Cigarette cut tobacco drying process quality multi-step prediction method based on MCF-TransGRU

Abstract

The invention discloses a cigarette cut tobacco drying process quality multi-step prediction method based on MCF-TransGRU, and the method comprises the steps: collecting the time sequence data of a thin plate cut tobacco drying process production line in a cigarette cut tobacco manufacturing workshop at a preset time, and forming a sample data set; preprocessing the sample data set to obtain a preprocessed data set; according to a time sequence, dividing the preprocessed data set into training set data, verification set data and test set data for model training and performance evaluation; an MCF-TransGRU multi-step prediction model is constructed through a multi-scale convolution fusion module, an improved Transform module and a gating circulation unit module based on a residual error; according to the training set data and the verification set data, the constructed MCF-TransGRU multi-step prediction model is trained, and hyper-parameters are adjusted; and utilizing the trained MCF-TransGRU multi-step prediction model to predict test set data / to-be-tested sheet cut tobacco drying process production line time sequence data. According to the method, quality index prediction results at multiple time points in the future can be provided through multi-step prediction, and guidance is provided for subsequent production regulation and control decisions.

Description

Technical Field

[0001] The invention relates to a multi-step prediction method for tobacco shred drying process quality based on MCF-TransGRU, and belongs to the field of tobacco processing quality index prediction. Background Art

[0002] The tobacco drying process is a key link in tobacco production. The process quality not only affects the taste and aroma of tobacco, but is also directly related to the market competitiveness of the product. The tobacco drying process involves multiple physical and chemical processes. During the operation, variables such as temperature, humidity, and airflow need to be constantly adjusted. Slight changes in these variables may lead to significant changes in the quality of the final product. Therefore, it is very important to accurately predict and control the process.

[0003] Since the parameters and environmental conditions in the process have strong time dependence, traditional prediction methods often have difficulty in fully capturing these complex time series information. Traditional time series prediction methods usually include autoregressive models, sliding average models, and basic regression analysis. These methods perform well in processing static data and linear relationships, but when faced with nonlinear and complex time series data, their prediction performance and accuracy are often limited. In particular, when dealing with multi-step prediction tasks, traditional methods are difficult to effectively capture long-term dependencies and nonlinear patterns in the data. Therefore, accurate multi-step prediction of the quality of tobacco drying process has become a bottleneck that restricts the further improvement of process quality and resource utilization, and needs to be solved urgently. Summary of the invention

[0004] The present invention provides a multi-step prediction method for the quality of tobacco shred drying process based on MCF-TransGRU, which is used to construct an MCF-TransGRU multi-step prediction model to realize the multi-step prediction of the time series of quality indicators of the tobacco shred drying process.

[0005] The technical solution of the present invention is:

[0006] According to a first aspect of the present invention, a multi-step prediction method for the quality of tobacco shredded tobacco drying process based on MCF-TransGRU is provided, comprising: collecting time series data of a thin-plate shredded tobacco drying process production line in a tobacco shredded tobacco making workshop at a preset time to form a sample data set; wherein the time series data includes process parameters and quality indicator data; preprocessing the sample data set to obtain a preprocessed data set; dividing the preprocessed data set into training set data, validation set data and test set data in chronological order for model training and performance evaluation; constructing an MCF-TransGRU multi-step prediction model through a multi-scale convolutional fusion module, an improved Transformer module and a residual-based gated recurrent unit module; training the constructed MCF-TransGRU multi-step prediction model based on the training set data and the validation set data, and adjusting hyperparameters; and predicting the test set data / time series data of the thin-plate shredded tobacco drying process production line to be tested using the trained MCF-TransGRU multi-step prediction model.

[0007] The sample data set is formed by: determining the process parameters and quality indicators according to the specific process of the thin plate drying process in the cigarette silk workshop; collecting time series data under preset time periods to form the sample data set according to the determined process parameters and quality indicators; wherein there are multiple process parameters and one quality indicator.

[0008] The preprocessing includes: firstly removing the head and tail of the sample data in the sample data set, then filling the missing values ​​in the data with the mean, and then eliminating the sample data using the 3sigma criterion; and normalizing the sample data set judged by the 3sigma criterion.

