Method for predicting, regulating and controlling moisture at outlet of thin-plate cut-tobacco drier and control system

Abstract

The invention provides a thin-plate cut-tobacco dryer outlet moisture prediction and regulation method and a control system. The method comprises the steps that a first data set of the thin-plate cut-tobacco dryer in the operation period is obtained, the first data set is a data set which is preprocessed and is within a first preset duration, and the first data set comprises inlet moisture, outlet moisture, environment humidity, cut-tobacco dryer temperature and cut-tobacco flow of the thin-plate cut-tobacco dryer; the first data set is input into a prediction model, a prediction result is output by the prediction model, and the prediction model comprises an input layer, an LSTM layer, an attention mechanism layer, a full connection layer and an output layer; and based on the prediction result, controlling an execution mechanism in the thin-plate cut-tobacco drier, and adjusting the outlet moisture, so that the outlet moisture is within a preset range corresponding to the generation process. Therefore, the prediction precision of the outlet moisture can be improved, the problem of possible outlet moisture abnormity in the future can be solved in time / in advance, and the control precision of the outlet moisture can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of tobacco shred production control, and in particular to a method and a control system for predicting and regulating moisture at an outlet of a thin-plate shred drying machine. Background Art

[0002] In the production process of tobacco shreds, the moisture content of tobacco shreds is one of the important process indicators, which directly affects the quality of finished tobacco shreds. As one of the equipment in the production process of tobacco shreds, the stable control of the moisture content at the outlet of the thin-plate shred dryer is of great significance to improving the quality of tobacco shreds. However, affected by various factors, the moisture content at the outlet of the shred dryer often fluctuates.

[0003] At present, the prediction of moisture content at the outlet of the tobacco drying machine mainly relies on traditional manual experience. This method is highly subjective and uncertain, and it is difficult to meet the requirements of modern tobacco production for high precision and stability. Although some prediction methods based on machine learning have been applied, there are still problems such as insufficient model accuracy and inability to effectively handle complex data features. For example, the impact of environmental factors on the tobacco drying process is not fully considered, resulting in a large deviation between the prediction results and the actual situation, affecting the precise control of the moisture content at the outlet. Summary of the invention

[0004] In view of this, the purpose of the embodiments of the present application is to provide a method and control system for predicting and regulating the outlet moisture of a thin-plate wire drying machine, which can improve the problem of large errors in outlet moisture prediction and affecting the precise control of outlet moisture.

[0005] In order to achieve the above technical objectives, the technical solutions adopted in this application are as follows:

[0006] In a first aspect, an embodiment of the present application provides a method for predicting and controlling moisture at the outlet of a thin-plate tofu drying machine, characterized in that the method comprises:

[0007] Acquire a first data set during the operation of the thin-plate tobacco drying machine, wherein the first data set is a data set that has been preprocessed and is within a first preset time period, and the first data set includes inlet moisture, outlet moisture, ambient humidity, drying machine temperature, and tobacco flow rate of the thin-plate tobacco drying machine;

[0008] The first data set is input into the prediction model, and the prediction model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer. The input layer receives the first data set and passes the first data set to the LSTM layer. The LSTM layer extracts time series features from the first data set. The attention mechanism layer performs attention weighting on each time step of the time series features, and based on the attention weighting, performs weighted summation on the time series features to obtain a feature vector. The fully connected layer fuses and maps the feature vectors and extracts global features. The output layer outputs a prediction result based on the global feature, and the prediction result includes the outlet moisture change data of the thin plate torrefied machine within a second preset time period in the future.

[0009] Based on the prediction result output by the prediction model, the actuator in the thin-plate tofu drying machine is controlled to adjust the outlet moisture content so that the outlet moisture content is within a preset range corresponding to the production process.

[0010] In combination with the first aspect, in some optional embodiments, before obtaining a first data set of the thin-plate tofu drying machine during operation, the method further includes:

[0011] Obtaining a second data set for model training and testing, the second data set includes inlet moisture, outlet moisture, ambient humidity, temperature of the tobacco drying machine, and flow rate of tobacco shreds, as well as a month label when the data is collected, and the second data set is divided into a training set, a validation set, and a test set;

[0012] Inputting the training set into a constructed neural network model, wherein the neural network model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer, so as to train the neural network model;

[0013] Inputting the validation set into the neural network model, and using a preset loss function to perform hyperparameter tuning of the neural network model to make the neural network model converge;

[0014] The test set is input into the neural network model to test the neural network model, and the tested neural network model is used as the prediction model.

