Moisture control method, device and equipment for leaf moistening machine and storage medium

Abstract

The invention discloses a moisture control method, device and equipment of a leaf moistening machine and a storage medium, and relates to the technical field of tobacco production equipment.The method comprises the steps that data are collected in real time through an inlet and outlet sensor of the leaf moistening machine, and a moisture dynamic model is constructed; based on the model, an improved volume Kalman filter is adopted to accurately estimate the tobacco leaf moisture content; meanwhile, disturbance such as environment humidity and flow is predicted by using an attention mechanism neural network, and a feedforward compensation signal is generated to realize advanced control. And then, the model, the estimated value and the compensation signal are fused, rolling optimization is performed through a model prediction control algorithm, optimal solutions of control quantities such as the water adding amount and the steam amount are solved, and multi-variable cooperative control is achieved. And finally, an instruction is issued to an execution mechanism, a closed loop is formed according to outlet water content feedback, errors are continuously corrected, it is ensured that the water content is stabilized near a target value, and the control precision and energy efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of tobacco production equipment technology, and in particular to a method, apparatus, equipment and storage medium for moisture control of a leaf moistening machine. Background Technology

[0002] With the tobacco industry's ever-increasing demands for product quality, the leaf-humidifying machine, as a key piece of equipment in the tobacco processing, directly impacts the processing efficiency through its moisture control. However, traditional leaf-humidifying machines rely primarily on experience-based adjustments and simple feedback control, such as PID control or manual adjustment, resulting in slow response and low accuracy. Especially under disturbances such as fluctuations in ambient temperature and humidity, and changes in tobacco leaf flow rate, the tobacco leaves are prone to becoming overly wet or overly dry, causing quality loss and energy waste.

[0003] Currently, although there are some control methods based on sensor-based collection of moisture content and temperature, these methods often neglect the prediction of environmental disturbances and nonlinear dynamic modeling, and cannot achieve forward-looking compensation. Summary of the Invention

[0004] The main objective of this application is to provide a moisture control method, device, equipment, and storage medium for a leaf-curing machine, in order to solve the problem that the moisture control of traditional leaf-curing machines in the prior art mainly relies on experience adjustment and simple feedback control, such as PID control or manual adjustment, which leads to slow response and low accuracy. In particular, under disturbances such as fluctuations in ambient temperature and humidity and changes in tobacco leaf flow, tobacco leaves are prone to becoming too wet or too dry, resulting in quality loss and energy waste.

[0005] To achieve the above objectives, this application provides the following technical solution: A method for controlling the moisture content of a leaf-curing machine, wherein sensors for collecting tobacco leaf data are installed at both the inlet and outlet of the leaf-curing machine, and the moisture control method includes: Step S1: Collect key parameter data of tobacco leaves in real time through sensors at the inlet and outlet, and construct a dynamic moisture model of the leaf humidifier based on the key parameter data and the physical property parameters of the tobacco leaves. Step S2: Based on the moisture dynamic model, construct an improved volumetric Kalman filter estimation model, and obtain the estimated moisture content of tobacco leaves during the wetting process through the improved volumetric Kalman filter estimation model; Step S3: Perturb the estimated moisture content using a neural network based on an attention mechanism, and generate a feedforward compensation signal based on a preset target moisture content. Step S4: Construct a leaf-moistening machine control model based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, and calculate the optimal solution of the control quantity of the leaf-moistening machine control model; Step S5: The optimal solution of the control quantity is allocated to the execution component of the leaf-wetting machine and the optimal solution of the control quantity is executed by the execution component.

[0006] Beneficial effects: This application first systematically collects key parameters of tobacco leaves in real time, such as moisture content, temperature, ambient humidity, and instantaneous flow rate, using multiple sensors installed at the inlet and outlet of the leaf-conditioning machine. Based on this multimodal data and the physical characteristics of the tobacco leaves themselves, a mathematical model is constructed that can accurately describe the dynamic changes in moisture during the leaf-conditioning process. This model lays a solid foundation for subsequent accurate estimation and control.

[0007] Based on the established moisture dynamics model, the scheme constructs an improved volumetric Kalman filter estimation model to obtain the estimated moisture content of tobacco leaves during the wetting process. This improved estimator, through adaptive noise adjustment technology, can effectively overcome the instability problem of traditional filtering methods when facing uncertainties in process noise and observation noise. This enables more accurate and stable real-time tracking of the critical state of internal moisture content of tobacco leaves, mitigating estimation errors caused by delays or fluctuations in direct sensor measurements, and providing a reliable basis for control decisions.

[0008] After obtaining an accurate moisture content estimate, the scheme further utilizes an attention-based neural network to predict disturbances affecting moisture control. The attention mechanism effectively captures the varying degrees of influence of different parameters (such as ambient humidity and flow rate changes) on the future from historical data sequences, thus enabling proactive prediction of disturbances. Based on the prediction results and the preset target moisture content, the system can generate a forward-looking feedforward compensation signal. This design allows the control system to proactively predict and counteract the impact of disturbances, rather than passively waiting for errors to occur before making corrections. This significantly enhances the system's robustness in the face of complex and changing operating conditions, avoiding large fluctuations in moisture content caused by environmental or material changes.

[0009] Subsequently, the scheme integrates the dynamic moisture model, real-time moisture content estimates, and feedforward compensation signals to construct a leaf-humidifying machine control model. The optimal solution for the control variables is then obtained through rolling optimization using a model predictive control algorithm. This method explicitly handles the coupling relationships between multiple variables and the constraints of the actuators. By optimizing a cost function that comprehensively considers tracking accuracy and control costs, it calculates the most reasonable control commands for water addition, steam addition, hot air volume, and dehumidification volume over a future period. This not only improves control accuracy but also achieves efficient and coordinated allocation among multiple actuators, optimizing overall energy efficiency.

[0010] Finally, the calculated optimal control values ​​are directly sent to the various execution components of the leaf humidifier, such as the atomizing water spray system and the direct injection steam system, to achieve real-time and precise control of the leaf humidifier's operation. Crucially, the solution feeds back the actual tobacco leaf moisture content detected at the outlet to the moisture content estimation module, forming a complete closed-loop control system. This closed-loop feedback mechanism continuously corrects model errors and unknown disturbances, ensuring that the entire system can stably maintain the tobacco leaf moisture content near the target value over a long period, achieving rapid and stable adjustment of the tobacco leaf moisture content.

[0011] As a further improvement to this application, step S1 involves real-time acquisition of key parameter data of the tobacco leaves through sensors at the inlet and outlet, and constructing a dynamic moisture model of the leaf-humidifying machine based on the key parameter data and the physical characteristic parameters of the tobacco leaves, including: Step S11: Real-time collection of the inlet moisture content of the tobacco leaves based on the moisture content sensor, temperature sensor, humidity sensor, and flow sensor at the inlet and outlet. Inlet temperature Ambient humidity Instantaneous flow and export moisture content outlet temperature ; Step S12, based on the inlet moisture content The inlet temperature The ambient humidity The instantaneous flow rate The aforementioned outlet moisture content The outlet temperature Based on the physical properties of the tobacco leaves, a dynamic moisture model of the leaf-moistening machine is constructed using equation (1): (1); in, Based on tobacco leaf moisture content and tobacco leaf temperature The state vector; It is the transpose matrix; The first derivative of the state vector; To control the input vector, including the amount of water added. Steam addition amount Hot air volume and tidal volume ; The disturbance vector includes the ambient humidity. and instantaneous flow ; This is the measurement output of the aforementioned moisture dynamics model. For process noise, It is observation noise; It is a nonlinear dynamic function; For observation functions.

