A method for controlling the outlet moisture of a drum airflow drying machine

Abstract

This invention discloses a method for controlling the outlet moisture content of a rotary airflow drying machine, comprising the following steps: acquiring and processing historical production data of the rotary airflow drying machine; dividing the historical production data according to production stages and constructing a prediction model for each production stage; acquiring and processing current production data of the rotary airflow drying machine to determine the current production stage; based on the determination result, substituting the current production data into the prediction model for the corresponding production stage to obtain the prediction result after time t; combining the prediction result and the current production data into a search combination; comparing the search combination with the historical production data of the corresponding production stage to obtain a control parameter combination with a historical process environment similar to the current production stage; extracting relevant control parameters from the control parameter combination and performing reverse control after time t. The method of this application reduces drying time and deviation.

Description

Technical Field

[0001] This invention relates to the technical field of cigarette manufacturing, and more specifically, to a method for controlling the outlet moisture content of a drum airflow drying machine. Background Technology

[0002] The rotary drum airflow drying machine is an online processing device that combines rotary drum drying and airflow drying modes. It is used in cigarette manufacturing production lines to dry and expand cigarette shreds, ensuring that the moisture content and temperature of the processed shreds meet production specifications and cigarette manufacturing requirements. Current technology uses a combination of manual experience and PID control to regulate the fan frequency and outlet moisture content. Specifically, PID control is used when the outlet moisture content is relatively stable, but manual intervention is required when it fluctuates significantly. However, this approach has several problems in actual production: First, large fluctuations in outlet moisture content during the steady-state phase affect the quality of the shredded cigarettes. Second, the drying time at the beginning of the process is long, reducing production efficiency. Third, during process control, when large fluctuations in outlet moisture necessitate manual intervention, the original PID control cannot converge quickly, and excessive manual intervention leads to an over-reliance on the experience of skilled operators for product quality. Fourth, PID control is a single-loop feedback control system with a long time lag and does not consider environmental factors such as temperature, humidity, season, and steam, making it an incomplete control system. Summary of the Invention

[0003] To address at least one aspect of the aforementioned problems, this invention provides a method for controlling the outlet moisture content of a drum airflow drying machine. Based on deep learning or random forest algorithms, predictive models for different production stages are constructed. A parameter optimization algorithm is used to optimize the weight coefficients of the control parameters in the predictive models, which helps improve the accuracy of predictions. Furthermore, combined with a locality-sensitive hashing algorithm, the optimal control combination from historical production data similar to the current production situation can be quickly and accurately identified for reverse control. The method of this application reduces drying time and deviation.

[0004] This application is achieved through the following technical solution: a method for controlling the outlet moisture of a drum airflow drying machine, comprising the following steps:

[0005] Step 1: Obtain historical production data on the drum airflow drying machine and process the data;

[0006] Step 2: Divide the historical production data from Step 1 according to the production stage;

[0007] Step 3: Using the historical production data segmented in Step 2, construct a prediction model for each corresponding production stage;

[0008] Step 4: Obtain and process the current production data of the drum airflow drying machine to determine the current production stage;

[0009] Step 5: Based on the judgment result in Step 4, substitute the current production data in Step 4 into the prediction model for the corresponding production stage constructed in Step 3 to obtain the prediction result after time t.

[0010] Step 6: Combine the prediction results from Step 5 with the current production data to form the search result combination;

[0011] Step 7: Compare the combination to be retrieved in Step 6 with the historical production data of the corresponding production stage in Step 2 to obtain the control parameter combination of the historical process environment similar to the current production stage;

[0012] Step 8: Extract the relevant control parameters from the control parameter combination in Step 7, and perform reverse control after time t.

[0013] Preferably, the production stage in step 3 includes a steady-state stage, and the step of building a prediction model in the steady-state stage includes the following steps:

[0014] Step a: Using the historical production data of the steady-state stage as defined in Step 2, train a steady-state stage prediction model based on the random forest algorithm;

[0015] Step b: Using the historical production data of the steady-state stage divided in step 2 and the prediction data in the steady-state stage prediction model in step a, optimize the weight coefficients of the control parameters in the steady-state stage prediction model based on the parameter optimization algorithm;

[0016] Step c: Using the historical production data of the steady-state stage divided in step 2 and the optimized steady-state stage prediction model in step b, fit the model based on the simulation algorithm and determine whether the model is reliable. If the model is determined to be reliable, determine the model as the prediction model for the steady-state stage; if the model is determined to be unreliable, return to steps a and b.