[0015] In combination with the first aspect, in some optional implementations, the preset loss function L is:

[0016]

[0017] Wherein, N refers to the total number of corresponding data in the second data set, and i is an integer from 1 to N. indicates the predicted value of saliva moisture, y iIt refers to the actual value of the mouth water content; γ is the smoothing coefficient, which is used to suppress the prediction mutation.

[0018] In combination with the first aspect, in some optional implementations, the LSTM layer includes multiple LSTM units, each of the LSTM units includes a forget gate, an input gate, and an output gate, and the LSTM layer extracts time series features from the first data set, including:

[0019] Based on the first data set, the corresponding information of the previous moment to be retained is determined by the forget gate in the LSTM layer, and the formula is:

[0020] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0021] f t is the information retained and output by the forget gate at the current time t, σ is the Sigmoid function, W f is the weight matrix of the forget gate, h t-1 is the hidden state at time t-1, x t is the input data at the current time t in the first data set; b f is the bias of the forget gate;

[0022] The input gate in the LSTM layer controls the input of new information at the current moment. The formula is:

[0023] i t =σ(W f ·[h t-1 ,x t ]+b i )

[0024]

[0025] In the formula, i t is the output information of the input gate, is a candidate memory unit, W f and W c are the corresponding weight matrices in the input gate, b i and b c are the corresponding biases in the input gate;

[0026] The output information at the current moment is determined by the output gate in the LSTM layer to extract the time series features from the first data set. The formula is:

[0027] o t =σ(W o·[h t-1 ,x t ]+b o )

[0028] h t =o t tanh(C t )

[0029] In the formula, o t is the output information of the output gate, W o is the weight matrix of the output gate, b o is the bias of the output gate, h t is the hidden state at the current moment, C t For memory unit.

[0030] In combination with the first aspect, in some optional implementations, the attention mechanism layer performs weighted summation of the time series features based on the attention weights for each time step of the time series features to obtain a feature vector, including:

[0031] Based on the time series features output by the LSTM layer, the energy fraction e of the data of the corresponding time step in the time series features is determined by the attention mechanism layer ij , expressed as:

[0032] e ij =score(h i ,h j )

[0033] In the formula, h i and h j They respectively refer to the output information of the LSTM layer at different time steps i and j, and score indicates the operation function using dot product or concatenation;

[0034] The energy score is converted into attention weight α through the Softmax function ij , expressed as:

[0035]

[0036] In the formula, α ij is the attention weight, T is the total number of time steps of the first data set, and k is an integer ranging from 1 to T;

[0037] Based on the attention weight, the time series features are weighted and summed, which is expressed as:

[0038]

[0039] In the formula, S iRepresents the feature vector obtained by weighted summation of the time series features.

[0040] In combination with the first aspect, in some optional implementations, based on the prediction result output by the prediction model, controlling the actuator in the thin-plate tofu drying machine to adjust the outlet moisture includes:

[0041] When the prediction result indicates that there is a period when the outlet moisture is not within the preset range within the second preset time period in the future, the period is regarded as an abnormal period, and before the abnormal period, a preset rolling optimization strategy is used to control the actuator in the thin plate wire drying machine to adjust the performance indicators of the thin plate wire drying machine so that the adjusted outlet moisture is within the preset range, and the performance indicators include the wire drying machine temperature and the tobacco flow rate.

[0042] In combination with the first aspect, in some optional implementations, the method further includes:

[0043] When the abnormal period exists, an early warning is issued, and the abnormal period and a third data set within a third preset time period before the abnormal period are recorded, wherein the third data set includes the outlet moisture.

[0044] In combination with the first aspect, in some optional embodiments, the temperature of the wire drying machine includes the temperature of the hot air entering the drying drum of the thin plate wire drying machine and the drum wall temperature of the drying drum.

