Cigarette index threshold generation method and device and storage medium

Abstract

The invention relates to a cigarette index threshold generation method and device and a storage medium, and the method comprises the steps: obtaining actual index data for a cigarette production process; inputting the actual index data into the constructed prediction model, and outputting predicted future data; the prediction model is generated by training according to collected multivariable time series data, and data features of the multivariable time series data are matched with data features of actual index data; and obtaining an extended data set according to the actual index data and the predicted future data, and performing real-time calculation according to the extended data set to generate a dynamic cigarette index threshold. According to the method and the device, the problem of low accuracy of setting the cigarette index threshold is solved, and the dynamic monitoring of the cigarette quality is realized.

Description

Technical Field

[0001] The present application relates to the field of cigarette index detection, and in particular, to a method, device, and storage medium for generating cigarette index thresholds. Background Art

[0002] With the continuous upgrading of the informatization of cigarette production enterprises and the manufacturing execution system of cigarette enterprises, basic data is continuously generated. Reasonably utilizing these data and deeply mining the information contained therein are of great help to improving the quality of cigarettes.

[0003] In the current production process of cigarette packing, the moisture and weight of cigarettes are key factors for measuring the qualification of cigarettes. The accurate prediction and timely regulation of these indexes have important impacts on product quality and production efficiency. However, at present, for the key quality indexes such as the moisture and weight of cigarettes in the cigarette production process, the alarm thresholds are mainly set according to manual experience, which is easily affected by subjective factors and it is difficult to ensure the applicability and accuracy for different production scenarios. Since the thresholds set by manual experience are usually static and cannot be dynamically adjusted according to the real-time production situation, when an abnormality occurs in the production process, it is necessary to wait until the problem has appeared or affected the production to trigger an alarm, resulting in a slow response speed and the inability to detect and respond to potential problems in a timely manner.

[0004] Currently, no effective solution has been proposed for the problem of low accuracy in setting cigarette index thresholds in the related art. Summary of the Invention

[0005] Embodiments of the present application provide a method, device, and storage medium for generating cigarette index thresholds to at least solve the problem of low accuracy in setting cigarette index thresholds in the related art.

[0006] In a first aspect, embodiments of the present application provide a method for generating a cigarette index threshold, the method including:

[0007] Obtaining actual index data for a cigarette production process;

[0008] Inputting the actual index data into a constructed prediction model and outputting predicted future data; the prediction model is generated by training based on collected multivariate time series data, and the data characteristics of the multivariate time series data match the data characteristics of the actual index data;

[0009] Obtaining an extended data set according to the actual index data and the predicted future data, and generating a dynamic cigarette index threshold by real-time calculation based on the extended data set.

[0010] In some embodiments, the method further includes:

[0011] Collect historical time-series data of indicators and historical time-series data of process parameters; the multi-variable time-series data includes the historical time-series data of indicators and the historical time-series data of process parameters, and the historical time-series data of process parameters is associated with the historical time-series data of indicators;

[0012] Perform difference calculation on the multi-variable time-series data to obtain a first estimation parameter;

[0013] For the historical time-series data of indicators and the historical time-series data of process parameters, calculate the results of a correlation function, and determine a second estimation parameter and a third estimation parameter according to the results of the correlation function;

[0014] Construct the prediction model based on the first estimation parameter, the second estimation parameter, and the third estimation parameter.

[0015] In some embodiments, the performing difference calculation on the multi-variable time-series data to obtain a first estimation parameter includes:

[0016] Generate a trend graph and a seasonal graph according to the multi-variable time-series data;

[0017] Perform stationarity detection on the multi-variable time-series data according to the trend graph and the seasonal graph to obtain a stationarity detection result, and perform difference calculation based on the stationarity detection result to obtain the first estimation parameter.

[0018] In some embodiments, the constructing the prediction model based on the first estimation parameter, the second estimation parameter, and the third estimation parameter includes:

[0019] Determine multiple candidate parameter groups based on the first estimation parameter, the second estimation parameter, and the third estimation parameter;

[0020] Traverse the particle parameters corresponding to each candidate parameter group;

[0021] Calculate the result of the fitness function of the traversed particle parameters, calculate the global optimal position based on the result of the fitness function, and determine the optimal parameter combination in the candidate parameter group based on the global optimal position;

[0022] Construct the prediction model according to the optimal parameter combination.

[0023] In some embodiments, the calculating the result of the fitness function of the traversed particle parameters and calculating the global optimal position based on the result of the fitness function includes:

[0024] Calculate the current particle flight speed according to the traversed particle parameters, the preset inertia weight, and the historical particle flight speed, and calculate the current particle position according to the current particle flight speed and the historical particle position;

[0025] Iteratively calculate the fitness function result of the particle parameters at the current particle flight speed and the current particle position until a preset iteration termination condition is reached, and determine the global optimal position.

[0026] In some embodiments, the collecting the historical index time series data and the historical process parameter time series data includes:

[0027] Obtain the collected historical index time series data;

[0028] Input the historical index time series data and the obtained preset process features into the trained feature screening model, and use the feature screening model to select the target process features associated with the historical index time series data from the preset process features;

[0029] Based on the target process features, obtain the historical process parameter time series data.

[0030] In some embodiments, the inputting the actual index data into the constructed prediction model and outputting the predicted future data includes:

[0031] Obtain the production condition information for the cigarette rod production process;

[0032] When the production condition information indicates stable conditions, input the actual index data into the constructed prediction model for processing, and output the predicted future data;

[0033] When the production condition information indicates changing conditions, reconstruct a new prediction model, input the actual index data into the new prediction model, and output the predicted future data.

[0034] In some embodiments, the generating the dynamic cigarette rod index threshold according to the real-time calculation of the extended data set includes:

[0035] Perform calculation and statistics according to the extended data set to generate a mean control chart;

[0036] Determine the mean data based on the mean control chart; determine the upper threshold according to the mean data and the preset upper control limit coefficient, and determine the lower threshold according to the mean data and the preset lower control limit coefficient;

[0037] The dynamic cigarette rod index threshold includes the upper threshold and the lower threshold.

