Adaptive delay-aware process quality continuous prediction method and model based on ada-net, terminal and storage medium

Abstract

The application relates to the technical field of intelligent manufacturing, in particular to an adaptive delay perception process quality continuous prediction method and model based on an ADA-Net, a terminal and a storage medium. Multivariate time series data of a loose rehydration process in a cigarette silk making workshop are collected to form a sample data set; the sample data is preprocessed; the preprocessed data is divided into a training set, a verification set and a test set in chronological order; based on the training set, lag correlation analysis is carried out, the optimal lag influence of each process parameter on the moisture of the outlet material is identified, and lag features and statistical enhancement features are constructed; on this basis, an ADA-Net prediction model integrating feature reconstruction, Transformer time series coding, time constraints and autoregressive recursive mechanism is built; the training set and the verification set are used to complete model training, and the outlet material moisture is continuously and multi-step predicted through a sliding window recursive method. The application aims to solve the problem of how to realize key quality index prediction in a continuous process production process.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing technology, and in particular to an adaptive delay-sensing process quality continuous prediction method, model, terminal and storage medium based on ADA-Net. Background Technology

[0002] Leveraging industrial big data to predict key quality indicators in advance has become an important direction for improving the intelligence level of production processes. In continuous process production, quality indicators are typically influenced by multiple process parameters such as raw material state, temperature, humidity, pressure, and flow rate, with complex nonlinear coupling relationships between these variables. Furthermore, the effects of different process parameters on quality indicators do not occur synchronously but exhibit differentiated, dynamic, and time-lag characteristics. Taking the loosening and rehydration process in cigarette manufacturing as an example, the moisture content of the outlet material is affected by factors such as the moisture content of the inlet material, hot air temperature, steam pressure, and cylinder wall temperature, exhibiting significant time lag, fluctuation, and non-stationarity, making it difficult to perceive and accurately predict quality variables in a timely manner.

[0003] While existing statistical modeling, machine learning, and deep learning methods have been applied to quality prediction in the process industry, they still do not adequately address issues such as multivariate differential lag response, dynamic time delay adaptability, and error accumulation and time misalignment in continuous multi-step rolling prediction.

[0004] Therefore, constructing a continuous multi-step quality prediction method with adaptive delay perception capability can help improve the accuracy of quality indicator prediction in advance and provide technical support for forward-looking quality perception, process parameter control and intelligent decision-making in the process industry. Summary of the Invention

[0005] The main objective of this application is to provide an adaptive delay-aware continuous prediction method for process quality based on ADA-Net, which aims to solve the problem of how to predict key quality indicators in continuous process production.

[0006] To achieve the above objectives, this application provides an adaptive delay-aware continuous prediction method for process quality based on ADA-Net, the method comprising:

[0007] Multivariate time-series data of the loosening and rehydration process in the cigarette processing workshop were collected to form a sample dataset.

[0008] The sample dataset is preprocessed to obtain the preprocessed dataset;

[0009] The preprocessed dataset is divided into training set data, validation set data and test set data according to the time sequence.

[0010] Based on the lag correlation analysis, the delayed impact features are extracted and combined with statistical enhancement features to construct the ADA-Net prediction model that integrates Transformer time series modeling, time constraints and autoregressive recursion mechanism;

[0011] The constructed ADA-Net model is trained based on the training set data and validation set data, and the hyperparameters are adjusted until they meet the expectations before the model is saved.

[0012] The trained ADA-Net prediction model was used to predict time series data of the loose rehydration process production line.

[0013] Optionally, the constituent sample dataset specifically includes:

[0014] Based on the specific process flow of the loose rehydration process in the cigarette making workshop, the process parameters and quality indicators are determined.

[0015] Based on the process parameters and quality indicators, time series data at a preset time are collected to form a sample dataset.

[0016] Optionally, the preprocessing includes:

[0017] Organize the field names, retaining only numeric variables for modeling purposes;

[0018] For any missing values ​​that may exist in the data, a whole row removal strategy is adopted to delete samples where any variable is missing, and retain data where all process parameters and quality indicators are complete;

[0019] The Min-Max normalization method was used for scaling transformation to complete the preprocessing.

