Flour processing constant temperature environment regulation system

Abstract

The application discloses a flour processing constant-temperature environment regulation system, and belongs to the technical field of constant-temperature environment regulation. The system comprises a temperature acquisition module, a variety propagation module, a hierarchical prediction module and an execution module. The temperature acquisition module acquires temperature data of monitoring points. The variety propagation module reads a reference heat generation coefficient in a heat generation coefficient table according to a raw material variety, calculates a difference between expected temperature rise and temperature drop as a feedforward compensation amount, decomposes high-frequency and low-frequency components during fine regulation and control, adjusts a controller output weight, simultaneously adopts a time series model to predict heat and dynamically fuses the predicted heat with initial compensation heat according to a temperature deviation change rate to obtain final compensation heat. The execution module converts the final compensation heat into a power regulation instruction output. The system can realize rapid adaptation of raw material varieties, accurate positioning of abnormal sources, multi-time scale decoupling control, and improvement of the stability and regulation efficiency of flour processing temperature.

Description

Technical Field

[0001] This invention relates to the field of constant temperature environment control technology, and in particular to a constant temperature environment control system for flour processing. Background Technology

[0002] During flour processing, the frictional heat generated in the grinding process and the endothermic effect caused by moisture evaporation cause the processing temperature to fluctuate continuously. A stable processing temperature is one of the key factors in ensuring flour quality (such as gluten network formation, starch gelatinization, color, and flour yield). To maintain temperature stability, existing flour production lines use temperature sensors to monitor key points (such as the mill bearings, wheat humidor, flour sieve surface, and finished product outlet), and use a PID controller as the core, combined with feedforward compensation based on the heat balance equation, to regulate the temperature.

[0003] However, in practical applications, existing technologies exhibit significant differences in the heat generation characteristics of different raw material varieties. Current feedforward compensation models cannot quickly adapt to variety changes, leading to large temperature fluctuations. Different wheat varieties (such as hard red wheat, soft white wheat, and spring wheat) can generate frictional heat differences of over 30% under the same grinding process parameters due to variations in grain hardness, protein content, and endosperm structure. Simultaneously, the moisture content and evaporation rate differ among varieties, causing variations in endothermic effects. Existing feedforward compensation methods often employ fixed heat generation coefficients or simple single-parameter online corrections. When switching raw material varieties on the production line, operators need to repeatedly adjust controller parameters, often taking hours or even days. During this period, processing temperatures frequently exceed the allowable range, causing a decline in flour quality and even producing defective products, severely impacting production efficiency and economic benefits.

[0004] Furthermore, heat is transferred step-by-step along the production line with the materials, exhibiting significant pure lag and multi-timescale characteristics. Conventional PID controllers cannot simultaneously suppress rapid disturbances and eliminate slow drift. In flour processing production lines, heat is primarily transferred step-by-step along the material flow direction (from grinding and sieving to flour mixing and packaging). Temperature fluctuations in the previous process need to pass through material conveying and retention stages before affecting the next process, thus introducing pure lag times of tens to minutes. Simultaneously, the temperature signal contains multiple timescale components: firstly, high-frequency disturbances with periods of several seconds to tens of seconds, mainly originating from instantaneous material flow fluctuations and ambient ventilation; secondly, low-frequency trend drift with periods of several minutes or more, mainly originating from progressive wear of the grinding mill bearings and seasonal changes in ambient temperature. Conventional PID controllers use the same set of gain parameters to process signals across all frequency bands, resulting in insufficient suppression of high-frequency disturbances and excessively slow elimination of low-frequency drift, making it difficult to meet the temperature control accuracy requirements for producing high-quality flour. Summary of the Invention

[0005] In response to the above situation, the present invention can achieve rapid adaptation to different raw material varieties and precise location and control of temperature anomalies through variety-adaptive methods, directional propagation analysis based on material flow direction, and hierarchical control.

[0006] The technical solution is a constant temperature environment control system for flour processing, including: a temperature acquisition module to acquire real-time temperature data at monitoring points;

[0007] The variety propagation module reads the baseline heat generation coefficient from the pre-stored variety heat generation coefficient table based on the raw material variety, calculates the expected temperature rise based on grinding power consumption and flour flow rate, and calculates the expected temperature drop based on ambient humidity and moisture evaporation rate. The difference between the expected temperature rise and the expected temperature drop is used as the feedforward compensation amount. A directed graph is established between monitoring points based on the material flow direction, with the material flow direction as a unidirectional edge. The heat contribution value transferred to the next process is calculated using the flour discharge temperature, flour flow rate, and flour specific heat capacity of the previous process. The ratio of this heat contribution value to the measured temperature rise of the next process is used as the heat propagation coefficient to determine abnormal propagation paths and update the heat propagation delay. After each adjustment, the baseline heat generation coefficient is updated by weighting the ratio of the cumulative compensation heat applied to the processing volume with the baseline heat generation coefficient of the corresponding variety in the variety heat generation coefficient table according to a set ratio.

[0008] The graded prediction module calculates an anomaly score based on the temperature deviation of the monitoring point and the heat transfer coefficient. This anomaly score is calculated by multiplying the temperature deviation rate between the current temperature and the historical average temperature of the same variety by the frequency weight of the monitoring point as the source node in the directed graph. Based on the comparison between the anomaly score and the dynamic threshold, the module selects to directly call the pre-tuned dedicated PID parameters of the monitoring point for regulation, not to regulate, or to initiate fine-grained regulation. During fine-grained regulation, the temperature signal of the monitoring point is decomposed into high-frequency and low-frequency components. The output weights of the fast response controller and the slow integral controller in the graded prediction module are adjusted according to the energy ratio of the high-frequency and low-frequency components, and then the preliminary compensation heat is calculated. At the same time, the compensation heat values ​​of multiple past regulation cycles are recorded, and a time series model is used to predict the compensation heat required for the current cycle. The predicted value and the preliminary compensation heat are dynamically fused based on the current temperature deviation change rate to obtain the final compensation heat.

[0009] The execution module converts the final heat compensation into power regulation commands to control the heating or cooling equipment.

[0010] Furthermore, the set ratio is 3:7, where 3 corresponds to the weight of the current measured value and 7 corresponds to the weight of the historical stored value; the method for establishing the variety heat generation coefficient table is as follows: when the new raw material is processed for the first time, the idle power consumption and temperature change of the grinding mill are measured under no-load conditions to calculate the basic temperature rise coefficient, and the temperature rise under the same process parameters is measured under load. The ratio of the load temperature rise to the basic temperature rise coefficient is taken as the benchmark heat generation coefficient of the variety and stored in the variety heat generation coefficient table; when the benchmark heat generation coefficient is greater than 1.2, the frequency weight of the monitoring point is increased by 20%.

