A method and system for intelligent temperature control in rice wine production
By employing local heterogeneous misalignment entropy calculation and adaptive sampling mechanism in the fermentation system, the problems of uneven reaction and data regulation lag in the fermentation temperature control system were solved, achieving precise temperature control and uniform fermentation during the fermentation process and improving the stability of the finished product quality.
Patent Information
- Application Number
- CN202611440141.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-11-17
AI Technical Summary
Existing fermentation temperature control systems cannot monitor the uniformity of the reaction of materials in the tank in real time, resulting in fluctuations in the sweetness, acidity and taste of different batches of finished products. Furthermore, the data acquisition and control parameters cannot be dynamically adjusted, leading to a waste of sampling resources or a delayed response.
By using local heterogeneous misalignment entropy calculation, segmented phase mapping analysis, and a two-level adaptive sampling mechanism in the fermentation system, the data acquisition cycle and weight are dynamically adjusted to generate temperature control decision judgment values, thereby achieving precise monitoring and control of the material reaction in the tank.
It improves the accuracy of temperature control decisions, reduces quality fluctuations, optimizes the utilization of sampling resources, adapts to the dynamic needs of different fermentation stages, and eliminates the need for additional testing equipment.
Smart Images

Figure CN123114136A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of temperature control technology in rice wine production, specifically to an intelligent temperature control method and system for rice wine production. Background Technology
[0002] Fermentation of rice wine relies on microbial metabolism to produce flavor substances and nutrients. Fermentation temperature directly determines the activity of the microbial community, the reaction rate, and the composition of the products, making it a core process parameter that determines the quality of the finished product and batch stability. The current industry-standard temperature control solution mainly consists of three steps: First, several contact temperature sensors are placed on the outer walls of the fermentation tank, such as the side walls and bottom. At the same time, fixed data acquisition cycles, temperature control thresholds, and operating parameters of the temperature control components are preset based on process experience. Second, the temperature values collected by the sensors on the tank wall are continuously read according to the preset fixed cycle, and the real-time readings are compared with the preset temperature process range. Third, when the collected surface temperature exceeds the set upper or lower limit, the corresponding cooling or heating components of the tank jacket are activated to adjust the overall temperature of the tank. Once the sensor readings fall back to the set range, the temperature control action is immediately stopped to maintain temperature stability during the fermentation process.
[0003] During the implementation of the above technical solution, at least the following technical problems were found: Firstly, all temperature sensors are located on the outer wall of the tank, and the collected values can only reflect the temperature of a thin layer of material near the tank wall, and cannot reach the core area of the material inside the tank. The fermentation reaction is driven by the metabolism of the microbial community inside the material. Affected by factors such as the concentration distribution of the material, the colonization location of the microbial community, and the intensity of natural convection, the reaction rate at different depths and radial locations inside the tank naturally varies. Using the surface temperature as the only control target, it is impossible to detect uneven local reactions that are too fast or too slow inside the tank. Over a long period of operation, the fermentation degree of the entire tank of material is often inconsistent, which directly leads to significant fluctuations in the sweetness, acidity, and taste of different batches of finished products. Secondly, the data acquisition cycle, control trigger threshold, and temperature control operating power are all fixed parameters preset before production and will not be automatically adjusted according to the dynamic changes in the fermentation process. In the early saccharification stage of fermentation, the activity of the microbial community is low, the heat production is low, and the temperature fluctuation is gentle. The fixed frequency of high-frequency sampling will generate a large amount of redundant data, occupying unnecessary storage and computing resources. After entering the peak stage of the main fermentation, the metabolism of the microbial community is active, the heat production rate is fast, and the temperature change is large. The fixed acquisition cycle is difficult to capture instantaneous temperature fluctuation signals, which causes the temperature control to always lag behind the temperature change and cannot adapt to the dynamic reaction rhythm inside the fermentation system. Therefore, we disclose an intelligent temperature control method and system for wine production. Summary of the Invention
[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent temperature control in rice wine production, comprising the following steps: Collect multiple sets of production source data throughout the entire process of brewing rice wine. The production source data includes core temperature data of fermentation tanks, temperature and humidity data of workshop environment, state data of fermentation materials, and progress data of fermentation stages. By combining the changing characteristics of various production source data with the matching collection frequency, an initial differentiated temperature collection cycle rule is formed; based on this cycle rule, batch data collection is carried out to obtain multiple batches of temperature-related data sample sets. Perform integration and verification on multiple batches of temperature-related data sample sets, and output the benchmark value for temperature data acquisition quality assessment; extract continuous temperature time series data during the temperature control rest period after the previous round of temperature regulation has completely stopped, and obtain the local heterogeneity misalignment entropy of the fermentation system through segmented phase mapping operation; Using the local heterogeneous misalignment entropy of the fermentation system as the core correction parameter, the differential temperature acquisition cycle rule, the sampling time grid arrangement, and the weight ratio of the quality assessment dimension are adjusted sequentially to finally generate the temperature control decision judgment value. The temperature control decision value is compared with the preset control threshold. If the threshold is exceeded, the corresponding level of temperature control strategy is activated to guide the fermentation system to converge toward homogenization. If the threshold is not exceeded, the current temperature control state is maintained, and the differentiated temperature acquisition cycle rules for the next round are updated simultaneously.
[0005] Furthermore, when collecting production source data, we first locate multiple data access points within the winemaking production scenario. After collecting multiple sets of raw production data from each point, we classify and aggregate them based on the identification features carried by the data. Among them, the raw data with temperature sensor labels are classified as the core temperature data of the fermentation tank, the data with environmental monitoring attributes correspond to the temperature and humidity data of the workshop environment, the data with material detection labels are determined as the fermentation material status data, and the data marked with process node markers are classified as fermentation stage progress data.
[0006] Furthermore, when generating the initial differentiated temperature acquisition cycle rules, the correspondence between temperature fluctuations and key fermentation nodes is first sorted out based on the core temperature data of the fermenter, and temperature node indexes and fermentation stage labels are built. The update frequency of the index and labels is used as the first acquisition cycle benchmark value. The workshop environment temperature and humidity data are adjusted based on the rate of temperature and humidity change per unit time to obtain the second acquisition cycle benchmark value. The fermentation material status data are set with reference to the update frequency of status indicators to set the third acquisition cycle benchmark value. The fermentation stage progress data is matched with the business dependence of the production process to obtain the fourth acquisition cycle benchmark value. After integrating four types of periodic benchmark values to form initial rules, the benchmark values of each period are then scaled differently according to the numerical range of the local heterogeneous misalignment entropy of the fermentation system. The higher the misalignment entropy value, the higher the collection frequency of various types of data, and finally an executable differentiated temperature collection periodic rule is obtained.
[0007] Furthermore, when acquiring multiple batches of temperature-related data sample sets, multiple data acquisition tasks are first generated based on the differentiated temperature acquisition cycle rules, and corresponding acquisition configuration parameters are matched for each task. The original traceability data sources corresponding to the four types of data are located according to the acquisition configuration parameters, and a basic sampling sequence is set for each data source. The asynchronous period of temperature fluctuation is predicted by combining the local heterogeneous misalignment entropy of the fermentation system. The higher the misalignment entropy value, the higher the sampling point density in the asynchronous period, and the corresponding lengthening of the sampling interval in the synchronous period. The total number of sampling points is controlled within the preset range, thereby forming a non-uniform sampling time grid. Preliminary data collection was conducted in batches for each original traceability data source according to a non-uniform sampling time grid. After obtaining the corresponding category of collected information, the four categories of collected information were integrated by aligning them along the time axis to form a multi-batch temperature-related data sample set.
[0008] Furthermore, the specific steps for calculating the temperature data acquisition quality assessment value are as follows: Data structure and content features were extracted from multiple batches of temperature-related data samples. The data results were classified into three categories: structured, semi-structured, and unstructured, based on their structural attributes. The information continuity assessment benchmark was then calculated accordingly. Five evaluation dimensions—information completeness, information accuracy, information timeliness, information clarity, and information relevance—are extracted from the data content characteristics. Based on the influence of each dimension on temperature control decisions, corresponding weight coefficient benchmark values are assigned, and the content information evaluation benchmark value is calculated. The weights of each dimension are dynamically adjusted by combining the local heterogeneity of the fermentation system with the misalignment entropy. The higher the misalignment entropy value, the greater the increase in the weight coefficients corresponding to information timeliness and information continuity. The weight coefficients of the other three dimensions decrease proportionally and synchronously, and the sum of the weight coefficients of the five dimensions remains constant. The weighted sum of the information continuity evaluation value and the content information evaluation value is used to obtain the final temperature data acquisition quality evaluation value.
[0009] Furthermore, when calculating the local heterogeneous misalignment entropy of the fermentation system, the time node when the previous round of temperature control completely stopped is first locked, and the complete time period after this node when the temperature value gradually returns to the steady state value is defined as the temperature control rest period; the continuous time-series temperature data with equal time intervals within this stage are extracted and divided into multiple sub-sequences of equal length according to time order, and each sub-sequence corresponds to the temperature feedback period of different spatial regions in the fermenter; Temperature regression curves for each sub-segment sequence are fitted separately, and the time nodes when each sub-segment regression curve reaches the preset proportional steady-state value are extracted to generate a semi-steady-state phase point sequence. The hysteresis difference between adjacent semi-steady-state phase points is calculated to form a phase hysteresis difference sequence. After normalizing the phase hysteresis difference sequence, the material reference physical property parameters of the current fermentation stage are substituted into the entropy calculation model to obtain the local heterogeneous misalignment entropy of the fermentation system. When generating temperature control decision values, a homogenization regulation correction coefficient is generated based on the local heterogeneity misalignment entropy of the fermentation system. The homogenization regulation correction coefficient is then multiplied by the temperature data acquisition quality assessment value to output the temperature control decision value.