[0045] In a second aspect, an embodiment of the present application further provides a control system, which includes a processor and a memory coupled to each other, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the control system executes the above method.

[0046] In combination with the second aspect, in some optional embodiments, the control system also includes a thin plate tofu drying machine controlled by the processor.

[0047] The invention adopting the above technical solution has the following advantages:

[0048] In the technical solution provided by the present application, a prediction model is used to comprehensively analyze various data such as inlet moisture, outlet moisture, ambient humidity, tobacco drying machine temperature, and tobacco flow rate within the first preset time period, and the attention mechanism layer is used to focus on key information that is conducive to the prediction of outlet moisture. In this way, the impact of environmental factors on outlet moisture can be more fully considered, which is conducive to improving the prediction accuracy of outlet moisture. Since the predicted outlet moisture is highly accurate, it is conducive to timely / pre-emptively solving the problem of abnormal outlet moisture that may exist in the future, improving the control accuracy of outlet moisture, and improving the hysteresis of moisture control. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present application may be further described by the non-limiting embodiments given in the accompanying drawings. It should be understood that the following drawings only illustrate certain embodiments of the present application and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings may be obtained based on these drawings without creative effort.

[0050] Figure 1 A schematic flow chart of a method for predicting and controlling moisture at the outlet of a thin-plate tofu drying machine provided in an embodiment of the present application.

[0051] Figure 2 A schematic diagram of the network structure of the prediction model provided in the embodiment of the present application. DETAILED DESCRIPTION

[0052] The present application will be described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that in the drawings or descriptions, similar or identical parts use the same figure numbers, and the implementation methods not shown or described in the drawings are forms known to ordinary technicians in the relevant technical field. In the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0053] An embodiment of the present application provides a control system, which includes a processor and a memory coupled to each other, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the control system executes each step of the following method for predicting and controlling moisture at the outlet of a thin-plate wire drying machine.

[0054] The control system also includes a thin-plate tobacco drying machine controlled by the processor. The structure of the thin-plate tobacco drying machine is conventional technology, and its functions are: first, drying, drying the tobacco with a high moisture content to a moisture content suitable for rolling into cigarettes. Second, improving the filling capacity, by drying the tobacco, the tobacco can be further loosened and deformed and curled, increasing elasticity, and making the leaf and stem shreds expand and shape, thereby improving the filling capacity of the tobacco. Third, baking, the baking process can make some green and odorous substances volatilize and discharge, so that the aroma is revealed and the smoke becomes mellow, so baking can improve the quality of tobacco to a certain extent. Fourth, mixing, through the movement and heating of the tobacco in the drying process, the different formula components and additives in the tobacco are more uniform, so that the components and taste of the same batch of cigarettes can be basically consistent.

[0055] Please refer to Figure 1 and Figure 2 The present application also provides a method for predicting and controlling the moisture content at the outlet of a thin-plate tofu drying machine (hereinafter referred to as the method for simplicity of description), which can be applied to the above-mentioned control system, and each step of the method can be executed or implemented by the control system. Among them, the method may include the following steps:

[0056] Step 110, obtaining a first data set during the operation of the thin-plate tobacco drying machine, wherein the first data set is a data set that has been preprocessed and is within a first preset time period, and the first data set includes inlet moisture, outlet moisture, ambient humidity, drying machine temperature, and tobacco flow rate of the thin-plate tobacco drying machine;

[0057] Step 120, inputting the first data set into the prediction model, the prediction model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer, the input layer receives the first data set, and passes the first data set to the LSTM layer, the LSTM (Long Short-Term Memory) layer extracts time series features from the first data set, the attention mechanism layer pays attention to each time step of the time series features, and based on the attention weights, performs weighted summation on the time series features to obtain a feature vector; the fully connected layer fuses and maps the feature vectors, and extracts global features; the output layer outputs a prediction result based on the global features, and the prediction result includes the outlet moisture change data of the thin plate tofu drying machine within a second preset time period in the future;

[0058] Step 130: Based on the prediction result output by the prediction model, the actuator in the thin-plate tofu drying machine is controlled to adjust the outlet moisture so that the outlet moisture is within a preset range corresponding to the production process.