[0011] Furthermore, the calculation method for the heat contribution value is as follows: the flour output temperature of the previous process minus the ambient temperature, multiplied by the flour flow rate and specific heat capacity of the previous process, and then divided by the material residence time of the current process; the heat propagation coefficient is calculated by dividing the heat contribution value by the measured temperature rise of the next process, and then by the normalized result of the flour flow rate. The range of the heat propagation coefficient is 0 to 1; when the heat propagation coefficient is lower than 0.3 for three consecutive sampling points, it is determined that there is an abnormal blockage in the corresponding propagation path. The system then reverses the path from the abnormal propagation path to the nearest upstream production equipment, outputs a cleaning and maintenance prompt, and marks the directed edge as blocked.

[0012] Furthermore, the dynamic threshold is set based on the historical maximum temperature deviation rate: when the anomaly score exceeds 60% of the historical maximum temperature deviation rate, it is judged as a severe anomaly, and the dedicated PID parameter of that monitoring point is directly called for regulation; when the anomaly score is lower than 30% of the historical maximum temperature deviation rate, it is judged as normal fluctuation, and no regulation is initiated; when the anomaly score is between 30% and 60% of the historical maximum temperature deviation rate, it is judged as a moderate anomaly, and fine-tuning is initiated; the dynamic maintenance method of the historical maximum temperature deviation rate is as follows: after each production batch, the maximum temperature deviation rate of each monitoring point in that batch is calculated, and the calculated maximum temperature deviation rate is compared with the stored historical maximum value. If it is greater than the historical maximum value, it is replaced; if three consecutive batches do not exceed 80% of the historical maximum value, the historical maximum value is reduced to 95%; the dedicated PID parameter is tuned separately for each monitoring point, and the regulation output of the dedicated PID parameter is directly output in the event of a severe anomaly, skipping empirical mode decomposition and time series model prediction.

[0013] Furthermore, the division between the high-frequency and low-frequency components is based on the following: empirical mode decomposition is performed on the original temperature signal of the monitoring point to obtain multiple intrinsic mode functions (IMFs). IMFs with an average period of less than 10 seconds are accumulated as high-frequency components, and IMFs with an average period of more than 30 seconds are accumulated as low-frequency components. IMFs with intermediate periods are discarded. The ratio of the proportional gain between the fast response controller and the slow integral controller in the graded prediction module is 5:1. When the proportion of high-frequency energy exceeds 70%, the fast response controller in the graded prediction module undertakes the main regulation task.

[0014] Furthermore, the time series model is an autoregressive moving average model with an order of 2, 1, and 2. The autoregressive part uses the compensation heat values ​​from the first two control periods, the differencing part uses first-order differencing, and the moving average part uses the prediction errors from the first two control periods. The model parameters of the autoregressive moving average model are re-estimated offline every 50 control periods using the least squares method. The temperature deviation change rate is obtained by differencing the temperature deviation values ​​of three consecutive sampling points, with a sampling period of 1 second. The dynamic fusion method is as follows: when the absolute value of the temperature deviation change rate is less than 0.1 degrees Celsius per minute, the predicted value weight is 0.3 and the preliminary compensation heat weight is 0.7; when the absolute value of the temperature deviation change rate is greater than 0.5 degrees Celsius per minute, the predicted value weight is 0.7 and the preliminary compensation heat weight is 0.3; when the temperature deviation change rate is between 0.1 degrees Celsius per minute and 0.5 degrees Celsius per minute, the predicted value weight and the preliminary compensation heat weight are linearly interpolated.

[0015] Furthermore, it also includes a batch switching auxiliary module, which is connected to the variety propagation module and is used to detect raw material variety switching signals. The detection method for raw material variety switching signals is as follows: when the real-time power consumption of the grinding mill changes by more than 20% of the rated power and the flour flow rate changes by more than 15% within 10 seconds, it is determined that the raw material variety has been switched. After the raw material variety is switched, an unloaded temperature rise test is automatically run once. The unloaded temperature rise test lasts for 30 seconds. The actual unloaded temperature rise coefficient obtained by the test is compared with the unloaded coefficient corresponding to the benchmark heat generation coefficient stored in the variety heat generation coefficient table. If the deviation exceeds 10%, a recalibration process is triggered. The load coefficient is recalculated based on the current measured unloaded coefficient, the benchmark heat generation coefficient of the corresponding variety in the variety heat generation coefficient table is updated, and the correction is continuously performed in the subsequent five production batches at a ratio of 3:7.

[0016] Furthermore, it also includes a power protection module, which is connected to the graded prediction module and the execution module. This module monitors the continuous operating time of the heating or cooling equipment. When the cumulative operating time exceeds the safe operating time, it automatically limits the power adjustment command output to the heating or cooling equipment to within 50% of the rated power. At the same time, it calculates the remaining unmet compensation heat, triggers the start of the backup auxiliary equipment, and distributes the remaining compensation heat to the backup auxiliary equipment. When the power limit is triggered, the update frequency of the heat transfer coefficient is synchronously reduced from once per control cycle to once every five control cycles. After the power limit is lifted, the original update frequency is restored.

[0017] Furthermore, it also includes a human-machine interface (HMI). The HMI uses a process flow diagram as a background and displays the temperature values, anomaly scores, heat propagation coefficients, and predicted heat accumulation values ​​of each monitoring point in real time. It also displays the anomaly propagation paths in the form of a directed graph, where the line width of the directed edges is inversely proportional to the heat propagation delay value, and the saturation of the node fill color is directly proportional to the anomaly score. When a propagation path is determined to be abnormally blocked, the propagation path is marked with a flashing red line width, and clicking on the propagation path automatically pops up a cleaning and maintenance prompt window. The cleaning and maintenance prompt window displays the location name of the upstream equipment and the blockage confidence level. The blockage confidence level is determined by the ratio of the number of sampling points with consecutive values ​​below 0.3 to the total number of sampling points.

