Cigarette loose end value prediction method and system and storage medium
By introducing physical information neural networks and tobacco processing knowledge into the detection of empty cigarette heads, a loss function based on physical information is constructed, which solves the problems of low efficiency and poor interpretability in traditional methods and achieves high-precision prediction and quality control of empty cigarette head values.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA TOBACCO ZHEJIANG IND CO LTD
- Filing Date
- 2026-01-27
- Publication Date
- 2026-05-08
AI Technical Summary
Existing methods for detecting empty cigarette ends are inefficient and prone to missing detections. Furthermore, traditional machine learning models struggle to incorporate physical laws, resulting in unreliable and poorly interpretable predictions.
A physical information neural network (PINN) is used in conjunction with tobacco industry process knowledge to construct a loss function based on physical information. The model is trained using multidimensional tobacco process parameter data to obtain the optimal feature set for prediction.
It improves the accuracy and reliability of predicting the number of empty cigarettes, and the model is highly interpretable, enabling it to predict the location and probability of empty cigarettes in advance, thereby reducing product quality fluctuations and costs.
Smart Images
Figure CN121998190A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cigarette manufacturing quality control technology, specifically to a method, system, and storage medium for predicting the empty value of cigarettes. Background Technology
[0002] In cigarette production, the "empty end value" is one of the core indicators for evaluating cigarette manufacturing quality. Empty ends are a common quality defect in cigarette production, meaning insufficient tobacco filling at the end of the cigarette, resulting in noticeable gaps. From an aesthetic perspective, empty ends significantly damage the overall image of the product, reducing consumer purchasing desire. From a usage perspective, empty ends may lead to faster tobacco burning and reduced draw resistance, further compromising the stability of the smoking process and the consistency of taste, severely impacting the consumer's smoking experience. Therefore, accurate and efficient prediction and control of the empty end value are crucial in the tobacco manufacturing industry. Current methods for detecting empty cigarette heads primarily rely on manual sampling and automated optical inspection based on machine vision. Manual methods are inefficient, subjective, and prone to missed detections. While machine vision methods improve efficiency, they are susceptible to interference from ambient light and cigarette placement, and struggle to predict unseen, inherent empty head trends. Combining empty head prediction technology with machine vision technology would allow for the prediction of the location and probability of empty cigarette heads before manual sampling and machine vision processes. This would enable inspectors to manage equipment more effectively, proactively removing defective products and improving both efficiency and accuracy. Furthermore, the predicted empty head values could serve as a basis for engineers to optimize production, reducing downtime for maintenance and waste from defective products.
[0003] Existing methods for predicting cigarette manufacturing quality have significant limitations. Most rely on statistical and machine learning techniques, with data-driven models such as Support Vector Machines (SVM), Random Forests, and deep learning networks widely used. While these methods improve predictive capabilities to some extent, they are essentially black-box models, purely dependent on statistical patterns in the data, and thus have obvious limitations. For example, predictions may contradict known physical laws and industry knowledge, rendering the models unreliable in practical applications. Furthermore, the performance of some models heavily depends on large amounts of high-quality labeled data, while high-quality defect samples are scarce in real-world industrial scenarios, resulting in insufficient generalization ability. On the other hand, traditional machine learning models suffer from poor interpretability; engineers struggle to understand the model's decision-making rationale and cannot connect the predictions to specific production process parameters to guide production optimization.
[0004] Physical Information Neural Networks (PINNs) are a new paradigm that has emerged in recent years, combining physical laws with deep learning. They guide the model's learning process by introducing governing equations, boundary conditions, or known physical relationships as constraints into the construction of the neural network, thereby addressing the problems of lack of physical information consistency, data dependence, and poor interpretability in traditional machine learning techniques. However, current technologies lack the ability to effectively encode the specific process knowledge and physical mechanisms of the tobacco industry into the PINN framework and design a complete, engineerable method for predicting cigarette blank values.
