Real-time tracking analysis method for temperature data in red ginseng steaming process

Abstract

The invention discloses a real-time tracking analysis method for temperature data in a red ginseng steaming process, and relates to the technical field of industrial process measurement and control. A temperature distribution diagram with high spatial resolution is constructed in combination with multi-sensor fusion analysis and a spatial weighted interpolation algorithm, so that the defect that the traditional process only depends on a few measuring points and cannot comprehensively sense the state of a temperature field is overcome, and the actual temperature distribution of each region in the red ginseng steaming process is intuitively and continuously depicted; on the basis, by identifying the non-uniform temperature area, spatial distribution information is directly introduced into multi-input multi-output control over the opening degree of the steam inlet valve and the rotating speed of the circulating fan, compared with simple threshold value control fed back by a single point or a few measuring points, coordinated adjustment can be conducted on the high-temperature area and the low-temperature area in a targeted mode, and the control accuracy is improved. And the temperature difference between different areas in the steam box is obviously reduced.

Description

Technical Field

[0001] This invention relates to the field of industrial process measurement and control technology, and in particular to a method for real-time tracking and analysis of temperature data during the steaming process of red ginseng. Background Technology

[0002] Red ginseng is a processed ginseng product obtained through steaming, drying, and other processes. One of the key steps in the processing of traditional Chinese medicine is the steaming process. The temperature and time conditions during the steaming process of red ginseng directly affect the transformation and content distribution of active ingredients such as ginsenosides. Therefore, temperature control during the steaming process has a significant impact on the quality stability of red ginseng products.

[0003] In existing technologies, the steaming of red ginseng generally employs equipment such as high-pressure steam ovens. Steam is continuously supplied to the oven to heat the layer of ginseng. Traditional processes rely heavily on operator experience, requiring manual adjustments based on steam pressure, exhaust conditions, and simple temperature displays. This makes it difficult to accurately and promptly monitor the actual temperature changes at different locations within the oven. In recent years, with the development of IoT and data acquisition technologies, solutions have emerged that incorporate temperature sensors within the oven to collect real-time temperature data during the steaming process. These sensors, combined with automated control devices, allow for the opening, closing, or power adjustment of steam valves or heating elements, thereby improving the controllability of the red ginseng steaming process.

[0004] However, in practical applications, uneven temperature distribution is still a common problem inside the steaming chamber of existing red ginseng steaming equipment. Due to factors such as steam flow path, condensate distribution, and red ginseng stacking method, there is often a large temperature gradient between the central and peripheral areas of the steaming chamber, and the thermal history experienced by red ginseng at different levels and locations varies significantly. On the one hand, when the temperature in a local area is too high, it may lead to excessive degradation of some saponin components; on the other hand, the temperature in a low area may cause insufficient conversion reactions of some saponins, thus causing fluctuations in quality indicators such as saponin content, color, and texture of the same batch of red ginseng.

[0005] Existing temperature monitoring and control methods mostly employ a limited number of fixed temperature measurement points, using simple on / off control or single-loop PID control based on the temperature values ​​of a single or a few measurement points and preset thresholds. These methods do not provide a comprehensive and dynamic analysis of the temperature field distribution inside the steam oven. Some solutions propose arranging multiple temperature measurement points inside the steam oven and processing the collected data to improve temperature control. However, limited by data processing capabilities, model building difficulties, and response speed, these methods remain insufficiently adaptable to sudden temperature changes within the steam oven, different loading volumes, or different specifications of red ginseng. They struggle to reflect the true temperature state of each area in a timely and accurate manner, resulting in limited compensation capabilities for temperature non-uniformity.

[0006] Therefore, existing technologies still lack a method that can comprehensively and in real-time track and analyze temperature data during the steaming process of red ginseng, thereby providing a reliable basis for improving the internal temperature distribution of the steamer and optimizing the steaming process parameters. Summary of the Invention

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] This invention provides a method for real-time tracking and analysis of temperature data during the steaming process of red ginseng, which solves the problem that traditional red ginseng steaming relies on a few measuring points and experience control, resulting in uneven temperature field inside the steaming chamber and difficulty in taking into account the differences in working conditions between batches.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for real-time tracking and analysis of temperature data during the steaming process of red ginseng, comprising: Step S1: Multiple temperature sensors are arranged inside the steaming box used for steaming red ginseng. The multiple temperature sensors are set in the central area, edge area and different height positions of the steaming box to form a distributed temperature sensor network. Step S2: During the steaming process of red ginseng, according to the preset sampling period, the multiple temperature sensors collect real-time temperature data at corresponding locations and send the temperature data to the data processing unit. Step S3: The data processing unit performs multi-sensor fusion analysis on the temperature data to obtain temperature distribution data at various spatial locations inside the steam oven, and generates a corresponding temperature distribution map. Based on the temperature distribution data, it determines the temperature non-uniform areas inside the steam oven. Step S4: Based on the determination of the temperature non-uniform area, the control unit calculates the adjustment amount of the steam supply parameters and controls the steam supply device of the steam chamber to adjust according to the adjustment amount, so as to reduce the temperature difference between different areas inside the steam chamber. Steps S2 to S4 are executed cyclically throughout the entire steaming process of red ginseng, which is used to achieve real-time closed-loop tracking and control of the temperature during the steaming process of red ginseng.

