Cigar tobacco stacking fermentation stack core temperature analysis method and device based on time sequence clustering

Abstract

The invention relates to a cigar tobacco stacking fermentation stack core temperature analysis method and device based on time sequence clustering, and the method comprises the following steps: data collection: monitoring and collecting the temperature of a stack core in a cigar tobacco stacking fermentation process; performing data processing, and performing interpolation by adopting a cubic spline interpolation method; establishing a time sequence clustering method, measuring temperature sequence similarity, and establishing a K-means + + clustering algorithm and an evaluation index; data analysis, clustering result feature analysis of the temperature data and cluster center feature curve analysis. A time sequence clustering analysis method is adopted to analyze the stack core temperature in the cigar tobacco fermentation process, a multi-dimensional temperature field stack core temperature rising model can be established according to the stack core temperature change rule, the current situation that the temperature time sequence dynamic analysis is insufficient is made up in a data driving mode, and the stability of the stack core temperature rising model is improved. Decision support can be provided for fermentation process operation such as stack turning or fermentation termination, and the fermentation process stability and the tobacco leaf homogenization level are improved.

Description

Technical Field

[0001] This invention relates to the field of tobacco processing, and in particular to a method and apparatus for analyzing the core temperature of cigar tobacco stack fermentation based on temporal clustering. Background Technology

[0002] Stacking fermentation is a key process for improving and enhancing the quality of cigar tobacco leaves. By stacking tobacco leaves according to specific sizes and methods, the synergistic effect of microbial metabolism and enzymatic reactions during stacking fermentation promotes the degradation of macromolecules and the transformation of small molecules, thereby improving physical properties, appearance, and sensory quality. However, the fermentation cycle and temperature control methods for cigar raw materials are not standardized due to factors such as the origin of the raw materials, varietal characteristics, harvest maturity, and existing fermentation levels. Current tobacco quality evaluation relies heavily on manual sensory assessment, leading to subjective biases in quality control and judgment. Existing research focuses primarily on static chemical composition detection, lacking a systematic analysis of the temporal evolution of the temperature field during dynamic stacking, particularly lacking quantitative characterization of temperature distribution characteristics under multiple stacking operations.

[0003] Therefore, providing a time-series clustering-based method for analyzing the core temperature of cigar tobacco stacks during fermentation can help establish a multi-dimensional temperature field core temperature rise model, analyze the temperature change patterns of different models, and provide guidance for operations such as turning over or terminating fermentation, thus possessing significant application value. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and apparatus for analyzing the core temperature of cigar tobacco stack fermentation based on temporal clustering.

[0005] To achieve this objective, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for analyzing the core temperature of cigar tobacco leaves during stack fermentation based on temporal clustering, the method comprising the following steps:

[0007] S1. Data Acquisition: Monitor and collect the temperature of the core of the cigar tobacco stack during the fermentation process.

[0008] S2. Data processing: Interpolation is performed using cubic spline interpolation.

[0009] S3. Establishment of time-series clustering methods: establishment of temperature series similarity measurement, K-means++ clustering algorithm and evaluation index;

[0010] S4. Data Analysis: Clustering results feature analysis and cluster center feature curve analysis of temperature data.

[0011] In this invention, the shortcomings of temperature time-series dynamic analysis are made up for by using a data-driven approach, providing decision support for improving the stability of fermentation process and the homogenization level of tobacco leaves. Furthermore, based on this method, a multi-dimensional temperature field core temperature rise model can be established to analyze the temperature change law of core under different models, which can provide real-time guidance for operations such as turning the stack or terminating fermentation.

[0012] Preferably, the time interval for data acquisition in step S1 is 1 hour per acquisition.

[0013] Preferably, the temperature sequence similarity measurement in step S3 employs a dynamic time warping algorithm, given two temperature sequences: X = (x1, x2, ..., x...). i ,...,x n ) and Y = (y1, y2, ..., y j ,...y m ),in The temperature sample value is represented by the value. The algorithm calculates the optimal path using dynamic programming, and its recursive formula is as follows:

[0014] D(i,j)=d(x i ,y j )+min{D(i-1,j),D(i,j-1),D(i-1,j-1)} Formula (1);

[0015] In the formula, d(x) i ,y j ) is x i and y j The distance metric function between them is given by D, which represents the cumulative distance matrix, and D(i,j) which represents the minimum distance required to align the first i points of the first time series with the first j points of the second time series. The boundary conditions are as follows:

[0016] D(1,1)=d(x1,y1) Equation (2);

[0017]

[0018] The final dynamic time-warped distance is D DTW (X,Y)=D(n,m).