[0059] The following is a detailed description of each step of this method:

[0060] Before step 110, the method may further include a model training test step. For example, before step 110, the method may further include:

[0061] Obtaining a second data set for model training and testing, the second data set includes inlet moisture, outlet moisture, ambient humidity, temperature of the tobacco drying machine, and flow rate of tobacco shreds, as well as a month label when the data is collected, and the second data set is divided into a training set, a validation set, and a test set;

[0062] Inputting the training set into a constructed neural network model, wherein the neural network model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer, so as to train the neural network model;

[0063] Inputting the validation set into the neural network model, and using a preset loss function to perform hyperparameter tuning of the neural network model to make the neural network model converge;

[0064] The test set is input into the neural network model to test the neural network model, and the tested neural network model is used as the prediction model.

[0065] In this embodiment, the second data set can be understood as data collected in advance by engineers for model training and testing, and can include at least one year's worth of data. As an example, the second data set can include data such as the inlet moisture, outlet moisture, ambient humidity, temperature of the tobacco drying machine, and tobacco flow rate of the thin-plate tobacco drying machine collected in the past 12 months.

[0066] The second data set is a data set obtained after the original data set is preprocessed. The preprocessing can filter out interference data and improve the validity of the data.

[0067] Preprocessing methods may include, but are not limited to, eliminating broken batch data, eliminating outliers based on process standards, and filling in missing values.

[0068] As an example, at the production site of the thin-plate tobacco drying machine, sensors are used to collect data such as the inlet moisture, outlet moisture, ambient humidity, temperature of the tobacco drying machine, and tobacco flow rate of the tobacco drying machine in real time. Data is collected at regular intervals (such as 1 minute), and data is collected continuously for 12 months to form an original data set. Next, the original data set is preprocessed to remove batch data during the material cut-off period, because the data is abnormal during the material cut-off period and has no reference value for prediction. According to the tobacco production process standards, a reasonable range of each data indicator is set, and abnormal values ​​that exceed the range are removed. For missing values, mean filling, linear interpolation, etc. are used to supplement. In addition, the values ​​of various types of data are normalized to facilitate the model to extract data features. After preprocessing, a second data set for model training and prediction is obtained.

[0069] After preprocessing, the second data set can be divided into a training set, a validation set, and a test set according to a preset ratio. The ratio can be flexibly set according to actual conditions, for example, 7:1.5:1.5.

[0070] The neural network model can be pre-built. During the training and testing of the neural network model using the training set, the training set data is input into the neural network model, and the mean square error loss between the outlet moisture result predicted by the neural network model and the actual outlet moisture value is calculated. Through the back propagation algorithm, the weights and biases of each layer of the neural network model are adjusted using optimizers such as Adam (the learning rate is set to 0.001), and the training is continuously iterated until the loss of the neural network model on the validation set no longer decreases or decreases very little, indicating that the model training converges, and a trained neural network model is obtained. Finally, the trained neural network model is tested and optimized using the test set. Model optimization can be understood as adjusting the hyperparameters of the model based on the test results during the test period. The hyperparameters may include, but are not limited to, parameters such as the number of attention heads and the learning rate. In this way, a neural network model with a high accuracy in predicting outlet moisture can be obtained, and the neural network model is used as a prediction model.

[0071] During model training and testing, the loss function uses a combination of MSE (mean square error) and a time smoothing regularization term, that is, the preset loss function L is:

[0072]

[0073] Wherein, N refers to the total number of corresponding data in the second data set, and i is an integer from 1 to N. indicates the predicted value of saliva moisture, y i It refers to the actual value of the outlet moisture; γ is the smoothing coefficient, which is used to suppress the prediction mutation. The smoothing coefficient γ can be flexibly set according to the actual situation, for example, the value is 0.1.