[0012] Beneficial effects: Step S11 enables multimodal, comprehensive monitoring of the tobacco leaf wetting process. By covering key points at the inlet and outlet, the sensor network can capture real-time changes in the tobacco leaf's state during processing, such as fluctuations in ambient humidity or instantaneous differences in tobacco leaf flow rate. This data provides a rich source of information for subsequent analysis. Compared to traditional methods that may rely on a single outlet sensor or manual recording, S11's comprehensive acquisition strategy avoids data blind spots and ensures the continuity and integrity of parameters. This not only improves the representativeness of the data but also enables the system to promptly detect external disturbances, such as changes in ambient temperature and humidity or uneven feeding, creating conditions for early warning and rapid response.

[0013] Step S12 transforms practical experience into a quantifiable mathematical framework, enabling the control system to accurately describe the dynamic behavior of the leaf moistening process. The model encompasses key variables such as tobacco leaf moisture content and temperature through state vectors, and incorporates control inputs and disturbance vectors to simulate the effects of water addition, steam generation, and environmental changes on moisture content. This modeling approach not only captures the system's nonlinear characteristics, such as the saturation effect of tobacco leaf hygroscopicity, but also considers noise and uncertainties, providing a theoretical foundation for subsequent filtering estimation and predictive control. Compared to simple linear approximations or empirical formulas in traditional methods, the dynamic model in S12 is more universal and adaptable, capable of handling variations under different operating conditions.

[0014] Steps S11 and S12 lay the initial intelligent foundation for closed-loop control. Through real-time data acquisition and precise modeling, the system is no longer a passive response, but can actively sense and predict changes, providing the prerequisites for disturbance compensation and optimized control in subsequent steps.

[0015] As a further improvement to this application, step S2 involves constructing an improved volumetric Kalman filter estimation model based on the aforementioned moisture dynamic model, and obtaining an estimated value of the moisture content of tobacco leaves during the wetting process using the improved volumetric Kalman filter estimation model, including: Step S21, the water dynamics model is discretized based on equation (2): (2); in, Let k be the state vector of the system at time k. Let be the state vector of the system at time k+1. This is the control input for the system at time k. Let k be the process noise vector of the system at time k. Let K be the observation noise vector of the system at time k. This is the observation vector of the system at time k+1; Step S22, define the material moisture absorption coefficient during the leaf wetting process according to equation (3): (3); in, The system parameters are unknown at time k. Let be the extended state vector of the system at time k; Step S23, according to equation (4), the extended state vector is extended into a cubic sampling point state vector: (4); in, Let be the cubic sampling point vector. Let be the square root of the covariance of the system at time k. Let be the state vector of the i-th cubic sampling point at time k in the system; Step S24, predict the propagation result of each cubic sampling point using equation (5): (5); in, Let i be the weight of the i-th cubic sampling point. Let be the predicted state vector of the i-th cubic sampling point at time k+1 of the system; Let the square root of the prediction covariance at time k+1 be the system value. Decomposition for QR Let be the predicted state vector of the system at time k+1; Step S25: Define the observation noise using equation (6), and adjust the observation noise in real time using adaptive noise estimation: (6); in, Here is the estimated observation noise value of the system at time k. This is the estimated observation noise value of the system at time k-1. This is the estimated value of the observation noise covariance at time k. This is the estimated value of the observation noise covariance at time k-1. Let k be the updated weight coefficients of the system at time k. Represents the innovation vector. This represents the transpose of the innovation vector. Let be the observation vector of the system at time k. This represents the observation vector of the prediction system at time k. This represents the transposed observation vector of the prediction system at time k. This represents the observation vector of the i-th sampling point at time k in the system. Let represent the transposed observation vector of the i-th sampling point at time k in the system; Step S26, update the observation vector using equation (7); (7); in, Let be the observation vector of the i-th sampling point at time k+1 of the system. Let i be the i-th cubic sampling point at time k+1 of the system. Let be the predicted observation vector of the system at time k+1. Let be the square root of the covariance of the system at time k+1. Let the cross-covariance of the state observations at time k+1 be the system's cross-covariance. To observe the square root of the noise covariance, Let be the state vector of the system at time k+1. Let be the transposed observation vector of the system at time k+1; Step S27, combining equations (5), (6), and (7) yields the final state update expression (8): (8); in, Let K+1 be the Kalman gain matrix of the system. Let be the state vector of the system at time k+1. Let be the predicted observation vector at time k+1 of the system; defined according to equation (1) The Including the estimated moisture content and temperature estimates .

[0016] Beneficial effects: Step S21 discretizes the water dynamics model, transforming the continuous-time nonlinear system into a discrete form, which facilitates real-time calculation by the digital controller.

[0017] Next, step S22 performs state extension, incorporating uncertain parameters such as the material's hygroscopic coefficient as part of the extended state vector. This helps capture dynamic characteristics that are difficult to model during the leaf wetting process, such as changes in the tobacco leaf's hygroscopicity, thereby reducing estimation bias caused by model mismatch. Through state extension, the estimator can simultaneously track changes in system state and parameters, improving the model's adaptability and generalization ability, especially maintaining estimation accuracy when tobacco varieties or processing conditions change. Step S23 generates a cubic sampling point state vector. As a key innovation of volumetric Kalman filtering, it approximates the probability distribution of the state through a set of symmetrical sampling points, handling nonlinear systems more efficiently than traditional methods.

[0018] Subsequently, step S24 predicts the propagation results for each sampling point, obtaining a preliminary predicted state value through weighted averaging. This step enhances the smoothness and consistency of the estimation, avoiding estimation fluctuations caused by abrupt disturbances. Furthermore, QR decomposition is introduced during the prediction process to maintain the numerical stability of the covariance matrix, preventing algorithm divergence and ensuring long-term operational reliability. Step S25 employs adaptive noise estimation to adjust the observed noise in real time, dynamically updating noise statistical characteristics to cope with changes in sensor performance or environmental interference. The adaptive mechanism automatically corrects noise parameters based on the innovation vector (i.e., the difference between prediction and measurement), reducing the conservatism of a fixed noise model and maintaining estimation accuracy even when noise is uncertain. Step S26 further updates the observation vector, fusing the predicted value with the actual measurement, and optimizing the estimation by calculating the state-observation cross-covariance. This step ensures efficient data utilization and avoids information loss.

[0019] Finally, step S27 completes the state update by weightedly fusing the predicted value and the measured information using the Kalman gain matrix, outputting the optimal state estimate, which includes the estimated moisture content of the tobacco leaves. This minimizes the estimation error, allowing the moisture content estimate to converge quickly to the true value with minimal overshoot. The synergistic effect of these sub-steps enables the estimator not only to handle nonlinear dynamics but also to continuously correct errors through closed-loop feedback (such as feeding the outlet moisture content back to the estimator in step S5), thereby significantly improving the stability and response speed of the control.

[0020] As a further improvement to this application, step S3 involves perturbating and predicting the estimated moisture content using a neural network based on an attention mechanism, and generating a feedforward compensation signal based on a preset target moisture content, including: Step S31: Obtain historical key parameter data and construct a multimodal time series. ,in, From arrive Estimated sequence of tobacco leaf moisture content at time points; From arrive Ambient humidity sequence over time; From arrive Ambient temperature sequence at any given time; From arrive The sequence of tobacco leaf flow rates at any given time; N is the historical step size; Step S32: Convert all the values ​​in the multimodal time series into a high-dimensional vector using equation (8). : (8); in, Let be the weight matrix of the multimodal time series; It is the bias vector; Step S33, define feedforward control according to equation (9): (9); in, For feedforward output, As the input to the high-dimensional vector, The ReLU activation function is used. , ; Step S34: Perturbation prediction is performed on the estimated water content using a neural network based on an attention mechanism, and the feedforward compensation signal is generated according to equation (10) based on the predicted perturbation and the target water content: (10); in, Forward compensation vector, The inverse function of the model. For the target moisture content, To estimate the moisture content, This is the integral of the predicted disturbance value.

[0021] Beneficial effects: Steps S31 to S34, as the core components of the attention-based neural network perturbation prediction, achieve forward-looking intelligent adjustment of the moisture control of the leaf-dripping machine through a series of processes, including multimodal time series construction, high-dimensional vector transformation, feedforward control definition, and compensation signal generation. These steps work synergistically to significantly improve the system's prediction accuracy, anti-interference capability, and response speed, overcoming the limitations of traditional methods that rely on hysteresis feedback.

