Cigarette production quality monitoring method, device, equipment, storage medium and program product

Abstract

The invention relates to a cigarette production quality monitoring method, device and equipment, a readable storage medium and a program product, and relates to the technical field of cigarette production. The method comprises the following steps: acquiring historical production data in a cigarette production process and corresponding cigarette quality standard deviation data; constructing an influence association model according to the historical production data and the cigarette quality standard deviation data; the influence association model represents the influence degree between each piece of historical production data and each piece of cigarette quality standard deviation data; according to the influence correlation model, constructing a quality deviation standard attribution analysis model and a quality deviation standard prediction model; obtaining a target influence factor according to the quality offset standard attribution analysis model; and processing the current production data according to the quality standard deviation prediction model to obtain a quality standard deviation prediction result, and taking the target influence factor and the quality standard deviation prediction result as cigarette production quality monitoring results. By adopting the method, the quality monitoring accuracy can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cigarette production, and in particular to a cigarette production quality monitoring method, device, equipment, readable storage medium and program product. BACKGROUND

[0002] In the process of cigarette production, the quality of the product is affected by various factors. The influence of these factors on the quality not only reduces the experience of consumers, but also leads to resource waste in the production process and increases the difficulty of quality control. In related technologies, the quality control of the cigarette production process mainly relies on manual monitoring and basic statistical analysis methods. In the production process, relevant parameters are monitored by sensors or manual sampling, and then statistical methods such as regression analysis or control charts are used to analyze the relationship between these relevant parameters and quality marks by production managers, so as to adjust the production conditions to ensure that the product quality meets the standards.

[0003] However, the cigarette quality monitoring method in related technologies lacks sufficient precision due to manual monitoring and statistical analysis methods, and cannot comprehensively and accurately identify the complex interaction effects between various production parameters, resulting in low precision of quality control. SUMMARY

[0004] Therefore, it is necessary to provide a cigarette production quality monitoring method, device, computer equipment, computer readable storage medium and computer program product capable of improving monitoring precision to solve the above technical problems.

[0005] In a first aspect, the present application provides a cigarette production quality monitoring method, comprising:

[0006] obtaining historical production data and corresponding cigarette quality mark data in the process of cigarette production;

[0007] constructing an influence correlation model according to the historical production data and the cigarette quality mark data; the influence correlation model represents the influence degree between each historical production data and each cigarette quality mark data;

[0008] constructing a quality mark attribution analysis model and a quality mark prediction model according to the influence correlation model;

[0009] obtaining a target influencing factor according to the quality mark attribution analysis model;

[0010] processing the current production data according to the quality mark prediction model to obtain a quality mark prediction result, and taking the target influencing factor and the quality mark prediction result as the cigarette production quality monitoring result.

[0011] In one of the embodiments, the constructing an influence correlation model according to the historical production data and the cigarette quality bias data comprises:

[0012] performing feature extraction processing on the historical production data to obtain production feature data;

[0013] constructing a correlation matrix between each production feature data and each cigarette quality bias data according to the correlation between the production feature data and the cigarette quality bias data;

[0014] processing each correlation matrix by using a multivariate interaction analysis algorithm to obtain an influence correlation model.

[0015] In one of the embodiments, the constructing a quality bias attribution analysis model and a quality bias prediction model according to the influence correlation model comprises:

[0016] processing an initial Bayesian network according to the influence correlation model to obtain a quality bias attribution analysis model;

[0017] performing weighted processing on each production feature data according to the influence correlation model to obtain a weighted feature set;

[0018] training an initial regression model according to the weighted feature set and each cigarette quality bias data until a preset training end condition is met to obtain a quality bias prediction model.

[0019] In one of the embodiments, the processing an initial Bayesian network according to the influence correlation model to obtain a quality bias attribution analysis model comprises:

[0020] according to the influence correlation model, taking each production feature data and each cigarette quality bias data as nodes of the initial Bayesian network, and taking the relationship between each production feature data and each cigarette quality bias data as edges of the initial Bayesian network to obtain a quality bias prediction model.

[0021] In one of the embodiments, the obtaining a target influence factor according to the quality bias attribution analysis model comprises:

[0022] determining a path between each production feature data and its corresponding cigarette quality bias data according to the quality bias attribution analysis model;

[0023] determining an attribution weight of each path;

[0024] determining a target path from all the paths according to the attribution weight of all the paths and a preset weight threshold, and determining the production feature data of the target path as a target influence factor.

[0025] In one embodiment, after processing the current production data according to the quality bias prediction model to obtain the quality bias prediction result, the method further comprises:

[0026] comparing the quality bias prediction result with a preset quality bias threshold range to obtain a comparison result;

[0027] generating a warning signal in a case where the comparison result is that the quality bias prediction result exceeds the quality bias threshold range.

[0028] In a second aspect, the present application further provides a cigarette production quality monitoring device, comprising:

[0029] a data acquisition module configured to acquire historical production data and corresponding cigarette quality bias data in a cigarette production process;

[0030] an association determination module configured to construct an influence association model according to the historical production data and the cigarette quality bias data; the influence association model representing an influence degree between each of the historical production data and each of the cigarette quality bias data;

[0031] a model construction module configured to construct a quality bias attribution analysis model and a quality bias prediction model according to the influence association model;

[0032] a factor determination module configured to obtain a target influence factor according to the quality bias attribution analysis model;

[0033] a quality monitoring module configured to process current production data according to the quality bias prediction model to obtain a quality bias prediction result, and use the target influence factor and the quality bias prediction result as a cigarette production quality monitoring result.

[0034] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0035] acquiring historical production data and corresponding cigarette quality bias data in a cigarette production process;

[0036] constructing an influence association model according to the historical production data and the cigarette quality bias data; the influence association model representing an influence degree between each of the historical production data and each of the cigarette quality bias data;

[0037] constructing a quality bias attribution analysis model and a quality bias prediction model according to the influence association model;

[0038] According to the quality bias attribution analysis model, a target influence factor is obtained.

[0039] According to the quality bias prediction model, current production data is processed to obtain a quality bias prediction result, and the target influence factor and the quality bias prediction result are taken as a cigarette production quality monitoring result.

[0040] In a fourth aspect, the present application further provides a computer readable storage medium, having a computer program stored thereon, the computer program being executed by a processor to implement the following steps:

[0041] obtaining historical production data and corresponding cigarette quality bias data in a cigarette production process;

[0042] According to the historical production data and the cigarette quality bias data, an influence correlation model is constructed; the influence correlation model represents an influence degree between each of the historical production data and each of the cigarette quality bias data;

[0043] According to the influence correlation model, a quality bias attribution analysis model and a quality bias prediction model are constructed;

[0044] According to the quality bias attribution analysis model, a target influence factor is obtained;

[0045] According to the quality bias prediction model, current production data is processed to obtain a quality bias prediction result, and the target influence factor and the quality bias prediction result are taken as a cigarette production quality monitoring result.