[0019] Preferably, the K-means++ clustering algorithm in step S3 specifically includes the following steps:

[0020] (1) Initial centroid selection: Randomly select a sample as the first initial cluster centroid μ1, and each sample X i Calculate its relationship with the current centroid set C = {μ1, μ2, ..., μ}. mThe squared dynamic time-warped distance of the nearest centroid in the matrix is ​​calculated, and the next centroid is selected according to the following probability formula. This process is repeated until K initial centroids are selected:

[0021]

[0022] In the formula, p i This represents the probability of choosing the initial centroid. Represents each sample X i The squared dynamic time-warped distance to the nearest centroid set C is This represents the sum of the squared dynamic time-normalized distances from all samples to their nearest centroids;

[0023] (2) Iterative optimization: For each sample X i Assign to the nearest cluster C k The division rules are as follows:

[0024]

[0025] In the formula, μ j Let D represent the j-th centroid. DTW (X i ,μ j ) is the sample X i With the center of mass μ j The dynamic time warp distance;

[0026] From cluster C k The process involves selecting a time series sample and iterating until the centroid no longer changes significantly or the maximum number of iterations is reached. The calculation formula is as follows:

[0027]

[0028] In the formula, X represents the candidate centroid, and C k Let the set of the k-th cluster be represented. This indicates the centroid after optimization and update.

[0029] Preferably, the evaluation index established in step S3 is based on the sum of squared errors within clusters based on dynamic time-warped distance. The compactness of clustering is measured by quantifying the dynamic time-warped distance between samples within a cluster and the cluster centroid. The calculation formula is as follows:

[0030]

[0031] In the formula, SSE DTW C represents the sum of squared errors within a cluster representing the dynamic time-warped distance. k Let μ represent the set of all time series samples in the k-th cluster. k Let D represent the centroid of the k-th cluster. DTW (X i,μ k ) is the sample X i With the center of mass μ k The dynamic time warp distance.

[0032] Preferably, the data analysis in step S4 further includes short-term dynamic change analysis of temperature within a sliding 24-hour window.

[0033] Preferably, the short-term dynamic change analysis of temperature within the sliding 24-hour window includes the following steps:

[0034] (1) Test and calculate the temperature difference change over 24 hours of sliding. The calculation formula is as follows:

[0035] ΔT i =x i -x i-24 , (i=25,26,....,n) Formula (9);

[0036] In the formula, ΔT i Indicates the temperature difference over 24 hours, x i Let x represent the i-th sample point in the temperature sequence. i-24 This represents the temperature value of the i-th temperature 24 hours ago;

[0037] (2) Calculate the average, maximum, and minimum temperature differences over 24 hours of sliding, using the following formulas:

[0038]

[0039] In the formula, ΔT mean ΔT represents the average temperature difference over a 24-hour sliding period. max ΔT represents the maximum temperature difference over a 24-hour sliding period. min This represents the minimum temperature difference over a 24-hour sliding period, where n represents the total number of temperature data points.

[0040] (3) Perform Simpson numerical integration on the data of temperature difference change over a 24-hour sliding period. The calculation formula is as follows:

[0041]

[0042] In the formula, I is the integral value of temperature, and Δt is the temperature sampling interval.

[0043] Preferably, the data analysis in step S4 specifically involves: calculating the statistical indicators of the temperature curves of each cluster, taking the average value to assess the differences between clusters, dividing the temperature change of the cigar tobacco stack during the fermentation process into different modes, combining the characteristic temperature curve of the cluster center and the sliding 24-hour temperature difference analysis, quantifying the dynamic characteristics of each mode, and analyzing the temperature change of the cigar tobacco stack during fermentation.

[0044] In a second aspect, the present invention provides a temperature analysis device for cigar tobacco stack fermentation core based on temporal clustering, the temperature analysis device being used to perform the temperature analysis method described in the first aspect.

[0045] Preferably, the device includes:

[0046] a. A data acquisition module monitors and collects the temperature of the core of the cigar tobacco stack during the fermentation process; the data acquisition time interval is 1 hour / time;

[0047] b. The data processing module uses cubic spline interpolation for interpolation;

[0048] c. Module for establishing time-series clustering methods, including the establishment of temperature sequence similarity measurement, K-means++ clustering algorithm, and evaluation metrics;

[0049] The temperature sequence similarity measure employs a dynamic time warping algorithm, given two temperature sequences: X = (x1, x2, ..., x...). i ,...,x n ) and Y = (y1, y2, ..., y j ,...y m ),in The temperature sample value is represented by the value. The algorithm calculates the optimal path using dynamic programming, and its recursive formula is as follows:

[0050] D(i,j)=d(x i ,y j )+min{D(i-1,j),D(i,j-1),D(i-1,j-1)} Formula (1);

[0051] In the formula, d(x) i ,y j ) is x i and y j The distance metric function between them is given by D, which represents the cumulative distance matrix, and D(i,j) which represents the minimum distance required to align the first i points of the first time series with the first j points of the second time series. The boundary conditions are as follows:

[0052] D(1,1)=d(x1,y1) Equation (2);

[0053]

[0054] The final dynamic time-warped distance is D DTW (X,Y)=D(n,m).