[0022] Step S31 systematically integrates multi-dimensional information such as estimated tobacco leaf moisture content, environmental humidity, environmental temperature, and flow rate by acquiring historical key parameter data and constructing a multimodal time series, forming a continuous time series input. Step S32 converts all values ​​in the multimodal time series into high-dimensional vectors through an embedding layer, using linear transformation and bias vectors to capture the implicit relationships between features. This processing maps the original data to a higher-dimensional space, enhancing the model's expressive power and enabling the neural network to learn nonlinear interactions between parameters, such as the coupling effect between tobacco leaf moisture content and environmental humidity, thereby extracting perturbation features more accurately. Step S33 further improves the model's complexity and fitting ability by defining a feedforward control structure and introducing nonlinear components such as the ReLU activation function, ensuring that the network can handle dynamic changes during the leaf wetting process. Step S34 uses a neural network with an attention mechanism to predict perturbations in the moisture content estimate. An encoder-decoder architecture processes the input sequence, adding positional encoding to preserve temporal information. A multi-head self-attention mechanism is used to monitor key points in historical data in parallel, enabling proactive prediction of perturbations such as changes in environmental humidity and flow fluctuations. Based on the prediction results and the target moisture content, the model's inverse function generates a feedforward compensation signal, allowing the system to proactively adjust control parameters such as water addition and steam output, rather than passively responding to errors, effectively offsetting the impact of anticipated perturbations.

[0023] As a further improvement to this application, step S4, constructing a leaf-moistening machine control model based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, and calculating the optimal solution of the control quantity of the leaf-moistening machine control model, includes: Step S41: Based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, construct a leaf-moistening machine control model, and define the optimization strategy of the leaf-moistening machine control model according to equation (11): (11); Where H is the prediction step size, The input weight matrix, For the control quantity weight matrix, For the control quantity increment weight matrix, Let H be the sum of the second norms of the weighted state vector Q from 0 to H. Let H be the summation of the second norm of the weighted control quantity R from 0 to H-1. Let S be the second norm of the control increment; Step S42, solve equation (11) to obtain the optimal solution of the control quantity. ,in, To achieve the optimal water addition, To achieve the optimal amount of steam added, To achieve the optimal hot air volume, This is the optimal tidal discharge rate.

[0024] Beneficial effects: Steps S41 and S42, as the core optimization steps of the leaf-conditioning machine moisture control method, achieve high-precision, coordinated, and real-time control of tobacco leaf moisture content by constructing a leaf-conditioning machine control model and solving for the optimal solution of the control variables. These steps, based on a dynamic moisture model, moisture content estimates, and feedforward compensation signals, establish a comprehensive optimization framework. Step S41 first constructs a leaf-moistening machine control model based on the dynamic moisture model, real-time moisture content estimate, and feedforward compensation signal, and defines an optimization strategy using equation (11). This optimization strategy is centered on a multi-objective cost function, minimizing the deviation between the state vector and the target reference value, the amplitude of the control quantity, and the incremental change of the control quantity. Among them, the cost function comprehensively considers the state tracking accuracy within the prediction step (weighted by weight matrix Q), the energy consumption of the control quantity (weighted by weight matrix R), and the smoothness of the control action (weighted by weight matrix S). This design ensures that the system not only focuses on the rapid convergence of the tobacco leaf moisture content to the target value, but also takes into account the energy consumption and action amplitude of the actuators (such as water addition, steam addition, hot air volume, and dehumidification volume), avoiding over-adjustment or actuator saturation problems. Step S41 integrates the feedforward compensation signal to incorporate disturbance prediction information into the optimization framework, enabling the controller to compensate for disturbances such as changes in environmental humidity or flow rate in advance, rather than passively responding to errors, thereby significantly improving the system's forward-looking control capability.

[0025] Step S42 uses a numerical optimization algorithm to solve the aforementioned cost function, obtaining the optimal solutions for the control variables, including the optimal water supply, steam supply, hot air supply, and humidification rate. The solution process typically employs a high-efficiency optimizer (such as a quadratic programming solver) to rapidly calculate the optimal control sequence under real-time constraints, ensuring the controller can operate online. This step achieves multivariate decoupling and collaborative allocation.

[0026] As a further improvement to this application, step S5 involves allocating the optimal solution of the control quantity to the execution component of the leaf-wetting machine and executing the optimal solution of the control quantity through the execution component. Afterwards, the process includes: Step S10: Real-time acquisition of optimized key parameter data of tobacco leaves through sensors at the inlet and outlet. Step S20: The optimized key parameter data is used as the execution subject input to step S2, and steps S2 to S5 are repeated to form a closed-loop control.

[0027] Beneficial effects: Steps S10 and S20, as the core components of the closed-loop control method in the leaf-humidifying machine's moisture control system, continuously optimize and dynamically adjust the entire control system by real-time acquisition of optimized key parameter data and feedback to the estimation module, thereby significantly improving the accuracy, stability, and adaptability of moisture control. In step S10, after executing the optimal solution for the control quantity in step S5, actual parameter data of the tobacco leaves, including moisture content, temperature, ambient humidity, and flow rate, are collected again by sensors at the inlet and outlet. This data reflects the actual effect of the current control action. Step S20 then uses the collected optimized data as input to re-inject into the improved volumetric Kalman filter estimation model of step S2, recalculates the estimated moisture content, and cyclically executes subsequent steps S3 to S5, forming a closed-loop control. This design ensures that the system is not adjusted only once, but can continuously correct the control strategy based on real-time feedback, effectively coping with the dynamic changes and uncertainties during the leaf-humidifying process.

[0028] As a further improvement to this application, step S5 involves allocating the optimal solution of the control quantity to the execution component of the leaf-wetting machine and executing the optimal solution of the control quantity through the execution component. Afterwards, the process includes: Step S100: Send the optimal solution of the control quantity to the external monitoring terminal.

[0029] Beneficial effects: Step S100 in the document serves as an extension of the moisture control method for the leaf humidifier. By sending the optimal solution of the control quantity calculated in step S5 to an external monitoring terminal in real time, it enables remote centralized monitoring and data sharing of the leaf humidifier's operating status. The main benefits of this step are improved system manageability and transparency, allowing operators or the central control system to track the execution of key control parameters such as water supply, steam supply, hot air supply, and dehumidification in real time, facilitating timely detection and intervention of anomalies.

[0030] To achieve the above objectives, this application also provides the following technical solutions: A moisture control device for a leaf-conditioning machine, the moisture control device being applied to the moisture control method described above, the moisture control device comprising: The moisture dynamic model construction module is used to collect key parameter data of tobacco leaves in real time through sensors at the inlet and outlet, and construct the moisture dynamic model of the leaf humidifier based on the key parameter data and the physical characteristic parameters of the tobacco leaves. The moisture content estimation module is used to construct an improved volumetric Kalman filter estimation model based on the moisture dynamic model, and obtain the moisture content estimate of tobacco leaves during the leaf wetting process through the improved volumetric Kalman filter estimation model. The feedforward compensation signal generation module is used to perform perturbation prediction on the moisture content estimate through a neural network based on an attention mechanism, and generate a feedforward compensation signal based on a preset target moisture content. The optimal solution calculation module for control quantity is used to construct a leaf-wetting machine control model based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, and to calculate the optimal solution for control quantity of the leaf-wetting machine control model. The optimal solution allocation module for control quantities is used to allocate the optimal solution for control quantities to the execution component of the leaf-wetting machine and execute the optimal solution for control quantities through the execution component.

[0031] To achieve the above objectives, this application also provides the following technical solutions: An electronic device includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the moisture control method as described above.

