A system and method for controlling production of a bean paste

Abstract

The application relates to the technical field of control systems, and discloses a soybean paste production control system and method, which are used for solving the quality fluctuation and pollution risk problems caused by insufficient precision control in the existing soybean paste fermentation process; through real-time collection of temperature, humidity, pH value and microbial activity indexes in the fermentation process, a dynamic parameter data set is formed; the deviation degree of the data set is analyzed, the matching degree with a standard fermentation curve is calculated, and potential unstable factors are identified; preliminary adjustment instructions are generated based on the deviation results, and the temperature and humidity are automatically fine-tuned to stabilize the environment; the change of microbial activity after adjustment is monitored, the control effect is verified through sensor feedback, and the activity recovery rate is calculated; the fermentation model parameters are optimized, the control algorithm is iteratively adjusted according to the recovery rate, and self-adaptive precision improvement is realized; a final control report is output, and the optimized parameters are fed back to the system, so that closed-loop precision control is realized.

Description

Technical Field

[0001] This invention relates to the field of control system technology, specifically to a control system and method for the production of fermented soybean paste. Background Technology

[0002] Doubanjiang (fermented broad bean paste) production control system and method are key technologies for the industrialization of traditional fermented foods. By precisely controlling raw material processing, fermentation environment and quality monitoring, the flavor stability and yield optimization of doubanjiang can be achieved, thereby improving food safety and economic benefits. In existing technologies, temperature control and time management methods are usually used to monitor the fermentation process. For example, brine injection and microbial activity can be optimized through closed post-ripening fermentation or modular production systems. This control method improves the quality consistency of doubanjiang to a certain extent, and is especially suitable for large-scale industrial production. It can reduce human error and support batch traceability.

[0003] However, existing control systems and methods for fermented broad bean paste production have significant shortcomings in terms of precision control during the fermentation process. Existing systems largely rely on static parameter settings or manual adjustments, resulting in insufficient real-time response to changes in temperature, humidity, pH, and microbial dynamics, easily leading to uneven fermentation and contamination risks. For example, while patent document CN112239720B (A Closed Post-fermentation Process for Pixian Broad Bean Paste) proposes temperature and time control, its scheme lacks precision when microbial activity fluctuates, causing significant batch-to-batch quality differences. Similarly, patent document CN108208595B (A Method for Producing Broad Bean Paste) employs steaming and fermentation steps, but fails to address precise monitoring under environmental interference, leading to increased contamination rates. These shortcomings not only reduce production efficiency but may also increase food safety risks and economic losses, especially in variable-temperature fermentation or large-scale production environments.

[0004] Therefore, there is an urgent need for a soybean paste production control system and method that can improve the precision control during the fermentation process in order to solve the above-mentioned problems in the existing technology. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a soybean paste production control system and method to solve the problem of insufficient precision control in existing methods.

[0007] (II) Technical Solution

[0008] To achieve the goal of improving precision control during the fermentation process mentioned in the background section, the present invention provides the following technical solution:

[0009] A method for controlling the production of fermented soybean paste includes:

[0010] S1. Real-time acquisition of key parameters during the fermentation process, including temperature, humidity, pH value and microbial activity indicators, to form a dynamic parameter dataset;

[0011] S2. Analyze the degree of deviation of the dynamic parameter dataset, calculate the degree of matching with the standard fermentation curve, and identify potential unstable factors.

[0012] S3. Based on the deviation analysis results, generate preliminary adjustment instructions and automatically fine-tune the temperature and humidity to stabilize the environment;

[0013] S4. Monitor changes in microbial activity after adjustment, verify the control effect through sensor feedback, and calculate the activity recovery rate;

[0014] S5. Optimize fermentation model parameters and iteratively adjust the control algorithm based on the recovery rate to achieve adaptive accuracy improvement;

[0015] S6. Output the final control report and feed back the optimized parameters to the system to achieve closed-loop precision control.

[0016] In a preferred embodiment, key parameters during the fermentation process are collected in real time, including temperature, humidity, pH value, and microbial activity indicators, forming a dynamic parameter dataset, including:

[0017] The temperature sensor collects temperature data at uniformly distributed points determined by finite element thermal conduction simulation. The data source is the voltage signal conversion caused by resistance change.

[0018] The humidity sensor collects humidity parameters, and the data source is the frequency signal caused by the change in capacitance.

[0019] pH electrode collects pH value parameters, and the data source is the potential difference measurement of glass electrode;

[0020] The microbial counter collects microbial activity indicators and counts bacterial community density in real time, with data sources including light scattering signals or microscopic counting.

[0021] The collected data is converted into standardized units to form a dynamic parameter dataset.

[0022] In a preferred embodiment, the degree of deviation of the dynamic parameter dataset is analyzed, the degree of matching with the standard fermentation curve is calculated, and potential instability factors are identified, including:

[0023] Extract the parameter value sequence of the current moment from the dynamic parameter dataset, compare it point by point with the preset standard fermentation curve, and calculate the degree of deviation;

[0024] Dominant factors are identified based on deviation vectors, and potential unstable factors are determined by contribution rates.

[0025] Calculate the degree of fit with the standard fermentation curve, and determine the degree of fit based on the ratio of the average deviation to the maximum permissible deviation;

[0026] Identify potential destabilizing factors by comparing the contribution of deviations parameter by parameter and marking factors with high contributions from a single parameter;

[0027] The rule engine is used to match historical patterns and output a list of factors.

[0028] In a preferred embodiment, based on the deviation analysis results, an initial adjustment command is generated to automatically fine-tune the temperature and humidity to stabilize the environment, including:

[0029] The deviation vector and factor labels are extracted from the deviation analysis results to generate preliminary adjustment instructions, with the instructions prioritized according to the severity of the factors.

[0030] Automatic temperature fine-tuning is achieved by driving the heating equipment to adjust the temperature, with the adjustment range controlled within a specified range and preset safety constraints.

[0031] The humidity is automatically fine-tuned by driving the humidification equipment to adjust the humidity, and the adjustment time is controlled within the specified time limit;

[0032] To verify the fine-tuning effect, the deviation before and after adjustment is compared by comparing feedback data, and the adjusted state is output after confirming that the environment is stable.

[0033] In a preferred embodiment, monitoring changes in the adjusted microbial activity, verifying the control effect through sensor feedback, and calculating the activity recovery rate include:

[0034] After the adjustment is implemented, microbial activity data is collected, and the control effect is verified by reading back through sensors. The feedback value is compared with the target value.

