Method and device for predicting moisture at loose and moistened leaf outlet
By dynamically updating the super parameters of the BP neural network prediction model, the problem of predicting the outlet moisture of loose moist leaves during cigarette silk making is solved, high-precision prediction is achieved, and the inherent quality of the finished tobacco silk is ensured.
Patent Information
- Application Number
- CN202411522862.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-29
- Publication Date
- 2025-02-07
AI Technical Summary
The prior art is difficult to effectively predict the loose moisturized leaves outlet moisture during the process of making cigarette silk, resulting in the impact of the internal quality of the finished tobacco wire.
The super parameters of the BP neural network prediction model are dynamically updated, and the prediction model is adjusted in real time according to the dynamic changes of actual equipment and materials to maintain high-precision prediction.
By dynamically updating the super parameters of the BP neural network model, the moisture in the loose moist leaves can be accurately predicted and the inherent quality of the finished tobacco is ensured.
Smart Images
Figure CN119398266A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of silk production, and more specifically, to a method and device for predicting the moisture content at the outlet of loose moistened leaves. Background Art
[0002] The control principle of the tobacco leaf moistening machine is mainly achieved by controlling the working status and parameters of the heater, water flow control system, fan and other equipment. Specifically, the control system of the tobacco leaf moistening machine is usually composed of a PLC controller, a touch screen, a sensor, an actuator, etc. By collecting and processing the temperature, humidity, pressure and other parameters fed back by the sensor, the working status and parameters of the heater, water flow control system, fan and other equipment are controlled, so as to control and adjust the key parameters such as temperature, humidity, airflow and so on in the tobacco leaf making process, so as to ensure the quality and taste of the cigarettes and improve production efficiency. In addition, a dehumidification system is set inside the cylinder to extract part of the water-containing and heat-containing exhaust gas for discharge to keep the temperature of the material constant and indirectly affect the moisture content of the material.
[0003] At present, the moisture content of loose moistened leaves at the tobacco-making workshop of a cigarette factory is generally set by process personnel or job operators based on experience. However, when the characteristics of cigarette packs and process control parameters change, the moisture content of loose moistened leaves set by experience is often unable to meet the drying moisture required for the key tobacco drying process. The temperature of the barrel wall needs to be adjusted to a large extent, which exceeds the control range required by the standard and affects the intrinsic quality of the finished tobacco.
[0004] In order to solve the problems caused by experience, a machine learning model with relatively static hyperparameters is usually used to predict the moisture content of loose moistened leaves at the outlet. However, since the characteristics of cigarette packs and equipment are in a dynamic change process, the relatively static model and hyperparameters cannot always maintain a high degree of fit. Summary of the invention
[0005] The present application provides a method and device for predicting the moisture content of loose moistened leaves at the outlet, which dynamically updates the super parameters of the BP neural network prediction model according to the dynamic changes of actual equipment and materials, maintains the high-precision prediction of the BP neural network prediction model, and ensures the intrinsic quality of the finished tobacco.
[0006] The present application provides a method for predicting the moisture content of loose leaf moistening outlets, comprising:
[0007] Receive real-time data related to the moisture content of loose leaf outlets that matches the BP neural network prediction model; wherein the super parameters of the BP neural network prediction model are dynamically updated according to the dynamic changes of equipment and materials;
[0008] The real-time data is input into the BP neural network prediction model to obtain the predicted value of the moisture content at the loose leaf outlet.
[0009] Preferably, constructing a BP neural network prediction model specifically includes:
[0010] Constructing a BP neural network initial model, wherein the BP neural network initial model has a first number of hidden layer nodes;
[0011] Based on the first data set, target optimization is performed on the super parameters to be optimized of the BP neural network initial model to obtain the optimal parameter values;
[0012] Bringing the optimal parameter value into the BP neural network initial model, and using the first data set for training to obtain the first BP neural network prediction model;
[0013] A second data set is obtained based on dynamic changes of equipment and materials, and the super parameters of the first BP neural network prediction model are dynamically updated using the second data set to obtain a BP neural network prediction model.