[0010] An intelligent temperature control system for rice wine production, the control system comprising: The first acquisition module collects multiple sets of production source data throughout the entire process of winemaking, including four categories: core temperature data of fermentation tanks, temperature and humidity data of workshop environment, state data of fermentation materials, and progress data of fermentation stages. The cycle generation module combines the changing characteristics of various production source data to match the collection frequency, forms the initial differentiated temperature collection cycle rules, and receives the local heterogeneous misalignment entropy of the fermentation system to complete the cycle parameter correction. The sampling execution module performs batch data acquisition according to the revised periodic rules and outputs multiple batches of temperature-related data sample sets. The quality assessment module performs integrated verification on multiple batches of temperature-related data sample sets, outputs the baseline value for temperature data acquisition quality assessment, and simultaneously receives the local heterogeneity misalignment entropy of the fermentation system to adjust the weight of the assessment dimensions. The heterogeneous entropy calculation module extracts continuous temperature time-series data throughout the entire time period after the temperature control action has completely stopped, and obtains the local heterogeneous misalignment entropy of the fermentation system through segmented phase mapping calculation. The decision generation module uses the local heterogeneity misalignment entropy of the fermentation system as the core correction parameter, and combines the adjusted temperature data acquisition quality assessment value to generate a temperature control decision judgment value. The temperature control decision judgment value is compared with the preset control threshold. If the threshold is exceeded, the corresponding level of temperature control strategy is activated to guide the fermentation system to converge toward homogenization. If the threshold is not exceeded, the current temperature control state is maintained and the rules for the next round of acquisition cycle are updated simultaneously.
[0011] Furthermore, the first acquisition module includes: The point positioning unit completes the positioning of multiple data access points in the winemaking production scenario and collects multiple sets of raw production data; Multiple data classification units, based on temperature sensor labels, environmental monitoring attributes, material detection labels, and process node markers, respectively, classify the raw production data into four categories of production source data.
[0012] Furthermore, the period generation module includes: The baseline period unit is matched with the baseline value of the collection period corresponding to the four types of data and integrated to form the initial rule; The periodic correction unit receives the local heterogeneous misalignment entropy of the fermentation system output by the heterogeneous entropy calculation module, performs differential scaling on the baseline values of each period according to the numerical range, and outputs the final differential temperature acquisition periodic rules to be executed. The sampling execution module includes: Basic sequence unit: A basic sampling sequence is set for each original traceability data source; The sample set acquisition unit, combined with the local heterogeneous misalignment entropy of the fermentation system, predicts asynchronous periods of temperature fluctuations, dynamically adjusts the sampling point density to form a non-uniform sampling time grid, and performs data collection and alignment integration according to the non-uniform sampling time grid, outputting multiple batches of temperature-related data sample sets.
[0013] Furthermore, the heterogeneous entropy calculation module includes: The phase is defined as a unit, locking in the time point at which the previous round of temperature control action completely stops and defining the complete temperature control rest period; The phase extraction unit extracts continuous time-series temperature data during the temperature control rest period and splits it into multiple sub-sequences of equal length. It then fits the temperature regression curve of each sub-sequence, extracts the semi-steady-state phase point, and generates a phase hysteresis sequence. The entropy output unit normalizes the phase lag difference sequence and calculates the local heterogeneous misalignment entropy of the fermentation system by combining it with the material baseline physical property parameters of the current fermentation stage. It then simultaneously transmits the entropy to the cycle generation module, sampling execution module, quality assessment module, and decision generation module. The quality assessment module receives the local heterogeneous misalignment entropy of the fermentation system and dynamically adjusts the weight ratio of each assessment dimension, outputting the final temperature data acquisition quality assessment value to the decision generation module.
[0014] This invention provides an intelligent temperature control method and system for rice wine production, which has the following beneficial effects: After each round of temperature control stops, the complete time period data of temperature regression to natural steady state is extracted and divided into multiple equal-length subsequences according to time. The temperature rebound curve of each segment is fitted. By comparing the time differences of the curves reaching the preset proportion steady state value, the local heterogeneous dislocation entropy of the fermentation system is calculated. The reaction uniformity of different regions inside the material is inferred by using the temporal differences of temperature rebound. Since the entire process does not require the addition of any detection sensors inside the material, nor does it require interruption of the fermentation process for sampling and detection, the internal reaction state information that cannot be covered by traditional surface temperature measurement can be obtained solely by relying on the time-series data produced by the original temperature measurement equipment in the tank. This allows the judgment basis of temperature control to go from the surface temperature value to the internal reaction uniformity, and the judgment dimension is more in line with the actual situation of fermentation. Secondly, the calculated heterogeneous misalignment entropy has a reverse effect on the entire process of data acquisition and quality assessment. For example, when the system is highly heterogeneous and temperature fluctuations are active, the acquisition cycle of various types of data is automatically shortened, and the sampling points are shifted and densified towards asynchronous periods with significant temperature fluctuations. At the same time, the weight of information timeliness and information continuity in quality assessment is increased. When the system is stable and homogeneous, the acquisition interval is appropriately lengthened, and the sampling density and weight of the corresponding assessment dimensions are reduced during stable periods. The entire acquisition strategy and assessment criteria can be dynamically and adaptively adjusted according to the fermentation state without manual parameter modification. Under the premise of controlling the overall acquisition load, it avoids the waste of sampling resources in the stable phase and solves the problem of delayed sampling response in the peak phase, effectively improving the accuracy of temperature control decision reference data. In addition, based on judgment criteria that are closer to the actual internal state, temperature control is no longer limited to maintaining the surface temperature within a set range. Instead, it guides the materials inside the tank to form orderly and natural convection through corresponding levels of temperature regulation, gradually reducing the differences in reaction progress in different areas, promoting the entire fermentation system to converge towards homogenization, reducing quality fluctuations caused by uneven local fermentation, and helping to stabilize the flavor and taste of different batches of fermented wine. Furthermore, it does not require the addition of a large number of testing equipment, making the implementation and modification costs controllable and adaptable to different scales of fermented wine production scenarios. Attached Figure Description
[0015] Figure 1 This is a general block diagram of the entire process of intelligent temperature control for rice wine in this invention; Figure 2 This is a flowchart illustrating the calculation of local heterogeneous dislocation entropy in the fermentation system of this invention. Figure 3 This is a diagram showing the relationship between misalignment entropy and scaling factor in this invention; Figure 4 This is a distribution diagram of the sampling point density along the rebound time in this invention. Detailed Implementation
[0016] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0017] Example 1: This example uses a jacketed fermentation tank for food processing as an example. Jacketed fermentation tanks are common equipment in the food fermentation field. Temperature control of the materials inside the tank is achieved by introducing hot and cold media through the outer jacket. It is currently standard equipment in rice wine production. Furthermore, current rice wine production generally employs a process of periodically collecting tank wall temperature data, comparing it to preset process thresholds, and starting and stopping temperature control components to stabilize the temperature at the measuring points within the required process range, providing a stable temperature foundation for the fermentation microorganisms. Although it has the advantages of simple implementation, stable operation, and low daily maintenance costs, there are still some issues in actual operation. Two technical issues exist: First, the sensors can only collect the surface temperature of the tank wall and cannot detect the uniformity of the reaction inside the material. Long-term operation can easily lead to uneven local fermentation, ultimately causing batch quality fluctuations. Second, the sampling cycle, evaluation criteria, and control logic are all preset fixed parameters, which cannot be dynamically adjusted with the fermentation process. This makes it difficult to match sampling resources and evaluation criteria with the actual needs of different stages. To address these shortcomings, we have made adjustments and improvements, namely through local heterogeneous misalignment entropy calculation of the fermentation system, segmented phase mapping analysis, a two-level adaptive sampling mechanism, and dynamic weighted quality assessment, to solve this problem. Figure 1 As shown, the specific content is as follows: Calibration is required before use, as it directly determines the baseline accuracy of all subsequent calculations. Conventional temperature control deployments typically prioritize rapid deployment, only performing factory-calibrated sensors and putting them into operation immediately after installation. While this shortens the commissioning cycle, factors such as sensor installation position deviations, tank wall thickness differences, and inconsistent jacket thermal conductivity create fixed systematic errors between different tanks. This reduces the lateral comparability of test results and indirectly lowers the accuracy of subsequent analyses. Therefore, to eliminate this problem and address the logical breakpoints lacking historical calculation data during the cold start phase, three pre-calibration steps are implemented, as follows: The first step is zero-point calibration of the empty tank. For fermenters with a defined structure, when the internal medium is completely homogeneous water, the temperature rebound phase difference after the temperature control stops is determined only by hardware factors such as the tank structure and sensor