Method for predicting temperature and humidity field of curing barn in tobacco leaf curing process

Abstract

The invention belongs to the field of intelligent tobacco leaf curing, and discloses a curing barn temperature and humidity field prediction method in the tobacco leaf curing process, which is characterized in that on the basis of time sequence measurement data of temperature and humidity of different areas of a curing barn in the tobacco leaf curing process, a neural network and COMSOL simulation are coupled, the internal temperature and humidity temporal and spatial change condition of the curing barn is predicted, and a Poisson fusion algorithm is combined to predict the temperature and humidity field of the curing barn. Discrete prediction of a neural network and spatial gradient distribution simulated by COMSOL are effectively combined, continuous temperature and humidity spatial distribution is achieved, and the result shows that the curing barn temperature and humidity field prediction method in the tobacco leaf curing process can accurately predict the change rule of temperature and humidity distribution in a curing barn; and a technical support is provided for regulation and control, loss reduction and efficiency improvement in the tobacco leaf curing process.

Description

Technical Field

[0001] The invention belongs to the field of intelligent tobacco leaf baking, and provides a method for simulating the temperature and humidity field in a baking room based on a neural network, namely, a method for predicting the temperature and humidity field in a baking room during the tobacco leaf baking process. Background Art

[0002] Tobacco leaf baking occupies a core position in the tobacco industry. It is not only a key step in converting tobacco leaves from agricultural products to industrial raw materials, but also a soul process for shaping tobacco quality. The control of temperature and humidity in tobacco leaf baking is extremely important because they directly affect the chemical composition conversion, appearance quality, physical properties and sensory quality of tobacco leaves, which in turn affect the quality and value of the final tobacco products. Accurate control of temperature and humidity is crucial to the formation of tobacco leaf quality.

[0003] Traditionally, the baking process relies on the judgment and manual adjustments of experienced technicians, but this method is difficult to achieve consistency and repeatability, especially in a large-scale production environment, which can lead to instability in tobacco leaf quality and waste of resources. During the baking process, the uneven distribution of temperature and humidity can cause uneven drying of tobacco leaves, which in turn affects the color, aroma and burning performance of tobacco leaves. With the advancement of computer technology and artificial intelligence, especially the development of neural networks, more advanced and automated methods can now be used to optimize temperature and humidity control during the baking process. Summary of the invention

[0004] The purpose of the present invention is to propose a method for predicting the temperature and humidity field in a tobacco baking room during tobacco baking based on the above-mentioned prior art conditions, so as to solve the problem of uneven temperature and humidity distribution in the baking room under the prior art.

[0005] The objective of the present invention is achieved through the following technical solutions:

[0006] A method for predicting the temperature and humidity field in a flue-curing room during the tobacco leaf curing process is based on the time series measurement data of temperature and humidity in different locations of the flue-curing room during the tobacco leaf curing process. The method couples the neural network with the COMSOL simulation to predict the temporal and spatial changes of temperature and humidity inside the flue-curing room. The Poisson fusion algorithm is combined to effectively merge the discrete prediction of the neural network with the spatial gradient distribution of the COMSOL simulation to achieve continuous spatial distribution of temperature and humidity. The specific steps are as follows:

[0007] S1: Place portable temperature and humidity sensors evenly in the baking room, with 6 sensors on each of the upper, middle and lower layers, and 2 sensors on the front, middle and back, for a total of 18 sensors;

[0008] S2: Establish a COMSOL model. The geometric model includes the following parts: dehumidification port, return air port, air inlet and door. The assumptions in the simulation are as follows: During the baking process, the tobacco leaves are treated uniformly, so that the tobacco leaves and air are regarded as porous media. During the curing stage, the liquid phase and gas phase in the simplified porous tobacco sample are in equilibrium;

[0009] S3: Neural network simulation modeling, using temperature, humidity, input area (i.e., location), and historical data as input to construct a baking state encoding network, an interval encoding network, and a history encoding network. In the prediction network, the feature vectors of the three networks are concatenated and a residual network is introduced to predict temperature and humidity.

[0010] The step S3 comprises the following steps:

[0011] S31: For the baked state encoding network and the interval encoding network, the method adopts three-layer and one-layer fully connected networks respectively.

[0012] S32: The temperature and humidity of different warehouse areas are predicted by using residual prediction, that is, the temperature and humidity status of other areas at the current time is determined by the temperature and humidity status of the known area and the residual value predicted by the neural network.

[0013] S33: This method uses the mean square error between the predicted temperature and humidity values ​​and the actual temperature and humidity values ​​as the optimization objective function L, and uses the Adam optimization algorithm to optimize the objective function.