[0035] The activity recovery rate is calculated based on the ratio of current activity to target activity, taking into account weighted enzyme activity and microbial density.

[0036] Verify the effectiveness of the control measures by confirming the effectiveness of the adjustments through deviation reduction.

[0037] In a preferred embodiment, the fermentation model parameters are optimized, the control algorithm is iteratively adjusted based on the recovery rate, and adaptive accuracy is improved, including:

[0038] Extract the current value from the recovery rate calculation results and mark the optimization requirements;

[0039] The input control algorithm model is iteratively adjusted, with the adjustment range kept within safe limits.

[0040] The control algorithm is iteratively adjusted based on the recovery rate, and the parameters are updated based on minimizing the error function to ensure that the accuracy is improved to above the threshold.

[0041] To achieve adaptive accuracy improvement, the effect of parameter application is verified through iterative validation.

[0042] Store the iteration log, including the number of iterations and the improvement rate.

[0043] In a preferred embodiment, a final control report is output, and optimized parameters are fed back to the system, including:

[0044] Key indicators are extracted from the optimization results to generate a final control report, which is then incorporated into the risk prediction for the next batch.

[0045] The optimized parameters are fed back to the system, and the stored parameters are updated through the database to support automatic loading in the next batch.

[0046] Verify the feedback effect and confirm the improved accuracy by simulating the next batch.

[0047] After generating the report, efficiency metrics are recorded to form a closed loop throughout the entire process.

[0048] In a preferred embodiment, a database rollback mechanism is automatically triggered to restore the process parameters to the stable version that passed the verification of the previous batch.

[0049] Simultaneously generate security alert information including rollback timestamp, abnormal parameter identifier, and deviation degree;

[0050] The rollback operation log, security alarm record, control data stream and real-time detection data of this batch are encapsulated together into a structured process record;

[0051] The records are stored tamper-proofly using blockchain technology and a unique mapping relationship is established between the records and the production batch number.

[0052] Based on the abnormal parameter distribution characteristics in the process records, parameter optimization suggestions are generated through machine learning algorithms, which serve as the constraint input for the next round of standard curve calibration.

[0053] On the other hand, the present invention provides a soybean paste production control system, comprising:

[0054] Parameter acquisition module: responsible for real-time acquisition of key parameters during the fermentation process, forming a dynamic parameter dataset, comprehensively sensing the environment and biological state, and supporting the real-time data foundation for subsequent analysis;

[0055] Deviation Analysis Module: Analyzes the degree of deviation in the dynamic parameter dataset, calculates the degree of matching with the standard fermentation curve, and identifies potential instability factors;

[0056] Command generation module: Generates preliminary adjustment commands based on deviation analysis results, and automatically fine-tunes temperature and humidity;

[0057] Activity monitoring module: Monitors changes in microbial activity after adjustment, verifies the control effect through sensor feedback, calculates the activity recovery rate, and provides control verification and abnormal alarms;

[0058] Model optimization module: Optimizes fermentation model parameters, iteratively adjusts the control algorithm based on the recovery rate, and adaptively improves accuracy and learns the system.

[0059] Report output module: Outputs the final control report and feeds back the optimized parameters to the system to complete closed-loop precision control and batch continuous optimization.

[0060] Compared with the prior art, the present invention provides a soybean paste production control system and method, which has the following beneficial effects:

[0061] 1. This invention achieves precise control of key factors such as temperature and humidity through real-time multi-parameter acquisition (e.g., temperature, humidity, pH, and microbial activity) in step S1, deviation analysis in step S2, combined with PID automatic fine-tuning in step S3 and recovery rate monitoring in step S4. Compared with existing methods, this reduces temperature control error and humidity fluctuation, improves microbial activity recovery rate, avoids enzyme activity inhibition and metabolic imbalance in complex fermentation environments, ensures improved batch quality consistency, significantly reduces fermentation failure rate caused by insufficient precision, and also reduces production waste loss rate. This significantly improves the precision control of the fermentation process and solves the problem of insufficient precision control in existing methods.

[0062] 2. This invention achieves adaptive adjustment of PID parameters by constructing an automated closed-loop system with S5 gradient descent iterative optimization and S6 closed-loop feedback mechanism, which can maintain stability under environmental disturbances caused by external temperature changes. Compared with existing manual adjustment technology, it effectively eliminates the risk of contamination and quality fluctuations caused by human error, significantly improves process robustness, and provides a stable and reliable control solution for large-scale industrialization.

[0063] 3. This invention optimizes resource allocation by introducing a closed-loop control throughout the entire process. In scenarios where insufficient precision control leads to prolonged fermentation cycles and increased energy consumption, it improves fermentation efficiency within the standard cycle, reduces manual intervention and labor costs, and lowers the scrap rate while improving quality. This results in reduced economic losses and, while maintaining product consistency, reduces overall energy consumption, thereby improving production efficiency and economic benefits and solving the problems of resource waste and increased costs caused by insufficient precision. Attached Figure Description

[0064] Figure 1 This is a flowchart of a method for controlling the production of fermented soybean paste according to the present invention;

[0065] Figure 2This is a schematic diagram of the structure of a soybean paste production control system according to the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Example 1: Figure 1 This invention provides a method for controlling the production of fermented soybean paste, comprising:

[0068] S1. Real-time acquisition of key parameters during the fermentation process, including temperature, humidity, pH value and microbial activity indicators, to form a dynamic parameter dataset;

[0069] S2. Analyze the degree of deviation of the dynamic parameter dataset, calculate the degree of matching with the standard fermentation curve, and identify potential unstable factors.

[0070] S3. Based on the deviation analysis results, generate preliminary adjustment instructions and automatically fine-tune the temperature and humidity to stabilize the environment;

[0071] S4. Monitor changes in microbial activity after adjustment, verify the control effect through sensor feedback, and calculate the activity recovery rate;

[0072] S5. Optimize fermentation model parameters, iteratively adjust the control algorithm based on the recovery rate, and perform adaptive accuracy improvement;

[0073] S6. Output the final control report and feed the optimized parameters back to the system.