[0038] In a second aspect, an apparatus for generating a threshold value of cigarette rod indicators according to an embodiment of the present application includes:

[0039] An acquisition module, configured to acquire actual indicator data for a cigarette rod production process;

[0040] A prediction module, configured to input the actual indicator data into a constructed prediction model and output predicted future data; the prediction model is generated by training based on collected multivariate time series data, and the data characteristics of the multivariate time series data match the data characteristics of the actual indicator data;

[0041] A threshold value generation module, configured to obtain an extended data set according to the actual indicator data and the predicted future data, and generate a dynamic cigarette rod indicator threshold value by performing real-time calculation according to the extended data set.

[0042] In a third aspect, an embodiment of the present application provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the method for generating a threshold value of cigarette rod indicators as described in the first aspect above.

[0043] Compared with the related art, the method, apparatus, and storage medium for generating a threshold value of cigarette rod indicators provided by the embodiments of the present application acquire actual indicator data for a cigarette rod production process; input the actual indicator data into a constructed prediction model and output predicted future data; the prediction model is generated by training based on collected multivariate time series data, and the data characteristics of the multivariate time series data match the data characteristics of the actual indicator data; obtain an extended data set according to the actual indicator data and the predicted future data, and generate a dynamic cigarette rod indicator threshold value by performing real-time calculation according to the extended data set.

[0044] Based on this, by collecting actual data, using a prediction model to predict future data, and generating a dynamic threshold value based on an extended data set, problems such as lagged warning response caused by setting indicator threshold values based on manual experience are avoided. Thus, both accurate prediction of product quality at a specific future moment can be achieved, and statistical process control can be incorporated, thereby achieving efficient and real-time monitoring and intelligent management of the cigarette rod production process, improving the accuracy of dynamic threshold value generation, effectively solving the problem of low accuracy in setting the threshold value of cigarette rod indicators, and realizing dynamic monitoring of cigarette rod quality.

[0045] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0047] Figure 1 is a hardware structure block diagram of a terminal for a method of generating a cigarette stick index threshold according to an embodiment of the present application;

[0048] Figure 2 is a flowchart of a method of generating a cigarette stick index threshold according to an embodiment of the present application;

[0049] Figure 3 is a schematic diagram of a data trend chart and a season chart according to an embodiment of the present application;

[0050] Figure 4 is a mean control chart of actual index data according to an embodiment of the present application;

[0051] Figure 5 is a schematic diagram of a model prediction comparison result according to an embodiment of the present application;

[0052] Figure 6 is a mean control chart of an extended data set according to an embodiment of the present application;

[0053] Figure 7 is a flowchart of another method of generating a cigarette stick index threshold according to an embodiment of the present application;

[0054] Figure 8 is a structure block diagram of a device for generating a cigarette stick index threshold according to an embodiment of the present application. Detailed implementation manners

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be described and explained below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments provided in the present application without making creative efforts belong to the scope of protection of the present application. In addition, it can also be understood that although the efforts made in this development process may be complex and time-consuming, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing or production changes based on the technical content disclosed in the present application are only conventional technical means and should not be understood as insufficient disclosure of the content of the present application.

[0056] References to "embodiments" in this application mean that specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment each time, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those of ordinary skill in the art will explicitly and implicitly understand that the embodiments described in this application can be combined with other embodiments without conflict.

[0057] Unless otherwise defined, technical terms or scientific terms involved in this application shall have the ordinary meaning as understood by those of ordinary skill in the technical field to which this application pertains. The words "a", "an", "one kind", "the" and similar words involved in this application do not indicate a limitation in quantity and can represent a singular or plural number. The terms "include", "comprise", "have" and any variations thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include steps or units not listed, or may further include other steps or units inherent to these processes, methods, products or devices. The words "connect", "be connected", "couple" and similar words involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "a plurality" involved in this application means greater than or equal to two. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may represent: A exists alone, A and B exist simultaneously, and B exists alone, these three situations. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0058] The method embodiments provided in this embodiment can be executed on a terminal, a computer, or a similar computing device. Taking running on a terminal as an example, Figure 1 is a hardware structure block diagram of a terminal for a method of generating a cigarette index threshold according to an embodiment of this application. As Figure 1 shown, the terminal may include one or more ( Figure 1 only one is shown in the figure) processors 102 (the processors 102 may include, but are not limited to, processing devices such as a microprocessor MCU or a field programmable gate array FPGA) and a memory 104 for storing data. Optionally, the above terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above terminal. For example, the terminal may further include more or fewer components than those shown in Figure 1 the figure, or have a structure different from that shown inFigure 1 The different configurations shown.

[0059] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for generating cigarette index thresholds in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof.

[0060] The transmission device 106 is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by a communication provider of the terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0061] This embodiment provides a method for generating cigarette index thresholds. Figure 2 It is a flowchart of a method for generating cigarette index thresholds according to an embodiment of the present application, as Figure 2 shown, and the process includes the following steps:

[0062] Step S210, obtain actual index data for the cigarette production process.

[0063] The actual index data refers to the key quality index data collected in real time during the production processes such as cigarette rolling and packing; for example, the actual index data can be cigarette moisture, weight, cigarette draw resistance, cigarette diameter, tobacco filling value, and / or tipping paper quality, etc., which are index data that can be included in the monitoring scope.

[0064] Step S220, input the actual index data into the constructed prediction model and output predicted future data; the prediction model is generated by training with the collected multi-variable time series data, and the data characteristics of the multi-variable time series data match the data characteristics of the actual index data.

[0065] The above prediction model is a model that has been pre-trained using machine learning or deep learning techniques with historical multivariate time series data (such as production data over a period of time). This model can learn the patterns in the data and make predictions about future data. In this embodiment, the prediction model can adopt models such as the Vector Autoregressive (VAR) / (Vector Autoregressive Moving Average, VARMA) series models, the Transformer model, or the global joint training model that supports joint training of multiple sequences, etc.