[0046] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, the computer program being executed by a processor to implement the following steps:

[0047] obtaining historical production data and corresponding cigarette quality bias data in a cigarette production process;

[0048] According to the historical production data and the cigarette quality bias data, an influence correlation model is constructed; the influence correlation model represents an influence degree between each of the historical production data and each of the cigarette quality bias data;

[0049] According to the influence correlation model, a quality bias attribution analysis model and a quality bias prediction model are constructed;

[0050] According to the quality bias attribution analysis model, a target influence factor is obtained;

[0051] According to the quality bias prediction model, current production data is processed to obtain a quality bias prediction result, and the target influence factor and the quality bias prediction result are taken as a cigarette production quality monitoring result.

[0052] The cigarette production quality monitoring method, device, computer equipment, computer readable storage medium and computer program product, the method obtains historical production data and corresponding cigarette quality mark bias data in the cigarette production process as data support, and constructs an influence correlation model according to the historical production data and the cigarette quality mark bias data, to clarify the interaction between the production data and the cigarette quality mark bias data, wherein the influence correlation model represents the influence degree between each historical production data and each cigarette quality mark bias data; further, according to the influence correlation model, a quality bias mark attribution analysis model and a quality bias mark prediction model are constructed; according to the quality bias mark attribution analysis model, a target influencing factor is obtained; and according to the quality bias mark prediction model, the current production data is processed to obtain a quality mark bias prediction result, and the target influencing factor and the quality mark bias prediction result are taken as the cigarette production quality monitoring result. The precise quality control based on the multi-factor interaction analysis and the quantitative influence model is realized, the change trend of the cigarette quality mark bias data can be predicted and monitored in real time, the production data that needs to be optimized can be automatically identified, and therefore the precision of the cigarette production quality is improved. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0054] Figure 1 A flowchart of a cigarette production quality monitoring method in an embodiment;

[0055] Figure 2 A flowchart of a feature extraction step of production data in an embodiment;

[0056] Figure 3 A flowchart of a model construction step in an embodiment;

[0057] Figure 4 A flowchart of a cigarette production quality control management and optimization method in an embodiment;

[0058] Figure 5 A schematic diagram of the principle of model construction in an embodiment;

[0059] Figure 6 A flowchart of a quality management and optimization step in an embodiment;

[0060] Figure 7 A structural block diagram of a cigarette production quality monitoring device in an embodiment;

[0061] Figure 8 Figure 1 is a schematic diagram of the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0062] For the purposes of the present application, the technical solutions and advantages thereof are more clearly apparent, the following further describes the present application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present application and are not intended to limit the present application.

[0063] As described in the background, the related art cigarette quality management method has the problem of low monitoring accuracy. The inventors have found that the cause of this problem is that, in the cigarette production process, the quality of the product is affected by multiple factors, especially the fluctuations in physical parameters such as temperature, humidity, filler content, moisture content, and end drop. These factors can cause significant differences in the physical quality of the finished cigarettes. For example, unstable temperature and humidity can directly affect the combustibility and taste of tobacco, while factors such as moisture content and filler content can cause differences in the appearance and draw resistance of the cigarettes. These changes not only affect the consumer experience, but also cause resource waste and increase the difficulty of quality control in the production process. Therefore, there is an urgent need for an efficient quality control management and optimization method that can ensure production quality while improving controllability and efficiency in the production process. In reality, quality control in the cigarette production process mainly relies on manual monitoring and basic statistical analysis methods. In the production process, relevant parameters such as temperature, humidity, filler content, moisture content, and end drop are monitored by sensors or manual sampling. Then, through regression analysis or control chart statistical methods, production managers analyze the relationship between these parameters and quality indicators, and then adjust the production conditions to ensure that the product quality meets the standards. However, such processing methods have several defects. First, the traditional manual monitoring and statistical analysis methods lack sufficient accuracy and cannot fully and accurately identify the complex interaction effects between various production parameters, resulting in insufficient accuracy of quality control. Second, existing technologies usually rely on static parameter adjustment and cannot reflect the dynamic changes of multiple factors in the production process in real time, resulting in poor consistency of product quality and ineffective prediction and control of quality fluctuations. Therefore, existing methods have obvious deficiencies in improving the accuracy of quality control, ensuring the consistency of finished product quality, and responding to production abnormalities in real time.

[0064] Based on the above reasons, the present application provides a cigarette production quality monitoring method aimed at improving monitoring accuracy.

[0065] In one exemplary embodiment, as Figure 1As shown, a cigarette production quality monitoring method is provided, and in this embodiment, the method is applied to a server for illustration. It should be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction of the terminal and the server. In this embodiment, the method includes the following steps:

[0066] In step S102, historical production data and corresponding cigarette quality deviation data in the cigarette production process are obtained.

[0067] The historical production data can be various types of parameter data recorded in the cigarette production process in a historical time period, such as temperature, humidity, filler content, water content, and end portion filament loss. These parameters are affected by various factors in the production process, and each parameter has a certain influence on the final quality of the production process. The temperature refers to the temperature control data of the key link in the production process, and the humidity reflects the influence of the environmental conditions on the production; the filler content refers to the proportion of the filler in the raw material, and the water content refers to the proportion of the water content in the tobacco leaf in the production process, which directly affects the quality and performance of the cigarette; the end portion filament loss refers to the amount of filamentous material at the end of the cigarette in the production process, which reflects the level of detail control in the production process.

[0068] The cigarette quality deviation data can be quality measurement data of the cigarette product after production, which is usually expressed in terms of quality deviation. These data reflect the deviation of the cigarette in terms of taste, shape, density, and suction resistance. In the actual production process, due to the differences in production lines, production equipment, and operation processes, the quality deviation data of each batch will be different, and these differences are inevitable, so it is necessary to specially analyze the quality deviation of different batches and different production lines for subsequent optimization and adjustment.

[0069] Optionally, the historical production data and the corresponding cigarette quality deviation data are obtained by the server through an automatic system from the data storage system of each production line. The data sources include sensors of production equipment, temperature and humidity control systems, and quality inspection systems, and real-time monitoring data and historical data of each production line are collected through an interface to the data storage system. In this way, not only the accuracy and completeness of the data can be ensured, but also real-time updating and dynamic monitoring of the data can be realized. During data acquisition, special attention should be paid to the differences between different batches of production data and quality deviation data. Each operation of the production line and each batch of materials and processes will affect the final product quality, so when processing and analyzing the data, the deviations between batches must be considered, and necessary standardization and normalization processing must be performed. This process is crucial for subsequent feature processing and model construction, ensuring the effectiveness and accuracy of the analysis results.

[0070] Step S104, constructing an influence correlation model according to the historical production data and the cigarette quality deviation data.

[0071] The influence correlation model represents the influence degree between each historical production data and each cigarette quality deviation data.

[0072] Optionally, the server performs correlation analysis on the historical production data and the cigarette quality deviation data, captures the linear and nonlinear relationships between each historical production data and each cigarette quality deviation data, and constructs the influence correlation model based on the linear and nonlinear relationships.