[0032] To achieve the above objectives, this application also provides the following technical solutions: A computer-readable storage medium storing program instructions that, when executed by a processor, can implement the moisture control method described above. Attached Figure Description

[0033] Figure 1 This is a schematic flowchart illustrating the steps of one embodiment of a moisture control method for a leaf-moistening machine according to this application; Figure 2 This is a functional module diagram of one embodiment of the moisture control device for a leaf-moistening machine according to this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the storage medium of this application. Detailed Implementation

[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0035] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0036] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0037] like Figure 1 As shown, a moisture control method for a leaf-curing machine is described. Sensors for collecting tobacco leaf data are installed at both the inlet and outlet of the leaf-curing machine. The moisture control method includes: Step S1: Collect key parameter data of tobacco leaves in real time through sensors at the inlet and outlet, and construct a dynamic moisture model of the leaf humidifier based on the key parameter data and the physical property parameters of the tobacco leaves.

[0038] Step S2: Based on the moisture dynamic model, construct an improved volumetric Kalman filter estimation model, and obtain the estimated moisture content of tobacco leaves during the leaf wetting process through the improved volumetric Kalman filter estimation model.

[0039] Step S3: Perturbation prediction of the moisture content estimate is performed by a neural network based on an attention mechanism, and a feedforward compensation signal is generated based on the preset target moisture content.

[0040] Step S4: Construct a leaf-moistening machine control model based on the moisture dynamic model, moisture content estimate, and feedforward compensation signal, and calculate the optimal solution of the control quantity of the leaf-moistening machine control model.

[0041] Step S5: The optimal solution of the control quantity is allocated to the execution component of the leaf-wetting machine and the optimal solution of the control quantity is executed by the execution component.

[0042] Beneficial effects: This application first systematically collects key parameters of tobacco leaves in real time, such as moisture content, temperature, ambient humidity, and instantaneous flow rate, using multiple sensors installed at the inlet and outlet of the leaf-conditioning machine. Based on this multimodal data and the physical characteristics of the tobacco leaves themselves, a mathematical model is constructed that can accurately describe the dynamic changes in moisture during the leaf-conditioning process. This model lays a solid foundation for subsequent accurate estimation and control.

[0043] Based on the established moisture dynamics model, the scheme constructs an improved volumetric Kalman filter estimation model to obtain the estimated moisture content of tobacco leaves during the wetting process. This improved estimator, through adaptive noise adjustment technology, can effectively overcome the instability problem of traditional filtering methods when facing uncertainties in process noise and observation noise. This enables more accurate and stable real-time tracking of the critical state of internal moisture content of tobacco leaves, mitigating estimation errors caused by delays or fluctuations in direct sensor measurements, and providing a reliable basis for control decisions.

[0044] After obtaining an accurate moisture content estimate, the scheme further utilizes an attention-based neural network to predict disturbances affecting moisture control. The attention mechanism effectively captures the varying degrees of influence of different parameters (such as ambient humidity and flow rate changes) on the future from historical data sequences, thus enabling proactive prediction of disturbances. Based on the prediction results and the preset target moisture content, the system can generate a forward-looking feedforward compensation signal. This design allows the control system to proactively predict and counteract the impact of disturbances, rather than passively waiting for errors to occur before making corrections. This significantly enhances the system's robustness in the face of complex and changing operating conditions, avoiding large fluctuations in moisture content caused by environmental or material changes.

[0045] Subsequently, the scheme integrates the dynamic moisture model, real-time moisture content estimates, and feedforward compensation signals to construct a leaf-humidifying machine control model. The optimal solution for the control variables is then obtained through rolling optimization using a model predictive control algorithm. This method explicitly handles the coupling relationships between multiple variables and the constraints of the actuators. By optimizing a cost function that comprehensively considers tracking accuracy and control costs, it calculates the most reasonable control commands for water addition, steam addition, hot air volume, and dehumidification volume over a future period. This not only improves control accuracy but also achieves efficient and coordinated allocation among multiple actuators, optimizing overall energy efficiency.

[0046] Finally, the calculated optimal control values ​​are directly sent to the various execution components of the leaf humidifier, such as the atomizing water spray system and the direct injection steam system, to achieve real-time and precise control of the leaf humidifier's operation. Crucially, the solution feeds back the actual tobacco leaf moisture content detected at the outlet to the moisture content estimation module, forming a complete closed-loop control system. This closed-loop feedback mechanism continuously corrects model errors and unknown disturbances, ensuring that the entire system can stably maintain the tobacco leaf moisture content near the target value over a long period, achieving rapid and stable adjustment of the tobacco leaf moisture content.

[0047] Further, in step S1, key parameter data of the tobacco leaves are collected in real time by sensors at the inlet and outlet, and a dynamic moisture model of the leaf-humidifying machine is constructed based on the key parameter data and the physical characteristic parameters of the tobacco leaves, including: Step S11: Real-time collection of the inlet moisture content of the tobacco leaves based on moisture content sensors, temperature sensors, humidity sensors, and flow sensors at the inlet and outlet. Inlet temperature Ambient humidity Instantaneous flow and export moisture content outlet temperature .

[0048] Step S12, based on inlet moisture content Inlet temperature Ambient humidity Instantaneous flow moisture content at export outlet temperature Based on the physical properties of the tobacco leaves, a dynamic moisture model of the leaf-moistening machine is constructed using equation (1): (1).

[0049] in, Based on tobacco leaf moisture content and tobacco leaf temperature The state vector; It is the transpose matrix; The first derivative of the state vector; To control the input vector, including the amount of water added. Steam addition amount Hot air volume and tidal volume ; The disturbance vector includes ambient humidity. and instantaneous flow ; This is the measurement output of the water dynamics model. For process noise, It is observation noise; It is a nonlinear dynamic function; For observation functions.

[0050] Beneficial effects: Step S11 enables multimodal, comprehensive monitoring of the tobacco leaf wetting process. By covering key points at the inlet and outlet, the sensor network can capture real-time changes in the tobacco leaf's state during processing, such as fluctuations in ambient humidity or instantaneous differences in tobacco leaf flow rate. This data provides a rich source of information for subsequent analysis. Compared to traditional methods that may rely on a single outlet sensor or manual recording, S11's comprehensive acquisition strategy avoids data blind spots and ensures the continuity and integrity of parameters. This not only improves the representativeness of the data but also enables the system to promptly detect external disturbances, such as changes in ambient temperature and humidity or uneven feeding, creating conditions for early warning and rapid response.

[0051] Step S12 transforms practical experience into a quantifiable mathematical framework, enabling the control system to accurately describe the dynamic behavior of the leaf moistening process. The model encompasses key variables such as tobacco leaf moisture content and temperature through state vectors, and incorporates control inputs and disturbance vectors to simulate the effects of water addition, steam generation, and environmental changes on moisture content. This modeling approach not only captures the system's nonlinear characteristics, such as the saturation effect of tobacco leaf hygroscopicity, but also considers noise and uncertainties, providing a theoretical foundation for subsequent filtering estimation and predictive control. Compared to simple linear approximations or empirical formulas in traditional methods, the dynamic model in S12 is more universal and adaptable, capable of handling variations under different operating conditions.

[0052] Steps S11 and S12 lay the initial intelligent foundation for closed-loop control. Through real-time data acquisition and precise modeling, the system is no longer a passive response, but can actively sense and predict changes, providing the prerequisites for disturbance compensation and optimized control in subsequent steps.

[0053] Further, in step S2, an improved volumetric Kalman filter estimation model is constructed based on the moisture dynamic model. The moisture content of tobacco leaves during the wetting process is estimated using this improved volumetric Kalman filter estimation model, including: Step S21, discretize the water dynamics model based on equation (2): (2).

[0054] in, Let k be the state vector of the system at time k. Let be the state vector of the system at time k+1. This is the control input for the system at time k. Let k be the process noise vector of the system at time k. Let K be the observation noise vector of the system at time k. Let be the observation vector of the system at time k+1.

