Threshing and redrying intelligent control method and system based on machine learning and medium
By using machine learning-based intelligent control methods, a positive data prediction model and a multivariate MPC controller were constructed, which solved the limitations and lack of foresight of single-variable control in the leaf re-drying process, achieving efficient and coordinated control effects and improving the stability and consistency of the production process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-03-31
AI Technical Summary
Existing PID control systems have limitations in single-variable control and lack foresight and optimization capabilities during the leaf re-drying process, resulting in difficulty in improving control quality, poor system adaptability, reliance on operator experience for frequent manual intervention, large fluctuations in control results, and difficulty in guaranteeing stability and consistency.
A machine learning-based intelligent control method is adopted. By constructing a forward data prediction model and a multivariable MPC controller, the output data of the controlled object is predicted and the optimal operation data is found in reverse. Combined with optimization algorithms and inverse objective functions, the prediction of operation data sequences and the generation of control commands for future time periods are realized.
It achieves dynamic adaptive capability, solves the oscillation problem of traditional PID control algorithm in multivariable processing, improves the efficiency and coordination of control, ensures that the production process always runs towards the optimal goal, and improves control quality and stability.
Smart Images
Figure CN121753958A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tobacco processing technology, specifically to a machine learning-based intelligent control method, system, and medium for leaf threshing and re-drying. Background Technology
[0002] Leaf threshing and re-drying is a core process in tobacco processing. It involves separating tobacco leaves from the stems (threshing) and then subjecting both leaves and stems to drying, cooling, and rehydration processes (re-drying) to obtain standardized raw materials suitable for cigarette production, long-term storage, and natural aging. Precise control at each stage has a decisive impact on the quality of tobacco processing, material loss, and the sensory quality of the product. Currently, domestic re-drying enterprises generally use control systems based on the classic PID (Proportional-Integral-Derivative) algorithm. This control system mainly faces the following technical bottlenecks:
[0003] 1. Limitations of Single-Variable Control: Traditional PID controllers perform independent closed-loop regulation on a single process variable (such as steam flow rate), which cannot effectively handle the strong coupling, nonlinearity, and large time delay characteristics between multiple variables (such as steam flow rate, water flow rate, and cylinder rotation speed of the leaf humidifier) and multiple controlled targets (such as outlet moisture content and outlet temperature of the leaf humidifier) in each stage of the process. Adjusting one variable will cause fluctuations in other controlled targets, resulting in the system being in a long-term cycle of "oscillation-stability-re-oscillation," making it difficult to further improve control quality.
[0004] 2. Lack of foresight and optimization capabilities: PID is a feedback control system that only takes action after a deviation occurs. It cannot make forward-looking predictions and optimization decisions based on incoming material conditions (such as inlet moisture, tobacco leaf flow rate, and grade changes) and the needs of downstream processes. The system has poor adaptability to dynamic changes in production rhythm and raw material characteristics, relying on frequent manual intervention based on operator experience, resulting in large fluctuations in control results and difficulty in guaranteeing stability and consistency. Summary of the Invention
[0005] The purpose of this invention is to provide a machine learning-based intelligent control method, system, and medium for leaf re-baking, which solves the problems of high limitations and lack of foresight and optimization capabilities in existing control methods.
[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:
[0007] In a first aspect, the present invention provides a machine learning-based intelligent control method for leaf removal and re-baking, the method comprising:
[0008] At the beginning of the preset prediction period, the current disturbance data and operation data of the controlled object are acquired. The current disturbance data and operation data are input into the pre-built positive data prediction model to predict the output data of the controlled object, and the predicted value of the output data of the controlled object is obtained.
[0009] Construct an inverse objective function for the operational data based on the predicted output data of the controlled object and the preset target output data value;
[0010] The optimal value of the operational data is obtained by solving the inverse objective function based on the optimization algorithm.
[0011] The optimal value of the operating data, the target value of the output data, and the current disturbance data are input into the multivariable MPC controller to predict the operating data for future periods, thus obtaining the operating data sequence for future periods.
[0012] Based on the operational data sequence for future time periods, control instructions for the controlled object are generated.