[0074] S1. Real-time acquisition of key parameters during the fermentation process, including temperature, humidity, pH value, and microbial activity indicators, to form a dynamic parameter dataset. Specific implementation details are as follows:

[0075] In the precision control of fermentation of fermented soybean paste based on real-time monitoring, key parameters during the fermentation process are first collected in real time, including temperature, humidity, pH value, and microbial activity indicators, forming a dynamic parameter dataset. This enables comprehensive perception of the fermentation environment and data preparation, ensuring the accuracy and timeliness of subsequent deviation analysis. Key parameters include temperature sensors for monitoring thermal balance, humidity sensors for controlling moisture evaporation, pH electrodes for assessing pH changes, and microbial activity indicators from microbial counters for quantifying bacterial density. The selection of parameter sources is based on the biochemical characteristics of the fermentation process. The acquisition frequency is uniformly set to 1Hz to balance real-time performance and data processing load. The frequency setting is based on the statistical distribution of dynamic changes in fermentation to ensure the capture of minute fluctuations without generating redundant data. The acquisition delay is controlled within 50ms, and a delay threshold is set according to the system's end-to-end response time requirements. Data is stored in a unified JSON format to achieve standardized integration of multiple parameters.

[0076] Temperature parameters are collected using temperature sensors. The temperature sensors are first deployed at multiple key points inside the fermenter, such as the bottom, middle and top of the tank. The uniform distribution points are determined by finite element heat conduction simulation. Temperature data is then collected, and the data source is the voltage signal conversion caused by resistance change. The sampling rate is 1Hz, and low-pass filtering is used. After collection, the data is converted into JSON fields, and metadata such as the point ID is attached to ensure that the temperature reflects the fermentation heat balance in real time.

[0077] Humidity parameters were collected using a humidity sensor installed near the fermenter vent. The humidity data was obtained from the frequency signal caused by capacitance changes, with a sampling rate of 1 Hz. Calibration was performed using a standard humidity generator. During the data collection, if the humidity fluctuation was greater than 5%, a potential evaporation anomaly was marked. The data was in JSON format to support subsequent pH correlation analysis.

[0078] pH parameters were collected using a pH electrode immersed in the fermentation broth. The data source was the potential difference measurement of a glass electrode, with a sampling rate of 1 Hz. After calibration with a buffer solution, the data was stored as JSON for evaluating the acidification process.

[0079] Microbial activity indicators were collected using a microbial counter. The microbial counter or online optical sensor counted the bacterial community density in real time, with data sourced from light scattering or microscopic counting. The sampling rate was 1 Hz, and the threshold was set to [value missing]. CFU / mL, when data > When the concentration is CFU / mL, the activity is considered normal. During data collection, the data is in JSON format, forming a dynamic parameter dataset for S2 deviation analysis to ensure the real-time performance and completeness of the analysis.

[0080] S2. Analyze the degree of deviation in the dynamic parameter dataset, calculate the degree of matching with the standard fermentation curve, and identify potential unstable factors. The specific implementation is as follows:

[0081] In the precision control process of fermentation of fermented soybean paste based on real-time monitoring, the dynamic parameter dataset formed in step S1 is analyzed to calculate the degree of deviation and the matching degree with the standard fermentation curve, and to identify potential unstable factors. This enables quantitative evaluation and early warning of anomalies in the fermentation process, ensuring the accuracy and pertinence of subsequent adjustment instructions. The dynamic parameter dataset includes multi-dimensional time series data of temperature, humidity, pH value and microbial activity indicators. The selection of data sources is based on the fermentation biochemical mechanism. For example, temperature deviation can easily lead to enzyme activity inhibition, unstable humidity may cause water imbalance, pH fluctuation directly affects microbial metabolism, and microbial activity serves as a comprehensive indicator to reflect the overall state. The analysis frequency is uniformly set to once per minute to match the dynamic changes in fermentation. The frequency setting is based on the statistical distribution of the fluctuation rate of fermentation parameters to ensure timely analysis without computational redundancy. The analysis delay is controlled within 100ms, and the delay threshold is set according to the requirements of the system's end-to-end response time. Through deviation calculation and matching degree evaluation, a logical chain from data input to factor identification is realized.

[0082] Real-time preprocessing of the dynamic parameter dataset: First, time series parameters are read from the JSON-formatted dataset in step S1. The length of each sequence is set to 60 points based on the sampling window, with a window sliding step of 30 points to cover the local dynamics of the fermentation cycle. Preprocessing includes noise filtering and outlier removal, using the Kalman filter algorithm, and the filtered state prediction formula is:

[0083] ;

[0084] in, Here is the state transition matrix. For the control matrix, To control the input vector and update the formula:

[0085] ;

[0086] in, For posterior state estimation, For Kalman gain, For the observation matrix, The measured values ​​are used to achieve a smooth sequence; outlier removal is calculated using the 3σ principle formula:

[0087] ;

[0088] If the value exceeds the limit, linear interpolation will be used to fill the gap. The interpolation formula is as follows:

[0089] ;

[0090] in, Let be the independent variable of the interpolation point. The independent variable is the one used for the previous valid point. For the next valid point, The dependent variable is the one that represents the previous valid point. The dependent variable for the next valid point;

[0091] Ensure a data loss rate of <2%; after preprocessing, the preprocessed data is finally standardized to the range of 0-1 to generate a preprocessed dataset. This dataset serves as a clean input source for bias calculation, avoiding noise interference with subsequent matching degree evaluation from the source and ensuring the reliability of the analysis results.

[0092] The deviation of the dynamic parameter dataset is calculated, with the deviation calculated for each parameter sequence relative to the standard fermentation curve. The preset curve is based on historical optimization data, such as the temperature standard curve which is a linear temperature increase of 30-40°C, and the Euclidean distance formula is used.

[0093] ;

[0094] in, The sequence length is 60. For a point in time, To preprocess the data, For the corresponding values ​​of the curve, calculate the deviation value of each parameter sequence; in the calculation, weights are assigned according to the parameter sensitivity, and the weighted deviation is calculated using the following formula:

[0095] ;

[0096] Where j is the parameter type, Assign weights to the sensitivity of the j-th parameter (e.g., temperature, humidity). The deviation value of the j-th parameter originates from the sub-step of calculating the deviation degree of the dynamic parameter dataset. It is calculated based on the Euclidean distance formula and according to historical stable batches, when the deviation < 0.05, the quality consistency rate is > 95%. Therefore, the deviation threshold is set to 0.05. If the deviation threshold is exceeded, the high deviation sequence is marked. This deviation value serves as the core quantitative indicator and is directly input into the downstream matching degree calculation module to objectively reflect the degree of deviation between the fermentation process and the ideal state, providing a data basis for process optimization.