[0055] Step S22, define the material moisture absorption coefficient during the leaf wetting process according to equation (3): (3).

[0056] in, The system parameters are unknown at time k. Let be the extended state vector of the system at time k.

[0057] Step S23, according to equation (4), the extended state vector is extended into a cubic sampling point state vector: (4).

[0058] in, For cubic sampling point vectors, Let be the square root of the covariance of the system at time k. Let be the state vector of the i-th cubic sampling point at time k in the system.

[0059] Step S24, predict the propagation result of each cubic sampling point using equation (5): (5).

[0060] in, Let i be the weight of the i-th cubic sampling point. Let be the predicted state vector of the i-th cubic sampling point at time k+1 of the system; Let the square root of the prediction covariance at time k+1 be the system value. Decomposition for QR Let be the predicted state vector of the system at time k+1.

[0061] Step S25: Define the observation noise using equation (6), and adjust the observation noise in real time using adaptive noise estimation: (6).

[0062] in, Here is the estimated observation noise value of the system at time k. This is the estimated observation noise value of the system at time k-1. This is the estimated value of the observation noise covariance at time k. This is the estimated value of the observation noise covariance at time k-1. Let k be the updated weight coefficients of the system at time k. Represents the innovation vector. This represents the transpose of the innovation vector. Let be the observation vector of the system at time k. This represents the observation vector of the prediction system at time k. This represents the transposed observation vector of the prediction system at time k. This represents the observation vector of the i-th sampling point at time k in the system. Let represent the transposed observation vector of the i-th sampling point at time k in the system.

[0063] Step S26, update the observation vector using equation (7).

[0064] (7).

[0065] in, Let be the observation vector of the i-th sampling point at time k+1 of the system. Let i be the i-th cubic sampling point at time k+1 of the system. Let be the predicted observation vector of the system at time k+1. Let be the square root of the covariance of the system at time k+1. Let the cross-covariance of the state observations at time k+1 be the system's cross-covariance. To observe the square root of the noise covariance, Let be the state vector of the system at time k+1. Let be the transposed observation vector of the system at time k+1.

[0066] Step S27, combining equations (5), (6), and (7) yields the final state update expression (8): (8).

[0067] in, Let K+1 be the Kalman gain matrix of the system. Let be the state vector of the system at time k+1. Let be the predicted observation vector at time k+1 of the system; defined according to equation (1) , Including moisture content estimates and temperature estimates .

[0068] Beneficial effects: Step S21 discretizes the water dynamics model, transforming the continuous-time nonlinear system into a discrete form, which facilitates real-time calculation by the digital controller.

[0069] Next, step S22 performs state extension, incorporating uncertain parameters such as the material's hygroscopic coefficient as part of the extended state vector. This helps capture dynamic characteristics that are difficult to model during the leaf wetting process, such as changes in the tobacco leaf's hygroscopicity, thereby reducing estimation bias caused by model mismatch. Through state extension, the estimator can simultaneously track changes in system state and parameters, improving the model's adaptability and generalization ability, especially maintaining estimation accuracy when tobacco varieties or processing conditions change. Step S23 generates a cubic sampling point state vector. As a key innovation of volumetric Kalman filtering, it approximates the probability distribution of the state through a set of symmetrical sampling points, handling nonlinear systems more efficiently than traditional methods.

[0070] Subsequently, step S24 predicts the propagation results for each sampling point, obtaining a preliminary predicted state value through weighted averaging. This step enhances the smoothness and consistency of the estimation, avoiding estimation fluctuations caused by abrupt disturbances. Furthermore, QR decomposition is introduced during the prediction process to maintain the numerical stability of the covariance matrix, preventing algorithm divergence and ensuring long-term operational reliability. Step S25 employs adaptive noise estimation to adjust the observed noise in real time, dynamically updating noise statistical characteristics to cope with changes in sensor performance or environmental interference. The adaptive mechanism automatically corrects noise parameters based on the innovation vector (i.e., the difference between prediction and measurement), reducing the conservatism of a fixed noise model and maintaining estimation accuracy even when noise is uncertain. Step S26 further updates the observation vector, fusing the predicted value with the actual measurement, and optimizing the estimation by calculating the state-observation cross-covariance. This step ensures efficient data utilization and avoids information loss.

[0071] Finally, step S27 completes the state update by weightedly fusing the predicted value and the measured information using the Kalman gain matrix, outputting the optimal state estimate, which includes the estimated moisture content of the tobacco leaves. This minimizes the estimation error, allowing the moisture content estimate to converge quickly to the true value with minimal overshoot. The synergistic effect of these sub-steps enables the estimator not only to handle nonlinear dynamics but also to continuously correct errors through closed-loop feedback (such as feeding the outlet moisture content back to the estimator in step S5), thereby significantly improving the stability and response speed of the control.

[0072] Further, in step S3, the moisture content estimate is perturbed and predicted using a neural network based on an attention mechanism, and a feedforward compensation signal is generated based on a preset target moisture content, including: Step S31: Obtain historical key parameter data and construct a multimodal time series. ,in, From arrive Estimated sequence of tobacco leaf moisture content at time points; From arrive Ambient humidity sequence over time; From arrive Ambient temperature sequence at any given time; From arrive The sequence of tobacco leaf flow at any given time; N is the historical step size.

[0073] Step S32: Convert all numerical values ​​in the multimodal time series into high-dimensional vectors using equation (8). : (8).

[0074] in, This is the weight matrix for the multimodal time series; This is the bias vector.

[0075] Step S33, define feedforward control according to equation (9): (9).

[0076] in, For feedforward output, As input to a high-dimensional vector, The ReLU activation function is used. , .

[0077] Step S34: Perturbation prediction is performed on the moisture content estimate using a neural network based on an attention mechanism. Based on the predicted perturbation and the target moisture content, a feedforward compensation signal is generated according to equation (10): (10).

[0078] in, Forward compensation vector, The inverse function of the model. For the target moisture content, To estimate the moisture content, This is the integral of the predicted disturbance value.

[0079] Beneficial effects: Steps S31 to S34, as the core components of the attention-based neural network perturbation prediction, achieve forward-looking intelligent adjustment of the moisture control of the leaf-dripping machine through a series of processes, including multimodal time series construction, high-dimensional vector transformation, feedforward control definition, and compensation signal generation. These steps work synergistically to significantly improve the system's prediction accuracy, anti-interference capability, and response speed, overcoming the limitations of traditional methods that rely on hysteresis feedback.

[0080] Step S31 systematically integrates multi-dimensional information such as estimated tobacco leaf moisture content, environmental humidity, environmental temperature, and flow rate by acquiring historical key parameter data and constructing a multimodal time series, forming a continuous time series input. Step S32 converts all values ​​in the multimodal time series into high-dimensional vectors through an embedding layer, using linear transformation and bias vectors to capture the implicit relationships between features. This processing maps the original data to a higher-dimensional space, enhancing the model's expressive power and enabling the neural network to learn nonlinear interactions between parameters, such as the coupling effect between tobacco leaf moisture content and environmental humidity, thereby extracting perturbation features more accurately. Step S33 further improves the model's complexity and fitting ability by defining a feedforward control structure and introducing nonlinear components such as the ReLU activation function, ensuring that the network can handle dynamic changes during the leaf wetting process. Step S34 uses a neural network with an attention mechanism to predict perturbations in the moisture content estimate. An encoder-decoder architecture processes the input sequence, adding positional encoding to preserve temporal information. A multi-head self-attention mechanism is used to monitor key points in historical data in parallel, enabling proactive prediction of perturbations such as changes in environmental humidity and flow fluctuations. Based on the prediction results and the target moisture content, the model's inverse function generates a feedforward compensation signal, allowing the system to proactively adjust control parameters such as water addition and steam output, rather than passively responding to errors, effectively offsetting the impact of anticipated perturbations.

