Tobacco shred-making leaf-moistening charging moisture early warning method and system, and storage medium

Abstract

The invention discloses a tobacco shred making, leaf moistening and charging moisture early warning method and system, and a storage medium. The method comprises the steps of establishing a tobacco-moistening and feeding moisture LSTM neural network model, collecting data related to tobacco-shred-making tobacco-moistening and feeding moisture early warning from a tobacco-moistening and feeding PLC through an OPC UA server, storing the data in a time sequence database table, extracting a training data set through a time sequence database, obtaining a tobacco-moistening and feeding moisture prediction model, and predicting tobacco-shred-making tobacco-moistening and feeding moisture early warning. The method comprises the following steps: extracting a latest collected information point value list from a time sequence database, calculating a leaf moistening and feeding moisture prediction value, calculating a leaf moistening and feeding moisture prediction deviation and a prediction deviation proportion, calculating a leaf moistening and feeding moisture early warning signal value, and issuing the early warning signal value to a leaf moistening and feeding PLC controller through an OPC UA server for early warning / alarm control. According to the method, the possibility of quality defects in the shred-making, leaf-moistening and charging production process can be reduced to the maximum extent, and a positive effect is brought to quality control of the shred-making, leaf-moistening and charging process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tobacco product moisture content control, and more particularly to a tobacco leaf conditioning and feeding moisture pre-warning method, a storage medium and a system. BACKGROUND

[0002] The leaf conditioning and feeding process is an important link in the tobacco leaf processing process, and the key processing equipment is a leaf feeding machine. The leaf feeding machine controls the set water feeding flow by comprehensively considering factors such as the inlet material flow, the inlet moisture, the applied material flow, and the temperature-increasing and humidity-increasing steam flow, so as to realize the control of the leaf conditioning and feeding moisture, and the core control method is PID.

[0003] PID control is a control strategy based on feedback, and its basic principle is to adjust the output signal according to the difference between the current state of the controlled object and the set value, so that the difference tends to zero. The PID controller calculates the output signal according to the weight combination of the proportional (P), integral (I), and differential (D) parts, so as to realize accurate control of the controlled object. This control method mainly relies on the deviation of the current and past output measurement values and the set value to determine the current control input, and does not have the ability to predict the future leaf conditioning and feeding moisture.

[0004] However, the leaf conditioning and feeding moisture is a key quality assessment index in the tobacco leaf processing process, and has a direct impact on the tobacco leaf quality. In the production process, when the tobacco leaf enters the feeding machine, if the outlet moisture value can be predicted, when the predicted value trend is abnormal, a pre-warning can be sent in time to allow the operator to intervene in advance, so that the possibility of quality defects in the production process can be minimized, thereby bringing a positive effect to the quality control of the tobacco leaf processing process. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a tobacco leaf conditioning and feeding moisture pre-warning method, a storage medium and a system.

[0006] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0007] The present application provides a tobacco leaf conditioning and feeding moisture pre-warning method, a storage medium and a system.

[0008] The method comprises the following steps:

[0009] (One), establish the leaf moisturizing and feeding water content LSTM neural network model Model_LSTM:

[0010] Model_LSTM includes:

[0011] (1), data standardization scaling layer MinMaxScaler:

[0012] MinMaxScaler receives two matrix data parameters with shapes of (n, 30) and (, 1) as input data, trains the relevant parameters of MinMaxScaler itself, and outputs the results to the following long short-term memory network layer LSTM1 after processing the input data.

[0013] Wherein, n is the number of training data set samples.

[0014] (2), long short-term memory network layer LSTM1:

[0015] LSTM1 receives the calculation output of the above data standardization scaling layer MinMaxScaler, trains the relevant weight parameters of LSTM1 itself, and outputs the results to the following long short-term memory network layer LSTM2.