[0017] Preferably, the production stage in step 3 further includes a material head stage and a material tail stage, and the step of building a prediction model in the material head stage or the material tail stage includes the following steps:

[0018] Step A: Using the historical production data of the material head stage or material tail stage as defined in Step 2, train a prediction model for the material head stage or the material tail stage based on a deep learning algorithm.

[0019] Step B: Using the historical production data of the material head stage or material tail stage divided in Step 2 and the prediction data in the material head stage prediction model or material tail stage prediction model in Step A, optimize the weight coefficients of the control parameters in the material head stage prediction model or material tail stage prediction model based on the parameter optimization algorithm.

[0020] Preferably, the production stage in step 3 further includes a preheating stage and a waiting stage. The step of building a prediction model in the preheating stage or the waiting stage includes using historical production data of the preheating stage or the waiting stage as defined in step 2, and training a prediction model for the preheating stage or the waiting stage based on a deep learning algorithm.

[0021] Preferably, the Locality Sensitive Hash algorithm is used for comparison in step 7.

[0022] Preferably, step 2 further includes establishing a high-dimensional optimal parameter combination space for each production stage based on the historical production data of each production stage after division; step 6 further includes converting the combination to be retrieved into a corresponding low-dimensional dense vector group; step 7 includes the following steps:

[0023] Step 7-1: Place the low-dimensional dense vector group from Step 6 into the optimal parameter combination space corresponding to Step 2 for retrieval;

[0024] Step 7-2: In the Euclidean distance space, retrieve the closest combinations of control parameters;

[0025] Step 7-3: Among the multiple different parameter combinations retrieved in Step 7-2, find the parameter combination that is closest to the set value in terms of current moisture content and has the smallest adjustment range, and determine it as the control parameter combination after time t.

[0026] Preferably, in step 2, when constructing the high-dimensional optimal parameter combination space of the material head stage, historical production data from the steady-state stage is used; in step 8, when performing reverse control in the material head stage, a step-type adjustment method is used to gradually adjust.

[0027] Preferably, in steps 2 and 4, an adaptive segmentation model is used to divide or determine the production stage of historical or current production data.

[0028] Preferably, both steps 1 and 4 employ a large time delay model for data processing to ensure that production data within the same group is aligned.

[0029] Preferably, the production data includes the PID output value of drying temperature, the detected value of drying temperature, the opening degree of the drying temperature regulating damper, the feedback value of the opening degree of the drying temperature regulating damper, the PID output value of final moisture content, the PID set value of steam flow rate, the PID measured value of steam flow rate, the PID output value of steam flow rate, the detected value of fresh air temperature, the opening degree of fresh air damper, the feedback value of fresh air damper opening, the opening degree of fresh air temperature regulating valve, the detected value of return air temperature, the instantaneous flow rate of material at the feed end, the cumulative amount of material at the feed end, the moisture content of material at the feed end, the temperature of material at the feed end, the instantaneous flow rate of material at the discharge end, the cumulative amount of material at the discharge end, the moisture content of material at the discharge end, the temperature of material at the discharge end, the outlet temperature of the combustion furnace, the process hot air volume, the exhaust air volume, the process air humidity, the process air temperature, the ambient humidity, the ambient temperature, the exhaust temperature, the exhaust moisture content, and the PID output value of the fan frequency.

[0030] The present invention provides a method for controlling the outlet moisture of a drum airflow drying machine, which has the following beneficial effects:

[0031] This application fully considers the large time-delay characteristics of the production data from the drum airflow drying machine. By employing a large time-delay model to align historical and current production data, real-time data can fully represent the equipment parameters during the processing of a particular batch of materials. An adaptive segmentation model is established based on workshop process indicators and process stage division standards to divide and determine production stages. Historical data corresponding to the production stage is used to build a prediction model based on a deep learning model or random forest model. Parameter optimization algorithms are combined to optimize the weight coefficients of the control parameters in the prediction model. Particularly in the steady-state stage, the reliability of the prediction model is assessed using a simulation model, which helps improve prediction accuracy. By employing a locality-sensitive hashing algorithm, a high-dimensional optimal parameter combination space is established using historical production data corresponding to the production stage. Placing the prediction results of the prediction model in this space allows for the rapid and accurate identification of the optimal control combination from historical production data similar to the current production situation for reverse control. The method of this application reduces drying time and deviation. Attached Figure Description

[0032] To better understand the above and other objects, features, advantages, and functions of the present invention, reference can be made to the embodiments shown in the accompanying drawings. The same reference numerals in the drawings refer to the same parts. Those skilled in the art should understand that the drawings are intended to schematically illustrate preferred embodiments of the invention and do not limit the scope of the invention in any way; the parts in the drawings are not drawn to scale.