[0066] Taking the VARMA model as an example, the construction process of the prediction model for the cigarette moisture index will be briefly described below. Considering that the cigarette moisture in the actual production process is affected by the outlet temperature, a VARMA model is introduced based on the univariate prediction of the Autoregressive Integrated Moving Average (ARIMA) model. Specifically, the synchronized multivariate time series data is input into the initial VARMA model for iterative training including stationarity processing and hyperparameter selection. That is, first, the non-stationary sequence of multivariate key quality indicators including cigarette moisture and outlet temperature is differenced to determine the minimum number of differences d to obtain a stationary sequence that meets the VARMA model. Then, based on the multivariate time series data of the stationary sequence, the Autocorrelation Function (ACF) and the Partial Autocorrelation Function (PACF) are calculated, and the ranges of p and q are initially judged in combination with their trailing or truncating characteristics. For example, if the ACF has a trailing tail and the PACF has a truncating tail, an AR model (corresponding to the p value) may be applicable; otherwise, an MA model (corresponding to the q value) may be applicable. During the training process, the model continuously adjusts its parameters to minimize the prediction error. Also, the accuracy of the training model is evaluated by calculating the error between the true value and the predicted value of the test set. If the error exceeds the threshold, the lag order is adjusted or external variables are introduced for retraining. Finally, a trained prediction model is constructed and generated.

[0067] It should also be noted that, to ensure the accuracy of prediction, the data characteristics of the multivariate time series data used to construct the prediction model (including data format, data type, and / or variable name, etc.) should match the data characteristics of the actual indicators. This is because during the model training process, the model has learned and adapted to the characteristics of these specific data, and any inconsistency may lead to deviations in the prediction results. For example, if the multivariate time series data involved in training includes time series data with the variable name of moisture index, it means that this prediction model is used to predict future moisture index, that is, the actual indicator data input into this prediction model should also include the actually collected moisture index data.

[0068] Once the actual indicator data is correctly input, the prediction model will start its prediction process. The model will use the rules and patterns learned internally to conduct in-depth analysis and processing on the input actual indicator data. These rules and patterns are obtained through learning a large amount of historical data during the model training stage, and they reflect the internal relationships and changing trends among the data. During the prediction process, the model will predict the data for a period of time in the future based on the characteristics of the input data and in combination with the rules and patterns it has learned. For example, the actual indicator data input into the prediction model includes multiple groups of moisture index data sampled within the first hour. Inputting this into the prediction model will output the predicted moisture index data for the next hour.

[0069] Through the above step S220, seamless docking between the prediction model and the actual production data can be ensured, achieving accurate prediction and closed-loop control of cigarette quality indicators.

[0070] Step S230: Obtain an extended data set based on the actual indicator data and the predicted future data, and calculate in real time according to the extended data set to generate a dynamic cigarette index threshold.

[0071] Merge the actual indicator data and the predicted future data to form a more comprehensive data set. This data set not only contains the current data but also the predicted data for a period of time in the future, so it is called an extended data set. Among them, the actual indicator data and the predicted future data can be directly fused in chronological order to obtain the extended data set. Or, when permitted by the embodiment, a corresponding weight assignment strategy can also be formulated to assign corresponding weight values to the actual indicator data and the predicted future data respectively, and perform weighted processing on the two types of data to obtain the extended data set; for example, the weight values assigned to these two types of data can be determined according to prior knowledge or manual setting, or by obtaining the model evaluation results of the above preset model and determining the weight values of the two types of data based on the model evaluation results, etc.

[0072] Subsequently, real-time data analysis and calculation are performed on the extended data set obtained from the above statistics. According to the results of the real-time calculation, the index thresholds for cigarette production are dynamically adjusted; the index thresholds may include upper limit values, lower limit values, or tolerance ranges of quality standards, etc.

[0073] In the above method for generating cigarette index thresholds, by collecting actual data, using a prediction model to predict future data, and generating dynamic thresholds based on the extended data set, problems such as lagged warning responses caused by setting index thresholds based on manual experience are avoided. It can not only accurately predict the product quality at a specific future moment but also incorporate statistical process control, thereby considering the overall distribution and fluctuations of the data, making the setting of the thresholds more reasonable, so as to achieve efficient and real-time monitoring and intelligent management of the cigarette production process, improve the accuracy of generating dynamic thresholds, effectively solve the problem of low accuracy in setting cigarette index thresholds, and achieve dynamic monitoring of cigarette quality.

[0074] In some embodiments, in the above method for generating cigarette index thresholds, constructing a prediction model may include the following steps:

[0075] First, collect historical index time-series data and historical process parameter time-series data; the multi-variable time-series data includes historical index time-series data and historical process parameter time-series data, and the historical process parameter time-series data is associated with the historical index time-series data.

[0076] Historical index time-series data refers to collecting various index data of cigarettes over a past period of time, such as cigarette moisture, weight, length, draw resistance, etc., and these data are arranged in a time series. At the same time, collect the process parameter data corresponding to these indexes, such as the outlet temperature corresponding to the cigarette moisture index, the material conveying speed corresponding to the weight index, the pressure of the wrapping equipment, etc., and these data are also arranged in a time series, that is, the above historical process parameter time-series data is obtained. In the actual production process, the adjustment of these historical process parameters will directly affect the qualification of the final index data.