[0005] Therefore, how to deeply integrate prior tobacco processing knowledge with physical information neural network quality prediction methods to achieve effective construction of physical information neural networks related to cigarette blank value, efficient screening of multi-dimensional process parameters, and parameter control based on process logic has become the core technical challenge for improving the prediction accuracy of cigarette blank value. Summary of the Invention
[0006] The purpose of this invention is to provide a method, system, and storage medium for predicting the empty value of cigarettes. It aims to solve the problems of lack of physical consistency, strong data dependence, and poor interpretability in the quality prediction process of existing methods by integrating process knowledge and physical information neural network quality prediction methods.
[0007] To achieve the above objectives, embodiments of the present invention provide a method for predicting the empty value of cigarettes, including: Obtain multi-dimensional cigarette manufacturing process parameter data; The multidimensional cigarette manufacturing process parameter data are preprocessed; Construct a cigarette empty value prediction model; Construct a loss function based on physical information; Based on the loss function based on physical information, the cigarette empty value prediction model is trained using preprocessed multidimensional cigarette manufacturing process parameter data to obtain the optimal feature set. The cigarette blank value prediction model corresponding to the optimal feature set is used to predict the cigarettes to be predicted in order to obtain the prediction results.
[0008] Optionally, preprocessing the multidimensional cigarette manufacturing parameter data includes: Detect and remove outliers; Perform data consistency verification; The missing process parameter features are filled in using a multiple interpolation method; Standard deviation is used to standardize the data and eliminate dimensional differences.
[0009] Optionally, constructing a loss function based on physical information includes: Construct a loss function based on physical information according to formula (1). (1) in, The loss function is based on physical information. Based on the predicted loss term, For port density loss term, For the distance loss at the compaction end, for The weighting coefficients, for The weighting coefficients.
[0010] Optionally, the basic prediction loss term is obtained according to formula (2): (2) in, Based on the predicted loss term, This represents the actual value of the empty cigarette count. This is the predicted value for short selling. The threshold for segmentation of the loss function.
[0011] Optionally, the port density loss term is obtained according to formula (3): (3) in, For port density loss term, This represents the average density of the first n segments before combustion in the multidimensional cigarette manufacturing process parameter data. This is the standard density value. This is the predicted value for short selling. The maximum allowed short position value. This is the proportionality coefficient.
[0012] Optionally, the compaction end distance loss term is obtained according to formula (4): (4) in, For the distance loss at the compaction end, This represents the deviation value of the distance between the compacted ends. This is the loss magnitude coefficient. This is a nonlinear influence coefficient. This is the predicted value for short selling. This represents the maximum allowed short position value.
[0013] Optionally, based on the loss function based on physical information, the preprocessed multidimensional cigarette manufacturing process parameter data is used to train the cigarette empty value prediction model to obtain the optimal feature set, including: Input the multidimensional cigarette manufacturing parameter dataset into the cigarette empty value prediction model; Calculate the global importance score for each feature in the multidimensional cigarette manufacturing parameter dataset; Features with global importance scores below a preset threshold are removed, the multidimensional cigarette manufacturing process parameter dataset is updated, and the number of remaining features is calculated. Determine whether the number of remaining features is less than or equal to a preset minimum value; If the number of remaining features is less than or equal to a preset minimum value, the model training is completed. If the number of remaining features is greater than the preset minimum value, return to the step of inputting the multidimensional cigarette process parameter dataset into the cigarette empty value prediction model; The predictive power of each model is evaluated using model evaluation metrics to obtain the optimal feature set.
[0014] Optionally, calculating the global importance score of each feature in the multidimensional cigarette manufacturing parameter dataset includes: Calculate the global importance score according to formula (5). (5) in, As a score for overall importance, Features In the sample On value, This represents the expected value of the sample.
[0015] On the other hand, the present invention also provides a cigarette empty value prediction system, the system including a processor configured to perform any of the methods described above.
[0016] In another aspect, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement any of the methods described above.