[0020] Optionally, the extraction of delay effect features based on lag correlation analysis includes:

[0021] Based on the training set data, the Pearson correlation coefficient between each process parameter and the moisture content of the outlet material under different lag steps was calculated, and the optimal lag step and maximum correlation strength of each variable were identified.

[0022] Strongly correlated variables are selected based on the correlation threshold, and differentiated lag features are constructed by combining the preset lag step size and the default short-term lag set. At the same time, the moving mean, moving standard deviation and first-order difference features are introduced to form delayed-perception input features.

[0023] Optionally, the ADA-Net prediction model, during execution, satisfies the following steps:

[0024] The multivariate time series samples after preprocessing and delay-aware feature reconstruction are used as input;

[0025] Based on the results of lag correlation analysis, the process parameters are subjected to differentiated lag expansion, so that different variables are time-aligned with the moisture content of the outlet material under their effective lag step. Combined with statistical enhancement features such as moving mean, moving standard deviation and first difference, a delayed sensing input feature space is formed.

[0026] A fixed-length input sequence is fed into a Transformer-based temporal coding module, which enhances the ability to express variable features and temporal order through linear embedding and positional coding.

[0027] A time-band structure constraint is introduced in the attention calculation process to make the model focus on the effective historical interval, reduce interference from irrelevant information at long distances, and adaptively measure the contribution of different historical moments to the current quality state through a dynamic weight allocation mechanism to strengthen the expression of key lag information.

[0028] An autoregressive recursive method is used for continuous multi-step prediction, which gradually predicts the moisture content of the outlet material at the next moment and fills the predicted value back into the target variable position of the input sequence. Subsequent prediction results are generated by updating through a sliding window, thereby simulating the rolling prediction process in continuous industrial production. The predicted result of the outlet material moisture content under the actual physical value is obtained after inverse normalization.

[0029] Furthermore, to achieve the above objectives, this application also provides a cigarette manufacturing process quality prediction model based on the ADA-Net model, comprising:

[0030] The collection module is used to collect multivariate time-series data of the loosening and rehydration process in the cigarette processing workshop to form a sample dataset;

[0031] The preprocessing module is used to preprocess the sample dataset to obtain the preprocessed dataset;

[0032] The partitioning module is used to divide the preprocessed dataset into training set data, validation set data, and test set data according to time sequence;

[0033] The feature reconstruction module is used to extract the delay effect features based on lag correlation analysis, and combine them with statistical enhancement features to construct the ADA-Net prediction model that integrates Transformer time series modeling, time constraints and autoregressive recursion mechanism;

[0034] The training module is used to train the constructed ADA-Net model based on the training set data and validation set data, and adjust the hyperparameters until the model meets the expectations and is then saved.

[0035] The prediction module is used to predict time series data of the loose rehydration process production line using the trained ADA-Net prediction model.

[0036] In addition, to achieve the above objectives, this application also provides a terminal, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to perform the adaptive delay-aware continuous prediction method for process quality based on the ADA-Net model as described above.

[0037] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the adaptive delay-aware continuous prediction method for process quality based on the ADA-Net model as described in any of the preceding claims.

[0038] This application has at least the following beneficial effects:

[0039] 1. The loosening and rehydration process in cigarette manufacturing involves multiple dynamically changing process parameters and key quality indicators. Furthermore, the impact of different process parameters on the moisture content of the effluent exhibits nonlinear coupling, dynamic time lag, and asynchronous response. Traditional prediction methods struggle to fully characterize the differentiated lag relationships between multiple variables, easily leading to insufficient accuracy and poor stability in continuous multi-step predictions, making it difficult to meet the needs of early quality perception and process control in actual production.

[0040] 2. By using lag correlation analysis to identify the optimal lag step size and correlation strength between different process variables and the moisture content of the outlet material, and combining differentiated lag expansion, default lag supplementation, and sliding statistical feature enhancement, a delayed-aware input feature space is constructed, which can more fully express the lag effect of process parameters on quality indicators. At the same time, a time band structure constraint and a dynamic weight allocation mechanism are introduced to make the model focus on key lag information within the effective historical interval, reduce interference from distant irrelevant historical information, and improve the model's ability to model complex time series dependencies and dynamic time lag relationships.