[0014] S4: Use Poisson fusion algorithm to merge the discrete prediction of neural network with the spatial gradient distribution simulated by COMSOL. Combine the predicted value of neural network with the changes in temperature and humidity field simulated by COMSOL;

[0015] In step S1, the flue-curing room is loaded with tobacco leaves by hanging rods, with a total of three layers of tobacco leaves hanging. Portable temperature and humidity sensors are evenly placed inside, with 6 sensors in each of the upper, middle and lower layers, and 2 sensors in the front, middle and back, for a total of 18 sensors. They are placed at the intersection of one quarter and one third of the length and width of the flue-curing room, respectively, to collect temperature and relative humidity data. Based on the average values ​​of A-1 and A-2 measured simultaneously, a standard curve for the upward airflow flue-curing room is drawn.

[0016] In step S32, a neural network simulation model is established for temperature and humidity, which is mainly composed of three encoding networks and one prediction network. The encoding network includes a baking state encoding network, an interval encoding network, and a history encoding network, and the temperature and humidity of different baking room areas are predicted by residual prediction, that is, the temperature and humidity state of other areas at the current time is determined by the temperature and humidity state of the known area and the residual value predicted by the neural network.

[0017] The formula is as follows

[0018]

[0019] In the above formula: and To predict the temperature and humidity of the vth region, add the residual value of the prediction network to the temperature and humidity of the kth region to get the predicted value of the temperature and humidity of the vth region. and

[0020] For the baked state encoding network and interval encoding network, the method adopts three-layer and one-layer fully connected networks respectively. For the history encoding network, in order to more effectively capture the characteristics of historical data, based on it, the temporal history information is extracted on the multi-layer SE-ResNet Blocks, and Squeeze and Exitation are used to extract deep features, where Squeeze captures the global feature h through global pooling. global ,Right now:

[0021]

[0022] where h G ∈R M×D is the D-dimensional feature h corresponding to the M moments in the historical temperature and humidity set G i G The Exitation operation captures the dependencies between channels through a gating mechanism with sigmoid activation, specifically:

[0023] F squeeze (h global )=Sigmoid(FC(ReLU(FC(h global ))

[0024] In step S4, the Poisson fusion method is used to optimize the results of COMSOL and neural network. The formula is as follows:

[0025]

[0026] It is represented by V, assuming that V is the gradient field of u, where u is the set of discrete points predicted by the neural network, S is the fused field of temperature and humidity, and Ω represents the fused field. is the area covered by the boundary. The field inside Ω is denoted by f, while the field outside Ω is denoted by f*.

[0027] The advantages of the present invention over the prior art are: the model uses part of the time series data that is spatially discrete, and adopts a neural network algorithm. To achieve continuous spatial distribution, a Poisson fusion algorithm is used to effectively merge the discrete prediction of the neural network with the spatial gradient distribution simulated by COMSOL. By accurately predicting and controlling temperature and humidity, the tobacco baking process is optimized, the baking efficiency and product quality are improved, and technical support is provided for the regulation, loss reduction and efficiency improvement of the tobacco leaf baking process. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a diagram of the prediction method model of the present invention.

[0029] Figure 2 It is the data collection position of the baking room sensor in the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The embodiment of the present invention uses part of the time series data with spatial discreteness and adopts a neural network algorithm. In order to achieve continuous spatial distribution, a Poisson fusion algorithm is used to effectively merge the discrete prediction of the neural network with the spatial gradient distribution simulated by COMSOL. By accurately predicting and controlling temperature and humidity, the tobacco baking process is optimized.

[0032] like Figure 1 As shown, a neural network assisted temperature and humidity field in a tobacco baking process includes the following steps:

[0033] 1) The dimensions of the curing room device 1 are 800cm long, 270cm wide and 350cm high. The curing room uses a hanging rod method to load tobacco leaves using a tobacco hanging rod component 3, with a total of three layers of tobacco leaves hanging. Portable temperature and humidity sensor components 2 are evenly placed inside, with 6 sensors in each of the upper, middle and lower layers, and 2 sensors in the front, middle and back, for a total of 18 sensors. The sensors are hung in the middle of the tobacco rod with cotton ropes and enter the curing room together with the loaded tobacco leaves. They are placed at the intersection of one quarter and one third of the length and width of the curing room, respectively, to obtain the real-time temperature and humidity in the curing room. Further, temperature and relative humidity data are collected. Based on the average values ​​of A-1 and A-2 measured at the same time, the standard curve of the upward airflow curing room is drawn.

[0034] 2) A COMSOL simulation model was established with the following simulation assumptions: During the curing process, tobacco leaves are treated uniformly, so that tobacco leaves and air are regarded as porous media. During the curing stage, the liquid and gas phases in the simplified porous tobacco sample are in equilibrium.