[0017] The beneficial effects of this invention are: The present invention proposes a method for predicting the empty value of cigarettes based on a physical information neural network. This method can deeply integrate the prior physical knowledge of the tobacco industry into the model learning process, so that the prediction results not only conform to the statistical laws of data but also follow the physical mechanism, which significantly improves the prediction accuracy and model reliability.
[0018] The iterative feature selection mechanism used in this invention can automatically identify the most relevant feature subsets, reducing model complexity and overfitting risk, and improving computational efficiency.
[0019] The prediction method provided by this invention has good interpretability, and the prediction results can be directly linked to specific physical characteristics, providing clear guidance for process improvement.
[0020] The embodiments of this invention achieve a prediction accuracy of over 95% on an independent test set, providing an innovative quality inspection method for the tobacco industry. This method can predict the location and probability of missing cigarettes in advance, helping operators to adjust processing techniques in a timely manner, preventing the processing from getting out of control, and is of great significance for reducing product quality fluctuations and processing costs, as well as reducing the rate of defective cigarettes.
[0021] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description
[0022] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 A flowchart of a cigarette blank value prediction method according to an embodiment of the present invention; Figure 2 This is a distribution diagram of cigarette empty count values according to one embodiment of the present invention; Figure 3 This is a distribution map of multi-dimensional input feature variables according to an embodiment of the present invention; Figure 4 A flowchart of a method for preprocessing multidimensional cigarette manufacturing process parameter data according to an embodiment of the present invention; Figure 5 A flowchart of a method for training a cigarette empty value prediction model using preprocessed multidimensional cigarette process parameter data to obtain the optimal feature set according to an embodiment of the present invention. Figure 6 This is a schematic diagram of verification loss according to one embodiment of the present invention; Figure 7 This is a graph showing the change of the loss function based on physical information during the iteration process according to an embodiment of the present invention; Figure 8 This is a feature selection sequence diagram during the iterative process according to an embodiment of the present invention; Figure 9 This is a graph showing the variation in feature importance within the optimal feature set according to one embodiment of the present invention. Figure 10 This is a comparison chart of prediction results from a cigarette empty value prediction model according to an embodiment of the present invention; Figure 11A bar chart showing the evaluation index of the prediction results of the cigarette empty value prediction model according to an embodiment of the present invention; Figure 12 This is a bar chart showing the contribution value of each feature in the optimal feature set according to an embodiment of the present invention. Detailed Implementation
[0023] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.
[0024] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.
[0025] like Figure 1 The diagram shows a flowchart of a cigarette empty value prediction method according to an embodiment of the present invention. Figure 1 The prediction method may include the following steps: In step S10, multidimensional cigarette manufacturing process parameter data are obtained; In step S11, the multidimensional cigarette manufacturing process parameter data is preprocessed; In step S12, a cigarette empty value prediction model is constructed; In step S13, a loss function based on physical information is constructed; In step S14, the cigarette empty value prediction model is trained using the preprocessed multidimensional cigarette process parameter data based on the loss function based on physical information, so as to obtain the optimal feature set. In step S15, the cigarette blank value prediction model corresponding to the optimal feature set is used to predict the cigarette to be predicted in order to obtain the prediction result.