[0097] For each preprocessed parameter sequence, its cosine similarity to the corresponding standard fermentation curve is calculated, and the matching degree is based on the cosine similarity. The cosine similarity formula is as follows:

[0098] ;

[0099] in, For preprocessing sequence vectors, As a standard curve vector, the matching degree of key fermentation parameters such as temperature and humidity is calculated independently during the calculation, and then the weighted average formula is used:

[0100] ;

[0101] The overall matching degree is obtained, and according to the production specifications, when the overall matching degree is >95%, the flavor consistency rate is >98%. The overall matching degree threshold is then set to 95%. If the overall matching degree is lower than the threshold, it is marked as a low-match sequence, and a deviation report is generated. In the calculation, if the parameter sequence lengths are inconsistent, the data dimensions are aligned by 60-point linear interpolation. This matching degree analysis serves as the core threshold judgment basis for identifying unstable factors, and is used to accurately screen abnormal parameters, providing data support for the quality control of the fermentation process.

[0102] When identifying potential destabilizing factors, the bias contribution of each parameter is first calculated based on the low-match sequence. The formula for the bias contribution is:

[0103] ;

[0104] Where k represents all parameters, The deviation value of the k-th parameter is calculated based on the Euclidean distance formula in the sub-step of calculating the degree of deviation of the dynamic parameter dataset. If the contribution of a parameter exceeds 20%, it is marked as a potential unstable factor. In the identification process, Granger causality test is used to verify the causal relationship between factors and to test the statistic. The test formula is:

[0105] ;

[0106] in, It is a lag phase. For the sample size, For the number of variables, The sum of squared residuals of the constrained model quantifies the total error of the constrained model fitting the data. The sum of squared residuals of the complete model is used to quantify the total error of the data fitting the complete model. For example, temperature fluctuations cause changes in pH value. Finally, the factors are divided into two types: environmental factors and biological factors, and a list of factors is output. This step is used as the closed-loop output of S2. The identification results are directly used in S3 to generate targeted adjustment instructions to ensure that control measures are accurately matched to unstable factors.

[0107] S3. Based on the deviation analysis results, generate preliminary adjustment instructions to automatically fine-tune temperature and humidity to stabilize the environment. The specific implementation is as follows:

[0108] In the real-time monitoring-based precision control of fermented soybean paste, the system responds to potential unstable factors identified in step S2, generates preliminary adjustment instructions based on deviation analysis results, and automatically fine-tunes temperature and humidity to stabilize the environment. This ensures thermal and humidity balance and optimized microbial metabolism during fermentation, avoiding flavor inconsistencies or contamination risks caused by deviation accumulation. Deviation analysis results include the deviation value and matching degree of each parameter. The data source selection is based on the preprocessing sequence of dynamic parameter datasets. For example, temperature deviations can easily lead to enzyme inactivation, and humidity deviations may cause uneven water evaporation. The preliminary adjustment instructions focus on these high-contribution factors, and the generation frequency is uniformly set to once per minute to match the analysis rhythm. The frequency setting is based on the statistical distribution of fermentation adjustment responses, ensuring timely instructions without over-adjustment, forming an automated chain from deviation input to environmental stability.

[0109] The preprocessing bias analysis results are first obtained by reading the result data from the bias report in step S2. Preprocessing includes anomaly filtering and weight allocation, using a median filter with a kernel size of 3 to remove isolated noise points. The median calculation formula is as follows:

[0110] ;

[0111] The neighborhood value is the set of neighboring data points around the current data point, which is all values ​​within the filter window (such as temperature points). The sorting function arranges the neighborhood values ​​in ascending order to generate an ordered list. Based on the historical distribution 3σ principle, a noise threshold of 0.01 is set. Then, the comprehensive deviation index is calculated using the following formula:

[0112] ;

[0113] The weights are determined based on sensitivity testing: temperature 0.4, humidity 0.3, pH 0.2, and microorganisms 0.1. Factors are prioritized in descending order of their indices. During preprocessing, according to production standards, if the matching degree is <90% and the risk is >10%, the matching degree threshold is set to 90%. If the matching degree is <90%, it is marked as a high-risk result and an alarm is added. After preprocessing, the results are stored in a temporary buffer to ensure data integrity. As a pre-processing step before instruction generation, targeted processing of deviation data provides a basis for subsequent precise adjustments, avoiding over-adjustment problems caused by generalized instructions.

[0114] Based on the deviation analysis results, preliminary adjustment instructions are generated. The controller first loads the pre-processed deviation data, including parameter type, deviation value, and matching degree, and generates targeted instructions for temperature and humidity. The target correction value is first calculated using the formula:

[0115] ;

[0116] The coefficient 0.1 is determined based on response testing. When the coefficient exceeds 0.2, the overshoot is less than 10%. For example, if the standard temperature is 30°C and the actual temperature is 28°C, the correction is +2°C. During the generation process, the rule base matches the instruction type according to the priority of factor contributions. The instruction format is JSON, and the duration is estimated using a heat conduction model. The formula is:

[0117] ;

[0118] The heating rate is 0.5°C / s. If multiple factors are involved, a joint command is generated to adjust the temperature and humidity simultaneously. After generation, the command is verified by simulating execution through a virtual model to ensure safety. This step serves as a prelude to fine-tuning. The generated command directly drives the heating and humidifying equipment, achieving a closed-loop connection from deviation analysis to environmental control.

[0119] The system automatically fine-tunes the temperature. It uses a 500W heating element (adjustable range 0-50°C) to execute control commands. The controller drives a relay to start and stop the heating, and it collects temperature sensor data in real time. The heating time is calculated using a time calculation formula:

[0120] ;

[0121] The temperature adjustment rate is 0.5°C / s, and the adjustment time is ensured to be <5s. If an overshoot >0.5°C is detected, the system will automatically reduce the heating power to 50% and dynamically stabilize the temperature through a feedback loop. After fine-tuning, the system will check whether the temperature fluctuation is ≤±0.2°C within a 10-second verification window (the average deviation is used to determine whether it meets the standard). Finally, the system will check the correlation between temperature adjustment and humidity to provide a basis for subsequent humidity fine-tuning and ensure coordinated control of environmental parameters.

[0122] The system automatically fine-tunes humidity levels. It uses a 300W humidifier (adjustable range 30%–80%RH) to execute control commands. The controller drives the pump to spray humidifier, collects humidity sensor data in real time, and calculates the humidification duration using a time calculation formula:

[0123] ;

[0124] The humidity adjustment rate is 5%RH / s, and the adjustment time is ensured to be <5s. During fine-tuning, if a fluctuation >2% is detected, the system automatically switches to an intermittent spray mode with a cycle of 2s. The system stabilizes the humidity within ±1%RH through a feedback loop. Based on historical production data, when the humidity fluctuation exceeds ±2%RH, the flavor consistency rate drops below 95% and the contamination risk increases by 15%. Therefore, the humidity deviation threshold is set to ±2%RH. After fine-tuning, the system checks whether the humidity deviation is <threshold within a 10-second verification window. Once the threshold is met, the adjustment ends. The fine-tuning results are fed back to the S4 monitoring module in real time, providing a basis for subsequent coordinated control of environmental parameters and ensuring stable and controllable humidity during the fermentation process.