[0077] During the data collection process, it is necessary to ensure the accuracy and integrity of the data and avoid introducing noise or outliers. Taking the moisture content of cut tobacco as an example, the process of data cleaning to ensure data reliability is described below. First, collect part of the moisture data of cigarette rods at 20:00 on November 20, 2023, including the data of the factor of the outlet temperature of cut tobacco drying with the greatest influence. Since there may be various abnormal situations in the actual production process, such as changes in raw material batches, equipment failures, environmental changes, etc., resulting in uneven data, the quartile method is used to identify and remove abnormal samples to ensure the stability and reliability of the data: Calculate the interquartile range (IQR) of the moisture sequence data, that is, the distance between the upper quartile and the lower quartile. Then determine the upper and lower limits of the outliers, namely the upper limit and the lower limit; determine that the data in the currently collected index time series data that is less than the lower limit or greater than the upper limit is an outlier and needs to be removed, and finally obtain the multivariate time series data after removing outliers and completing data cleaning.

[0078] In addition, different from the traditional ARIMA, VARMA uses vector input. Still taking the cigarette rod moisture index to be predicted as an example, the input vector is as shown in the formula: Y t =[Y t,1 ,Y t,2 T . In the above input vector representation formula, Y t,1 represents the cigarette rod moisture, and Y t,2 represents the outlet temperature.

[0079] Next, perform a difference calculation on the multivariate time series data to obtain the first estimation parameter; calculate the correlation function results for the historical index time series data and the historical process parameter time series data, and determine the second and third estimation parameters according to the correlation function results; construct a prediction model based on the first, second, and third estimation parameters.

[0080] Specifically, the above multivariate time series data (including historical index time series data and historical process parameter time series data) can be divided according to a certain ratio, such as divided into a training set and a test set according to a ratio of 8:2. During the model training process, input the training set for iterative training. Among them, perform a difference calculation on the historical index time series data and the historical process parameter time series data in the training set. The difference calculation is a commonly used time series analysis method for eliminating the trend component in the data, making it more stable and facilitating subsequent analysis. Through the difference calculation, the first estimation parameter d can be obtained, which represents the number of differences required to transform a non-stationary sequence into a stationary sequence. For example, judge the stationarity of the above input vector Y t . If Y t is not stationary, then for Y t ​Perform differencing. Assume Y t After d times of differencing, a stationary sequence Z is obtained t , then Z t The expression of is: Z t =(1 - B) d Y t , where (1 - B) d is the differencing operator, B is the backshift operator, and Z t represents the stationary VARMA sequence.

[0081] After obtaining the stationary time series, calculate the autocorrelation function ACF and partial autocorrelation function PACF of the sequence (i.e., the above correlation function results). The correlation function is used to measure the strength of the linear relationship between two variables, and its results can help determine which process parameters have a significant impact on the index data. Among them, the autocorrelation function ACF represents the correlation between the time series and itself at different lags, representing the correlation between each time point in the sequence and its lagged value. The calculation method of the autocorrelation function is as follows:

[0082]

[0083] In the above formula, ACF(k) represents the autocorrelation coefficient at lag k, x t represents the value of the time series at time point t, represents the mean of the time series, and n represents the length of the time series.

[0084] The partial autocorrelation function PACF is the partial autocorrelation function or the partial autocorrelation function. Basically, instead of finding the correlation between the lag and the current like ACF, it finds the correlation between the residual and the next lagged value, as shown in the following formula:

[0085]

[0086] Among them, PACF(k) represents the partial autocorrelation coefficient at lag k, Υ(k) represents the partial autocovariance at lag k, and Υ(0) represents the variance of the sequence. Combining ACF and PACF can help determine whether differencing is needed and the number of differencing times. At the same time, combining the characteristics of the function's tailing or truncating can initially determine the ranges of p and q. For example, if ACF tails and PACF truncates, an AR model may be applicable, corresponding to the second estimated parameter p value; p represents the number of autoregressive terms, which determines how many past time steps of its own the model considers. Conversely, an MA model may be applicable, corresponding to the third estimated parameter q value; q represents the number of moving average terms, which determines the impact of random shocks in the past how many time steps on the current variable.

[0087] Based on the first estimation parameter (differential calculation result), the second estimation parameter (a part of the correlation function result), and the third estimation parameter (another part of the correlation function result), a prediction model is constructed. When constructing the model, it is necessary to select an appropriate model structure and parameters to ensure that the model can accurately capture the patterns and trends in the data.

[0088] After that, the constructed prediction model is verified using the above-divided test set. By comparing the differences between the predicted results and the actual results, the accuracy and reliability of the model are evaluated. According to the verification results, necessary adjustments and optimizations are made to the model to improve its prediction performance. Specifically, verification calculations can be performed through indicators such as root mean square error (RMSE) and mean absolute error (MAE), and the formulas are as follows:

[0089]

[0090] In the above formula, y pred is the predicted value of the test set, and y is the true value of the test set. The smaller the values of RMSE and MAE, the smaller the error and the higher the accuracy of the model.

[0091] Compared with the related art, in which a univariate ARIMA model is used and only the time series characteristics of cigarette rod indicators (such as moisture) themselves are considered, resulting in a way that affects the accuracy of threshold setting, through the above embodiments of the present application, by introducing process parameters as multivariate inputs and constructing a prediction model, the correlation between cigarette rod indicators and process parameters can be effectively captured, which helps to more accurately reflect the actual dynamics of the production process.

[0092] In some of these embodiments, the above-mentioned differential calculation of multivariate time series data to obtain the first estimation parameter may further include the following steps:

[0093] Generate a trend chart and a seasonal chart based on the multivariate time series data; perform a stationarity test on the multivariate time series data according to the trend chart and the seasonal chart to obtain a stationarity test result, and perform differential calculation based on the stationarity test result to obtain the first estimation parameter.

[0094] Among them, the trend chart is used to show the overall trend of the data changing over time. By plotting the time series chart of each variable, it can be intuitively seen whether there is a long-term upward or downward trend in the data. Seasonal chart: The seasonal chart is used to reveal the seasonal fluctuations in the data. It is usually generated by grouping the data according to a specific period (such as days, weeks, months, etc.) and calculating the average value, so as to help identify the periodic changes in the data.