[0055] The K-means++ clustering algorithm specifically includes the following steps:

[0056] (1) Initial centroid selection: Randomly select a sample as the first initial cluster centroid μ1, and each sample X i Calculate its relationship with the current centroid set C = {μ1, μ2, ..., μ}. m The squared dynamic time-warped distance of the nearest centroid in the matrix is ​​calculated, and the next centroid is selected according to the following probability formula. This process is repeated until K initial centroids are selected:

[0057]

[0058] In the formula, p i This represents the probability of choosing the initial centroid. Represents each sample X i The squared dynamic time-warped distance to the nearest centroid set C is This represents the sum of the squared dynamic time-normalized distances from all samples to their nearest centroids;

[0059] (2) Iterative optimization: For each sample X i Assign to the nearest cluster C k The division rules are as follows:

[0060]

[0061] In the formula, μ j Let D represent the j-th centroid. DTW (X i ,μ j ) is the sample X i With the center of mass μ j The dynamic time warp distance;

[0062] From cluster C k The process involves selecting a time series sample and iterating until the centroid no longer changes significantly or the maximum number of iterations is reached. The calculation formula is as follows:

[0063]

[0064] In the formula, X represents the candidate centroid, and C k Let the set of the k-th cluster be represented. This indicates the centroid after optimization and update.

[0065] The evaluation metric is established based on the sum of squared errors within clusters based on dynamic time-warped distance. The compactness of the clustering is measured by quantifying the dynamic time-warped distance between samples within a cluster and the cluster centroid. The calculation formula is as follows:

[0066]

[0067] In the formula, SSE DTW C represents the sum of squared errors within a cluster representing the dynamic time-warped distance.k Let μ represent the set of all time series samples in the k-th cluster. k Let D represent the centroid of the k-th cluster. DTW (X i ,μ k ) is the sample X i With the center of mass μ k The dynamic time warp distance.

[0068] d. Data analysis module: clustering result feature analysis and cluster center characteristic curve analysis of temperature data;

[0069] The data analysis also includes short-term dynamic change analysis of temperature within a sliding 24-hour window.

[0070] Preferably, the short-term dynamic change analysis of temperature within the sliding 24-hour window includes the following steps:

[0071] (1) Test and calculate the temperature difference change over 24 hours of sliding. The calculation formula is as follows:

[0072] ΔT i =x i -x i-24 , (i=25,26,....,n) Formula (9);

[0073] In the formula, ΔT i Indicates the temperature difference over 24 hours, x i Let x represent the i-th sample point in the temperature sequence. i-24 This represents the temperature value of the i-th temperature 24 hours ago;

[0074] (2) Calculate the average, maximum, and minimum temperature differences over 24 hours of sliding, using the following formulas:

[0075]

[0076] In the formula, ΔT mean ΔT represents the average temperature difference over a 24-hour sliding period. max ΔT represents the maximum temperature difference over a 24-hour sliding period. min This represents the minimum temperature difference over a 24-hour sliding period, where n represents the total number of temperature data points.

[0077] (3) Perform Simpson numerical integration on the data of temperature difference change over a 24-hour sliding period. The calculation formula is as follows:

[0078]

[0079] In the formula, I is the integral value of temperature, and Δt is the temperature sampling interval;

[0080] The data analysis specifically involves: calculating the statistical indicators of the temperature curves of each cluster, taking the average value to assess the differences between clusters, dividing the temperature change of the cigar tobacco stack during the fermentation process into different modes, combining the characteristic temperature curves of the cluster center and the sliding 24-hour temperature difference analysis, quantifying the dynamic characteristics of each mode, and analyzing the temperature change of the cigar tobacco stack during fermentation.

[0081] Thirdly, the present invention provides a computer-readable storage medium storing a computer program or instructions thereon, the computer program or instructions being used to cause a processor to execute and implement the temperature analysis method as described in the first aspect.

[0082] Compared with the prior art, the present invention has the following beneficial effects:

[0083] The temperature analysis method provided by this invention can establish a multi-dimensional temperature field core temperature rise model, analyze the temperature change patterns of cores under different models, and provide real-time guidance for operations such as turning the stack or terminating fermentation. Furthermore, compared to traditional methods relying on subjective experience and static analysis, the time-series clustering-based core temperature analysis method for cigar tobacco stack fermentation provided by this invention compensates for the shortcomings of dynamic time-series temperature analysis in a data-driven manner, providing data-driven decision support for improving fermentation process stability and tobacco homogenization. Attached Figure Description

[0084] Figure 1 It is a graph showing the intra-cluster squared error and its variation based on the dynamic time warping distance;

[0085] Figure 2 This is a distribution diagram of the number of clusters in the AE stacking fermentation process;

[0086] Figure 3 This is a time-series curve of the stacked fermentation temperature of cluster A;

[0087] Figure 4 This is a time-series curve of the stacked fermentation temperature of cluster B;