[0009] The MCF-TransGRU multi-step prediction model is used to take the sample data in the preprocessed data set as input according to different prediction step sizes, and send them to different parallel one-dimensional convolution kernel combinations for multi-scale feature extraction. The features extracted at each scale will be spliced ​​and fused as the output of the multi-scale convolution fusion module; then the improved Transformer module is input to further extract deeper features; the output of the improved Transformer module is input to the GRU network, and the residual module adds the input of the GRU network to the output after the nonlinear transformation of the GRU network, so that the GRU network can fit complex mapping functions more easily, which is used to capture the global temporal features in the sequence as the output of the residual-based gated recurrent unit module RGM; the output of the residual-based gated recurrent unit module passes through two fully connected layers, and the number of output neurons of the second fully connected layer is set according to the prediction step size, which is used to output the multi-step prediction results; the final output of the model is the prediction result of the quality index of the tobacco drying process at multiple time points in the future, and the prediction result is finally denormalized to obtain the final data that meets the requirements.

[0010] The improved Transformer module includes an improved Transformer layer. The improved Transformer layer uses a standard Transformer layer as a framework. On the framework, a simple linear layer is used to replace the Decoder structure in the standard Transformer layer.

[0011] According to a second aspect of the present invention, a multi-step prediction system for tobacco shred drying process quality based on MCF-TransGRU is provided, comprising: a collection module, used to collect time series data of a thin-plate shred drying process production line in a tobacco shred making workshop at a preset time to form a sample data set; wherein the time series data includes process parameters and quality indicator data; a preprocessing module, used to preprocess the sample data set to obtain a preprocessed data set; a division module, used to divide the preprocessed data set into training set data, validation set data and test set data in chronological order for model training and performance evaluation; a construction module, used to construct an MCF-TransGRU multi-step prediction model through a multi-scale convolution fusion module, an improved Transformer module and a residual-based gated recurrent unit module; a training module, used to train the constructed MCF-TransGRU multi-step prediction model based on the training set data and the validation set data, and adjust hyperparameters; a prediction module, used to predict the test set data / the time series data of the thin-plate shred drying process production line to be tested using the trained MCF-TransGRU multi-step prediction model.

[0012] According to a third aspect of the present invention, there is provided a terminal comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to execute any one of the above-mentioned multi-step prediction methods for the quality of tobacco drying process based on MCF-TransGRU.

[0013] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, which includes a stored program, which, when executed by a processor, enables the processor to implement any of the above-mentioned multi-step prediction methods for the quality of tobacco drying process based on MCF-TransGRU.

[0014] The beneficial effects of the present invention are as follows: the tobacco shred drying process involves multiple process parameters and quality indicators, and there are complex spatiotemporal relationships between these parameters and indicators. Traditional prediction methods are difficult to effectively capture this complex relationship, resulting in low multi-step prediction accuracy. The adjustment of process parameters has a direct and complex impact on the final quality indicators, and existing methods are difficult to simultaneously consider the comprehensive impact of multiple process parameters on quality indicators. The present invention captures the temporal characteristics of process parameters through convolution kernels of different scales, uses Transformer to adaptively capture long-term dependencies and key time step information, and finally integrates process parameter features through gated recurrent units, and realizes multi-step prediction through fully connected layers. The model is trained and hyperparameters are adjusted to enhance the learning ability of time series relationships. Through multi-step prediction, the model can provide quality indicator prediction results at multiple time points in the future, providing guidance for subsequent production control decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the process of the present invention;

[0016] Figure 2 A schematic diagram of the structure of the MCF-TransGRU multi-step prediction model constructed by the present invention;

[0017] Figure 3 This is a structural diagram of the multi-scale convolution fusion module of the present invention;

[0018] Figure 4 This is a diagram of the improved Transformer structure of the present invention;

[0019] Figure 5 The structure diagram of the residual-based gated recurrent unit module of the present invention;

[0020] Figure 6 This is a comparison diagram of the predicted value and the actual value of the moisture content of the outlet material when the prediction step length is 4 in the embodiment of the present invention;

[0021] Figure 7This is a comparison chart between the predicted value and the actual value of the moisture content of the outlet material when the prediction step length is 20 in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily without conflict.