[0125] Verify and record the fine-tuning effect. After fine-tuning, the system collects 5 seconds of real-time data and calculates the stability rate using the following formula:

[0126] ;

[0127] If the stability rate is >95%, it is marked as successful and logged. If it fails, it is fed back to step S2 to re-analyze the deviation and generate a correction instruction. The collected stability rate data is directly input into the S4 activity monitoring module to provide a basis for the assessment of microbial metabolic status.

[0128] S4. Monitor changes in microbial activity after adjustment, verify the control effect through sensor feedback, and calculate the activity recovery rate. The specific implementation is as follows:

[0129] In the real-time monitoring-based precision control of fermented soybean paste, the fermentation environment is tracked after the execution of the initial adjustment command generated in step S3. The changes in microbial activity after adjustment are monitored, and the control effect is verified through sensor feedback. The activity recovery rate is calculated to quantify the effectiveness of the adjustment and identify potential residual problems, ensuring the stability and flavor consistency of the fermentation process and avoiding batch failures or quality declines due to insufficient control. The changes in microbial activity after adjustment include indicators such as microbial density, enzyme activity level, and metabolite concentration. The selection of data sources is based on the principles of fermentation biology. For example, microbial activity directly reflects the control effect, while the recovery rate is used as a comprehensive quantitative tool to evaluate the overall impact of temperature and humidity fine-tuning. The monitoring frequency is uniformly set to once every 10 seconds to match the adjustment response time. The frequency setting is based on the statistical distribution of microbial metabolic rate to ensure timely monitoring without generating data. The monitoring delay is controlled within 50ms, realizing an automated chain from data acquisition to recovery rate calculation.

[0130] After preprocessing and adjusting the fermentation data, firstly, raw feedback data is collected from the fermentation environment after step S3. Preprocessing includes baseline correction and noise removal, according to the formula:

[0131] ;

[0132] Where i is the current window point, a moving average filter with a window size of 5 is used to smooth short-term fluctuations, and the filtering is based on historical noise distribution. Then, anomaly detection is performed using the Z-score method, with the formula:

[0133] ;

[0134] like If the value is greater than 3, it is marked as an anomaly and replaced with the median of the neighborhood data. The median is the middle value of the sorted neighborhood data to ensure data reliability. After preprocessing, the data is normalized to the range of 0-1 and a preprocessed dataset is generated. The preprocessed data is used as input to avoid noise interference in the recovery rate calculation.

[0135] Real-time acquisition of adjusted microbial activity data is achieved through ViableCell Counter sensors deployed at the bottom, middle, and top of the tank, immersed in the fermentation broth. The data source is optical density measurement, which is converted into CFU (cell population density) values ​​based on light scattering intensity. The conversion formula is as follows:

[0136] ;

[0137] in, It is an abbreviation for Optical Density, which quantifies the degree of absorption or scattering of light after it passes through a sample; and The calibration coefficient is determined by fitting a standard curve. Simultaneously, enzyme activity is detected by fluorescent labeling. The data acquisition rate is 0.1 Hz. Each record includes a timestamp, site ID, bacterial density, and enzyme activity value. During acquisition, if signal drift >2% is detected, the system automatically triggers the calibration procedure to ensure data reliability. The acquired real-time data is directly input into the recovery rate calculation module for comparison with the target activity value, to evaluate the adjustment effect, and to guide subsequent control strategies.

[0138] To calculate the activity recovery rate, first calculate the current average activity level using the following formula:

[0139] ;

[0140] in, The sequence length is 30, and the target activity at the corresponding time point of the standard fermentation curve is compared. The baseline recovery rate is then obtained using the recovery rate calculation formula, which is:

[0141] ;

[0142] Then, a weight correction is introduced. The calculation takes this weight correction into account to obtain the corrected recovery rate, as shown in the formula:

[0143] ;

[0144] The deviation index comes from the deviation analysis results of step S2. For example, if the correction basis is that the recovery rate is underestimated by 5% when the deviation is >0.1, and according to the quality specification, the risk is >8% when the corrected recovery rate is <90%, then the corrected recovery rate threshold is set to 90%. If the corrected recovery rate is lower than the threshold, it is marked as a low recovery state and an alarm signal is generated. After calculation, the recovery rate is stored in the log for S5 optimization. The recovery rate is used as a quantitative indicator to evaluate the effectiveness of the adjustment.

[0145] The control effect was verified through sensor feedback. Real-time data (sequence length 30 points) was collected over a 5-minute window after adjustment using deployed temperature, humidity, pH, and activity sensors. The control effect was verified by calculating the stability index, using the following formula:

[0146] ;

[0147] The target ranges are: temperature ±1°C, humidity ±2%RH, pH ±0.1, and activity ±10%. Based on production testing, when the index > 95%, the batch success rate is > 98%, so the index threshold is set to 95%. When the index > the threshold, it is marked as effective control. During validation, if the index < the threshold, the root cause is analyzed, and secondary adjustment suggestions are generated, i.e., the parameter changes that need to be adjusted. The formula is:

[0148] ;

[0149] Feedback data is transmitted back to the controller via the internal bus, with a verification cycle of 10 seconds. The results are directly input to S5 for iterative adjustment of the control algorithm to ensure the continuous stability of the fermentation environment.

[0150] When handling low recovery rate alarms, if the activity recovery rate is <90%, the system triggers a multi-level alarm mechanism, including audible and visual alarms and remote notifications. After the alarm is triggered, the system immediately suspends the fermentation process and records the event. The maximum number of alarm events per batch is 10. If this number is exceeded, the batch will be automatically terminated to prevent the spread of deviations and quality loss. This step logically supplements the verification. Through active intervention and event recording, it ensures that the fermentation process is traceable and controllable under abnormal conditions.

[0151] Once verification is complete, a report file is generated, displaying core metrics in real time through the user interface. The report includes additional recommendations and is simultaneously stored in the historical database to provide data support for S5 parameter iteration.