[0016] The initialization parameters of LSTM1 include:

[0017] Whether to return the entire sequence return_sequences: true

[0018] The dimension of the output space units: 56

[0019] (3), long short-term memory network layer LSTM2:

[0020] LSTM2 receives the calculation output of the above data long short-term memory network layer LSTM1, trains the relevant weight parameters of LSTM2 itself, and outputs the results to the following threshold cycle unit network layer GRU.

[0021] The initialization parameters of LSTM2 include:

[0022] Whether to return the entire sequence return_sequences: true

[0023] The dimension of the output space units: 56

[0024] (4), threshold cycle unit network layer GRU:

[0025] The GRU receives the calculation output of the LSTM2 layer, trains its own weight parameters, and outputs the results to the Batch Normalization layer described below.

[0026] The initialization parameters of GRU include:

[0027] Whether to return the entire sequence return_sequences: value true

[0028] The dimension of the output space units: value 56

[0029] (5) Batch Normalization:

[0030] BatchNormalization receives the calculated output of the gated recurrent unit network layer GRU mentioned above, trains the relevant weight parameters of BatchNormalization itself, and outputs the results to the regularization dropout layer described below.

[0031] (6) Regularization Dropout layer:

[0032] Dropout receives the calculated output of the batch normalization layer BatchNormalization, trains the relevant weight parameters of Dropout itself, and outputs the results to the fully connected neural network layer Dense described below.

[0033] The initialization parameters of Dropout include:

[0034] Discard rate: set to 0.2

[0035] (7) Fully connected neural network layer Dense:

[0036] Dense receives the calculated output of the above regularized dropout layer Dropout and trains the relevant weight parameters of Dense itself.

[0037] The initialization parameters of Dense include:

[0038] The dimension of the output space, units: takes the value 1.

[0039] In summary, by inputting two matrix data parameters of shape (n, 30) and (, 1) into Model_LSTM, and calculating through the neural network layers (1) to (7) of the above Model_LSTM, a model for predicting the moisture content of leaf moistening feed can be finally obtained.

[0040] (2) During the production process, the OPC UA server collects the values ​​of the following information points from the Runye feeding PLC controller in a 1-second cycle and stores them in the time series database table:

[0041] X1 inlet electronic scale flow

[0042] X2 instantaneous feeding ratio

[0043] X3 SP02 liquid atomization pressure actual

[0044] X4 SP04 barrel heating steam source pressure

[0045] X5 SP01 valve island compressed air source 1 pressure

[0046] X6 SP03A pump liquid pressure

[0047] X7 inlet hot air temperature PV value

[0048] X8 return air temperature PV value

[0049] X9 heated wet steam flow PV value

[0050] X10 main water flow control PV value

[0051] X11 water atomization steam pressure PV value

[0052] X12 humidifying water temperature PV value

[0053] X13 radiator steam pressure PV value

[0054] X14 cylinder wall heating steam pressure PV value

[0055] X15 discharge chamber wall steam pressure PV value

[0056] X16 preheating cycle hot air temperature PV value

[0057] X17 actual steam flow

[0058] X18 damper opening

[0059] X19 main steam flow

[0060] X20 drum heating wet steam flow

[0061] X21 front section wall temperature

[0062] X22 Middle section wall temperature

[0063] X23 rear section wall temperature

[0064] X24 Radiator steam temperature detection value X25 Cylinder wall heating steam temperature detection value

[0065] X26 Discharge chamber wall heating steam temperature detected value

[0066] X27 Radiator steam pressure

[0067] X28 Feeder outlet temperature

[0068] X29 Hot air temperature detected value

[0069] X30 Return air temperature detected value

[0070] Y Outlet material moisture content.

[0071] (Three), through the time series database, extract the training data set X_Test, Y_Test in the production process in the last at least one month:

[0072] X_Test = [[X1, X2, …, X30]0, [X1, X2, …, X30]1, …, [X1, X2, …, X30] n ],

[0073] Y_Test = [Y0, Y1, …, Y n ],n is the number of training data set samples-1.