[0018] Furthermore, the human-machine interface also allows operators to manually modify the baseline heat generation coefficient value in the variety heat generation coefficient table. Manual modification requires the input of a process authorization code, and the modification record, along with the operator's account and timestamp, is stored in an encrypted log. When manual modification occurs, the system automatically triggers the recalibration process in the batch switching auxiliary module, using the manually modified value as the baseline heat generation coefficient for the current variety and resetting the correction starting point for the subsequent five production batches. Simultaneously, the manual modification record is marked as a human intervention sample. After collecting 50 human intervention samples, the initial parameters of the autoregressive moving average model are retrained offline using the stochastic gradient descent method, with the learning rate of the stochastic gradient descent method set to 0.01.

[0019] Due to the adoption of the above technical solutions, the present invention has the following advantages compared with the prior art;

[0020] 1. A product variety heat generation coefficient table is established in the variety propagation module. For each raw material variety, a baseline heat generation coefficient is obtained by comparing the ratio of no-load and load tests, eliminating the influence of individual differences between different grinding mills on the coefficient calibration. After each adjustment, the ratio of the actual applied cumulative compensation heat to the processing volume is updated by weighting the baseline heat generation coefficient in the table at a ratio of 3:7, ensuring that the feedforward compensation can respond to batch fluctuations while maintaining the stability of variety characteristics. When a raw material variety switching signal is detected, the batch switching auxiliary module automatically runs a no-load temperature rise test, compares the measured no-load coefficient with the stored value, and triggers recalibration if the deviation exceeds a certain threshold, continuously correcting it in subsequent production batches. This invention can quickly complete the automatic adaptation of feedforward parameters after variety switching, which helps reduce manual intervention time and maintain the temperature within the process requirements range, thereby solving the problems of large differences in the heat generation characteristics of different raw material varieties and poor adaptability of the feedforward model in the prior art.

[0021] 2. In the product propagation module, a directed graph is established based on the material flow direction. The heat contribution value is calculated using the flour discharge temperature, flour flow rate, and specific heat capacity of the previous process. The ratio of this value to the measured temperature rise of the subsequent process is used as the heat propagation coefficient. When this coefficient consistently falls below a preset lower limit, an abnormal blockage is identified, and a maintenance prompt is output to the upstream equipment. Simultaneously, the directed edge is marked as blocked to prevent subsequent mis-controls. The graded prediction module further calculates the degree of abnormality score based on temperature deviation and the heat propagation coefficient, and compares it with a dynamic threshold based on the historical maximum deviation rate, classifying it into three levels of control strategies: severe abnormality, moderate abnormality, and normal fluctuation. In the case of moderate abnormality, the temperature signal is separated into high-frequency and low-frequency components through empirical mode decomposition. The output weights of the fast response controller and the slow integral controller are adjusted according to the energy ratio. Simultaneously, historical compensation heat values ​​are recorded, and the ARIMA model is used to predict the compensation heat required for the current cycle. The predicted value and the preliminary compensation heat are dynamically merged based on the temperature deviation change rate. It can apply corresponding control strategies to disturbances at different time scales, which is conducive to quickly suppressing high-frequency disturbances, eliminating low-frequency drift without overshoot, and realizing rapid location of fault sources based on the abnormality judgment of propagation coefficient. This solves the problems of heat propagation lag and multi-time scale disturbances in the existing technology, and the difficulty of conventional PID to balance fast suppression and slow integration. Attached Figure Description

[0022] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0023] The foregoing and other technical contents, features and effects of the present invention are described in conjunction with the appendix below. Figure 1 The detailed description of the embodiments will make this clear. All structural details mentioned in the following embodiments are based on the accompanying drawings.

[0024] Example 1: Based on existing technology, this example provides a constant temperature environment control system for flour processing. This system includes a temperature acquisition module, a variety propagation module, a grading prediction module, and an execution module. The temperature acquisition module is used to acquire real-time temperature data from monitoring points in the flour processing production line. Here, "monitoring points" refers to key locations on the flour processing production line where temperature sensors are installed, including but not limited to the mill bearings, the inside of the wheat humidor, the surface of the flour sieve, and the finished product outlet. Setting multiple monitoring points aims to capture the heat distribution at different locations on the production line, providing a spatial data foundation for subsequent anomaly propagation analysis. The variety propagation module is connected to the temperature acquisition module. This module first reads the corresponding baseline heat generation coefficient from a pre-stored variety heat generation coefficient table based on the variety of the currently processed raw material. The "variety heat generation coefficient table" is a pre-established database that stores the heat generation characteristic parameters of different wheat varieties (hard red wheat, soft white wheat, and spring wheat) under standard milling processes. The significance of this coefficient is: the temperature rise multiple of this variety relative to the baseline variety under unit milling power consumption. Reading the baseline heat generation coefficient allows the feedforward compensation to automatically adjust with variety changes, avoiding temperature deviations caused by fixed coefficients. Then, the variety propagation module calculates the expected temperature rise based on the real-time power consumption of the grinder and the flour flow rate, and simultaneously calculates the expected temperature drop based on the ambient humidity and the flour moisture evaporation rate. The "expected temperature rise" is the theoretical temperature rise value obtained by multiplying the grinding power consumption by the baseline heat generation coefficient and then dividing by the flour heat capacity. The "expected temperature drop" is the theoretical temperature drop value obtained by multiplying the moisture evaporation rate by the latent heat of vaporization of water and then dividing by the flour heat capacity.

[0025] The difference between the expected temperature rise and the expected temperature drop is used as a feedforward compensation. This compensation is applied in advance to subsequent controls to offset the effects of grinding heat generation and moisture absorption on temperature, thereby reducing the lag in feedback regulation. The product propagation module also establishes a directed graph between monitoring points based on the material flow direction on the production line, with the direction of material flow as a unidirectional edge. A "directed graph" is a mathematical structure in which each monitoring point is a node, and the direction of material flow from the previous process to the next process is the direction of the edge. Because heat in flour processing is mainly transferred with material transport rather than air radiation, limiting it to unidirectional edges is more in line with physical reality and avoids misjudging reverse heat transfer. The heat contribution value transferred to the next process is calculated using the flour discharge temperature, flour flow rate, and flour specific heat capacity of the previous process. The "heat contribution value" represents the portion of the heat carried by the previous process due to temperature rise that is actually transferred to the next process. Its calculation formula is: (Flour discharge temperature of the previous process - Ambient temperature) × (Flour flow rate of the previous process) × (Flour specific heat capacity) ÷ (Material residence time of the current process). This formula considers heat transfer driven by temperature difference, material flow rate and heat capacity, and heat accumulation caused by residence time. The ratio of this heat contribution value to the measured temperature rise of the subsequent process is used as the heat transfer coefficient. The "heat transfer coefficient" is dimensionless, ranging from 0 to 1, and reflects the actual impact of the temperature change of the previous process on the subsequent process. A coefficient close to 1 indicates high heat transfer efficiency, with almost all the temperature rise of the previous process being conducted to the subsequent process; a coefficient close to 0 indicates heat loss or material blockage. This coefficient is used to determine abnormal propagation paths and update the heat propagation delay. The "heat propagation delay" refers to the time required for a temperature change in the previous process to trigger a temperature response in the subsequent process, which is determined by the material conveying speed and equipment residence time. After each adjustment, the product propagation module updates the baseline heat generation coefficient by weighting the ratio of the actual applied cumulative compensation heat to the processing volume with the baseline heat generation coefficient of the corresponding product in the product heat generation coefficient table according to a set ratio. The "cumulative compensation heat" refers to the total heat output of the heating or cooling equipment from the start of the adjustment until the temperature recovers to the set value. Dividing this by the processing volume yields the compensation heat required per unit processing volume, reflecting the actual heat generation characteristics of this variety during this processing. A weighted average of this actual value and historical stored values ​​is then applied to ensure the baseline heat generation coefficient responds to batch fluctuations while maintaining long-term stability.