[0073] Step S106, constructing a quality deviation attribution analysis model and a quality deviation prediction model according to the influence correlation model.

[0074] The quality deviation attribution analysis model can be a model quantifying the causal path and influence degree of the production data on the cigarette quality deviation data.

[0075] The quality deviation prediction model can be a deep learning model for predicting the cigarette quality deviation data in a future time period.

[0076] Optionally, the server constructs the quality deviation attribution analysis model by using a Bayesian network as a causal inference algorithm based on the influence correlation model, quantifies the causal path and influence degree of the historical production data on the cigarette quality deviation data, and trains a deep learning model based on the influence correlation model, the historical production data and the corresponding cigarette quality deviation data to obtain the quality deviation prediction model.

[0077] Step S108, obtaining a target influence factor according to the quality deviation attribution analysis model.

[0078] The target influence factor can be a production data that has a greater influence on the cigarette quality deviation data and meets a preset influence degree among multiple production data, and can be used as a basis for monitoring the quality of cigarette production.

[0079] Optionally, the server identifies the historical production data that has a greater influence on the cigarette quality deviation data according to the quality deviation attribution analysis model, and regards the historical production data meeting the preset condition as the target influence factor. These target influence factors will be the main adjustment targets in the subsequent optimization process of cigarette production to improve the effect of quality control.

[0080] Step S110, processing the current production data according to the quality deviation prediction model to obtain a quality deviation prediction result, and taking the target influence factor and the quality deviation prediction result as a cigarette production quality monitoring result.

[0081] Optionally, the server inputs the current production data into the quality bias prediction model to obtain predicted cigarette quality bias data, takes the predicted cigarette quality bias data as a quality bias prediction result, and takes the target influencing factor and the quality bias prediction result as a cigarette production quality monitoring result, so as to enable a technician to monitor the quality of cigarette production and understand the production data that needs to be optimized to affect the quality of cigarette production.

[0082] Specifically, the server continuously monitors and collects current production data in the production process by integrating sensors and a production management system. The current production data includes the latest values of key production parameters such as temperature, humidity, filler content, moisture content, and end filament loss. It is worth mentioning that the production data herein includes but is not limited to these, and the addition of other data is still applicable in this application. The collected current production data X is preprocessed in real time and kept in a consistent format and scale with historical data to ensure that it can be directly input into the quality bias prediction model. In the real-time data input link, the preprocessed current production data X is input into the quality bias prediction model P in real time to generate a quality bias prediction value at the current time . Specifically, the calculation formula of the quality bias prediction value is as follows: , wherein P(X 实时 ) represents the quality bias value obtained by processing and predicting the real-time production feature data by a weighted ensemble regression method. In the prediction execution link, the quality bias prediction model P generates an accurate quality bias prediction result according to the input current production data, in combination with the main effects and interaction effects of each production feature in the influence correlation model M. This prediction result not only reflects the direct influence of each production feature on the quality bias at the current time, but also considers the synergistic effect between the features, thereby achieving comprehensive prediction of the change trend of the quality bias.

[0083] In the above cigarette production quality monitoring method, the method obtains historical production data and corresponding cigarette quality bias data in the cigarette production process as data support, and constructs an influence correlation model according to the historical production data and the cigarette quality bias data to clarify the interaction between the production data and the cigarette quality bias data, wherein the influence correlation model represents the influence degree between each historical production data and each cigarette quality bias data. Further, according to the influence correlation model, a quality bias attribution analysis model and a quality bias prediction model are constructed; according to the quality bias attribution analysis model, a target influencing factor is obtained; and according to the quality bias prediction model, current production data is processed to obtain a quality bias prediction result, and the target influencing factor and the quality bias prediction result are taken as a cigarette production quality monitoring result. This method realizes precise quality control based on multi-factor interaction analysis and quantitative influence model, can predict and monitor the change trend of the cigarette quality bias data in real time, automatically identifies the production data that needs to be optimized, and thus improves the accuracy of the quality of cigarette production.

[0084] In an exemplary embodiment, step S104 constructs an impact correlation model based on historical production data and cigarette quality standardization data, including:

[0085] Feature extraction is performed on historical production data to obtain production feature data; based on the correlation between each production feature data and each cigarette quality standard deviation data, a correlation matrix between each production feature data and each cigarette quality standard deviation data is constructed; a multivariate interaction analysis algorithm is used to process each correlation matrix to obtain an influence correlation model.

[0086] Optionally, such as Figure 2 The diagram shows a flowchart of the feature extraction steps for production data. The server performs feature extraction processing on historical production data, extracting production feature data such as temperature, humidity, end content, moisture content, and end-filament discharge from the historical production data. The production feature data includes the distribution and trend of each feature, which is used to reflect the dynamic change process of each production parameter. The production feature data is processed and optimized to generate the feature dataset corresponding to the production feature data.

[0087] Specifically, the server's basic feature extraction for temperature data involves calculating the mean, variance, skewness, and kurtosis statistics for a production batch or a time period to describe the distribution and fluctuation characteristics of temperature during the production process. Let the temperature sequence be T = {T1, T2, ..., Tn}, then the mean... and variance Calculated using the following formulas respectively: , skewness and kurtosis Calculated using the following formulas respectively: , .

[0088] In this embodiment, the basic features of humidity, end content, moisture content and end-filament loss are extracted using the same method and are denoted as mean, variance, skewness and kurtosis, respectively, as H, A, W and F.

[0089] To capture the changing trends of various production parameters, this embodiment applies time series analysis to perform trend analysis on the data for each feature. Taking temperature as an example, the moving average method is used to smooth the temperature series, eliminate short-term fluctuations, and extract the long-term trend. The formula for calculating the moving average value (MAT) is: (t=1, 2, ..., n), where k is the window size of the moving average, and MAT(t) represents the smoothed temperature value at time t. This process eliminates short-term fluctuations, allowing MAT(t) to better reflect the long-term trend of temperature.

[0090] In this embodiment, autoregressive models (AR models) are used to model each production parameter to quantify its autocorrelation and dynamic characteristics. Taking temperature as an example, the form of AR(1) model is: where ΦT is the autoregressive coefficient representing the relationship strength between the current time temperature and the previous time temperature, and ε t is a white noise term. By estimating the autoregressive coefficient, the temperature change pattern over time and its inertia characteristics are described. In the AR(1) model, the autoregressive coefficient Φ T reflects the inertia or trend strength of temperature data over time. Therefore, the trend degree τ T is defined as the absolute value of Φ T , representing the long-term trend strength of temperature data. The calculation formula of the trend degree is: If τ T is large, it means that the temperature change trend is more obvious, and if τ T is small, it means that the inertia of temperature change is weak and the change trend is not obvious.

[0091] For the content rate, moisture content, and end filament loss characteristics, in this embodiment, corresponding autoregressive models are also established to reveal their dynamic change rules and mutual dependence. Through the above method, the extracted production feature data not only covers the static distribution characteristics of each production parameter, but also fully reflects its dynamic change process in the production process, providing a solid data foundation for subsequent feature interaction analysis and influence correlation model construction.