[0010] As a preferred embodiment of the real-time temperature data tracking and analysis method for the steaming process of red ginseng described in this invention, in step S1, the multiple temperature sensors are set in the central area of ​​the steaming chamber, the area near the steam inlet, the area near the steam outlet, and the upper and lower layers of red ginseng material areas.

[0011] As a preferred embodiment of the real-time tracking and analysis method for temperature data during the steaming process of red ginseng described in this invention, in step S3, the multi-sensor fusion analysis includes: performing time-domain filtering and outlier removal on the temperature data collected by different temperature sensors, and calculating the temperature value of each grid point in the temperature distribution map based on a spatial weighted interpolation algorithm; In step S3, after time-domain filtering and outlier removal of the data collected by the temperature sensors, the effective space inside the steam chamber is divided into regular grids. For each grid point in the temperature distribution map, multiple temperature sensors with a spatial distance not exceeding a preset threshold are selected as interpolation objects. Interpolation weights are set according to the decreasing spatial distance between the grid point and each temperature sensor, and the interpolation weights are corrected by combining the sensor reliability coefficient obtained based on historical operating performance. The temperature value of the grid point is calculated in a weighted average manner to form a temperature distribution map with high spatial resolution, which is used for subsequent determination of temperature non-uniform areas and adjustment of steam supply parameters.

[0012] As a preferred embodiment of the real-time tracking and analysis method for temperature data during the steaming process of red ginseng described in this invention, the time-domain filtering adopts at least one of Kalman filtering, moving average filtering, or exponentially weighted moving average filtering.

[0013] As a preferred embodiment of the real-time temperature data tracking and analysis method for the steaming process of red ginseng described in this invention, the control unit in step S4 adopts a multi-input multi-output control algorithm to calculate the control quantity used to adjust the opening degree of the steam inlet valve and the speed of the circulating fan based on the deviation between the temperature of each monitoring position in the temperature distribution data and the corresponding target temperature.

[0014] As a preferred embodiment of the real-time tracking and analysis method for temperature data during the steaming process of red ginseng described in this invention, the multi-input multi-output control algorithm is combined with a reinforcement learning algorithm to update the control parameters of the multi-input multi-output control algorithm according to the temperature uniformity evaluation index, so as to adapt to different red ginseng loading amounts and loading methods. The reinforcement learning algorithm constructs a temperature uniformity evaluation index based on the temperature deviation of each monitoring location relative to the corresponding target temperature in the temperature distribution map. At each control sampling moment, the temperature uniformity evaluation index, together with the changes in the steam inlet valve opening and the circulating fan speed control, constitute a reward function. Taking each batch of red ginseng steaming process as a learning iteration cycle, the instantaneous rewards within that batch are accumulated to obtain a batch-level cumulative reward value. Based on the difference between the cumulative reward value and the preset benchmark reward level, as well as the parameter update direction estimated from the relationship between temperature deviation and control quantity changes, the control parameters in the multi-input multi-output control algorithm are adaptively updated between batches, so that the control parameters gradually converge to a parameter region that adapts to different red ginseng loading amounts and loading methods.

[0015] As a preferred embodiment of the real-time tracking and analysis method for temperature data in the steaming process of red ginseng described in this invention, step S3 further includes: predicting the temperature change trend of each monitoring location based on a time series analysis algorithm, and adjusting the steam supply parameters in step S4 in advance based on the predicted future temperature distribution data.

[0016] As a preferred embodiment of the real-time tracking and analysis method for temperature data in the steaming process of red ginseng described in this invention, in step S4, when adjusting the steam supply parameters, a temperature gradient constraint is set, wherein the temperature gradient constraint limits the temperature difference between adjacent monitoring positions in the temperature distribution data to not exceed a preset gradient threshold.

[0017] As a preferred embodiment of the real-time temperature data tracking and analysis method for the steaming process of red ginseng described in this invention, the method further includes: performing correlation analysis between the temperature distribution map and a pre-established red ginseng saponin conversion model; correcting the target temperature curve or target temperature uniformity index based on the degree of saponin conversion obtained from the correlation analysis; and using the corrected target for determining the temperature non-uniform region in step S3 and adjusting the steam supply parameters in step S4.

[0018] As a preferred embodiment of the real-time tracking and analysis method for temperature data during the steaming process of red ginseng described in this invention, the method further includes: storing the collected temperature data, temperature distribution map, and corresponding steam supply parameters during each batch of red ginseng steaming process to form a historical steaming database. The historical steaming database is used for subsequent optimization of the multi-sensor fusion analysis strategy in step S3 and the steam supply parameter adjustment strategy in step S4.