[0074] MSE can effectively measure the average error between the predicted value and the true value, accurately capture the prediction deviation, and make the model output closer to the actual value. In the prediction of moisture at the outlet of the thin-plate tofu drying machine, it can enable the model to learn the key features and rules in the data and improve the accuracy of the prediction. The time smoothing regularization term ensures that the prediction results remain smooth in the time series and avoid abnormal fluctuations. In the scenario of moisture prediction at the outlet of the thin-plate tofu drying machine, the moisture changes in the tobacco are consistent in the time dimension. This regularization term can make the model prediction results conform to this change trend, reduce unreasonable predictions caused by noise or local fluctuations, and enhance the model's ability to capture moisture change trends, thereby improving prediction accuracy.

[0075] The time smoothing regularization term constrains the model parameters, limits their drastic changes during the training process, prevents the model from overfitting the noise and outliers in the training data, and makes the model more focused on the long-term trends and stable features of the learning data, improves the generalization ability of the model, and can stably output reliable prediction results when facing new production data. Through the synergy of MSE and the time smoothing regularization term, the model can better balance the degree of fit to the training data and the reasonable grasp of the data trend during the optimization process, avoid falling into the local optimal solution, and enhance the stability and robustness of the model.

[0076] In step 110, the first data set is obtained by preprocessing the collected raw data. The preprocessing method is similar to the preprocessing method of the second data set and will not be repeated here. Various types of data in the raw data can be collected by devices such as temperature sensors, humidity sensors, and weight sensors. The first preset time can be 5 minutes, 10 minutes, etc. before the current moment, so that the data in the first data set can be updated in real time, and avoid the collection time being too long, resulting in a large amount of data and a large amount of prediction calculations.

[0077] In this embodiment, the "moisture content" in "outlet moisture content" and "inlet moisture content" refers to the moisture content of the tobacco. For example, the outlet moisture content refers to the moisture content of the tobacco at the outlet of the thin-plate drying machine; the inlet moisture content refers to the moisture content of the tobacco at the inlet of the thin-plate drying machine. The current moisture content of the tobacco is collected in a conventional manner and will not be described here.

[0078] In step 120, the LSTM layer includes a plurality of LSTM units, each of which includes a forget gate, an input gate, and an output gate. The LSTM layer extracts time series features from the first data set, including:

[0079] Step 1211, based on the first data set, the corresponding information of the previous moment to be retained is determined by the forget gate in the LSTM layer, and the formula is:

[0080] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0081] f t is the information retained and output by the forget gate at the current time t, σ is the Sigmoid function, W f is the weight matrix of the forget gate, h t-1 is the hidden state at time t-1, x t is the input data at the current time t in the first data set; b f is the bias of the forget gate; hidden state ht-1 It carries the key information of the previous moment of the time series, which includes the characteristics and patterns of past data. When predicting the outlet moisture of the thin-plate drying machine, the hidden state may be the comprehensive feature representation of multiple factors such as the flow rate, temperature, and inlet moisture of the tobacco at the previous moment after being processed by the LSTM layer. This information is crucial for the forget gate decision at the current moment, because the forget gate uses it to decide how much information from the previous moment to retain.

[0082] Step 1212, controlling the input of new information at the current moment through the input gate in the LSTM layer, the formula is:

[0083] i t =σ(W f ·[h t-1 ,x t ]+b i )

[0084]

[0085] In the formula, i t is the output information of the input gate, is a candidate memory unit, W f and W c are the corresponding weight matrices in the input gate, b i and b c are the corresponding biases in the input gate;

[0086] Step 1213, determine the output information at the current moment through the output gate in the LSTM layer to extract the time series features from the first data set, and the formula is:

[0087] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0088] h t =o t tanh(C t )

[0089] In the formula, o t is the output information of the output gate, W o is the weight matrix of the output gate, b o is the bias of the output gate, h t is the hidden state at the current moment, C t For memory unit.

[0090] In step 1211, if the data input to the forget gate is the first data information, the first data information is retained. For the data information input at the subsequent moment, the forget gate determines how much information of the previous moment to retain.

[0091] In the LSTM unit, the input data is selectively memorized and updated through the collaborative work of the forget gate, input gate, and output gate. Through the layer-by-layer progression of multiple LSTM units, the long-term dependency features in the time series data are effectively extracted to capture the trend characteristics of the change in moisture content of the tobacco.