[0092] Finally, in this embodiment, all the extracted production feature data are integrated into a comprehensive feature data set, including mean, variance, skewness, kurtosis, and trend degree, as the basis for further analysis and modeling. This feature data set accurately describes the distribution characteristics and dynamic change trend of key parameters in the production process, providing a guarantee for the effectiveness and precision of subsequent quality control and optimization steps.

[0093] In some embodiments, the systematic processing and optimization of the plurality of production feature data is to process the data of multiple production batches to generate a high-quality feature data set, ensuring the effectiveness and accuracy of subsequent analysis and modeling. In this embodiment, data processing and optimization include data cleaning, standardization processing, feature selection, and dimensionality reduction process.

[0094] The data cleaning phase aims to eliminate noise and outliers in the production feature data, ensuring the integrity and consistency of the data. In this embodiment, a statistical method-based outlier detection mechanism is adopted. Specifically, for each production feature data set, its interquartile range (IQR) is calculated, i.e., IQR = Q3 - Q1, where Q1 and Q3 represent the first quartile and the third quartile, respectively. Any data point below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR is considered an outlier and is removed or replaced to reduce interference with subsequent analysis.

[0095] In the data standardization processing phase, the purpose is to eliminate the dimensional difference between different features, so that they are in the same order of magnitude, facilitating subsequent correlation analysis and model construction. In this embodiment, the Z-score standardization method is adopted to standardize each production feature data, and its formula is as follows: where X represents the original data, μ is the mean of the feature, and σ is the standard deviation of the feature. The standardized data Z has a zero mean and a unit variance, ensuring that the features are compared on the same scale.

[0096] The feature selection and dimensionality reduction process aims to reduce redundant features, improve the expression ability of the feature data set, and improve the computational efficiency of the model. In this embodiment, the principal component analysis (PCA) method is used to reduce the dimensionality of the standardized feature data. PCA converts the original features into a set of new uncorrelated features (principal components) through linear transformation, with decreasing variance. Specifically, first, the feature covariance matrix is constructed: where Xi is the feature vector of the i-th sample, and μ is the mean vector of all sample features. Next, the eigenvalues and eigenvectors of the covariance matrix are calculated, and the eigenvectors corresponding to the largest k eigenvalues are selected to form the dimensionality reduction matrix W. Then, the original data is projected onto the dimensionality reduction matrix to obtain the reduced feature data: Y = XW. Through PCA dimensionality reduction, the main variation information of the data can be effectively preserved, while the dimensionality of the feature space is reduced, improving the efficiency of subsequent analysis and modeling.

[0097] In some embodiments, in addition to PCA, other feature selection methods such as variance selection method, mutual information method, etc. can also be combined to further optimize the quality of the feature data set. In this embodiment, PCA is selected as the main dimensionality reduction method to balance the interpretability and computational efficiency of the features.

[0098] Finally, the feature data after data cleaning, standardization processing and dimensionality reduction optimization is integrated into a comprehensive feature data set. The feature data set includes the standardized statistical features of various production parameters and the reduced principal components, fully reflecting the distribution characteristics and dynamic change trend of the key parameters in the production process, providing a high-quality data basis for subsequent correlation analysis and model construction.

[0099] Through the above data processing and optimization steps, the feature data set generated in this embodiment not only eliminates the noise and outliers in the data, unifies the feature scale, but also reduces the complexity and redundancy of the data through dimensionality reduction, ensuring the effectiveness and reliability of the feature data set in subsequent analysis.

[0100] Further, the server uses correlation analysis to analyze the correlation between each production feature data and each cigarette quality index data, constructs an association matrix between the single production feature data and each cigarette quality index data, and the association matrix reveals the independent influence degree of each production feature on the quality index. Based on the association matrix, the server uses a multivariate interaction analysis method to jointly analyze each production feature data, generates an influence association model, and the influence association model quantitatively represents the influence degree and interaction between each production feature and the quality index, which lays the foundation for identifying the production data to be regulated and optimized in the cigarette production process.

[0101] Specifically, in some embodiments, the correlation analysis includes a processing and analysis method combining multidimensional dependence evaluation and partial correlation analysis. First, the multidimensional dependence evaluation method is used to comprehensively capture the linear and nonlinear relationship between the production features and the quality index by combining the Pearson correlation coefficient and the mutual information. Specifically, the comprehensive dependence metric Ψ(X i , Y) between each production feature X i and the quality index Y is defined, and the calculation formula is: wherein ρX i , Y represents the Pearson correlation coefficient between the i-th production feature and the quality index, MI(X i , Y) represents the mutual information, and λ is the weight coefficient for balancing the contribution of linear and nonlinear dependence.

[0102] In this embodiment, the value of λ is determined by cross-validation method to ensure that the comprehensive dependence metric can accurately reflect the overall correlation between the production features and the quality index.

[0103] Secondly, partial correlation analysis is performed to eliminate the influence of other features on the correlation between the specific production feature and the quality index. Let the specific production feature be X i , the quality index be Y, and the remaining feature set be Z, then the partial correlation coefficient is defined as the correlation degree between X i and Y under the condition of controlling other features Z. The calculation formula is: wherein ρ Xi，Z and ρ Y，Z represent the correlation coefficient vectors between X i and Z and Y and Z, respectively.

[0104] By partial correlation analysis, the interference of other features on the correlation between X i and Y is effectively eliminated, ensuring that the obtained correlation measure more truly reflects the independent influence of X i on Y.

[0105] The results of multi-dimensional dependency evaluation and partial correlation analysis are combined to construct the final correlation matrix. Each element R i of the correlation matrix R represents the comprehensive correlation between the i-th production feature and the quality bias, and the calculation formula is: In this embodiment, the quality bias data involves different quality indicators of multiple batches, so the dimension of the correlation matrix R is m x n, where m is the number of production features and n is the number of quality bias indicators. Each R i value reveals the independent influence degree of the i-th production feature on the quality bias, and the numerical range reflects the strength and direction of the comprehensive correlation.

[0106] Through the above correlation analysis method, the correlation matrix generated in this embodiment not only combines the measurement of linear and nonlinear relationships, but also eliminates the interference of other features through partial correlation analysis, ensuring that the correlation matrix can accurately reflect the independent influence degree of each production feature on the quality bias. This innovative method provides solid foundation data for subsequent multi-element interaction analysis and the construction of influence correlation models.

[0107] In this embodiment, the correlation matrix has accurately reflected the independent influence degree of each production feature on the quality bias, and this step further utilizes the matrix to evaluate the interaction between production features through a multi-element interaction analysis method, and construct an influence correlation model that comprehensively reflects the main effect and interaction effect.

[0108] In some embodiments, the multi-element interaction analysis method includes a processing and analysis method combining feature priority ranking and interaction effect construction. Based on the correlation measure in the correlation matrix R, the production features are prioritized to determine the feature combination that should be considered in joint analysis. Specifically, let the correlation coefficient of the i-th production feature and the quality bias in the correlation matrix R be R i , then according to the descending order of ∣R i ∣, the production features with higher correlation are preferentially selected for interaction effect analysis. In this embodiment, the kurtosis of the end content, the skewness of the end filament amount, the trend of the moisture content, and the skewness of the end content are ranked lower.