[0019] The beneficial effects of this invention are as follows: By constructing a distributed temperature sensor network covering the center, edges, and upper and lower material areas inside the steaming chamber, and combining multi-sensor fusion analysis and spatial weighted interpolation algorithms to construct a high spatial resolution temperature distribution map, this invention overcomes the shortcomings of traditional processes that rely on only a few measuring points and cannot comprehensively perceive the temperature field. This allows for a direct and continuous depiction of the actual temperature distribution in each area during the steaming process of red ginseng. Furthermore, by identifying areas of uneven temperature, this invention directly incorporates spatial distribution information into the multi-input multi-output control of the steam inlet valve opening and the circulating fan speed. Compared to simple threshold control based on single-point or a few measuring point feedback, this invention can specifically coordinate and adjust high-temperature and low-temperature areas, significantly reducing temperature differences between different areas within the steaming chamber. Simultaneously, by setting temperature gradient constraints, it avoids excessive control actions that could cause new temperature fluctuations, thereby improving the overall uniformity of the temperature field and the stability of the steaming process. Furthermore, this invention introduces a reinforcement learning mechanism based on temperature uniformity evaluation and control quantity changes. The complete steaming process of each batch of red ginseng is considered as a learning iteration cycle. Without altering the existing equipment hardware structure, the parameters of the multi-input multi-output control algorithm are adaptively updated batch-to-batch using historical batch temperature distribution maps, control quantities, and evaluation indicators. This allows the control strategy to gradually converge to a parameter range more suitable for the current steamer structure, red ginseng loading volume, and loading method, solving the problems of fixed control parameters and poor adaptability to changes in loading conditions in traditional methods. Simultaneously, this invention can combine time series analysis to predict temperature change trends at each monitoring location, enabling feedforward adjustments to future temperature distribution and improving response speed to sudden disturbances. It also links the temperature distribution map with the red ginseng saponin conversion model, correcting the target temperature curve or temperature uniformity index through feedback. This transforms temperature control from simply pursuing numerical values ​​to process optimization that considers both component conversion paths and final quality attributes, ultimately improving the repeatability of the red ginseng steaming process and the consistency of product quality across different batches. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0021] Figure 1 This is a flowchart illustrating the real-time temperature data tracking and analysis method for the red ginseng steaming process in the embodiment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0023] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0024] For example, the terms “first” and “second” used in this application are only used to distinguish and describe similar objects, to differentiate the first object from another object, and are not used to describe a specific order or sequence, nor should they be interpreted as indicating or implying relative importance.

[0025] This application proposes a method for real-time tracking and analysis of temperature data during the steaming process of red ginseng, combined with... Figure 1 As shown, the method includes: Step S1: Multiple temperature sensors are arranged inside the steaming box used for steaming red ginseng. The multiple temperature sensors are set in the central area, edge area and different height positions of the steaming box to form a distributed temperature sensor network. Step S2: During the steaming process of red ginseng, multiple temperature sensors collect real-time temperature data at corresponding locations according to a preset sampling period, and send the temperature data to the data processing unit. Step S3: The data processing unit performs multi-sensor fusion analysis on the temperature data to obtain the temperature distribution data of each spatial location inside the steam oven and generates the corresponding temperature distribution map. Based on the temperature distribution data, the temperature non-uniform area inside the steam oven is determined. Step S4: Based on the determination of the temperature uneven area, the control unit calculates the adjustment amount of the steam supply parameters and controls the steam supply device of the steam chamber to adjust according to the adjustment amount, so as to reduce the temperature difference between different areas inside the steam chamber. Among them, steps S2 to S4 are executed cyclically throughout the entire steaming process of red ginseng, which is used to realize real-time closed-loop tracking and control of the temperature during the steaming process of red ginseng. In this embodiment, the steaming oven used for steaming red ginseng can be a high-pressure steam sterilization or steaming equipment commonly used in existing production lines. Multiple temperature sensors are preferably thermocouples or platinum resistance thermometers, with a range covering the temperature range required for steaming red ginseng, such as from room temperature to 150 degrees Celsius. Each temperature sensor is connected to the acquisition interface of the data processing unit via a shielded cable or industrial fieldbus. The data processing unit can be an industrial controller or an industrial computer. The control unit can be implemented by a control program on the data processing unit, or by a control program that interacts with the data processing unit. Specifically, the preset sampling period can be determined based on the combined thermal inertia characteristics of the steaming oven volume, the amount of red ginseng loaded, and the equipment's heating rate. For a medium-sized steaming oven, the sampling period can be set in the range of 1 to 10 seconds, preferably in the range of 2 to 5 seconds, so as to ensure the capture of temperature changes without excessively increasing the computational burden. For example, with a sampling period of two seconds and 16 to 32 sensors, the data processing unit can complete the acquisition of all measurement point data, multi-sensor fusion analysis, and control quantity calculation within each sampling period. In the same period, it can output the updated steam inlet valve opening and circulating fan speed to the steam supply device, achieving the real-time requirements of closed-loop temperature control. Optionally, if one or more temperature sensors experience communication interruption, acquisition failure, or a measurement value significantly exceeding the reasonable operating range of the equipment within a sampling period, the data processing unit can mark the corresponding measurement point within that sampling period as missing or abnormal. This missing or abnormal value can then be removed in subsequent time-domain filtering and outlier removal steps, or temporarily replaced with the most recent valid measurement value, allowing subsequent fusion analysis and control calculations to still be based on the remaining valid measurement point data. Furthermore, to prevent all sensor data from being judged as abnormal or missing in several consecutive sampling cycles, which would lead to distorted control decisions, this embodiment can fix the steam inlet valve opening and circulating fan speed at the safe value of the previous control cycle or the safety level set by the process when it is detected that there are no effective measurement points in multiple consecutive samplings, and trigger fault prompts or shutdown logic, which can be handled by operators or upper-level systems, thereby ensuring the safety of the production process under extreme abnormal conditions.