[0013] Preferably, the controlled object is a leaf humidifier, leaf threshing machine, or leaf drying machine. When the controlled object is a leaf humidifier, the interference data includes: the inlet moisture content, inlet temperature, ambient temperature, and ambient humidity of the leaf humidifier. The operation data includes: the steam flow rate, water flow rate, and rotational speed of the leaf humidifier drum. The output data includes: the outlet moisture content and outlet temperature of the leaf humidifier.
[0014] Preferably, the method further includes: constructing a positive data prediction model, including:
[0015] Acquire historical operation data, historical output data, and corresponding historical interference data of the controlled object;
[0016] The historical operation data and historical interference data are preprocessed and feature constructed to obtain a feature set, which includes: univariate features, lag features, sliding window features, difference features and interaction features.
[0017] A training set is constructed based on the feature set and historical output data. The pre-set machine learning model is trained using the training set to obtain the trained model, which is then used as a positive data prediction model.
[0018] Preferably, the machine learning model is a GradientDescent model, a RandomForest model, or an XGBoost model.
[0019] Preferably, constructing an inverse objective function for the operation data based on the predicted output data value of the controlled object and the preset target output data value includes:
[0020] The first difference function is determined based on the predicted value and target value of the output data of the controlled object.
[0021] The second difference function is determined based on the operational data and the corresponding nominal values.
[0022] By weighting the first and second difference functions, the inverse objective function is obtained.
[0023] Preferably, the optimization algorithm is a differential evolution algorithm, which solves the inverse objective function based on the optimization algorithm to obtain the optimal value of the operational data, including:
[0024] Initialize the control parameters of the differential evolution algorithm, including: initial population size, scaling factor, maximum number of iterations, and crossover probability;
[0025] An initial population of the initial size is randomly generated based on the constraints of the inverse objective function, and each individual in the initial population is a solution of the inverse objective function; wherein, the constraints of the inverse objective function include: upper and lower limits of the operating data and a limit on the rate of change.
[0026] Using the initial population as the current population, calculate the fitness of each individual in the current population, and select the best and worst individuals from the current population based on their fitness.
[0027] Determine whether the preset termination condition is met. If so, take the solution of the inverse objective function corresponding to the best individual in the current population as the optimal value of the operation data.
[0028] If not, based on the elite mutation strategy and the reverse crossover strategy, perform mutation and crossover operations on each individual in the current population in turn to obtain a new population. Use the new population to update the current population, recalculate the fitness of each individual in the current population, and select the best and worst individuals until the preset termination condition is met.
[0029] Preferably, the elite mutation strategy is:
[0030] Obtain the optimal individual from each iteration and the individual corresponding to the minimum value of the inverse objective function in the current iteration;
[0031] Based on the best individual from each iteration, the individual corresponding to the minimum value of the inverse objective function in the current iteration, and the scaling factor, perform mutation on each individual in the current population to obtain the mutated individuals.
[0032] Preferably, the reverse crossover strategy is as follows:
[0033] Perform a crossover operation on the mutated individuals in the current population to obtain intermediate individuals;
[0034] Obtain the best and worst individuals from each iteration;
[0035] The intermediate individuals are back-fused with the best and worst individuals from each iteration to obtain the crossover individuals.
[0036] Secondly, the present invention provides a machine learning-based intelligent control system for leaf removal and re-drying, used to implement the above-mentioned machine learning-based intelligent control method for leaf removal and re-drying, the system comprising:
[0037] The first prediction module is used to acquire the current disturbance data and operation data of the controlled object at the beginning of the preset prediction period, input the current disturbance data and operation data into the pre-built positive data prediction model to predict the output data of the controlled object, and obtain the predicted value of the output data of the controlled object.
[0038] The function construction module is used to construct the inverse objective function of the operation data based on the predicted value of the output data of the controlled object and the preset target value of the output data.
[0039] The function solving module is used to solve the inverse objective function based on optimization algorithms to obtain the optimal value of the operational data;
[0040] The second prediction module is used to input the optimal value of the operation data, the target value of the output data, and the current interference data and operation data into the multivariable MPC controller to predict the operation data for future periods, and obtain the operation data sequence for future periods.
[0041] The instruction generation module is used to generate control instructions for the controlled object based on the operation data sequence for future time periods.
