Intelligent control method for tobacco airflow cut-tobacco dryer

Abstract

The invention provides an intelligent control method for a tobacco airflow cut-tobacco dryer, which comprises the steps of constructing an environment model and constructing a strategy algorithm, and specifically comprises the following steps: S1, simulating a cut-tobacco drying process of the tobacco airflow cut-tobacco dryer by adopting the environment model, driving a neural network through historical data to complete virtual modeling, and predicting future outlet moisture; and S2, finding an optimal action (water adding amount) through a strategy algorithm according to the prediction of the future outlet moisture by the environment model, and making a corresponding decision to achieve the purpose of control. The moisture control effect can be improved, rapid stabilization of moisture at the initial stage of cut tobacco drying is achieved, and manual intervention is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of tobacco pneumatic cut tobacco dryers, and particularly relates to an intelligent control method for a tobacco pneumatic cut tobacco dryer. Background Art

[0002] A tobacco pneumatic cut tobacco dryer is a device for expanding and drying cut tobacco. During the drying process of cut tobacco, the heated high-temperature process gas enters the drying pipe through the air inlet pipe and quickly blows the incoming cut tobacco at the bottom of the drying pipe into the drying pipe, where the cut tobacco is quickly expanded and dried. Subsequently, the dried cut tobacco and the process gas are separated in a cyclone separator. The process gas is recovered through the return air pipe, and the cut tobacco is discharged from the discharging device installed at the bottom of the separator. A water adding device is installed at a position on the air inlet pipe close to the feed inlet. There are strict requirements for the discharged moisture of the cut tobacco during the drying process. Currently, a method of controlling the discharged moisture by adjusting the water addition amount through a PID controller is widely used. Due to the time-delay characteristic of the drying process of cut tobacco, and the great influence of interference factors such as the incoming amount of cut tobacco, the incoming moisture of cut tobacco, the temperature of the process gas, and the flow rate of the process gas on the discharged moisture, the moisture control effect of the existing PID control method is not ideal. Especially in the initial stage of drying cut tobacco, a long stabilization time is required, and usually manual intervention is needed, resulting in high manual labor intensity and low control accuracy. Summary of the Invention

[0003] To solve the above problems, the present invention discloses an intelligent control method for a tobacco pneumatic cut tobacco dryer, which can improve the moisture control effect, achieve the rapid stabilization of the moisture in the initial stage of drying cut tobacco, and avoid manual intervention.

[0004] The specific scheme is as follows:

[0005] An intelligent control method for a tobacco pneumatic cut tobacco dryer includes constructing an environment model and constructing a strategy algorithm, specifically:

[0006] S1. Simulate the drying process of the tobacco pneumatic cut tobacco dryer using the environment model, complete virtual modeling through historical data-driven neural network, and predict the future outlet moisture;

[0007] S2. According to the prediction of the future outlet moisture by the environment model, find the optimal action (water addition amount) through the strategy algorithm, and make corresponding decisions to achieve the control purpose.

[0008] Further, when training the environment model, first let the environment model generate a predicted trajectory. After obtaining the predicted trajectory, update the environment model according to the error between the predicted trajectory and the true trajectory. Among them, the error between the predicted trajectory and the true trajectory is calculated with the help of a discriminant network, and until the discriminant network can no longer distinguish whether the predicted trajectory generated by the environment model is real or generated, it indicates that the environment model has realized the simulation of the tobacco air-flow drying process of the cigarette cut filler dryer; then test the environment model, and only passing the test can fully indicate the correctness of the constructed environment model; after the environment model passes the test, build the strategy algorithm based on it.

[0009] Further, the environment model is a transfer model, with the moisture content of the cut filler at the outlet as the observation value obs, the water addition amount as the action amount act, and the negative pressure at the inlet, the inlet moisture content, the process gas temperature, and the process gas flow rate as the external variables factor, to predict the moisture content of the cut filler at the next moment; there is a certain time delay from the inlet to the outlet of the cut filler drying, and this time period is denoted as T; the moisture content of the cut filler at the outlet of the tobacco air-flow drying machine at time t is determined by all the action amounts and external variables from time t - T to time t.

[0010] Further, the generation steps of the predicted trajectory are as follows:

[0011] S11. Select the moisture content of the cut filler at a certain moment t from the historical data as the initial observation value obs t , and at the same time obtain the action amounts act t-T:t and external variables factor t-T:t at this moment and the previous T - 1 moments, and input these quantities into the environment model to obtain the predicted result of the moisture content of the cut filler at the next moment

[0012] S12. Use this predicted result as the observed value of the moisture content of the cut filler at the next moment, and at the same time obtain the action amounts and external variables that can be correctly matched and stacked with the next moment from the historical data, and input these quantities into the environment model to obtain the predicted result;

[0013] S13. Continuously repeat step S12 until the preset trajectory length is reached to obtain the predicted trajectory;

[0014] The formula expression of the above process is as follows:

[0015]

[0016] Among them, obs t represents the observed value at time t, act t-T:t represents the action amounts from time t - T to time t, factor t-T:t represents the external variables from time t - T to time t, Represents the predicted value of the observed quantity at time t, and L is the set prediction trajectory length.