[0041] 3. An autoregressive recursive prediction mechanism is adopted, which gradually generates multi-step prediction results through predicted value backfilling and sliding window updates, making the prediction process closer to the rolling prediction scenario in continuous industrial production. This enables continuous multi-step advance prediction of the moisture content of exported materials, providing data support for online monitoring, forward-looking quality perception, and process parameter control of the loosening and rehydration process in cigarette manufacturing, and helping to improve product quality stability and the level of intelligent production processes. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the adaptive delay-sensing continuous prediction method for process quality based on the ADA-Net model involved in the embodiments of this application.

[0043] Figure 2This is a schematic diagram of the ADA-Net prediction model structure involved in the embodiments of this application;

[0044] Figure 3 This is a structural diagram of the delay-aware feature reconstruction module involved in the embodiments of this application;

[0045] Figure 4 This is a schematic diagram of delay-aware timing modeling of time-band structure constraints involved in an embodiment of this application;

[0046] Figure 5 This is a schematic diagram illustrating the autoregressive recursive modeling and continuous rolling prediction involved in the embodiments of this application;

[0047] Figure 6 This is a comparison chart of the predicted and actual values ​​of moisture content in exported materials involved in the embodiments of this application;

[0048] Figure 7 This is a scatter plot showing the fitting between the predicted and actual values ​​of moisture content forecasting for exported materials in the embodiments of this application.

[0049] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0051] First Embodiment

[0052] like Figure 1-5 As shown, according to a first aspect of this application, a method for continuous prediction of process quality based on the ADA-Net model is provided, comprising:

[0053] S10: Collect multivariate time-series data of the loosening and rehydration process in the cigarette processing workshop to form a sample dataset;

[0054] Furthermore, based on the specific process of the loosening and rehydration process in cigarette manufacturing, a sample dataset is constructed by collecting multivariate time-series data during the operation of the production line. The sample dataset includes process parameter data and quality index data. There are multiple process parameters, covering production process variables such as temperature, pressure, flow rate, and environmental parameters. There is one quality index, namely the actual moisture content of the outlet material.

[0055] S20, preprocess the sample dataset to obtain the preprocessed dataset;

[0056] S30. According to the time sequence, the preprocessed dataset is divided into training set data, validation set data and test set data;

[0057] Furthermore, the preprocessing first organizes the field names, retaining only numerical variables for modeling to ensure the consistency of multivariate time series data. To address potential missing values, a row-wide removal strategy is employed, deleting samples with missing values ​​for any variable and retaining valid data with complete process parameters and quality indicators. Subsequently, the data is divided into training, validation, and test sets according to time sequence. Finally, a Min-Max normalization method is used for scaling.

[0058] The calculation formula for the Min-Max normalization method is as follows:

[0059]

[0060] in, For normalized variables, The original data, The maximum value, It is the minimum value.

[0061] The lag correlation feature screening and reconstruction includes: based on training set data, calculating the Pearson correlation coefficient between each process parameter and the moisture content of the outlet material under different lag steps, and identifying the optimal lag step and maximum correlation strength for each variable. Subsequently, strongly correlated variables are screened according to the correlation threshold, and differentiated lag features are constructed by combining the preset lag step and the default short-term lag set. At the same time, moving mean, moving standard deviation, and first-order difference features are introduced to form delayed-aware input features.

[0062] S40, based on lag correlation analysis to extract the delay effect features, combined with statistical enhancement features, to construct the ADA-Net prediction model that integrates Transformer time series modeling, time constraints and autoregressive recursion mechanism;

[0063] S50: Based on the training set data and validation set data, the constructed ADA-Net model is trained and the hyperparameters are adjusted until they meet the expectations, and then the model is saved.

[0064] S60 uses the trained ADA-Net prediction model to predict time series data of the loose rehydration process production line.

[0065] Furthermore, the ADA-Net prediction model takes the preprocessed and delayed sensing feature reconstructed multivariate time series samples as input. First, it performs differentiated lag expansion on the process parameters based on the lag correlation analysis results, so that different variables are time-aligned with the moisture content of the outlet material under their effective lag step length. Then, it combines statistical enhancement features such as moving mean, moving standard deviation and first difference to form a delayed sensing input feature space.