[0026] In one embodiment, in step S1, multiple temperature sensors are set in the central area of ​​the steam chamber, the area near the steam inlet, the area near the steam outlet, and the upper and lower layers of red ginseng material areas. In one embodiment, step S3 includes: performing time-domain filtering and outlier removal on temperature data collected by different temperature sensors, and calculating the temperature value of each grid point in the temperature distribution map based on a spatial weighted interpolation algorithm. In step S3, after time-domain filtering and outlier removal of the data collected by the temperature sensors, the effective space inside the steam chamber is divided into regular grids. For each grid point in the temperature distribution map, multiple temperature sensors with a spatial distance not exceeding a preset threshold are selected as interpolation objects. The interpolation weights are set according to the decreasing spatial distance between the grid point and each temperature sensor. The interpolation weights are then corrected by combining the sensor reliability coefficients obtained based on historical operating performance. The temperature value of the grid point is calculated in a weighted average manner to form a temperature distribution map with high spatial resolution, which is used for subsequent determination of temperature non-uniform areas and adjustment of steam supply parameters. In step S3, after completing the time-domain filtering and outlier removal of the data collected by each temperature sensor, the effective working space inside the steam oven is discretized into a regular grid, with each grid point corresponding to a pixel unit in the temperature distribution map. Combining the spatial position of each temperature sensor in the steam oven coordinate system, in each sampling period, spatial weighted interpolation is performed on each grid point using the measurement results of adjacent sensors to construct a high-resolution temperature distribution map from the discrete measurement point temperature. Within each sampling period, for any number in the temperature distribution map... For a given grid point, the temperature value of that grid point can be calculated using a distance-based spatial weighted interpolation algorithm. The interpolation relationship can be written as: in, The temperature distribution map is numbered as follows. The interpolated temperature value of the grid points in the current sampling period, in °C. Indicates the number is Temperature data from the temperature sensor during the current sampling period, after time-domain filtering and outlier removal, in °C. Indicates the number is Temperature sensor numbered The spatial interpolation weights of the grid points are dimensionless coefficients. This indicates the number of temperature sensors involved in the interpolation calculation at this grid point. Indicates the number is The grid points and their numbers are The spatial distance of the temperature sensor in the steam oven coordinate system, in meters. The distance weighting index is a real number greater than 0, obtained through experimental calibration of the red ginseng steaming process. This represents the index parameter of the grid points in the temperature distribution map. Indicates the index parameters of the temperature sensor; In practice, for each grid point, the spatial distance between each sensor is calculated in advance based on the geometry and layout of the steam oven. The smaller the distance, the higher the corresponding weight. The larger the value, the more dependent the grid point temperature becomes on spatially proximate measurement data; distance weighting index As the value increases, the influence of nearby measuring points increases, and the interpolation results become more sensitive to sensor layout in local areas. This makes it suitable for steaming conditions where sensors are densely arranged and local temperature gradients are large. (Distance weighting index) When the value decreases, the influence of distant measuring points on the interpolation results increases, which is beneficial for maintaining the overall continuity of the temperature field when the sensors are relatively sparsely arranged. In practical applications, a maximum participation distance based on the size of the steam oven and the spacing between sensors can be given. Among the sensors whose distance exceeds this threshold, only some sensors are selected to participate in the interpolation, so as to reduce the amount of computation and reduce the interference of distant measuring points on local temperature estimation. By repeating the above interpolation calculation for all grid points, a complete temperature distribution map is obtained in each sampling period, which provides temperature information with high spatial resolution for subsequent identification of temperature non-uniform regions and adjustment of multi-input multi-output control algorithms. Specifically, the above implementation method describes the spatial weighted interpolation algorithm based on the construction process of the temperature distribution map during the steaming of red ginseng: by dividing the working space inside the steaming chamber into a regular grid, the originally discrete sensor measurements are transformed into temperature estimates for each grid cell, expanding the temperature field from a small number of measurement points into a distribution map with continuous spatial resolution; the interpolation process introduces a distance-based weight design, making the temperature of the grid points more dependent on the measurement results of sensors that are spatially close, which conforms to the physical characteristic that the temperature inside the steaming chamber changes relatively smoothly with space; the setting of the distance weight index provides a means to adjust the local sensitivity and global smoothness, and can be combined with different red ginseng loading methods and sensor arrangement density, and appropriate values ​​can be selected through empirical calibration, thereby controlling the temperature estimation error within a predetermined range under complex steaming conditions; by filtering and limiting the number of sensors participating in the interpolation by distance, the interference of measurements far from the grid points on the local temperature estimation can be reduced, while reducing the computational burden, so that the interpolation method takes into account both real-time performance and spatial resolution, providing directly usable basic temperature distribution data for subsequent identification of temperature non-uniform areas and adjustment of steam supply parameters; Specifically, when implementing the aforementioned spatial weighted interpolation algorithm, the grid size when dividing the effective space inside the steamer into a regular grid can be selected based on the steamer's geometry, the thickness of the ginseng stack, and the density of temperature sensors. For ginseng steamers with a volume in the range of several cubic meters, the side length of a single grid can be set between 5 cm and 20 cm, ensuring that each grid point is covered by several temperature sensors under most operating conditions. For example, when the grid side length is 10 cm and the actual spacing between temperature sensors is approximately 30 cm, to balance the number of sensors involved in the interpolation with the computational load, the maximum participation distance can be set between 30 cm and 1 meter, preferably around 50 cm, so that each grid point typically has 3-8 sensors involved in the interpolation. The sensor reliability coefficients used in the above interpolation process can be obtained by the data processing unit by statistically analyzing the deviation between the measured value of each sensor and the reference temperature obtained by interpolation based on surrounding sensors within a predetermined historical time window. The reliability coefficients of sensors with stable and small deviations are set to high values ​​close to one, while the reliability coefficients of sensors with large long-term deviations or frequent anomalies are set to low values ​​close to zero. This automatically reduces the impact of suspicious measurement points on grid temperature estimation during weighted averaging. Optionally, to ensure that the interpolation results of each grid point have basic reliability, a preset lower limit can be set for the number of sensors participating in the interpolation, such as requiring at least two or three sensors to participate. When the number of available sensors within the preset maximum participation distance of a certain grid point is less than the lower limit, the data processing unit can appropriately relax the maximum participation distance of that grid point, or degenerate to directly using the measured value of the nearest sensor as the temperature estimate of that grid point. This ensures that a complete temperature distribution map can still be generated even when the sensor deployment is relatively sparse or when some sensors temporarily fail. Similarly, when the reliability coefficient of a certain sensor is consistently lower than a preset threshold over a long historical window, the data processing unit can temporarily not use the sensor's measurement value in the interpolation calculation, but only retain it in the original data record for maintenance personnel to diagnose, thereby reducing the systematic impact of inaccurate sensors on temperature field estimation, while not changing the overall framework of multi-sensor fusion.