[0017] Furthermore, the test includes:

[0018] (1) Deduction test, generating a predicted trajectory close to the real trajectory;

[0019] (2) Response curve test, when each variable changes, the predicted observed value also changes correctly;

[0020] (3) Data distribution test, when the environmental model is accurate enough to simulate the process of drying tobacco strands with hot air, the data distribution of the output generated by the model based on historical data is similar to the data distribution of the real output.

[0021] Furthermore, the strategy algorithm adopts the MPC control algorithm, and its specific implementation method is as follows:

[0022] S21. When action decision needs to be made at time t, first randomly sample N water addition action plans from the given range by the method of uniform sampling

[0023] S22. Obtain the observed value obs of the moisture content at the outlet of the tobacco strands at the current moment t the action quantity act from time t - T to time t - 1 t-T:t-1 the external variable factor from time t - T to time t t-T:t ;

[0024] S23. For each of the N action plans, perform deduction by combining the obtained data. During the deduction process, the action quantity and external variable at time t and after remain unchanged, that is

[0025] When the deduction length reaches the pre-set length L MPC at this time, the deduction of the current action plan is completed, and a predicted trajectory is obtained; when the deduction of all plans is completed, N predicted trajectories are obtained;

[0026] S24. Input the N predicted trajectories and the target value into the defined optimization function, that is, select the optimal action plan

[0027] The formula expression of the above process is as follows:

[0028]

[0029] where obs t represents the observed quantity at time t, Denote the action quantity of the n-th action quantity scheme from time t-T to time t, factor t-T:t Denote the external variables from time t-T to time t, Denote the predicted value of the observable quantity deduced at time t for the n-th action quantity scheme, L MPC Is the set deduced trajectory length, traj n , n = 1, 2, …, N denote the n-th prediction and deduction trajectory, target t Denote the target value of the moisture content at the outlet of the cut tobacco at time t, opt_func denotes a predefined optimization function, Denote the optimal action scheme selected by the MPC strategy at time t.

[0030] The beneficial effects of the present invention are as follows:

[0031] By using the control scheme based on the environmental model, the system can accurately predict the change of the moisture content at the outlet of the cut tobacco in a future period of time, and dynamically adjust the current control strategy according to the prediction result, so as to achieve accurate and stable moisture control. This method uses a neural network model to model the cut tobacco drying process, fully excavates the laws in historical data, predicts the moisture change trend, and accurately adjusts the water addition amount accordingly. Compared with the traditional PID control method, the control scheme based on the environmental model can effectively cope with the disturbances caused by factors such as the feeding amount, feeding moisture, process gas temperature, and process gas flow rate, and significantly improves the anti-interference ability of the system. Although there is a certain time delay in the cut tobacco drying process, by predicting the moisture change in advance, the system can flexibly adjust the control actions to ensure that the moisture always remains within the ideal range, thus ensuring the stability of the cut tobacco drying process and the product quality. Due to the high prediction accuracy, this method almost eliminates the need for manual intervention, greatly improving the automation level and control accuracy of the system. Overall, the control method based on the environmental model not only optimizes the moisture control effect, but also enhances the adaptability of the system to external disturbances, and has strong practicality and broad application prospects. Specific implementation manners

[0032] The following further clarifies the present invention in combination with specific implementation manners. It should be understood that the following specific implementation manners are only used to illustrate the present invention and not to limit the scope of the present invention.

[0033] An intelligent control method for a tobacco pneumatic cut tobacco dryer of the present invention includes two parts: constructing an environmental model and constructing a strategy algorithm. The environmental model is a simulation of the cut tobacco drying process of the tobacco pneumatic cut tobacco dryer, and virtual modeling is completed by driving a neural network with historical data; the strategy algorithm finds the optimal action (water addition amount) based on the prediction of the future outlet moisture by the environmental model, and makes corresponding decisions to achieve the control purpose.

[0034] Among them, the constructed environmental model is a transition model. Taking the moisture content of the cut tobacco at the outlet as the observed quantity obs, the water addition amount as the action quantity act, and the negative pressure at the feed inlet, the moisture content of the feed, the process gas temperature, and the process gas flow rate as external variables factor, it predicts the moisture content of the cut tobacco at the next moment. There is a certain time delay from the feed to the discharge of the cut tobacco drying, and this duration is denoted as T. The moisture content of the cut tobacco at the outlet of the cut tobacco dryer at time t is not determined only by the action quantity and external variables at the single moment of t - T, but is determined by the comprehensive action of all variables during this period of the entire drying process. Therefore, the action quantity and external variables input into the environmental model are stacked in the time dimension, that is, all the action quantities and external variables from time t - T to time t are used.