[0066] The formula for calculating the coefficients in the lag correlation analysis is as follows:

[0067]

[0068]

[0069] The maximum lag step size is The process variables are The quality variable is The maximum lag correlation strength is .

[0070] Subsequently, the fixed-length input sequence is fed into the Transformer-based temporal coding module, which enhances the ability to express variable features and temporal order through linear embedding and positional coding.

[0071] The formula for dividing process variables is as follows:

[0072]

[0073]

[0074] The formula for the delayed-aware input feature is expressed as follows:

[0075] in, This represents the set of differentially lagged features of strongly correlated variables. This represents the set of short-term lagged characteristics of weakly correlated variables.

[0076] Next, a time-band structure constraint is introduced into the attention calculation process to make the model focus on the effective historical interval, reduce interference from irrelevant information from a distance, and adaptively measure the contribution of different historical moments to the current quality state through a dynamic weight allocation mechanism, thereby strengthening the expression of key lagging information.

[0077] The calculation formula for the band constraint is as follows:

[0078]

[0079] Among them, when "Time" indicates a historical moment. Can participate in the current moment Information exchange; when When the time is specified, it indicates that the current moment is masked and does not participate in the modeling of the current state.

[0080] The calculation formula for the dynamic weight allocation is as follows:

[0081] in, This indicates the weight of the influence of that historical moment on the current moment.

[0082] The formula for calculating the time series at the current moment is as follows:

[0083] in, Representing historical moments The value vector.

[0084] The model then uses an autoregressive recursive approach to make continuous multi-step predictions and backfills the predicted values ​​into the target variable position of the input sequence. It then generates subsequent prediction results by updating through a sliding window, thereby simulating the rolling prediction process in continuous industrial production. Finally, the prediction result of the moisture content of the export material under the actual physical value is obtained through inverse normalization.

[0085] The formula for updating the input via the sliding window is as follows:

[0086] Among them, the historical sequence is , No. Step prediction .

[0087] Specifically, the ADA-Net model training employs a composite loss function as the optimization objective. This composite loss function is based on Huber loss and incorporates full-sequence prediction loss, final-step prediction weighted loss, target value weighted error term, and prediction range penalty term to improve the stability of multi-step predictions, especially the final-step prediction. During training, the AdamW optimizer is used to update the model parameters, and a learning rate scheduling strategy is combined to improve the model's convergence stability.

[0088] Furthermore, the final parameters of the model are set as follows: input step size is 60, prediction step size is 5, training epochs are 300, initial learning rate is set to 8e-5, batch size is 64, and weight decay coefficient is 1×10⁻⁵. -6The model employs a Transformer-based autoregressive prediction structure for multivariate time-series modeling. Input features are first mapped to a 768-dimensional representation space via linear embedding, and positional encoding is introduced to enhance temporal sequence representation. The temporal encoding part uses 12 attention heads, 4 encoding layers, a 3072-dimensional feedforward neural network, a Dropout factor of 0.02, and GeLU activation. A temporal banding constraint is introduced into the attention mechanism. The maximum receptive range on the left side of the constraint is set to 60, adaptively truncated based on the input sequence length, while the receptive range on the right side is set to 0, allowing the model to focus on valid historical information prior to the current moment. The loss function uses a composite loss form, based on Huber loss, and includes full-sequence prediction loss, a final-step weighted loss, a target value weighted error term, and a prediction range penalty term. The full-sequence loss weight is 1.0, the final-step loss weight is 2.0, the Huber parameter is 0.2, and the prediction range penalty coefficient is 0.02. The model ultimately uses an autoregressive recursive approach to progressively output the predicted moisture content of the export material for the next 5 steps.

[0089] Specifically, the evaluation metrics include percentage absolute error (MAPE), mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²). 2 ).

[0090] Furthermore, the calculation formula for the evaluation index is as follows:

[0091]

[0092]

[0093]

[0094]

[0095] in, This is a predicted value; average value; The actual value is represented by n; the total sample size is n. The smaller the MAPE, MSE, and MAE values, the closer the predicted value is to the actual value; R0 2 The closer to 1, the higher the model fit.