[0081] Further, in step S4, a control model for the leaf-moistening machine is constructed based on the moisture dynamic model, the estimated moisture content, and the feedforward compensation signal, and the optimal solution of the control quantity for the leaf-moistening machine control model is calculated, including: Step S41: Construct a leaf-moistening machine control model based on the moisture dynamic model, moisture content estimate, and feedforward compensation signal, and define the optimization strategy of the leaf-moistening machine control model according to equation (11): (11).

[0082] Where H is the prediction step size, The input weight matrix, For the control quantity weight matrix, For the control quantity increment weight matrix, Let H be the sum of the second norms of the weighted state vector Q from 0 to H. Let H be the summation of the second norm of the weighted control quantity R from 0 to H-1. Let S be the second norm of the weighted control increment.

[0083] Step S42, solve equation (11) to obtain the optimal solution of the control quantity. ,in, To achieve the optimal water addition, To achieve the optimal amount of steam added, To achieve the optimal hot air volume, This is the optimal tidal discharge rate.

[0084] Beneficial effects: Steps S41 and S42, as the core optimization steps of the leaf-conditioning machine moisture control method, achieve high-precision, coordinated, and real-time control of tobacco leaf moisture content by constructing a leaf-conditioning machine control model and solving for the optimal solution of the control variables. These steps, based on a dynamic moisture model, moisture content estimates, and feedforward compensation signals, establish a comprehensive optimization framework. Step S41 first constructs a leaf-moistening machine control model based on the dynamic moisture model, real-time moisture content estimate, and feedforward compensation signal, and defines an optimization strategy using equation (11). This optimization strategy is centered on a multi-objective cost function, minimizing the deviation between the state vector and the target reference value, the amplitude of the control quantity, and the incremental change of the control quantity. Among them, the cost function comprehensively considers the state tracking accuracy within the prediction step (weighted by weight matrix Q), the energy consumption of the control quantity (weighted by weight matrix R), and the smoothness of the control action (weighted by weight matrix S). This design ensures that the system not only focuses on the rapid convergence of the tobacco leaf moisture content to the target value, but also takes into account the energy consumption and action amplitude of the actuators (such as water addition, steam addition, hot air volume, and dehumidification volume), avoiding over-adjustment or actuator saturation problems. Step S41 integrates the feedforward compensation signal to incorporate disturbance prediction information into the optimization framework, enabling the controller to compensate for disturbances such as changes in environmental humidity or flow rate in advance, rather than passively responding to errors, thereby significantly improving the system's forward-looking control capability.

[0085] Step S42 uses a numerical optimization algorithm to solve the aforementioned cost function, obtaining the optimal solutions for the control variables, including the optimal water supply, steam supply, hot air supply, and humidification rate. The solution process typically employs a high-efficiency optimizer (such as a quadratic programming solver) to rapidly calculate the optimal control sequence under real-time constraints, ensuring the controller can operate online. This step achieves multivariate decoupling and collaborative allocation.

[0086] Further, in step S5, the optimal solution of the control quantity is allocated to the execution component of the leaf-curing machine and the optimal solution of the control quantity is executed by the execution component. Afterwards, the process includes: Step S10: Real-time acquisition of optimized key parameter data of tobacco leaves using sensors at the inlet and outlet.

[0087] Step S20: The optimized key parameter data is used as the input to step S2, and steps S2 to S5 are repeated to form a closed-loop control.

[0088] Beneficial effects: Steps S10 and S20, as the core components of the closed-loop control method in the leaf-humidifying machine's moisture control system, continuously optimize and dynamically adjust the entire control system by real-time acquisition of optimized key parameter data and feedback to the estimation module, thereby significantly improving the accuracy, stability, and adaptability of moisture control. In step S10, after executing the optimal solution for the control quantity in step S5, actual parameter data of the tobacco leaves, including moisture content, temperature, ambient humidity, and flow rate, are collected again by sensors at the inlet and outlet. This data reflects the actual effect of the current control action. Step S20 then uses the collected optimized data as input to re-inject into the improved volumetric Kalman filter estimation model of step S2, recalculates the estimated moisture content, and cyclically executes subsequent steps S3 to S5, forming a closed-loop control. This design ensures that the system is not adjusted only once, but can continuously correct the control strategy based on real-time feedback, effectively coping with the dynamic changes and uncertainties during the leaf-humidifying process.

[0089] Further, in step S5, the optimal solution of the control quantity is allocated to the execution component of the leaf-curing machine and the optimal solution of the control quantity is executed by the execution component. Afterwards, the process includes: Step S100: Send the optimal solution of the control quantity to the external monitoring terminal.

[0090] Beneficial effects: Step S100 in the document serves as an extension of the moisture control method for the leaf humidifier. By sending the optimal solution of the control quantity calculated in step S5 to an external monitoring terminal in real time, it enables remote centralized monitoring and data sharing of the leaf humidifier's operating status. The main benefits of this step are improved system manageability and transparency, allowing operators or the central control system to track the execution of key control parameters such as water supply, steam supply, hot air supply, and dehumidification in real time, facilitating timely detection and intervention of anomalies.

[0091] like Figure 2 As shown, this embodiment provides an example of a moisture control device for a leaf-moistening machine. In this embodiment, the moisture control device is applied to the moisture control method as described in the above embodiment.

[0092] Specifically, the moisture control device includes a moisture dynamic model construction module 1, a moisture content estimation module 2, a feedforward compensation signal generation module 3, a control quantity optimal solution calculation module 4, and a control quantity optimal solution allocation module 5, which are connected electrically or by signal in sequence.

[0093] The system comprises the following modules: Moisture Dynamic Model Construction Module 1, which collects key parameter data of tobacco leaves in real time through sensors at the inlet and outlet, and constructs a moisture dynamic model of the leaf-wetting machine based on the key parameter data and the physical characteristics of the tobacco leaves; Moisture Content Estimation Module 2, which constructs an improved volumetric Kalman filter estimation model based on the moisture dynamic model, and obtains the moisture content estimate of the tobacco leaves during the leaf-wetting process through the improved volumetric Kalman filter estimation model; Feedforward Compensation Signal Generation Module 3, which performs perturbation prediction on the moisture content estimate using a neural network based on an attention mechanism, and generates a feedforward compensation signal based on a preset target moisture content; Control Quantity Optimal Solution Calculation Module 4, which constructs a leaf-wetting machine control model based on the moisture dynamic model, moisture content estimate, and feedforward compensation signal, and calculates the optimal solution of the control quantity of the leaf-wetting machine control model; and Control Quantity Optimal Solution Allocation Module 5, which allocates the optimal solution of the control quantity to the execution components of the leaf-wetting machine and executes the optimal solution of the control quantity through the execution components.

[0094] Furthermore, the moisture dynamic model construction module 1 specifically includes a first moisture dynamic model construction unit and a second moisture dynamic model construction unit that are electrically or signal-connected in sequence; the second moisture dynamic model construction unit is electrically or signal-connected to the moisture content estimation module 2.

[0095] The first moisture dynamic model construction unit is used to collect the inlet moisture content of tobacco leaves in real time based on moisture content sensors, temperature sensors, humidity sensors, and flow sensors at the inlet and outlet. Inlet temperature Ambient humidity Instantaneous flow and export moisture content outlet temperature .

[0096] The second moisture dynamic model construction unit is used to construct a model based on inlet moisture content. Inlet temperature Ambient humidity Instantaneous flow moisture content at export outlet temperature Based on the physical properties of the tobacco leaves, a dynamic moisture model of the leaf-moistening machine is constructed using equation (1): (1).

[0097] in, Based on tobacco leaf moisture content and tobacco leaf temperature The state vector; It is the transpose matrix; The first derivative of the state vector; To control the input vector, including the amount of water added. Steam addition amount Hot air volume and tidal volume ; The disturbance vector includes ambient humidity. and instantaneous flow ; This is the measurement output of the water dynamics model. For process noise, It is observation noise; It is a nonlinear dynamic function; For observation functions.