[0074] (Four), input X_Test, Y_Test to Model_LSTM, after the training calculation of each neural network layer in the above step (one), finally get a leaf moistening and feeding moisture prediction model Model_LSTM_P:

[0075] Model_LSTM_P = Model_LSTM (X_Test, Y_Test).

[0076] (Five), extract the latest 120 rows of information point value list X_Predict from the time series database:

[0077] X_Predict = [[X1, X2, …, X30]0, [X1, X2, …, X30]1, …, [X1, X2, …, X30] 119 ]。

[0078] (Six), input X_Predict to Model_LSTM_P, calculate the leaf moistening and feeding moisture prediction value Y_Predict, and send Y_Predict to the time series database table for storage:

[0079] Y_Predict = Model_LSTM_P (X_Predict).

[0080] (VII) Calculate the predicted deviation Y_D and the predicted deviation ratio Y_DP of the moisture content of the moistening leaf feed:

[0081] Y_D=Y_Predict-Y_C, where Y_C is the center value of the moisture specification of the moistening leaf feed.

[0082] When Y_D=0:

[0083] Y_DP=0,

[0084] When Y_D>0:

[0085] Y_DP=100*Y_D / (Y_USL-Y_C) where Y_USL is the upper limit of the moisture content of the moistening material.

[0086] When Y_D<0:

[0087] Y_DP=100*Y_D / (Y_C-Y_LSL) where Y_LSL is the lower limit of the moisture specification for the leaf moistening feed.

[0088] (8) Calculate the leaf moistening and feeding moisture warning signal value Y_Alarm:

[0089] When -75≤Y_DP≤75, Y_Alarm takes the value 0;

[0090] When -100≤Y_DP<-75 or 75<Y_DP≤100, Y_Alarm takes the value 1;

[0091] When Y_DP>100 or Y_DP<-100, Y_Alarm takes the value 2.

[0092] (IX) Y_Alarm is sent to the Runye feeding PLC controller through the OPC UA server. The PLC controller performs early warning / alarm control based on the Y_Alarm value. The control rules are as follows:

[0093] (a) When Y_Alarm=2, the cabinet siren sounds, the warning light flashes, and a red alarm message box pops up on the monitoring screen;

[0094] (b) When Y_Alarm=1, the alarm light of the electric cabinet flashes and a yellow warning message box pops up on the monitoring screen;

[0095] (c) When Y_Alarm=0, the cabinet siren sound is turned off, the warning light stops flashing, and the warning / alarm message box on the monitoring screen disappears.

[0096] During the batch production process of the moistening leaf feeding process, the above steps (5) to (9) are repeated with a cycle of 1 second.

[0097] When there are complex uncertainties such as time variation and changes in the process of equipment related to the leaf moistening and feeding in silk making, which cause the prediction accuracy of the leaf moistening and feeding moisture prediction model Model_LSTM_P to be lower than 90%, thereby causing an erroneous early warning / alarm signal, the above steps (three) to (four) can be repeated to perform rolling optimization processing on the leaf moistening and feeding moisture prediction model Model_LSTM_P.

[0098] Model_LSTM_P prediction accuracy = the ratio of the number of points where the difference between the predicted value Y_Predict and the true value Y for the leaf moistening moisture in the past month is within the range of ±0.2 to the total number of predicted value samples.

[0099] A computer-readable storage medium stores a computer program, which can be executed by a processor to implement the steps of the moisture early warning method for silk-making leaf moistening and adding materials described in the present invention.

[0100] A moisture early warning system for silk-making leaf moistening and feeding, comprising a time series database, an OPC UA server and a loose moisture regain PLC controller; the time series database and the OPC UA server are installed and configured in a silk-making centralized control system environment, the loose moisture regain PLC controller is a device provided by a loose moisture regain machine, the time series database, the OPC UA server and the loose moisture regain PLC controller are all interconnected via a workshop Ethernet; and the system also comprises a computer-readable storage medium according to the present invention, which is arranged in the loose moisture regain PLC controller.