installation location. This fixed phase difference is the zero-point reference value. In specific operation, the empty tank is filled with room temperature water, the jacket heating component is started to raise the water temperature to 30°C at a uniform rate, and the temperature is maintained for about 30 minutes until the overall temperature of the tank is uniform. Then, the temperature control output is stopped, and the temperature rebound data is continuously collected through the PT100 platinum resistance sensor on the tank wall until the temperature completely returns to a steady state. The rebound phase difference of each segment is extracted from the complete time series data, and the average value is taken as the zero-point reference. All subsequent misalignment entropy calculations will first deduct the zero-point reference value to eliminate the systematic errors caused by hardware installation and the tank itself, and ensure that the entropy calculations of different tanks and different batches have a unified comparison benchmark. The second item is the import of formula parameters, used to avoid interference from material property differences in heterogeneity judgment. During the fermentation process of rice wine, material properties such as viscosity and sugar content change significantly with the formula and fermentation stage. Since these properties directly determine the temperature conduction rate, directly comparing phase differences while ignoring property differences can easily misjudge heat transfer differences caused by properties as differences in reaction heterogeneity. Therefore, for the current rice wine formula, corresponding baseline misalignment entropy thresholds, material baseline property parameters, and control thresholds need to be imported according to the three core fermentation stages: saccharification, primary fermentation, and post-fermentation. The material baseline property parameters include four core indicators: material viscosity, sugar content, density, and solids content, all of which can be calibrated from experimental test data of historical batches of the same formula. In actual use, the tank heat transfer coefficient will be combined with existing empirical heat transfer formulas to convert the property parameters into the corresponding stage's baseline hysteresis difference Δt_. Ref, or the standard phase difference of a perfectly homogeneous normal material after temperature control stops, is calculated based on well-known empirical formulas for unsteady-state heat transfer. First, the convective heat transfer coefficient of the material inside the tank is calculated using its viscosity, density, thermal conductivity, specific heat capacity, and other physical properties, combined with structural features such as the tank's inner diameter and liquid level. Then, the characteristic response time of the temperature rebound of the homogeneous material is obtained by substituting these parameters into the unsteady-state heat transfer characteristic time formula, which is the physical property reference hysteresis. This conversion method is a common calculation logic in existing heat transfer engineering, mainly used for theoretical estimation when there are no readily available materials or on-site calibration is not possible. When conditions permit, experimental calibration should be preferred, using the average of three repeated tests of homogeneous material as the final reference value for higher accuracy. The converted physical property reference hysteresis will offset the influence of the material's own properties on the heat transfer rate in subsequent entropy calculations, retaining only the phase difference caused by uneven microbial community reactions. The baseline misalignment entropy threshold represents the maximum acceptable degree of heterogeneity at the corresponding stage. It is determined using historical batch quality data. When the misalignment entropy is below this value, the batch quality is within the acceptable range and no additional control is needed. Exceeding this value indicates that uneven local fermentation may affect quality, requiring homogenization intervention. The control threshold is a judgment standard matched with the decision-making value, calculated by multiplying the baseline misalignment entropy threshold by the corresponding level's standard correction coefficient. Under standard data quality (Q=1), a misalignment entropy of 0.3 corresponds to k_d=1.2, therefore the mild control threshold is 0.3×1.2=0.36; a misalignment entropy of 0.6 corresponds to k_d=1.8, therefore the moderate control threshold is... The degree of control threshold is 0.6 × 1.8 = 1.08; the misalignment entropy of 0.8 corresponds to k_d = 2.5, so the severe control threshold is 0.8 × 2.5 = 2.0. At the same time, it is divided into three levels of thresholds: mild, moderate and severe, corresponding to three levels of control actions. Taking the common glutinous rice wine as an example, the reference viscosity in the main fermentation stage is 120 mPa・s, the reference sugar content is 18°Bx, and the calculated adjacent phase difference reference hysteresis difference Δt_ref is 5.75 seconds, that is, the standard time difference between two adjacent semi-steady-state phase points under homogeneous normal material. The corresponding reference misalignment entropy threshold is set to 0.3. If this value is exceeded, homogenization control will be initiated. The third item is the initial operating parameter calibration. Based on historical production data, the four types of data collection cycle benchmarks, initial weights for quality assessment, and graded control trigger thresholds for the first round of operation are determined to ensure that the system can execute with reasonable parameters in the first round of operation without waiting for iterative optimization. Among them, the graded control thresholds correspond one-to-one with three heterogeneous intervals. Under standard data quality, the mild control threshold corresponds to a misalignment entropy of 0.3, the moderate threshold corresponds to 0.6, and the severe threshold corresponds to 0.8, respectively matching three levels of control intensity. For new formulas without historical data, the general glutinous rice wine benchmark parameters are used as initial values, and a rapid calibration mode is activated. The sampling frequency is increased in the first three control cycles to quickly collect data and iteratively correct the benchmark parameters. The exclusive parameter calibration is completed within 8 hours, and the system smoothly transitions to normal operation mode. During cold start, the first temperature control action follows the traditional temperature deviation threshold trigger logic. When the tank wall temperature exceeds the process range, the corresponding control is activated. After the control stops, the first complete temperature control rest period is generated, and the first misalignment entropy is calculated. From the next collection cycle, the system enters a full-link dynamic adaptation cycle.
[0018] After the three calibrations are completed, the parameters are automatically stored in the system's benchmark library, and can be continuously iterated and optimized based on actual production data during subsequent operation.
[0019] Data Acquisition, Classification, and Adaptive Cycle Generation: After pre-calibration, routine data acquisition is required. Due to the complex communication environment in industrial settings, the accuracy of data classification directly affects the reliability of subsequent analysis. Traditional address mapping classification methods are simple to implement and sufficient for stable operating conditions, but they are prone to data crosstalk when encountering situations such as message mistransmission or equipment replacement. Therefore, we adopt a dual classification mechanism of hard judgment of message identifier bits plus feature value range verification to ensure data accuracy from both identifier and value range dimensions. Firstly, an industrial gateway is used to connect all data access points within the production scenario. These points cover four dimensions: fermenters, workshop environment, laboratory testing, and production control. Specifically, this includes six PT100 platinum resistance temperature sensors across three layers on the fermenter wall, wall-mounted temperature and humidity transmitters in the workshop, interfaces for the laboratory LIMS system, and process node interfaces for the production MES system. LIMS, or Laboratory Information Management System, is a common laboratory data management system used to uniformly store and manage sample testing data. MES, or Manufacturing Execution System, is a common control system in industrial production used to track the progress of processes and the status of materials throughout the entire production process. All points communicate via industrial Ethernet to control data transmission latency, ensuring timely real-time data acquisition. After the raw data is collected, it is classified and aggregated according to a unified dual-verification rule. Sensor messages with the TEMP function code first undergo temperature value rationality verification, limiting the range to 10℃~45℃ within the fermentation process temperature range. Once verified, they are included in the core temperature data of the fermentation tank, recording temperature readings at different locations on the tank wall in real time. Gateway data with the ENV identifier, after temperature and humidity range verification, corresponds to the workshop's environmental temperature and humidity data, simultaneously collecting both ambient temperature and relative humidity. Laboratory test data with the MAT identifier, after physicochemical index range verification, is determined to be fermentation material state data, covering four core physicochemical indicators: sugar content, viscosity, pH value, and solids content. MES push data with the PROC mark... After logical verification of process nodes, it is confirmed that the process proceeds in the order of saccharification, primary fermentation, and post-ripening. After no node jump anomalies occur, the data is classified as fermentation stage progress data, and the current fermentation stage is marked in real time. Abnormal data that fails verification is not directly discarded, but is stored separately in an abnormal data pool and marked with an abnormality type, while triggering a data retransmission request. If three consecutive retransmissions still fail, it is judged as a data acquisition failure, triggering an abnormal data fallback mechanism, thereby reducing the interference of abnormal data on subsequent analysis. All classified data is appended with a unified Unix timestamp, which is uniformly synchronized by the gateway's clock synchronization module to ensure that the time base of the four types of data is completely consistent, providing a unified time scale for subsequent end-to-end timing alignment and phase difference calculation. The categorized data is first stored in a temporary buffer, waiting for the sampling module to read it according to the periodic rules. However, since the entire fermentation process of rice wine is a dynamic biochemical process, the intensity of temperature fluctuations varies significantly at different stages, making it difficult to balance operational efficiency and sampling accuracy with a fixed collection cycle. The industry practice is to uniformly set a compromise frequency, which is simple in operation logic and controllable in load, but the problems of data redundancy during the stable period and insufficient sampling during the active period always exist. To address this, this solution adopts a two-level mechanism that combines an initial baseline and dynamic scaling to generate the collection cycle. This is the first level of the two-level adaptive sampling mechanism, namely the batch-level daily sampling layer, which realizes that the collection frequency automatically adapts to the reaction state. This layer optimizes the daily monitoring collection frequency for the batch level, that is, how often to complete a complete multi-source data collection, for daily status monitoring.