[0035] 3) A neural network simulation model is established for temperature and humidity, which is mainly composed of three encoding networks and one prediction network. The encoding network includes a baking state encoding network, an interval encoding network, and a history encoding network, and the temperature and humidity of different baking room areas are predicted by residual prediction, that is, the temperature and humidity status of other areas at the current time is determined by the temperature and humidity status of the known area and the residual value predicted by the neural network.

[0036] 4) The results from COMSOL and the neural network are optimized using a Poisson fusion method to synchronize the spatial gradient data from the simulation with the accurate temporal predictions of the neural network in discrete space, thereby producing high-resolution predictions and timing of temperature and humidity.

[0037] It should be noted that the method for predicting the temperature and humidity field in a flue-curing barn during the tobacco leaf curing process involved in this embodiment is applicable to all types of flue-curing barns, such as heat pump intensive, biomass type, airflow rising type, airflow descending type, etc.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting temperature and humidity field in a tobacco baking room during baking, characterized in that: This method is based on the time series measurement data of temperature and humidity in different locations of the flue-curing barn during the tobacco leaf curing process. It couples the neural network with COMSOL simulation to predict the temporal and spatial changes of temperature and humidity inside the flue-curing barn. Combined with the Poisson fusion algorithm, the discrete prediction of the neural network is effectively merged with the spatial gradient distribution of COMSOL simulation to achieve continuous spatial distribution of temperature and humidity. It can be effectively applied to multiple types of flue-curing barns. The specific steps are as follows: S1: Place portable temperature and humidity sensors evenly in each baking room; S2: Establish a COMSOL model. The geometric model includes the following parts: dehumidification port, return air port, air inlet and door. The assumptions in the simulation are as follows: During the baking process, the tobacco leaves are treated uniformly, so that the tobacco leaves and air are regarded as porous media. During the curing stage, the liquid phase and gas phase in the simplified porous tobacco sample are in equilibrium; S3: Neural network simulation modeling, using temperature, humidity, location, and historical data as input to construct a baking state encoding network, an interval encoding network, and a history encoding network. In the prediction network, the feature vectors of the three networks are concatenated and a residual network is introduced to predict temperature and humidity. S4: The discrete predictions of the neural network are merged with the spatial gradient distribution simulated by COMSOL using the Poisson fusion algorithm, while taking into account the variations in the temperature and humidity fields simulated by COMSOL.

2. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 1, characterized in that: In step S1, the specific placement of the temperature and humidity sensors is: 6 sensors in each of the upper, middle and lower layers, and 2 sensors in the front, middle and back respectively, for a total of 18 sensors.

3. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 1, characterized in that: The step S3 comprises the following steps: S31: For the baking state encoding network and the interval encoding network, the method uses a three-layer and a one-layer fully connected network respectively; S32: Predicting the temperature and humidity of different warehouse areas by using residual prediction, that is, determining the temperature and humidity status of other areas at the current time by using the temperature and humidity status of the known area and the residual value predicted by the neural network; S33: This method uses the mean square error between the predicted temperature and humidity values ​​and the actual temperature and humidity values ​​as the optimization objective function L, and uses the Adam optimization algorithm to optimize the objective function.

4. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 1 or 2, characterized in that: The sensors are placed at one third of the length and one quarter of the width of the oven.

5. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 3, characterized in that: In S32, the temperature and humidity state of other areas at the current time is determined by the temperature and humidity state of the known area and the residual value predicted by the neural network, and the formula is as follows: In the above formula: and To predict the residual value of the temperature and humidity predicted by the prediction network for the vth region, add it to the temperature and humidity of the kth region to get the predicted value of the temperature and humidity of the vth region and 6. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 1, characterized in that: For the history encoding network, in order to more effectively capture the characteristics of historical data, based on it, the temporal history information is extracted on the multi-layer SE-ResNet Blocks, and Squeeze and Exitation are used to extract deep features, where Squeeze captures the global feature h through global pooling. global ,Right now: where h G ∈R M×D is the D-dimensional feature h corresponding to the M moments in the historical temperature and humidity set G i G The Exitation operation captures the dependencies between channels through a gating mechanism with sigmoid activation, specifically: F squeeze (h global )=Sigmoid(FC(ReLU(FC(h global ))。 7. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 1, characterized in that: In step S4, the Poisson fusion method is used to optimize the results of COMSOL and neural network. The formula is as follows: It is represented by V, assuming that V is the gradient field of u, where u is the set of discrete points predicted by the neural network, S is the fused field of temperature and humidity, and Ω represents the fused field. is the area covered by the boundary; the field inside Ω is denoted by f, and the field outside Ω is denoted by f*.

8. The method for predicting temperature and humidity field in a tobacco leaf curing room during the curing process according to claim 1, characterized in that: This prediction method is applicable to all types of curing rooms, such as heat pump intensive, biomass, airflow rising, airflow descending, etc.

Images