[0101] Beneficial effects of the present invention:

[0102] The present invention provides a moisture early warning method for silk-making leaf moistening feed based on an LSTM neural network prediction model. During the production process, the moisture deviation value of the leaf moistening feed can be predicted in real time. When the predicted deviation value exceeds 75% of the process specification range, an early warning message is issued in advance. When the predicted deviation value exceeds 100% of the process specification range, an alarm message is issued in advance to prompt the operator, so that the operator can make timely corrections in advance when the moisture of the leaf moistening feed may be abnormal, thereby minimizing the possibility of quality defects in the production process of the leaf moistening feed, and having a positive effect on the quality control of the silk-making leaf moistening feed process. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0104] Figure 1 A schematic diagram of the system structure provided by an embodiment of the present invention.

[0105] Figure 2 Schematic diagram of the Model_LSTM structure of the leaf moistening and feeding moisture LSTM neural network model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0106] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0107] like Figure 1 As shown, the system structure of a moisture early warning method for silk-making leaf moistening and feeding based on an LSTM neural network prediction model includes: a time series database 11, an OPC UA server 12, and a leaf moistening and feeding PLC controller 13.

[0108] In this embodiment, the time series database 11, the OPC UA server 12, and the Runye feeding PLC controller 13 are interconnected through the workshop Ethernet.

[0109] The specific implementation method is:

[0110] (1) Establish the LSTM neural network model Model_LSTM:

[0111] Model_LSTM structure is as follows Figure 2 Shown, including:

[0112] (1) Data normalization scaling layer MinMaxScaler 21:

[0113] MinMaxScaler receives two matrix data parameters of shape (n,30) and (,1) as input data, trains the relevant parameters of MinMaxScaler itself, processes the input data, and outputs the results to the following long short-term memory network layer LSTM1 22;

[0114] Where n is the number of samples in the training data set;

[0115] (2) Long short-term memory network layer LSTM1 22:

[0116] LSTM1 receives the calculation output of the above-mentioned data normalization and scaling layer MinMaxScaler 21, trains the relevant weight parameters of LSTM1 itself, and outputs the result to the following long short-term memory network layer LSTM2 23;

[0117] The initialization parameters of LSTM1 include:

[0118] Whether to return the entire sequence return_sequences: the value is true,

[0119] The dimension of the output space units: the value is 56;

[0120] (3) Long short-term memory network layer LSTM2 23:

[0121] LSTM2 receives the calculation output of the long short-term memory network layer LSTM1 22 of the above data, trains the relevant weight parameters of LSTM2 itself, and outputs the result to the following gated recurrent unit network layer GRU 24;

[0122] The initialization parameters of LSTM2 include:

[0123] Whether to return the entire sequence return_sequences: the value is true,

[0124] The dimension of the output space units: the value is 56;

[0125] (4) Gated recurrent unit network layer GRU 24:

[0126] The GRU receives the calculation output of the long short-term memory network layer LSTM2 23, trains the relevant weight parameters of the GRU itself, and outputs the results to the following batch normalization layer BatchNormalization 25;

[0127] The initialization parameters of GRU include:

[0128] Whether to return the entire sequence return_sequences: the value is true,

[0129] The dimension of the output space units: the value is 56;

[0130] (5) Batch Normalization 25:

[0131] BatchNormalization receives the calculation output of the gated recurrent unit network layer GRU 24, trains the relevant weight parameters of BatchNormalization itself, and outputs the result to the regularization dropout layer Dropout 26 described below;

[0132] (6) Regularization dropout layer Dropout 26:

[0133] Dropout receives the calculation output of the batch normalization layer BatchNormalization 25, trains the relevant weight parameters of Dropout itself, and outputs the result to the fully connected neural network layer Dense 27 described below;

[0134] The initialization parameters of Dropout include:

[0135] Discard rate: set to 0.2.