[0027] In one embodiment, the time-domain filtering employs at least one of Kalman filtering, moving average filtering, or exponentially weighted moving average filtering. In this embodiment, the specific configuration parameters of the time-domain filter can be set according to the dynamic characteristics of temperature changes and sensor noise levels during the steaming process of red ginseng. For example, when using a moving average filter, the length of the sliding window can be set to include 5-30 sampling points. For a typical operating condition with a sampling period of two seconds, the sliding window length is preferably set to 10 to 20 sampling points to balance the ability to suppress random noise and the timely response to changes in process temperature. When using an exponentially weighted moving average filter, its attenuation coefficient can be set between 0.7 and 0.98. A larger attenuation coefficient places more emphasis on historical data, which is beneficial for obtaining a smooth temperature curve during long-term isothermal stages; a smaller attenuation coefficient places more emphasis on recent measurements, which is suitable for heating and cooling stages and avoids introducing excessive lag through filtering. Furthermore, when using Kalman filtering, the covariance of measurement noise can be estimated based on the nominal accuracy of the temperature sensor and the field test results, and the covariance of process noise can be estimated based on the thermal inertia of the steam oven and the amplitude of control quantity changes. The initial estimated error covariance can be set as a diagonally dominant matrix based on a temperature uncertainty of a few degrees Celsius to ensure that the filter has sufficient robustness in the initial stage. Optionally, during the system startup phase or when resuming operation from maintenance, since historical data is insufficient to directly support the setting of the predetermined window length or filtering parameters, the data processing unit can first use a moving average with a shorter window length or directly use unfiltered data for fusion analysis in the first few sampling periods. After accumulating a certain number of sampling points, it can then switch to the pre-set filtering configuration to reduce the risk of distortion in the filtering results in the initial stage. Similarly, when it is detected that the deviation between the filtered output of a certain sensor and its original measured value repeatedly exceeds the reasonable range over a period of time, it can be considered that the sensor has drift or intermittent faults. The data processing unit can temporarily stop applying time-domain filtering to the sensor and hand over its measured value to the outlier removal and spatial interpolation module to maintain the stability of the overall temperature field estimation.