[0042] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described intelligent control method for leaf picking and re-baking based on machine learning.
[0043] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described intelligent control method for leaf re-baking based on machine learning.
[0044] The beneficial effects of this invention are mainly reflected in:
[0045] 1. This invention constructs a forward data prediction model. At the beginning of each preset prediction cycle, the current disturbance data and operation data are input into the forward data prediction model to predict the output data prediction value of the controlled object. Then, the optimal operation data, i.e. the best value of the operation data, is found by using the inverse objective function. Therefore, in each cycle, this method will re-execute the inverse optimization based on the latest operating data, so that the control method has a dynamic adaptive capability of "adapting to circumstances" and always guides the production process toward the optimal goal.
[0046] 2. After obtaining the optimal operating data, this invention uses a multivariable MPC controller to predict the trajectory of changes in variables such as operating data over a future period (prediction time domain). This solves the problem of oscillation in traditional PID control algorithms when dealing with multiple variables, and makes the control action efficient and coordinated.
[0047] 3. This invention utilizes the seamless integration of a positive data prediction model and a multivariate MPC controller. The positive data prediction model is responsible for intelligent decision-making of set values (optimal values of operating data), while the multivariate MPC controller is responsible for accurately executing the set values. Together, they constitute the core technological foundation for realizing intelligent and lean production of leaf picking and re-drying. Attached Figure Description
[0048] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:
[0049] Figure 1 This is a flowchart of a machine learning-based intelligent control method for leaf removal and re-baking provided in one embodiment of the present invention;
[0050] Figure 2 This is a block diagram of a machine learning-based intelligent control system for leaf re-drying provided in one embodiment of the present invention. Detailed Implementation
[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.
[0052] Example 1
[0053] Figure 1This is a flowchart of a machine learning-based intelligent control method for leaf re-baking provided in one embodiment of the present invention. Figure 1 As shown, this embodiment provides a machine learning-based intelligent control method for leaf re-drying. The application scenario of this method includes a control layer, a data acquisition layer, and an execution layer. The control layer deploys a controller and a server, which are communicatively connected. The data acquisition layer includes several sensors used to monitor the operating data of the secondary leaf re-drying process, including interference data and operational data. The sensors can be connected to the controller or the server, or the controller can upload the operating data to the server. The execution layer includes several actuators connected to the controller, which controls the actuators to complete corresponding control actions. Based on this scenario, the control method of this embodiment includes steps S10 to S50.
[0054] Step S10: At the initial moment of the preset prediction period, acquire the current disturbance data and operation data of the controlled object, input the current disturbance data and operation data into the pre-built positive data prediction model to predict the output data of the controlled object, and obtain the predicted value of the output data of the controlled object.
[0055] In this embodiment, the preset prediction period can be 5 minutes, and step S10 is executed once every 5 minutes.
[0056] In this embodiment, the controlled object is a leaf-moistening machine, a leaf-beating machine, or a leaf-baking machine, meaning that the control method of this embodiment is used in the leaf-moistening stage, the leaf-beating stage, and the leaf-baking stage.
[0057] When the controlled object is a leaf-curing machine, the interference data includes, but is not limited to, the inlet moisture content, inlet temperature, ambient temperature, and ambient humidity of the leaf-curing process. The inlet moisture content can be collected using a moisture meter, the inlet and ambient temperatures can be collected using temperature sensors, and the ambient humidity can be collected using a humidity sensor. Operational data includes, but is not limited to, steam flow rate, water flow rate, and the rotational speed of the leaf-curing cylinder. In this embodiment, the actuator includes a steam valve, a water inlet valve, and a servo motor, all of which are connected to the controller. The output data of the controlled object includes, but is not limited to, the outlet moisture content and outlet temperature of the leaf-curing process. The outlet moisture content is collected using a moisture meter, and the outlet temperature is collected using a temperature sensor.