[0136] (7) Fully connected neural network layer Dense 27:

[0137] Dense receives the calculated output of the above regularized dropout layer Dropout 26 and trains its own related weight parameters;

[0138] The initialization parameters of Dense include:

[0139] The dimension of the output space units: the value is 1,

[0140] In summary, by inputting two matrix data parameters of shape (n, 30) and (, 1) into Model_LSTM and calculating through the neural network layers (1) to (7) of the above Model_LSTM, a model for predicting the moisture content of leaf moistening feed can be finally obtained.

[0141] (2) During the production process, the OPC UA server collects the values ​​of the following information points from the Runye feeding PLC controller in a 1-second cycle and stores them in the time series database table:

[0142] X1 inlet electronic scale flow

[0143] X2 instantaneous feeding ratio

[0144] X3 SP02 liquid atomization pressure actual

[0145] X4 SP04 barrel heating steam source pressure

[0146] X5 SP01 valve island compressed air source 1 pressure

[0147] X6 SP03A pump liquid pressure

[0148] X7 inlet hot air temperature PV value

[0149] X8 return air temperature PV value

[0150] X9 heated wet steam flow PV value

[0151] X10 main water flow control PV value

[0152] X11 water atomization steam pressure PV value

[0153] X12 humidifying water temperature PV value

[0154] X13 radiator steam pressure PV value

[0155] X14 cylinder wall heating steam pressure PV value

[0156] X15 discharge chamber wall steam pressure PV value

[0157] X16 preheating cycle hot air temperature PV value

[0158] X17 actual steam flow

[0159] X18 damper opening

[0160] X19 main steam flow

[0161] X20 drum heating wet steam flow

[0162] X21 front section wall temperature

[0163] X22 Middle section wall temperature

[0164] X23 rear section wall temperature

[0165] X24 Radiator steam temperature detection value

[0166] X25 cylinder wall heating steam temperature detection value

[0167] X26 discharge chamber wall heating steam temperature detection value

[0168] X27 Radiator Steam Pressure

[0169] X28 feeder outlet temperature

[0170] X29 hot air temperature detection value

[0171] X30 Return air temperature detection value

[0172] Y outlet material moisture content.

[0173] (3) Extract the training data sets X_Test and Y_Test from the production process of at least the last month through the time series database 11:

[0174] X_Test=[[X1,X2,…,X30]0,[X1,X2,…,X30]1,…,[X1,X2,…,X30] n ],

[0175] Y_Test=[Y0,Y1,…,Y n]n is the number of samples in the training data set - 1.

[0176] (IV) Input X_Test and Y_Test into Model_LSTM. After the training calculation of each neural network layer in step (I) above, a leaf moistening feed moisture prediction model Model_LSTM_P is finally obtained:

[0177] Model_LSTM_P=Model_LSTM(X_Test,Y_Test).

[0178] (5) Extract the latest 120 rows of information point value list X_Predict from the time series database 11:

[0179] X_Predict=[[X1,X2,…,X30]0,[X1,X2,…,X30]1,…,[X1,X2,…,X30] 119 ].

[0180] (6) Input X_Predict into Model_LSTM_P, calculate the leaf moistening feed moisture prediction value Y_Predict, and send Y_Predict to the time series database table for storage:

[0181] Y_Predict=Model_LSTM_P(X_Predict).

[0182] (VII) Calculate the predicted deviation Y_D and the predicted deviation ratio Y_DP of the moisture content of the moistening leaf feed:

[0183] Y_D = Y_Predict - Y_C, where Y_C is the center value of the moisture specification of the moistening leaf feed;

[0184] When Y_D=0:

[0185] Y_DP=0,

[0186] When Y_D>0:

[0187] Y_DP=100*Y_D / (Y_USL-Y_C) where Y_USL is the upper limit of the moisture content of the leaf moistening feed;

[0188] When Y_D<0,

[0189] Y_DP=100*Y_D / (Y_C-Y_LSL) where Y_LSL is the lower limit of the moisture specification for the leaf moistening feed.