[0026] In such Figure 1In the illustrated method for predicting cigarette blank values, step S10 is used to acquire multidimensional cigarette process parameter data. In this embodiment, this can be achieved by collecting cigarette sample datasets through high-precision sensors on the cigarette rolling and packaging production line to obtain cigarette microwave density data and single cigarette data during the rolling and packaging process. In this example, a total of 22,698 valid samples were collected. This cigarette sample dataset includes the cigarette blank value ('f156') as the target variable and multidimensional cigarette process parameter data as independent variables. The input independent variables include two types of physical information related to the cigarette manufacturing process: the density value sequence at the burning end of the cigarette ('s 1st segment', 's 2nd segment', 's 3rd segment', 's 4th segment', 's 5th segment', 's 6th segment') and the distance deviation value at the compacted end of the cigarette ('f96'). It also includes the cigarette serial number ('f2'), machine speed during cigarette production ('f4'), negative pressure of the VE large fan ('f34'), pressure of the VE small fan ('f35'), VE primary air separation pressure ('f36'), amount of recycled tobacco ('f38'), actual position of the suction ribbon ('f39'), actual weight of the cigarette stick ('f95'), cigarette segment value 1 ('f97'), cigarette segment value 2 ('f98'), and cigarette segment value 3 ('f97'). The input features include 26 other process parameters related to cigarette rolling and packaging, such as cigarette segment value 4 ('f100'), cigarette segment value 5 ('f101'), circumference value ('f104'), tobacco temperature ('f105'), tobacco moisture ('f106'), MAX total air source pressure ('f131'), MAX large fan negative pressure ('f132'), MAX paper cutting roller negative pressure ('f133'), cigarette empty end detection value ('f156'), cigarette OTIS detection value ('f161'), suction resistance value ('f173'), ventilation degree ('f174'), air tightness value ('f175'), waste rejection word ('f180'), needle roller feeding coefficient ('f197'), etc., for a total of 32 input features. The final input data is a high-quality dataset with a size of 32 × 22698 (feature dimension × number of samples), and the data results are as follows: the distribution of empty cigarette counts is as follows. Figure 2 As shown, the distribution of multi-dimensional input feature variables is as follows: Figure 3 As shown.
[0027] Step S11 is used to preprocess the multidimensional cigarette manufacturing process parameter data. In this embodiment, the specific method for preprocessing the multidimensional cigarette manufacturing process parameter data in step S11 can be of various forms known to those skilled in the art. In one example of the present invention, step S11 may include, for example... Figure 4 The steps shown are described in this. Figure 4 In this context, step S11 may include: In step S20, outliers are detected and removed; In step S21, a data consistency check is performed; In step S22, the missing process parameter features are filled in using the multiple interpolation method; In step S23, the data is standardized using standard deviation to eliminate dimensional differences.
[0028] In such Figure 4 In the method shown, step S20 is used for outlier detection and removal. Specifically, in this example, it can be based on the "waste removal" item in the multidimensional cigarette process parameter data to remove invalid cigarette samples that do not meet industry testing standards. Step S21 is used for data consistency verification. In this example, it can be used to check the cigarette serial number and production time correlation features, remove samples with duplicate serial numbers or contradictory time logic, and ensure the uniqueness and consistency of the dataset. Step S22 is used to fill in the missing process parameter features using multiple interpolation. A regression model is constructed with the features without missing features as predictors and the missing features as target variables. After iterating the interpolation 5 times, the mean is taken as the filled value. Step S23 is used for data standardization. Standard deviation standardization is used to eliminate dimensional differences. In this example, it can be standardized using formula (6) for example: (6) in, The result is the standard deviation. These are the original eigenvalues. The mean of this feature for the training set, The standard deviation of this feature in the training set.
[0029] Step S12 is used to construct a cigarette empty value prediction model. In this embodiment, a deep neural network model can be constructed, which includes an input layer, multiple hidden layers, and an output layer. The number of neurons in the input layer corresponds to the feature dimension of the multidimensional cigarette manufacturing parameters, and it receives standardized data preprocessed in step S11, with the data type being floating-point. The hidden layers consist sequentially of a linear transformation layer, a batch normalization layer, a LeakyReLU activation function layer, and a Dropout layer. The output layer is a single neuron that outputs the cigarette empty value as the prediction result. Specifically, in this example, the first hidden layer uses 256 neurons, the second hidden layer uses 128 neurons, and the third hidden layer uses 64 neurons. The momentum parameter of the batch normalization layer is set to 0.99, the negative slope parameter of the LeakyReLU activation function is set to 0.1, and the dropout rate of the Dropout layer is set to 0.3.