[0092] In step 120, the attention mechanism layer performs weighted summation of the time series features based on the attention weights for each time step of the time series features to obtain a feature vector, including:

[0093] Step 1221, based on the time series features output by the LSTM layer, determine the energy fraction e of the data of the corresponding time step in the time series features through the attention mechanism layer ij , expressed as:

[0094] e ij =score(h i ,h j )

[0095] In the formula, h i and h j They respectively refer to the output information of the LSTM layer at different time steps i and j, and score indicates the operation function using dot product or concatenation;

[0096] Step 1222, convert the energy score into attention weight α through Softmax function ij , expressed as:

[0097]

[0098] In the formula, α ij is the attention weight, T is the total number of time steps of the first data set, and k is an integer ranging from 1 to T;

[0099] Step 1223, based on the attention weight, weighted sum is performed on the time series features, expressed as:

[0100]

[0101] In the formula, S i Represents the feature vector obtained by weighted summation of the time series features.

[0102] Understandably, the attention mechanism layer can adaptively assign weights to highlight features that are more important to the prediction results. The attention mechanism layer receives the output of the LSTM layer, calculates the attention weight of each time step, and performs weighted summation of the LSTM layer output according to the attention weight to obtain a weighted feature representation, thereby focusing on the key information for outlet moisture prediction and improving prediction accuracy.

[0103] The fully connected layer further fuses and maps the feature vectors output by the attention mechanism layer. Through a series of weight matrices and biases, the feature vectors are mapped to extract more abstract and representative global features. Assume that the input of the fully connected layer is x qt , the subscript q refers to the fully connected layer, t refers to the time step, and the weight matrix is ​​W q , with a bias of b q , then the output y of the fully connected layer qt =W q x qt +b q The fully connected layer can contain multiple hidden layers, and the feature representation is optimized by continuously adjusting the weights and biases.

[0104] The output layer can obtain the final prediction result through linear transformation and activation function based on the output of the fully connected layer. For the outlet moisture prediction, a linear activation function is used to directly output the predicted outlet moisture value of the thin-plate tofu drying machine in the future.

[0105] In step 130, based on the prediction result output by the prediction model, the actuator in the thin-plate tofu drying machine is controlled to adjust the outlet moisture, including:

[0106] When the prediction result indicates that there is a period in the future second preset time period when the outlet moisture is not within the preset range, the period is regarded as an abnormal period, and before the abnormal period, the actuator in the thin-plate tobacco drying machine is controlled to adjust the performance index of the thin-plate tobacco drying machine through the preset rolling optimization strategy, so that the adjusted outlet moisture is within the preset range, and the performance index includes the tobacco drying machine temperature and tobacco flow rate. The preset range can be flexibly determined according to actual conditions.

[0107] The actuator may be, but is not limited to, a hot air blower, a mechanism for adjusting the flow of tobacco at the inlet of a thin-plate drying machine, or a mechanism for adjusting the temperature of the drying machine (such as an electric valve). The formula for the preset rolling optimization strategy may be as follows:

[0108]

[0109] In the formula, q i and r jare the weighted coefficients of error and control action at different moments, i = 1, 2, ..., P; j = 1, 2, ..., M, P and M can be determined according to actual conditions; Y P (k+i) is the output value predicted by the prediction model at time k+i (predicted outlet moisture), Y r (k+i) is the reference value at time k+i (the set value corresponding to the production process); U is the input variable value, and J refers to the control quantity.

[0110] Understandably, by adopting the preset rolling optimization strategy, the output value and reference value of the prediction model can be used to calculate the control amount after a period of time in the future, and the future control amount can be determined by optimizing a certain performance indicator. The performance indicator can be the drying machine temperature and tobacco flow rate of the tobacco drying machine. In this way, it is helpful to solve the abnormal outlet moisture problem that may exist in the future in a timely / premature manner.

[0111] In this embodiment, the temperature of the wire drying machine may include: the temperature of the hot air entering the drying drum of the thin-plate wire drying machine, and the temperature of the wall of the drying drum. The wall temperature can be adjusted by the temperature and flow rate of the hot steam blown into the cavity of the wall.