[0026] The graded prediction module is connected to the variety propagation module. This module calculates anomaly scores based on the temperature deviation and heat transfer coefficient of the monitoring points. "Temperature deviation" refers to the difference between the current temperature and the historical average temperature of the same variety. "Historical average temperature of the same variety" is the stored average temperature of the variety under the same operating conditions. Dividing the temperature deviation by this average yields the temperature deviation rate, used to eliminate the influence of dimensions. The anomaly score is calculated by multiplying the temperature deviation rate by the frequency weight of the monitoring point as a source node in the directed graph. The "source node frequency weight" refers to the number of directed edges pointing to that node in the directed graph divided by the total number of edges in the directed graph. This weight reflects the influence of the point in the heat propagation network: more incoming edges indicate a greater influence from upstream, but may also indicate that the point itself is a heat accumulation point. This multiplication prioritizes the handling of anomalies at historical heat source points. Based on the comparison between the anomaly score and the dynamic threshold, the graded prediction module chooses to directly call the pre-tuned dedicated PID parameters for the monitoring point for regulation, not regulate, or initiate fine-grained regulation. The "dynamic threshold" is a fixed percentage based on the historical maximum temperature deviation rate, rather than a fixed absolute value, thus allowing it to adaptively adjust to production line aging or seasonal changes. When the score exceeds 60% of the historical maximum deviation rate, it is considered a severe anomaly, requiring the fastest possible intervention. Therefore, dedicated PID parameters are directly invoked, skipping complex decomposition predictions to ensure rapid response. When the score is below 30%, it is considered normal fluctuation, and no control is initiated to avoid frequent actuator movements that could reduce lifespan. When the score falls between these two values, it is considered a moderate anomaly, and refined control is activated.

[0027] During refined control, the hierarchical prediction module decomposes the temperature signal at the monitoring point into high-frequency and low-frequency components. The "high-frequency component" corresponds to temperature fluctuations with a period of less than 10 seconds, mainly originating from instantaneous material flow fluctuations and environmental disturbances; the "low-frequency component" corresponds to temperature trends with a period of more than 30 seconds, mainly originating from gradual equipment wear and daily environmental changes. The output weights of the fast-response controller and the slow-integral controller in the hierarchical prediction module are adjusted according to the energy ratio of the high-frequency and low-frequency components. The "fast-response controller" uses a pure proportional control law to quickly suppress high-frequency disturbances; the "slow-integral controller" uses a proportional-integral control law to eliminate low-frequency drift. The purpose of adjusting the output weights is to allow the controller better suited to handle disturbances in that frequency band to take on the main task, thereby improving the overall control effect. Then, the preliminary compensation heat is calculated. The hierarchical prediction module simultaneously records the compensation heat values ​​from multiple past control cycles and uses a time series model to predict the compensation heat required for the current cycle. The "time series model" chosen is the ARIMA model because it can simultaneously capture the autocorrelation and trend of the heat series. During prediction, the model uses the extrapolation trend of historical compensation heat values ​​to estimate the current required heat. When temperature changes drastically (high rate of change), the historical trend has higher reference value, thus increasing the weight of the predicted value. The hierarchical prediction module dynamically merges the predicted value with the preliminary compensation heat based on the current temperature deviation change rate to obtain the final compensation heat. The "temperature deviation change rate" refers to the rate of change of temperature deviation over time, reflecting the degree to which the system deviates from a steady state. The fusion method is: the larger the change rate, the higher the weight of the predicted value; the smaller the change rate, the higher the weight of the preliminary compensation heat. This is because when the temperature changes rapidly, the preliminary compensation heat calculated based on the method may lag, while prediction based on historical trends is better at capturing the change pattern; when the temperature changes slowly, the method calculation is more reliable. The execution module is connected to the hierarchical prediction module, and the execution module converts the final compensation heat into a power adjustment command and outputs it to the heating or cooling equipment. The "heating or cooling equipment" refers to electric heaters, steam valves, cooling water valves, and fan actuators. The above structure solves the problems of slow raw material switching parameter setting and heat transfer lag leading to control instability in existing technologies.