[0058] When the controlled object is a leaf-beating machine, the interference data includes, but is not limited to, the inlet moisture content, inlet temperature, ambient temperature, ambient humidity, material flow rate, and incoming material grade and type for the leaf-beating process. Operational data includes, but is not limited to, the speed of the beating roller, the equipment feeding speed, the air volume of the air distribution system, and the gap between the beating pin and the bottom blade. The actuators include: a variable frequency motor for the beating roller, a variable frequency motor for the feeder, a variable frequency motor for the air distribution fan or a damper regulating valve, and a servo motor or pneumatic adjustment mechanism (adjusting the gap between the beating pin and the bottom blade can change the intensity of leaf beating). The output data of the controlled object includes, but is not limited to, the moisture content of the discharged leaves, the percentage of large and medium-sized leaves, the percentage of stems in the leaves, and the discharge temperature for the leaf-beating process.
[0059] When the controlled object is a wafer baking machine, the interfering data includes, but is not limited to, the inlet moisture content, inlet temperature, ambient temperature, ambient humidity, material flow rate / layer thickness, incoming material grade and type, and steam pressure fluctuations during the wafer baking process. Operating data includes, but is not limited to, hot air temperature, exhaust air volume, conveyor belt / drum speed, and circulating air volume; the actuators include: heater steam regulating valve, exhaust fan variable frequency motor or exhaust damper, conveyor belt / drum drive variable frequency motor, and circulating fan variable frequency motor. The output data of the controlled object includes, but is not limited to, the outlet moisture content, outlet temperature, and blade color during the wafer baking process.
[0060] In this embodiment, the controlled object is a leaf-moistening machine. The initial time of the preset prediction cycle can be the last moment of the previous cycle or the first moment of the current cycle. Therefore, every 5 minutes, the predicted output data of the leaf-moistening machine in this cycle is predicted. The predicted output data of the leaf-moistening machine includes the predicted outlet moisture content and the predicted outlet temperature.
[0061] In this embodiment, the positive data prediction model is constructed using machine learning algorithms. It is trained using historical data from the leaf-moistening process as sample data to obtain the positive data prediction model. The positive data prediction model is then used to predict the output data prediction value of the controlled object.
[0062] As a further optimization of this embodiment, the method further includes: constructing a positive data prediction model, including:
[0063] Step a10: Obtain the historical operation data, historical output data, and corresponding historical interference data of the controlled object; in this embodiment, the historical operation data, historical output data, and corresponding historical interference data are the historical data, which can be data from the past year.
[0064] The historical operating data in this embodiment includes historical steam flow rate, historical water flow rate, and historical humidifier rotation speed. The historical output data includes historical outlet moisture content and historical outlet temperature. The historical interference data includes historical inlet moisture content, historical inlet temperature, historical ambient temperature, and historical ambient humidity.
[0065] Step a20: Preprocess and construct features for historical operation data and historical interference data to obtain a feature set, which includes: univariate features, lag features, sliding window features, difference features, and interaction features.
[0066] In this embodiment, the preprocessing of historical operation data and historical interference data includes: time alignment processing and data cleaning.
[0067] Time alignment is one aspect: Since different sensors have different sampling periods, network latency, and database recording times, directly merging them can lead to incorrect causal relationships. To address this, the timestamp of the DCS (Distributed Control System) is selected as the reference time axis. Low-frequency data is aligned to the reference time axis by interpolation (such as linear interpolation) or forward padding. At the same time, high-frequency data is aggregated according to a fixed time window (such as every 5 seconds) (taking the mean and median) to reduce data volume and noise.
[0068] Data cleaning involves removing missing or outlier values from the data to improve its quality.
[0069] The univariate features in this embodiment are constructed as follows: draw histograms or kernel density estimation plots of the historical data to obtain their distribution patterns, which include features such as normality, skewness, and bimodality; calculate the basic statistics of each historical data, which include features such as mean, standard deviation, and maximum / minimum values.
[0070] The lag feature in this embodiment is constructed as follows: steam flow rate and outlet moisture data are plotted on the same time axis. Lag time and dynamic relationship can be extracted from the time axis. For example, if it is found that outlet moisture lags behind steam flow rate by about 60 seconds, then "steam_flow_lag_60s" is created and "steam_flow_lag_60s" is used as the lag feature.
[0071] The sliding window feature in this embodiment is constructed by calculating the statistical characteristics of variables over a period of time, such as the average steam flow rate over the past 2 minutes and the standard deviation of inlet moisture over the past 5 minutes.
