Batch steady state identification method and system for threshing and redrying production process

Abstract

The application discloses a batch steady state identification method and system for a threshing and redrying production process, relates to the technical field of tobacco, and acquires original time series of key process parameters of a threshing and redrying production line through a manufacturing execution system, adopts a Binary Segmentation algorithm to perform recursive segmentation on a pretreated target time series to detect a significant mutation point, divides a time sequence subinterval of the target time series based on the significant mutation point to obtain a plurality of subintervals, calculates a batch global median, and determines a mean lower threshold and a standard deviation upper threshold according to a process tolerance, determines the mean and the standard deviation of each time sequence subinterval one by one to identify a steady state period, constructs an individual-moving range control chart for a steady state period set, performs forward progressive scanning and reverse progressive scanning on the first and last steady state periods based on an adaptive control limit, and locates accurate time boundaries of the batch steady state period.

Description

Technical Field

[0001] This application relates to the field of tobacco technology, and in particular to a batch steady-state identification method and system for the leaf re-drying production process. Background Technology

[0002] Leaf threshing and re-drying is a crucial pre-treatment step in cigarette production, and its processing quality directly affects the stability of subsequent tobacco processing and cigarette rolling processes, as well as the sensory quality of the final product. In recent years, with the tobacco industry's deepening implementation of the concepts of "homogenized processing" and "refined manufacturing," data-driven analysis and optimization have become important means to improve technological levels. Currently, leaf threshing and re-drying production lines are generally equipped with advanced process control and production management systems, which can collect and store massive amounts of process data such as temperature, humidity, flow rate, and pressure in real time for key processes such as leaf moistening, threshing, and re-drying. To facilitate quality traceability and process evaluation, this data is usually organized and managed using "production batches" as the basic unit.

[0003] However, in actual production operations, due to the combined effects of factors such as periodic equipment start-ups and shutdowns, material switching, dynamic adjustments to process parameters, environmental temperature and humidity disturbances, and equipment response characteristics, the collected process data often exhibits significant fluctuations and non-stationary characteristics. Specifically, a complete production batch typically includes two distinct time periods, and the transition between these two periods is accompanied by significant changes in statistical characteristics: the first is the non-steady-state period (transition process), which covers stages such as batch start-up, material replacement, parameter adjustment, and fault recovery. During this period, process parameters are in dynamic adjustment or system response, and data fluctuates dramatically, making it difficult to characterize the preset stable process state. The second is the steady-state period (stationary process), which refers to the stage after the system reaches equilibrium, where variables fluctuate around the set value or mean only due to random noise interference. Data from this period best reflects the batch's process achievement level and equipment operating status.

[0004] Accurately identifying and extracting batch data during steady-state periods has dual significance for deepening production analysis and process optimization: From a data analysis perspective, steady-state data eliminates dynamic interference from transient processes, more accurately reflecting the inherent fluctuation patterns of process parameters and process capability. Based on this, the calculated process capability index (such as Cpk) and the constructed statistical process control model or quality prediction model are more representative and reliable. From a process evaluation perspective, removing the influence of non-steady-state data such as start-up and material change can effectively avoid misjudgments of batch process quality, making cross-batch comparisons more objective and accurate, and providing a consistent data benchmark for process parameter standardization and optimization evaluation.

[0005] The current practice of using fixed rules to extract data from the middle of a batch (assuming the middle period is the steady state) is difficult to adapt to the dynamic changes of different batches and processes. At the same time, this method is difficult to accurately delineate the transition boundary between the non-steady state and the steady state, which may result in the extracted data still containing interference from the transition process. Summary of the Invention

[0006] The main purpose of this application is to provide a batch steady-state identification method and system for the leaf re-drying production process, in order to solve the problem that the current practice of assuming the middle period as the steady state in the existing technology is difficult to adapt to the dynamic changes of different batches and different processes. At the same time, the method is difficult to accurately delineate the transition boundary between the non-steady state and the steady state, which may result in the interference of the transition process in the intercepted data.

[0007] To achieve the above objectives, this application provides the following technical solution: A batch steady-state identification method for the leaf re-drying production process, the batch steady-state identification method comprising: Step S1: Collect the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line; Step S2: Perform data preprocessing on the original time series and construct a complete and equally spaced target time series; Step S3: The significant mutation points of the target time series are calculated by traversing through the Binary Segmentation algorithm, and the target time series is recursively segmented until there are no significant mutation points, thus obtaining a set of mutation point locations. Step S4: The set of mutation point locations, together with the start and end points of the original time series, constitute a complete segmentation boundary to divide the target time series into several continuous and non-overlapping time series sub-intervals. Step S5: Calculate the mean and standard deviation of each time series sub-interval. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period. Step S6: Calculate the adaptive control limits by establishing an individual-moving range control chart for the sub-intervals marked as steady-state periods. Perform forward progressive scanning and backward progressive scanning on the sub-intervals of the first and last steady-state periods respectively. Use the rule that no more than one-third of the data points exceed the adaptive control limits as the fault tolerance rule to determine the precise start time and precise end time of each steady-state period.

[0008] Beneficial effects of steps S1 to S6: This method achieves automatic identification and precise location of steady-state periods in batches through full-process processing of batch data from the leaf re-drying production line. In the data acquisition phase, the raw time series of multiple parameters for the entire batch process is directly obtained through the Manufacturing Execution System, ensuring the integrity and traceability of the data source. In the data preprocessing phase, outlier replacement and missing value imputation eliminate acquisition noise, and equal-interval resampling unifies the time reference of each parameter, providing high-quality input data for subsequent analysis. In the mutation point detection phase, the Binary Segmentation algorithm adaptively locates significant structural change points in the time series using a recursive segmentation method, without the need for a pre-set segmentation window, effectively overcoming the insufficient adaptability of traditional fixed-window methods to batch length variations. In the sub-interval partitioning phase, the set of mutation point locations and the batch boundary together constitute a complete segmentation boundary, ensuring the continuity and completeness of the time series sub-intervals and ensuring no production period is missed. In the steady-state discrimination phase, the lower limit threshold of the mean is dynamically calculated based on the global median of the batch, and combined with the upper limit threshold of the standard deviation, a dual constraint judgment is applied to each sub-interval, avoiding the insufficient adaptability of fixed thresholds to differences in operating conditions of different batches. In the precise boundary localization stage, the individual-moving range control chart provides adaptive control limits. The forward and backward progressive scanning combined with the one-third fault tolerance rule ensures the robustness of steady-state period identification while accurately locating the start and end times of each steady-state period to the sampling point level.

[0009] In this process, step S1 ensures the integrity and consistency of the original time series with multiple parameters by associating the batch identifier of the manufacturing execution system with the entire process data; step S2 eliminates acquisition noise by replacing outliers and filling missing values, and unifies the time reference by resampling at equal intervals; step S3 achieves adaptive localization of changes in the internal structure of the batch by recursively detecting significant mutation points using the Binary Segmentation algorithm; step S4 ensures the continuity and completeness of the time series sub-interval division by using the set of mutation point locations and the batch boundary together to form a complete segmentation boundary; step S5 dynamically calculates the discrimination threshold based on the global median of the batch, and achieves adaptive identification of steady-state periods through dual constraints of mean and standard deviation; step S6 calculates adaptive control limits by using individual-moving range control charts, and accurately locates the start and end times of each steady-state period to the sampling point level by combining forward and backward progressive scanning.

[0010] As a further improvement to this application, step S1 involves collecting the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line, including: Step S1.1: Obtain the raw data of each process node of the leaf-re-drying production line from the manufacturing execution system of the leaf-re-drying production line; Step S1.2: Based on the batch identifier on the leaf re-drying production line, filter all the original data to find the production data belonging to the same complete batch. The batch identifier is used to associate the entire production data of a batch on the leaf re-drying production line from feeding to finished product warehousing. Step S1.3: Determine the types of key process parameters to be collected based on the preset process parameter list; Step S1.4: Extract the sampled values ​​corresponding to each key process parameter type from the selected production data of the same batch in order of timestamps to construct the original time series of each key process parameter; Step S1.5: Associate and store the original time series of each key process parameter with the corresponding process parameter type identifier to obtain the original time series data of multiple key process parameters for a complete batch on the leaf re-drying production line.

[0011] Beneficial effects of steps S1.1 to S1.5: This series of steps uses batch identifiers from the Manufacturing Execution System (MES) to link the entire process data from material feeding to finished product warehousing, constructing a complete and traceable multi-parameter raw time series data foundation. This provides the correct data range and complete batch context for subsequent preprocessing, mutation point detection, and steady-state identification.