[0028] In Example 2, based on Example 1, the variety propagation module sets the weighted average ratio to 3:7, where 3 corresponds to the weight of the current measured value and 7 corresponds to the weight of the historical stored value. This ratio is based on the following: statistical analysis of a large amount of flour production data revealed that the coefficient of variation of the heat generation coefficient between different batches of the same variety is approximately 15%, while the difference between different varieties can reach over 30%. Therefore, a higher weight (70%) is given to historical stored values ​​to maintain the stability of variety characteristics, while a 30% weight is given to the measured values ​​to allow the system to track batch fluctuations. The method for establishing the variety heat generation coefficient table is as follows: for each batch of new raw materials processed for the first time, the power consumption and temperature change of the grinding mill are measured under no-load conditions, and the basic temperature rise coefficient is calculated accordingly. "No-load condition" refers to the operating state when there is no material in the grinding mill; at this time, the power consumption is only used to overcome mechanical friction, and the temperature rise is entirely caused by mechanical wear. Then, the temperature rise under the same process parameters is measured under load conditions, and the ratio of the load temperature rise to the basic temperature rise coefficient is taken as the benchmark heat generation coefficient for that variety and stored in the variety heat generation coefficient table. This ratio eliminates individual deviations caused by wear and model differences among different grinding mills, making the heat generation coefficient table universally applicable across equipment of the same model. When the baseline heat generation coefficient is greater than 1.2, it indicates that the grinding heat generation of this variety is significantly higher than that of conventional varieties (a ratio of 1.2 corresponds to a heat generation capacity 20% higher than the average). In this case, the frequency weight of this monitoring point is increased by 20%. This increases the weight of high-heat-generating points in the anomaly scoring, causing the control system to prioritize these locations prone to overheating. The heat contribution value is calculated as follows: subtract the ambient temperature from the flour output temperature of the previous process, multiply by the flour flow rate and specific heat capacity of the previous process, and then divide by the material residence time of the current process. The result is the heat contribution value transferred to the next process. The specific heat capacity of flour is approximately 1.8–2.0 kJ / (kg·K), with 1.9 kJ / (kg·K) taken as the standard value. The material residence time is estimated through equipment geometry and material flow rate. For example, for a conveyor belt with a length of 2 meters and a conveying speed of 0.1 meters / second, the residence time is approximately 20 seconds. The heat transfer coefficient is calculated by dividing the heat contribution value by the measured temperature rise in the subsequent process, and then by the normalized result of the flour flow rate. The coefficient ranges from 0 to 1. The purpose of normalization is to eliminate the influence of differences in flour flow rate in different processes on the coefficient amplitude.

[0029] When the heat transfer coefficient is below 0.3 for three consecutive sampling points, an abnormal blockage is identified in the corresponding propagation path. Intermittent sensor noise is filtered out at these three consecutive sampling points (corresponding to 3-15 seconds). "Abnormal blockage" may include screen blockage, pipe condensation, or material accumulation. In this case, the system reverses its direction along the abnormal propagation path to locate the nearest upstream production equipment, outputs a cleaning and maintenance prompt, and marks the directed edge as blocked. The reverse location method is as follows: find the first upstream node connected to the blocked path along the reverse direction of the directed graph; the equipment corresponding to this node is the source of the fault. After being marked as blocked, this path is temporarily ignored in subsequent heat transfer coefficient calculations to prevent heat loss due to blockage from being mistakenly identified as a downstream equipment malfunction. The dynamic threshold of the graded prediction module is set based on the historical maximum temperature deviation rate: when the anomaly score exceeds 60% of the historical maximum temperature deviation rate, it is judged as a serious anomaly, and the dedicated PID parameter of the monitoring point is directly called for regulation; when the anomaly score is lower than 30% of the historical maximum temperature deviation rate, it is judged as normal fluctuation, and no regulation is initiated; when the anomaly score is between 30% and 60%, it is judged as a moderate anomaly, and fine regulation is initiated.

[0030] The dynamic maintenance method for the historical maximum temperature deviation rate is as follows: After each production batch, the maximum temperature deviation rate of each monitoring point within that batch is calculated. The calculated maximum temperature deviation rate is compared with the stored historical maximum value. If it is greater than the historical maximum value, it is replaced. If three consecutive batches do not exceed 80% of the historical maximum value, the historical maximum value is attenuated to 95%. The attenuation coefficient of 95% is chosen because: under normal circumstances, the normal fluctuation range expansion rate caused by equipment aging or environmental changes is about 0.5% to 1% per batch, and about 1.5% to 3% cumulatively over three batches. Therefore, an attenuation of 5% will neither lower the threshold too quickly, leading to false alarms, nor will it cause the threshold to remain too high for a long time, resulting in a loss of sensitivity. Dedicated PID parameters are individually tuned for each monitoring point. The tuning method can be the critical proportional gain method or the attenuation curve method. In the event of severe anomalies, the control output of the dedicated PID parameters is directly output, skipping empirical mode decomposition and time series model prediction. This ensures a response time of less than 1 second, meeting the needs of emergency intervention. Regarding the decomposition of the temperature signal, the original temperature signal from the monitoring point is subjected to empirical mode decomposition (EMD) to obtain multiple intrinsic mode functions (IMFs). IMFs with an average period of less than 10 seconds are accumulated as high-frequency components, and IMFs with an average period of more than 30 seconds are accumulated as low-frequency components. IMFs with intermediate periods are discarded. The endpoint effect of EMD is handled using the mirror extension method, with the number of extension points being 5% of the original signal length. The decomposition stops when the number of remaining extreme points is less than 3. The average period of each IMF is calculated through zero-crossing detection. The basis for discarding intermediate period IMFs is that the energy in this frequency band is less than 10% of the total energy, and its frequency band overlaps with common interferences in flour processing (such as ambient temperature fluctuations and uneven instantaneous material flow), thus avoiding erroneous adjustments by not participating in control. The proportional gain ratio of the fast response controller to the slow integral controller in the graded prediction module is 5:1.

[0031] The fast response controller employs a proportional control law, with its proportional coefficient Kp_fast set to five times that of the slow integral controller's proportional coefficient Kp_slow. This is because high-frequency disturbances require rapid and significant correction, while low-frequency trends require slow integration for elimination; a higher proportional coefficient accelerates the response speed. When the proportion of high-frequency energy exceeds 70%, the fast response controller assumes the primary regulation task, meaning its output weights exceed those of the slow integral controller. A high-frequency energy proportion exceeding 70% indicates that current temperature fluctuations are dominated by rapid disturbances; therefore, the fast controller takes the lead in suppressing these fluctuations more quickly. The time series model uses an autoregressive moving average model with orders 2, 1, and 2. The autoregressive part uses the compensation heat values ​​from the first two regulation periods, the differencing part uses first-order differencing, and the moving average part uses the prediction errors from the first two regulation periods. The order 2, 1, and 2 is chosen based on AIC criterion analysis of the compensation heat sequences from 20 different flour production lines. The analysis revealed that (2, 1, 2) has the smallest average AIC value, indicating that this order achieves the best balance between model complexity and fitting accuracy. The model parameters are re-estimated offline every 50 control cycles using the least squares method to adapt to seasonal changes in production line conditions. The rate of change of temperature deviation is obtained by differentiating the temperature deviation values ​​of three consecutive sampling points, with a sampling period of 1 second.