[0072] The differential feature in this embodiment is constructed by calculating the rate of change of the operational data and using the rate of change of the operational data as the differential feature.
[0073] The interactive feature in this embodiment is constructed as follows: parameters that have a combined effect on outlet temperature or outlet moisture are statistically analyzed. For example, the combined effect of steam and water may have a special influence on temperature. Therefore, "steam_flow * water_flow" is constructed and used as the interactive feature.
[0074] Step a30: Construct a training set based on the feature set and historical output data, train the preset machine learning model with the training set to obtain the trained model, and use the trained model as the positive data prediction model.
[0075] In this embodiment, the feature set needs to be divided into a training set, a validation set, and a test set. The training set consists of the first 70% of the data, the validation set consists of the middle 15% of the data, and the test set consists of the last 15% of the data. Historical output data is used as the training label to train, validate, and test the machine learning model in order to obtain a trained model.
[0076] In this embodiment, the machine learning model is: GradientDescent model, RandomForest model, or XGBoost model.
[0077] Step S20: Construct the inverse objective function of the operation data based on the predicted output data of the controlled object and the preset target output data value.
[0078] In this embodiment, an inverse objective function for the operation data is constructed based on the predicted output data value of the controlled object and the preset target output data value, including:
[0079] Step S201: Determine the first difference function based on the predicted output data and the target output data of the controlled object.
[0080] In this embodiment, the expression for the first difference function is:
[0081] ;
[0082] In the formula, For the first difference function, The target value for the outlet temperature. The predicted export temperature is , This is the target value for export moisture content. This is the predicted moisture content for export.
[0083] Step S202: Determine the second difference function based on the operation data and the nominal value corresponding to the operation data.
[0084] ;
[0085] In the formula, The second difference function, For steam flow rate, This is the nominal value of the steam flow rate. To increase water flow, This is the nominal value of the water flow rate. The rotational speed of the blade-wetting cylinder, This refers to the nominal value of the rotational speed of the blade-lubricating cylinder; among which, the nominal value corresponding to the operating data is the benchmark setting or reference value that can meet the process requirements under normal and stable operating conditions and has been verified through long-term practice.
[0086] Step S203: Weight the first difference function and the second difference function to obtain the inverse objective function.
[0087] In this embodiment, the expression for the inverse objective function is:
[0088] ;
[0089] In the formula, The inverse objective function is... As the first weighting coefficient, This is the second weighting coefficient.
[0090] Step S30: Solve the inverse objective function based on the optimization algorithm to obtain the optimal value of the operation data.
[0091] In this embodiment, the optimization algorithm can be a particle swarm optimization algorithm, a genetic algorithm, or a differential evolution algorithm. This embodiment preferably uses a differential evolution algorithm. The steps for solving the inverse objective function based on the optimization algorithm are as follows:
[0092] Step A1: Initialize the control parameters of the differential evolution algorithm, including the initial population size, scaling factor, maximum number of iterations, and crossover probability.
[0093] Step A2: Randomly generate an initial population of the initial size based on the constraints of the inverse objective function, wherein each individual in the initial population is a solution of the inverse objective function; wherein the constraints of the inverse objective function include: upper and lower limits of the operating data and a limit on the rate of change.
[0094] In this embodiment, the initial population is generated randomly within the constraints, which reduces the convergence speed of the solution. Therefore, this embodiment uses the following method to generate the initial population:
[0095] First, within the constraints of the inverse objective function, two populations are randomly generated to obtain the first population and the second population. Then, the fitness of each individual in the two populations is calculated, and the fitness of all individuals in each population is added together to obtain the total fitness. Finally, the total fitness of the two populations is compared, and the population with the smaller total fitness is selected as the initial population.
[0096] This embodiment, through the aforementioned generation strategy, effectively avoids stagnation in subsequent iterations, especially when prone to getting trapped in local optima. Furthermore, this method significantly improves the convergence speed of the solution process, enabling the algorithm to reach the ideal solution in a shorter time, thereby greatly improving overall computational efficiency and the timeliness of problem-solving.
[0097] Step A3: Using the initial population as the current population, calculate the fitness of each individual in the current population, and select the best and worst individuals from the current population based on their fitness.