[0014] Preferably, the method for predicting the moisture content at the outlet of loose moistened leaves also includes optimizing the network structure of the BP neural network prediction model based on dynamic changes of equipment and materials.
[0015] Preferably, a simulated annealing algorithm is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model.
[0016] Preferably, obtaining the data set required for training specifically includes:
[0017] Obtain historical data sets related to loose leaf outlet moisture;
[0018] Preprocess historical data to obtain feature data sets;
[0019] Perform dimensionality reduction processing on the features in the feature data set to obtain a first data set.
[0020] The present application also provides a device for predicting the moisture content at the outlet of loose leaf moistening, comprising a receiving module and a predicting module;
[0021] The receiving module is used to receive real-time data related to the moisture content of loose leaf moistening outlets that matches the BP neural network prediction model; wherein the super parameters of the BP neural network prediction model are dynamically updated according to the dynamic changes of equipment and materials;
[0022] The prediction module is used to input real-time data into the BP neural network prediction model to obtain the predicted value of the moisture content at the loose leaf outlet.
[0023] Preferably, the prediction device for the moisture at the outlet of loose leaves further comprises a model building module, which comprises an initial model building module, an optimization module, a first training module and a super parameter updating module;
[0024] The initial model building module is used to build a BP neural network initial model, and the BP neural network initial model has a first number of hidden layer nodes;
[0025] The optimization module is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model based on the first data set to obtain the optimal parameter value;
[0026] The first training module is used to bring the optimal parameter value into the BP neural network initial model, and use the first data set for training to obtain the first BP neural network prediction model;
[0027] The super parameter updating module is used to obtain a second data set based on the dynamic changes of equipment and materials, and dynamically update the super parameters of the first BP neural network prediction model using the second data set to obtain the BP neural network prediction model.
[0028] Preferably, the device for predicting the moisture content at the outlet of loose moistened leaves also includes a model optimization module, which is used to optimize the network structure of the BP neural network prediction model based on dynamic changes of equipment and materials.
[0029] Preferably, the optimization module is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model using a simulated annealing algorithm.
[0030] Preferably, the model building module further includes a data set acquisition module, which includes a historical data set acquisition module, a preprocessing module and a dimensionality reduction module;
[0031] The historical data set acquisition module is used to acquire the historical data set related to the moisture at the loose leaf moistening outlet;
[0032] The preprocessing module is used to preprocess the historical data to obtain the feature data set;
[0033] The dimension reduction module is used to perform dimension reduction processing on the features in the feature data set to obtain a first data set.
[0034] Other features and advantages of the present application will become apparent from the following detailed description of exemplary embodiments of the present application with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the application and, together with the description, serve to explain the principles of the application.
[0036] Figure 1 A flow chart of a method for predicting the moisture content of loose leaf outlets provided in this application;
[0037] Figure 2 Structural diagram of the device for predicting moisture at the loose leaf moistening outlet provided in this application. DETAILED DESCRIPTION
[0038] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present application.
[0039] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.
[0040] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.
[0041] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0042] The present application provides a method and device for predicting the moisture content of loose moistened leaves at the outlet, which dynamically updates the super parameters of the BP neural network prediction model according to the dynamic changes of actual equipment and materials, maintains the high-precision prediction of the BP neural network prediction model, and ensures the intrinsic quality of the finished tobacco.
[0043] like Figure 1 As shown, the method for predicting the moisture content of loose leaf outlet provided by the present application includes:
[0044] S110: Receive real-time data related to the moisture at the loose leaf moistening outlet that matches the BP neural network prediction model.
[0045] As an embodiment, the data related to the moisture content at the loose leaf moistening outlet include 11 variables, including the first cigarette pack type in the current leaf moistening machine cylinder, the second cigarette pack type in the cylinder, the third cigarette pack type in the cylinder, the water addition flow rate, the steam flow rate, the steam pressure, the hot air temperature, the outlet material temperature, the ambient temperature, the dehumidification negative pressure, and the hot air speed.
[0046] Among them, the first cigarette package grade in the leaf moistening machine barrel is the material type of the cigarette package that enters the leaf moistening machine barrel first. The cigarette package origin, grade and material number are used as distinguishing parameters, and each material type is assigned a unique integer; similarly, the second cigarette package type in the barrel and the third cigarette package type in the barrel are the material types of the second and third cigarette packages entering the leaf moistening machine barrel respectively.