[0032] The dynamic fusion method is as follows: when the absolute value of the temperature deviation change rate is less than 0.1 degrees Celsius per minute, the predicted value weight is 0.3, and the initial compensation heat weight is 0.7; when the absolute value of the temperature deviation change rate is greater than 0.5 degrees Celsius per minute, the predicted value weight is 0.7, and the initial compensation heat weight is 0.3; when the temperature deviation change rate is between the two, the weights are linearly interpolated. The boundary values ​​of 0.1℃ / min and 0.5℃ / min are determined based on the step response experiment of the flour temperature control system: when the temperature change rate is less than 0.1℃ / min, the PID feedback is sufficient to eliminate the deviation, and an excessively high predicted value weight will introduce noise; when the change rate is greater than 0.5℃ / min, the PID response lags significantly, and it is necessary to rely on the predicted value for early compensation. This embodiment also includes a batch switching auxiliary module, which is connected to the variety propagation module and is used to detect the raw material variety switching signal. The detection method is: when the real-time power consumption of the grinder changes by more than 20% of the rated power within 10 seconds and the flour flow rate changes by more than 15%, it is determined that the raw material variety has been switched. These two thresholds avoid random fluctuations in normal production (power consumption fluctuations within 5% and flow rate fluctuations within 10%), while quickly capturing sudden changes in process parameters caused by product switching. After the switch is completed, the system automatically runs an no-load temperature rise test for 30 seconds. 30 seconds is sufficient for the grinding mill to reach a stable temperature under no-load operation without excessively occupying production time. The actual no-load temperature rise coefficient obtained from the test is compared with the no-load coefficient corresponding to the benchmark heat generation coefficient stored in the product heat generation coefficient table. If the deviation exceeds 10%, a recalibration process is triggered. The load coefficient is recalculated based on the current measured no-load coefficient, and the benchmark heat generation coefficient of the corresponding product in the product heat generation coefficient table is updated. This is continuously corrected at a 3:7 ratio in the subsequent five production batches. The five batches are set based on the following: the quality fluctuations of the first five batches of a newly changed product are relatively large, requiring enhanced temperature control monitoring; after five batches, the process tends to stabilize, and the regular update frequency is restored. This embodiment also includes a power protection module, which is connected to the graded prediction module and the execution module to monitor the continuous operating time of the heating or cooling equipment. When the cumulative operating time exceeds the safe operating time (e.g., the equipment manual specifies continuous operation for no more than 30 minutes), the power adjustment command output to the heating or cooling equipment will be automatically limited to within 50% of the rated power. Simultaneously, the remaining unmet compensation heat will be calculated, triggering the startup of the backup auxiliary equipment, and the remaining compensation heat will be distributed to the backup auxiliary equipment. This prevents overheating damage to the main equipment without affecting the temperature control effect.

[0033] When power limitation is triggered, the system synchronously reduces the update frequency of the heat propagation coefficient from once per control cycle to once every five control cycles. This is because the main equipment is already under power limitation, reducing the significance of rapid updates to the propagation coefficient and decreasing computational burden. The original update frequency is restored after the power limitation is lifted. This embodiment also includes a human-machine interface (HMI). The HMI uses a process flow diagram as a background and displays the temperature values, anomaly scores, heat propagation coefficients, and predicted cumulative heat values ​​at each monitoring point in real time. Anomaly propagation paths are overlaid in a directed graph, where the line width of the directed edges is inversely proportional to the heat propagation delay (a larger delay results in a thinner line width, indicating a weaker impact). The saturation of the node fill color is directly proportional to the anomaly score (a higher score results in a darker color). When a propagation path is determined to be abnormally blocked, it is marked with a flashing red line width. Clicking on the propagation path automatically pops up a cleaning and maintenance prompt window, displaying the location name of the upstream equipment and the blockage confidence level. This blockage confidence level is determined by the ratio of the number of sampling points with consecutive values ​​below 0.3 to the total number of sampling points. For example, if three consecutive sampling points are below 0.3 and the total number of sampling points is 3, the confidence level is 100%; if three out of six consecutive sampling points are below 0.3, the confidence level is 50%. This visualization method allows operators to intuitively locate fault points without analyzing large amounts of data. The human-machine interface also allows operators to manually modify the baseline heat generation coefficient value in the variety heat generation coefficient table. Manual modification requires inputting a process authorization code, and the modification record, along with the operator's account and timestamp, is stored in an encrypted log. When a manual modification occurs, the system automatically triggers the recalibration process in the batch switching auxiliary module, using the manually modified value as the baseline heat generation coefficient for the current variety and resetting the correction starting point for the subsequent five production batches. Simultaneously, the manual modification record is marked as a human intervention sample. After collecting 50 human intervention samples, the initial parameters of the autoregressive moving average model are retrained offline using stochastic gradient descent with a learning rate of 0.01. This self-learning mechanism allows the system to gradually absorb the experience of senior operators and continuously optimize the model's predictive capabilities.

[0034] In practical application, based on existing technology, the temperature acquisition module's sensors are first installed at key monitoring points in the flour processing production line. These monitoring points include, but are not limited to, the mill bearings, the wheat humidor, the flour sieve surface, and the finished product outlet. After installation, a no-load test and a load test are run for each raw material to be processed, recording the mill's no-load power consumption and temperature rise, as well as its load power consumption and temperature rise. The system automatically calculates the baseline heat generation coefficient for each raw material and stores it in a product heat generation coefficient table. After the above preparations are completed, the system can be put into automatic operation.

[0035] When the production line detects a raw material variety switching signal (i.e., the real-time power consumption of the grinder changes by more than 20% of the rated power within 10 seconds and the flour flow rate changes by more than 15%), the batch switching auxiliary module automatically starts a 30-second no-load temperature rise test. It compares the measured no-load coefficient with the no-load coefficient stored in the variety heat generation coefficient table. If the deviation exceeds 10%, it triggers a recalibration process, recalculates the load coefficient based on the current measured no-load coefficient, updates the benchmark heat generation coefficient of the corresponding variety, and continuously corrects it at a ratio of 3:7 in the subsequent five production batches.

[0036] During production, the temperature acquisition module continuously sends real-time temperature data to the variety propagation module and the grading prediction module. The variety propagation module calculates the expected temperature rise in real time based on the current variety's baseline heat generation coefficient, combined with grinding power consumption and flour flow rate. Simultaneously, it calculates the expected temperature drop based on ambient humidity and moisture evaporation rate, generating a feedforward compensation amount. This compensation amount is applied to the execution module before temperature disturbances occur, effectively reducing the lag time of feedback adjustment.