[0096] Verification of Examples

[0097] Based on the first embodiment, referring to Figure 6-7 In this embodiment, the method proposed above is verified. Specifically, the experimental data comes from the production line time-series data collected on-site during the cigarette manufacturing loosening and rehydration process. The sampling period is 1 second, and 33,470 data points are selected as the sample dataset. An example of the dataset is shown in Table 1.

[0098] Table 1. Partial Original Data of the Dataset

[0099]

[0100] Second, the field names were first organized, retaining only numerical variables for modeling to ensure consistency in multivariate time series data. To address potential missing values, a row-wide removal strategy was employed, deleting samples with missing values ​​for any variable and retaining valid data for all process parameters and quality indicators. Subsequently, the data was divided into training, validation, and test sets according to time sequence. Finally, Min-Max normalization was used for scaling.

[0101] Third, the preprocessed dataset is divided into training, validation, and test sets in a 6:2:2 ratio. The partitioning maintains a chronological order to simulate real-world prediction scenarios.

[0102] IV. The constructed ADA-Net prediction model includes: a delay-aware feature reconstruction module, a temporal band structure constraint module, and an autoregressive recursive prediction module, such as... Figure 2 As shown, the constructed ADA-Net model is trained using the training and validation sets, and its hyperparameters are adjusted. The trained model is then used to predict test set data or actual production time series data, and the model's performance is validated using multiple evaluation metrics.

[0103] To verify the predictive performance of the model, the ADA-Net model provided in this application was compared with traditional algorithms such as KAN, TCN, CNN-BigRu, GRU-Attention, and seq2seq-LSTM-LSTM. The deep learning model parameters were configured according to the model settings in this application. Finally, percentage error (MAPE), mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²) were used as the statistical measures. 2 As shown in Table 2, the predicted and actual values ​​of the model test set in this application are plotted as comparative analysis indicators.

[0104] Table 2. Experimental Prediction Results

[0105]

[0106] Overall, the ADA-Net model demonstrated the best prediction performance in MAE, MSE, MAPE, and R. 2The model outperformed other comparative models in all four metrics, demonstrating its strong adaptability to complex and multi-scale process industry data. In the specific scenario of predicting the moisture content of the output material in the loose rehydration process, it can more comprehensively model the relationship between features and time series, thereby achieving more accurate quality prediction.

[0107] Furthermore, this embodiment also relates to a cigarette manufacturing process quality prediction model based on the ADA-Net model, including:

[0108] The collection module is used to collect multivariate time-series data of the loosening and rehydration process in the cigarette processing workshop to form a sample dataset;

[0109] The preprocessing module is used to preprocess the sample dataset to obtain the preprocessed dataset;

[0110] The partitioning module is used to divide the preprocessed dataset into training set data, validation set data, and test set data according to time sequence;

[0111] The feature reconstruction module is used to extract the delay effect features based on lag correlation analysis, and combine them with statistical enhancement features to construct the ADA-Net prediction model that integrates Transformer time series modeling, time constraints and autoregressive recursion mechanism;

[0112] The training module is used to train the constructed ADA-Net model based on the training set data and validation set data, and adjust the hyperparameters until the model meets the expectations and is then saved.

[0113] The prediction module is used to predict time series data of the loose rehydration process production line using the trained ADA-Net prediction model.

[0114] Furthermore, this embodiment also relates to a terminal, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor is configured as described above for the adaptive delay-aware continuous prediction method for process quality based on the ADA-Net model.

[0115] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the embodiments of the above methods.

[0116] Therefore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the adaptive delay-aware continuous prediction method for process quality based on the ADA-Net model as described in the above embodiments.

[0117] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0118] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.

[0119] 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.

[0120] 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.

[0121] 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.

[0122] 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 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0123] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0124] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An adaptive delay-sensing continuous prediction method for process quality based on ADA-Net, characterized in that, The method includes the following steps: Multivariate time-series data of the loosening and rehydration process in the cigarette processing workshop were collected to form a sample dataset. The sample dataset is preprocessed to obtain the preprocessed dataset; The preprocessed dataset is divided into training set data, validation set data, and test set data according to the time sequence. Based on the lag correlation analysis, the delayed impact features are extracted and combined with statistical enhancement features to construct the ADA-Net prediction model, which integrates Transformer time series modeling, time constraints and autoregressive recursion mechanism. The constructed ADA-Net model is trained based on the training set data and validation set data, and the hyperparameters are adjusted until they meet the expectations before the model is saved. The trained ADA-Net prediction model was used to predict time series data of the loose rehydration process production line.