[0030] Step S13 is used to construct a loss function based on physical information. Based on the model constructed in step S12, a physical constraint loss function based on cigarette manufacturing process information is configured. This physical constraint loss function is composed of a weighted sum of the basic prediction loss term, the port density loss term, and the compaction end distance loss term. Specifically, it can be constructed using, for example, formula (1): (1) in, The loss function is based on physical information. Based on the predicted loss term, For port density loss term, For the distance loss at the compaction end, for The weighting coefficients, for The weighting coefficients. In this example, the weighting coefficients... It can be set to 0.1. It can be set to 0.01.
[0031] In this example, the basic prediction loss term is obtained according to formula (2): (2) in, Based on the predicted loss term, This represents the actual value of the empty cigarette count. This is the predicted value for short selling. The loss function is segmented with threshold values. Based on the predicted values obtained from the network in step S12, the basic prediction loss term is calculated to guide the model in correcting biases during training, ensuring that the prediction results are consistent with the true values.
[0032] In this example, the port density loss term is obtained according to formula (3): (3) in, For port density loss term, This represents the average density of the first n segments before combustion in the multidimensional cigarette manufacturing process parameter data. The standard density value is automatically extracted and corrected based on the training data. Set the maximum allowed short position value according to the actual situation. This is the scaling factor. In this example, the scaling factor is... It can be set to 0.05; maximum allowed short position value. It can be set to 80; the calculation method for the port density loss term can be to calculate the average density value of the first 6 segments of the cigarette combustion end obtained from step S11 cleaning, and then calculate the port density loss term. The port density loss term reflects the overall compactness of the port area. If the model prediction result violates the physical law that "the higher the density, the lower the void value", this constraint term will produce a positive loss. By controlling the magnitude of the loss value, it can be ensured that the prediction result is consistent with the physical mechanism.
[0033] In this example, the compaction end distance loss term is obtained according to formula (4): (4) in, For the distance loss at the compaction end, This represents the deviation value of the distance between the compacted ends. This is the loss magnitude coefficient; This is a nonlinear influence coefficient, and its value is obtained by fitting experimental process data on the impact of compaction end deviation on the void value. The absolute value of the compaction end distance obtained in step S11 (cleaning) is taken to calculate the compaction end distance loss term. This term reflects the magnitude of the compaction end position deviation of the cigarette by taking the absolute value of the compaction end distance. If the model prediction result violates the physical law that "under the same density distribution, the larger the compaction end distance deviation, the higher the void value," then this constraint term will generate a positive loss. By controlling the magnitude of the loss value, the model is guided to learn a mapping relationship that conforms to physical common sense. In this example, the loss magnitude coefficient... It can be set to 0.02, the nonlinear influence coefficient. It can be set to 0.1.
[0034] Step S14 is used to train the cigarette empty value prediction model using preprocessed multidimensional cigarette manufacturing process parameter data based on a loss function based on physical information, in order to obtain the optimal feature set. In this embodiment, the specific method for training the cigarette empty value prediction model in step S14 can be of various forms known to those skilled in the art. In one example of the present invention, step S14 may include, for example... Figure 5 The steps shown are described in this. Figure 5 In this context, step S14 may include: In step S30, the multidimensional cigarette manufacturing process parameter dataset is input into the cigarette empty value prediction model; In step S31, the global importance score of each feature in the multidimensional cigarette manufacturing process parameter dataset is calculated; In step S32, features with global importance scores below a preset threshold are removed, the multidimensional cigarette manufacturing process parameter dataset is updated, and the number of remaining features is calculated. In step S33, it is determined whether the number of remaining features is less than or equal to a preset minimum value; In step S34, if the number of remaining features is less than or equal to a preset minimum value, the model training is completed. If the number of remaining features is greater than the preset minimum value, return to the step of inputting the multidimensional cigarette process parameter dataset into the cigarette empty value prediction model; In step S35, the predictive ability of each model is evaluated using model evaluation metrics to obtain the optimal feature set.