[0047] In general, the materials inside the cylinder of the leaf moistening machine will only mix three cigarette packs in sequence, and the types of these three cigarette packs are known on the production line. According to the accumulation of the belt scale in front of the leaf moistening machine, the type of cigarette pack can be inferred after a delay. As production proceeds, the cigarette packs in the cylinder of the leaf moistening machine continuously update the value of the cigarette pack type according to the first-in-first-out rule, and the material type settings of the cigarette packs are shown in the following table:
[0048] Cigarette Pack Material Number Material type setting integer ABCXXX001 1 ABCXXX002 2 ABCXXX003 3 …… …… ABCXXX00N N
[0049] In the above table, ABCXXX represents the material number of a specific cigarette package, which is defined internally by the enterprise; N represents the number of material types.
[0050] Preferably, the data related to the moisture content at the outlet of loose leaf moistening matched with the BP neural network prediction model include the first cigarette pack type, the second cigarette pack type, and the third cigarette pack type in the cylinder of the leaf moistening machine, and also include multiple (≤8) variables obtained after dimensionality reduction processing of the eight variables of water addition flow, steam flow, steam pressure, hot air temperature, outlet material temperature, ambient temperature, dehumidification negative pressure, and hot air speed. For details, please see the following description of model construction.
[0051] S120: Input the real-time data into the BP neural network prediction model to obtain the predicted value of the moisture at the loose leaf moistening outlet.
[0052] Among them, the input data of the BP neural network prediction model is the data related to the moisture content of the loose leaf moistening outlet that matches the BP neural network prediction model, and the output data is the predicted value of the moisture content of the loose leaf moistening outlet.
[0053] It should be noted that the super parameters of the BP neural network prediction model are dynamically updated according to the dynamic changes of equipment and materials, so that the parameters of the model are adaptively updated as the dynamic changes of equipment and materials occur, maintaining a high degree of fit of the model, improving the prediction accuracy of the BP neural network prediction model, and ensuring the intrinsic quality of the finished tobacco.
[0054] As an embodiment, constructing a BP neural network prediction model specifically includes:
[0055] P1: Construct an initial BP neural network model, where the initial BP neural network model has a first number of hidden layer nodes.
[0056] P2: Based on the first data set, the hyper parameters to be optimized of the BP neural network initial model are optimized to obtain the optimal parameter values.
[0057] As an embodiment, the continuous weight ω ij as the hyperparameter to be optimized.
[0058] Specifically, in order to obtain the optimal weights of the BP neural network, the square sum error function of the predicted value and the actual value of the BP neural network is used as the target optimization function:
[0059]
[0060] Among them, y i is the predicted value calculated according to the above formula, is the actual value given by the sample, and J is the target optimization function value. The solution of this optimization problem is defined as the connection weight ω ij .
[0061] Preferably, a simulated annealing algorithm is used to target and optimize the super parameters to be optimized of the BP neural network initial model. The simulated annealing algorithm seeks a global optimal solution in a larger search space in time and space through a probabilistic algorithm.
[0062] When the simulated annealing algorithm is used for optimization, at each first temperature T k Multiple local optimal solutions are searched at the first temperature T k The thermal equilibrium is reached at the first temperature T k The number of internal cycles is greater than the set value) and then the temperature is lowered to the second temperature T k+1 (T k+1 <T k ), at the second temperature T k+1 The local optimal solution is searched for multiple times until the temperature reaches the threshold G. The local optimal solution at this time is the global optimal solution with a probability of 1. The global optimal solution is the optimal connection weight of the BP neural network.
[0063] For each temperature T k , assuming that the current solution is ω(l) (the solution of the lth search step), the search process of each search step is as follows:
[0064] S1: Use the following random perturbation model to generate multiple new solutions ω′(l):
[0065] ω′(l)=ω(l)+ηε (2)
[0066] Among them, η is the disturbance amplitude parameter, ε is the random disturbance variable, and the random disturbance model can be selected to obey the common probability density function similar to Cauchy, Gaussian, and uniform distribution. Different random disturbance variables will produce different new solutions.