[0112] In this embodiment, the first data set acquired in real time is input into the trained prediction model, and the model estimates the change in the outlet moisture of the thin-plate tofu drying machine in the future (such as the next 10 minutes) through feature extraction and calculation of each network layer. Production personnel can adjust the operating parameters of the thin-plate tofu drying machine in advance according to the prediction results, such as the temperature of the tofu drying machine, the amount of steam, and the flow rate of tobacco. For example, if the predicted outlet moisture value is too high, the temperature of the tofu drying machine or the amount of steam can be appropriately increased, otherwise the temperature can be reduced or the amount of steam can be reduced, thereby achieving precise control of the outlet moisture and ensuring the stability of the tobacco quality.

[0113] The present invention fully considers various key factors in the operation of the thin-plate tobacco drying machine, including inlet moisture, ambient humidity, etc., and effectively extracts and analyzes these complex data through a deep learning model, which is more conducive to improving the accuracy of outlet moisture prediction compared to traditional prediction methods. In addition, the LSTM layer is used to process time series data, which can capture the long-term dependence characteristics of tobacco moisture changes. Combined with the attention mechanism layer, the key features are further highlighted, making the model more accurate in grasping the trend of outlet moisture changes.

[0114] After step 120 or after step 130, the method may further include:

[0115] When the abnormal period exists, an early warning is issued, and the abnormal period and a third data set within a third preset time period before the abnormal period are recorded, wherein the third data set includes the outlet moisture.

[0116] Understandably, for the abnormal period, various data corresponding to the predicted abnormal period can be recorded to facilitate engineers to analyze the abnormal cause later. Among them, the third preset time length can be flexibly determined according to the actual situation, and the third data set can include, but is not limited to, ambient humidity, outlet moisture, etc.

[0117] In the control system, the processor may be a general-purpose processor. For example, the processor may be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and may implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application.

[0118] The memory may be, but is not limited to, a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc. In this embodiment, the memory may be used to store the prediction model, the first data set, the second data set, the prediction result, etc. Of course, the storage module may also be used to store a program, and the processing module executes the program after receiving the execution instruction.

[0119] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working process of the control system described above can refer to the corresponding process of each step in the aforementioned method, and will not be elaborated here.

[0120] Through the description of the above implementation methods, technical personnel in this field can clearly understand that the present application can be implemented by hardware, and can also be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a control system, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0121] In the embodiments provided in the present application, it should be understood that the disclosed system and method can also be implemented in other ways. The system and method embodiments described above are merely schematic, for example, the flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a part of a module, a program segment or a code, and a part of the module, program segment or code includes one or more executable instructions for implementing the specified logical function. It should also be noted that each box in the block diagram and / or the flow chart, and the combination of the boxes in the block diagram and / or the flow chart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. In addition, each functional module in each embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0122] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting and controlling moisture at the outlet of a thin-plate tofu drying machine, characterized in that: The method comprises: Acquire a first data set during the operation of the thin-plate tobacco drying machine, wherein the first data set is a data set that has been preprocessed and is within a first preset time period, and the first data set includes inlet moisture, outlet moisture, ambient humidity, drying machine temperature, and tobacco flow rate of the thin-plate tobacco drying machine; The first data set is input into a prediction model, and the prediction model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer. The input layer receives the first data set and passes the first data set to the LSTM layer. The LSTM layer extracts time series features from the first data set. The attention mechanism layer performs attention weighting on each time step of the time series features, and based on the attention weighting, performs weighted summation on the time series features to obtain a feature vector. The fully connected layer fuses and maps the feature vectors and extracts global features. The output layer outputs a prediction result based on the global features, and the prediction result includes the outlet moisture change data of the thin plate torrefied machine within a second preset time period in the future. Based on the prediction result output by the prediction model, the actuator in the thin-plate tofu drying machine is controlled to adjust the outlet moisture content so that the outlet moisture content is within a preset range corresponding to the production process.