[0020] The first round of operation directly calls the pre-calibrated initial cycle benchmark, which is the core temperature data of the fermenter. The first acquisition cycle benchmark is the temperature node index compiled from historical batches and the update frequency of fermentation stage tags. In this scenario, the initial benchmark is set to 5 minutes / time, perfectly matching the changing rhythm of fermentation stage switching. The interval for workshop environmental temperature and humidity data is dynamically adjusted based on the temperature change rate per unit time; the faster the temperature change rate, the shorter the acquisition interval. The initial second benchmark is 15 minutes / time, and the channel interval is automatically shortened when the temperature change rate exceeds 0.5℃ / h. The fermentation material status data is matched with the frequency of on-site offline detection, setting the third benchmark to 2 hours / time to keep pace with the rhythm of manual sampling and detection. The fermentation stage progress data, combined with the business dependence of production control, sets the fourth benchmark to 30 minutes / time to ensure the timely identification of stage switching. Starting from the second round of operation, the local heterogeneous misalignment entropy of the fermentation system calculated during the previous round of temperature control rest period, such as... Figure 2 As shown, the scaling ratios of the four types of periodic benchmark values will be adjusted. Among them, the local heterogeneity misalignment entropy of the fermentation system is used to quantify the degree of synchronization of fermentation reactions in different spatial regions within the tank. The theoretical value range is 0 to 1, where 0 represents complete homogeneity and the reaction rates in each region are completely consistent, and 1 represents complete heterogeneity and the reaction differences between regions are extremely large. The normal operating range in actual production is 0.1 to 0.8. Its scaling needs to be combined with the previous round. That is, the higher the misalignment entropy detected in the previous round, the worse the uniformity of the reaction in the tank, and the higher the probability of temperature fluctuations in this round, requiring a higher sampling frequency to capture changes. Conversely, when the system is stable, the sampling frequency can be reduced to control the load.
[0021] Based on the misalignment entropy value, the scaling rules are divided into three levels as follows: Within the mildly heterogeneous interval (0–0.3), the scaling factor is set to 0.9–1.0, basically maintaining the initial sampling frequency. Upon entering the moderately heterogeneous interval (0.3–0.6), the scaling factor narrows to 0.6–0.9. In the severely heterogeneous interval (above 0.6), the scaling factor further decreases to 0.4–0.6. Within each interval, linear interpolation is used to calculate the specific scaling factor. Linear interpolation is a common method in numerical computation. In this scheme, the interpolation for the three intervals needs to be determined based on the interval values. Specifically, within the 0–0.3 interval, k(S) = 1.0 – S / 0.3 × 0.1; 0. Within the range of 3 to 0.6, k(S) = 0.9 - (S - 0.3) / 0.3 × 0.3; within the range of 0.6 to 0.8, k(S) = 0.6 - (S - 0.6) / 0.2 × 0.2; when S ≥ 0.8, k(S) takes the minimum value of 0.4; piecewise linear interpolation is used to achieve a continuous and smooth transition of parameters, avoiding parameter jumps during gear switching; the thresholds for all three gears are jointly calibrated by the signal capture rate and system load data of historical batches, achieving an optimal balance between sampling accuracy and operating cost; this dynamic scaling correspondence can be uniformly described by a quantitative formula: T_i = T_i0 × k (S); In the formula, T_i is the actual acquisition cycle after adjustment for the i-th type of data, in minutes / time; T_i0 is the initial cycle baseline value for the i-th type of data, in minutes / time, pre-calibrated by the fermentation stage and data characteristics; k(S) is the cycle scaling factor corresponding to the misalignment entropy, dimensionless, with a value range of [0.4, 1.0], and is a monotonically decreasing piecewise linear function of the misalignment entropy S, with a value range constrained between 0.4 and 1.0, corresponding to the aforementioned three ranges; taking core temperature data as an example, the initial baseline T_i0 = 5 min / time, when the misalignment entropy S = 0.42, i.e., moderate heterogeneity, the corresponding scaling factor k(S) = 0.78, substituting into the formula, we can get the actual acquisition cycle T_i = 5 × 0.78 = 3.9 min / time; Figure 3 As shown, after the scaling calculation is completed, the final executable cycle rule will be sent to the sampling execution module as the basis for daily monitoring sampling; the batch-level cycle optimization is the macro layer of the two-level adaptive sampling mechanism, which is responsible for controlling the overall amount of data in daily monitoring; the subsequent rest period grid optimization is the micro layer, which is responsible for improving the capture accuracy of key features. The two share the same set of misalignment entropy parameters as the adjustment basis, and play a role at different time scales to jointly achieve the optimal allocation of sampling resources.
[0022] Sample set construction and data quality assessment: After determining the batch-level acquisition rhythm, further optimization of the point distribution for high-frequency sampling during the temperature-controlled rest period is performed to improve the capture efficiency of key features. This is the second level of the two-level adaptive sampling mechanism, namely the rest period feature sampling layer. Conventional uniform sampling is highly efficient in processing stable signals, but the heterogeneous features of the rebound process are concentrated in the early stage, and uniform distribution will lead to insufficient sampling density in key periods. Therefore, this scheme introduces a non-uniform sampling time grid to improve the feature capture rate without changing the total sampling amount.
[0023] Non-uniform sampling time grid is an optimization method specifically designed for high-frequency temperature sampling during temperature-controlled rest periods. Its principle is a customized application of the variable sampling rate approach in signal processing. Under the premise of a constant total sampling volume, sampling points are allocated according to the density of signal features. Sampling density is increased during periods of rapid signal change and numerous features, while sampling density is reduced during periods of stable signal and fewer features, thus concentrating sampling resources on effective signals. It is important to clarify that this grid only applies to high-frequency sampling of the core fermenter temperature during temperature-controlled rest periods to accurately capture rebound phase characteristics. Data on workshop environmental temperature and humidity, fermentation material status, and fermentation stage progress, as well as temperature data during daily monitoring periods, are still sampled uniformly according to differentiated periodic rules to ensure a clear data hierarchy and controllable load. High-frequency sampling during temperature-controlled rest periods uses a base frequency of 1 second / sample, automatically triggered by the main control stop signal, independent of the minute-level cycle of daily monitoring. The non-uniform sampling time grid only optimizes the point distribution at this base frequency, without changing the total sampling duration during rest periods, ensuring that the two sampling systems do not interfere with each other.
[0024] After receiving the periodic rules for daily monitoring, they are broken down into corresponding data collection tasks. Each task is matched with a data source address, communication protocol, parsing rules, and verification standards to locate the original traceability data source for each of the four types of data, that is, the source device or system that generated the data, ensuring that all collected data is traceable and verifiable. Specifically, a basic uniform sampling sequence is first generated for each data source to clarify the collection order and basic interval within a single batch. Based on this, the grid is dynamically adjusted for high-frequency sampling during the rest period. The non-uniform sampling time grid is generated using an algorithm combining baseline even distribution and margin tilting. First, a baseline uniform interval is determined based on the total number of sampling points, resulting in a baseline grid. Then, the adjustable sampling point margin is calculated using the previous round of misalignment entropy as a weighting coefficient; the higher the misalignment entropy, the larger the adjustable margin. Finally, all margins are distributed exponentially to the first 40% of the rebound period. The closer to the rebound start point, the higher the sampling point density, ensuring density in high-signal areas while avoiding excessive sparseness in later sampling points. During adjustment, the algorithm strictly controls the total number of sampling points within a preset range of 85%–115% of the baseline uniform sampling, preventing significant fluctuations in the total data volume. The main impact on subsequent processing efficiency lies in ensuring the stability of the data processing load while optimizing the sampling distribution. This is easier to understand with specific numerical examples. Taking temperature sampling during a single temperature control pause as an example, the basic uniform sampling setup includes 100 sampling points with a 1-second sampling interval. When the misalignment entropy is 0.42 (moderate heterogeneity), the total number of points is controlled at 95. Specifically, 60 sampling points are allocated during the early asynchronous phase of the rebound, with the sampling interval compressed to approximately 0.6 seconds, while 35 points are allocated during the later stable phase, with the sampling interval extended to approximately 1.8 seconds. Compared to uniform sampling, the sampling density during critical periods increases by approximately 67%, while the total data volume decreases slightly, achieving a concentrated allocation of sampling resources towards effective signals.
[0025] After completing high-frequency sampling during the rest period according to the adjusted non-uniform time grid, time-series alignment is performed on the four types of daily monitoring data based on a unified timestamp. The alignment adopts the fixed time window method, which is a common processing method for integrating multi-source time-series data. In this scheme, whole minutes are used as the time base, and each window covers a continuous time range of 5 minutes, consistent with the daily acquisition cycle of the core temperature. The arithmetic mean within the window is taken for high-frequency temperature data, and the latest valid value within the window is taken for low-frequency material and stage data. Finally, each window outputs a set of aligned multi-dimensional data to ensure that data from different sampling frequencies can be integrated into the same time dimension. If there is no valid data in the window, it is marked as a time-series breakpoint and included in the continuity score. After alignment, different types of data within the same time window are packaged into a group of samples, and after batch integration, multiple batches of temperature-related data sample sets are formed and sent to the quality assessment module for verification. The two-level sampling optimization and cooperation not only adapt to the overall monitoring rhythm in the batch dimension, but also optimize the capture rate of key features in the rest period dimension, solving the contradiction of redundancy in the stable period and insufficient activity period in fixed sampling.