[0037] The graded prediction module calculates an anomaly score based on temperature deviation and heat transfer coefficient, and compares it with a dynamic threshold based on the historical maximum temperature deviation rate. When the score exceeds 60% of the historical maximum deviation rate, the system determines it as a severe anomaly and directly calls the dedicated PID parameters for that location for rapid adjustment; when the score is below 30%, it is determined as normal fluctuation, and no adjustment is initiated to avoid frequent actuator actions; when the score is between the two, fine-grained adjustment is initiated. The fine-grained adjustment process includes: performing empirical mode decomposition on the temperature signal at that location, separating high-frequency components with a period of less than 10 seconds and low-frequency components with a period of more than 30 seconds, adjusting the output weights of the fast response controller and the slow integral controller according to the energy ratio of high frequency to low frequency (proportional coefficient ratio of 5:1), and calculating the initial compensation heat; simultaneously, based on the compensation heat values ​​of the past 50 adjustment cycles, the ARIMA(2,1,2) model is used to predict the compensation heat required for the current cycle, and the predicted value and the initial compensation heat are dynamically fused according to the temperature deviation change rate (the dividing values ​​are 0.1℃ / min and 0.5℃ / min) to obtain the final compensation heat. The execution module converts the final compensated heat into a power adjustment command and outputs it to the heating or cooling equipment. After adopting this hierarchical prediction fusion strategy, the system's suppression time for high-frequency disturbances can be shortened to less than 3 seconds, and there is no overshoot in the adjustment of low-frequency drift. The temperature control accuracy meets the requirements for high-quality flour production (typical process tolerances: mill bearing ±1.5℃, wheat humidor ±2.0℃, finished flour outlet ±0.8℃).

[0038] The product propagation module simultaneously monitors the heat transfer efficiency between each process in real time based on the heat transfer coefficient. When the heat transfer coefficient of a certain propagation path is lower than 0.3 for three consecutive sampling points, the system determines that there is an abnormal blockage in the path, automatically reverses the location to the nearest upstream production equipment, and marks the blocked path with a red flashing directed edge on the human-machine interface. Clicking on it will bring up the location name of the upstream equipment and the blockage confidence level.

[0039] The power protection module monitors the continuous operating time of heating or cooling equipment. When the cumulative operating time exceeds the safe operating time specified in the equipment manual, it automatically limits the output power to within 50% of the rated power and triggers the backup auxiliary equipment to share the remaining compensation heat demand, thereby avoiding overheating damage to the main equipment.

[0040] The human-machine interface allows operators to manually modify the baseline heat generation coefficient value in the variety heat generation coefficient table based on practical experience. Modification requires an authorization code, and the modification is recorded and stored in an encrypted log. After collecting 50 manual intervention samples, the system uses stochastic gradient descent (learning rate 0.01) to retrain the initial parameters of the ARIMA model offline, allowing the model to gradually incorporate the operator's experience-based judgment logic.

[0041] The above description is a further detailed explanation of the invention in conjunction with specific embodiments, and it should not be considered that the specific embodiments of the invention are limited to this. For those skilled in the art to which this invention pertains and related fields, any extensions, operation methods, and data substitutions made based on the technical solution concept of this invention should fall within the protection scope of this invention.

Claims

1. A constant temperature environment control system for flour processing, characterized in that, include: The temperature acquisition module obtains real-time temperature data from the monitoring points. The variety propagation module reads the baseline heat generation coefficient from the pre-stored variety heat generation coefficient table based on the raw material variety, calculates the expected temperature rise by combining the grinding power consumption and flour flow rate, and calculates the expected temperature drop based on the ambient humidity and moisture evaporation rate. The difference between the expected temperature rise and the expected temperature drop is used as the feedforward compensation amount. Furthermore, a directed graph is established between monitoring points based on the material flow direction, with the material flow direction as a unidirectional edge. The heat contribution value transferred to the next process is calculated using the flour discharge temperature, flour flow rate, and flour specific heat capacity of the previous process. The ratio of this heat contribution value to the measured temperature rise of the next process is used as the heat propagation coefficient, which is used to determine abnormal propagation paths and update the heat propagation delay. After each adjustment is completed, the baseline heat generation coefficient is updated by weighting the ratio of the cumulative compensation heat applied to the processing volume with the baseline heat generation coefficient of the corresponding variety in the variety heat generation coefficient table according to the set ratio. The graded prediction module calculates an anomaly score based on the temperature deviation of the monitoring point and the heat transfer coefficient. This anomaly score is calculated by multiplying the temperature deviation rate between the current temperature and the historical average temperature of the same variety by the frequency weight of the monitoring point as the source node in the directed graph. Based on the comparison between the anomaly score and the dynamic threshold, the module selects to directly call the pre-tuned dedicated PID parameters of the monitoring point for regulation, not to regulate, or to initiate fine-grained regulation. During fine-grained regulation, the temperature signal of the monitoring point is decomposed into high-frequency and low-frequency components. The output weights of the fast response controller and the slow integral controller in the graded prediction module are adjusted according to the energy ratio of the high-frequency and low-frequency components, and then the preliminary compensation heat is calculated. At the same time, the compensation heat values ​​of multiple past regulation cycles are recorded, and a time series model is used to predict the compensation heat required for the current cycle. The predicted value and the preliminary compensation heat are dynamically fused based on the current temperature deviation change rate to obtain the final compensation heat. The execution module converts the final heat compensation into power regulation commands to control the heating or cooling equipment.

2. The flour processing constant temperature environment control system according to claim 1, characterized in that, The set ratio is 3:7, where 3 corresponds to the weight of the current measured value and 7 corresponds to the weight of the historical stored value. The method for establishing the heat generation coefficient table of the variety is as follows: when the new raw material is processed for the first time, the idle power consumption and temperature change of the grinding mill are measured under no-load conditions to calculate the basic temperature rise coefficient. The temperature rise under the same process parameters is measured under load. The ratio of the load temperature rise to the basic temperature rise coefficient is taken as the benchmark heat generation coefficient of the variety and stored in the heat generation coefficient table of the variety. When the baseline heat generation coefficient is greater than 1.2, the frequency weight of the monitoring point is increased by 20%.

3. The flour processing constant temperature environment control system according to claim 2, characterized in that, The calculation method for the heat contribution value is as follows: the flour output temperature of the previous process minus the ambient temperature, multiplied by the flour flow rate and specific heat capacity of the previous process, and then divided by the material residence time of the current process; the heat propagation coefficient is equal to the heat contribution value divided by the measured temperature rise of the next process, and then divided by the normalized result of the flour flow rate. The value range of the heat propagation coefficient is 0 to 1; when the heat propagation coefficient is lower than 0.3 for three consecutive sampling points, it is determined that there is an abnormal blockage in the corresponding propagation path. The system reverses the positioning from the abnormal propagation path to the nearest upstream production equipment, outputs a cleaning and maintenance prompt, and marks the directed edge as blocked.