[0023] Example 1: Figure 1-7 As shown, according to the first aspect of the embodiment of the present invention, a multi-step prediction method for the quality of tobacco shred drying process based on MCF-TransGRU is provided, including: collecting time series data of the thin plate shred drying process production line in the tobacco shred workshop at a preset time to form a sample data set; wherein the time series data includes process parameters and quality index data; preprocessing the sample data set to obtain a preprocessed data set; dividing the preprocessed data set into training set data, validation set data and test set data in chronological order for model training and performance evaluation; constructing an MCF-TransGRU multi-step prediction model through a multi-scale convolution fusion module MCF, an improved Transformer module and a residual-based gated recurrent unit module RGM; training the constructed MCF-TransGRU multi-step prediction model based on the training set data and the validation set data, and adjusting the hyperparameters; using the trained MCF-TransGRU multi-step prediction model to predict the test set data, and verifying the model through a variety of evaluation indicators. Further, the trained MCF-TransGRU multi-step prediction model is used to predict the time series data of the thin plate shred drying process production line to be tested.

[0024] Furthermore, the sample data set is formed by: determining the process parameters and quality indicators according to the specific process of the thin plate drying process in the cigarette silk making workshop; based on the determined process parameters and quality indicators, collecting time series data under preset time to form the sample data set; wherein, there are multiple process parameters and one quality indicator.

[0025] Furthermore, the process parameters include moisture content of inlet material, dehumidification damper opening, dehumidification air volume, hot air temperature, HT steam pressure, rear chamber damper opening, inlet thin plate steam pressure, front chamber damper opening, cylinder wall temperature pressure, and inlet material flow rate; the quality indicator is the moisture content of outlet material.

[0026] Furthermore, the preprocessing includes: first removing the head and tail of the sample data in the sample data set, then filling the missing values ​​in the data with the mean, and then eliminating the sample data using the 3sigma criterion; and normalizing the sample data set judged by the 3sigma criterion.

[0027] Furthermore, the MCF-TransGRU multi-step prediction model is used to take the sample data in the preprocessed data set as input according to different prediction step sizes, and send them to different parallel one-dimensional convolution kernel combinations for multi-scale feature extraction. The one-dimensional convolution layer of each scale uses convolution kernels of different sizes, which can extract features from different time ranges at the same time, thereby capturing the characteristics of the data more comprehensively; the features extracted at each scale will be spliced ​​and fused as the output of the multi-scale convolution fusion module; after being processed by the multi-scale convolution fusion module, the one-dimensional time series signal is converted into a multi-channel time series signal, and then input into the improved Transformer module to further extract deeper features; the improved Transformer module The output of the ransformer module is input into the GRU network. The residual module adds the input of the GRU network to the output after the nonlinear transformation of the GRU network, so that the GRU network can fit complex mapping functions more easily, which is used to capture the global temporal features in the sequence. As the output of the residual-based gated recurrent unit module RGM, after the residual-based gated recurrent unit module, the model can learn higher-level abstract features and better adapt to the complex time relationship in the tobacco drying process; the output of the residual-based gated recurrent unit module RGM passes through two layers of fully connected layers, and the number of output neurons of the second fully connected layer is set according to the prediction step size, which is used to output multi-step prediction results. The final output of the model is the prediction result of the quality index of the tobacco drying process at multiple time points in the future. The prediction result is finally denormalized to obtain the final data that meets the requirements. For example, if the prediction step size is set to 20, the input step size is a multiple of the prediction step size, specifically 40, so that the quality index of the next 20 moments can be predicted based on the process parameters and quality index data of the previous 40 moments.

[0028] The multi-scale convolution fusion module expression is as follows:

[0029]

[0030] MC i F(x)=concat(MC1,MC2,MC3)

[0031] In the formula, Represents a one-dimensional convolution operation; MC1, MC2, and MC3 represent three scale features; x is the input; and ReLU is used as the activation function.