[0095] Specifically, please refer toFigure 3 , the figure respectively shows a trend graph and a seasonal graph generated from multivariate time series data. The abscissa is used to represent the sampling points of the time series, and the ordinate represents the index of cigarette quality. Among them, Figure 3 the curve graph in the upper half is the trend graph. In the trend graph, the original data series and the "trend term" representing the data are respectively shown. The trend term is used to depict the long-term change trend of the index (such as the overall upward, downward or stable trend). Figure 3 the curve graph in the lower half is the seasonal graph. In the seasonal graph, the original data series and the "seasonal component" or mean reference line representing the data are respectively shown, which is used to assist in observing the seasonal fluctuation law that repeats within a fixed period (such as every day, each shift) after the data is detrended. In addition, in the trend graph and the seasonal graph, the curve representing the original data series can visually present the actual fluctuation of the cigarette index in the time dimension, including comprehensive information such as trend, season, and randomness. For example, if the data in this period shows a significant numerical span, the curve form does not extend smoothly, but continuously shows peaks and valleys, indicating that the overall fluctuation amplitude of the data in the time series is large. And in the seasonal graph, it fluctuates violently around the mean reference line, not only frequently deviating from the mean reference line, but also with a high degree of deviation, indicating that the fluctuation intensity of the data within the seasonal cycle is large and the stability is poor.

[0096] After generating the trend graph and the seasonal graph, it is necessary to perform stationarity detection on the multivariate time series data based on the trend graph and the seasonal graph. Taking Figure 3 as an example, at this time, according to the above data analysis of the illustrated curve, it can be determined that the multivariate time series data including the cigarette moisture index and the outlet temperature has great volatility. Further stationarity tests are required. Stationarity means that the statistical characteristics (such as mean, variance) of the data do not change with time. Stationarity detection is an important prerequisite for time series analysis because many time series analysis methods assume that the data is stationary.

[0097] In this embodiment, a stationarity detection method of the Augmented Dickey-Fuller (ADF) test can be adopted. The ADF test is a quantitative test index for the stationarity of time series. Its main idea is to judge whether there is a unit root in the time series. If there is a unit root, the time series is a non-stationary series. The ADF test first proposes a null hypothesis (there is a unit root, i.e., non-stationary) and an alternative hypothesis (stationary), and then performs the test to obtain the ADF statistic and the critical value. If the statistic is less than the critical value, the null hypothesis is rejected and it is considered stationary; if it is greater, the null hypothesis cannot be rejected and it is considered non-stationary. Finally, according to the test results, its difference is calculated until the time series is stationary. For a multi-variable time series, the ADF test is performed on each variable separately. The ADF test statistic of the original series is greater than any critical value, and the P-value is much greater than 0.05. Therefore, the hypothesis of having a unit root is rejected, that is, it cannot be confirmed that the time series data is stationary. The ADF statistic after differencing is [value not provided in the original], the P-value is almost 0, and the ADF is much less than the critical value of 1%. This indicates that there is no unit root in the differenced series, that is, the time series data is stationary and the next step of model fitting can be carried out.

[0098] Through the above embodiments, trend charts and seasonal charts can be generated based on multi-variable time series data, stationarity detection can be performed, and the first estimation parameter can be obtained through difference calculation, providing a solid foundation for subsequent construction of a prediction model and generation of dynamic thresholds, thereby facilitating the improvement of the accuracy of setting dynamic cigarette index thresholds.

[0099] In some of these embodiments, constructing a prediction model based on the first estimation parameter, the second estimation parameter, and the third estimation parameter may further include the following steps:

[0100] Based on the first estimation parameter, the second estimation parameter, and the third estimation parameter, determine multiple candidate parameter groups; traverse the particle parameters corresponding to each candidate parameter group; calculate the fitness function results of the traversed particle parameters, based on the fitness function results, calculate the global optimal position, and based on the global optimal position, determine the optimal parameter combination in the candidate parameter group; construct a prediction model according to the optimal parameter combination.

[0101] It should be noted that the three parameters p, d, and q of the prediction model play a decisive role in the final prediction effect of the model. Excessively high p or q will cause the prediction model to overfit, and excessively low p or q may cause the prediction model to underfit. Moreover, higher p and q increase the complexity of the model, making the model more difficult to interpret. And a higher d value may lead to over-differencing and loss of long-term trend information of the data. Therefore, selecting appropriate p, d, and q values is the key to constructing the model.

[0102] To determine the optimal combination of p, d, and q parameters, the Particle Swarm Optimization (PSO) algorithm is adopted. The PSO algorithm is a heuristic algorithm that mimics the foraging behavior of bird flocks and searches for the optimal solution in the search space by continuously updating the positions and velocities of particles. Taking p, d, and q as the positions of the particles, the optimal parameter combination is found by minimizing the Akaike Information Criterion (AIC).

[0103] First, the particle swarm needs to be initialized. Each particle represents a combination of (p, d, q), where the value ranges of p, d, and q are set according to the actual situation and experience. The set ranges are p ∈ [0, 5], d ∈ [0, 2], and q ∈ [0, 5], thus obtaining the above-mentioned candidate parameter groups. The initial positions of the particles are randomly generated within these ranges, and at the same time, a random velocity is assigned to each particle.

[0104] During the iterative process of the PSO algorithm, for each combination of (p, d, q) represented by a particle, the VARMA model is used to fit the time series data, and its AIC value is calculated as the fitness of the particle, obtaining the results of the fitness function. By comparing the current fitness of each particle with its historical best fitness, the individual historical optimal position of the particle is updated; at the same time, by comparing the global optimal fitness with the current fitness of all particles, the global optimal position is updated, and the combination of (p, d, q) corresponding to this global optimal position will be the desired optimal parameter combination. Secondly, for the evaluation of fitness, the Akaike Information Criterion of VARMA can be used as the fitness of the particle, and the combination of (p, d, q) that minimizes the AIC is found with this as the goal.