2. The adaptive delay-aware continuous prediction method for process quality based on ADA-Net according to claim 1, characterized in that, The sample dataset is specifically comprised of: Based on the specific process flow of the loose rehydration process in the cigarette making workshop, the process parameters and quality indicators are determined. Based on the process parameters and quality indicators, time series data at a preset time are collected to form a sample dataset.

3. The adaptive delay-aware continuous prediction method for process quality based on ADA-Net according to claim 1, characterized in that, The preprocessing includes: Organize the field names, retaining only numeric variables for modeling purposes; For any missing values ​​that may exist in the data, a whole row removal strategy is adopted to delete samples where any variable is missing, and retain data where all process parameters and quality indicators are complete; The Min-Max normalization method was used for scaling transformation to complete the preprocessing.

4. The adaptive delay-aware continuous prediction method for process quality based on ADA-Net according to claim 1, characterized in that, The extraction of delay effect features based on lag correlation analysis includes: Based on the training set data, the Pearson correlation coefficient between each process parameter and the moisture content of the outlet material under different lag steps was calculated, and the optimal lag step and maximum correlation strength of each variable were identified. Strongly correlated variables are selected based on the correlation threshold, and differentiated lag features are constructed by combining the preset lag step size and the default short-term lag set. At the same time, the moving mean, moving standard deviation and first-order difference features are introduced to form delayed-perception input features.

5. The adaptive delay-aware continuous prediction method for process quality based on ADA-Net according to claim 1, characterized in that, The ADA-Net prediction model, during execution, satisfies the following steps: The multivariate time series samples after preprocessing and delay-aware feature reconstruction are used as input; Based on the results of lag correlation analysis, the process parameters are subjected to differentiated lag expansion, so that different variables are time-aligned with the moisture content of the outlet material under their effective lag step. The moving mean, moving standard deviation and first difference are combined as statistical enhancement features to form a delayed sensing input feature space. A fixed-length input sequence is fed into a Transformer-based temporal coding module, which enhances the ability to express variable features and temporal order through linear embedding and positional coding. A time-band structure constraint is introduced in the attention calculation process to make the model focus on the effective historical interval, reduce interference from irrelevant information at long distances, and adaptively measure the contribution of different historical moments to the current quality state through a dynamic weight allocation mechanism to strengthen the expression of key lag information. An autoregressive recursive method is used for continuous multi-step prediction, which gradually predicts the moisture content of the outlet material at the next moment and fills the predicted value back into the target variable position of the input sequence. Subsequent prediction results are generated by updating through a sliding window, thereby simulating the rolling prediction process in continuous industrial production. The predicted result of the outlet material moisture content under the actual physical value is obtained after inverse normalization.

6. A quality prediction model for cigarette manufacturing process, characterized in that, include: The collection module is used to collect multivariate time-series data of the loosening and rehydration process in the cigarette processing workshop to form a sample dataset; The preprocessing module is used to preprocess the sample dataset to obtain the preprocessed dataset; The partitioning module is used to divide the preprocessed dataset into training set data, validation set data, and test set data according to time sequence; The feature reconstruction module is used to extract the delay effect features based on lag correlation analysis, and combine them with statistical enhancement features to construct the ADA-Net prediction model that integrates Transformer time series modeling, time constraints and autoregressive recursion mechanism; The training module is used to train the constructed ADA-Net model based on the training set data and validation set data, and adjust the hyperparameters until the model meets the expectations and is then saved. The prediction module is used to predict time series data of the loose rehydration process production line using the trained ADA-Net prediction model.

7. A terminal, characterized in that: The method includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor is configured as the adaptive delay-aware continuous prediction method for process quality based on the ADA-Net model as described in any one of claims 1-5.

8. A computer-readable storage medium storing a program, characterized in that: When the program is executed by the processor, it enables the processor to implement the adaptive delay-aware continuous prediction method for process quality based on the ADA-Net model as described in any one of claims 1-5.

Images