[0088] Figure 5 This is a time-series curve of the stacking fermentation temperature of cluster C;

[0089] Figure 6 This is a time-series curve of the stacking fermentation temperature of cluster D;

[0090] Figure 7 This is a time-series curve of the stacking fermentation temperature of cluster E;

[0091] Figure 8 This is a graph showing the characteristic temperature curve of cluster A center and the sliding 24-hour temperature difference curve;

[0092] Figure 9This is a graph showing the characteristic temperature curve of cluster B center and the sliding 24-hour temperature difference curve;

[0093] Figure 10 This is a graph showing the characteristic temperature curve of cluster C center and the sliding 24-hour temperature difference curve;

[0094] Figure 11 This is a graph showing the characteristic temperature curve of cluster D center and the sliding 24-hour temperature difference curve;

[0095] Figure 12 It is a graph showing the characteristic temperature curve of cluster E center and the sliding 24-hour temperature difference curve. Detailed Implementation

[0096] The technical solution of the present invention will be further illustrated below through specific embodiments. Those skilled in the art should understand that the embodiments described are merely illustrative of the present invention and should not be construed as limiting the invention in any way.

[0097] Example

[0098] This embodiment provides a method for analyzing the core temperature of cigar tobacco leaves during stack fermentation based on temporal clustering, as detailed below:

[0099] S1, Data Acquisition

[0100] Cigar filler tobacco leaves from Hainan, Sichuan, Yunnan, and Hubei provinces were selected as the research subjects. The tobacco leaves were stacked in piles measuring 1.5m × 1.5m × 1.5m (length × width × height), with a single pile weighing 600kg. A temperature sensor probe was placed at the center of the tobacco pile to monitor the temperature changes in the core of the pile during fermentation. Sampling intervals were 1 hour, and the specific time of each pile turning was recorded. The experimental environmental conditions were set at a temperature of 27℃ and a relative humidity of 70%.

[0101] S2, Data Processing:

[0102] Cubic spline interpolation is used to interpolate time series data.

[0103] S3, Establishment of Temporal Clustering Method

[0104] ① Given two temperature sequences: X=(x1,x2,...,x i ,...,x n ) and Y = (y1, y2, ..., y j ,...y m ),in This represents the temperature sample value. The algorithm calculates the optimal path using dynamic programming, and its recursive formula is:

[0105] D(i,j)=d(x i ,y j)+min{D(i-1,j),D(i,j-1),D(i-1,j-1)}

[0106] Where d(x) i ,y j ) is x i and y j The distance metric function between them is given by D, which represents the cumulative distance matrix, and D(i,j) which represents the minimum distance required to align the first i points of the first time series with the first j points of the second time series. The boundary conditions are as follows:

[0107] D(1,1)=d(x1,y1)

[0108]

[0109] The final dynamic time warping (DTW) distance is D. DTW (X,Y)=D(n,m).

[0110] ② Initial centroid selection: Randomly select a sample as the first initial cluster centroid μ1, and each sample X i Calculate its relationship with the current centroid set C = {μ1, μ2, ..., μ}. m}, calculate the squared DTW distance of the nearest centroid, and select the next centroid according to the following probability formula, repeating until K initial centroids are selected:

[0111]

[0112] Represents each sample X i The squared DTW distance to the nearest centroid set C is This represents the sum of the squared DTW distances from all samples to their nearest centroids.

[0113] Iterative optimization: For each time series X i Assign to the nearest cluster C k The division rules are as follows:

[0114]

[0115] In the formula, μ j Let D represent the j-th centroid. DTW (X i ,μ j ) is the sample X i With the center of mass μ j DTW distance.

[0116] The new centroid is calculated using the DBA algorithm from cluster C. kThe process involves selecting a time series sample and iterating until the centroid no longer changes significantly or the maximum number of iterations is reached. The calculation method is as follows:

[0117]

[0118] In the formula, X represents the candidate centroid, and C k Let the set of the k-th cluster be represented. This indicates the updated centroid.

[0119] ③ Evaluation Metric: Define the sum of squared errors within clusters (SSE) based on DTW distance. This quantifies the DTW distance between samples within a cluster and the cluster centroid, measuring the compactness of the cluster. The calculation formula is as follows:

[0120]

[0121] In the formula, C k Let μ represent the set of all time series samples in the k-th cluster. k The centroid of the k-th cluster, D DTW (X i ,μ k ) is a temperature sequence X i With the center of mass μ k DTW distance.

[0122] To quantify the short-term dynamic changes in temperature during fermentation, a method based on a sliding 24-hour window is proposed to calculate the temperature difference change ΔT over the sliding 24 hours. i Temperature difference changes are continuously captured in hourly increments to ensure high temporal resolution and avoid missing fluctuation information during fixed-time-period analysis. The calculation formula is as follows:

[0123] ΔT i =x i -x i-24 (i = 25, 26, ..., n)

[0124] Based on this, the average temperature difference ΔT over a 24-hour sliding period was calculated. mean Maximum value ΔT max and minimum value ΔT min These represent the overall trend and extreme fluctuation characteristics of temperature difference, respectively. The calculation formulas for the above indicators are as follows:

[0125]

[0126] Where n represents the total number of temperature data points.

