historical fermentation trajectory data, determine the trajectory balance benchmark value that conforms to the preset process path; The trajectory balance benchmark value is compared with the real-time collected temperature and humidity data to obtain the deviation sequence. Based on the deviation sequence, a trajectory plan that is adjusted for each time period is generated, and the trajectory plan is output as the basis for the fine-tuning scheme.

7. The method according to claim 6, characterized in that, Step S5 involves historical verification and secondary calibration of the fermentation environment fine-tuning scheme, generating new control instructions including: The temperature and humidity target value sequence corresponding to the trajectory planning in the current fine-tuning scheme is time-series aligned and similarity matched with the fermentation trajectory data of the same historical stage. Based on the matching deviation, the probability of abnormal deviation of the fermentation rate after the implementation of the scheme is calculated. If the probability exceeds the preset safety threshold, the temperature and humidity control parameters in the fine-tuning scheme are calibrated a second time to address the interaction between temperature and humidity. The fine-tuning scheme is updated using the calibrated parameters to generate a fermentation environment fine-tuning instruction that includes specific temperature and humidity target values ​​and execution timing. The fine-tuning instructions are compared and verified with the valid instruction set in the historical control records. After confirming that they meet the process constraints and stability requirements, the final executable control instructions are output.

8. The method according to claim 1, characterized in that, Step S6, which involves outputting a temperature and humidity control signal according to the control command and tracking the fermentation rate trajectory, includes: According to the control command, output specific temperature and humidity control signals to the control system; The real-time state of the fermentation environment is adjusted through the aforementioned control signal, and the fermentation rate trajectory is tracked in real time. Based on historical fermentation process parameter records, the predicted range of the fermentation rate trajectory is updated, where the predicted range is the allowable range of fermentation rate fluctuations. The real-time tracked fermentation rate trajectory is compared with the updated prediction interval to determine whether it conforms to the preset path. If the deviation does not meet the requirements, a correction amount is calculated based on the degree of deviation, and a deviation correction signal is generated. The deviation correction signal is then fed back to the control system for adjustment.

Images