[0032] Furthermore, the improved Transformer module includes an improved Transformer layer, which uses the standard Transformer layer as a framework, and on this framework, a simple linear layer replaces the Decoder structure in the standard Transformer layer. Since the sample data set size of the thin plate drying process in the cigarette silk workshop is an order of magnitude larger than that of the text or image data set, directly using the standard Transformer model may lead to the problem of excessive model complexity, thereby causing cumulative errors and resulting in excessive deviations in the prediction results. To address this problem, the present invention improves the traditional Transformer model, such as Figure 4 The improved Transformer module only uses the Enconder structure of the standard Transformer and replaces the original complex Decoder structure with a linear layer. In this way, the number of model parameters is reduced and the complexity of the model is reduced, thereby effectively controlling the accumulation of errors and improving the accuracy of the prediction results and the training speed of the model. Figure 4 As shown in the figure, the output of the multi-scale convolutional fusion module is used for input embedding to convert the input into a feature vector that can be processed by the model. By adding time step information, the model can capture the sequential relationship in the sequence. The multi-head self-attention mechanism unit in the Transformer layer is used to regard the encoded representation of the input as a set of key-value pairs, where the dimensions of the key and value are the length of the process parameter input time series. The Transformer adopts a scaled dot product attention mechanism, and the output is the weighted sum of the values, where the weight of each value is determined by the dot product of the query and all keys.

[0033] Furthermore, the attention mechanism is expressed as follows:

[0034]

[0035] In the formula, Q (Query) is the query matrix, K (Key) is the key matrix, V (Value) is the value matrix, d k is the dimension of the key vector and softmax is the activation function used to normalize the similarity scores into a probability distribution.

[0036] The GRU network not only inherits the advantages of LSTM, but also simplifies the model structure. By reducing one gating unit, it has faster training speed and less memory usage. Furthermore, the residual module is introduced into GRU, which further improves the stability and performance of the model when processing long sequence data. Figure 5 As shown. In the GRU network, x t and h tare input data and GRU unit output respectively; r t is the reset gate, z t is the update gate, r t and z t Jointly control the calculation and update of h from the previous hidden state t-1 to the new hidden state h t Compared with the three gated units of LSTM: input gate, forget gate, and output gate, GRU merges the input gate and forget gate into an update gate, and uses a reset gate to reduce the complexity of the model, making GRU more efficient in processing time series data. The GRU expression is as follows:

[0037]

[0038] Where: x t is the input vector input to the memory unit at time t; W rx , W rh , W zx , W zh , W hx , W hh is the weight matrix; b r 、b z 、b h is the bias vector; is the candidate hidden state; ⊙ is the dot product operation at the corresponding position of the element; σ and tanh represent the activation function.

[0039] Specifically, the constructed MCF-TransGRU multi-step prediction model is trained, the Optuna framework is used for automatic hyperparameter optimization, and the root mean square error (RMSE) at the 4th second on the validation set is used as the optimization target, aiming to minimize the RMSE value.

[0040] Furthermore, the final parameters are set to a model input step size of 40. In the multi-scale convolution fusion module, three scales of one-dimensional convolution kernels are constructed, namely (3, 5, 7), with a step size of 1, 128 hidden layers, and ReLU activation function; the D_model of the improved Transformer module is set to 128, the number of heads of the multi-head attention mechanism is 4, the Dropout is 0.1, and the activation function is ReLU; the number of GRU neurons in the residual-based gated recurrent unit module is 128, the activation function is tanh, and two fully connected layers are connected, with the number of neurons being (128, 128) and (128, 20) respectively; the model learning rate is 0.001, the Batchsize is 32, and the optimizer is Adam.

[0041] Specifically, the multiple evaluation indicators include root mean square error (RMSE), mean absolute error (MAE) and goodness of fit (R2 ) The parameter sizes of the three evaluation criteria are used as the measure for evaluating the effectiveness of the model.

[0042] Furthermore, the evaluation index calculation formula is as follows:

[0043]

[0044] Where: n is the data volume of the predicted moisture value of the export material; y i The actual value of the moisture content of the exported material; is the predicted value of moisture content of the export material; y is the average value of the actual moisture content of the export material.

[0045] The following is an explanation based on experimental data:

[0046] 1. The collected tow-bread drying data in the production workshop were generated from December 15, 2023 to December 20, 2023, with a total of 87,238 records, as shown in Table 1. The data set includes: inlet material moisture, dehumidification damper opening, dehumidification air volume, hot air temperature, HT steam pressure, rear chamber damper opening, inlet sheet steam pressure, front chamber damper opening, cylinder wall temperature pressure, inlet material flow rate, outlet material moisture, a total of 10 process parameters and 1 quality index.

[0047] Table 1 Part of the original data of the dataset

[0048]

[0049] Second, first remove the head and tail data in the sample data set. The removal rules are shown in Table 2. Then fill the missing values ​​in the sample data with the mean. Finally, use the 3sigma criterion to eliminate the sample data. Normalize the sample data set judged by the 3sigma criterion.