[0190] (8) Calculate the leaf moistening and feeding moisture warning signal value Y_Alarm:

[0191] When -75≤Y_DP≤75, Y_Alarm takes the value 0.

[0192] When -100≤Y_DP<-75 or 75<Y_DP≤100, Y_Alarm takes the value 1.

[0193] When Y_DP>100 or Y_DP<-100, Y_Alarm takes the value 2.

[0194] (IX) Y_Alarm is sent to the Runye feeding PLC controller 13 via the OPC UA server 12. The PLC controller performs early warning / alarm control based on the Y_Alarm value. The control rules are as follows:

[0195] (a) When Y_Alarm=2, the cabinet siren sounds, the warning light flashes, and a red alarm message box pops up on the monitoring screen;

[0196] (b) When Y_Alarm=1, the alarm light of the electric cabinet flashes and a yellow warning message box pops up on the monitoring screen;

[0197] (c) When Y_Alarm=0, the cabinet siren sound is turned off, the warning light stops flashing, and the warning / alarm message box on the monitoring screen disappears.

[0198] During the batch production process of the moistening leaf feeding process, the above steps (5) to (9) are repeated with a cycle of 1 second.

[0199] When there are uncertain and complex factors such as time variation and changes in the process of equipment related to silk-making leaf moistening and feeding, which cause the prediction accuracy of the leaf moistening and feeding moisture prediction model Model_LSTM_P to be lower than 90%, thereby causing an erroneous early warning / alarm signal to be generated, the above steps (three) to (four) can be repeated to perform rolling optimization processing on the leaf moistening and feeding moisture prediction model Model_LSTM_P.

[0200] The prediction accuracy of Model_LSTM_P is equal to the ratio of the number of points where the difference between the predicted value Y_Predict and the true value Y of the leaf moistening moisture in the past month is within the range of ±0.2 to the total number of predicted value samples.

[0201] In this embodiment, the moisture warning method for silk-making leaf moistening feed of the present invention can predict the moisture deviation of the leaf moistening feed nearly 200 seconds later during the production process. When the predicted deviation value exceeds 75% of the process specification range, a warning message is issued in advance. When the predicted deviation value exceeds 100% of the process specification range, an alarm message is issued in advance, so that the operator can make timely corrections in advance when the moisture of the leaf moistening feed may be abnormal, thereby minimizing the possibility of quality defects in the production process of silk-making leaf moistening feed, and having a positive effect on the quality control of the silk-making leaf moistening feed process.

Claims

1. A moisture early warning method for silk-making leaf moistening and feeding, characterized in that: The method comprises a time series database, an OPC UA server and a loose moisture recovery PLC controller; the time series database and the OPC UA server are installed and configured in a silk making centralized control system environment; the loose moisture recovery PLC controller is a device provided by the loose moisture recovery machine; the time series database, the OPC UA server and the loose moisture recovery PLC controller are all interconnected via a network; the method comprises the following steps: S1 establishes the leaf moistening and feeding moisture LSTM neural network model Model_LSTM, which includes: (1) Data normalization scaling layer MinMaxScaler, (2) Long short-term memory network layer LSTM1, (3) Long short-term memory network layer LSTM2, (4) Gated recurrent unit network layer GRU, (5) Batch Normalization layer, (6) Regularization dropout layer, (7) Fully connected neural network layer Dense; S2 collects data related to moisture warning of silk-making leaf moistening and feeding from the leaf moistening and feeding PLC controller through the OPC UA server and stores it in the time series database table; S3 extracts the training data sets X_Test and Y_Test from the production process of at least the last month through the time series database; S4 inputs X_Test and Y_Test into Model_LSTM, and finally obtains a leaf moistening feed moisture prediction model Model_LSTM_P after the training calculation of each neural network layer in step S1; S5 extracts the latest collected information point value list X_Predict from the time series database; S6 inputs X_Predict into Model_LSTM_P, calculates the leaf moistening feed moisture prediction value Y_Predict, and sends Y_Predict to the time series database table for storage; S7 calculates the leaf moistening feed moisture prediction deviation Y_D and the prediction deviation ratio Y_DP; S8 calculates the leaf moistening and feeding moisture warning signal value Y_Alarm; S9 sends Y_Alarm to the Runye feeding PLC controller through the OPC UA server, and the PLC controller performs early warning / alarm control based on the Y_Alarm value.