[0028] In one embodiment, in step S4, the control unit uses a multiple input multiple output control algorithm to calculate the control quantity used to adjust the opening of the steam inlet valve and the speed of the circulating fan based on the deviation between the temperature of each monitored location and the corresponding target temperature in the temperature distribution data. In one embodiment, the multiple input multiple output control algorithm is combined with a reinforcement learning algorithm to update the control parameters of the multiple input multiple output control algorithm according to the temperature uniformity evaluation index, so as to adapt to different red ginseng loading amounts and loading methods. The reinforcement learning algorithm constructs a temperature uniformity evaluation index based on the temperature deviation of each monitoring location relative to the corresponding target temperature in the temperature distribution map. At each control sampling moment, the temperature uniformity evaluation index, together with the change amplitude of control quantities such as the steam inlet valve opening and the circulating fan speed, constitute a reward function. Taking each batch of red ginseng steaming process as a learning iteration cycle, the instantaneous rewards within the batch are accumulated to obtain the batch-level cumulative reward value. Based on the difference between the cumulative reward value and the preset benchmark reward level, as well as the parameter update direction estimated from the relationship between temperature deviation and control quantity change, the control parameters in the multi-input multi-output control algorithm are adaptively updated between batches, so that the control parameters gradually converge to the parameter region that adapts to different red ginseng loading amounts and loading methods. The update of control parameters is specifically a parameter adaptation process based on the reward function and batch iteration, as follows: Step S41, at each control sampling time Within this framework, based on the temperature distribution map obtained in step S3, a temperature uniformity evaluation index is calculated for the selected grid region, for example, using the mean square deviation of the grid temperature relative to the target temperature: in, Indicates at the sampling time The temperature uniformity evaluation index is a numerical value; the smaller the value, the more uniform the temperature inside the steamer. It is a dimensionless quantity. This indicates the number of grid points participating in the uniformity evaluation. Indicates at the sampling time The temperature distribution map is numbered as follows: Temperature values ​​of the grid points, in °C. Indicates at the sampling time The target temperature curve corresponding to this curve provides the target temperature in °C. This indicates the index of a grid point in the temperature distribution map. Indicates the discrete-time index of control and sampling; Step S42, at each sampling time Within this framework, the temperature uniformity evaluation result and the change in control quantity are both incorporated into the reward function. The reinforcement learning algorithm drives the temperature field towards a more uniform and stable state by maximizing the cumulative reward; for example, the following approach could be used: , in, Indicates at the sampling time The instant reward value is a scalar. This represents the weighting coefficient for the temperature uniformity evaluation index, a real number greater than 0, which is determined experimentally to adjust the importance of temperature uniformity in the reward system. This represents the penalty weighting coefficient for changes in the control variable; it is a real number greater than 0, used to limit excessively rapid changes in steam valve opening and circulating fan speed. Indicates at the sampling time The control quantity change vector corresponds to the change amplitude of multiple input control quantities such as steam inlet valve opening and circulating fan speed between two adjacent control moments, and is a column vector. The square of the Euclidean norm of the control quantity change vector is used to measure the magnitude of the overall control adjustment. Step S43: Considering the characteristics of the red ginseng steaming process, the complete steaming process of each batch of red ginseng can be regarded as a reinforcement learning iteration process. All immediate rewards collected within that batch are normalized and accumulated to obtain the batch-level cumulative reward. in, Indicates the first The cumulative reward value corresponding to each batch of steamed red ginseng is a scalar quantity. Indicates the first The number of control and sampling steps that occur during the entire steaming process for each batch. Indicates the sampling time within this batch Instant reward value, This indicates the time index within that batch. An index indicating the batch number of steamed red ginseng; Step S44: The control parameters in the multi-input multi-output control algorithm are uniformly represented in vector form. For example, the proportional coefficient, integral coefficient, and feedforward coefficient between steam pressure and temperature deviation of each actuator are expanded into control parameter vectors. After each steaming batch is completed, the control parameter vector for the next batch can be adaptively updated based on the cumulative reward, for example, using the following formula: , in, Indicates the first The multi-input multi-output control parameter vector used in each steaming batch includes parameters from multiple control channels, such as the steam inlet valve opening adjustment channel and the circulating fan speed adjustment channel. Indicates after the first After the reinforcement learning update for the first steaming batch, in the second... The vector of control parameters to be used in each steaming batch. The learning rate, a real number between 0 and 1, represents the step size for updating the control parameters and is used to adjust the step size of each batch of parameter adjustments. Indicates the first Cumulative rewards earned from each steaming batch. This represents the baseline bonus value, which can be derived from empirical process formulas or the average bonus level of high-performing batches in a historical steaming database. It is used to determine the performance of the current batch relative to the baseline. Indicates the first The parameter update direction vector corresponding to each steaming batch can be calculated based on the correlation between temperature deviation and control variable changes within that batch, or the strategy gradient estimation results. Indicates the batch index of steaming; In actual operation, the control system operates according to the current multi-input multi-output control parameter vector during each batch of steaming. Calculate control variables such as steam valve opening and circulating fan speed, and record temperature distribution maps, temperature uniformity evaluation indicators, and changes in control variables in real time during the process, obtaining the reward at each sampling moment. Cumulative rewards will be calculated after the batch ends. And according to the above formula Updated; after multiple batches of red ginseng steaming process, the reinforcement learning algorithm gradually adjusts the control parameter vector to a region that is more suitable for the current steamer structure, red ginseng loading amount and loading method, so that the area and duration of temperature unevenness in subsequent batches are reduced, improving the uniformity of the temperature field and the repeatability of the process during red ginseng steaming. Specifically, the implementation method described here presents a complete technical chain from the construction of evaluation indicators to the iteration of parameter vectors: First, based on the sampling time, several grid points are selected from the temperature distribution map to form a uniformity evaluation indicator reflecting the overall temperature field dispersion, quantifying the temperature non-uniformity as a scalar sequence that changes over time; then, at each sampling time, the uniformity indicator is combined with the change in control quantity to form a reward function that both constrains the temperature distribution and limits large-amplitude control actions, thus taking into account product quality and actuator lifespan during the optimization process; further, in terms of process rhythm, each batch of red ginseng steaming process is used as an iteration cycle, and the cumulative reward is obtained by normalizing and accumulating the rewards of the entire batch, so that the parameter updates are aligned with the actual production batches, facilitating analysis in conjunction with the historical steaming database; finally, the parameters of multiple channels within the multi-input multi-output controller are uniformly mapped to a parameter vector, and the vector is adjusted batch by batch using an iterative formula composed of the learning rate and update direction, so that the control strategy gradually converges to a parameter region that better balances temperature uniformity and control smoothness in multiple rounds of steaming, providing an implementable algorithmic path for adaptive operating condition control; In this embodiment, the grid points used in the calculation of the temperature uniformity evaluation index can be selected from representative locations covering the red ginseng material area on the temperature distribution map. These representative grid points can be evenly distributed in the central area and near the edge of the steamer, or the density can be appropriately increased in areas with historically large temperature fluctuations to enhance the sensitivity of the evaluation index to temperature non-uniformity in key areas. Specifically, the target temperature curve used in the evaluation process can be selected as the standard heating, holding, and cooling curves given by empirical processes, or as the target temperature trajectory optimized by combining the red ginseng saponin conversion model. The target temperature curve is plotted with the sampling time on the horizontal axis corresponding to the actual control cycle. For example, the weighting coefficient used in the reward function to balance the temperature uniformity evaluation result and the amplitude of control quantity changes can be tuned through multiple batches of steaming verification during the trial production stage. Its typical order of magnitude can be in the range of 0.1-10, adjusted according to the actual product quality and the allowable frequency of actuator actions, so that the control strategy neither excessively pursues perfect temperature uniformity leading to overly frequent control actions, nor fails to effectively reduce temperature non-uniformity due to excessive suppression of control action changes. Furthermore, when iteratively updating the control parameter vector using batch-level cumulative rewards, pre-defined safety upper and lower limits can be set for each control parameter. For example, the parameters of the proportional channel can be limited to a range that ensures system stability, and the integral-related parameters can be limited to a reasonable range that prevents integral saturation. After each batch update, parameters exceeding the safety range are truncated, thereby preventing the controller from entering an unstable or excessively slow response state during the reinforcement learning exploration phase. Optionally, the learning rate parameter can be gradually reduced with the increase of iteration batches. A relatively large learning rate is used in the initial batches to accelerate parameter convergence. After observing that the cumulative reward changes tend to stabilize, the learning rate is gradually reduced so that the changes in control parameters tend to be gradual when approaching the optimal region, reducing performance fluctuations between batches. Similarly, when the cumulative reward of several consecutive batches is significantly lower than the baseline reward value and the statistical results show no improvement trend, the data processing unit can trigger a protection strategy to temporarily freeze the parameter updates of subsequent batches and restore the control parameter vector to the historically better-performing reference value. At the same time, the records of the aforementioned low-reward batches are retained for subsequent manual analysis, thereby ensuring the adaptive capability of reinforcement learning while maintaining the safe and reliable operation of the production process.