[0127] Simpson's rule is used for numerical integration to calculate the integral value of temperature changing with time. The formula is as follows:

[0128]

[0129] Where I is the integral value of temperature, and Δt is the temperature sampling interval.

[0130] S4, Data Analysis

[0131] ① The algorithm parameters are determined.

[0132] The optimal number of clusters was determined using the elbow rule and combined with production experience. The curve showing the sum of squared errors within clusters (SSE) calculated based on DTW distance as a function of the number of clusters is shown below. Figure 1 As shown in the figure, the monotonically decreasing trend conforms to the basic laws of cluster analysis. As the k value increases from 2 to 8, the SSE gradually decreases, indicating that increasing the number of clusters can significantly improve the similarity of temperature sequences within clusters. A significant inflection point appears when k = 5; further increasing the k value significantly slows the downward trend of SSE, and the marginal benefit decreases substantially. Considering the diversity and complexity of fermentation temperature changes in actual production, dividing the data into 5 categories can reveal the variation patterns of different stack fermentation temperatures.

[0133] Other parameter settings are as follows: initial centroid selection is 8 runs, maximum number of iterations in the clustering process is 50, maximum number of iterations for centroid update is 25, and convergence threshold is 1e-6.

[0134] ② Cluster analysis of the stacking fermentation process

[0135] To investigate the variation patterns of tobacco stack temperature curves during multiple stack turning processes, cluster analysis was performed on the temperature sequences of each stack. First, a similarity metric matrix was constructed based on DTW (Digital Transformer-Warshall), and then the K-means++ algorithm was used to cluster this matrix, ultimately identifying five typical fermentation temperature variation patterns. To assess the differences between categories, statistical indices were calculated for all tobacco stack temperature curves in each cluster, and the average value was used as the representative evaluation index for that cluster. Specific results are shown in Table 1, where the distribution of the number of fermentation processes in each category of tobacco stacks is as follows: Figure 2 As shown, all fermentation temperature curves for each cluster type were plotted, and the curve closest to the centroid in each cluster type was selected as the cluster center characteristic curve, as shown. Figure 3-7 As shown.

[0136] Table 1

[0137] Clusters A B C D E Fermentation time (h) 189.8 185.6 196.7 163.8 193 Initial temperature (°C) 24.9 26.7 27.1 34.2 31.9 Outbound temperature (°C) 26.5 29.4 38.9 38.7 30.7 Range (°C) 2.0 3.2 12.4 6.3 3.7 Variance (°C) 0.5 1.3 18.1 4.1 1.4 <![CDATA[ΔT max (℃)]]> 0.635 1.160 4.194 2.684 0.915 <![CDATA[ΔT min (℃)]]> -0.180 -0.160 0.025 -0.938 -0.815 <![CDATA[ΔT mean (℃)]]> 0.256 0.327 1.673 0.965 -0.003 I(℃·h) 4913.2 5193.3 6592.7 5922.3 5930.0

[0138] Based on the data in the table and the results in the graph, we can conclude that:

[0139] Cluster A: The average initial stacking temperature of this cluster is 24.9℃, the outgoing temperature is 26.5℃, the temperature range is 2.0℃, and the variance is 0.5℃. 2These values ​​are the lowest among all cluster types, indicating small temperature variations and high stability. The numerical temperature integral is 4913.2℃·h, indicating relatively low heat accumulation. ΔT mean The value is 0.256℃, indicating that the average temperature change over 24 hours is extremely small; ΔT max The value is 0.635℃, indicating a relatively low upper limit for temperature difference fluctuations. For example... Figure 3 As shown, the temperature curve of this cluster is approximately horizontal with minimal fluctuation. The stacking fermentation process mainly occurs during the first stacking, followed by the second and third stacking. This distribution indicates that this temperature pattern is most common in the early stages of fermentation, with a small number involving the fourth and fifth stackings, which is related to the abnormal smokestacks in the data that exhibit sustained low temperatures and no temperature rise. Overall, this cluster exhibits a low-temperature stable pattern with weak temperature increases over a relatively long period. From a biological perspective, this pattern is inferred to be related to the microbial community not being fully activated and having weak metabolic activity, resulting in limited heat release.

[0140] Cluster B: The initial temperature of this cluster was 26.7℃, and the temperature rose to 29.4℃ upon leaving the warehouse. The range was 3.2℃, and the variance was 1.3℃. 2 The reading shows a slow upward trend in temperature. The numerical temperature integral is 5193.3℃·h, indicating a slight increase in heat accumulation compared to cluster A. ΔT mean It is 0.327℃, slightly higher than cluster A; ΔT max The value is 1.160℃, indicating that the temperature fluctuation has increased. For example... Figure 4 As shown, the temperature curve for this cluster is generally steeper than that for cluster A. The temperature initially rises slowly, reflecting the gradual activation of the microbial community, increased metabolic activity, and more significant heat release compared to cluster A. Although the temperature increase is still relatively small, the improvement compared to cluster A suggests that the fermentation process is transitioning from a low-temperature stable state to an active state.