[0033] Figure 1 A flowchart is shown below illustrating a method for controlling the outlet moisture content of a drum airflow drying machine according to an embodiment of the present invention;

[0034] Figure 2The diagram illustrates a predictive model classification for each production stage of a method for controlling outlet moisture in a drum airflow drying machine according to an embodiment of the present invention. Detailed Implementation

[0035] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0036] The term "comprising" and its variations as used herein signify open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0037] To at least partially address one or more of the aforementioned problems and other potential issues, embodiments of this disclosure propose a method for controlling the outlet moisture of a drum airflow drying machine, such as... Figure 1 and Figure 2 As shown, it includes the following steps:

[0038] Step 1: Obtain historical production data on the drum airflow drying machine and process the data;

[0039] Step 2: Divide the historical production data from Step 1 according to the production stage;

[0040] Step 3: Using the historical production data segmented in Step 2, construct a prediction model for each corresponding production stage;

[0041] Step 4: Obtain and process the current production data of the drum airflow drying machine to determine the current production stage;

[0042] Step 5: Based on the judgment result in Step 4, substitute the current production data in Step 4 into the prediction model for the corresponding production stage constructed in Step 3 to obtain the prediction result after time t.

[0043] Step 6: Combine the prediction results from Step 5 with the current production data to form the search result combination;

[0044] Step 7: Compare the combination to be retrieved in Step 6 with the historical production data of the corresponding production stage in Step 2 to obtain the control parameter combination of the historical process environment similar to the current production stage;

[0045] Step 8: Extract the relevant control parameters from the control parameter combination in Step 7, and perform reverse control after time t.

[0046] Specifically, step 1: Obtain historical production data of the drum airflow drying machine and process the data. Preferably, a large time delay model is used to process the historical production data to align the production data in the same group. The production data includes the PID output value of drying temperature, the detected value of drying temperature, the opening of the drying temperature regulating damper, the feedback value of the opening of the drying temperature regulating damper, the PID output value of final moisture, the PID set value of steam flow, the PID measured value of steam flow, the PID output value of steam flow, the detected value of fresh air temperature, the opening of the fresh air damper, the feedback value of the fresh air damper opening, the opening of the fresh air temperature regulating valve, the detected value of return air temperature, the instantaneous flow rate of material at the feed end, the cumulative amount of material at the feed end, the moisture content of material at the feed end, the temperature of material at the feed end, the instantaneous flow rate of material at the discharge end, the cumulative amount of material at the discharge end, the moisture content of material at the discharge end, the temperature of material at the discharge end, the outlet temperature of the combustion furnace, the process hot air volume, the exhaust air volume, the process air humidity, the process air temperature, the ambient humidity, the ambient temperature, the exhaust temperature, the exhaust moisture content, and the PID output value of the fan frequency.

[0047] Step 2: Divide the historical production data from Step 1 into production stages. Preferably, an adaptive segmentation model is established to divide the historical production data according to the workshop process indicators and process stage division standard documents. It also includes using the locality-sensitive hash algorithm to establish a high-dimensional optimal parameter combination space for each production stage based on the historical production data of each production stage after division. Preferably, when constructing the high-dimensional optimal parameter combination space of the material head stage, the historical production data of the steady-state stage is used.

[0048] Step 3: Using the historical production data divided in Step 2, construct a prediction model for each production stage. The prediction model uses the moisture content of the outlet material as the output and other production data as the input. Specifically, the production stages include the start-up self-inspection stage, the preheating stage, the material head stage, the steady-state stage, the material tail stage, and the waiting stage.

[0049] Step 3-1: Using the historical production data of the preheating stage as defined in Step 2, train a preheating stage prediction model based on a deep learning algorithm;

[0050] Step 3-2-A: Using the historical production data of the material head stage as defined in Step 2, train a prediction model for the material head stage based on a deep learning algorithm;

[0051] Step 3-2-B: Using the historical production data of the material head stage divided in Step 2 and the prediction data in the material head stage prediction model in Step 3-2-A, optimize the weight coefficients of the control parameters in the material head stage prediction model based on the parameter optimization algorithm;

[0052] Step 3-3-a: Using the historical production data of the steady-state stage divided in Step 2, train a steady-state stage prediction model based on the random forest algorithm; preferably, a top-down greedy regression tree is used when training the prediction model using the random forest algorithm.