[0012] Step S1.1 involves acquiring raw data from each process node through the manufacturing execution system of the leaf re-drying production line. This system, as the data source of the production line, provides a unified source of production data, ensuring internal consistency between data timestamps, process nodes, and sampled values. Step S1.2 involves filtering the entire process data of the same complete batch based on batch identifiers, achieving full-process data association coverage from material input to finished product warehousing, ensuring that subsequent analysis is conducted on a single continuous production batch and avoiding cross-batch data confusion. Step S1.3 involves determining the types of key process parameters to be collected based on a preset list of process parameters, capturing core parameters directly related to product quality and process stability, avoiding irrelevant noise introduced by full data collection, and establishing appropriate data dimensions for subsequent analysis. Step S1.4 involves extracting the sampled values ​​corresponding to each key process parameter type in the order of timestamps to construct the original time series, ensuring the temporal correctness and data order integrity of each parameter time series. Step S1.5 involves associating and storing the original time series of each key process parameter with the corresponding process parameter type identifier, ensuring that parameter type confusion does not occur during subsequent transmission and processing, and maintaining data traceability.

[0013] As a further improvement to this application, step S2 involves preprocessing the original time series and constructing a complete and equally spaced target time series, including: Step S2.1: Identify and mark outliers point by point in the original time series of each key process parameter; Step S2.2: Replace the sampling points marked as outliers. When there is temporal continuity between the outlier and the adjacent normal sampling points, the replacement value is calculated using the adjacent normal sampling values ​​through linear interpolation. When the outlier is an isolated sampling point, the average of the adjacent normal sampling values ​​is used as the replacement value. Step S2.3: Detect missing values ​​in the original time series of each key process parameter. If the actual number of samples is less than the theoretical number of samples in the time series, fill the missing position by linear interpolation of the adjacent normal sample values. Step S2.4: After filling and repairing, the time series of each key process parameter is resampled at different sampling intervals. Linear interpolation is used to unify each time series to a preset equal-interval sampling frequency to obtain a complete target time series with equal-interval sampling.

[0014] Beneficial effects of steps S2.1 to S2.4: This series of steps, through the coordinated processing of three stages—outlier identification and replacement, missing value detection and imputation, and equal-interval resampling—transforms the collected raw time series into a complete and equally spaced target time series, providing high-quality data input for mutation point detection.

[0015] In step S2.1, outlier identification and marking are performed point by point. This locates all potential outliers while preserving the integrity of the original data, providing accurate processing targets for subsequent replacement and avoiding interference from outliers in subsequent statistical calculations. In step S2.2, for sampling points marked as outliers, linear interpolation or mean replacement is selected based on their temporal continuity. For time-continuous outliers, linear interpolation ensures a smooth transition between the replacement value and adjacent data, while mean replacement reduces the interference of random fluctuations on the replacement results for isolated outliers. In step S2.3, missing value detection is performed on the original time series of each key process parameter, distinguishing between missing values ​​and outliers to prevent data loss caused by incorrect processing of missing values ​​by the outlier handling module. In step S2.4, the time series of each key process parameter after filling and repair are resampled at equal intervals. The sampling frequency of different parameters is unified to the preset equal interval sampling frequency, eliminating the time axis misalignment problem caused by inconsistent sampling frequencies and creating the preconditions for subsequent joint analysis of various parameters.

[0016] As a further improvement to this application, step S3 involves using the Binary Segmentation algorithm to traverse and calculate the significant abrupt change points of the target time series, and recursively segmenting the target time series until no significant abrupt change points remain, thereby obtaining a set of abrupt change point locations, including: Step S3.1: Quantify the degree of fitting improvement in dividing the target time series into left and right sub-segments at the candidate split points by the log-likelihood ratio statistic. Calculate the log-likelihood ratio statistic point by point for all candidate split points on the target time series that satisfy the minimum split interval. Step S3.2: Traverse all candidate split points to locate the candidate split point that maximizes the log-likelihood ratio statistic as the optimal split point for the current interval. Step S3.3: Define a significance threshold based on a chi-square distribution with 2 degrees of freedom, and set the value of the significance threshold based on a preset confidence level; Step S3.4: Compare the maximum log-likelihood ratio statistic corresponding to the current optimal segmentation point with the significance threshold. If the current maximum statistic exceeds the significance threshold, then determine the current optimal segmentation point as a significant mutation point and add the corresponding position to the mutation point position set. Step S3.5: Repeat the traversal calculation and significance determination of steps S3.1 to S3.4 for the left and right sub-segments generated by each segmentation. Stop the recursive segmentation if there are no candidate segmentation points exceeding the significance threshold in a certain sub-segment, and output a set of mutation point locations including all significant mutation point locations.

[0017] Beneficial effects of steps S3.1 to S3.5: This series of steps uses the Binary Segmentation algorithm to recursively calculate and determine the significance of the log-likelihood ratio statistic, enabling adaptive localization of points of structural change within a time series without the need for a pre-set segmentation window. This effectively overcomes the problem of insufficient adaptability of traditional fixed-window methods to changes in batch length.

[0018] In this process, step S3.1 quantifies the improvement in fit of candidate segmentation points using the log-likelihood ratio statistic, transforming segmentation quality into a quantifiable statistical indicator and ensuring a unified standard for evaluating all candidate segmentation points. Step S3.2 iterates through all candidate segmentation points to locate the optimal segmentation point that maximizes the log-likelihood ratio statistic, ensuring that each segmentation step selects the segmentation position with the highest improvement in fit within the current interval using the statistic maximization criterion. Step S3.3 defines a significance threshold based on a chi-square distribution with 2 degrees of freedom, setting the threshold value based on a preset confidence level, transforming the criteria for determining abrupt changes from a qualitative description into a quantifiable and verifiable statistical criterion. Step S3.4 compares the maximum statistic corresponding to the optimal segmentation point with the significance threshold and outputs significant abrupt changes, ensuring that only segmentation points with statistical significance are included in the abrupt change location set, avoiding false segmentations caused by sampling noise. Step S3.5 recursively performs traversal calculations and significance determination on the left and right sub-segments generated by the segmentation, achieving complete detection of all significant structural change points within the time series without the need for a preset segmentation window.

[0019] As a further improvement of this application, step S4 involves using the set of mutation point locations together with the start and end points of the original time series to form a complete segmentation boundary, thereby dividing the target time series into several continuous and non-overlapping time series sub-intervals, including: Step S4.1: Obtain the starting sampling time of the target time series as the starting time point and the ending sampling time of the target time series as the ending time point; Step S4.2: Combine the sampling times corresponding to all mutation points in the mutation point location set with the start time point and the end time point to obtain a boundary time set including all segmentation boundaries; Step S4.3: Sort the set of boundary moments in ascending order of time, remove duplicate boundary moments, and obtain an ordered and non-repeating complete set of segmentation boundaries. Step S4.4: Based on the time interval formed by two adjacent boundary moments in the complete segmentation boundary set, the target time series is segmented and divided into several continuous and non-overlapping time series sub-intervals. Step S4.5: Record the start and end times and duration of each time series sub-interval to obtain the set of time series sub-intervals after division.

[0020] Beneficial effects of steps S4.1 to S4.5: This series of steps uses the set of mutation point locations and batch boundaries to form a complete set of segmentation boundaries, which divides the target time series into continuous and non-overlapping intervals, providing a structured data segmentation basis for subsequent steady-state period identification.

[0021] In this process, step S4.1 obtains the start and end sampling times of the target time series as batch boundaries, clarifying the complete time span for subsequent partitioning and avoiding the problem of missing non-mutation boundaries at the beginning and end of the batch in the mutation point set; step S4.2 merges the sampling times corresponding to mutation points with the start and end time points, ensuring that the partition boundary set covers the entire time span of the batch, laying the foundation for complete partitioning; step S4.3 sorts the boundary time set in ascending order and removes duplicate values, obtaining an ordered and non-repeating complete partition boundary set, avoiding the problem of multiple partitions at the same time due to boundary repetition; step S4.4 segments the target time series according to the time interval formed by adjacent boundary times, ensuring that the data association between adjacent sub-intervals is severed at the partition boundaries, making the statistical feature calculations of each sub-interval independent; step S4.5 records the start and end times and duration of each time series sub-interval and outputs the partitioned sub-interval set, enabling the subsequent steady-state discrimination module to directly obtain the structured information of each sub-interval without repeating boundary calculations.

[0022] As a further improvement to this application, in step S5, the mean and standard deviation of each time series sub-interval are calculated. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period, including: Step S5.1: Obtain the data points of all time-series sub-intervals in the time-series sub-interval set, and calculate the median of all data points as the median of all data in the batch; Step S5.2: Calculate the lower limit threshold of the mean based on the preset process tolerance value. The lower limit threshold of the mean is equal to the median of all data in the batch minus the preset process tolerance value. Step S5.3: Calculate the upper limit threshold of the standard deviation based on the preset process tolerance value. The upper limit threshold of the standard deviation is equal to the preset process tolerance value divided by 3. Step S5.4: Calculate the mean and standard deviation of each time series sub-interval one by one. If the mean of the current time series sub-interval is not less than the lower limit threshold of the mean and the standard deviation of the current time series sub-interval does not exceed the upper limit threshold of the standard deviation, then the current time series sub-interval is determined to be a steady state period; otherwise, it is determined to be a non-steady state period. Step S5.5: Summarize the judgment results of all time series sub-intervals to obtain the start and end times, mean, standard deviation, and steady-state labeling status of each time series sub-interval.