[0141] Cluster C: The initial temperature of this cluster was 27.1℃, and the temperature upon leaving the warehouse reached as high as 38.9℃. The range was 12.4℃, and the variance was 18.1℃. 2 These values ​​are the highest among all cluster types, indicating significant temperature fluctuations. The numerical temperature integral is 6592.7℃·h, indicating a substantial increase in heat accumulation. ΔT mean The temperature was 1.673℃, much higher than other clusters; ΔT max The temperature was 4.194℃, indicating significant temperature fluctuations; ΔT min The value is 0.025℃, indicating no obvious downward trend. (For example...) Figure 5 As shown, the temperature curve of this cluster exhibits a significantly faster rate of warming and a steeper slope, with a higher proportion of secondary, tertiary, and quaternary stacking. This cluster demonstrates a rapid warming pattern, with the rapid decomposition of organic matter within the tobacco leaves leading to a large release of heat. The drastic temperature fluctuations and high warming amplitude are related to the vigorous reproduction of the microbial community and the rapid transformation of chemical components.

[0142] Cluster D: The initial temperature of this cluster was 34.2℃, and the outlet temperature was 38.7℃, the highest among all clusters. The range decreased to 6.3℃, and the variance was 4.1℃. 2 The value is second only to cluster C, indicating that the temperature fluctuates within the high-temperature range and the warming trend gradually slows down. The numerical temperature integral is 5922.3℃·h, reflecting the heat accumulation in the high-temperature range. ΔT mean The value is 0.965℃, indicating that the temperature difference is still relatively large, while ΔT min A negative value indicates that the temperature decreased during the fermentation process. For example... Figure 6 As shown, the temperature curve of this cluster tends to stabilize or slightly decrease after the temperature rises, mainly due to the fourth and fifth stacking processes, concentrated in the later stage of fermentation. This cluster exhibits a high-temperature stabilization pattern. The reasons for the high initial temperature may include incomplete stacking operations leading to local heat accumulation, rapid recovery of microbial metabolism after stacking, or insufficient heat dissipation in the fermentation chamber under high summer temperatures.

[0143] Cluster E: The initial temperature of this cluster was 31.9℃, and the temperature dropped to 30.7℃ upon leaving the warehouse. The range was 3.7℃, and the variance was 1.4℃. 2 This indicates that the temperature is stabilizing or decreasing. The numerical temperature integral is 5930.0℃·h, and the heat accumulation is close to cluster D. ΔT mean The value is -0.003℃, close to zero, reflecting an extremely small temperature difference; ΔT max The temperature is 0.915℃, with a limited fluctuation range; ΔT min The value is -0.815℃, indicating a clear downward trend in the temperature difference fluctuation. For example... Figure 7 As shown, the temperature curve of this cluster gradually decreased in fluctuation, showing a steady downward trend. It mainly consisted of three, four, and five stacking processes, with no first or second stacking processes. This cluster exhibited a cooling and decay pattern, with the temperature gradually decreasing from a higher value. The negative mean temperature difference indicates a significant reduction in microbial metabolism, indicating that the fermentable substances inside the tobacco leaves were nearly exhausted, and the fermentation activity tended to be calm.

[0144] Cluster analysis revealed five typical patterns of core temperature changes during tobacco stack fermentation. Cluster A exhibited a low-temperature stability pattern with minimal temperature fluctuations. Cluster B showed a gradual temperature increase, indicating a slow heating pattern. Cluster C exhibited a rapid temperature increase, indicating a rapid heating pattern. Cluster D tended to stabilize at high temperatures, indicating a high-temperature stability pattern. Cluster E showed a gradual temperature decrease, indicating a cooling and decay pattern. These characteristics reflect the diversity of temperature dynamics during fermentation, providing a basis for process optimization.

[0145] ③ Based on the clustering results, analyze the central characteristic curve of each cluster, i.e., the core temperature curve closest to the centroid, and plot the central characteristic temperature curve of each cluster with temperature change and sliding 24-hour temperature difference curves, such as... Figure 8-12 As shown.

[0146] As can be seen from the figure:

[0147] like Figure 8 As shown, the characteristic curve of cluster A shows a fermentation time of 193 hours, with the temperature slowly rising from 25.1℃ to 26.3℃. The overall trend of the curve is extremely flat, with no obvious peaks and minimal fluctuations. The maximum temperature difference over the 24-hour sliding period is 0.3℃, and the minimum is 0℃, with minimal fluctuations. This low-temperature stability mode indicates low microbial activity and limited heat release, suggesting that the microbial community is in the initial stage of establishment, or that fermentation is being limited by initial conditions. To address this low-temperature stability characteristic, moisture can be added in the early stages of fermentation to promote microbial activity and prevent the fermentation process from becoming too slow.