[0050] Table 2 Rules for extracting moisture content of export materials, head and tail data

[0051]

[0052] Taking the selected quality characteristic item of export material moisture ≤ 13.5% as an example, in the thin plate drying process, one day is taken as a batch, and the first data in the current batch of sample data that meets the export material moisture > 10% is taken as the material head node, and the critical point 2.5 minutes after the material head node is delayed is taken as the critical position of the material head; the first data in the current batch of sample data that meets the export material moisture < 10% is taken as the material tail node, and the critical point 2.5 minutes after the material tail node is moved forward is taken as the critical position of the material tail, and the data between the critical positions of the material head and the material tail are retained.

[0053] Furthermore, the normalization process refers to mapping the sample data to between [0, 1], and the normalization formula is:

[0054]

[0055] Where: X′ represents the normalized data; max(X) and min(X) represent the maximum and minimum values ​​of the sample data before normalization in the sample data set; X represents the original sample data.

[0056] 3. Divide the preprocessed data set into a training set (52,344 records), a validation set (17,447 records), and a test set (17,447 records) in a ratio of 6:2:2. Keep the time order when dividing (that is, the time of the training set is earlier than the validation set, and the time of the validation set is earlier than the test set) to simulate the prediction scenario in actual production.

[0057] 4. Construct the MCF-TransGRU multi-step prediction model. The MCF-TransGRU multi-step prediction model includes: multi-scale convolution fusion module, Transformer module and residual-based gated recurrent unit module, such as Figure 2 As shown in the figure, the constructed MCF-TransGRU multi-step prediction model is trained based on the training set and the validation set, and the hyperparameters are adjusted to enable it to deeply learn the relationship and pattern between time series and achieve efficient prediction performance; the trained MCF-TransGRU multi-step prediction model is used to predict the test set data, and the model is verified through a variety of evaluation indicators to achieve accurate multi-step prediction of the quality indicators of the tobacco drying process.

[0058] The prediction effect of the verification model is shown in Table 3. The MCF-TransGRU multi-step prediction model provided by the present invention is compared with five deep learning algorithms, namely, traditional GRU, CNN-LSTM, Seq2Seq, TCN-Attention, and CNN-GRU-Attention. The parameter configuration of the deep learning model is based on the model settings of the present invention. Finally, the root mean square error (RMSE), mean absolute error (MAE), and goodness of fit (R 2 ) as comparative analysis indicators, as shown in Table 3, and the optimal result part of the test set data when the prediction step length is 4, 8, 12, 16 and 20 in the model of the present invention is selected to draw a plot, as shown in Table 3. Figure 6-7 shown.

[0059] Table 3 Prediction results of different models with different step lengths

[0060]

[0061]

[0062] The results show that the multi-step prediction method of tobacco drying process quality based on MCF-TransGRU in the present invention shows the best prediction effect at each prediction time. Whether in short-term prediction or long-term prediction, it has the best prediction effect in RMSE, MAE and R 2 The performance of the three indicators is better than that of other comparison models. This result proves that the MCF-TransGRU multi-step prediction model has strong adaptability and prediction ability in processing complex time series data. In the specific scenario of predicting the moisture content of the material at the outlet of the thin-plate drying process, its high-precision prediction ability can provide strong support for the refined control in the actual production process.

[0063] In order to verify the rationality and effectiveness of the design of the MCF-TransGRU multi-step prediction model provided by the present invention, the moisture content of the material at the outlet of the thin plate drying process was taken as the prediction target, and an ablation experiment was established. The comparison models of the ablation experiment of the present invention are multi-scale CNN-Transformer, Transformer-GRU, CNN-TransGRU, and multi-scale CNN-GRU (in the comparison model, the Transformer involved is the improved Transformer module proposed in the present invention). The hyperparameters of each ablation part are set the same, and the above models are trained under the same experimental conditions. The prediction results of the ablation experiment with a prediction step length of 4 seconds to 20 seconds are shown in Table 4.