[0023] Beneficial effects of steps S5.1 to S5.5: This series of steps dynamically calculates the discrimination threshold based on the global median of the batch, and automatically marks each time series sub-interval as steady-state and non-steady-state through the dual constraints of mean and standard deviation, thereby realizing adaptive steady-state identification of differences in operating conditions of different batches.

[0024] Specifically, step S5.1 acquires data points from all time-series sub-intervals and calculates the median of all data in the batch, using the batch's global statistics as the criterion to avoid misjudgments caused by fluctuations in statistics within a single sub-interval; step S5.2 calculates the lower limit threshold of the mean based on the preset process tolerance value, using the batch median minus the process tolerance value as the lower bound for mean discrimination, ensuring that the steady-state discrimination standard is compatible with the actual process level of the batch; step S5.3 calculates the upper limit threshold of the standard deviation based on the preset process tolerance value, using the process tolerance value divided by 3 as the upper bound for standard deviation discrimination, using the 3σ principle to ensure that the fluctuation range of the steady-state period does not exceed the allowable dispersion of the process; step S5.4 makes a judgment through the dual constraints of the mean not being lower than the lower limit threshold and the standard deviation not exceeding the upper limit threshold, controlling production deviation while constraining the degree of production dispersion, avoiding misjudgments caused by a single indicator; step S5.5 summarizes the judgment results of all time-series sub-intervals, outputting the start and end times, mean, standard deviation, and steady-state marker status of each sub-interval, providing structured input data for the subsequent precise boundary positioning module.

[0025] As a further improvement of this application, step S6 involves establishing an individual-moving range control chart for the sub-intervals marked as steady-state periods to calculate adaptive control limits, and performing forward and backward progressive scans on the sub-intervals of the first and last steady-state periods respectively, using the rule that no more than one-third of the data points exceed the adaptive control limits as a fault tolerance rule, to determine the precise start and precise end times of each steady-state period, including: Step S6.1: Calculate the moving range between adjacent sampling points in the time series sub-interval of the steady-state period, and merge the individual value sequence of each steady-state sub-interval with the corresponding moving range sequence to obtain the individual-moving range data of each steady-state sub-interval; Step S6.2: Summarize the individual values ​​of all steady-state sub-intervals and calculate the overall mean to serve as the center line of the individual control chart; Summarize the moving range of all steady-state sub-intervals and calculate the overall mean to serve as the center line of the moving range control chart. Step S6.3: Calculate the adaptive upper control limit and adaptive lower control limit of the individual value based on the center line of the individual control chart, the center line of the moving range control chart, and the 3σ principle; Step S6.4: Perform a forward progressive scan from the starting time point of the first steady-state sub-interval to the ending time point, sequentially expanding the range of data points participating in the statistical calculation, and determining whether the newly added data points exceed the adaptive upper control limit and the adaptive lower control limit of the individual value. Step S6.5: If the proportion of data points exceeding the control limit to the total number of data points in the current time period exceeds one-third, update the earliest occurrence time of the data points exceeding the control limit to the precise start time of the current steady-state time period. Step S6.6: Perform a reverse progressive scan from the end time point of the last steady-state sub-interval towards the start time point, sequentially expanding the range of data points participating in the statistical calculation, and determining whether the newly added data points exceed the individual value adaptive upper control limit and the individual value adaptive lower control limit; Step S6.7: If the proportion of data points exceeding the control limit to the total number of data points in the current time period exceeds one-third, update the latest occurrence time of the data points exceeding the control limit to the precise termination time of the current steady-state time period. Step S6.8: Repeat steps S6.4 to S6.7 to scan all sub-intervals marked as steady-state periods. Based on the scanning results, adjust the precise start time and precise end time of each steady-state period to obtain the precise time boundary of each steady-state period.

[0026] Beneficial effects of steps S6.1 to S6.8: This series of steps provides adaptive control limits through individual-moving range control charts. Combined with forward and backward progressive scanning and one-third tolerance rules, the start and end times of each steady-state period are precisely located at the sampling point level, achieving accurate time boundary output while ensuring robust recognition.

[0027] In step S6.1, the moving range between adjacent sampling points during the steady-state period is calculated and merged with the individual value sequence to construct a complete individual-moving range dataset, providing the necessary data structure for subsequent control limit calculation. Step S6.2 summarizes the individual values ​​and moving ranges of all steady-state sub-intervals and calculates the overall mean, using the overall data statistics of the steady-state period as the control chart centerline, ensuring that the control limits represent the true level of variation in the steady-state process. Step S6.3 calculates the adaptive upper control limit and adaptive lower control limit based on the control chart centerline and the 3σ principle, allowing the control limits to be dynamically adjusted according to the actual data situation without the need for manually preset fixed thresholds. Step S6.4 performs a forward progressive scan on the first steady-state sub-interval, sequentially expanding the data point range and determining whether newly added data points exceed the control limits, achieving fine-grained analysis from the inside of the steady-state period towards the boundary. Step S6.5 updates the precise start time of the first steady-state period using the rule that one-third of the data points exceed the control limit, avoiding boundary misjudgment caused by random fluctuations of a few edge data points. Step S6.6 performs a reverse progressive scan on the sub-interval of the last steady-state period, expanding the data range sequentially from the end time point to the start time point and judging whether the newly added data points exceed the control limit, forming a symmetrical boundary search mechanism with the forward scan. Step S6.7 updates the precise end time of the last steady-state period using the rule that one-third of the data points exceed the control limit, ensuring that the boundary judgment criteria are consistent in the symmetrical direction. Step S6.8 repeats the forward and reverse scans on all steady-state period sub-intervals, adjusts the precise start and end times of each steady-state period based on the scan results, and outputs the complete precise time boundaries of each steady-state period.

[0028] To achieve the above objectives, this application also provides the following technical solutions: A batch steady-state identification system for the leaf re-drying production process, wherein the batch steady-state identification system is applied to the batch steady-state identification method described above, and the batch steady-state identification system includes: The original time series acquisition module for leaf re-drying is used to collect the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line. The target time series acquisition module is used to preprocess the original time series and construct a complete and equally spaced target time series. The mutation point location set acquisition module is used to traverse and calculate the significant mutation points of the target time series through the Binary Segmentation algorithm, and recursively segment the target time series until there are no significant mutation points, thereby obtaining the mutation point location set. The mutation point location set partitioning module is used to form a complete partitioning boundary by combining the mutation point location set with the start and end points of the original time series, so as to divide the target time series into several continuous and non-overlapping time series sub-intervals. The steady-state marking module for time-series sub-intervals is used to calculate the mean and standard deviation of each time-series sub-interval. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period. The precise start and end time determination module is used to calculate the adaptive control limits by establishing an individual-moving range control chart for sub-intervals marked as steady-state periods, and to perform forward progressive scanning and backward progressive scanning on the sub-intervals of the first and last steady-state periods respectively, with the fault tolerance rule being that no more than one-third of the data points exceed the adaptive control limits, to determine the precise start and precise end times of each steady-state period.

[0029] To achieve the above objectives, this application also provides the following technical solutions: An electronic device includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the batch steady-state identification method for the leaf re-drying production process as described above.

[0030] To achieve the above objectives, this application also provides the following technical solutions: A computer-readable storage medium storing program instructions, which, when executed by a processor, enable the batch steady-state identification method for the leaf re-drying production process described above. Attached Figure Description

[0031] Figure 1 This is a schematic flowchart illustrating the steps of an embodiment of the batch steady-state identification method in the leaf re-drying production process of this application; Figure 2 This is a schematic diagram of the functional modules of a batch steady-state identification system for the leaf re-drying production process according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the storage medium of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0033] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (e.g., as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0034] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0035] It should be noted that, due to the limited types and number of symbols or letters that can represent specific meanings, for embodiments with many formulas or codes, there may be situations where symbols or letters cannot meet the usage requirements. Therefore, the interpretation of formula symbols in the steps or sub-steps of the embodiments is only valid for the current step or sub-step.

[0036] If the same symbol has different interpretations in different steps or sub-steps, the interpretation in the current step or sub-step shall prevail; if the same symbol appears in different steps or sub-steps, but no interpretation is given in subsequent steps or sub-steps after its first appearance, the interpretation in the first step or sub-step shall be used.

[0037] For example, "i=1,2,...,n" is a writing convention and a well-known meaning. If "i=1,2,...,n" has already appeared once in an embodiment and needs to be used again in formulas with different scenarios and meanings in the future, then the meaning of "i=1,2,...,n" should be interpreted according to the corresponding formulas. If "i=1,2,...,n" is not used for the first time and is changed to a symbol or letter with a non-well-known meaning, such as "p=1,2,...,q", to avoid repetition, it is more likely to cause confusion, ambiguity, and unclear problems.

[0038] like Figure 1 As shown, this embodiment provides an example of a batch steady-state identification method for the leaf re-drying production process. In this embodiment, the batch steady-state identification method includes the following steps: Step S1: Collect the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line.

[0039] Furthermore, step S1 specifically includes the following steps: Step S1.1: Obtain the raw data of each process node of the leaf-drying production line from the manufacturing execution system of the leaf-drying production line.