[0098] Step A4: Determine whether the preset termination condition is met. If so, take the solution of the inverse objective function corresponding to the best individual in the current population as the optimal value of the operation data. In this embodiment, the preset termination condition can be the maximum number of iterations, that is, when the current number of iterations reaches the maximum number of iterations, it indicates that the preset termination condition is met.
[0099] Step A5: If not, perform mutation and crossover operations on each individual in the current population in sequence based on the elite mutation strategy and the reverse crossover strategy to obtain a new population.
[0100] Update the current population with the new population, and repeat steps A3-A5 until the preset termination condition is met.
[0101] As a further optimization of this embodiment, the elite mutation strategy is as follows:
[0102] Step A5-01: Obtain the optimal individual from each iteration and the individual corresponding to the minimum value of the inverse objective function in the current iteration;
[0103] Step A5-02: Perform mutation on each individual in the current population based on the best individual in previous iterations, the individual corresponding to the minimum value of the inverse objective function in the current iteration, and the scaling factor to obtain the mutated individuals.
[0104] In this embodiment, the functional expression of the mutated individual is:
[0105] ;
[0106] In the formula, For the mutated individual j, For individual j before the mutation, This refers to the individual corresponding to the minimum value of the inverse objective function in the current iteration. It is the optimal individual in each iteration.
[0107] The elite mutation strategy in this embodiment makes full use of the richer information resources within the population, avoids wasting some high-quality individuals, improves the global search capability of the algorithm, and avoids getting trapped in local optima. Furthermore, the elite mutation strategy can guide the population to evolve towards a better solution, thereby accelerating the convergence speed of the algorithm.
[0108] As a further optimization of this embodiment, the reverse crossover strategy is as follows:
[0109] Step A5-11: Perform a crossover operation on the mutated individuals in the current population to obtain intermediate individuals.
[0110] Step A5-12: Obtain the best and worst individuals from each iteration.
[0111] Step A5-13: Perform reverse fusion of the intermediate individuals with the best and worst individuals from each iteration to obtain the crossover individuals.
[0112] ;
[0113] In the formula, For the intermediate individual i before fusion, For the intermediate individual i after fusion, It is the worst individual in all iterations. The fusion coefficient has a value range of [0.5, 0.9].
[0114] This embodiment employs a reverse crossover strategy, where individuals that are opposed to the current individual effectively expand the algorithm's search range, explore more potential regions in the search space, and avoid getting trapped in local optima. Furthermore, this individual increases the diversity of solutions in the population, preventing premature convergence and thus improving the algorithm's optimization ability.
[0115] Therefore, this embodiment constructs a forward data prediction model. At the beginning of each preset prediction cycle, the current disturbance data and operation data are input into the forward data prediction model to predict the output data prediction value of the controlled object. Then, the inverse objective function is used to find the optimal operation data, i.e., the best value of the operation data. Therefore, in each cycle, the method will re-execute the inverse optimization based on the latest operating condition data, so that the control method has a dynamic adaptive capability of "going with the flow" and always guides the production process toward the optimal goal.
[0116] Step S40: Input the optimal value of the operation data, the target value of the output data of the controlled object, and the current disturbance data and operation data into the multivariable MPC controller to predict the operation data for future periods and obtain the operation data sequence for future periods.
[0117] In this embodiment, the multivariable MPC controller is a dynamic mathematical model (such as a state-space model) that describes how the process changes over time, such as the steam valve opening sequence and the water supply valve opening sequence for the next 1 second, 2 seconds, ..., 10 seconds. The multivariable MPC controller employs a "rolling optimization" strategy. At the next sampling time, it recalculates the entire optimization based on the new measurements and outputs a new first control action; this allows it to continuously correct prediction errors and cope with unknown disturbances.
[0118] This embodiment utilizes a multivariable MPC controller to predict the trajectory of changes in variables such as operational data over a future period (prediction time domain). This solves the oscillation problem that traditional PID control algorithms have when dealing with multiple variables, giving the control action advantages such as high efficiency and coordination.