[0105] More specifically, in an optional embodiment, for the results of the fitness function of the particle parameters traversed by the above algorithm, based on the results of the fitness function, calculating the global optimal position may further include the following steps:

[0106] According to the traversed particle parameters, the preset inertia weight, and the historical particle flight speed, calculate the current particle flight speed, and according to the current particle flight speed and the historical particle position, calculate the current particle position; iteratively calculate the results of the fitness function of the particle parameters under the current particle flight speed and the current particle position until the preset iteration termination condition is reached, and determine the global optimal position.

[0107] Among them, in each iteration, the update of the particle's velocity and position follows the following formula:

[0108]

[0109] In the above formula, is the velocity of particle i at time t, is the position of particle i at time t (i.e., p, d, q combination), pbest i is the personal historical best position of particle i, gbest is the global historical best position, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers in the range of [0, 1]. After updating the particle position, boundary processing needs to be performed on p, d, and q to ensure that their values do not exceed the set range.

[0110] Through multiple iterations of training, it finally converges to a global optimal position. At this time, the (p, d, q) combination is (4, 0, 7). Using this optimal parameter combination can make the VARMA model achieve a better fitting effect, avoid overfitting and underfitting problems, and improve the prediction performance of the model.

[0111] Through the above embodiments, combining the PSO algorithm with the AIC criterion, automatically search for the optimal parameters. Through swarm intelligence optimization, it avoids the subjectivity and inefficiency of traditional methods, and is especially suitable for scenarios with large amounts of cigarette production data and complex parameter spaces. Thus, it not only improves the training accuracy of the prediction model, but also provides more reliable and accurate information support for production decisions.

[0112] In some of these embodiments, the above-mentioned acquisition of historical indicator time-series data and historical process parameter time-series data may further include the following steps:

[0113] Obtain the acquired historical indicator time-series data; input the historical indicator time-series data and the obtained preset process features into the trained feature screening model, and use the feature screening model to select the target process features associated with the historical indicator time-series data from the preset process features; based on the target process features, obtain the historical process parameter time-series data.

[0114] Specifically, extract the time-series data containing key historical indicators such as moisture indicators from the production database. These data should be arranged in the order of time series for subsequent analysis. At the same time, obtain multiple possible process parameters, including temperature, pressure, flow rate, etc., a total of 12 or more. These preset process features will be used as candidate features and input into the feature screening model. In addition, a trained feature screening model needs to be prepared beforehand. This model can be based on machine learning or deep learning methods, such as the Extreme Gradient Boosting (XgBoost) algorithm, random forest, etc., to evaluate the correlation between each process feature and the target indicator. The training data of the model should include historical indicator time-series data, corresponding process parameter time-series data, and known feature labels that have a significant impact on the target indicator.

[0115] Then, the obtained historical indicator time-series data and the preset process characteristics are input into the trained feature screening model together. Using the feature screening model, target process characteristics associated with the historical indicator time-series data (such as moisture indicator time-series data) are selected from the preset process characteristics. These target characteristics should have relatively high correlation or importance scores. Based on the selected target process characteristics, the historical process parameter time-series data corresponding to these characteristics are extracted from the production database. These time-series data will be used for subsequent model training or analysis.

[0116] Through the above embodiments, the adaptive selection of historical process parameter time-series data associated with historical indicator time-series data is realized, so that in the process of generating the cigarette index threshold, the relevant process parameters that may affect the cigarette index can be automatically, efficiently and accurately determined, and input into the model together with the cigarette index for training, thus effectively improving the automation degree of model training and also being conducive to improving the model accuracy.

[0117] In some of these embodiments, the above step of inputting the actual indicator data into the constructed prediction model and outputting the predicted future data may further include the following steps:

[0118] Obtain the production condition information for the cigarette production process; when the production condition information indicates stable conditions, input the actual indicator data into the constructed prediction model for processing and output the predicted future data; when the production condition information indicates changing conditions, reconstruct a new prediction model, input the actual indicator data into the new prediction model, and output the predicted future data.

[0119] In this step, first collect the production condition information related to the cigarette production process, such as equipment status, raw material batches, production environment parameters, etc. These information are used to evaluate the stability of the current production conditions. Then analyze the production condition information to judge whether the current production is under stable conditions. Stable conditions usually mean that key factors such as production equipment, raw materials and production environment remain relatively consistent without significant changes.

[0120] If the production condition information indicates stable conditions, then under stable conditions, use the constructed prediction model (which is trained based on data under historical stable conditions) to process the actual indicator data. Input the actual indicator data into this model and output the predicted future data.

[0121] If the production condition information indicates changing conditions, such as equipment upgrades, replacement of raw material batches, or production environment adjustments, etc., then under such changing conditions, it is necessary to retrain the model parameters and construct a new prediction model. The construction process of the new model can be as follows: collect new historical data, and this data should reflect the production situation under the current changing conditions. Use the new historical data to train the model parameters to generate a new prediction model. Finally, input the actual index data into the new prediction model and output the predicted future data.

[0122] Through the above embodiments, it is possible to more flexibly respond to changes in production conditions, ensure the accuracy and applicability of the prediction model, and thus provide reliable prediction support for cigarette production.

[0123] In some of these embodiments, the above real-time calculation based on the extended dataset to generate the dynamic cigarette index threshold may further include the following steps:

[0124] Perform calculation and statistics based on the extended dataset to generate a mean control chart; determine the mean data based on the mean control chart; determine the upper threshold according to the mean data and the preset upper control limit coefficient, and determine the lower threshold according to the mean data and the preset lower control limit coefficient; the dynamic cigarette index threshold includes the upper threshold and the lower threshold.

[0125] The mean control chart is a preventive management method that can draw the standard upper and lower threshold values for historical indicators and monitor process changes. By analyzing the change trend of the data distribution center (X-bar chart) and the within-group fluctuation (R chart), and combining the multivariate characteristics of the extended dataset, the mean control chart can simultaneously monitor the process mean shift and abnormal fluctuations. For example, when the cigarette weight shows a mean drift due to equipment vibration, the mean control chart can trigger an alarm in real time to avoid batch quality problems.