[0035] In such Figure 5 In the method shown, step S30 is used to train the cigarette empty value prediction model using a multi-dimensional cigarette manufacturing parameter dataset to improve prediction accuracy. Specifically, in this example, the number of training iterations per round can be set to 200, the model can use the Adam optimizer, and the learning rate can be 0.001.
[0036] Step S31 is used to calculate the global importance score of each feature in the multidimensional cigarette manufacturing process parameter dataset. Specifically, in this example, the global importance score can be calculated, for example, using formula (5): (5) in, As a score for overall importance, Features In the sample On value, This represents the expected value of the sample. A higher global importance score indicates a greater contribution of the feature to the model's prediction results.
[0037] In step S32, features with global importance scores below a preset threshold are removed, the multidimensional cigarette manufacturing process parameter dataset is updated, and the number of remaining features is calculated. In this example, an adaptive thresholding method can be used to prioritize the removal of completely non-contributing features with an importance score of 0. The performance of the cigarette empty value prediction model changes accordingly, and the loss is verified as follows. Figure 6 As shown. Then, those with scores below the threshold are removed. The low importance of the feature, among which The design uses the lower quartile, retaining the core feature subset and then retraining the model using the remaining feature subset.
[0038] Steps S33 and S34 are used to cyclically execute the process until the number of remaining features reaches a preset minimum value. In this example, the preset minimum value for the number of remaining features can be 1, and the loss function constructed based on physical information changes during the iteration process as follows: Figure 7 As shown, the overall feature selection process is as follows: Figure 8 As shown.
[0039] In training the cigarette empty value prediction model, multiple models and their corresponding feature combinations were obtained by continuously eliminating features. To obtain the optimal feature set, step S35 is needed to evaluate the predictive ability of each model using model evaluation metrics to obtain the model with the highest score. The feature combination corresponding to this model is the optimal feature set. In this example, by comparing the evaluation metrics of the training process, including the validation set loss function value and the prediction accuracy metric, the feature combination with the best prediction accuracy ['f174', 'f175', 'segment 1', 'f98', 'f97', 'f100', 'f161', 'f101', 'f173'] is determined as the final feature set. The change in feature importance in the optimal feature set is as follows: Figure 9 As shown.
[0040] Step S15 is used to predict the cigarette's empty value using the cigarette empty value prediction model corresponding to the optimal feature set, thereby obtaining the prediction result. In this example, the physical feature data of the cigarette to be predicted is input into the trained cigarette empty value prediction model corresponding to the optimal feature set, and the empty value prediction result and evaluation index of the cigarette are output. The line graph comparing the prediction result and the actual value is shown below. Figure 10 As shown, the evaluation indicators for the model prediction results are as follows: Figure 11 As shown, RMSE is 6.5710, MAE is 4.6774, MAPE is 4.1596%, and R² is 0.9585. Meanwhile, as... Figure 12 The output shows the contribution value of each feature, providing interpretive analysis of the prediction results. This helps operators make quality warnings and process adjustments based on the prediction results and feature analysis.
[0041] On the other hand, the present invention also provides a cigarette empty value prediction system, the system including a processor configured to perform any of the methods described in the cigarette empty value prediction method.
[0042] In another aspect, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement any of the methods described in the cigarette empty value prediction method.
[0043] The beneficial effects of this invention are: The present invention proposes a method for predicting the empty value of cigarettes based on a physical information neural network. This method can deeply integrate the prior physical knowledge of the tobacco industry into the model learning process, so that the prediction results not only conform to the statistical laws of data but also follow the physical mechanism, which significantly improves the prediction accuracy and model reliability.
[0044] The iterative feature selection mechanism used in this invention can automatically identify the most relevant feature subsets, reducing model complexity and overfitting risk, and improving computational efficiency.
[0045] The prediction method provided by this invention has good interpretability, and the prediction results can be directly linked to specific physical characteristics, providing clear guidance for process improvement.