[0026] After the sample set is formed, quality verification needs to be completed before proceeding to subsequent calculations. Data quality directly determines the reliability of decisions; low-quality data can easily lead to erroneous control. Fixed-weight evaluation systems are logically unified and computationally efficient, suitable for stable operating conditions, but cannot match the different data dimension requirements at different stages of fermentation. We adopt a dynamic weight quality evaluation method, allowing the evaluation criteria to change synchronously with the reaction state. After obtaining the sample set, we integrate and verify the acquisition quality from both structural format and data content perspectives, outputting a temperature data acquisition quality evaluation value. This value is used to correct the accuracy of judgments in subsequent decisions, avoiding interference from low-quality data in temperature control decisions. In terms of structure, data is categorized into three types based on data attributes: structured data, semi-structured data, and unstructured data. The proportion of time-series breakpoints is statistically analyzed, and an information continuity benchmark score is calculated, such as... Figure 4 As shown; the score is calculated using a linearly decreasing formula: Q_cont = 1 - N_break / N_total; In the formula, Q_cont is the information continuity benchmark score, which is dimensionless and ranges from [0, 1]. N_break is the number of time-series breakpoints within the statistical period, and N_total is the total number of sampling points. The score range is 0 to 1, with higher values indicating better time-series integrity, lower breakpoint percentages, and higher continuity scores. Five evaluation criteria are set for the content dimension. Among them, information completeness is judged based on the field missing rate, with higher scores for more complete fields; information accuracy is the deviation rate between the reference and calibration values, with higher scores for smaller deviations; information timeliness is determined by the collection delay time, with higher scores for shorter delays; and information clarity is also considered. The score is determined by the signal-to-noise ratio (SNR) of the data; the higher the SNR, the higher the score. The relevance of information is determined by its correlation with the temperature control target; the higher the correlation, the greater the reference value. Initial weight coefficients are first assigned according to the degree of influence of each dimension on temperature control decisions. The initial weights for the five dimensions are 0.25, 0.25, 0.2, 0.15, and 0.15, respectively, to calculate the baseline score for the content information. All weight adjustments for all dimensions are based on this initial weight as a fixed benchmark, and are adjusted proportionally in conjunction with the previous round's misalignment entropy level. This avoids the accumulation of weights from the previous round as a benchmark, thus preventing the problems of continuous weight shift and distortion of the evaluation scale.
[0027] The adjustment method is based on the fact that the higher the misalignment entropy and the faster the temperature change in this round, the higher the reference value of information timeliness and continuity for decision-making, and the greater the corresponding increase in weight. The other three dimensions are reduced synchronously according to the initial weight ratio, and the sum of the five weights is kept constant at 1 throughout the process to ensure that the evaluation scale is always consistent. In order to ensure that the sum of the weights remains constant after adjustment, a normalization formula derived from the minimum-maximum normalization idea is used to achieve dynamic adjustment. Minimum-maximum normalization is a standard method in the field of data processing. The principle is to map the original data to the interval [0, 1] through linear transformation to eliminate the difference in data dimensions.
[0028] ; In the formula, w_j is the actual weight of the j-th evaluation dimension after adjustment, and the sum of the weights of all dimensions is always equal to 1; w_j0 is the initial weight baseline value of the j-th evaluation dimension; α_j is the adjustment coefficient of the j-th dimension, with positive values for timeliness and continuity dimensions, and zero values for the other dimensions, so that only the weight of timeliness-related dimensions is increased when the misalignment entropy increases. Specifically, only the information timeliness in the content dimension has an adjustment coefficient α=1.0, while the adjustment coefficients for the other four content dimensions are 0, so that only the weight of the timeliness dimension is increased when the misalignment entropy increases; information continuity... The continuity score is an independent indicator at the structural layer and does not participate in the weighting of content dimensions. It is ultimately weighted and merged with the content score at a 3:7 ratio. S represents the current misalignment entropy value. The denominator is the sum of all weighted dimensions, and normalization ensures that the sum of the five weights is always equal to 1. For example, the initial weights are completeness 0.25, accuracy 0.25, timeliness 0.2, clarity 0.15, and relevance 0.15. When the heterogeneity S = 0.42, after normalization, the timeliness weight is approximately 0.26. That is, with S... Taking a value of 0.42 as an example, the calculation is as follows: Completeness 0.25 × (1 + 0) = 0.25, Accuracy 0.25 × (1 + 0) = 0.25, Timeliness 0.2 × (1 + 1.0 × 0.42) = 0.284, Clarity 0.15 × (1 + 0) = 0.15, Relevance 0.15 × (1 + 0) = 0.15; The total numerator is 0.25 + 0.25 + 0.284 + 0.15 + 0.15 = 1.084; After normalization, the weight for timeliness is 0.284 / 1.084 ≈ 0.2 The continuity weight is approximately 0.22, and the remaining 0.52 weights of the other three items are distributed proportionally, keeping the total weights at 1. This evaluation standard better aligns with the decision-making needs under active conditions. Finally, the adjusted information continuity evaluation value and the content information evaluation value are weighted and summed in a 3:7 ratio to obtain the final temperature data acquisition quality evaluation value. The 3:7 ratio is used because the content dimension directly reflects the quality of the data itself, while continuity only reflects temporal completeness. Focusing on content quality as the core and continuity as a supplement is more in line with the evaluation logic.
[0029] The data for the quality assessment is processed in two streams: the daily monitoring batch sample set is sent to the cycle generation module for parameter adaptation and overall status monitoring; the high-frequency temperature data during the resting period, after quality verification, is sent separately to the heterogeneous entropy calculation module as the core input. The fermentation material status data, fermentation stage progress data, and environmental temperature and humidity data are simultaneously input as auxiliary parameters for property calibration and stage matching, thus realizing a classification system to support the whole-chain analysis and ensuring the consistency of data sources.
[0030] Heterogeneous Dislocation Entropy Calculation: Resting period data, after quality verification, is ultimately used in the calculation process. Conventional temperature control schemes always aim for surface temperature to meet standards, assuming that a normal surface temperature equates to overall normality. They never explore the implicit information of the rebound process after temperature control is withdrawn, thus failing to overcome the bottleneck of uneven local fermentation. Our proposed segmented phase mapping method is the core algorithm for calculating dislocation entropy, indirectly inferring internal uniformity solely from tank wall time-series data. Its principle is based on the heat transfer characteristics of the biochemical reaction system. When external temperature control intervention is completely withdrawn, the process of the system temperature rebounding to its natural steady state is entirely determined by the heat production rate and heat transfer characteristics of each internal region. If the bacterial community reaction rate is consistent and heat production is uniform across different regions, the temperature rebound rhythm of each region will be synchronized with minimal phase difference. If a local reaction is too fast and heat production is excessive, the rebound rhythm of the corresponding region will be advanced, exhibiting a phase difference in time. By analyzing the phase difference in the rebound process, the uniformity of the internal reaction can be deduced in reverse.
[0031] The entire calculation process uses temperature-controlled rest periods as the data basis, segmented phase mapping as the core logic, and physical property parameter calibration as the accuracy guarantee, proceeding step by step to obtain the final result.
[0032] First, it's necessary to clarify the trigger boundary for the temperature-controlled rest period. The rest period is determined only when the complete main control cycle ends, the temperature control output signal is completely reset to zero, and the system enters the natural rebound phase. The main control cycle refers to the continuous control process initiated to correct overall temperature deviations. Intermittent starts and stops in graded control are actions within the control execution phase; brief stops do not trigger the rest period determination, avoiding frequent calculations that could distort results. This time period is the only valid data window for calculating misalignment entropy. The main fermentation stage typically lasts 30–60 seconds, and the saccharification and post-ripening stages approximately 60–120 seconds, significantly shorter than the daily control cycle and thus not adding extra system load. The calculation begins... The time point at which the previous main control completely stopped, i.e., the moment when the PLC temperature control output signal was set to 0, is used as the starting point to define the temperature control rest period. The rest period ends when the temperature fluctuation amplitude is less than 0.1℃ for three consecutive sampling cycles. PLC, or Programmable Logic Controller, is a standard device in the industrial control field, used to realize logic control and timing control in industrial settings. It is the general control core of the fermenter temperature control system. The analysis is performed after the temperature control stops to completely eliminate manual intervention from external control, allowing the timing characteristics of the temperature rebound to purely reflect the internal heat transfer and intrinsic characteristics of the system, thus ensuring the accuracy of the analysis results.