[0148] like Figure 9 As shown, the central characteristic curve of cluster B indicates a fermentation time of 193 hours, with the temperature rising from 26.5℃ to 29.7℃. The curve shows a slow upward trend, with an average temperature difference of 0.415℃ over a 24-hour sliding period, a maximum of 0.6℃, and a minimum of 0.2℃. The temperature fluctuation is slightly higher than that of cluster A. This model reflects a gradual increase in microbial activity, a transition to the active fermentation phase, and a gradual increase in heat release.

[0149] like Figure 10 As shown, the central characteristic curve of cluster C indicates a fermentation time of 192 hours. The temperature rapidly increased from 24.1℃ to 38.4℃, with a sliding 24-hour average temperature difference of 1.704℃. The temperature dropped rapidly from 3.8℃ at the beginning of fermentation (24 hours) to 2.5℃ at 30 hours, then fluctuated between 2℃ and 2.6℃, falling below 2℃ after 122 hours. The temperature difference reached 0℃ at 190 hours and remained there until the end of fermentation. The temperature curve shows a rapid temperature rise followed by a stabilization, corresponding to the peak of microbial reproduction, significant heat release, and intense fermentation activity.

[0150] like Figure 11 As shown, the central characteristic curve of cluster D indicates a fermentation time of 169 hours, with the temperature rising from 32.6℃ to 39.7℃. The average temperature difference over a 24-hour sliding period was 1.037℃. After decreasing to 0.3℃ at 120 hours, the temperature level plateaued, reaching 0℃ at 162 hours, at which point turning the stack could be considered. This pattern exhibits a high-temperature stabilization characteristic, indicating that fermentation activity gradually weakens and the temperature stabilizes at a relatively high level.

[0151] like Figure 12 As shown, the central characteristic curve of cluster E indicates a fermentation time of 217 hours, with the temperature decreasing from 31.1℃ to 29℃. Within the initial 30 hours, the temperature rose from 31.1℃ to 31.5℃ and then dropped to 31.2℃, experiencing a slight increase before a continuous decline. The average temperature difference over the 24-hour sliding period was -0.287℃, initially 0.3℃, and remained negative after 30 hours, indicating a continuous cooling trend. This signifies the near end of fermentation and a significant reduction in microbial metabolic activity. In production, if the monitored temperature difference remains consistently negative, premature cessation of fermentation can be considered to save process time and avoid over-fermentation.

[0152] Cluster center characteristic curve analysis revealed the temporal dynamics of temperature changes in each cluster. Cluster A exhibited low temperature and minimal fluctuations, reflecting a low-temperature stability pattern and indicating that microbial activity had not yet fully commenced. Cluster B showed a slow temperature increase, indicating a gradual warming pattern and signifying the gradual activation of fermentation. Cluster C experienced a rapid temperature increase followed by stabilization, reflecting a rapid warming pattern and corresponding to the peak of microbial reproduction. Cluster D's temperature leveled off within the high-temperature range, exhibiting a high-temperature stability pattern and indicating weakened fermentation activity. Cluster E showed a continuous temperature decrease, reflecting a cooling and decay pattern, indicating that fermentation was nearing its end. These characteristics were quantified using temperature curves and a sliding 24-hour temperature difference, providing a reference for fermentation process regulation.

[0153] The applicant declares that the technical solution of this invention is illustrated by the above embodiments, but this invention is not limited to the above embodiments, that is, it does not mean that this invention must rely on the above embodiments to be implemented. Those skilled in the art should understand that any improvements to this invention, equivalent substitutions of raw materials for the products of this invention, addition of auxiliary components, selection of specific methods, etc., all fall within the protection scope and disclosure scope of this invention.

[0154] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0155] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable way without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A method for analyzing the core temperature of cigar tobacco leaves during stack fermentation based on temporal clustering, characterized in that, The method includes the following steps: S1. Data Acquisition: Monitor and collect the temperature of the core of the cigar tobacco stack during the fermentation process. S2. Data processing: Interpolation is performed using cubic spline interpolation. S3. Establishment of time-series clustering methods: establishment of temperature series similarity measurement, K-means++ clustering algorithm and evaluation index; S4. Data Analysis: Clustering results feature analysis and cluster center feature curve analysis of temperature data.

2. The analytical method according to claim 1, characterized in that, The data collection time interval in step S1 is 1 hour per time.