[0040] Preferably, the leaf-picking and re-drying production line is equipped with a Manufacturing Execution System (MES) to record production data for each process node in real time. Each record in the MES database contains three fields: timestamp, process node identifier, and sampled value of process parameters. When retrieving raw data, the data record of the most recent complete batch is queried in reverse chronological order to confirm that the batch contains a complete work order record from the first leaf-picking process of feeding materials to the final process of drying leaf fibers and packaging the finished product.

[0041] Step S1.2: Based on the batch identifier on the leaf re-drying production line, filter all raw data belonging to the same complete batch. The batch identifier is used to associate the entire production data of a batch on the leaf re-drying production line from material input to finished product warehousing.

[0042] Preferably, the leaf re-drying production line uses batches as the basic production unit. Each batch is assigned a unique batch identifier, which is associated with the entire process of work order data from material input to finished product warehousing. Using the batch identifier as the primary key, the production data records of all workstation nodes under that batch are queried from the Manufacturing Execution System and stored in ascending order of timestamp.

[0043] Step S1.3: Determine the types of key process parameters to be collected based on the preset process parameter list.

[0044] Preferably, the key process parameters that have the greatest impact on product quality and process stability during the leaf-picking and re-drying process are: the cylinder wall temperature during the rehydration stage of the leaf-picking process, the shred width during the leaf-shredding process, the drying zone temperature during the leaf-shredding process, and the cold end temperature during the leaf-shredding process. A predefined list of process parameter types is used as the basis for data collection, and the sampled values ​​corresponding to the above types are selected from the raw data.

[0045] Preferably, depending on additional process control requirements, the parameters that can be collected can be expanded to include moisture content, feed liquid flow rate, and humidity level in the drying zone.

[0046] Step S1.4: Extract the sampled values ​​corresponding to each key process parameter type from the selected production data of the same batch in order of timestamp to construct the original time series of each key process parameter.

[0047] Preferably, for the selected production data of the same batch, the data records of each process parameter type are traversed in ascending order of time, with the timestamp as the primary index, and the sampled values ​​of each parameter type are extracted sequentially to construct an independent time series for each process parameter type. Each data point in the time series contains the sampling time and the corresponding sampled value.

[0048] Step S1.5: Associate and store the original time series of each key process parameter with the corresponding process parameter type identifier to obtain the original time series data of multiple key process parameters for a complete batch on the leaf re-drying production line.

[0049] Preferably, the original time series of each process parameter is paired and stored with the corresponding process parameter type identifier. The storage structure is in key-value pair form, where the key is the process parameter type identifier and the value is the timestamp-sample value sequence set corresponding to that parameter type, thus obtaining the original time series data of multiple key process parameters for a complete batch on the leaf re-drying production line.

[0050] Beneficial effects of steps S1.1 to S1.5: This series of steps uses batch identifiers from the Manufacturing Execution System (MES) to link the entire process data from material feeding to finished product warehousing, constructing a complete and traceable multi-parameter raw time series data foundation. This provides the correct data range and complete batch context for subsequent preprocessing, mutation point detection, and steady-state identification.

[0051] Step S1.1 involves acquiring raw data from each process node through the manufacturing execution system of the leaf re-drying production line. This system, as the data source of the production line, provides a unified source of production data, ensuring internal consistency between data timestamps, process nodes, and sampled values. Step S1.2 involves filtering the entire process data of the same complete batch based on batch identifiers, achieving full-process data association coverage from material input to finished product warehousing, ensuring that subsequent analysis is conducted on a single continuous production batch and avoiding cross-batch data confusion. Step S1.3 involves determining the types of key process parameters to be collected based on a preset list of process parameters, capturing core parameters directly related to product quality and process stability, avoiding irrelevant noise introduced by full data collection, and establishing appropriate data dimensions for subsequent analysis. Step S1.4 involves extracting the sampled values ​​corresponding to each key process parameter type in the order of timestamps to construct the original time series, ensuring the temporal correctness and data order integrity of each parameter time series. Step S1.5 involves associating and storing the original time series of each key process parameter with the corresponding process parameter type identifier, ensuring that parameter type confusion does not occur during subsequent transmission and processing, and maintaining data traceability.

[0052] Step S2: Perform data preprocessing on the original time series and construct a complete and equally spaced target time series.

[0053] Furthermore, step S2 specifically includes the following steps: Step S2.1: Identify and mark outliers point by point in the original time series of each key process parameter.

[0054] Preferably, outlier identification is performed point-by-point on the original time series of each key process parameter. The outlier detection window is set as a neighborhood window containing N sampling points before and after it. In this embodiment, N is 15, meaning each neighborhood window covers 15 sampling points before and after it, containing a total of 31 data points (including the center point). The local mean and local standard deviation within the neighborhood window are calculated. When a sampled value deviates from the local mean by more than three times the local standard deviation, the sampled point is determined to be an outlier and marked.

[0055] Preferably, the size of the neighborhood window is set according to the sampling period. When the sampling period is 10 seconds, the neighborhood window covers the production data for 2.5 minutes before and after the sampling period, corresponding to 30 sampling points, and N is 15.

[0056] Preferably, outlier detection can also use the Grubbs criterion or the Dixon criterion instead of the 3σ criterion. The Grubbs criterion calculates the critical value based on the sequence mean and standard deviation, and is suitable for single-point outlier detection under the assumption of normal distribution.

[0057] Step S2.2: Replace the sampling points marked as outliers. When there is temporal continuity between the outlier and the adjacent normal sampling points, the replacement value is calculated using the adjacent normal sampling values ​​before and after by linear interpolation. When the outlier is an isolated sampling point, the average of the adjacent normal sampling values ​​before and after is used as the replacement value.

[0058] Preferably, the temporal continuity of each sampling point marked as an outlier is determined. All sampling points marked as outliers are traversed; if at least one sampling point before and after the outlier is not marked as an outlier, then the outlier is determined to have temporal continuity with the adjacent normal sampling point, and a replacement value is calculated using linear interpolation.

[0059] The linear interpolation replacement formula is as follows: .

[0060] in, and These are the sampled values ​​of the adjacent normal sampling points before and after the anomaly point. , , For the corresponding sampling time, This is the replacement value.

[0061] Preferably, if there are sample points marked as anomalous among the adjacent sample points before and after the outlier (i.e., there are no normal sample points before and after the outlier), then the outlier is determined to be an isolated outlier, and the mean of the adjacent normal sample values ​​is used to replace it. .

[0062] Step S2.3: Perform missing value detection on the original time series of each key process parameter. If the actual number of samples is less than the theoretical number of samples in the time series, then fill the missing position with linear interpolation of the adjacent normal sample values.

[0063] Preferably, missing value detection is performed on the time series of each key process parameter after filling and repair. The theoretical number of samples covering the time span and sampling period of the time series is calculated, and the theoretical number of samples is compared with the actual number of data points received. If the actual number of samples is less than the theoretical number of samples, the missing positions are marked.

[0064] Step S2.4: After filling and repairing, the time series of each key process parameter is resampled at different sampling intervals. Linear interpolation is used to unify each time series to a preset equal-interval sampling frequency to obtain a complete target time series with equal-interval sampling.

[0065] Preferably, the time series after filling and repair is subjected to sampling interval resampling processing. A preset standard sampling period is used as the unified sampling interval. In this embodiment, the standard sampling period is set to 10 seconds, that is, the standard sampling frequency is 0.1 Hz. Linear interpolation is used to unify each time series to an equal sampling frequency of 10 seconds. The positions where missing values ​​have been filled are also included in the interpolation calculation to obtain a complete and equally spaced target time series.

[0066] Beneficial effects of steps S2.1 to S2.4: This series of steps, through the coordinated processing of three stages—outlier identification and replacement, missing value detection and imputation, and equal-interval resampling—transforms the collected raw time series into a complete and equally spaced target time series, providing high-quality data input for mutation point detection.

[0067] In step S2.1, outlier identification and marking are performed point by point. This locates all potential outliers while preserving the integrity of the original data, providing accurate processing targets for subsequent replacement and avoiding interference from outliers in subsequent statistical calculations. In step S2.2, for sampling points marked as outliers, linear interpolation or mean replacement is selected based on their temporal continuity. For time-continuous outliers, linear interpolation ensures a smooth transition between the replacement value and adjacent data, while mean replacement reduces the interference of random fluctuations on the replacement results for isolated outliers. In step S2.3, missing value detection is performed on the original time series of each key process parameter, distinguishing between missing values ​​and outliers to prevent data loss caused by incorrect processing of missing values ​​by the outlier handling module. In step S2.4, the time series of each key process parameter after filling and repair are resampled at equal intervals. The sampling frequency of different parameters is unified to the preset equal interval sampling frequency, eliminating the time axis misalignment problem caused by inconsistent sampling frequencies and creating the preconditions for subsequent joint analysis of various parameters.

[0068] Step S3: The significant mutation points of the target time series are calculated by traversing through the Binary Segmentation algorithm, and the target time series is recursively segmented until there are no significant mutation points, thus obtaining the set of mutation point locations.