[0109] In some embodiments, there are m production features, and the top p production features with higher correlation are selected as the main analysis objects through ranking. This step ensures that the multi-element interaction analysis focuses on feature combinations that have a significant impact on the quality bias, improving the explanatory power and prediction accuracy of the model.

[0110] Subsequently, interaction effect construction is performed by generating interaction terms between selected production features to evaluate their joint impact on quality mark bias. Let X i and X j be two different production features selected, their interaction effect term X i X j is defined as: .

[0111] In this embodiment, the multivariate interaction analysis method is further implemented by the following steps:

[0112] Interaction effect generation for feature pairs: According to the priority ranking results, interaction terms Φ i for all possible production feature pairs (X j , X ij ) are constructed to form an interaction feature set.

[0113] Weight determination for interaction effects: Using the correlation measures R i and R j in the correlation matrix R, each interaction term is assigned a weight ω ij , whose calculation formula is: where γ is a proportionality constant used to adjust the scale of interaction effect weights to ensure appropriate weight allocation in the impact correlation model.

[0114] Construction of impact correlation model: Combining the main effects and weighted interaction effects, a comprehensive impact correlation model M is generated, whose mathematical expression is: where θ i is the main effect coefficient of production feature X i , ω ij is the interaction effect weight of production feature pair (X i , X j ), Φ ij is the interaction term, and P is the number of selected main production features.

[0115] In this embodiment, the main effect coefficient θ i is determined based on the correlation matrix. Specifically, θi is determined by: where η is a proportionality constant used to adjust the scale of main effect coefficients to coordinate with the interaction effect weights ω ij . The value of the proportionality constant η is determined by cross-validation method to ensure that the main effect coefficients accurately reflect the independent impact of each production feature on quality mark bias.

[0116] To further improve the accuracy and interpretability of the model, this embodiment employs ridge regression to estimate the parameters of the influence correlation model. Ridge regression, by adding a regularization term to the least squares method, effectively prevents multicollinearity and improves the model's stability. Its objective function is defined as:

[0117] ,

[0118] Where Y k Let X be the quality bias value of the k-th sample. ik and Φ ij，k Let β be the i-th production feature value of the k-th sample and the value of the (i, j)-th interaction term. i and β ij λ represents the regression coefficients of the main effect and the interaction effect, respectively, and λ is the regularization parameter, which is determined by cross-validation to balance the model's fit and complexity.

[0119] The impact correlation model M, optimized by ridge regression, can quantitatively represent the main effects of each production characteristic on the quality standard deviation and their interactions. The specific expression is as follows: In this embodiment, the influence correlation model M reflects not only each production feature X i Independent effect β on quality scale deviation Y i It also covers production characteristic pairs (X) i ×X j The interaction between β ij This fully reflects the complex relationship between production parameters and quality deviations.

[0120] Furthermore, to ensure the robustness and generalization ability of the model, cross-validation is used in this embodiment to determine the regression coefficient β. i and β ij The optimal value is obtained, and regularization techniques, such as Lasso regression, are used to reduce the risk of overfitting and improve the predictive performance of the model.

[0121] Using the multivariate interaction analysis method described above, the influence correlation model generated in this embodiment can quantitatively represent the degree of influence and interaction between each production feature and the quality benchmark. This model not only reflects the direct impact of independent features on the quality benchmark, but also reveals the synergistic effects between features, providing a more comprehensive and accurate basis for quality control.

[0122] In one exemplary embodiment, such as Figure 3 As shown, step S106 constructs a quality bias attribution analysis model and a quality bias prediction model based on the influence correlation model, including:

[0123] In step S302, the initial Bayesian network is processed according to the influence correlation model to obtain a quality bias attribution analysis model.

[0124] The Bayesian network is a graph model used to represent and reason about uncertainty, commonly used in decision, statistical inference and machine learning, decomposes the joint distribution of a random variable into a product of local conditional distributions, and encodes the causal or dependent relationship between variables with a directed acyclic graph.

[0125] Optionally, the server processes the initial Bayesian network based on the influence correlation model, and the initial Bayesian network processing process includes determination of network structure, estimation of conditional probability distribution, and verification and optimization of the model, to obtain a quality bias attribution analysis model.

[0126] In step S304, each production feature data is weighted according to the influence correlation model to obtain a weighted feature set.

[0127] Optionally, the server weights each production feature data based on the influence correlation model to obtain a weighted feature set. Specifically, the server assigns a weight w i to each production feature data according to the influence degree of the production feature data on the cigarette quality bias data in the influence correlation model. , where R i represents the correlation measure between the i th production feature data and the cigarette quality bias data in the correlation matrix R, and m is the total number of production feature data. Through the above formula, the weight w i reflects the relative importance of each production feature data in the overall influence.

[0128] In step S306, the initial regression model is trained according to the weighted feature set and each cigarette quality bias data until the preset training end condition is met to obtain a quality bias prediction model.

[0129] Optionally, the server can use a variety of initial regression models, including linear regression, support vector regression and neural network regression model, to train the weighted feature set respectively to obtain respective prediction results. Let be the prediction value of the i th sample by the j th regression model, and the prediction result of the integrated model is calculated by the following formula: , where aj is the weight coefficient of the j th regression model, and n is the total number of regression models. The weight coefficient aj is determined by cross-validation method to optimize the overall prediction performance of the integrated model.

[0130] The prediction fusion part combines the interaction between each production feature in the influence correlation model M, and further improves the accuracy and stability of the prediction by weightedly fusing the prediction results of the regression models. Specifically, let Φ ij be the interaction term between the production features X i and X j , and the fused prediction result is defined as: where C is a set of production feature pairs with significant interaction effects, β ij is the weight coefficient of the interaction term Φ ij , which is determined by the least squares method or other optimization algorithms to ensure that the fused prediction result can fully reflect the synergistic effect between the production features.

[0131] The final generated quality bias prediction model P is expressed as: In this embodiment, the weighted ensemble regression method uses a multi-step process of feature weighting, model integration, and prediction fusion to fully utilize the main effect and interaction effect information in the influence correlation model, and constructs a quality bias prediction model that can accurately predict the quality bias value. This model not only reflects the current influence of each production feature, but also dynamically captures the synergistic effect between the features, ensuring the accuracy and timeliness of the prediction results.

[0132] In some embodiments, the constructed quality bias prediction model P can be combined with the quality bias reverse attribution analysis model B to achieve comprehensive quality management and optimization. In this embodiment, the prediction model generated by the weighted ensemble regression method provides a scientific and accurate basis for quality bias prediction, supports real-time adjustment and optimization in the production process, and improves the quality control level of cigarette production.