3. The analytical method according to claim 1, characterized in that, Step S3 describes a temperature sequence similarity measurement using a dynamic time warping algorithm, given two temperature sequences: X = (x1, x2, ..., x...). i ,...,x n ) and Y = (y1, y2, ..., y j ,...y m ),in The temperature sample value is represented by the value. The algorithm calculates the optimal path using dynamic programming, and its recursive formula is as follows: D(i,j) = d(x i , y j ) + min{D(i - 1, j), D(i, j - 1), D(i - 1, j - 1)} Equation (1); In the formula, d(x) i ,y j ) is x i and y j The distance metric function between them is given by D, which represents the cumulative distance matrix, and D(i,j) which represents the minimum distance required to align the first i points of the first time series with the first j points of the second time series. The boundary conditions are as follows: D(1,1)=d(x1,y1) Equation (2); The final dynamic time-warped distance is D DTW (X,Y)=D(n,m).

4. The analytical method according to claim 1, characterized in that, Step S3 of the K-means++ clustering algorithm specifically includes the following steps: (1) Initial centroid selection: Randomly select a sample as the first initial cluster centroid μ1, and each sample X i Calculate its relationship with the current centroid set C = {μ1, μ2, ..., μ}. m The squared dynamic time-warped distance of the nearest centroid in the matrix is ​​calculated, and the next centroid is selected according to the following probability formula. This process is repeated until K initial centroids are selected: In the formula, p i This represents the probability of choosing the initial centroid. Represents each sample X i The squared dynamic time-warped distance to the nearest centroid set C is This represents the sum of the squared dynamic time-normalized distances from all samples to their nearest centroids; (2) Iterative optimization: For each sample X i Assign to the nearest cluster C k The division rules are as follows: In the formula, μ j Let D represent the j-th centroid. DTW (X i ,μ j ) is the sample X i With the center of mass μ j The dynamic time warp distance; From cluster C k The process involves selecting a time series sample and iterating until the centroid no longer changes significantly or the maximum number of iterations is reached. The calculation formula is as follows: In the formula, X represents the candidate centroid, and C k Let the set of the k-th cluster be represented. This indicates the centroid after optimization and update.

5. The analytical method according to claim 1, characterized in that, The evaluation metric established in step S3 is based on the sum of squared errors within clusters according to the dynamic time-warped distance. The compactness of the clustering is measured by quantifying the dynamic time-warped distance between the samples within the cluster and the cluster centroid. The calculation formula is as follows: In the formula, SSE DTW C represents the sum of squared errors within a cluster representing the dynamic time-warped distance. k Let μ represent the set of all time series samples in the k-th cluster. k Let D represent the centroid of the k-th cluster. DTW (X i ,μ k ) is the sample X i With the center of mass μ k The dynamic time warp distance.

6. The analytical method according to claim 1, characterized in that, The data analysis in step S4 also includes short-term dynamic change analysis of temperature within a sliding 24-hour window; Preferably, the short-term dynamic change analysis of temperature within the sliding 24-hour window includes the following steps: (1) Test and calculate the temperature difference change over 24 hours of sliding. The calculation formula is as follows: ΔT i = x i - x i-24 , (i = 25, 26,...., n) Equation (9); In the formula, ΔT i Indicates the temperature difference over 24 hours, x i Let x represent the i-th sample point in the temperature sequence. i-24 This represents the temperature value of the i-th temperature 24 hours ago; (2) Calculate the average, maximum, and minimum temperature differences over 24 hours of sliding, using the following formulas: In the formula, ΔT mean ΔT represents the average temperature difference over a 24-hour sliding period. max ΔT represents the maximum temperature difference over a 24-hour sliding period. min This represents the minimum temperature difference over a 24-hour sliding period, where n represents the total number of temperature data points. (3) Perform Simpson numerical integration on the data of temperature difference change over a 24-hour sliding period. The calculation formula is as follows: In the formula, I is the integral value of temperature, and Δt is the temperature sampling interval.

7. The analytical method according to claim 6, characterized in that, The data analysis in step S4 specifically involves: calculating the statistical indicators of the temperature curves of each cluster, taking the average value to assess the differences between clusters, dividing the temperature change of the cigar tobacco stack during fermentation into different modes, combining the characteristic temperature curves of the cluster center and the sliding 24-hour temperature difference analysis, quantifying the dynamic characteristics of each mode, and analyzing the temperature change of the cigar tobacco stack during fermentation.

8. A device for analyzing the temperature of cigar tobacco stack fermentation core based on temporal clustering, characterized in that, The temperature analysis device is used to perform the temperature analysis method according to any one of claims 1-7.

9. The apparatus according to claim 8, characterized in that, The device includes: The data acquisition module monitors and collects the temperature of the core of the cigar tobacco stack during the fermentation process. The data processing module uses cubic spline interpolation for interpolation. The module for establishing time-series clustering methods includes the establishment of temperature sequence similarity measurement, K-means++ clustering algorithm, and evaluation metrics. The data analysis module includes clustering result feature analysis and cluster center feature curve analysis of temperature data.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program or instructions that are used to cause a processor to execute the temperature analysis method as described in any one of claims 1-7.

Images