[0064] Table 4 Prediction results of ablation experiments

[0065]

[0066]

[0067] As can be seen from Table 4, the multi-scale feature extraction mechanism of the MCF-TransGRU multi-step prediction model is a significant advantage. In the MCF-TransGRU multi-step prediction model, the multi-scale convolution fusion module (MCF) is used to extract features of different scales, which is crucial for capturing the complex dynamic changes in the moisture content of the outlet material of the thin-plate drying process. When the multi-scale convolution layer is replaced by an ordinary convolution layer (such as in CNN-TransGRU, the network is a deformation of the present invention, that is, the MCF in the model of the present invention is replaced by CNN), although the model still has a certain prediction ability, its performance is significantly reduced, especially in longer-term predictions. This shows that the introduction of multi-scale convolution enables the model to capture the multi-scale features in the data more comprehensively, thereby improving the prediction accuracy. Secondly, the application of the Transformer structure in MCF-TransGRU also greatly enhances the model's ability to capture long-term dependencies. In the ablation experiment, the multi-scale convolution fusion was removed, and only the combination of Transformer and GRU (such as in Transformer-GRU) was retained. The performance of the model was significantly worse than that of the complete MCF-TransGRU. This shows that although the Transformer has unique advantages in handling long-term dependencies, when used alone, it cannot fully capture the multi-scale features in the data, especially in short-term and medium-term predictions, and its performance is relatively unstable. The addition of the multi-scale convolutional fusion module makes up for this deficiency, allowing MCF-TransGRU to maintain stable high prediction performance at all time scales. In addition, the introduction of GRU provides MCF-TransGRU with stronger temporal information processing capabilities. Unlike traditional RNNs, GRU effectively alleviates the gradient vanishing problem in long time series through a gating mechanism. In the ablation experiment, after removing GRU (such as in MCF-Transformer), the prediction performance of the model dropped significantly, especially in long-term predictions (such as 16 seconds and 20 seconds), this problem is more significant. This shows that GRU plays a key role in MCF-TransGRU. By combining with Transformer and multi-scale convolutional fusion modules, GRU effectively enhances the model's processing capabilities for long time series. The performance of the two ablation models, MCF-GRU and CNN-TransGRU, is relatively good, but compared with MCF-TransGRU, they still have certain deficiencies in processing complex time series data. MCF-GRU lacks the Transformer module, so it is relatively weak in capturing long-term dependencies in the data, resulting in a decrease in accuracy in long-term predictions. CNN-TransGRU lacks a multi-scale convolution module. Although it has a strong ability to extract local features, its ability to capture global features is not as good as MCF-TransGRU, which is also reflected in the difference in prediction accuracy.In summary, the significant advantage of the MCF-TransGRU multi-step prediction model lies in its comprehensive use of the ability of multi-scale convolution to capture multi-scale features of data, the effective processing of long-term dependencies by Transformer, and the enhancement of GRU's ability to express and remember temporal information. Together, they play to their respective strengths, enabling the entire model to maintain excellent prediction performance at different time scales.

[0068] According to a second aspect of an embodiment of the present invention, a multi-step prediction system for tobacco shred drying process quality based on MCF-TransGRU is provided, comprising: a collection module, used to collect time series data of a thin-plate shred drying process production line in a tobacco shred making workshop at a preset time to form a sample data set; wherein the time series data includes process parameters and quality indicator data; a preprocessing module, used to preprocess the sample data set to obtain a preprocessed data set; a division module, used to divide the preprocessed data set into training set data, validation set data and test set data in chronological order for model training and performance evaluation; a construction module, used to construct an MCF-TransGRU multi-step prediction model through a multi-scale convolution fusion module, an improved Transformer module and a residual-based gated recurrent unit module; a training module, used to train the constructed MCF-TransGRU multi-step prediction model based on the training set data and the validation set data, and adjust hyperparameters; a prediction module, used to predict the test set data / the time series data of the thin-plate shred drying process production line to be tested using the trained MCF-TransGRU multi-step prediction model.

[0069] According to a third aspect of an embodiment of the present invention, there is provided a terminal comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to execute any one of the above-mentioned multi-step prediction methods for tobacco drying process quality based on MCF-TransGRU.

[0070] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, which, when executed by a processor, enables the processor to implement any of the above-mentioned multi-step prediction methods for the quality of tobacco shredded drying process based on MCF-TransGRU.