[0119] Step S50: Based on the operation data sequence of future time periods, generate control instructions for the controlled object. In this embodiment, the control instructions for the controlled object include: control instructions for the steam valve, control instructions for the water supply valve, and control instructions for the servo motor, so as to adjust the steam flow rate, water supply flow rate, and the rotation speed of the humidifier.
[0120] This embodiment utilizes the seamless integration of a positive data prediction model and a multivariate MPC controller. The positive data prediction model is responsible for intelligent decision-making of the set value (the optimal value of the operating data), while the multivariate MPC controller is responsible for accurately executing the set value. Together, they constitute the core technological foundation for realizing intelligent and lean production of leaf picking and re-drying.
[0121] Example 2
[0122] Figure 2 This is a machine learning-based intelligent control system for leaf harvesting and re-drying provided in one embodiment of the present invention. For example... Figure 2 As shown, this embodiment provides a machine learning-based intelligent control system for leaf removal and re-drying. The system is used to implement the machine learning-based intelligent control method for leaf removal and re-drying in Embodiment 1. The system includes:
[0123] The first prediction module is used to acquire the current disturbance data and operation data of the controlled object at the beginning of the preset prediction period, input the current disturbance data and operation data into the pre-built positive data prediction model to predict the output data of the controlled object, and obtain the predicted value of the output data of the controlled object.
[0124] The function construction module is used to construct the inverse objective function of the operation data based on the predicted value of the output data of the controlled object and the preset target value of the output data.
[0125] The function solving module is used to solve the inverse objective function based on optimization algorithms to obtain the optimal value of the operational data;
[0126] The second prediction module is used to input the optimal value of the operation data, the target value of the output data, and the current interference data and operation data into the multivariable MPC controller to predict the operation data for future periods, and obtain the operation data sequence for future periods.
[0127] The instruction generation module is used to generate control instructions for the controlled object based on the operation data sequence for future time periods.
[0128] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described intelligent control method for leaf re-baking based on machine learning.
[0129] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described machine learning-based intelligent control method for leaf re-baking.
[0130] This embodiment constructs a forward data prediction model. At the initial moment of each preset prediction cycle, the current disturbance data and operation data are input into the forward data prediction model to predict the output data prediction value of the controlled object. Then, the optimal operation data, i.e., the best value of the operation data, is found by using the inverse objective function. Therefore, in each cycle, this method will re-execute the inverse optimization based on the latest operating data, giving the control method a dynamic adaptive capability that "adapts to circumstances" and always guides the production process toward the optimal goal. Secondly, after obtaining the optimal operation data, the multivariable MPC controller is used to predict the change trajectory of variables such as operation data in the future (prediction time domain). This can solve the oscillation problem of traditional PID control algorithms when dealing with multiple variables, making the control action efficient and coordinated. Furthermore, the seamless integration of the forward data prediction model and the multivariable MPC controller, with the forward data prediction model responsible for intelligent decision-making of the setpoint (optimal value of the operation data) and the multivariable MPC controller responsible for precise execution of the setpoint, together constitute the core technological foundation for realizing intelligent and lean production of leaf picking and re-drying.
[0131] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0132] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.
[0133] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A machine learning-based intelligent control method for leaf re-baking, characterized in that, The method includes: At the beginning of the preset prediction period, the current disturbance data and operation data of the controlled object are acquired. The current disturbance data and operation data are input into the pre-built positive data prediction model to predict the output data of the controlled object, and the predicted value of the output data of the controlled object is obtained. Construct an inverse objective function for the operational data based on the predicted output data of the controlled object and the preset target output data value; The optimal value of the operational data is obtained by solving the inverse objective function based on the optimization algorithm. The optimal value of the operating data, the target value of the output data, and the current disturbance data are input into the multivariable MPC controller to predict the operating data for future periods, thus obtaining the operating data sequence for future periods. Based on the operational data sequence for future time periods, control instructions for the controlled object are generated.
2. The intelligent control method for leaf re-baking based on machine learning according to claim 1, characterized in that, The controlled object is a leaf humidifier, leaf threshing machine, or leaf drying machine. When the controlled object is a leaf humidifier, the interference data includes: the inlet moisture, inlet temperature, ambient temperature, and ambient humidity of the leaf humidifier. The operation data includes: the steam flow rate, water flow rate, and rotation speed of the leaf humidifier drum. The output data includes: the outlet moisture and outlet temperature of the leaf humidifier.