[0133] Through the above innovative weighted ensemble regression method, the quality bias prediction model generated in this embodiment not only effectively utilizes the independent influence information in the correlation matrix, but also improves the prediction performance and interpretability of the model through multi-model integration and interaction effect fusion, providing a solid quantitative support for quality control and optimization.

[0134] In an exemplary embodiment, step S302 processes the initial Bayesian network according to the influence correlation model to obtain a quality bias attribution analysis model, including:

[0135] According to the influence correlation model, the production feature data and the cigarette quality bias data are used as the nodes of the initial Bayesian network, and the relationship between the production feature data and the cigarette quality bias data is used as the edges of the initial Bayesian network, to obtain a quality bias prediction model.

[0136] Optionally, the server obtains the quality bias prediction model by taking each production feature data and each cigarette quality bias data as a node of the initial Bayesian network and taking the relationship between each production feature data and each cigarette quality bias data as an edge of the initial Bayesian network according to the influence correlation model. Specifically, the server first determines the initial structure of the Bayesian network based on the correlation degree between each production feature and the cigarette quality bias in the influence correlation model M. Specifically, if the influence correlation model M has a production feature X i with a greater influence degree on the cigarette quality bias Y, a directed edge from X i to Y is established in the Bayesian network. In addition, if the influence correlation model has an interaction between a production feature X i and X j , a directed edge between X i and X j is established in the Bayesian network to represent the potential causal relationship.

[0137] After the structure of the Bayesian network is determined, the estimation of the conditional probability distribution is performed. In the embodiment, the maximum likelihood estimation method is adopted to calculate the conditional probability distribution of each node according to the historical production data. Let X = {X1, X2, …, Xm} be the production feature set, Y be the cigarette quality bias, and the conditional probability distribution P(Y|X) of the Bayesian network be estimated by the correlation matrix R in the influence correlation model M. The specific formula is as follows: where P(Y) is the prior probability distribution of the cigarette quality bias data, P(X|Y) is the conditional probability distribution of each production feature given the cigarette quality bias, and P(X) is the marginal probability distribution of the production feature. By maximizing the likelihood function: where θ represents the parameter set in the Bayesian network, N is the sample number, Y(k) and X(k) are the cigarette quality bias value and the production feature value of the kth sample, respectively, and the parameter θ is estimated to maximize the likelihood function L(θ).

[0138] In the embodiment, the parameter estimation process of the Bayesian network combines the correlation degree of each production feature in the influence correlation model, ensuring that the model can accurately reflect the causal influence of each feature on the cigarette quality bias. At the same time, by introducing a regularization technique such as Lasso regression, overfitting of the model is prevented, and the generalization ability and stability of the model are improved. The optimization objective function after regularization is defined as: where λ is the regularization parameter, which is determined by the cross-validation method to balance the fitting degree and complexity of the model.

[0139] After completing the parameter estimation, model validation and optimization are performed. Through the cross-validation method, the performance of the Bayesian network on the training set and the validation set is evaluated, and the model structure and parameters are adjusted to improve the prediction accuracy and robustness. In this embodiment, indicators such as entropy gain and Bayesian information criterion are used to optimize the structure of the Bayesian network to ensure the scientificity and practicality of the model. The constructed cigarette quality label bias reverse attribution analysis model is represented by a Bayesian network BN, and its mathematical expression is: where V is a node set including various production features and cigarette quality labels, E is an edge set representing the causal relationship between nodes, and P is a conditional probability distribution set quantifying the causal influence between nodes.

[0140] In this embodiment, the cigarette quality label bias reverse attribution analysis model B can intuitively display the causal path and its impact of production features on cigarette quality label bias through the Bayesian network BN. For example, if the production feature X1 has a significant impact on the cigarette quality label bias y, a directed edge from X1 to Y will be established in the Bayesian network, and the impact will be quantified through the conditional probability distribution P(Y|X1). Similarly, if there is a synergistic effect between production features X1 and X2, a directed edge between X1 and X2 will be established in the Bayesian network, and the joint effect will be quantified through the conditional probability distribution P(Y|X1, X2).

[0141] The cigarette quality label bias reverse attribution analysis model constructed by the Bayesian network can accurately identify and quantify the causal path and its impact of production features on cigarette quality label bias in this embodiment, providing a scientific basis for subsequent quality management and optimization steps, and ensuring efficient and accurate quality control in the cigarette production process.

[0142] In one exemplary embodiment, step S108 obtains target influencing factors according to the quality label bias attribution analysis model, including:

[0143] According to the quality label bias attribution analysis model, the paths between each production feature data and its corresponding cigarette quality label bias data are determined, the attribution weight of each path is determined, and the target path is determined from all paths according to the attribution weight of all paths and a preset weight threshold, and the production feature data of the target path is determined as the target influencing factor.

[0144] Alternatively, the server identifies the paths between each production feature data and its corresponding cigarette quality label bias data according to the quality label bias attribution analysis model, calculates the attribution weight of each path, and further determines the target path from all paths according to the attribution weight of all paths and a preset weight threshold, and determines the production feature data of the target path as the target influencing factor.

[0145] Specifically, in some embodiments, all paths from the nodes X i representing production features to the nodes Y representing quality bias data are identified by traversing the Bayesian network BN of the quality bias attribution analysis model. These paths can include direct paths and indirect paths through other intermediate feature nodes. Let P be a path from an arbitrary production feature X i to a quality bias Y, the impact measure of the path is defined as the product of the conditional probabilities of the edges on the path, i.e., where A and B represent adjacent nodes in the path. By calculating the impact measure of all paths, the overall contribution of each path to the quality bias can be evaluated.

[0146] In some embodiments, for each path P, its attribution weight W(P) is calculated, which is defined as the ratio of the path impact measure and the path length L(P) to balance the influence and complexity of the path: In this embodiment, the path length L(P) represents the number of edges in the path. The path with a larger attribution weight W(P) is considered as a path that significantly affects the quality bias.

[0147] In some embodiments, the attribution weights of all paths are aggregated to calculate the total attribution weight W(X i ) of each production feature X i to the quality bias Y, which is formulated as: where represents the set of all paths from the production feature X i to the quality bias Y. By sorting the total attribution weights W(X i ) of all production features X i , the production features that have a greater impact on the quality bias can be identified. In this embodiment, there are m production features, by calculating the total attribution weight W(X i ) of each feature and ranking them in descending order of weight values, the top p production features with higher weight values are selected as the key influencing factors. These key influencing factors will be the main adjustment targets in the subsequent production optimization process to improve the effectiveness of quality control.

[0148] In addition, to further verify the effectiveness of the identified key influencing factors, in this embodiment, a sensitivity analysis method is used to perform single-factor and multi-factor sensitivity tests on the key production features. Specifically, different value ranges of the key production features X i are set, and by simulating the changes of the quality bias Y under different values, the sensitivity of each feature to the quality bias is evaluated. The results of the sensitivity test are used to confirm the actual influence of the key influencing factors and guide the adjustment of specific production process parameters.