[0152] S5. Optimize fermentation model parameters, iteratively adjust the control algorithm based on the recovery rate, and perform adaptive accuracy improvement. Specifically, this is implemented as follows:

[0153] In the precision control of fermentation of fermented soybean paste based on real-time monitoring, the activity recovery rate calculated in step S4 is utilized to optimize fermentation model parameters. The control algorithm is iteratively adjusted according to the recovery rate to achieve adaptive accuracy improvement, thereby continuously improving the control strategy and ensuring the long-term stability and quality consistency of the fermentation process. This avoids accuracy degradation caused by slight environmental changes or microbial adaptation. The fermentation model parameters include the gain coefficient and threshold setting of the PID controller. The data source selection is based on sensor feedback and recovery rate quantification in step S4. For example, the recovery rate is used as a core indicator to reflect the control effect, while iterative adjustment optimizes parameters for low recovery scenarios to improve the accuracy of the next batch. The optimization frequency is uniformly set to once per batch to match the fermentation cycle. The frequency setting is based on the statistical distribution of the model convergence rate to ensure efficient optimization without overcomputation. The optimization delay is controlled within 500ms, realizing an automated chain from parameter input to adaptive improvement.

[0154] To preprocess the recovery rate data, firstly, the recovery rate sequence is read from the validation report in step S4. Preprocessing includes trend smoothing and outlier removal. A moving average filter is then calculated using the exponential formula:

[0155] ;

[0156] Where is the smoothing coefficient, and based on historical fluctuation tests, when If the allergy rate is >0.3% and the allergy rate is <5%, then set... , For time points The exponential moving average (i.e., the smoothed result of the previous time point) is used, and the average noise level is effectively reduced by 2% to remove short-term fluctuations. Anomaly detection is then performed using the isolated forest algorithm, with 100 trees and a sample ratio of 0.1. An anomaly detection threshold of 0.05 is set based on an accuracy test of >95%. If the anomaly detection rate exceeds the threshold, the median (the 4th value in the sorted neighborhood) is used instead. After preprocessing, the recovery rate data is used to generate a preprocessed recovery rate dataset, standardized to the 0-1 range, using the following formula:

[0157] ;

[0158] Preprocessed data is used as input to avoid noise interfering with the iterative process;

[0159] Generate an optimization objective function based on the recovery rate sequence obtained from preprocessing. (Normalization, Construct an objective function that minimizes the bias loss.

[0160] ;

[0161] Wherein, parameter vector The initial values ​​are 0.5, 0.1, and 0.2, and L2 regularization is introduced to suppress overfitting. Based on the overfitting test, when λ < 0.05, the generalization error is < 5%, so the coefficient λ = 0.01 is set. At the same time, stability constraints are incorporated. (Experimental tests show that oscillations exceed 10% when outside the range), and the gradient is calculated using the formula during the generation phase:

[0162] ;

[0163] The specific derivative is approximated using finite difference to ensure simplicity and computability. To ensure optimizability, the Hessian matrix of the objective function is required to be positive definite in the feasible region, making the objective function strictly convex and having a unique minimum value. After generation, the objective function is stored as a callable function Python script with initial parameters attached. In subsequent iterations, these parameters are used as optimization criteria and combined with gradient descent to update the parameters.

[0164] The gradient descent algorithm iterates through the parameters, starting with a variant of standard stochastic gradient descent (SGD) from the initial parameters. Based on the convergence test showing non-convergence with a learning rate > 0.05 and a convergence rate < 20%, the learning rate is set to 0.01. Furthermore, based on the computational efficiency > 10 and time > 1 second, the maximum number of iterations is set to 10. In each iteration, the gradient of the objective function is first calculated, and the parameters are updated according to the following formula:

[0165] ;

[0166] in, To update parameters, For learning rate, For the gradient, for example, the update of the scaling gain Kp is:

[0167] ;

[0168] Furthermore, to accelerate convergence and suppress oscillations in high curvature directions, a momentum term is introduced based on the momentum calculation formula, which is:

[0169] ;

[0170] The update is completed using a parameter update formula, which is compared with the stability test. Setting it to 09 can reduce oscillation by approximately 15%, so set it accordingly. The learning rate is set to 0.9. If the loss in a certain iteration increases by more than 5% compared to the previous iteration, the learning rate is adaptively halved until it is no lower than 0.001 to avoid oscillations or divergence in the near-optimal range. After execution, the iteration log is recorded and...

[0171] ;

[0172] As an accuracy indicator, the accuracy improvement is guaranteed to be >10%. After the iteration is completed, the output parameters are used as the input of the subsequent control algorithm to achieve closed-loop correction and verification.

[0173] Based on historical process data replay, a virtual model is established and offline verified to ensure prediction accuracy >90%, enabling supervised prediction of the actual fermentation process. After each iteration, the simulated recovery rate is calculated using the same formula as S4, and the consistency rate is >99% when the recovery rate improvement >10%. The recovery rate threshold is set to 10%. If the improvement relative to the previous version is >10%, the verification is considered successful, and the differences of key parameters before and after are compared. A difference <0.1 is considered as parameter stability. If the threshold is not reached, the iteration number is increased by 5 (maximum 20) on the current basis, or the optimizer is switched to continue searching for better parameters. After the verification is completed, an effect report is automatically generated and written to the database. At the same time, the verification results of this step are used as input for S6 to form the evidence chain for the final report and release decision.

[0174] When the system determines that the accuracy improvement of the current round relative to the previous round is less than 10%, it immediately triggers a "low gain" alarm, including the terminal LED flashing yellow at a frequency of 0.5 Hz and sending an alarm message to the control center via MQTT for remote linkage and parameter tracking. During the alarm period, the system automatically pauses the optimization loop and records the event in the local SQLite database. The maximum number of events in the same batch is 5. If the threshold is exceeded, the system will automatically roll back to the initial parameter set, thereby preventing invalid iterations under low gain conditions and ensuring the stability, auditability and traceability of the strategy.

[0175] During the output optimization parameter integration process, the system first freezes the optimal PID parameters for this batch and outputs the final parameter set in JSON format as a portable data exchange object. Simultaneously, it generates an optimization summary PDF (including the Loss convergence curve and parameter change tables for each round) for manual review to support traceable experimental records and compliance audits. This parameter set is then written back to the control algorithm: PID controller coefficient / register updates are completed within the real-time task cycle, ensuring deterministic execution with an end-to-end update time of <100 ms, and recording the version and timestamp for rollback and comparison. Meanwhile, the output of S5 serves as feedback input to S6 for adaptive improvement, and this parameter set is used as the initial baseline for S2 analysis in the next batch.