[0098] Furthermore, the moisture content estimation module 2 specifically includes a first moisture content estimation unit, a second moisture content estimation unit, a third moisture content estimation unit, a fourth moisture content estimation unit, a fifth moisture content estimation unit, a sixth moisture content estimation unit, and a seventh moisture content estimation unit that are electrically or signal-connected in sequence; the first moisture content estimation unit is electrically or signal-connected to the second moisture dynamic model construction unit, and the seventh moisture content estimation unit is electrically or signal-connected to the feedforward compensation signal generation module 3.

[0099] The first moisture content estimation unit is used to discretize the moisture dynamic model based on equation (2): (2).

[0100] in, Let k be the state vector of the system at time k. Let be the state vector of the system at time k+1. This is the control input for the system at time k. Let k be the process noise vector of the system at time k. Let K be the observation noise vector of the system at time k. Let be the observation vector of the system at time k+1.

[0101] The second moisture content estimation unit is used to define the material hygroscopic coefficient during the leaf wetting process according to equation (3): (3).

[0102] in, The system parameters are unknown at time k. Let be the extended state vector of the system at time k.

[0103] The third moisture content estimation unit is used to extend the extended state vector into a cubic sampling point state vector according to equation (4): (4).

[0104] in, For cubic sampling point vectors, Let be the square root of the covariance of the system at time k. Let be the state vector of the i-th cubic sampling point at time k in the system.

[0105] The fourth moisture content estimation unit is used to predict the propagation results of each cubic sampling point using equation (5): (5).

[0106] in, Let i be the weight of the i-th cubic sampling point. Let be the predicted state vector of the i-th cubic sampling point at time k+1 of the system; Let the square root of the prediction covariance at time k+1 be the system value. Decomposition for QR Let be the predicted state vector of the system at time k+1.

[0107] The fifth moisture content estimation unit is used to define the observation noise through equation (6) and adjust the observation noise in real time through adaptive noise estimation: (6).

[0108] in, Here is the estimated observation noise value of the system at time k. This is the estimated observation noise value of the system at time k-1. This is the estimated value of the observation noise covariance at time k. This is the estimated value of the observation noise covariance at time k-1. Let k be the updated weight coefficients of the system at time k. Represents the innovation vector. This represents the transpose of the innovation vector. Let be the observation vector of the system at time k. This represents the observation vector of the prediction system at time k. This represents the transposed observation vector of the prediction system at time k. This represents the observation vector of the i-th sampling point at time k in the system. Let represent the transposed observation vector of the i-th sampling point at time k in the system.

[0109] The sixth moisture content estimation unit is used to update the observation vector through equation (7).

[0110] (7).

[0111] in, Let be the observation vector of the i-th sampling point at time k+1 of the system. Let i be the i-th cubic sampling point at time k+1 of the system. Let be the predicted observation vector of the system at time k+1. Let be the square root of the covariance of the system at time k+1. Let the cross-covariance of the state observations at time k+1 be the system's cross-covariance. To observe the square root of the noise covariance, Let be the state vector of the system at time k+1. Let be the transposed observation vector of the system at time k+1.

[0112] The seventh moisture content estimation unit is used to combine equations (5), (6), and (7) to obtain the final state update expression (8): (8).

[0113] in, Let K+1 be the Kalman gain matrix of the system. Let be the state vector of the system at time k+1. Let be the predicted observation vector at time k+1 of the system; defined according to equation (1) , Including moisture content estimates and temperature estimates .

[0114] Furthermore, the feedforward compensation signal generation module 3 specifically includes a first feedforward compensation signal generation unit, a second feedforward compensation signal generation unit, a third feedforward compensation signal generation unit, and a fourth feedforward compensation signal generation unit that are electrically or signal-connected in sequence; the first feedforward compensation signal generation unit is electrically or signal-connected to the seventh moisture content estimation unit, and the fourth feedforward compensation signal generation unit is electrically or signal-connected to the control quantity optimal solution calculation module 4.

[0115] The first feedforward compensation signal generation unit is used to acquire historical key parameter data and construct a multimodal time series. ,in, From arrive Estimated sequence of tobacco leaf moisture content at time points; From arrive Ambient humidity sequence over time; From arrive Ambient temperature sequence at any given time; From arrive The sequence of tobacco leaf flow at any given time; N is the historical step size.

[0116] The second feedforward compensation signal generation unit is used to convert all the values ​​in the multimodal time series into a high-dimensional vector using equation (8). : (8).

[0117] in, This is the weight matrix for the multimodal time series; This is the bias vector.

[0118] The third feedforward compensation signal generation unit is used to define feedforward control according to equation (9): (9).

[0119] in, For feedforward output, As input to a high-dimensional vector, The ReLU activation function is used. , .

[0120] The fourth feedforward compensation signal generation unit is used to predict the perturbation of the water content estimate through a neural network based on an attention mechanism, and to generate a feedforward compensation signal according to equation (10) based on the predicted perturbation and the target water content: (10).

[0121] in, Forward compensation vector, The inverse function of the model. For the target moisture content, To estimate the moisture content, This is the integral of the predicted disturbance value.

[0122] Furthermore, the control quantity optimal solution calculation module 4 specifically includes a first control quantity optimal solution calculation unit and a second control quantity optimal solution calculation unit that are electrically or signal-connected sequentially; the first control quantity optimal solution calculation unit is electrically or signal-connected to the fourth feedforward compensation signal generation unit, and the second control quantity optimal solution calculation unit is... The first control variable optimal solution calculation unit is used to construct the leaf-moistening machine control model based on the moisture dynamic model, moisture content estimate, and feedforward compensation signal, and to define the optimization strategy of the leaf-moistening machine control model according to equation (11): (11).

[0123] Where H is the prediction step size, The input weight matrix, For the control quantity weight matrix, For the control quantity increment weight matrix, Let H be the sum of the second norms of the weighted state vector Q from 0 to H. Let H be the summation of the second norm of the weighted control quantity R from 0 to H-1. Let S be the second norm of the weighted control increment.

[0124] The second control quantity optimal solution calculation unit is used to solve equation (11) to obtain the optimal solution of the control quantity. ,in, To achieve the optimal water addition, To achieve the optimal amount of steam added, To achieve the optimal hot air volume, This is the optimal tidal discharge rate.

[0125] Furthermore, the moisture control device also includes an optimized key parameter data acquisition module and a closed-loop control module that are electrically or signal-connected in sequence; the optimized key parameter data acquisition module is electrically or signal-connected to the control quantity optimal solution allocation module 5.

[0126] Among them, the optimized key parameter data acquisition module is used to collect the optimized key parameter data of tobacco leaves in real time through sensors at the inlet and outlet; the closed-loop control module is used to take the optimized key parameter data as the main body of execution, input the moisture content estimation value acquisition module 2, and repeatedly execute the moisture content estimation value acquisition module 2 to the control quantity optimal solution allocation module 5 to form closed-loop control.

[0127] Furthermore, the moisture control device also includes a control quantity optimal solution sending module that is electrically or signalally connected to the control quantity optimal solution allocation module 5. This module is used to send the control quantity optimal solution to an external monitoring terminal.

[0128] It should be noted that this embodiment is a functional module embodiment based on the above method embodiment. For the preferred, extended, limited, exemplified, and principle-explained parts of this embodiment, please refer to the above embodiments. This embodiment will not repeat them here.

[0129] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Figure 3 As shown, the electronic device 6 includes a processor 61 and a memory 62 coupled to the processor 61.

[0130] The memory 62 stores program instructions for implementing the federated learning-based collaborative energy-saving method for government data clusters in any of the above embodiments.

[0131] The processor 61 is used to execute program instructions stored in the memory 62 for collaborative energy saving of government data clusters based on federated learning.

[0132] The processor 61 can also be referred to as a CPU (Central Processing Unit). The processor 61 may be an integrated circuit chip with signal processing capabilities. The processor 61 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0133] Furthermore, Figure 4 This is a schematic diagram of the structure of a storage medium according to an embodiment of this application. See also: Figure 4 In this embodiment of the application, the storage medium 7 stores program instructions 71 capable of implementing all the above methods. These program instructions 71 can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in each embodiment of the application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0134] In the several embodiments provided in this application, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, signal, or other forms.