[0053] Step 3-3-b: Using the historical production data of the steady-state stage divided in Step 2 and the prediction data in the steady-state stage prediction model in Step 3-3-a, optimize the weight coefficients of the control parameters in the steady-state stage prediction model based on the parameter optimization algorithm;

[0054] Step 3-3-c: Using the historical production data of the steady-state stage divided in Step 2 and the optimized steady-state stage prediction model in Step 3-3-b, fit the model based on the simulation algorithm and determine whether the model is reliable. If the model is determined to be reliable, determine that the model is the prediction model for the steady-state stage; if the model is determined to be unreliable, return to Step 3-3-a and Step 3-3-b.

[0055] Step 3-4-A: Using the historical production data of the tail stage as defined in Step 2, train a prediction model for the tail stage based on a deep learning algorithm.

[0056] Step 3-4-B: Using the historical production data of the tail stage divided in Step 2 and the prediction data in the tail stage prediction model in Step 3-2-A, optimize the weight coefficients of the control parameters in the tail stage prediction model based on the parameter optimization algorithm.

[0057] Steps 3-5: Using the historical production data of the waiting-for-materials stage as defined in Step 2, train a prediction model for the waiting-for-materials stage based on a deep learning algorithm.

[0058] Step 4: Obtain and process the current production data of the drum airflow drying machine. Preferably, a large time delay model is used to process the current production data to align the production data in the same group. The processed current production data is extracted or derived, and an adaptive segmentation model is used to determine the current production stage based on the current production data.

[0059] Step 5: Based on the judgment result in Step 4, substitute the current production data from Step 4 into the prediction model for the corresponding production stage constructed in Step 3 to obtain the prediction result after time t; specifically,

[0060] Step 5-1: When the current production data indicates that the machine is in the self-test stage, no parameters are recommended; use the start-up parameters of the drum airflow drying machine.

[0061] Step 5-2: When the current production data indicates that it is in the preheating stage, substitute the current production data into the preheating stage prediction model in step 3-1 to obtain the prediction result after time t.

[0062] Step 5-3: When the current production data indicates that it is in the material head stage, substitute the current production data into the optimized material head stage prediction model in step 3-2-B to obtain the prediction result after time t.

[0063] Step 5-4: When the current production data indicates that it is in a steady state, substitute the current production data into the steady state prediction model in step 3-3-c to obtain the prediction result after time t.

[0064] Step 5-5: When the current production data indicates that it is in the tail stage, substitute the current production data into the optimized tail stage prediction model in step 3-4-B to obtain the prediction result after time t.

[0065] Steps 5-6: When the current production data indicates that the production is in the waiting stage, substitute the current production data into the waiting stage prediction model in Step 3-5 to obtain the prediction result after time t.

[0066] Step 6: Combine the prediction results from Step 5 with the current production data to form a searchable combination, and convert it into a corresponding low-dimensional dense vector group.

[0067] Step 7: Compare the combination to be retrieved in Step 6 with the historical production data of the corresponding production stage in Step 2 to obtain the control parameter combination of the historical process environment similar to the current production stage; specifically,

[0068] Step 7-1: Place the low-dimensional dense vector group from Step 6 into the optimal parameter combination space corresponding to Step 2 for retrieval;

[0069] Step 7-2: In the Euclidean distance space, retrieve the closest combinations of control parameters; wherein, a threshold for the Euclidean distance or a number of combinations of control parameters can be set to retrieve the closest combinations of control parameters.

[0070] Step 7-3: Among the multiple different parameter combinations retrieved in Step 7-2, find the parameter combination that is closest to the set value (i.e., the absolute value is the smallest and the adjustment range is the smallest) in terms of the current moisture content. This combination will be used as the control parameter combination after time t.

[0071] Step 8: Extract relevant control parameters from the control parameter combination in Step 7, such as the fan frequency PID output value, and perform reverse control after time t. Preferably, when performing reverse control in the material head stage, a step-type adjustment method is used to gradually adjust the parameters.

[0072] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand this document.