[0126] The generation process of the mean control chart is described below. First, plot the mean control chart for the actual cigarette moisture values collected within the historical time period to determine the threshold values of the upper and lower limits; for example, according to the production batch, sample 5 cigarettes per minute, continuously sample 13 groups, with a total of 65 data, and this series of data is the actual index data. Secondly, record the data and calculate the average and range of each group, including the total average and total range of the 13 groups (i.e., the above mean data).

[0127] Next, calculate the upper and lower limits of its threshold in combination with the upper control limit coefficient and the lower control limit coefficient. The upper control limit coefficient is used to calculate the upper control limit threshold, and the lower control limit coefficient is used to calculate the lower control limit threshold. Specifically, the upper threshold UCL: UCL = D4 × R, and the lower threshold LCL: LCL = D3 × R. Where D3 (i.e., the above-mentioned lower control limit coefficient) and D4 (i.e., the above-mentioned upper control limit coefficient) are fixed constants, which can be obtained by looking up the table; R is the calculated total average range. In the generated mean control chart, each data point represents the average value of a sample. If a certain data point exceeds the upper and lower limits, or shows a specific pattern (such as continuous increase or decrease), it indicates that there may be a problem with the process. Based on this, a historical mean control chart is drawn, such as Figure 4 shown. It can be seen from the figure that the thresholds of the cigarette moisture content during this period are the upper limit value UCL = 18.5061 and the lower limit value LCL = 18.2429 respectively.

[0128] Moreover, combine the VARMA prediction model to predict the cigarette moisture content value at a future time and recalculate the upper and lower limit thresholds of its mean control chart. The predicted future data obtained by using the prediction model to predict the next five groups of samples is as Figure 5 shown, Figure 5 in which the solid line segment shows a series of actual index data of actual sampling, and the dashed line segment shows the predicted future data obtained based on the prediction model, specifically including five groups of future sample data of 18.23, 18.39, 18.32, 18.37, and 18.50. Then, the mean control chart statistically obtained from the actual index data and the predicted future data is as Figure 6 shown. At this time, the recalculated cigarette moisture thresholds are the upper limit UCL = 19.0 and the lower limit LCL = 17.75 respectively. Since the collection frequency of the cigarette moisture content is relatively high and the fluctuation is large, the upper threshold is about 2.6% higher than the upper threshold statistically obtained from the actual index data collected during the historical time period in Figure 4 , and the lower threshold is about 2.7% lower. By combining the predicted future moisture value of VARMA with the historical moisture value and dynamically setting the upper and lower limit thresholds of the mean control chart through predictive statistical process control, the early warning ability has been significantly improved.

[0129] Through the above embodiments, by expanding the dataset (integrating historical data and predicted data) to calculate the mean control chart in real time and dynamically adjusting the upper and lower limit thresholds in combination with preset coefficients (such as D3, D4), it can quickly respond to changes in production conditions (such as equipment wear, raw material fluctuations). Compared with the traditional fixed threshold, this method can avoid misjudgment caused by working condition drift (such as too loose or too strict rejection criteria) and significantly improve the monitoring sensitivity.

[0130] The following describes and illustrates the present application through specific embodiments. Figure 7 is a flowchart of another method for generating cigarette index thresholds according to an embodiment of the present application, such asFigure 7 As shown, the process includes the following steps:

[0131] Step S701, observe the sequence characteristics of the multivariate time series data.

[0132] Step S702, determine whether it passes the stationarity test; if so, continue to execute the subsequent step S704.

[0133] Step S703, if the judgment result of the above step S702 is no, perform difference calculation, and return to the above step S702 to continue the judgment.

[0134] Step S704, calculate the ACF and PACF.

[0135] Step S705, iteratively train the VARMA model.

[0136] Step S706, calculate the p, d, and q parameters in the estimation model; among them, use PSO to optimize the p, d, q parameter combination.

[0137] Step S707, determine whether it passes the model test. If not, return to the above step S705 to continue the identification.

[0138] Step S708, if the judgment result of the above step S707 is yes, use the above VARMA model to predict the future trend of the cigarette moisture, and finally generate the dynamic cigarette index threshold.

[0139] It should be noted that the steps shown in the above process or the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0140] This embodiment also provides a device for generating a cigarette index threshold. This device is used to implement the above embodiment and the preferred implementation manner, and those that have been described will not be repeated. As used below, terms such as "module", "unit", "sub-unit", etc. can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0141] Figure 8 is a structural block diagram of a device for generating a cigarette index threshold according to an embodiment of the present application. As Figure 8 shown, the device includes: an acquisition module 81, a prediction module 82, and a threshold generation 83 module; where:

[0142] An acquisition module 81 for acquiring actual index data for the cigarette production process; a prediction module 82 for inputting the actual index data into a constructed prediction model and outputting predicted future data; the prediction model is generated by training based on collected multivariate time series data, and the data characteristics of the multivariate time series data match the data characteristics of the actual index data; a threshold generation module 83 for obtaining an extended data set based on the actual index data and the predicted future data, and generating a dynamic cigarette index threshold by real-time calculation based on the extended data set.

[0143] In some embodiments, the above-mentioned device for generating the cigarette index threshold further includes a training module; the training module is used for collecting historical index time series data and historical process parameter time series data; the multivariate time series data includes historical index time series data and historical process parameter time series data, and the historical process parameter time series data is associated with the historical index time series data; the training module is further used for performing differential calculation on the multivariate time series data to obtain a first estimation parameter; the training module is further used for calculating the correlation function results for the historical index time series data and the historical process parameter time series data, and determining a second estimation parameter and a third estimation parameter based on the correlation function results; the training module is further used for constructing a prediction model based on the first estimation parameter, the second estimation parameter and the third estimation parameter.