[0033] After extracting continuous temperature time-series data at equal time intervals within the rest period, the system is divided into five equal-length subsequences in chronological order. An exponential temperature regression curve is fitted to each subsequence. This curve is a commonly used temperature change fitting model in heat transfer, where the process of an object regressing to a steady state without external intervention follows an exponential change law. The least squares method is used to fit a single exponential model to each subsequence, with the fitting form T(t) = T_∞ + (T_0 - T_∞) × e^(-t / τ), where T(t) is the material temperature at time t (in °C); is the output value of the fitted curve; T_∞ is the final steady-state temperature of the temperature rebound (in °C), the equilibrium temperature finally reached by the material after temperature control stops; T_0 is the initial temperature at the moment temperature control stops (in °C), i.e., the starting temperature of the rebound process; e is the natural constant; t is the time variable of the rebound process (in seconds); and τ is the temperature... The rebound time constant, measured in seconds, characterizes the rate of temperature rebound; a smaller value indicates a faster rebound. After fitting, a complete rebound curve expression is obtained, used for subsequent phase point extraction. For a vertical cylindrical fermenter, during the heat transfer from the center to the wall, it sequentially passes through five typical heat transfer zones: the central zone, the inner middle layer, the middle layer, the outer middle layer, and the near-wall zone. The temperature changes in these different zones exhibit natural temporal differences. Taking the natural cooling rebound after heating stops as an example, the material near the wall zone exchanges heat with the tank jacket faster, resulting in a faster temperature drop, corresponding to the earlier semi-steady-state phase point in the time sequence. The material in the central zone exchanges heat more slowly, leading to a delayed temperature drop, corresponding to a later semi-steady-state phase point in the time sequence. The lag difference between adjacent phase points reflects the degree of delay in heat transfer from the center to the wall. A larger delay indicates a more significant difference in heat generation rates between different zones and a higher degree of internal reaction heterogeneity. This correspondence also applies to the scenario of temperature rebound after cooling stops: The material near the wall exchanges heat with the jacket more quickly, and its temperature rises first, corresponding to the first semi-steady-state phase point in the time sequence. The material in the central region exchanges heat slowly, and its temperature rises later, corresponding to the later semi-steady-state phase point in the time sequence. The phase difference analysis logic is consistent in both temperature control scenarios, and both can be used to inversely deduce the uniformity of the internal reaction. Therefore, by dividing the time sequence into five segments, five spatial regions can be indirectly mapped, allowing for the indirect acquisition of the reaction state in different regions without the need to intrude into the material's interior to place sensors. The five-segment division is based on a 5m... 3The optimized diameter-to-depth ratio of the fermentation tank allows for adjustments to the number of segments based on the actual structure of tanks of different volumes. This addresses the industry problem that surface temperature measurement can only reflect the surface state, achieving non-invasive perception of the internal state through pure data processing. After fitting an exponential temperature regression curve to each subsequence, the time points at which each curve reaches a preset steady-state value are extracted, generating a semi-steady-state phase point sequence. The exponential temperature regression curve is a commonly used temperature change fitting model in heat transfer, and the process of an object regressing to a steady state without external force intervention conforms to an exponential change law. After fitting, the time points when each curve segment reaches the preset steady-state value are extracted to generate a semi-steady-state phase point sequence. The semi-steady-state phase points are extracted using the tangent method, a common method in signal feature extraction. The specific steps are as follows: First, the first derivative of the fitted curve is calculated to obtain the slope change sequence, and the maximum slope value at the rebound initiation point is found. Then, the time axis is traversed backwards, and when the instantaneous slope decays to 50% of the maximum slope, the corresponding moment is recorded as the semi-steady-state phase point for that segment. This ratio matches the preset steady-state value, ensuring that the extracted phase point corresponds to the characteristic moment when the temperature rebounds to the halfway point, exhibiting strong noise resistance and adaptability to industrial field acquisition environments. The semi-steady-state point is selected as the phase marker. The temperature rebound curve exhibits an exponential trend, and the slope of the curve is largest and the change most significant at approximately 50% of the steady-state value. The phase difference between different regions has the highest distinguishability; if the ratio is too low, it is easily affected by acquisition noise, while if the ratio is too high, the signal tends to be flat and the distinguishability decreases significantly; the selectable range of the preset steady-state ratio value is 30% to 70%, with 50% preferred in the main fermentation stage, 40% lower in the saccharification stage, and 60% higher in the post-ripening stage. It can be adjusted according to the actual fermentation stage to obtain the best distinguishing effect; the time difference between two adjacent semi-steady-state phase points is calculated sequentially to form a phase lag difference sequence; the larger the lag difference value, the worse the temperature rebound synchronization in different regions, corresponding to a greater degree of unevenness in the internal reaction. High; The lag difference sequence is normalized using min-max normalization to eliminate numerical magnitude differences; min-max normalization is a well-known standard method in the field of data processing. In this scheme, the normalization formula for the phase lag difference sequence is: Δt'_i = (Δt_i - Δt_min) / (Δt_max - Δt_min), where Δt'_i is the normalized i-th group of relative phase differences, dimensionless, with a value range of [0,1], eliminating absolute numerical differences and retaining only the discrete distribution characteristics of the sequence; Δt_i is the original phase lag of the i-th group of adjacent semi-steady-state phase points. The difference is expressed in seconds; Δt_min is the minimum value in the phase lag difference sequence, in seconds; Δt_max is the maximum value in the phase lag difference sequence, in seconds; after normalization, the sequence is mapped to the [0,1] interval to eliminate absolute numerical differences and retain only the discrete distribution characteristics of the sequence for subsequent entropy value mapping; then, the material reference physical property parameters of the current fermentation stage are substituted into the entropy value calculation model, and the zero-point reference obtained from the pre-calibration is substituted in for error correction, finally obtaining the local heterogeneous misalignment entropy of the fermentation system; the core calculation logic of misalignment entropy can be abstracted into the following expression: S=f[(Δt_i-Δt_i0) / Δt_ref]; In the formula, S is the local heterogeneous dislocation entropy of the fermentation system, with a value range of [0,1]. The higher the value, the worse the reaction uniformity in the tank; Δt_i is the original phase lag difference between adjacent semi-steady-state phase points in the i-th group, in seconds; Δt_i0 is the reference lag difference obtained from the zero-point calibration of the empty tank, in seconds, used to eliminate system errors caused by hardware structure and sensor installation; Δt_ref is the reference lag difference of material properties in the corresponding fermentation stage, in seconds, obtained from the experimental calibration of the corresponding formula and stage, used to eliminate the influence of material viscosity, sugar content and other intrinsic properties on the heat transfer rate; the fractional part calculates the relative phase difference after eliminating the influence of system errors and properties. This reflects the synchronicity differences purely caused by uneven bacterial community responses. The normalized entropy mapping function f(·) is implemented using piecewise linear mapping combined with information entropy weighting. First, the coefficient of variation of the relative phase difference sequence is calculated. When the coefficient of variation is ≤0.1, the entropy value is mapped to the interval 0–0.2; when the coefficient of variation is between 0.1 and 0.3, it is mapped to the interval 0.2–0.6; and when the coefficient of variation is ≥0.3, it is mapped to the interval 0.6–1.0. Simultaneously, information entropy weights are introduced to correct the dispersion of the sequence distribution, ultimately outputting the misalignment entropy value in the 0–1 interval. Essentially, this transforms the dispersion of the phase difference into a quantitative score of uniformity, which is the core transformation from temporal characteristics to physical meaning. Using the previous example, the average of the five phase lag differences is approximately 5.75 seconds. After deducting the zero-point baseline of 0.5 seconds, and dividing by the lag difference of 12.5 seconds based on the physical property baseline of the main fermentation stage, the misalignment entropy S = 0.42 is obtained after entropy mapping, which is consistent with the judgment result of moderate heterogeneity. The calculation process introduces the material baseline physical property parameters precisely to eliminate the influence of the material's own properties on the heat transfer rate. For example, the higher the viscosity of the material, the slower the temperature conduction, and the larger the baseline phase difference. By normalizing and correcting through the baseline coefficient, the interference of physical property differences on the judgment of the degree of heterogeneity can be avoided, ensuring that the entropy value only reflects the uniformity of the reaction itself. After obtaining the misalignment entropy, a homogenization control correction coefficient is first generated based on its value, and then the correction coefficient is multiplied by the temperature data acquisition quality assessment value to finally generate the temperature control decision judgment value. The homogenization control correction coefficient k_d(S) is calibrated using a piecewise method, corresponding one-to-one with the three-level control intervals: Within the 0–0.3 range, the coefficient linearly increases from 1.0 to 1.2, corresponding to a slight fine-tuning; within the 0.3–0.6 range, it linearly increases from 1.2 to 1.8, corresponding to a moderate disturbance; and within the range above 0.6, it linearly increases from 1.8 to 2.5, corresponding to a severe intensification. The coefficients are calibrated based on historical batch control effect data to ensure optimal control intensity matching under different misalignment entropies. The calculation of the final decision judgment value follows the following quantitative relationship: D = S × k_d(S) × Q; In the formula, D is the temperature control decision judgment value, which is ultimately used to compare with the control threshold and match the control level as a core judgment parameter; S is the local heterogeneity misalignment entropy of the fermentation system, reflecting the true degree of heterogeneity of the system; k_d(S) is the homogenization control correction coefficient, with a value range of [1.0, 2.5]; it is also a monotonically increasing piecewise linear function of the misalignment entropy S, used to match different levels of control intensity; Q is the comprehensive evaluation value of temperature data acquisition quality, with a value range of [0, 1]; it is used as a confidence coefficient to correct the decision value. The worse the data quality, the lower the confidence of the decision value; this formula uses the data quality evaluation value to correct the confidence of the heterogeneity result. The higher the data quality, the more confident the misalignment entropy result is, and the closer the decision value is to the original heterogeneity level; data quality The worse the quality, the lower the confidence level, and the more conservative the decision value, thus avoiding mis-regulation caused by low-quality data. For example, when the misalignment entropy S=0.42, substituting it into the formula for the medium-severity range yields the correction coefficient, k_d(S)=1.2+(0.42-0.3) / 0.3×0.6=1.44. When the data quality assessment value Q=0.9, the decision judgment value D=0.42×1.44×0.9≈0.544, which falls within the medium-severity regulation range of 0.36 to 1.08, consistent with the grading rules. After comparing with the preset threshold, medium-severity regulation is triggered. At the same time, the calculated misalignment entropy is synchronously distributed to each unit to adjust the full-link operation parameters for the next round, forming a cyclical iterative logic of the previous round's calculation results, the next round's parameter adaptation, and then the current round's calculation update.