Claims

1. A method for controlling the outlet moisture content of a drum airflow drying machine, characterized in that: Includes the following steps: Step 1: Obtain historical production data on the drum airflow drying machine and process the data; Step 2: Divide the historical production data from Step 1 into production stages, and establish a high-dimensional optimal parameter combination space for each production stage based on the historical production data of each production stage after the division. Step 3: Using the historical production data segmented in Step 2, construct a prediction model for each corresponding production stage; Step 4: Obtain and process the current production data of the drum airflow drying machine to determine the current production stage; Step 5: Based on the judgment result in Step 4, substitute the current production data in Step 4 into the prediction model for the corresponding production stage constructed in Step 3 to obtain the prediction result after time t. Step 6: Combine the prediction results from Step 5 with the current production data to form a searchable combination, and transform the searchable combination into a corresponding low-dimensional dense vector group; Step 7: Compare the combination to be retrieved in Step 6 with the historical production data of the corresponding production stage in Step 2 using the Locality Sensitive Hashing algorithm to obtain the control parameter combination of the historical process environment similar to the current production stage; Step 7 includes the following steps: Step 7-1: Place the low-dimensional dense vector group from Step 6 into the optimal parameter combination space corresponding to Step 2 for retrieval; Step 7-2: In the Euclidean distance space, retrieve the closest combinations of control parameters; Step 7-3: Among the multiple different parameter combinations retrieved in Step 7-2, find the parameter combination that is closest to the set value in terms of current moisture content and has the smallest adjustment range, and determine it as the control parameter combination after time t. Step 8: Extract the relevant control parameters from the control parameter combination in Step 7, and perform reverse control after time t.

2. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 1, characterized in that: The production stage in step 3 includes a steady-state stage, and the steps for building a prediction model in the steady-state stage include the following: Step a: Using the historical production data of the steady-state stage as defined in Step 2, train a steady-state stage prediction model based on the random forest algorithm; Step b: Using the historical production data of the steady-state stage divided in step 2 and the prediction data in the steady-state stage prediction model in step a, optimize the weight coefficients of the control parameters in the steady-state stage prediction model based on the parameter optimization algorithm; Step c: Using the historical production data of the steady-state stage divided in step 2 and the optimized steady-state stage prediction model in step b, fit the model based on the simulation algorithm and determine whether the model is reliable. If the model is determined to be reliable, determine the model as the prediction model for the steady-state stage; if the model is determined to be unreliable, return to steps a and b.

3. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 2, characterized in that: The production stage in step 3 also includes a material head stage and a material tail stage. The steps for building a prediction model in the material head stage or the material tail stage include the following steps: Step A: Using the historical production data of the material head stage or material tail stage as defined in Step 2, train a prediction model for the material head stage or the material tail stage based on a deep learning algorithm. Step B: Using the historical production data of the material head stage or material tail stage divided in Step 2 and the prediction data in the material head stage prediction model or material tail stage prediction model in Step A, optimize the weight coefficients of the control parameters in the material head stage prediction model or material tail stage prediction model based on the parameter optimization algorithm.

4. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 3, characterized in that: The production stage in step 3 also includes a preheating stage and a waiting stage. The steps for building a prediction model in the preheating stage or the waiting stage include using historical production data from the preheating stage or the waiting stage as defined in step 2, and training a prediction model for the preheating stage or the waiting stage based on a deep learning algorithm.

5. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 3, characterized in that: In step 2, when constructing the high-dimensional optimal parameter combination space of the material head stage, historical production data from the steady-state stage is used; in step 8, when performing reverse control in the material head stage, a step-type adjustment method is used to gradually adjust.

6. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 1, characterized in that: In both steps 2 and 4, an adaptive segmentation model is used to divide or determine the production stage of historical or current production data.

7. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 1, characterized in that: Both steps 1 and 4 employ a large time delay model for data processing to ensure that production data within the same group is aligned.

8. The method for controlling the outlet moisture of a drum airflow drying machine according to claim 1, characterized in that: The production data includes the PID output value of drying temperature, the detected value of drying temperature, the opening degree of the drying temperature regulating damper, the feedback value of the opening degree of the drying temperature regulating damper, the PID output value of final moisture content, the PID set value of steam flow, the PID measured value of steam flow, the PID output value of steam flow, the detected value of fresh air temperature, the opening degree of fresh air damper, the feedback value of fresh air damper opening, the opening degree of fresh air temperature regulating valve, the detected value of return air temperature, the instantaneous flow rate of material at the feed end, the cumulative amount of material at the feed end, the moisture content of material at the feed end, the temperature of material at the feed end, the instantaneous flow rate of material at the discharge end, the cumulative amount of material at the discharge end, the moisture content of material at the discharge end, the temperature of material at the discharge end, the outlet temperature of the combustion furnace, the process hot air volume, the exhaust air volume, the process air humidity, the process air temperature, the ambient humidity, the ambient temperature, the exhaust temperature, the exhaust moisture content, and the PID output value of the fan frequency.

Images