[0176] S6. Output the final control report and feed the optimized parameters back to the system. The specific implementation is as follows:

[0177] In the precision control process of fermentation of fermented soybean paste based on real-time monitoring, the fermentation model parameters optimized in step S5 are processed to output a final control report. The optimized parameters are then fed back to the system to achieve closed-loop precision control. This process summarizes batch effectiveness, stores experience data, and supports the application of the next batch, ensuring continuous improvement and quality stability of the fermentation process. It also avoids accuracy degradation caused by parameter loss or manual transmission. The core indicators are "recovery rate, stability index, and parameter change". The data are taken from the iteration results of S5 and the multi-source sensor verification set, which are used to quantitatively summarize the overall control effect of this batch. The parameter feedback is the core of the closed-loop mechanism and is used for the initial setting of the next batch. Based on the statistical distribution of the batch summary rate, the output frequency is uniformly set to once per batch to match the production cycle, ensuring efficient output without delay. The output delay is controlled within 300ms. Through the coordination of PDF generation library and database interface, an automated chain from parameter input to closed-loop optimization is realized.

[0178] The preprocessing of optimization parameters and validation data begins with reading the optimization parameter set from the iteration log of step S5, performing consistency checks and format conversions, and hashing the parameter strings using the SHA-256 algorithm. If the hash does not match the initial hash, the string is marked as invalid to ensure a parameter error rate of <0.1%. Subsequently, the validation data from step S4 is merged, and the overall performance index is calculated using the comprehensive performance index formula:

[0179] ;

[0180] The weights of 0.6 and 0.4 are derived from the principle of prioritizing recovery testing, and based on the principle that a batch success rate of >99% is achieved when the quality target index is >95%, the index threshold is set to 95%. During preprocessing, if any parameter deviates from the initial value...

[0181] ;

[0182] If the value exceeds 0.05, an exponential moving average smoothing is applied to the dev sequence to reduce iterative noise. After preprocessing, the data is standardized and output as a report template to generate a preprocessed dataset. The preprocessed data is used as input to avoid noise interfering with the accuracy of the report.

[0183] Generate the final control report by loading the preprocessed dataset using a PDF generation library. The report structure includes a title page, a summary page summarizing recovery rates >90% and improvements >10%, a data chart plotting the loss curve with iteration count on the x-axis, loss value on the y-axis, smoothed spline interpolation, and 10 nodes, and a parameter table listing Kp, Ki, Kd, ​​initial optimization values, and rate of change. The rate of change is calculated using the following formula:

[0184] ;

[0185] The report also includes a quantitative indicator of average recovery rate, which is calculated using the following formula:

[0186] ;

[0187] in, The system sets a threshold alarm. If the improvement is less than 10%, further optimization is required. Then, a PDF file is output and a digital signature is attached to ensure that the report cannot be tampered with. After generation, the integrity of the report is verified. If it fails, it is regenerated. The final report is used as the system archive output and provides a consistent and verifiable evidence carrier for parameter tuning and closed-loop feedback.

[0188] The optimized parameters are fed back to the system, representing the optimized parameters estimated in the preprocessed dataset for this batch. The system writes back to the database via the database interface and updates the system database. The target table structure includes an auto-incrementing primary key id, a timestamp, parameter fields, and a batch identifier. The system updates according to instructions, implementing persistent storage logic of overwriting if it exists and inserting if it does not. During the feedback process, the performance improvement rate corresponding to this optimization is simultaneously appended to the parameter table as metadata. After the update, the table is queried and verified to ensure the accuracy of the parameter writing. The feedback parameters are used as the initial values ​​for the next batch in the S1-S5 loop to improve adaptability.

[0189] To achieve closed-loop precision control and support subsequent batch applications, the closed loop forms a cycle from S6 to S1 through parameter feedback. At the end of each batch, after summarizing performance indicators in S6, the optimal control parameters are written back to the database. When the next batch starts (S1), the latest PID parameters are loaded from the database as the initial settings for the PID controller in S3, and register updates are completed within <50 ms to ensure seamless switching. During operation, the average boost rate is calculated using the following formula:

[0190] ;

[0191] in, If the historical average of the "improvement rate" of the most recent 5 batches is less than 10%, global optimization is triggered. The learning rate is increased by 0.001 to accelerate convergence and ensure that the improvement rate of subsequent batches is consistently greater than 10%. At the same time, online monitoring is implemented for consistency between batches, and the consistency is calculated according to the consistency formula. If the consistency is greater than 95%, the batch is marked as successful and enters parameter solidification, which is used as the starting parameter for the next batch. The consistency calculation formula is as follows:

[0192] ;

[0193] The above strategy logically constitutes a closed-loop process of "parameter inheritance → online evaluation → condition optimization → result consolidation → reuse in the next batch", which realizes stable alignment and continuous production optimization between batches;

[0194] Handling abnormal parameter feedback, when a control parameter is detected Exceeding the permitted range ( or When this occurs, the system immediately enters the fail-safe procedure sequence. On one hand, it triggers a buzzer alarm; on the other hand, it records the anomaly information in a structured log and rolls back the controller parameters to the default safe configuration (e.g., ...). Other parameters are restored synchronously according to default values); at the same time, the process is locked before manual confirmation and the start of the next batch of tasks is paused until manual confirmation, in order to prevent invalid parameters from spreading; after the feedback is completed, a confirmation message is generated and stored in the audit log. When overflow occurs, the old record is deleted to ensure traceability. The confirmation is output as an end marker for S6 closure and supports auditing.

[0195] In this embodiment, during fermentation, the system first synchronously collects and standardizes key parameters such as temperature, humidity, pH, and microbial activity at a uniform frequency, forming a traceable dynamic dataset. Then, it compares the current multi-parameter curve with the standard fermentation curve point by point, identifying the dominant factors causing instability based on the magnitude and contribution rate of the deviation. Based on this, it issues real-time temperature and humidity fine-tuning commands (including priority and safety constraints), and uses sensors to read back and compare the deviations before and after adjustment within a short verification window, calculating quantitative indicators such as activity recovery rate and stability to confirm the control effect. If the recovery is insufficient, the verification results and deviation report are sent back to the control model, iteratively correcting the control parameters to make the next round of adjustment more aligned with process requirements. At the end of the batch, the system generates a control report containing core indicators and trend charts, and solidifies the optimized parameters into the database as the initial benchmark for the next batch, achieving parameter inheritance and process closure, ensuring continuous convergence and stable consistency of the fermentation process across multiple batches.