[0067] S2: Calculate the target optimization function value corresponding to each new solution ω′(l).
[0068] Specifically, each of the above new solutions ω′(l) is used as the connection weight of the BP neural network initial model, and the BP neural network initial model predicts the training samples to obtain the predicted value y i , and then calculate the target optimization function value J according to formula (1).
[0069] S3: Calculate the probability of taking each new solution ω′(l) as the next new solution ω(l+1).
[0070] The probability of accepting ω′(l) as the next new solution ω(l+1) is:
[0071]
[0072] S4: Use the roulette method to select one of the ω′(l) as the next new solution ω(l+1). The roulette method is a prior art and will not be described in detail here.
[0073] P3: Bring the optimal parameter values into the BP neural network initial model, and use the first data set for training to obtain the first BP neural network prediction model.
[0074] Based on the above, in this application, in the actual production process, the construction of the BP neural network prediction model also includes:
[0075] P4: Based on the dynamic changes of equipment and materials, a second data set is obtained, and the super parameters of the first BP neural network prediction model are dynamically updated using the second data set to obtain a BP neural network prediction model.
[0076] The learning ability and generalization ability of the BP neural network are contradictory to each other. Generally, the error tends to decrease gradually, while the estimation error will gradually increase. Based on this consideration, in order to take into account the learning ability and generalization ability of the BP neural network, preferably, in the prediction method of the moisture content at the outlet of loose moist leaves provided by the present application, after P3 (i.e., the model building stage), it also includes optimizing the network structure of the BP neural network prediction model.
[0077] As an embodiment, optimizing the network structure of the BP neural network prediction model specifically includes:
[0078] Q1: Modify the number of hidden layer nodes of the BP neural network prediction model to multiple second numbers to obtain multiple second BP neural network prediction models, wherein the first number is different from the second number, and the number of hidden layer nodes can be increased or decreased.
[0079] Q2: Use the first data set to train each second BP neural network prediction model.
[0080] Q3: The second BP neural network prediction model with the best prediction results is used as the optimized BP neural network prediction model.
[0081] Preferably, obtaining a data set required for training (such as a first data set and a second data set) specifically includes:
[0082] R1: Obtain historical data sets related to the moisture at the loose leaf outlet.
[0083] As an embodiment, historical data of the above 11 variables are collected as a historical data set.
[0084] R2: Preprocess the historical data to obtain the feature data set.
[0085] As an embodiment, preprocessing includes removing data outliers and normalizing the data of eight variables, namely, water addition flow, steam flow, steam pressure, hot air temperature, outlet material temperature, ambient temperature, dehumidification negative pressure, and hot air speed.
[0086] R3: Perform dimensionality reduction processing on the features in the feature data set to obtain the first data set.
[0087] Preferably, the Spearman correlation coefficient method is used to perform dimensionality reduction processing on the features in the feature data set to reduce redundancy.
[0088] Specifically, the formula for calculating the spearman correlation coefficient is:
[0089]
[0090] In the above formula, ρ is the spearman correlation coefficient, d i is the difference in the rank value of the i-th data pair, and n is the total number of samples.
[0091] In this application, the Spearman correlation coefficient ρ between eight variables, namely, water flow rate, steam flow rate, steam pressure, hot air temperature, outlet material temperature, ambient temperature, dehumidification negative pressure, and hot air speed, is calculated. For each variable, d i is the difference between the rank values of this variable and other variables.
[0092] If |ρ|≤0.3, the variable is discarded, thereby reducing the dimension of the data. The data set composed of the data of the variable after dimension reduction is the data set required for training.
[0093] In the actual production process, the data of the reduced-dimensional variables is directly collected as real-time data.
[0094] On the basis of the above, the method for predicting the moisture content of loose moistened leaves at the outlet provided in the present application also includes optimizing the network structure of the BP neural network prediction model based on the dynamic changes of equipment and materials. Please refer to steps Q1-Q3 for details. The difference is that the data set used to train the model is the most recent second data set obtained based on the dynamic changes of equipment and materials.