[0046] The embodiments of this invention achieve a prediction accuracy of over 95% on an independent test set, providing an innovative quality inspection method for the tobacco industry. This method can predict the location and probability of missing cigarettes in advance, helping operators to adjust processing techniques in a timely manner, preventing the processing from getting out of control, and is of great significance for reducing product quality fluctuations and processing costs, as well as reducing the rate of defective cigarettes.
[0047] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0048] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0049] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0050] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0051] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0052] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0053] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0054] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0055] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for predicting the empty value of cigarettes, characterized in that, The prediction method includes: Obtain multi-dimensional cigarette manufacturing process parameter data; The multidimensional cigarette manufacturing process parameter data are preprocessed; Construct a cigarette empty value prediction model; Construct a loss function based on physical information; Based on the loss function based on physical information, the cigarette empty value prediction model is trained using preprocessed multidimensional cigarette manufacturing process parameter data to obtain the optimal feature set. The cigarette blank value prediction model corresponding to the optimal feature set is used to predict the cigarettes to be predicted in order to obtain the prediction results.
2. The prediction method according to claim 1, characterized in that, Preprocessing of the multidimensional cigarette manufacturing process parameter data includes: Detect and remove outliers; Perform data consistency verification; The missing process parameter features are filled in using a multiple interpolation method; Standard deviation is used to standardize the data and eliminate dimensional differences.
3. The prediction method according to claim 1, characterized in that, Constructing loss functions based on physical information includes: Construct a loss function based on physical information according to formula (1). ，（1） in, The loss function is based on physical information. Based on the predicted loss term, For port density loss term, For the distance loss at the compaction end, for The weighting coefficients, for The weighting coefficients.
4. The prediction method according to claim 3, characterized in that, The basic prediction loss term is obtained according to formula (2): ，（2） in, Based on the predicted loss term, This represents the actual value of the empty cigarette count. This is the predicted value for short selling. The threshold for segmentation of the loss function.
5. The prediction method according to claim 3, characterized in that, The port density loss term is obtained according to formula (3): ，（3） in, For port density loss term, This represents the average density of the first n segments before combustion in the multidimensional cigarette manufacturing process parameter data. This is the standard density value. This is the predicted value for short selling. The maximum allowed short position value. This is the proportionality coefficient.
6. The prediction method according to claim 3, characterized in that, The compaction end distance loss term is obtained according to formula (4): ，（4） in, For the distance loss at the compaction end, This represents the deviation value of the distance between the compacted ends. This is the loss magnitude coefficient. This is a nonlinear influence coefficient. This is the predicted value for short selling. This represents the maximum allowed short position value.
7. The prediction method according to claim 1, characterized in that, Based on the loss function based on physical information, the cigarette empty value prediction model is trained using preprocessed multidimensional cigarette manufacturing parameter data to obtain the optimal feature set, including: Input the multidimensional cigarette manufacturing parameter dataset into the cigarette empty value prediction model; Calculate the global importance score for each feature in the multidimensional cigarette manufacturing parameter dataset; Features with global importance scores below a preset threshold are removed, the multidimensional cigarette manufacturing process parameter dataset is updated, and the number of remaining features is calculated. Determine whether the number of remaining features is less than or equal to a preset minimum value; If the number of remaining features is less than or equal to a preset minimum value, the model training is completed. If the number of remaining features is greater than the preset minimum value, return to the step of inputting the multidimensional cigarette process parameter dataset into the cigarette empty value prediction model; The predictive power of each model is evaluated using model evaluation metrics to obtain the optimal feature set.
8. The prediction method according to claim 7, characterized in that, Calculating the global importance score of each feature in the multidimensional cigarette manufacturing parameter dataset includes: Calculate the global importance score according to formula (5). ，（5） in, As a score for overall importance, Features In the sample On value, This represents the expected value of the sample.
9. A cigarette empty value prediction system, characterized in that, The system includes a processor configured to perform the method as described in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a processor, implement the method as described in any one of claims 1 to 8.