2. The early warning method according to claim 1, characterized in that: The step S1 further includes: (1) Data normalization scaling layer MinMaxScaler MinMaxScaler receives two matrix data parameters of shape (n, 30) and (, 1) as input data, trains its own parameters, processes the input data, and outputs the results to the long short-term memory network layer LSTM1; where n is the number of samples in the training dataset; (2) Long short-term memory network layer LSTM1 LSTM1 receives the calculation output of the data normalization scaling layer MinMaxScaler, trains the relevant weight parameters of LSTM1 itself, and outputs the result to the long short-term memory network layer LSTM2; The initialization parameters of LSTM1 include: Whether to return the entire sequence return_sequences: the value is true, The dimension of the output space units: the value is 56; (3) Long short-term memory network layer LSTM2 LSTM2 receives the calculated output of the data long short-term memory network layer LSTM1, trains the relevant weight parameters of LSTM2 itself, and outputs the result to the gated recurrent unit network layer GRU. The initialization parameters of LSTM2 include: Whether to return the entire sequence return_sequences: the value is true, The dimension of the output space units: the value is 56; (4) Gated Recurrent Unit Network Layer GRU GRU receives the calculation output of the long short-term memory network layer LSTM2, trains the GRU's own related weight parameters, and outputs the results to the batch normalization layer BatchNormalization. The initialization parameters of GRU include: Whether to return the entire sequence return_sequences: the value is true, The dimension of the output space units: the value is 56; (5) Batch Normalization BatchNormalization receives the calculated output of the gated recurrent unit network layer GRU, trains the relevant weight parameters of BatchNormalization itself, and outputs the result to the regularization dropout layer Dropout; (6) Regularization Dropout Dropout receives the calculated output of the batch normalization layer BatchNormalization, trains the relevant weight parameters of Dropout itself, and outputs the result to the fully connected neural network layer Dense. The initialization parameters of Dropout include: Discard rate: set to 0.

2. (7) Fully connected neural network layer Dense Dense receives the calculated output of the regularized dropout layer Dropout and trains the relevant weight parameters of Dense itself. The initialization parameters of Dense include: The dimension of the output space, units: takes the value 1.

3. The early warning method according to claim 1, characterized in that: The data related to the moisture warning for silk making and leaf moistening in step S2 include: Inlet electronic scale flow rate X1, instantaneous feeding ratio X2, actual SP02 liquid atomization pressure X3, SP04 barrel heating steam source pressure X4, SP01 valve island compressed air source 1 pressure X5, SP03A pump liquid pressure X6, inlet hot air temperature PV value X7, return air temperature PV value X8, heating wet steam flow PV value X9, main water flow control PV value X10, water atomization steam pressure PV value X11, moistening water temperature PV value X12, radiator steam pressure PV value X13, cylinder wall heating steam pressure PV value X14, discharge chamber wall steam pressure P V value X15, preheating cycle hot air temperature PV value X16, actual steam flow X17, dehumidification damper opening X18, main steam flow X19, drum heating wet steam flow X20, front section cylinder wall temperature X21, middle section cylinder wall temperature X22, rear section cylinder wall temperature X23, radiator steam temperature detection value X24, cylinder wall heating steam temperature detection value X25, discharge chamber wall heating steam temperature detection value X26, radiator steam pressure X27, feeder outlet temperature X28, hot air temperature detection value X29, return air temperature detection value X30 and outlet material moisture content Y.