[0069] Furthermore, step S3 specifically includes the following steps: Step S3.1: Quantify the degree of fitting improvement of dividing the target time series into left and right sub-segments at the candidate split points by the log-likelihood ratio statistic. Calculate the log-likelihood ratio statistic for each candidate split point on the target time series that satisfies the minimum split interval.

[0070] Preferably, the log-likelihood ratio statistic is calculated point-by-point for all candidate segmentation points on the target time series that satisfy the minimum segmentation interval. The minimum segmentation interval is defined as containing at least 60 sampling points, i.e., the duration of the minimum segmentation interval is 600 seconds.

[0071] Among them, candidate segmentation points The formula for calculating the log-likelihood ratio statistic is: .

[0072] Where n is the total number of sampling points in the current interval. and The number of sampling points in the left and right sub-intervals of the dividing point. This is the variance estimate for the current interval before it is divided. and Variance estimation for the left and right sub-intervals after segmentation: .

[0073] in, These are the sample means of the current interval and its left and right sub-intervals, respectively.

[0074] Step S3.2: Traverse all candidate split points to locate the candidate split point that maximizes the log-likelihood ratio statistic as the optimal split point for the current interval.

[0075] Preferably, all candidate split points within the current interval are traversed, and the log-likelihood ratio statistic for each candidate split point is calculated. , take The candidate segment point that reaches the maximum value is taken as the optimal segment point of the current interval, and this maximum value is recorded.

[0076] Step S3.3: Define a significance threshold based on a chi-square distribution with 2 degrees of freedom, and set the value of the significance threshold based on a preset confidence level.

[0077] Preferably, the significance threshold is determined based on a chi-square distribution with 2 degrees of freedom. A pre-set confidence level of 95% is used, and the critical value corresponding to the 95% confidence level with 2 degrees of freedom is obtained from the chi-square distribution table as 5.99. This value is then used as the significance threshold. .

[0078] Step S3.4: Compare the maximum log-likelihood ratio statistic corresponding to the current optimal segmentation point with the significance threshold. If the current maximum statistic exceeds the significance threshold, the current optimal segmentation point is determined to be a significant mutation point and the corresponding position is added to the mutation point position set.

[0079] Preferably, the maximum log-likelihood ratio statistic corresponding to the optimal segmentation point of the current interval is used. Significance threshold Compare. If If the optimal split point is determined to be a significant mutation point, its position is added to the mutation point position set; if If there is no significant abrupt change in the current interval, then the segmentation of the interval is stopped.

[0080] Step S3.5: Repeat the traversal calculation and significance determination of steps S3.1 to S3.4 for the left and right sub-segments generated by each segmentation. Stop the recursive segmentation if there are no candidate segmentation points exceeding the significance threshold in a certain sub-segment, and output a set of mutation point locations including all significant mutation point locations.

[0081] Preferably, steps S3.1 to S3.4 are repeated for the left and right sub-segments generated by each segmentation. When no candidate segment exceeds the saliency threshold in a sub-segment, the recursive segmentation of that sub-segment is stopped. The final output is a set of mutation point locations containing all significant mutation point locations.

[0082] Beneficial effects of steps S3.1 to S3.5: This series of steps uses the Binary Segmentation algorithm to recursively calculate and determine the significance of the log-likelihood ratio statistic, enabling adaptive localization of points of structural change within a time series without the need for a pre-set segmentation window. This effectively overcomes the problem of insufficient adaptability of traditional fixed-window methods to changes in batch length.

[0083] In this process, step S3.1 quantifies the improvement in fit of candidate segmentation points using the log-likelihood ratio statistic, transforming segmentation quality into a quantifiable statistical indicator and ensuring a unified standard for evaluating all candidate segmentation points. Step S3.2 iterates through all candidate segmentation points to locate the optimal segmentation point that maximizes the log-likelihood ratio statistic, ensuring that each segmentation step selects the segmentation position with the highest improvement in fit within the current interval using the statistic maximization criterion. Step S3.3 defines a significance threshold based on a chi-square distribution with 2 degrees of freedom, setting the threshold value based on a preset confidence level, transforming the criteria for determining abrupt changes from a qualitative description into a quantifiable and verifiable statistical criterion. Step S3.4 compares the maximum statistic corresponding to the optimal segmentation point with the significance threshold and outputs significant abrupt changes, ensuring that only segmentation points with statistical significance are included in the abrupt change location set, avoiding false segmentations caused by sampling noise. Step S3.5 recursively performs traversal calculations and significance determination on the left and right sub-segments generated by the segmentation, achieving complete detection of all significant structural change points within the time series without the need for a preset segmentation window.

[0084] Step S4: Combine the set of mutation point locations with the start and end points of the original time series to form a complete segmentation boundary, so as to divide the target time series into several continuous and non-overlapping time series sub-intervals.

[0085] Furthermore, step S4 specifically includes the following steps: Step S4.1: Obtain the starting sampling time of the target time series as the starting time point and the ending sampling time of the target time series as the ending time point.

[0086] Preferably, the first sampling moment of the target time series is taken as the starting time point, and the last sampling moment of the target time series is taken as the ending time point. The starting time point and the ending time point together constitute the two endpoints of the batch time span.

[0087] Step S4.2: Combine the sampling times corresponding to all mutation points in the mutation point location set with the start time point and the end time point to obtain the boundary time set including all segmentation boundaries.

[0088] Preferably, the sampling times corresponding to all mutation points in the mutation point location set are merged with the start and end time points to obtain a boundary time set containing all segmentation boundaries. Each time in the boundary time set corresponds to a segmentation boundary location on the target time series.

[0089] Step S4.3: Sort the set of boundary moments in ascending order of time, remove duplicate boundary moments, and obtain an ordered and non-repeating complete set of segmentation boundaries.

[0090] Preferably, the set of boundary moments is arranged in ascending chronological order, and duplicate moments caused by abrupt changes occurring at the start or end time points are removed, resulting in an ordered and non-repeating complete set of segmentation boundaries. Each pair of adjacent boundary moments in the complete set of segmentation boundaries constitutes a sub-interval.

[0091] Step S4.4: Based on the time interval formed by two adjacent boundary moments in the complete segmentation boundary set, the target time series is segmented and divided into several continuous and non-overlapping time series sub-intervals.

[0092] Preferably, the target time series is segmented based on the time interval formed by two adjacent boundary moments in the complete set of segmentation boundaries. Each time interval corresponds to a continuous time series sub-interval, and all time series sub-intervals cover the complete time span of the target time series and do not overlap with each other.

[0093] Step S4.5: Record the start and end times and duration of each time series sub-interval to obtain the set of time series sub-intervals after division.

[0094] Preferably, the start time, end time, and duration of each time series sub-interval are recorded, and the resulting set of time series sub-intervals is output. Each time series sub-interval includes its start and end times, duration, and corresponding original sampled data.

[0095] Beneficial effects of steps S4.1 to S4.5: This series of steps uses the set of mutation point locations and batch boundaries to form a complete set of segmentation boundaries, which divides the target time series into continuous and non-overlapping intervals, providing a structured data segmentation basis for subsequent steady-state period identification.

[0096] In this process, step S4.1 obtains the start and end sampling times of the target time series as batch boundaries, clarifying the complete time span for subsequent partitioning and avoiding the problem of missing non-mutation boundaries at the beginning and end of the batch in the mutation point set; step S4.2 merges the sampling times corresponding to mutation points with the start and end time points, ensuring that the partition boundary set covers the entire time span of the batch, laying the foundation for complete partitioning; step S4.3 sorts the boundary time set in ascending order and removes duplicate values, obtaining an ordered and non-repeating complete partition boundary set, avoiding the problem of multiple partitions at the same time due to boundary repetition; step S4.4 segments the target time series according to the time interval formed by adjacent boundary times, ensuring that the data association between adjacent sub-intervals is severed at the partition boundaries, making the statistical feature calculations of each sub-interval independent; step S4.5 records the start and end times and duration of each time series sub-interval and outputs the partitioned sub-interval set, enabling the subsequent steady-state discrimination module to directly obtain the structured information of each sub-interval without repeating boundary calculations.

[0097] Step S5: Calculate the mean and standard deviation of each time series sub-interval. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period.

[0098] Furthermore, step S5 specifically includes the following steps: Step S5.1: Obtain the data points of all time series sub-intervals in the time series sub-interval set, and calculate the median of all data points as the median of all data in the batch.

[0099] Preferably, the original sampled data of all divided time-series sub-intervals are obtained, and the data points of all time-series sub-intervals are merged into a complete dataset. The median of all sampled values ​​in this complete dataset is then calculated as the median of all data in the batch. The median, as a global location statistic for the batch, is insensitive to outliers and has good robustness as a benchmark for steady-state discrimination.

[0100] Step S5.2: Calculate the lower limit threshold of the mean based on the preset process tolerance value. The lower limit threshold of the mean is equal to the median of all data in the batch minus the preset process tolerance value.