2. The method according to claim 1, characterized in that Before obtaining a first data set during operation of the thin-plate tofu drying machine, the method further includes: Obtaining a second data set for model training and testing, the second data set includes inlet moisture, outlet moisture, ambient humidity, temperature of the tobacco drying machine, and flow rate of tobacco shreds, as well as a month label when the data is collected, and the second data set is divided into a training set, a validation set, and a test set; Inputting the training set into a constructed neural network model, wherein the neural network model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer, so as to train the neural network model; Inputting the validation set into the neural network model, and using a preset loss function to perform hyperparameter tuning of the neural network model to make the neural network model converge; The test set is input into the neural network model to test the neural network model, and the tested neural network model is used as the prediction model.

3. The method according to claim 2, characterized in that The preset loss function L is: Wherein, N refers to the total number of corresponding data in the second data set, i is an integer from 1 to N, indicates the predicted value of saliva moisture, y i It refers to the actual value of the mouth water content; γ is the smoothing coefficient, which is used to suppress the prediction mutation.

4. The method according to claim 1, characterized in that The LSTM layer includes a plurality of LSTM units, each of which includes a forget gate, an input gate, and an output gate. The LSTM layer extracts time series features from the first data set, including: Based on the first data set, the corresponding information of the previous moment to be retained is determined by the forget gate in the LSTM layer, and the formula is: f t =σ(W f ·[h t-1 ,x t ]+b f ) f t is the information retained and output by the forget gate at the current time t, σ is the Sigmoid function, W f is the weight matrix of the forget gate, h t-1 is the hidden state at time t-1, x t is the input data at the current time t in the first data set; b f is the bias of the forget gate; The input gate in the LSTM layer controls the input of new information at the current moment. The formula is: i t =σ(W f ·[h t-1 ,x t ]+b i ) In the formula, i t is the output information of the input gate, is a candidate memory unit, W f and W c are the corresponding weight matrices in the input gate, b i and b c are the corresponding biases in the input gate; The output information at the current moment is determined by the output gate in the LSTM layer to extract the time series features from the first data set. The formula is: the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t ·tanh(C t ) In the formula, o t is the output information of the output gate, W o is the weight matrix of the output gate, b o is the bias of the output gate, h t is the hidden state at the current moment, C t For memory unit.

5. The method according to claim 1, characterized in that The attention mechanism layer pays attention to each time step of the time series feature, and based on the attention weight, performs weighted summation on the time series feature to obtain a feature vector, including: Based on the time series features output by the LSTM layer, the energy fraction e of the data of the corresponding time step in the time series features is determined by the attention mechanism layer ij , expressed as: e ij =score(h i ,h j ) In the formula, h i and h j They respectively refer to the output information of the LSTM layer at different time steps i and j, and score indicates the operation function using dot product or concatenation; The energy score is converted into attention weight α through the Softmax function ij , expressed as: In the formula, α ij is the attention weight, T is the total number of time steps of the first data set, and k is an integer ranging from 1 to T; Based on the attention weight, the time series features are weighted and summed, which is expressed as: In the formula, S i Represents the feature vector obtained by weighted summation of the time series features.

6. The method according to claim 1, characterized in that Based on the prediction result output by the prediction model, controlling the actuator in the thin-plate tofu drying machine to adjust the outlet moisture includes: When the prediction result indicates that there is a period when the outlet moisture is not within the preset range within the second preset time period in the future, the period is regarded as an abnormal period, and before the abnormal period, a preset rolling optimization strategy is used to control the actuator in the thin plate wire drying machine to adjust the performance indicators of the thin plate wire drying machine so that the adjusted outlet moisture is within the preset range, and the performance indicators include the wire drying machine temperature and the tobacco flow rate.

7. The method according to claim 6, characterized in that The method further comprises: When the abnormal period exists, an early warning is issued, and the abnormal period and a third data set within a third preset time period before the abnormal period are recorded, wherein the third data set includes the outlet moisture.

8. The method according to claim 6, characterized in that The wire drying machine temperature includes the temperature of the hot air entering into the drying drum of the thin plate wire drying machine and the drum wall temperature of the drying drum.

9. A control system, characterized in that: The control system comprises a processor and a memory coupled to each other, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the control system executes the method according to any one of claims 1 to 8.

10. The control system according to claim 9, characterized in that: The control system also includes a thin-plate tofu drying machine controlled by the processor.

Images