4. The early warning method according to claim 3, characterized in that: The training data sets X_Test and Y_Test in step S3 are: X_Test=[[X1,X2,…,X30]0,[X1,X2,…,X30]1,…,[X1,X2,…,X30] n ]; Y_Test=[Y0,Y1,…,Y n ], where n is the number of samples in the training dataset - 1.

5. The early warning method according to claim 4, characterized in that: The leaf moistening and feeding moisture prediction model Model_LSTM_P in step S4 is: Model_LSTM_P=Model_LSTM(X_Test,Y_Test); In step S5, the latest 120 rows of information point value list X_Predict extracted from the time series database are: X_Predict = [[X1, X2, ..., X30]0, [X1, X2, ..., X30]1, ..., [X1, X2, ..., X30] 119 ]; In step S6, X_Predict is input into Model_LSTM_P, and the predicted value of moisture of the leaf moistening material Y_Predict is calculated as: Y_Predict=Model_LSTM_P(X_Predict).

6. The control method according to claim 5, characterized in that: The predicted deviation Y_D and predicted deviation ratio Y_DP of the moisture content of the moistening material calculated in step S7 are: Y_D=Y_Predict-Y_C, where Y_C is the center value of the moisture specification of the moistening leaf feed. When Y_D=0: Y_DP=0, When Y_D>0: Y_DP=100*Y_D / (Y_USL-Y_C) where Y_USL is the upper limit of the moisture content of the moistening material. When Y_D<0: Y_DP=100*Y_D / (Y_C-Y_LSL) where Y_LSL is the lower limit of the moisture specification for the leaf moistening feed.

7. The control method according to claim 6, characterized in that: The step S8 calculates the leaf moistening feed moisture warning signal value Y_Alarm as: When -75≤Y_DP≤75, Y_Alarm takes the value 0; When -100≤Y_DP<-75 or 75<Y_DP≤100, Y_Alarm takes the value 1; When Y_DP>100 or Y_DP<-100, Y_Alarm takes the value 2.

8. The control method according to claim 7, wherein: In step S9, the PLC controller performs early warning / alarm control according to the Y_Alarm value, and the control rules include: (a) When Y_Alarm=2, the cabinet siren sounds, the warning light flashes, and a red alarm message box pops up on the monitoring screen; (b) When Y_Alarm=1, the alarm light of the electric cabinet flashes and a yellow warning message box pops up on the monitoring screen; (c) When Y_Alarm=0, the cabinet siren sound is turned off, the warning light stops flashing, and the warning / alarm message box on the monitoring screen disappears; When the prediction accuracy of the leaf moistening and feeding moisture prediction model Model_LSTM_P is lower than 90% due to time variation or changes in the process of the silk-making leaf moistening and feeding equipment, repeat the above steps S3 to S4 to perform rolling optimization on the leaf moistening and feeding moisture prediction model Model_LSTM_P; The prediction accuracy of Model_LSTM_P is the ratio of the number of points where the difference between the predicted value Y_Predict of the water content of the moistening material in the past month and the true value Y is within the range of ±0.2 to the total number of predicted value samples.

9. A computer-readable storage medium having a computer program stored thereon, wherein the computer program can be executed by a processor to implement the steps of the moisture early warning method for silk-making leaf moistening and adding materials as described in any one of claims 1 to 8.

10. A moisture early warning system for silk-making leaf moistening and feeding, characterized in that: The system includes a time series database, an OPC UA server and a loose regain PLC controller; the time series database and the OPC UA server are installed and configured in the silk making centralized control system environment, the loose regain PLC controller is a device that comes with the loose regain machine, and the time series database, the OPC UA server and the loose regain PLC controller are all interconnected through the workshop Ethernet; and it also includes a computer-readable storage medium as described in claim 9 that is arranged in the loose regain PLC controller.

Images