[0034] Homogenization control and iteration: After obtaining the decision judgment value, the control target is upgraded, that is, the transformation from maintaining temperature to maintaining homogeneity; traditional temperature control only pursues surface temperature stability, which can meet the basic fermentation conditions, but cannot solve the quality fluctuation caused by uneven local reaction; this solution adopts a hierarchical homogenization control strategy, with homogenization convergence as the control target, and achieves material mixing by driving natural convection through temperature difference. The principle behind this control method stems from the thermal convection characteristics of fluids: When there is a temperature difference between different areas inside the tank, the materials will generate density differences due to thermal expansion and contraction. The less dense, higher-temperature materials rise, while the denser, lower-temperature materials sink, thus forming natural convection, which drives the materials to mix and gradually reduces the difference in reaction progress between different areas. By controlling the amplitude and rhythm of temperature adjustment, the intensity of convection can be controlled to achieve different degrees of homogenization. After obtaining the decision judgment value, it is compared with the preset graded control threshold. The preset control threshold is the benchmark threshold under standard data quality, which is converted from the benchmark misalignment entropy threshold combined with the standard correction coefficient. It matches the scale of the decision judgment value after quality correction to ensure the accuracy of the judgment logic under normal data quality. Under standard data quality (Q=1), the decision judgment thresholds corresponding to the three levels of control are 0.36 and 1, respectively. 08 and 2.0 correspond to the critical values of misalignment entropy of 0.3, 0.6, and 0.8, respectively. Values below 0.36 are considered normal and require no adjustment. Values between 0.36 and 1.08 trigger moderate adjustment, while values above 1.08 trigger severe adjustment. A severe heterogeneity fallback rule is also set: when the original misalignment entropy value reaches the severe range of 0.6 or higher, regardless of the data quality assessment results, the corresponding level of homogenization adjustment is forcibly triggered to prevent low-quality data from lowering decision values and to shield high-risk conditions, thus ensuring operational reliability from a rule-based perspective. If the threshold is exceeded, the current degree of heterogeneity is deemed too high, and the corresponding level of adjustment strategy is initiated to guide the system towards homogenization. If the threshold is not exceeded, the current temperature control state is maintained, and the misalignment entropy calculated in this round is simultaneously stored in the parameter library for parameter adaptation in the next round.
[0035] The actual operation adopts a three-level graded control mechanism. The control intensity increases progressively with the increase of the misalignment entropy. The direction of temperature adjustment is determined by the deviation between the current tank temperature and the process setpoint. If the current temperature is lower than the setpoint, a heating disturbance is used; if it is higher than the setpoint, a cooling disturbance is used. In the intermittent start-stop mode, the control direction is kept consistent with the original temperature control, and temperature fluctuations are created by the on-off rhythm to drive convection. In the case of mild heterogeneity, i.e., misalignment entropy of 0 to 0.3, only the setpoint temperature is finely adjusted by ±0.3℃, relying on the small temperature difference to drive natural convection, gradually reducing the temperature range. Regional reaction differences have minimal impact on energy consumption and are suitable for minor daily corrections. For moderate heterogeneity, i.e., when the dislocation entropy is 0.3 to 0.6, an intermittent start-stop temperature difference disturbance strategy is adopted, with the temperature control component operating in a cycle of 5 minutes on and 10 minutes off to create periodic temperature fluctuations and enhance the convection effect, which is suitable for moderate non-uniformity corrections. For severe heterogeneity, i.e., when the dislocation entropy is above 0.6, the circulating turbulence pump at the bottom of the tank is linked to the temperature control to accelerate material mixing and quickly promote the homogenization and convergence of the system, which is suitable for scenarios with large deviations.
[0036] Verification based on actual operational data: When moderate heterogeneity (dislocation entropy of 0.42) occurs during the main fermentation stage of glutinous rice wine, an intermittent perturbation strategy is adopted to run a control cycle. The dislocation entropy detected in the next round can be reduced to about 0.28, a decrease of about 33%, which corresponds to a significant improvement in the homogeneity of the system and a clear control effect. Considering the possibility of parameter iteration stagnation under long-term stable conditions, if no main control action is triggered for 4 consecutive hours, a micro-temperature calibration control is automatically initiated. This control is performed by single small temperature rise and fall, with the execution direction preferably selected in the opposite direction to the current temperature trend. If the temperature is stable, a symmetrical small fluctuation of first rising and then falling is adopted to ensure that the overall temperature average remains unchanged. The amplitude is controlled within ±0.2℃ and the duration does not exceed 2 minutes, which is far below the temperature fluctuation tolerance range of normal fermentation and will not have an observable impact on the activity of the microbial community and the fermentation process.
[0037] After the control is stopped, a new temperature control rest period is generated. The data is directly used to update the misalignment entropy parameter without triggering additional graded control. This ensures that even with long-term stable operation, the parameters can iterate synchronously with the changes in physical properties during the fermentation stage. During operation, the progress data of the fermentation stage is read in real time, and the current stage's benchmark parameter library is automatically matched. The entropy calculation model and control threshold are dynamically calibrated to eliminate the interference of material property differences at different fermentation stages on heterogeneity judgment. Data from multiple fermenters in the workshop can be aggregated to the cloud. Through big data training, benchmark models for different formulas and batches are optimized, and the judgment accuracy of a single tank is continuously improved through iteration, forming a positive cycle of single tank operation, cluster data aggregation, model optimization, and single tank accuracy improvement.
[0038] Example 2: A modular deployment is adopted between the workshop edge computing gateway and the central control room server. Each module transmits data via the MQTT industrial IoT protocol. It consists of a first acquisition module, a period generation module, a sampling execution module, a quality assessment module, a heterogeneous entropy calculation module, a decision generation module, and a judgment execution module, as detailed below: The first acquisition module is the data entry point for the entire system. Deployed on the industrial data gateway, it consists of a point location unit and four sets of parallel classification units. The point location unit is responsible for scanning various sensors, gateways, and system interfaces in the workshop, establishing a data access point table, and continuously collecting raw production data. The four sets of classification units adopt a parallel processing mode, classifying and collecting four types of data based on temperature sensor labels, environmental monitoring attributes, material detection labels, and process node markers. Parallel processing can improve data processing efficiency and avoid single-link blockage affecting overall timeliness.
[0039] The cycle generation module is divided into a reference cycle unit and a cycle correction unit. The former is responsible for matching the acquisition cycle reference values corresponding to the four types of data and integrating them to form the initial rules. The latter receives the misalignment entropy parameters output by the heterogeneous entropy calculation module, performs differential scaling on each cycle reference value according to the numerical range, and outputs the final differential temperature acquisition cycle rules to be executed. The sampling execution module consists of a basic sequence unit, a grid adjustment unit, and a sampling integration unit. The basic sequence unit sets a basic sampling sequence for each original traceability data source. The grid adjustment unit adapts the sampling point distribution during the rest period by combining the misalignment entropy and dynamically adjusts the density to form a non-uniform sampling time grid. After completing the collection according to the grid, the sampling integration unit aligns and integrates various types of data and outputs multiple batches of temperature-related data sample sets. The heterogeneous entropy calculation module internally consists of four execution units: a stage delineation unit, a sub-segmentation unit, a phase extraction unit, and an entropy output unit. Specifically, the stage delineation unit is responsible for locking the time node of the previous round of main control cessation and delineating the temperature control rest period that conforms to the rules; the sub-segmentation unit extracts the continuous time-series temperature data within the rest period and splits it into multiple equal-length sub-segment sequences; the phase extraction unit fits the temperature regression curve of each sub-segment, extracts the semi-steady-state phase point, and generates a phase lag difference sequence; after completing the normalization process and entropy calculation, the entropy output unit synchronously sends the misaligned entropy parameters to the period generation module, sampling execution module, quality assessment module, and decision generation module to ensure the consistency of parameters throughout the entire chain. After receiving the misalignment entropy parameter, the quality assessment module dynamically adjusts the weight ratio of the five assessment dimensions based on the initial weight, calculates the final temperature data acquisition quality assessment value, and sends it to the decision generation module. The decision generation module uses misalignment entropy as the core correction parameter and combines it with the adjusted quality assessment value to generate temperature control decision judgment value. The judgment and execution module completes the threshold comparison and control action trigger, and at the same time updates the rules for the next round of collection cycle, driving the entire system to operate in a closed loop.
[0040] It adopts industrial-grade protection standards, is suitable for the temperature, humidity and dust environment of the workshop, and can be directly connected to the PLC control system of the existing fermentation tank.
[0041] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.
[0042] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0043] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes 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.
Claims
1. A method for intelligent temperature control in the production of rice wine, characterized in that, The control method includes the following steps: Collect multiple sets of production source data throughout the entire process of brewing rice wine. The production source data includes core temperature data of fermentation tanks, temperature and humidity data of workshop environment, state data of fermentation materials, and progress data of fermentation stages. By combining the changing characteristics of various production source data with the matching collection frequency, an initial differentiated temperature collection cycle rule is formed; based on this cycle rule, batch data collection is carried out to obtain multiple batches of temperature-related data sample sets. Perform integration and verification on multiple batches of temperature-related data sample sets, and output the benchmark value for temperature data acquisition quality assessment; extract continuous temperature time series data during the temperature control rest period after the previous round of temperature regulation has completely stopped, and obtain the local heterogeneity misalignment entropy of the fermentation system through segmented phase mapping operation; Using the local heterogeneous misalignment entropy of the fermentation system as the core correction parameter, the differential temperature acquisition cycle rule, the sampling time grid arrangement, and the weight ratio of the quality assessment dimension are adjusted sequentially to finally generate the temperature control decision judgment value. The temperature control decision value is compared with the preset control threshold. If the threshold is exceeded, the corresponding level of temperature control strategy is activated to guide the fermentation system to converge toward homogenization. If the threshold is not exceeded, the current temperature control state is maintained, and the differentiated temperature acquisition cycle rules for the next round are updated simultaneously.