[0101] Preferably, the lower limit threshold of the mean is calculated based on a preset process tolerance value. In this embodiment, the process tolerance of the key process parameters of the leaf re-drying production line is set to 6.0 degrees Celsius, meaning that the maximum allowable deviation of the process parameters from the set value is ±6.0 degrees Celsius. The lower limit threshold of the mean is calculated using the following formula: .

[0102] in, The median of all data in the batch. To preset process tolerance values, This is the lower limit threshold of the mean. In this embodiment, If the median of a certain batch is Then the lower limit threshold of the mean is .

[0103] Step S5.3: Calculate the upper limit threshold of the standard deviation based on the preset process tolerance value. The upper limit threshold of the standard deviation is equal to the preset process tolerance value divided by 3.

[0104] Preferably, the upper limit threshold of the standard deviation is calculated based on the preset process tolerance value. The upper limit threshold of the standard deviation is calculated according to the following formula: .

[0105] In this embodiment, The calculated upper limit threshold for standard deviation is: Using one-third of the process tolerance as the upper limit of the standard deviation reflects the constraint on the range of variation of the steady-state process under the 3σ principle.

[0106] Step S5.4: Calculate the mean and standard deviation of each time series sub-interval one by one. If the mean of the current time series sub-interval is not less than the lower limit threshold of the mean and the standard deviation of the current time series sub-interval does not exceed the upper limit threshold of the standard deviation, then the current time series sub-interval is determined to be a steady state period; otherwise, it is determined to be a non-steady state period.

[0107] Preferably, the mean and standard deviation of each time series sub-interval are calculated one by one. When the mean of the time series sub-interval is not less than the lower limit threshold of the mean determined in step S5.2, and the standard deviation of the current time series sub-interval does not exceed the upper limit threshold of the standard deviation determined in step S5.3, the time series sub-interval is determined to be a steady-state period and marked; otherwise, it is determined to be a non-steady-state period and marked.

[0108] The determination formula is as follows: .

[0109] in, The mean of the sub-interval. denoted as the standard deviation of the sub-interval.

[0110] Step S5.5: Summarize the judgment results of all time series sub-intervals to obtain the start and end times, mean, standard deviation, and steady-state labeling status of each time series sub-interval.

[0111] Beneficial effects of steps S5.1 to S5.5: This series of steps dynamically calculates the discrimination threshold based on the global median of the batch, and automatically marks each time series sub-interval as steady-state and non-steady-state through the dual constraints of mean and standard deviation, thereby realizing adaptive steady-state identification of differences in operating conditions of different batches.

[0112] Step S5.1 acquires data points from all time-series sub-intervals and calculates the median of all data in the batch, using the batch's global statistics as the criterion to avoid misjudgments caused by fluctuations in statistics of a single sub-interval. Step S5.2 calculates the lower limit threshold of the mean based on the preset process tolerance value, using the batch median minus the process tolerance value as the lower bound for mean discrimination, ensuring that the steady-state discrimination standard is adapted to the actual process level of the batch. Step S5.3 calculates the upper limit threshold of the standard deviation based on the preset process tolerance value, using the process tolerance value divided by 3 as the upper bound for standard deviation discrimination, using the 3σ principle to ensure that the fluctuation range of the steady-state period does not exceed the allowable dispersion of the process. Step S5.4 makes a judgment through the dual constraints of the mean not being lower than the lower limit threshold and the standard deviation not exceeding the upper limit threshold, controlling production deviation while constraining the degree of production dispersion, avoiding misjudgments caused by a single indicator. Step S5.5 summarizes the judgment results of all time-series sub-intervals and outputs the start and end times, mean, standard deviation, and steady-state marker status of each sub-interval, providing structured input data for the subsequent precise boundary positioning module.

[0113] Step S6: Calculate the adaptive control limits by establishing an individual-moving range control chart for the sub-intervals marked as steady-state periods. Perform forward and backward progressive scans on the sub-intervals of the first and last steady-state periods, respectively. Determine the precise start and end times of each steady-state period by using the rule that no more than one-third of the data points exceed the adaptive control limits.

[0114] Furthermore, step S6 specifically includes the following steps: Step S6.1: Calculate the moving range between adjacent sampling points in the time series sub-interval of the steady-state period, and merge the individual value sequence of each steady-state sub-interval with the corresponding moving range sequence to obtain the individual-moving range data of each steady-state sub-interval.

[0115] Preferably, for each time series sub-interval marked as a steady-state period, the moving range between adjacent sampling points within that sub-interval is calculated. The formula for calculating the moving range is: .

[0116] in, and The sampled values ​​are from adjacent sampling points. The corresponding moving range is represented by the individual value sequence of each steady-state sub-interval. The individual value sequence of each steady-state sub-interval is paired and merged with the corresponding moving range sequence to obtain the individual-moving range data for each steady-state sub-interval.

[0117] Step S6.2: Summarize the individual values ​​of all steady-state sub-intervals and calculate the overall mean to serve as the center line of the individual control chart; summarize the moving range of all steady-state sub-intervals and calculate the overall mean to serve as the center line of the moving range control chart.

[0118] Preferably, the overall mean of the individual values ​​from all steady-state sub-intervals is calculated, and this overall mean is used as the center line of the individual control chart. The overall mean of the moving ranges from all steady-state sub-intervals is also calculated, and this overall mean is used as the center line of the moving range control chart. The center line of the control chart represents the expected level of the steady-state process under conditions of no abnormal disturbances.

[0119] The formula for the centerline of an individual control chart is as follows: .

[0120] The formula for the centerline of the moving range control chart is as follows: .

[0121] Where k is the number of steady-state subintervals, Let j be the number of sampling points in the j-th steady-state sub-interval. It represents the sum of the moving ranges of the j-th steady-state sub-interval.

[0122] Step S6.3: Calculate the adaptive upper control limit and adaptive lower control limit of individual values ​​based on the center line of the individual control chart, the center line of the moving range control chart, and the 3σ principle.

[0123] Preferably, based on the center line of the individual control chart Center line of moving range control chart Following the 3σ principle, adaptive upper and lower control limits are calculated for individual values. The adaptive control limits are dynamically calculated based on the actual mean and variation level of each batch of data, eliminating the need for manually preset fixed thresholds.

[0124] The formula for the individual value control limit is as follows: .

[0125] in. This is the ratio of the expected value of the moving range to the standard deviation when the sample size is 2. In this embodiment, adaptive upper control limits are used. Adaptive lower control limit selection .

[0126] Step S6.4: Perform a forward progressive scan from the starting time point of the first steady-state sub-interval to the ending time point, sequentially expanding the range of data points participating in the statistical calculation, and determining whether the newly added data points exceed the individual value adaptive upper control limit and the individual value adaptive lower control limit.

[0127] Preferably, a forward progressive scan is performed from the start time point of the first steady-state sub-interval towards the end time point. The initial scan window contains the first 3 data points of the first steady-state period, and then one data point is added at each step thereafter. The mean and standard deviation of the expanded data subset are recalculated, and it is determined whether the newly added data points exceed the adaptive upper control limit and adaptive lower control limit of the individual values ​​determined in step S6.3.

[0128] Step S6.5: If the proportion of data points exceeding the control limit to the total number of data points in the current time period exceeds one-third, the earliest occurrence time among the data points exceeding the control limit is updated to the precise start time of the current steady-state time period.

[0129] Preferably, during the forward progressive scanning of the first steady-state time interval, data points exceeding the control limits are accumulated. When the proportion of data points exceeding the control limits to the total number of data points in the current scanning window exceeds one-third, the earliest occurrence time of the data points exceeding the control limits is recorded as the precise start time of the first steady-state time interval. The precise start time of the first steady-state time interval will be earlier than or equal to the original start time of that sub-interval.

[0130] Step S6.6: Perform a reverse progressive scan from the end time point of the last steady-state sub-interval towards the start time point, sequentially expanding the range of data points participating in the statistical calculation, and determining whether the newly added data points exceed the individual value adaptive upper control limit and the individual value adaptive lower control limit.

[0131] Preferably, a reverse progressive scan is performed from the end time point of the last steady-state sub-interval towards the starting time point. The initial scan window contains the last 3 data points of the last steady-state sub-interval, and then one data point is added at each step thereafter. The mean and standard deviation of the expanded data subset are recalculated, and it is determined whether the newly added data points exceed the adaptive upper control limit and adaptive lower control limit of the individual values ​​determined in step S6.3.

[0132] Step S6.7: If the proportion of data points exceeding the control limit to the total number of data points in the current time period exceeds one-third, the latest occurrence time of the data points exceeding the control limit is updated to the precise termination time of the current steady-state time period.

[0133] Preferably, during the reverse progressive scanning of the last steady-state sub-interval, data points exceeding the control limits are accumulated. When the proportion of data points exceeding the control limits to the total number of data points in the current scanning window exceeds one-third, the latest occurrence time of the data points exceeding the control limits is recorded as the precise termination time of the last steady-state period. The precise termination time of the last steady-state period will be later than or equal to the original termination time of that sub-interval.

[0134] Step S6.8: Repeat steps S6.4 to S6.7 to scan all sub-intervals marked as steady-state periods. Based on the scanning results, adjust the precise start time and precise end time of each steady-state period to obtain the precise time boundary of each steady-state period.