[0029] In one embodiment, step S3 further includes: predicting the temperature change trend of each monitoring location based on a time series analysis algorithm, and adjusting the steam supply parameters in step S4 in advance based on the predicted future temperature distribution data. In one embodiment, when adjusting the steam supply parameters in step S4, a temperature gradient constraint is set, which limits the temperature difference between adjacent monitoring locations in the temperature distribution data to not exceed a preset gradient threshold. In one embodiment, the method further includes: performing a correlation analysis between the temperature distribution map and a pre-established red ginseng saponin conversion model; correcting the target temperature curve or target temperature uniformity index based on the degree of saponin conversion obtained from the correlation analysis; and using the corrected target for determining the temperature non-uniform region in step S3 and adjusting the steam supply parameters in step S4. In one embodiment, the method further includes: during each batch of steaming of red ginseng, storing the collected temperature data, temperature distribution map and corresponding steam supply parameters to form a historical steaming database, which is used to optimize the multi-sensor fusion analysis strategy in step S3 and the steam supply parameter adjustment strategy in step S4. In this embodiment, the historical steaming database can be stored by the data processing unit in the form of structured data records. Each data record can at least include a sampling time identifier, temperature data from each temperature sensor after time-domain filtering and outlier removal, an index or compressed description of the temperature distribution map constructed for the corresponding sampling period, steam supply parameters such as the steam inlet valve opening and circulating fan speed at that time, and the temperature uniformity evaluation index and immediate or cumulative reward value for that batch at the corresponding time. Specifically, in order to control storage space usage while ensuring data integrity, the data processing unit can adjust the storage frequency according to the steaming process stage. For example, during the heating and cooling stages, complete records can be stored for each sampling period, while during the long-term constant temperature stage, records can be stored once every several sampling periods. At the same time, spatial downsampling or feature extraction can be performed on the temperature distribution map, retaining only key statistics or low-dimensional feature vectors for subsequent analysis. Furthermore, the historical steaming database can be organized by batch. The records under each batch are associated with the red ginseng loading amount, loading method, target temperature curve number, and quality inspection results after the actual steaming is completed, which facilitates subsequent comparative analysis and offline parameter optimization between different operating conditions and different quality levels. Optionally, during production breaks or periods of low load on the data processing unit, the data processing unit can access data from several representative batches in the historical steaming database to re-evaluate the calculation rules for grid size, maximum participation distance, and sensor reliability coefficients used in spatial weighted interpolation, or recalculate the baseline reward value and initial control parameter vector in the reinforcement learning module. This allows the online control strategy to continuously adapt to new production data and changes in operating conditions during long-term operation. Similarly, when storage space is detected to be nearing its limit, the data processing unit can clean up historical batches according to chronological order or contribution to parameter tuning, prioritizing the retention of batches with representative control effects or significant differences in operating conditions. Batches that are older and have high degree of operating condition repetition are compressed, archived, or stored in summary form. This maintains the database's coverage of diverse steaming conditions under limited storage resources, without affecting the real-time performance and stability of the method in this embodiment during field operation.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0031] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for real-time tracking and analysis of temperature data during the steaming process of red ginseng, characterized in that, include: Step S1: Multiple temperature sensors are arranged inside the steaming box used for steaming red ginseng. The multiple temperature sensors are set in the central area, edge area and different height positions of the steaming box to form a distributed temperature sensor network. Step S2: During the steaming process of red ginseng, according to the preset sampling period, the multiple temperature sensors collect real-time temperature data at corresponding locations and send the temperature data to the data processing unit. Step S3: The data processing unit performs multi-sensor fusion analysis on the temperature data to obtain temperature distribution data at various spatial locations inside the steam oven, and generates a corresponding temperature distribution map. Based on the temperature distribution data, it determines the temperature non-uniform areas inside the steam oven. Step S4: Based on the determination of the temperature non-uniform area, the control unit calculates the adjustment amount of the steam supply parameters and controls the steam supply device of the steam chamber to adjust according to the adjustment amount, so as to reduce the temperature difference between different areas inside the steam chamber. Steps S2 to S4 are executed cyclically throughout the entire steaming process of red ginseng, which is used to achieve real-time closed-loop tracking and control of the temperature during the steaming process of red ginseng.

2. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1, characterized in that, In step S1, the multiple temperature sensors are set in the central area of ​​the steam chamber, the area near the steam inlet, the area near the steam outlet, and the upper and lower layers of red ginseng material area.

3. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1 or 2, characterized in that, In step S3, the multi-sensor fusion analysis includes: performing time-domain filtering and outlier removal on the temperature data collected by different temperature sensors, and calculating the temperature value of each grid point in the temperature distribution map based on a spatial weighted interpolation algorithm; In step S3, after time-domain filtering and outlier removal of the data collected by the temperature sensors, the effective space inside the steam chamber is divided into regular grids. For each grid point in the temperature distribution map, multiple temperature sensors with a spatial distance not exceeding a preset threshold are selected as interpolation objects. Interpolation weights are set according to the decreasing spatial distance between the grid point and each temperature sensor, and the interpolation weights are corrected by combining the sensor reliability coefficient obtained based on historical operating performance. The temperature value of the grid point is calculated in a weighted average manner to form a temperature distribution map with high spatial resolution, which is used for subsequent determination of temperature non-uniform areas and adjustment of steam supply parameters.

4. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 3, characterized in that, The time-domain filtering employs at least one of Kalman filtering, moving average filtering, or exponentially weighted moving average filtering.

5. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1, characterized in that, The control unit in step S4 uses a multiple-input multiple-output (MIMO) control algorithm to calculate the control quantities used to adjust the steam inlet valve opening and the circulating fan speed based on the deviation between the temperature at each monitoring location and the corresponding target temperature in the temperature distribution data.

6. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 5, characterized in that, The multi-input multi-output control algorithm, combined with reinforcement learning algorithm, updates the control parameters of the multi-input multi-output control algorithm according to the temperature uniformity evaluation index, so as to adapt to different red ginseng loading and loading methods. The reinforcement learning algorithm constructs a temperature uniformity evaluation index based on the temperature deviation of each monitoring location relative to the corresponding target temperature in the temperature distribution map. At each control sampling moment, the temperature uniformity evaluation index, together with the changes in the steam inlet valve opening and the circulating fan speed control, constitute a reward function. Taking each batch of red ginseng steaming process as a learning iteration cycle, the instantaneous rewards within that batch are accumulated to obtain a batch-level cumulative reward value. Based on the difference between the cumulative reward value and the preset benchmark reward level, as well as the parameter update direction estimated from the relationship between temperature deviation and control quantity changes, the control parameters in the multi-input multi-output control algorithm are adaptively updated between batches, so that the control parameters gradually converge to a parameter region that adapts to different red ginseng loading amounts and loading methods.

7. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1, characterized in that, Step S3 also includes: predicting the temperature change trend of each monitoring location based on the time series analysis algorithm, and adjusting the steam supply parameters in step S4 in advance based on the predicted future temperature distribution data.

8. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1, characterized in that, In step S4, when adjusting the steam supply parameters, a temperature gradient constraint is set, which limits the temperature difference between adjacent monitoring locations in the temperature distribution data to not exceed a preset gradient threshold.

9. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1, characterized in that, The method further includes: performing correlation analysis between the temperature distribution map and the pre-established red ginseng saponin conversion model; correcting the target temperature curve or target temperature uniformity index based on the degree of saponin conversion obtained from the correlation analysis; and using the corrected target for determining the temperature non-uniform region in step S3 and adjusting the steam supply parameters in step S4.

10. The method for real-time tracking and analysis of temperature data during the steaming process of red ginseng as described in claim 1, characterized in that, The method further includes: storing the collected temperature data, temperature distribution map and corresponding steam supply parameters during each batch of steaming of red ginseng to form a historical steaming database. The historical steaming database is used to optimize the multi-sensor fusion analysis strategy in step S3 and the steam supply parameter adjustment strategy in step S4.

Images