[0095] Based on the above, the present application also provides a device for predicting the moisture content of the loose leaf moistening outlet. Figure 2 As shown, the device for predicting the moisture at the outlet of loose leaf moistening includes a receiving module 210 and a prediction module 220.
[0096] The receiving module 210 is used to receive real-time data related to the moisture content of the loose leaf moistening outlet that matches the BP neural network prediction model, wherein the super parameters of the BP neural network prediction model are dynamically updated according to the dynamic changes of the equipment and materials.
[0097] The prediction module 220 is used to input real-time data into the BP neural network prediction model to obtain the predicted value of the moisture at the loose leaf moistening outlet.
[0098] Preferably, the prediction device for the moisture at the loose leaf outlet also includes a model construction module 230, and the model construction module 230 includes an initial model construction module 2301, an optimization module 2302, a first training module 2303 and a hyper-parameter updating module 2304.
[0099] The initial model building module 2301 is used to build a BP neural network initial model, and the BP neural network initial model has a first number of hidden layer nodes.
[0100] The optimization module 2302 is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model based on the first data set to obtain the optimal parameter values.
[0101] The first training module 2303 is used to bring the optimal parameter value into the BP neural network initial model, and use the first data set for training to obtain the first BP neural network prediction model.
[0102] The super parameter updating module 2304 is used to obtain a second data set based on dynamic changes in equipment and materials, and dynamically update the super parameters of the first BP neural network prediction model using the second data set to obtain a BP neural network prediction model.
[0103] Preferably, the prediction device for the moisture at the loose leaf outlet also includes a model optimization module 240, which is used to optimize the network structure of the BP neural network prediction model during the model construction stage or to optimize the network structure of the BP neural network prediction model based on dynamic changes of equipment and materials.
[0104] Preferably, the model optimization module 240 includes a hidden layer modification module 2401 , a second training module 2402 , and a model determination module 2403 .
[0105] The hidden layer modification module 2401 is used to modify the number of hidden layer nodes of the BP neural network prediction model to multiple second numbers to obtain multiple second BP neural network prediction models, wherein the first number is different from the second number.
[0106] The second training module 2402 is used to train each second BP neural network prediction model using the second data set.
[0107] The model determination module 2403 is used to use the second BP neural network prediction model with the best prediction result as the optimized BP neural network prediction model.
[0108] Preferably, the optimization module 2302 is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model using a simulated annealing algorithm.
[0109] Preferably, the model building module 230 also includes a data set acquisition module 2305 , and the data set acquisition module 2305 includes a historical data set acquisition module 23051 , a preprocessing module 23052 and a dimensionality reduction module 23053 .
[0110] The historical data set acquisition module 23051 is used to obtain the historical data set related to the moisture at the loose leaf moistening outlet.
[0111] The preprocessing module 23052 is used to preprocess the historical data to obtain a feature data set.
[0112] The dimension reduction module 23053 is used to perform dimension reduction processing on the features in the feature data set to obtain a first data set.
[0113] Preferably, the dimension reduction module 23053 is used to perform dimension reduction processing on the features in the feature data set using the Spearman correlation coefficient method.
[0114] The beneficial effects of this application are as follows:
[0115] 1. This application combines multi-objective optimization strategies with actual equipment and material dynamic changes to automatically adjust the super parameters and network structure of the prediction model, so that the model maintains high-precision prediction capabilities.
[0116] 2. This application uses the Spearman correlation coefficient method to reduce redundancy and correlation of preprocessed data, which better achieves data dimensionality reduction and ensures the accuracy of model fitting.
[0117] 3. The BP neural network algorithm used in this application does not have problems such as too many parameters, slow learning speed and poor nonlinear ability, and can stably predict results, especially in small sample industrial applications, it can show fast and stable characteristics, thus meeting actual needs.
[0118] 4. This application uses a simulated annealing optimization algorithm to seek the optimal solution for the connection weights of the BP neural network, overcoming the defect that the connection weights of the traditional BP neural network are prone to fall into local minima.