[0135] Preferably, the forward and backward progressive scan operations of steps S6.4 to S6.7 are repeated for all sub-intervals marked as steady-state periods. The first and last steady-state periods undergo complete forward and backward scans, while intermediate steady-state periods only undergo data verification without boundary updates. Finally, the precise time boundaries of all steady-state periods are output based on the scan results.

[0136] Beneficial effects of steps S6.1 to S6.8: This series of steps provides adaptive control limits through individual-moving range control charts. Combined with forward and backward progressive scanning and one-third tolerance rules, the start and end times of each steady-state period are precisely located at the sampling point level, achieving accurate time boundary output while ensuring robust recognition.

[0137] In step S6.1, the moving range between adjacent sampling points during the steady-state period is calculated and merged with the individual value sequence to construct a complete individual-moving range dataset, providing the necessary data structure for subsequent control limit calculation. Step S6.2 summarizes the individual values ​​and moving ranges of all steady-state sub-intervals and calculates the overall mean, using the overall data statistics of the steady-state period as the control chart centerline, ensuring that the control limits represent the true level of variation in the steady-state process. Step S6.3 calculates the adaptive upper control limit and adaptive lower control limit based on the control chart centerline and the 3σ principle, allowing the control limits to be dynamically adjusted according to the actual data situation without the need for manually preset fixed thresholds. Step S6.4 performs a forward progressive scan on the first steady-state sub-interval, sequentially expanding the data point range and determining whether newly added data points exceed the control limits, achieving fine-grained analysis from the inside of the steady-state period towards the boundary. Step S6.5 updates the precise start time of the first steady-state period using the rule that one-third of the data points exceed the control limit, avoiding boundary misjudgment caused by random fluctuations of a few edge data points. Step S6.6 performs a reverse progressive scan on the sub-interval of the last steady-state period, expanding the data range sequentially from the end time point to the start time point and judging whether the newly added data points exceed the control limit, forming a symmetrical boundary search mechanism with the forward scan. Step S6.7 updates the precise end time of the last steady-state period using the rule that one-third of the data points exceed the control limit, ensuring that the boundary judgment criteria are consistent in the symmetrical direction. Step S6.8 repeats the forward and reverse scans on all steady-state period sub-intervals, adjusts the precise start and end times of each steady-state period based on the scan results, and outputs the complete precise time boundaries of each steady-state period.

[0138] In summary, the overall beneficial effects of steps S1 to S6 in this embodiment are as follows: This method achieves automatic identification and precise location of steady-state periods in batches through full-process processing of batch data from the leaf re-drying production line. In the data acquisition phase, the raw time series of multiple parameters for the entire batch process is directly obtained through the Manufacturing Execution System, ensuring the integrity and traceability of the data source. In the data preprocessing phase, outlier replacement and missing value imputation eliminate acquisition noise, and equal-interval resampling unifies the time reference of each parameter, providing high-quality input data for subsequent analysis. In the mutation point detection phase, the Binary Segmentation algorithm adaptively locates significant structural change points in the time series using a recursive segmentation method, without the need for a pre-set segmentation window, effectively overcoming the insufficient adaptability of traditional fixed-window methods to batch length variations. In the sub-interval partitioning phase, the set of mutation point locations and the batch boundary together constitute a complete segmentation boundary, ensuring the continuity and completeness of the time series sub-intervals and ensuring no production period is missed. In the steady-state discrimination phase, the lower limit threshold of the mean is dynamically calculated based on the global median of the batch, and combined with the upper limit threshold of the standard deviation, a dual constraint judgment is applied to each sub-interval, avoiding the insufficient adaptability of fixed thresholds to differences in operating conditions of different batches. In the precise boundary localization stage, the individual-moving range control chart provides adaptive control limits. The forward and backward progressive scanning combined with the one-third fault tolerance rule ensures the robustness of steady-state period identification while accurately locating the start and end times of each steady-state period to the sampling point level.

[0139] In this process, step S1 ensures the integrity and consistency of the original time series with multiple parameters by associating the batch identifier of the manufacturing execution system with the entire process data; step S2 eliminates acquisition noise by replacing outliers and filling missing values, and unifies the time reference by resampling at equal intervals; step S3 achieves adaptive localization of changes in the internal structure of the batch by recursively detecting significant mutation points using the Binary Segmentation algorithm; step S4 ensures the continuity and completeness of the time series sub-interval division by using the set of mutation point locations and the batch boundary together to form a complete segmentation boundary; step S5 dynamically calculates the discrimination threshold based on the global median of the batch, and achieves adaptive identification of steady-state periods through dual constraints of mean and standard deviation; step S6 calculates adaptive control limits by using individual-moving range control charts, and accurately locates the start and end times of each steady-state period to the sampling point level by combining forward and backward progressive scanning.

[0140] like Figure 2 As shown, this embodiment provides an example of a batch steady-state identification system for the leaf re-drying production process. In this embodiment, the batch steady-state identification system is applied to the batch steady-state identification method described above.

[0141] Specifically, the batch steady-state identification system includes, in sequence, an electrically or communicatively connected module 1 for acquiring the original time series of leaf re-drying, a module 2 for acquiring the target time series, a module 3 for acquiring the set of mutation point locations, a module 4 for dividing the set of mutation point locations, a module 5 for marking the steady-state of time series sub-intervals, and a module 6 for determining the precise start and end times.

[0142] The module 1 for acquiring the original time series of leaf-picking and re-drying is used to collect the original time series of multiple key process parameters for a complete batch on the leaf-picking and re-drying production line; the module 2 for acquiring the target time series is used to preprocess the original time series and construct a complete and equally spaced target time series; and the module 3 for acquiring the set of mutation point locations is used to obtain the Binary data. The Segmentation algorithm iterates through and calculates significant abrupt changes in the target time series, recursively segmenting the target time series until no significant abrupt changes remain, obtaining a set of abrupt change location positions. The abrupt change location set partitioning module 4 is used to construct a complete partition boundary together with the start and end points of the original time series, dividing the target time series into several continuous and non-overlapping time series sub-intervals. The time series sub-interval steady-state marking module 5 is used to calculate the mean and standard deviation of each time series sub-interval. If the sub-interval mean is not less than the median of all data in the batch minus the preset process tolerance value, and the sub-interval standard deviation does not exceed the preset process tolerance value divided by 3, it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period. The precise start and end time determination module 6 is used to calculate the adaptive control limits by establishing an individual-moving range control chart for the sub-intervals marked as steady-state periods, and performs forward progressive scanning and backward progressive scanning on the sub-intervals of the first and last steady-state periods, respectively, with the tolerance rule that no more than one-third of the data points exceed the adaptive control limits, to determine the precise start and precise end times of each steady-state period.

[0143] like Figure 3 As shown, the electronic device 7 includes a processor 71 and a memory 72 coupled to the processor 71.

[0144] The memory 72 stores program instructions for implementing the batch steady-state identification method for the leaf re-drying production process of any of the above embodiments.

[0145] The processor 71 is used to execute program instructions stored in the memory 72 to perform batch steady-state identification in the leaf re-drying production process.

[0146] The processor 71 can also be referred to as a CPU (Central Processing Unit). The processor 71 may be an integrated circuit chip with signal processing capabilities. The processor 71 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0147] Furthermore, Figure 4 This is a schematic diagram of the structure of a storage medium according to an embodiment of this application. See also: Figure 4 The storage medium 8 in this embodiment stores program instructions 81 capable of implementing all the above methods. These program instructions 81 can be stored in the storage medium as a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in each embodiment of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0148] In the several embodiments provided in this application, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the batch steady-state identification of a unit is only one logical function of batch steady-state identification. In actual implementation, there may be other batch steady-state identification methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces. The indirect coupling or communication connection between systems or units may be electrical, mechanical, signal, or other forms.

[0149] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A batch steady-state identification method for the leaf re-drying production process, characterized in that, The batch steady-state identification method includes: Step S1: Collect the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line; Step S2: Perform data preprocessing on the original time series and construct a complete and equally spaced target time series; Step S3: The significant mutation points of the target time series are calculated by traversing through the Binary Segmentation algorithm, and the target time series is recursively segmented until there are no significant mutation points, thus obtaining a set of mutation point locations. Step S4: The set of mutation point locations, together with the start and end points of the original time series, constitute a complete segmentation boundary to divide the target time series into several continuous and non-overlapping time series sub-intervals. Step S5: Calculate the mean and standard deviation of each time series sub-interval. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period. Step S6: Calculate the adaptive control limits by establishing an individual-moving range control chart for the sub-intervals marked as steady-state periods. Perform forward progressive scanning and backward progressive scanning on the sub-intervals of the first and last steady-state periods respectively. Use the rule that no more than one-third of the data points exceed the adaptive control limits as the fault tolerance rule to determine the precise start time and precise end time of each steady-state period.