[0071] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A multi-step prediction method for tobacco cut drying process quality based on MCF-TransGRU, characterized in that: include: Collect the time series data of the thin-plate drying process production line in the cigarette cutting workshop at a preset time to form a sample data set; wherein the time series data includes process parameters and quality index data; Preprocessing the sample data set to obtain a preprocessed data set; Divide the preprocessed data set into training set data, validation set data, and test set data in chronological order for model training and performance evaluation; The MCF-TransGRU multi-step prediction model is constructed through a multi-scale convolutional fusion module, an improved Transformer module, and a residual-based gated recurrent unit module; Based on the training set data and validation set data, the constructed MCF-TransGRU multi-step prediction model is trained and the hyperparameters are adjusted; The trained MCF-TransGRU multi-step prediction model is used to predict the test set data / time series data of the thin plate drying process production line to be tested.

2. The multi-step prediction method for tobacco cut drying process quality based on MCF-TransGRU according to claim 1 is characterized in that: The sample data set is formed by: determining the process parameters and quality indicators according to the specific process of the thin plate drying process in the cigarette silk workshop; collecting time series data under preset time periods to form the sample data set according to the determined process parameters and quality indicators; wherein there are multiple process parameters and one quality indicator.

3. The multi-step prediction method for tobacco cut drying process quality based on MCF-TransGRU according to claim 1 is characterized in that: The preprocessing includes: firstly removing the head and tail of the sample data in the sample data set, then filling the missing values ​​in the data with the mean, and then eliminating the sample data using the 3sigma criterion; and normalizing the sample data set judged by the 3sigma criterion.

4. The multi-step prediction method for tobacco cut drying process quality based on MCF-TransGRU according to claim 1 is characterized in that: The MCF-TransGRU multi-step prediction model is used to take the sample data in the preprocessed data set as input according to different prediction step sizes, and send them to parallel different one-dimensional convolution kernel combinations for multi-scale feature extraction. The features extracted at each scale will be spliced ​​and fused as the output of the multi-scale convolution fusion module; then the improved Transformer module is input to further extract deeper features; the output of the improved Transformer module is input to the GRU network, and the residual module adds the input of the GRU network to the output after the nonlinear transformation of the GRU network, so that the GRU network can fit complex mapping functions more easily, which is used to capture the global temporal features in the sequence as the output of the residual-based gated recurrent unit module RGM; the output of the residual-based gated recurrent unit module passes through two fully connected layers, and the number of output neurons of the second fully connected layer is set according to the prediction step size, which is used to output the multi-step prediction result; The final output of the model is the prediction results of the quality indicators of the tobacco cut drying process at multiple time points in the future. The prediction results are finally denormalized to obtain the final data that meets the requirements.

5. The multi-step prediction method for tobacco cut-drying process quality based on MCF-TransGRU according to claim 4 is characterized in that: The improved Transformer module includes an improved Transformer layer. The improved Transformer layer uses a standard Transformer layer as a framework. On the framework, a simple linear layer is used to replace the Decoder structure in the standard Transformer layer.

6. A multi-step prediction system for tobacco cut drying process quality based on MCF-TransGRU, characterized in that: include: A collection module is used to collect the time series data of the thin plate drying process production line in the cigarette cutting workshop at a preset time to form a sample data set; wherein the time series data includes process parameters and quality index data; A preprocessing module is used to preprocess the sample data set to obtain a preprocessed data set; The partitioning module is used to partition the preprocessed data set into training set data, validation set data, and test set data in chronological order for model training and performance evaluation; A construction module for constructing an MCF-TransGRU multi-step prediction model through a multi-scale convolutional fusion module, an improved Transformer module, and a residual-based gated recurrent unit module; The training module is used to train the constructed MCF-TransGRU multi-step prediction model based on the training set data and the validation set data, and adjust the hyperparameters; The prediction module uses the trained MCF-TransGRU multi-step prediction model to predict the test set data / the time series data of the thin plate drying process production line to be tested.

7. A terminal, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the processor is configured to execute the multi-step prediction method for the quality of the tobacco shredded tobacco drying process based on MCF-TransGRU as described in any one of claims 1 to 5.

8. A computer-readable storage medium, the computer-readable storage medium comprising a stored program, characterized in that: When the program is executed by a processor, the processor implements the multi-step prediction method for the quality of tobacco shredded tobacco drying process based on MCF-TransGRU as described in any one of claims 1 to 5.

Images