4. The flour processing constant temperature environment control system according to claim 3, characterized in that, The dynamic threshold is set based on the historical maximum temperature deviation rate: when the anomaly score exceeds 60% of the historical maximum temperature deviation rate, it is judged as a serious anomaly, and the dedicated PID parameter of that monitoring point is directly called for regulation; when the anomaly score is lower than 30% of the historical maximum temperature deviation rate, it is judged as normal fluctuation, and no regulation is initiated; when the anomaly score is between 30% and 60% of the historical maximum temperature deviation rate, it is judged as a moderate anomaly, and fine-tuning is initiated; the dynamic maintenance method of the historical maximum temperature deviation rate is as follows: after each production batch, the maximum temperature deviation rate of each monitoring point in that batch is calculated, and the calculated maximum temperature deviation rate is compared with the stored historical maximum value. If it is greater than the historical maximum value, it is replaced; if three consecutive batches do not exceed 80% of the historical maximum value, the historical maximum value is reduced to 95%; the dedicated PID parameter is tuned separately for each monitoring point, and the regulation output of the dedicated PID parameter is directly output in the case of serious anomaly, skipping empirical mode decomposition and time series model prediction.

5. The flour processing constant temperature environment control system according to claim 3, characterized in that, The division between high-frequency and low-frequency components is based on the following: empirical mode decomposition is performed on the original temperature signal of the monitoring point to obtain multiple intrinsic mode functions (IMFs). IMFs with an average period of less than 10 seconds are accumulated as high-frequency components, and IMFs with an average period of more than 30 seconds are accumulated as low-frequency components. IMFs with intermediate periods are discarded. The ratio of the proportional gain between the fast response controller and the slow integral controller in the graded prediction module is 5:

1. When the proportion of high-frequency energy exceeds 70%, the fast response controller in the graded prediction module undertakes the main regulation task.

6. The flour processing constant temperature environment control system according to claim 5, characterized in that, The time series model is an autoregressive moving average model with orders of 2, 1, and 2. The autoregressive part uses the compensation heat values ​​from the first two control periods, the differencing part uses first-order differencing, and the moving average part uses the prediction errors from the first two control periods. The model parameters of the autoregressive moving average model are re-estimated offline every 50 control periods using the least squares method. The temperature deviation change rate is obtained by differencing the temperature deviation values ​​of three consecutive sampling points, with a sampling period of 1 second. The dynamic fusion rules are as follows: when the absolute value of the temperature deviation change rate is less than 0.1 degrees Celsius per minute, the predicted value weight is 0.3 and the preliminary compensation heat weight is 0.7; when the absolute value of the temperature deviation change rate is greater than 0.5 degrees Celsius per minute, the predicted value weight is 0.7 and the preliminary compensation heat weight is 0.3; when the temperature deviation change rate is between 0.1 degrees Celsius per minute and 0.5 degrees Celsius per minute, the predicted value weight and the preliminary compensation heat weight are linearly interpolated.

7. The flour processing constant temperature environment control system according to claim 2, characterized in that, It also includes a batch switching auxiliary module, which is connected to the variety propagation module and is used to detect raw material variety switching signals. The detection method for raw material variety switching signals is as follows: when the real-time power consumption of the grinding mill changes by more than 20% of the rated power and the flour flow rate changes by more than 15% within 10 seconds, it is determined that the raw material variety has been switched. After the raw material variety is switched, an idle temperature rise test is automatically run once. The idle temperature rise test lasts for 30 seconds. The actual idle temperature rise coefficient obtained by the test is compared with the idle coefficient corresponding to the benchmark heat generation coefficient stored in the variety heat generation coefficient table. If the deviation exceeds 10%, a recalibration process is triggered. The load coefficient is recalculated based on the current measured idle coefficient, the benchmark heat generation coefficient of the corresponding variety in the variety heat generation coefficient table is updated, and the correction is continuously carried out in the subsequent five production batches at a ratio of 3:

7.

8. The flour processing constant temperature environment control system according to claim 1, characterized in that, It also includes a power protection module, which is connected to the graded prediction module and the execution module. The power protection module is used to monitor the continuous operating time of the heating or cooling equipment. When the cumulative operating time exceeds the safe operating time, it automatically limits the power adjustment command output to the heating or cooling equipment to within 50% of the rated power. At the same time, it calculates the remaining unmet compensation heat, triggers the start of the backup auxiliary equipment, and distributes the remaining compensation heat to the backup auxiliary equipment. When the power limit is triggered, the update frequency of the heat transfer coefficient is reduced synchronously from once per control cycle to once every five control cycles. After the power limit is lifted, the original update frequency is restored.

9. The flour processing constant temperature environment control system according to claim 3, characterized in that, It also includes a human-machine interface (HMI), which uses a process flow diagram as a background and displays the temperature values, anomaly scores, heat propagation coefficients, and predicted heat accumulation values ​​of each monitoring point in real time. The HMI also displays the anomaly propagation paths in the form of a directed graph, where the line width of the directed edges is inversely proportional to the heat propagation delay value, and the saturation of the node fill color is directly proportional to the anomaly score. When a propagation path is determined to be abnormally blocked, the propagation path is marked with a flashing red line width, and clicking on the propagation path will automatically pop up a cleaning and maintenance prompt window. The cleaning and maintenance prompt window displays the location name of the upstream equipment and the blockage confidence level. The blockage confidence level is determined by the ratio of the number of sampling points with consecutive values ​​below 0.3 to the total number of sampling points.

10. The flour processing constant temperature environment control system according to claim 7 or 9, characterized in that, The human-machine interface also allows operators to manually modify the baseline heat generation coefficient value in the variety heat generation coefficient table. Manual modification requires inputting a process authorization code, and the modification record, along with the operator's account and timestamp, is stored in an encrypted log. When manual modification occurs, the system automatically triggers the recalibration process in the batch switching auxiliary module, using the manually modified value as the baseline heat generation coefficient for the current variety and resetting the correction starting point for the subsequent five production batches. At the same time, the manual modification record is marked as a human intervention sample. After collecting 50 human intervention samples, the initial parameters of the autoregressive moving average model are retrained offline using the stochastic gradient descent method, with the learning rate of the stochastic gradient descent method set to 0.01.

Images