[0149] In an exemplary embodiment, after the quality bias prediction result is obtained by processing the current production data according to the quality bias prediction model in step S110, the method further includes:

[0150] comparing the quality bias prediction result with a preset quality bias threshold range to obtain a comparison result; and generating a warning signal in a case where the comparison result is that the quality bias prediction result exceeds the quality bias threshold range.

[0151] Optionally, the server compares the quality bias prediction result with a preset quality bias threshold range to obtain a comparison result; and generates a warning signal in a case where the comparison result is that the quality bias prediction result exceeds the quality bias threshold range, indicating that the quality bias prediction result exceeds a controllable range, and alarms according to a preset alarm mode, such as audible and visual alarm, to prompt a technical personnel to respond to the quality fluctuation in the production process in a timely manner and ensure the stability of the cigarette production quality.

[0152] In addition, in the embodiment, the quality bias prediction result of the quality bias prediction model is also fed back to the production management system to form a closed-loop control mechanism. Through continuous monitoring and dynamic adjustment, the quality bias in the production process is ensured to be maintained within the set target range, and the stability of the production process and the consistency of the product quality are improved.

[0153] In an exemplary embodiment, as shown in Figure 4 a cigarette production quality control management and optimization method is provided, including:

[0154] In step S1, obtain the to-be-processed data, which includes cigarette quality bias data and production data (historical production data), and the production data includes temperature, humidity, filler content, moisture content, and end portion filament loss in the cigarette production;

[0155] In step S2, perform feature processing and construction on the to-be-processed data to obtain a feature data set;

[0156] Specifically, basic feature data (production feature data) of temperature, humidity, filler content, moisture content, and end portion filament loss is extracted from the production data, and the basic feature data includes the distribution and variation trend of each feature, which is used to reflect the dynamic change process of each production parameter; and the basic feature data is processed and optimized to generate the feature data set.

[0157] In step S3, perform correlation analysis on the feature data set and the cigarette quality bias data and construct an influence correlation model;

[0158] Specifically, correlation analysis is used to analyze the feature dataset and cigarette quality standard deviation data to obtain the correlation matrix between a single production feature and the quality standard deviation. The correlation matrix reveals the degree of independent influence of each production feature on the quality standard deviation. Based on the correlation matrix, a multivariate interaction analysis method is used to jointly analyze each production feature to generate an influence correlation model. The influence correlation model quantitatively represents the degree of influence and interaction between each production feature and the quality standard deviation.

[0159] Step S4: Construct a quality standard bias inverse attribution analysis model (quality standard bias attribution analysis model) and a quality standard bias positive prediction model (quality standard bias prediction model) through the influence correlation model.

[0160] Specifically, such as Figure 5 As shown, a schematic diagram of the model construction principle is provided. Based on the influence association model, a Bayesian network is used as the causal inference algorithm to construct a quality standard deviation inverse attribution analysis model, which quantifies the causal path and degree of influence of production features on quality standard deviation. Based on the influence association model, a weighted ensemble regression method is used to construct a quality standard deviation positive prediction model, which is used to predict the quality standard deviation value in real time and reflect the current influence of each production feature.

[0161] Step S5: Perform quality management and optimization based on the quality standard deviation reverse attribution analysis model and the quality standard deviation positive prediction model.

[0162] Specifically, such as Figure 6 As shown, a flowchart illustrating the quality management and optimization steps is provided. A reverse attribution analysis model for quality benchmarks is used to analyze the impact path of production characteristics on quality benchmarks, identify key influencing factors, and provide guidance for optimization. A forward prediction model for quality benchmarks is used to predict the changing trends of quality benchmarks in real time, combined with actual production data, for monitoring and early warning. Based on the analysis results of the reverse attribution and forward prediction models, the production scheduling system is optimized, and production parameters are automatically adjusted to ensure the accuracy and consistency of quality control.

[0163] In some implementations, this step optimizes the production scheduling system and automatically adjusts production parameters based on the analysis results of reverse attribution and forward prediction models, ensuring the accuracy and consistency of quality control. This step includes three stages: data integration, optimized decision-making, and production parameter adjustment. It fully utilizes the analytical results of the aforementioned reverse attribution and forward prediction models to achieve dynamic optimization and precise control of the production process.

[0164] In this embodiment, the quality bias reverse attribution analysis model quantifies the causal paths and impact of production features on quality bias through Bayesian networks, identifying production features and their interactions that have a significant impact on quality bias. Meanwhile, the quality bias forward prediction model, combined with real-time production data, can accurately predict the current and future quality bias values, reflecting the immediate impact of each production feature.

[0165] In the data integration stage, the key influencing factors identified by the reverse attribution analysis model are integrated with the quality bias prediction results generated by the forward prediction model. Specifically, the reverse attribution analysis provides detailed information on the production features that affect quality bias and their causal relationships, while the forward prediction provides a quality bias trend prediction based on current production parameters. By matching and integrating the data from both models, a comprehensive quality control information base is formed.

[0166] In the optimization decision-making stage, based on the integrated quality control information, production scheduling optimization strategies are developed. Using the production feature influence paths determined by the reverse attribution analysis model, combined with the quality bias prediction values provided by the forward prediction model, the potential impact of current production parameter settings on quality bias is evaluated. Based on this evaluation result, the system automatically generates production parameter adjustment schemes to ensure that key parameters in the production process can be optimized to the best state, thereby minimizing quality bias.

[0167] In the production parameter adjustment stage, based on the adjustment schemes generated in the optimization decision-making stage, real-time adjustments are automatically made to production equipment and process parameters. Specific operations include dynamic adjustment of key production parameters such as temperature, humidity, end drop, moisture content, and end drop, to adapt to the predicted quality bias trend. This automatic adjustment process, through seamless integration with the production scheduling system, ensures the timeliness and accuracy of production parameter adjustment, avoiding delays and errors caused by human intervention.

[0168] In some embodiments, the optimized production scheduling system also integrates a real-time monitoring and feedback mechanism. The system continuously monitors changes in various parameters during the production process and their impact on quality bias, updating the input data for the prediction model and attribution analysis model in real time. Through closed-loop control, the system can dynamically adjust production parameters based on the latest production data and prediction results, ensuring continuous optimization and stability of quality control.

[0169] The step automatically optimizes the production scheduling system by comprehensively utilizing the analysis results of the reverse attribution and forward prediction model, realizes real-time adjustment and precise control of the production parameters. Through the organic combination of data integration, optimized decision making and production parameter adjustment, it ensures that the quality index deviation in the cigarette production process is always maintained within the predetermined target range, improves the consistency of product quality and the stability of the production process. In some embodiments, this step not only improves the automation level of quality control, but also significantly enhances the efficiency and scientificity of production management through intelligent production parameter optimization. In this embodiment, the systematic production scheduling optimization process provides solid technical support for efficient and precise quality control of cigarette production.

[0170] In this embodiment, a cigarette production quality control management and optimization method and system realize precise quality control based on multi-factor interaction analysis and quantitative influence model, can predict and monitor the trend of quality index deviation in real time, automatically identify and optimize key production parameters. By comprehensively utilizing reverse attribution and forward prediction model, not only the precision of quality control is improved, but also the consistency of finished product quality is ensured, the controllability and efficiency of the production process are improved, and the problems of insufficient quality control precision and large finished product quality fluctuation in the prior art are solved.