2. The intelligent temperature control method for rice wine production according to claim 1, characterized in that, When collecting production source data, first locate multiple data access points in the winemaking production scenario, collect multiple sets of raw production data through each point, and then classify and collect them according to the identification features carried by the data. Among them, the raw data with temperature sensor labels are classified as the core temperature data of the fermentation tank, the data with environmental monitoring attributes correspond to the temperature and humidity data of the workshop environment, the data with material detection labels are determined as the fermentation material status data, and the data marked with process node markers are classified as fermentation stage progress data.
3. The intelligent temperature control method for rice wine production according to claim 1, characterized in that, When generating the initial differentiated temperature acquisition cycle rules, the correspondence between temperature fluctuations and key fermentation nodes is first sorted out based on the core temperature data of the fermenter, and temperature node index and fermentation stage label are built. The update frequency of the index and label is used as the benchmark value of the first acquisition cycle. The temperature and humidity data of the workshop environment are collected by adjusting the collection interval based on the rate of change of temperature and humidity per unit time to obtain the second collection cycle benchmark value; the fermentation material status data are set with reference to the update frequency of status indicators to set the third collection cycle benchmark value; the fermentation stage progress data are matched with the business dependence of the production process to obtain the fourth collection cycle benchmark value. After integrating four types of periodic benchmark values to form initial rules, the benchmark values of each period are then scaled differently according to the numerical range of the local heterogeneous misalignment entropy of the fermentation system. The higher the misalignment entropy value, the higher the collection frequency of various types of data, and finally an executable differentiated temperature collection periodic rule is obtained.
4. The intelligent temperature control method for rice wine production according to claim 1, characterized in that, When acquiring multiple batches of temperature-related data sample sets, multiple data acquisition tasks are first generated based on the differentiated temperature acquisition cycle rules, and corresponding acquisition configuration parameters are matched for each task. The original traceability data sources corresponding to the four types of data are located according to the acquisition configuration parameters, and a basic sampling sequence is set for each data source. The asynchronous period of temperature fluctuation is predicted by combining the local heterogeneous misalignment entropy of the fermentation system. The higher the misalignment entropy value, the higher the sampling point density in the asynchronous period, and the corresponding lengthening of the sampling interval in the synchronous period. The total number of sampling points is controlled within the preset range, thereby forming a non-uniform sampling time grid. Preliminary data collection was conducted in batches for each original traceability data source according to a non-uniform sampling time grid. After obtaining the corresponding category of collected information, the four categories of collected information were integrated by aligning them along the time axis to form a multi-batch temperature-related data sample set.
5. The intelligent temperature control method for rice wine production according to claim 1, characterized in that, The specific steps for calculating the quality assessment value of temperature data acquisition are as follows: Data structure and content features were extracted from multiple batches of temperature-related data samples. The data results were classified into three categories: structured, semi-structured, and unstructured, based on their structural attributes. The information continuity assessment benchmark was then calculated accordingly. Five evaluation dimensions—information completeness, information accuracy, information timeliness, information clarity, and information relevance—are extracted from the data content characteristics. Based on the influence of each dimension on temperature control decisions, corresponding weight coefficient benchmark values are assigned, and the content information evaluation benchmark value is calculated. The weights of each dimension are dynamically adjusted by combining the local heterogeneity of the fermentation system with the misalignment entropy. The higher the misalignment entropy value, the greater the increase in the weight coefficients corresponding to information timeliness and information continuity. The weight coefficients of the other three dimensions decrease proportionally and synchronously, and the sum of the weight coefficients of the five dimensions remains constant. The weighted sum of the information continuity evaluation value and the content information evaluation value is used to obtain the final temperature data acquisition quality evaluation value.
6. The intelligent temperature control method for rice wine production according to claim 1, characterized in that, When calculating the local heterogeneous misalignment entropy of the fermentation system, first lock the time node when the previous round of temperature control action completely stopped running, and define the complete time period after the node when the temperature value gradually returns to the steady state value as the temperature control rest period; extract the continuous time-series temperature data of equal time intervals within this stage, and divide it into multiple sub-sequences of equal length according to the time sequence, with each sub-sequence corresponding to the temperature feedback period of different spatial regions in the fermenter. Fit the temperature regression curve of each sub-segment sequence separately, extract the time nodes when the regression curve of each sub-segment reaches the preset ratio steady state value, and generate a semi-steady state phase point sequence. Calculate the hysteresis difference between adjacent semi-steady-state phase points to form a phase hysteresis difference sequence; after normalizing the phase hysteresis difference sequence, substitute it into the entropy calculation model along with the material reference physical property parameters of the current fermentation stage to obtain the local heterogeneous misalignment entropy of the fermentation system; When generating temperature control decision values, a homogenization regulation correction coefficient is generated based on the local heterogeneity misalignment entropy of the fermentation system. The homogenization regulation correction coefficient is then multiplied by the temperature data acquisition quality assessment value to output the temperature control decision value.
7. A temperature intelligent control system for rice wine production, characterized in that, The control system includes: The first acquisition module collects multiple sets of production source data throughout the entire process of winemaking, including four categories: core temperature data of fermentation tanks, temperature and humidity data of workshop environment, state data of fermentation materials, and progress data of fermentation stages. The cycle generation module combines the changing characteristics of various production source data to match the collection frequency, forms the initial differentiated temperature collection cycle rules, and receives the local heterogeneous misalignment entropy of the fermentation system to complete the cycle parameter correction. The sampling execution module performs batch data acquisition according to the revised periodic rules and outputs multiple batches of temperature-related data sample sets. The quality assessment module performs integrated verification on multiple batches of temperature-related data sample sets, outputs the baseline value for temperature data acquisition quality assessment, and simultaneously receives the local heterogeneity misalignment entropy of the fermentation system to adjust the weight of the assessment dimensions. The heterogeneous entropy calculation module extracts continuous temperature time-series data throughout the entire time period after the temperature control action has completely stopped, and obtains the local heterogeneous misalignment entropy of the fermentation system through segmented phase mapping calculation. The decision generation module uses the local heterogeneity misalignment entropy of the fermentation system as the core correction parameter, and combines the adjusted temperature data acquisition quality assessment value to generate a temperature control decision judgment value. The temperature control decision judgment value is compared with the preset control threshold. If the threshold is exceeded, the corresponding level of temperature control strategy is activated to guide the fermentation system to converge toward homogenization. If the threshold is not exceeded, the current temperature control state is maintained and the rules for the next round of acquisition cycle are updated simultaneously.
8. The intelligent temperature control system for rice wine production according to claim 7, characterized in that, The first acquisition module includes: The point positioning unit completes the positioning of multiple data access points in the winemaking production scenario and collects multiple sets of raw production data; Multiple data classification units, based on temperature sensor labels, environmental monitoring attributes, material detection labels, and process node markers, respectively, classify the raw production data into four categories of production source data.
9. The intelligent temperature control system for brewing rice wine according to claim 7, characterized in that, The period generation module includes: The baseline period unit is matched with the baseline value of the collection period corresponding to the four types of data and integrated to form the initial rule; The periodic correction unit receives the local heterogeneous misalignment entropy of the fermentation system output by the heterogeneous entropy calculation module, performs differential scaling on the baseline values of each period according to the numerical range, and outputs the final differential temperature acquisition periodic rules to be executed. The sampling execution module includes: Basic sequence unit: A basic sampling sequence is set for each original traceability data source; The sample set acquisition unit, combined with the local heterogeneous misalignment entropy of the fermentation system, predicts asynchronous periods of temperature fluctuations, dynamically adjusts the sampling point density to form a non-uniform sampling time grid, and performs data collection and alignment integration according to the non-uniform sampling time grid, outputting multiple batches of temperature-related data sample sets.
10. The intelligent temperature control system for rice wine production according to claim 7, characterized in that, The heterogeneous entropy calculation module includes: The phase is defined as a unit, locking in the time point at which the previous round of temperature control action completely stops and defining the complete temperature control rest period; The phase extraction unit extracts continuous time-series temperature data during the temperature control rest period and splits it into multiple sub-sequences of equal length. It then fits the temperature regression curve of each sub-sequence, extracts the semi-steady-state phase point, and generates a phase hysteresis sequence. The entropy output unit normalizes the phase lag difference sequence and calculates the local heterogeneous misalignment entropy of the fermentation system by combining it with the material baseline physical property parameters of the current fermentation stage. It then simultaneously transmits the entropy to the cycle generation module, sampling execution module, quality assessment module, and decision generation module. The quality assessment module receives the local heterogeneous misalignment entropy of the fermentation system and dynamically adjusts the weight ratio of each assessment dimension, outputting the final temperature data acquisition quality assessment value to the decision generation module.