2. The batch steady-state identification method according to claim 1, characterized in that, Step S1: Collect the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line, including: Step S1.1: Obtain the raw data of each process node of the leaf-re-drying production line from the manufacturing execution system of the leaf-re-drying production line; Step S1.2: Based on the batch identifier on the leaf re-drying production line, filter all the original data to find the production data belonging to the same complete batch. The batch identifier is used to associate the entire production data of a batch on the leaf re-drying production line from feeding to finished product warehousing. Step S1.3: Determine the types of key process parameters to be collected based on the preset process parameter list; Step S1.4: Extract the sampled values ​​corresponding to each key process parameter type from the selected production data of the same batch in order of timestamps to construct the original time series of each key process parameter; Step S1.5: Associate and store the original time series of each key process parameter with the corresponding process parameter type identifier to obtain the original time series data of multiple key process parameters for a complete batch on the leaf re-drying production line.

3. The batch steady-state identification method according to claim 1, characterized in that, Step S2 involves preprocessing the original time series data and constructing a complete and equally spaced target time series, including: Step S2.1: Identify and mark outliers point by point in the original time series of each key process parameter; Step S2.2: Replace the sampling points marked as outliers. When there is temporal continuity between the outlier and the adjacent normal sampling points, the replacement value is calculated using the adjacent normal sampling values ​​through linear interpolation. When the outlier is an isolated sampling point, the average of the adjacent normal sampling values ​​is used as the replacement value. Step S2.3: Detect missing values ​​in the original time series of each key process parameter. If the actual number of samples is less than the theoretical number of samples in the time series, fill the missing position by linear interpolation of the adjacent normal sample values. Step S2.4: After filling and repairing, the time series of each key process parameter is resampled at different sampling intervals. Linear interpolation is used to unify each time series to a preset equal-interval sampling frequency to obtain a complete target time series with equal-interval sampling.

4. The batch steady-state identification method according to claim 1, characterized in that, Step S3: The significant abrupt change points of the target time series are calculated by traversing the BinarySegmentation algorithm, and the target time series is recursively segmented until no significant abrupt change points are found, resulting in a set of abrupt change point locations, including: Step S3.1: Quantify the degree of fitting improvement in dividing the target time series into left and right sub-segments at the candidate split points by the log-likelihood ratio statistic. Calculate the log-likelihood ratio statistic point by point for all candidate split points on the target time series that satisfy the minimum split interval. Step S3.2: Traverse all candidate split points to locate the candidate split point that maximizes the log-likelihood ratio statistic as the optimal split point for the current interval. Step S3.3: Define a significance threshold based on a chi-square distribution with 2 degrees of freedom, and set the value of the significance threshold based on a preset confidence level; Step S3.4: Compare the maximum log-likelihood ratio statistic corresponding to the current optimal segmentation point with the significance threshold. If the current maximum statistic exceeds the significance threshold, then determine the current optimal segmentation point as a significant mutation point and add the corresponding position to the mutation point position set. Step S3.5: Repeat the traversal calculation and significance determination of steps S3.1 to S3.4 for the left and right sub-segments generated by each segmentation. Stop the recursive segmentation if there are no candidate segmentation points exceeding the significance threshold in a certain sub-segment, and output a set of mutation point locations including all significant mutation point locations.

5. The batch steady-state identification method according to claim 1, characterized in that, Step S4: The set of mutation point locations, together with the start and end points of the original time series, constitute a complete segmentation boundary to divide the target time series into several continuous and non-overlapping time series sub-intervals, including: Step S4.1: Obtain the starting sampling time of the target time series as the starting time point and the ending sampling time of the target time series as the ending time point; Step S4.2: Combine the sampling times corresponding to all mutation points in the mutation point location set with the start time point and the end time point to obtain a boundary time set including all segmentation boundaries; Step S4.3: Sort the set of boundary moments in ascending order of time, remove duplicate boundary moments, and obtain an ordered and non-repeating complete set of segmentation boundaries. Step S4.4: Based on the time interval formed by two adjacent boundary moments in the complete segmentation boundary set, the target time series is segmented and divided into several continuous and non-overlapping time series sub-intervals. Step S4.5: Record the start and end times and duration of each time series sub-interval to obtain the set of time series sub-intervals after division.

6. The batch steady-state identification method according to claim 1, characterized in that, Step S5: Calculate the mean and standard deviation of each time series sub-interval. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period, including: Step S5.1: Obtain the data points of all time-series sub-intervals in the time-series sub-interval set, and calculate the median of all data points as the median of all data in the batch; Step S5.2: Calculate the lower limit threshold of the mean based on the preset process tolerance value. The lower limit threshold of the mean is equal to the median of all data in the batch minus the preset process tolerance value. Step S5.3: Calculate the upper limit threshold of the standard deviation based on the preset process tolerance value. The upper limit threshold of the standard deviation is equal to the preset process tolerance value divided by 3. Step S5.4: Calculate the mean and standard deviation of each time series sub-interval one by one. If the mean of the current time series sub-interval is not less than the lower limit threshold of the mean and the standard deviation of the current time series sub-interval does not exceed the upper limit threshold of the standard deviation, then the current time series sub-interval is determined to be a steady state period; otherwise, it is determined to be a non-steady state period. Step S5.5: Summarize the judgment results of all time series sub-intervals to obtain the start and end times, mean, standard deviation, and steady-state labeling status of each time series sub-interval.

7. The batch steady-state identification method according to claim 1, characterized in that, Step S6: An individual-moving range control chart is established for the sub-intervals marked as steady-state periods to calculate adaptive control limits. Forward and backward progressive scans are performed on the sub-intervals of the first and last steady-state periods, respectively. Using the rule that no more than one-third of the data points exceed the adaptive control limits as a tolerance rule, the precise start and end times of each steady-state period are determined, including: Step S6.1: Calculate the moving range between adjacent sampling points in the time series sub-interval of the steady-state period, and merge the individual value sequence of each steady-state sub-interval with the corresponding moving range sequence to obtain the individual-moving range data of each steady-state sub-interval; Step S6.2: Summarize the individual values ​​of all steady-state sub-intervals and calculate the overall mean to serve as the center line of the individual control chart; Summarize the moving range of all steady-state sub-intervals and calculate the overall mean to serve as the center line of the moving range control chart. Step S6.3: Calculate the adaptive upper control limit and adaptive lower control limit of the individual value based on the center line of the individual control chart, the center line of the moving range control chart, and the 3σ principle; Step S6.4: Perform a forward progressive scan from the starting time point of the first steady-state sub-interval to the ending time point, sequentially expanding the range of data points participating in the statistical calculation, and determining whether the newly added data points exceed the adaptive upper control limit and the adaptive lower control limit of the individual value. Step S6.5: If the proportion of data points exceeding the control limit to the total number of data points in the current time period exceeds one-third, update the earliest occurrence time of the data points exceeding the control limit to the precise start time of the current steady-state time period. Step S6.6: Perform a reverse progressive scan from the end time point of the last steady-state sub-interval towards the start time point, sequentially expanding the range of data points participating in the statistical calculation, and determining whether the newly added data points exceed the individual value adaptive upper control limit and the individual value adaptive lower control limit; Step S6.7: If the proportion of data points exceeding the control limit to the total number of data points in the current time period exceeds one-third, update the latest occurrence time of the data points exceeding the control limit to the precise termination time of the current steady-state time period. Step S6.8: Repeat steps S6.4 to S6.7 to scan all sub-intervals marked as steady-state periods. Based on the scanning results, adjust the precise start time and precise end time of each steady-state period to obtain the precise time boundary of each steady-state period.

8. A batch steady-state identification system for a leaf re-drying production process, wherein the batch steady-state identification system is applied to the batch steady-state identification method as described in any one of claims 1 to 7, characterized in that, The batch steady-state identification system includes: The original time series acquisition module for leaf re-drying is used to collect the original time series of multiple key process parameters for a complete batch on the leaf re-drying production line. The target time series acquisition module is used to preprocess the original time series and construct a complete and equally spaced target time series. The mutation point location set acquisition module is used to traverse and calculate the significant mutation points of the target time series through the Binary Segmentation algorithm, and recursively segment the target time series until there are no significant mutation points, thereby obtaining the mutation point location set. The mutation point location set partitioning module is used to form a complete partitioning boundary by combining the mutation point location set with the start and end points of the original time series, so as to divide the target time series into several continuous and non-overlapping time series sub-intervals. The steady-state marking module for time-series sub-intervals is used to calculate the mean and standard deviation of each time-series sub-interval. If the mean of the sub-interval is not less than the median of all data in the batch minus the preset process tolerance value, and the standard deviation of the sub-interval does not exceed the preset process tolerance value divided by 3, then it is marked as a steady-state period; otherwise, it is marked as a non-steady-state period. The precise start and end time determination module is used to calculate the adaptive control limits by establishing an individual-moving range control chart for sub-intervals marked as steady-state periods, and to perform forward progressive scanning and backward progressive scanning on the sub-intervals of the first and last steady-state periods respectively, with the fault tolerance rule being that no more than one-third of the data points exceed the adaptive control limits, to determine the precise start and precise end times of each steady-state period.

9. An electronic device, characterized in that, The method includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the batch steady-state identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed by a processor, enable the batch steady-state identification method as described in any one of claims 1 to 7.

Images