3. The intelligent control method for leaf re-baking based on machine learning according to claim 1, characterized in that, The method further includes: constructing a positive data prediction model, including: Acquire historical operation data, historical output data, and corresponding historical interference data of the controlled object; The historical operation data and historical interference data are preprocessed and feature constructed to obtain a feature set, which includes: univariate features, lag features, sliding window features, difference features and interaction features. A training set is constructed based on the feature set and historical output data. The pre-set machine learning model is trained using the training set to obtain the trained model, which is then used as a positive data prediction model.
4. The intelligent control method for leaf re-baking based on machine learning according to claim 3, characterized in that, The machine learning model is either GradientDescent, RandomForest, or XGBoost.
5. The intelligent control method for leaf re-baking based on machine learning according to claim 1, characterized in that, Construct an inverse objective function for the operational data based on the predicted output data of the controlled object and the preset target output data value, including: The first difference function is determined based on the predicted value and target value of the output data of the controlled object. The second difference function is determined based on the operational data and the corresponding nominal values. By weighting the first and second difference functions, the inverse objective function is obtained.
6. The intelligent control method for leaf re-baking based on machine learning according to claim 5, characterized in that, The optimization algorithm is a differential evolution algorithm. Based on the optimization algorithm, the inverse objective function is solved to obtain the optimal value of the operational data, including: Initialize the control parameters of the differential evolution algorithm, including: initial population size, scaling factor, maximum number of iterations, and crossover probability; An initial population of the initial size is randomly generated based on the constraints of the inverse objective function, and each individual in the initial population is a solution of the inverse objective function; wherein, the constraints of the inverse objective function include: upper and lower limits of the operating data and a limit on the rate of change. Using the initial population as the current population, calculate the fitness of each individual in the current population, and select the best and worst individuals from the current population based on their fitness. Determine whether the preset termination condition is met. If so, take the solution of the inverse objective function corresponding to the best individual in the current population as the optimal value of the operation data. If not, based on the elite mutation strategy and the reverse crossover strategy, perform mutation and crossover operations on each individual in the current population in turn to obtain a new population. Use the new population to update the current population, recalculate the fitness of each individual in the current population, and select the best and worst individuals until the preset termination condition is met.
7. The intelligent control method for leaf re-baking based on machine learning according to claim 6, characterized in that, The elite mutation strategy is as follows: Obtain the optimal individual from each iteration and the individual corresponding to the minimum value of the inverse objective function in the current iteration; Based on the best individual from each iteration, the individual corresponding to the minimum value of the inverse objective function in the current iteration, and the scaling factor, perform mutation on each individual in the current population to obtain the mutated individuals.
8. The intelligent control method for leaf re-baking based on machine learning according to claim 6, characterized in that, The reverse crossover strategy is as follows: Perform a crossover operation on the mutated individuals in the current population to obtain intermediate individuals; Obtain the best and worst individuals from each iteration; The intermediate individuals are back-fused with the best and worst individuals from each iteration to obtain the crossover individuals.
9. A machine learning-based intelligent control system for leaf removal and re-drying, used to implement the machine learning-based intelligent control method for leaf removal and re-drying as described in any one of claims 1-8, characterized in that, The system includes: The first prediction module is used to acquire the current disturbance data and operation data of the controlled object at the beginning of the preset prediction period, input the current disturbance data and operation data into the pre-built positive data prediction model to predict the output data of the controlled object, and obtain the predicted value of the output data of the controlled object. The function construction module is used to construct the inverse objective function of the operation data based on the predicted value of the output data of the controlled object and the preset target value of the output data. The function solving module is used to solve the inverse objective function based on optimization algorithms to obtain the optimal value of the operational data; The second prediction module is used to input the optimal value of the operation data, the target value of the output data, and the current interference data and operation data into the multivariable MPC controller to predict the operation data for future periods, and obtain the operation data sequence for future periods. The instruction generation module is used to generate control instructions for the controlled object based on the operation data sequence for future time periods.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the machine learning-based intelligent control method for leaf re-baking as described in any one of claims 1-8.