[0119] 5. In order to improve the generalization ability of BP network, this application utilizes the strategy optimization of BP network training to further improve the accuracy of model prediction from two aspects: the selection of training samples and the number of hidden layer nodes.
[0120] Although some specific embodiments of the present application have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are only for illustration, not for limiting the scope of the present application. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present application. The scope of the present application is defined by the appended claims.
Claims
1. A method for predicting the moisture content of loose leaf moistening outlet, characterized in that: include: Receive real-time data related to the moisture content of loose leaf moistening outlets that matches the BP neural network prediction model; wherein the super parameters of the BP neural network prediction model are dynamically updated according to the dynamic changes of equipment and materials; The real-time data is input into the BP neural network prediction model to obtain the predicted value of the moisture content at the loose leaf moistening outlet.
2. The method for predicting the moisture content of loose leaf outlet according to claim 1, characterized in that: Constructing the BP neural network prediction model specifically includes: Constructing a BP neural network initial model, wherein the BP neural network initial model has a first number of hidden layer nodes; Based on the first data set, target optimization is performed on the super parameters to be optimized of the BP neural network initial model to obtain the optimal parameter values; Bringing the optimal parameter value into the BP neural network initial model, and training it with the first data set to obtain a first BP neural network prediction model; A second data set is obtained based on dynamic changes of equipment and materials, and the super parameters of the first BP neural network prediction model are dynamically updated using the second data set to obtain the BP neural network prediction model.
3. The method for predicting the moisture content of loose leaf moistening outlet according to claim 2, characterized in that: It also includes optimizing the network structure of the BP neural network prediction model based on dynamic changes of equipment and materials.
4. The method for predicting the moisture content of loose leaf outlet according to claim 2, characterized in that: The simulated annealing algorithm is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model.
5. The method for predicting the moisture content of loose leaf moistening outlet according to claim 2, characterized in that: Get the data set required for training, including: Obtain historical data sets related to loose leaf outlet moisture; Preprocessing the historical data to obtain a feature data set; Perform dimensionality reduction processing on the features in the feature data set to obtain the first data set.
6. A device for predicting the moisture content of loose leaf outlets, characterized in that: It includes a receiving module and a prediction module; The receiving module is used to receive real-time data related to the moisture content of the loose leaf moistening outlet that matches the BP neural network prediction model; wherein the super parameters of the BP neural network prediction model are dynamically updated according to the dynamic changes of equipment and materials; The prediction module is used to input the real-time data into the BP neural network prediction model to obtain the predicted value of the moisture content at the loose leaf moistening outlet.
7. The device for predicting the moisture content at the outlet of loose leaves according to claim 6, characterized in that: It also includes a model construction module, which includes an initial model construction module, an optimization module, a first training module, and a super parameter update module; The initial model building module is used to build a BP neural network initial model, and the BP neural network initial model has a first number of hidden layer nodes; The optimization module is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model based on the first data set to obtain the optimal parameter value; The first training module is used to bring the optimal parameter value into the BP neural network initial model, and use the first data set for training to obtain a first BP neural network prediction model; The super parameter updating module is used to obtain a second data set based on dynamic changes of equipment and materials, and use the second data set to dynamically update the super parameters of the first BP neural network prediction model to obtain the BP neural network prediction model.
8. The device for predicting the moisture content at the outlet of loose leaves according to claim 7, characterized in that: It also includes a model optimization module, which is used to optimize the network structure of the BP neural network prediction model based on dynamic changes of equipment and materials.
9. The device for predicting the moisture content at the outlet of loose leaves according to claim 7, characterized in that: The optimization module is used to perform target optimization on the super parameters to be optimized of the BP neural network initial model by using a simulated annealing algorithm.
10. The device for predicting the moisture content at the outlet of loose leaves according to claim 7, characterized in that: The model building module also includes a data set acquisition module, which includes a historical data set acquisition module, a preprocessing module and a dimensionality reduction module; The historical data set acquisition module is used to acquire the historical data set related to the moisture at the loose leaf moistening outlet; The preprocessing module is used to preprocess the historical data to obtain a feature data set; The dimension reduction module is used to perform dimension reduction processing on the features in the feature data set to obtain the first data set.