[0035] Because there is a certain time delay in the cut tobacco drying process, simply predicting the moisture content of the cut tobacco at the next moment and performing control based on this result may not achieve the best control effect. To improve the control effect, it is required that the environmental model has a certain deduction ability, that is, to deduce the moisture content of the cut tobacco at a future period of time. If the environmental model trained by using traditional supervised learning algorithms is used for deduction, once an error occurs, the error will continuously amplify during the deduction process, and performing control based on such an incorrect result will inevitably lead to a poor control result. Therefore, when training the environmental model of the present invention, first let the environmental model generate a prediction trajectory, and the generation steps of this prediction trajectory are as follows:

[0036] S11. Select the moisture content of the cut tobacco at a certain moment (for example, time t) from the historical data as the initial observed value obs t , and at the same time obtain the action quantity act at this moment and the previous T - 1 moments t-T:t and the external variable factor t-T:t , input these quantities into the environmental model, and obtain the prediction result of the moisture content of the cut tobacco at the next moment

[0037] S12. Take this prediction result as the observed value of the moisture content of the cut tobacco at the next moment. At the same time, obtain the action quantity and external variables that can be correctly matched and stacked with the next moment from the historical data, and input these quantities into the environmental model to obtain the prediction result;

[0038] S13. Continuously repeat step S12 until the preset trajectory length is reached to obtain the prediction trajectory.

[0039] The formula expression of the above process is as follows:

[0040]

[0041] Among them, obs t represents the observed quantity at time t, act t-T:t represents the action quantity from time t - T to time t, factort-T:t Denote the external variables from time t-T to time t. Denote the predicted value of the observed quantity at time t, and L is the set prediction trajectory length.

[0042] After obtaining the prediction trajectory, update the environmental model according to the error between the prediction trajectory and the real trajectory. Although the environmental model trained in this way may not perform better than the model trained by the supervised learning algorithm in single-step prediction, the error during the deduction process will be much smaller than that of the model trained by the supervised learning algorithm. The error between the prediction trajectory and the real trajectory is calculated with the help of a discriminant network. The main responsibility of the discriminant network is to distinguish the two trajectories, that is, correctly identify whether the input trajectory is real or generated by the environmental model; the main responsibility of the environmental model is to achieve accurate prediction as much as possible, that is, generate a prediction trajectory as close as possible to the real trajectory. During the training process, the two compete with each other and continuously play games until reaching the Nash equilibrium. At this time, the discriminant network cannot distinguish whether the prediction trajectory generated by the environmental model is real or generated, indicating that the environmental model has approximately realized the simulation of the tobacco air-flow drying process. After the training is completed, the environmental model will also be tested. Only by passing the test can it fully demonstrate the correctness of the constructed environmental model. First, the deduction test, the generated prediction trajectory is as close as possible to the real trajectory; second, the response curve test, when each variable changes, the predicted observed value should also change correctly. Third, the data distribution test, when the environmental model is accurate enough to realize the simulation of the tobacco air-flow drying process, the data distribution of the output generated by the model based on historical data should be approximately the same as the real output data distribution.

[0043] When the environmental model passes all the above tests, a strategy algorithm can be constructed based on this environmental model. The strategy algorithm adopts the MPC control algorithm, and the specific implementation method is as follows:

[0044] S21. When action decision needs to be made at time t, first randomly sample N water addition action plans from the given range by the method of uniform sampling

[0045] S22. Obtain the observed value obs of the moisture content at the cigarette silk outlet at the current moment t and the action quantity act from time t-T to time t-1 t-T:t-1 and the external variable factor from time t-T to time t t-T:t ;

[0046] S23. For each of the N action plans, conduct deduction by combining the obtained data. During the deduction process, the action quantity and external variable at time t and after remain unchanged, that is

[0047] When the deduction length reaches the pre-set length L MPC When the current action plan The deduction is completed and a prediction trajectory is obtained. When all the schemes are deduced, N prediction trajectories are obtained;

[0048] S24: Input N predicted trajectories and target values ​​into the defined optimization function to select the optimal action plan

[0049] The formula for the above process is as follows:

[0050]

[0051] where obs t represents the observed value at time t, It represents the action amount of the nth action amount scheme from time tT to time t, factor t-T:t Represents the external variables from time tT to time t, represents the predicted value of the observation quantity derived from the nth action quantity scheme at time t, L MPC is the set deduction trajectory length, traj n ,n=1,2,…,N represents the nth prediction trajectory, target t represents the target value of the moisture content of the tobacco outlet at time t, opt_func represents the predefined optimization function, Represents the optimal action plan selected by the MPC strategy at time t.