[0171] It should be understood that although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0172] Based on the same inventive concept, the embodiments of the present application also provide a cigarette production quality monitoring device for implementing the above-mentioned cigarette production quality monitoring method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more cigarette production quality monitoring device embodiments provided below can refer to the limitations of the cigarette production quality monitoring method in the above text, which will not be repeated here.

[0173] In an exemplary embodiment, as Figure 7As shown, a cigarette production quality monitoring device 700 is provided, comprising a data acquisition module 701, an association determination module 702, a model construction module 703, a factor determination module 704, and a quality monitoring module 705, wherein:

[0174] The data acquisition module 701 is configured to acquire historical production data and corresponding cigarette quality deviation data in a cigarette production process.

[0175] The association determination module 702 is configured to construct an influence association model according to the historical production data and the cigarette quality deviation data. The influence association model represents the influence degree between each historical production data and each cigarette quality deviation data.

[0176] The model construction module 703 is configured to construct a quality deviation attribution analysis model and a quality deviation prediction model according to the influence association model.

[0177] The factor determination module 704 is configured to obtain a target influence factor according to the quality deviation attribution analysis model.

[0178] The quality monitoring module 705 is configured to process current production data according to the quality deviation prediction model to obtain a quality deviation prediction result, and use the target influence factor and the quality deviation prediction result as a cigarette production quality monitoring result.

[0179] Further, in an embodiment, the association determination module 702 is further configured to perform feature extraction processing on the historical production data to obtain production feature data, construct an association matrix between each production feature data and each cigarette quality deviation data according to the correlation between the production feature data and the cigarette quality deviation data, and process each association matrix using a multivariate interaction analysis algorithm to obtain the influence association model.

[0180] Further, in an embodiment, the model construction module 703 is further configured to process an initial Bayesian network according to the influence association model to obtain the quality deviation attribution analysis model, perform weighted processing on each production feature data according to the influence association model to obtain a weighted feature set, and train an initial regression model according to the weighted feature set and each cigarette quality deviation data until a preset training end condition is met to obtain the quality deviation prediction model.

[0181] Further, in an embodiment, the model construction module 703 is further configured to use each production feature data and each cigarette quality deviation data as a node of an initial Bayesian network, and use the relationship between each production feature data and each cigarette quality deviation data as an edge of the initial Bayesian network, to obtain the quality deviation prediction model according to the influence association model.

[0182] Further, in one embodiment, the factor determination module 704 is further configured to determine paths between each production feature data and its corresponding cigarette quality mark bias data according to the quality mark bias attribution analysis model; determine attribution weights of each path; determine target paths from all paths according to attribution weights of all paths and a preset weight threshold, and determine production feature data of the target paths as target impact factors.

[0183] Further, in one embodiment, the quality monitoring module 705 is further configured to compare the quality mark bias prediction result with a preset quality mark bias threshold range to obtain a comparison result; and generate a warning signal in a case that the comparison result is that the quality mark bias prediction result exceeds the quality mark bias threshold range.

[0184] The modules in the cigarette production quality monitoring device 700 described above can be implemented wholly or partially by software, hardware and combinations thereof. The modules described above can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the modules.

[0185] In one exemplary embodiment, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in Figure 8 The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store historical production data, cigarette quality mark bias data, current production data and the like. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with terminals outside through network connection. The computer program is executed by the processor to implement a cigarette production quality monitoring method.

[0186] Those skilled in the art can understand that Figure 8 The structure shown in the above

[0187] In an embodiment, a computer device is also provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0188] In an embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0189] In an embodiment, a computer program product is provided, which comprises a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0190] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, 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. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0191] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0192] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method of monitoring the quality of cigarette production, characterized by, The method comprises: obtaining historical production data and corresponding cigarette quality mark deviation data in a cigarette production process; constructing an influence correlation model according to the historical production data and the cigarette quality mark deviation data; the influence correlation model represents the influence degree between each historical production data and each cigarette quality mark deviation data; constructing a quality deviation mark attribution analysis model and a quality deviation mark prediction model according to the influence correlation model; obtaining a target influence factor according to the quality deviation mark attribution analysis model; processing current production data according to the quality deviation mark prediction model to obtain a quality mark deviation prediction result, and taking the target influence factor and the quality mark deviation prediction result as a cigarette production quality monitoring result.

2. The method of claim 1, wherein, The method comprises: performing feature extraction processing on the historical production data to obtain production feature data; constructing an association matrix between each production feature data and each cigarette quality mark deviation data according to the correlation between each production feature data and each cigarette quality mark deviation data; processing each association matrix by using a multivariate interaction analysis algorithm to obtain an influence correlation model.

3. The method of claim 2, wherein, The method comprises: processing an initial Bayesian network according to the influence correlation model to obtain a quality deviation mark attribution analysis model; performing weighted processing on each production feature data according to the influence correlation model to obtain a weighted feature set; training an initial regression model according to the weighted feature set and each cigarette quality mark deviation data until a preset training end condition is met to obtain a quality deviation mark prediction model.

4. The method of claim 3, wherein, The method comprises: according to the influence correlation model, taking each production feature data and each cigarette quality mark deviation data as nodes of the initial Bayesian network, and taking the relationship between each production feature data and each cigarette quality mark deviation data as edges of the initial Bayesian network to obtain a quality deviation mark prediction model.

5. The method of claim 4, wherein, The method comprises: determining the path between each production feature data and its corresponding cigarette quality mark deviation data according to the quality deviation mark attribution analysis model; determining the attribution weight of each path; determining a target path from all the paths according to the attribution weight of all the paths and a preset weight threshold, and determining the production feature data of the target path as a target influence factor.

6. The method of claim 1, wherein, After processing current production data according to the quality deviation mark prediction model to obtain a quality mark deviation prediction result, the method further comprises: comparing the quality mark deviation prediction result with a preset quality mark deviation threshold range to obtain a comparison result; generating an early warning signal in the case that the comparison result is that the quality mark deviation prediction result exceeds the quality mark deviation threshold range.

7. A cigarette production quality monitoring device, characterized in that, The device comprises: A data acquisition module is configured to acquire historical production data and corresponding cigarette quality label deviation data in a cigarette production process; An association determination module is configured to construct an influence association model according to the historical production data and the cigarette quality label deviation data, wherein the influence association model represents an influence degree between each historical production data and each cigarette quality label deviation data; A model construction module is configured to construct a quality label deviation attribution analysis model and a quality label deviation prediction model according to the influence association model; A factor determination module is configured to obtain a target influence factor according to the quality label deviation attribution analysis model; A quality monitoring module is configured to process current production data according to the quality label deviation prediction model to obtain a quality label deviation prediction result, and take the target influence factor and the quality label deviation prediction result as a cigarette production quality monitoring result.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images