[0144] It should be noted that the above-mentioned each module can be a functional module or a program module, and can be implemented either by software or by hardware. For the modules implemented by hardware, the above-mentioned each module can be located in the same processor; or the above-mentioned each module can also be located in different processors in any combined form. The specific examples in this embodiment can refer to the examples described in the above-mentioned embodiments and optional implementation manners, and will not be elaborated in this embodiment.

[0145] This embodiment also provides an electronic device, including a memory and a processor, where a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0146] Optionally, the above-mentioned electronic device may further include a transmission device and an input / output device, where the transmission device is connected to the above-mentioned processor, and the input / output device is connected to the above-mentioned processor.

[0147] Optionally, in this embodiment, the above-mentioned processor may be configured to execute the following steps through a computer program:

[0148] S1, acquiring actual index data for the cigarette production process.

[0149] S2, input the actual index data into the constructed prediction model and output the predicted future data; the prediction model is generated by training with the collected multivariate time series data, and the data characteristics of the multivariate time series data match the data characteristics of the actual index data.

[0150] S3, obtain an extended data set based on the actual index data and the predicted future data, and calculate in real time according to the extended data set to generate a dynamic cigarette index threshold.

[0151] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation manners, and will not be repeated here.

[0152] In addition, in combination with the method for generating the cigarette index threshold in the above embodiments, the embodiments of the present application can be implemented by providing a storage medium. A computer program is stored on the storage medium; when the computer program is executed by a processor, the method for generating any one of the cigarette index thresholds in the above embodiments is implemented.

[0153] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties.

[0154] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include Read-Only Memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0155] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0156] The above embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for generating a threshold value of cigarette rod indexes, characterized in that, The method includes: Obtaining actual index data for the cigarette production process; Inputting the actual index data into a constructed prediction model and outputting predicted future data; the prediction model is generated by training based on collected multivariate time series data, and the data characteristics of the multivariate time series data match the data characteristics of the actual index data; Based on the actual index data and the predicted future data, obtaining an extended data set, and calculating in real time based on the extended data set to generate a dynamic cigarette index threshold.

2. The generation method according to claim 1, wherein The method further includes: Collecting historical index time series data and historical process parameter time series data; the multivariate time series data includes the historical index time series data and the historical process parameter time series data, and the historical process parameter time series data is associated with the historical index time series data; Performing difference calculation on the multivariate time series data to obtain a first estimation parameter; Calculating the correlation function results for the historical index time series data and the historical process parameter time series data, and determining a second estimation parameter and a third estimation parameter based on the correlation function results; Based on the first estimation parameter, the second estimation parameter, and the third estimation parameter, constructing the prediction model.

3. The generation method according to claim 2, wherein The performing difference calculation on the multivariate time series data to obtain a first estimation parameter includes: Generating a trend graph and a season graph based on the multivariate time series data; Performing stationarity detection on the multivariate time series data according to the trend graph and the season graph to obtain a stationarity detection result, and performing difference calculation based on the stationarity detection result to obtain the first estimation parameter.

4. The generating method according to claim 2, wherein The constructing the prediction model based on the first estimation parameter, the second estimation parameter, and the third estimation parameter includes: Based on the first estimation parameter, the second estimation parameter, and the third estimation parameter, determining multiple candidate parameter groups; Traversing the particle parameters corresponding to each candidate parameter group; Calculating the fitness function result of the traversed particle parameters, based on the fitness function result, calculating the global optimal position, and based on the global optimal position, determining the optimal parameter combination in the candidate parameter group; Constructing the prediction model according to the optimal parameter combination.

5. The generation method according to claim 4, wherein The calculating the fitness function result of the traversed particle parameters and calculating the global optimal position based on the fitness function result includes: Calculating the current particle flight speed according to the traversed particle parameters, a preset inertia weight, and the historical particle flight speed, and calculating the current particle position according to the current particle flight speed and the historical particle position; Iteratively calculating the fitness function result of the particle parameters at the current particle flight speed and the current particle position until a preset iteration termination condition is reached, and determining the global optimal position.

6. The generation method according to claim 2, wherein The collecting historical index time series data and historical process parameter time series data includes: Obtaining the collected historical index time series data; Input the historical indicator time-series data and the obtained preset process characteristics into the trained feature screening model, and use the feature screening model to select target process characteristics associated with the historical indicator time-series data from the preset process characteristics; Obtain the historical process parameter time-series data based on the target process characteristics.

7. The generation method according to claim 1, wherein The inputting the actual indicator data into the constructed prediction model and outputting the predicted future data includes: Obtain the production condition information for the cigarette rod production process; When the production condition information indicates stable conditions, input the actual indicator data into the constructed prediction model for processing and output the predicted future data; When the production condition information indicates changing conditions, reconstruct a new prediction model, input the actual indicator data into the new prediction model, and output the predicted future data.

8. The generation method according to any one of claims 1 to 7, characterized in that The generating the dynamic cigarette rod indicator threshold according to the real-time calculation of the extended data set includes: Perform calculation and statistics based on the extended data set to generate a mean control chart; Determine the mean data based on the mean control chart; determine the upper threshold according to the mean data and a preset upper control limit coefficient, and determine the lower threshold according to the mean data and a preset lower control limit coefficient; The dynamic cigarette rod indicator threshold includes the upper threshold and the lower threshold.

9. A generating device for cigarette index thresholds, characterized in that, Includes: An acquisition module for acquiring the actual indicator data for the cigarette rod production process; A prediction module for inputting the actual indicator data into the constructed prediction model and outputting the predicted future data; the prediction model is generated by training based on the collected multi-variable time-series data, and the data characteristics of the multi-variable time-series data match the data characteristics of the actual indicator data; A threshold generation module for obtaining an extended data set based on the actual indicator data and the predicted future data, and generating a dynamic cigarette rod indicator threshold according to the real-time calculation of the extended data set.

10. A storage medium, characterized in that, A computer program is stored in the storage medium, wherein the computer program is configured to execute the method for generating the cigarette rod indicator threshold according to any one of claims 1 to 8 when running.

Images