[0052] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above-mentioned technical features. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also regarded as the protection scope of the present invention.

Claims

1. An intelligent control method for a tobacco airflow drying machine, characterized in that: The following steps are involved: S1. Use the environmental model to simulate the tobacco airflow drying process and predict the future outlet moisture; S2. Based on the prediction of future export moisture by the environmental model, the optimal amount of water added is found through the strategy algorithm, and corresponding decisions are made for control.

2. The intelligent control method for tobacco airflow drying machine according to claim 1, characterized in that: When training the environmental model, first let the environmental model generate a predicted trajectory. After obtaining the predicted trajectory, the environmental model is updated according to the error between the predicted trajectory and the true trajectory. The error between the predicted trajectory and the true trajectory is calculated with the help of a discriminant network, and until the discriminant network cannot distinguish whether the predicted trajectory generated by the environmental model is real or generated, it indicates that the environmental model has realized the simulation of the tobacco airflow drying machine's drying process; then the environmental model is tested. Only by passing the test can the correctness of the constructed environmental model be fully demonstrated; after the environmental model passes the test, a strategy algorithm is constructed based on it.

3. The intelligent control method for tobacco airflow drying machine according to claim 2, characterized in that: The environmental model is a transfer model, with the outlet moisture of tobacco as the observation obs, the amount of water added as the action act, the negative pressure of the feed inlet, the feed moisture, the process gas temperature, and the process gas flow rate as the external variable factor, and predicts the outlet moisture of tobacco at the next moment; there is a certain time delay in tobacco drying from feeding to discharging, and this time is recorded as T; the outlet moisture of tobacco of the tobacco airflow dryer at time t is determined by all the action quantities and external variables from time tT to time t.

4. The intelligent control method for a tobacco airflow drying machine according to claim 2, characterized in that: The steps for generating the predicted trajectory are: S11. Select the outlet moisture of the cut tobacco at a certain time t from the historical data as the initial observation value obs t , and obtain the action amount act at this moment and the previous T-1 moment t-T:t and external variable factor t-T:t , input these quantities into the environmental model to obtain the prediction result of the moisture content of the tobacco outlet at the next moment S12, using the prediction result as the observed value of the moisture content of the cut tobacco outlet at the next moment, and obtaining the action quantity and external variables that can be correctly matched and stacked at the next moment from the historical data, and inputting these quantities into the environmental model to obtain the prediction result; S13, continuously repeating step S12 until the preset trajectory length is reached to obtain the predicted trajectory; The formula for the above process is as follows: where obs t represents the observed value at time t, act t-T:t Indicates the amount of action from time tT to time t, factor t-T:t Represents the external variables from time tT to time t, It represents the predicted value of the observation at time t, and L is the set predicted trajectory length.

5. The intelligent control method for a tobacco airflow drying machine according to claim 2, characterized in that: The tests include: (1) Inference test: the generated predicted trajectory is close to the actual trajectory; (2) Response curve test, when each variable changes, the predicted observed value also changes correctly; (3) Data distribution test: When the environmental model is accurate enough to simulate the tobacco airflow drying process, the data distribution of the output generated by the model based on historical data is similar to the data distribution of the actual output.

6. The intelligent control method for a tobacco airflow drying machine according to claim 2, characterized in that: The strategy algorithm adopts the MPC control algorithm, and its specific implementation is as follows: S21. When an action decision needs to be made at time t, N water addition action plans are randomly sampled from a given range by uniform sampling. S22. Obtain the observed value obs of the moisture content of the tobacco outlet at the current moment t 、The amount of action act from time tT to time t-1 t-T:t-1 、External variable factor from time tT to time t t-T:t ; S23. For each of the N action plans, the acquired data is combined for deduction. During the deduction process, the action amount and external variables at and after time t remain unchanged, that is, factor t+i =factor t ,i=1,2,…,L MPC -1; when the deduction length reaches the pre-set length L MPC When the current action plan When the deduction of all plans is completed, a prediction trajectory is obtained; when the deduction of all plans is completed, N prediction trajectories are obtained; S24, input the N predicted trajectories and target values ​​into the defined optimization function, that is, select the optimal action plan The formula for the above process is as follows: where obs t represents the observed value at time t, It represents the action amount of the nth action amount scheme from time tT to time t, factor t-T:t Represents the external variables from time tT to time t, represents the predicted value of the observation quantity derived from the nth action quantity scheme at time t, L MPC is the set deduction trajectory length, traj n ,n=1,2,…,N represents the nth prediction trajectory, target t represents the target value of the moisture content of the tobacco outlet at time t, opt_func represents the predefined optimization function, It represents the optimal action plan selected by the MPC strategy at time t.

Images