[0196] Example 2: Figure 2 A schematic diagram of a soybean paste production control system according to the present invention is provided. The soybean paste production control system includes:

[0197] Parameter acquisition module: responsible for real-time acquisition of key parameters during the fermentation process, forming a dynamic parameter dataset, comprehensively sensing the environment and biological state, and supporting the real-time data foundation for subsequent analysis;

[0198] Deviation Analysis Module: Analyzes the degree of deviation in the dynamic parameter dataset, calculates the degree of matching with the standard fermentation curve, and identifies potential instability factors;

[0199] Command generation module: Generates preliminary adjustment commands based on deviation analysis results, and automatically fine-tunes temperature and humidity;

[0200] Activity monitoring module: Monitors changes in microbial activity after adjustment, verifies the control effect through sensor feedback, calculates the activity recovery rate, and provides control verification and abnormal alarms;

[0201] Model optimization module: Optimizes fermentation model parameters, iteratively adjusts the control algorithm based on the recovery rate, and adaptively improves accuracy and learns the system.

[0202] Report output module: Outputs the final control report and feeds back the optimized parameters to the system to complete closed-loop precision control and batch continuous optimization.

[0203] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0204] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0205] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0206] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0207] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0208] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0209] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0210] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0211] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling production of a bean paste, characterized by, Comprising: S1, real-time acquisition of key parameters in the fermentation process, including temperature, humidity, pH value and microbial activity index, forming a dynamic parameter data set; S2, analyze the deviation degree of the dynamic parameter data set, calculate the matching degree with the standard fermentation curve, and identify potential unstable factors; S3, based on the deviation analysis result, generate preliminary adjustment instruction, automatically fine-tune temperature and humidity to stabilize the environment; S4, monitor the change of microbial activity after adjustment, verify the control effect through sensor feedback, and calculate the activity recovery rate, including collecting microbial activity data after adjustment, verifying the control effect through sensor feedback, and comparing the feedback value with the target value; Calculate the activity recovery rate, determine the recovery rate based on the ratio of current activity and target activity, consider the weight of enzyme activity and microbial density; Verify the control effect by reducing the deviation to confirm the effectiveness of the adjustment; S5, optimize the fermentation model parameters, adjust the control algorithm iteratively according to the recovery rate, and improve the adaptive accuracy, including extracting the current value from the recovery rate calculation result, and marking the optimization requirement; Input control algorithm model for iterative adjustment, adjustment range control within safety limit; According to the recovery rate, iteratively adjust the control algorithm, update the parameters based on the minimum error function to ensure the accuracy is improved above the threshold; Adaptive accuracy improvement, verify the parameter application effect through iteration; Store the iteration log, including the number of iterations and the improvement rate; S6, output the final control report, and feed back the optimized parameters to the system.

2. The control method of claim 1, wherein Real-time acquisition of key parameters in the fermentation process, including temperature, humidity, pH value and microbial activity index, forming a dynamic parameter data set, including: Temperature sensor collects temperature data at evenly distributed points determined by finite element heat conduction simulation, data source is voltage signal conversion caused by resistance change; Humidity sensor collects humidity parameters, data source is frequency signal caused by capacitance change; pH electrode collects pH value parameters, data source is potential difference measurement of glass electrode; Microbial counter collects microbial activity index, real-time counts microbial density, data source is light scattering signal or microscopic counting; Convert the collected data into standardized units to form a dynamic parameter data set.

3. The control method of claim 1, wherein Analyze the deviation degree of the dynamic parameter data set, calculate the matching degree with the standard fermentation curve, and identify potential unstable factors, including: Extract the parameter value sequence at the current time from the dynamic parameter data set, compare it with the preset standard fermentation curve point by point, and calculate the deviation degree; Identify the dominant factors based on the deviation vector, determine the potential unstable factors by contribution rate; Calculate the matching degree with the standard fermentation curve, determine the matching degree based on the ratio of average deviation and maximum allowed deviation; Identify potential unstable factors by comparing the deviation contribution of each parameter, mark the factors with high single parameter contribution; Use rule engine to match historical patterns and output factor list.

4. The control method of claim 1, wherein Based on the deviation analysis result, generate preliminary adjustment instruction, automatically fine-tune temperature and humidity to stabilize the environment, including: Extract the deviation vector and factor label from the deviation analysis result, generate preliminary adjustment instruction, the instruction priority is sorted according to the factor severity; Automatic fine-tuning for temperature, adjusting the temperature by driving the heating equipment, with the adjustment range controlled within the specified range, with preset safety constraints; Automatic fine-tuning for humidity, adjusting the humidity by driving the humidification equipment, with the adjustment time controlled within the specified time limit; Verify the fine-tuning effect by comparing the deviation before and after adjustment through feedback data, and output the adjustment status after confirming the stability of the environment.

5. The control method of claim 1, wherein Output the final control report and feed back the optimization parameters to the system, including: Extract key indicators from optimization results to generate a final control report, incorporating risk prediction for the next batch; Optimization parameters are fed back to the system, with database updates to store parameters, supporting automatic loading for the next batch; Verify the feedback effect by simulating the next batch to confirm the accuracy improvement; After outputting the report, record the efficiency indicators to form a closed loop throughout the process.

6. The control method of claim 5, wherein Including: Automatically trigger the database rollback mechanism to restore the process parameters to the stable version verified by the previous batch; Synchronously generate safety alert information containing rollback timestamp, abnormal parameter identification and deviation degree; Rollback operation logs, safety alerts and control data streams of this batch, real-time detection data are packaged together as structured process records; Record the tamper-proof storage through blockchain storage technology, and establish a unique mapping relationship with the production batch number; Based on the abnormal parameter distribution characteristics in the process record, generate parameter optimization suggestions through machine learning algorithms, as input constraints for the next round of standard curve calibration.

7. A miso production control system for implementing the miso production control method according to any one of claims 1 to 6, characterized by Including: Parameter acquisition module: responsible for real-time acquisition of key parameters in the fermentation process, forming a dynamic parameter dataset, fully sensing the environment and biological state, supporting real-time data basis for subsequent analysis; Deviation analysis module: analyze the deviation degree of dynamic parameter dataset, calculate the matching degree with the standard fermentation curve, and identify potential unstable factors; Instruction generation module: generate preliminary adjustment instructions based on deviation analysis results, automatically fine-tune temperature and humidity; Activity monitoring module: monitor the activity change of microorganisms after adjustment, verify the control effect through sensor feedback, calculate the activity recovery rate, and control verification and abnormal alarm; Model optimization module: optimize fermentation model parameters, iteratively adjust control algorithms according to recovery rate, and adaptively improve accuracy and system learning; Report output module: output the final control report and feed back the optimization parameters to the system, complete closed-loop precision control and batch continuous optimization.

Images