[0135] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for controlling the moisture content of a leaf-curing machine, wherein sensors for collecting tobacco leaf data are installed at both the inlet and outlet of the leaf-curing machine, characterized in that, The moisture control method includes: Step S1: Collect key parameter data of tobacco leaves in real time through sensors at the inlet and outlet, and construct a dynamic moisture model of the leaf humidifier based on the key parameter data and the physical property parameters of the tobacco leaves. Step S2: Based on the moisture dynamic model, construct an improved volumetric Kalman filter estimation model, and obtain the estimated moisture content of tobacco leaves during the wetting process through the improved volumetric Kalman filter estimation model; Step S3: Perturb the estimated moisture content using a neural network based on an attention mechanism, and generate a feedforward compensation signal based on a preset target moisture content. Step S4: Construct a leaf-moistening machine control model based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, and calculate the optimal solution of the control quantity of the leaf-moistening machine control model; Step S5: The optimal solution of the control quantity is allocated to the execution component of the leaf-wetting machine and the optimal solution of the control quantity is executed by the execution component.

2. The moisture control method according to claim 1, characterized in that, Step S1 involves collecting key parameter data of the tobacco leaves in real time using sensors at the inlet and outlet, and constructing a dynamic moisture model of the leaf-humidifying machine based on the key parameter data and the physical property parameters of the tobacco leaves, including: Step S11: Real-time collection of the inlet moisture content of the tobacco leaves based on the moisture content sensor, temperature sensor, humidity sensor, and flow sensor at the inlet and outlet. Inlet temperature Ambient humidity Instantaneous flow and export moisture content outlet temperature ; Step S12, based on the inlet moisture content The inlet temperature The ambient humidity The instantaneous flow rate The aforementioned outlet moisture content The outlet temperature Based on the physical properties of the tobacco leaves, a dynamic moisture model of the leaf-moistening machine is constructed using equation (1): （1）； in, Based on tobacco leaf moisture content and tobacco leaf temperature The state vector; It is the transpose matrix; The first derivative of the state vector; To control the input vector, including the amount of water added. Steam addition amount Hot air volume and tidal volume ; The disturbance vector includes the ambient humidity. and instantaneous flow ; This is the measurement output of the aforementioned moisture dynamics model. For process noise, It is observation noise; It is a nonlinear dynamic function; For observation functions.

3. The moisture control method according to claim 1, characterized in that, Step S2, constructing an improved volumetric Kalman filter estimation model based on the moisture dynamic model, and obtaining the estimated moisture content of tobacco leaves during the leaf wetting process through the improved volumetric Kalman filter estimation model, including: Step S21, the water dynamics model is discretized based on equation (2): （2）； in, Let k be the state vector of the system at time k. Let be the state vector of the system at time k+1. This is the control input for the system at time k. Let k be the process noise vector of the system at time k. Let K be the observation noise vector of the system at time k. This is the observation vector of the system at time k+1; Step S22, define the material moisture absorption coefficient during the leaf wetting process according to equation (3): （3）； in, The system parameters are unknown at time k. Let be the extended state vector of the system at time k; Step S23, according to equation (4), the extended state vector is extended into a cubic sampling point state vector: （4）； in, Let be the cubic sampling point vector. Let be the square root of the covariance of the system at time k. Let be the state vector of the i-th cubic sampling point at time k in the system; Step S24, predict the propagation result of each cubic sampling point using equation (5): （5）； in, Let i be the weight of the i-th cubic sampling point. Let be the predicted state vector of the i-th cubic sampling point at time k+1 of the system; Let the square root of the prediction covariance at time k+1 be the system value. Decomposition for QR Let be the predicted state vector of the system at time k+1; Step S25: Define the observation noise using equation (6), and adjust the observation noise in real time using adaptive noise estimation: （6）； in, Here is the estimated observation noise value of the system at time k. This is the estimated observation noise value for the system at time k-1. This is the estimated value of the observation noise covariance at time k. This is the estimated value of the observation noise covariance at time k-1. Let k be the updated weight coefficients of the system at time k. Represents the innovation vector. This represents the transpose of the innovation vector. Let be the observation vector of the system at time k. This represents the observation vector of the prediction system at time k. This represents the transposed observation vector of the prediction system at time k. This represents the observation vector of the i-th sampling point at time k in the system. Let represent the transposed observation vector of the i-th sampling point at time k in the system; Step S26, update the observation vector using equation (7); （7）； in, Let be the observation vector of the i-th sampling point at time k+1 of the system. Let i be the i-th cubic sampling point at time k+1 of the system. Let be the predicted observation vector of the system at time k+1. Let be the square root of the covariance of the system at time k+1. Let the cross-covariance of the state observations at time k+1 be the system's cross-covariance. To observe the square root of the noise covariance, Let be the state vector of the system at time k+1. Let be the transposed observation vector of the system at time k+1; Step S27, combining equations (5), (6), and (7) yields the final state update expression (8): （8）； in, Let K+1 be the Kalman gain matrix of the system. Let be the state vector of the system at time k+1. Let be the predicted observation vector of the system at time k+1; defined according to equation (1). The Including the estimated moisture content and temperature estimates .

4. The moisture control method according to claim 1, characterized in that, Step S3 involves perturbating and predicting the estimated moisture content using a neural network based on an attention mechanism, and generating a feedforward compensation signal based on a preset target moisture content, including: Step S31: Obtain historical key parameter data and construct a multimodal time series. ,in, From arrive Estimated sequence of tobacco leaf moisture content at time points; From arrive Ambient humidity sequence over time; From arrive Ambient temperature sequence at any given time; From arrive The sequence of tobacco leaf flow rates at any given time; N is the historical step size; Step S32: Convert all the values ​​in the multimodal time series into a high-dimensional vector using equation (8). : （8）； in, Let be the weight matrix of the multimodal time series; It is the bias vector; Step S33, define feedforward control according to equation (9): （9）； in, For feedforward output, As the input to the high-dimensional vector, The ReLU activation function is used. , ; Step S34: Perturbation prediction is performed on the estimated water content using a neural network based on an attention mechanism, and the feedforward compensation signal is generated according to equation (10) based on the predicted perturbation and the target water content: （10）； in, Forward compensation vector, The inverse function of the model. For the target moisture content, To estimate the moisture content, This is the integral of the predicted disturbance value.

5. The moisture control method according to claim 1, characterized in that, Step S4: Based on the moisture dynamic model, the estimated moisture content, and the feedforward compensation signal, construct a leaf-moistening machine control model, and calculate the optimal solution of the control quantity for the leaf-moistening machine control model, including: Step S41: Based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, construct a leaf-moistening machine control model, and define the optimization strategy of the leaf-moistening machine control model according to equation (11): （11）； Where H is the prediction step size, The input weight matrix, For the control quantity weight matrix, For the control quantity increment weight matrix, Let H be the sum of the second norms of the weighted state vector Q from 0 to H. Let H be the summation of the second norm of the weighted control quantity R from 0 to H-1. Let S be the second norm of the weighted control increment; Step S42, solve equation (11) to obtain the optimal solution of the control quantity. ,in, To achieve the optimal water addition, To achieve the optimal amount of steam added, To achieve the optimal hot air volume, This is the optimal tidal discharge rate.

6. The moisture control method according to claim 1, characterized in that, Step S5 involves allocating the optimal solution of the control quantity to the execution component of the leaf-wetting machine and executing the optimal solution of the control quantity through the execution component. Following this, the process includes: Step S10: Real-time acquisition of optimized key parameter data of tobacco leaves through sensors at the inlet and outlet. Step S20: The optimized key parameter data is used as the execution subject input to step S2, and steps S2 to S5 are repeated to form a closed-loop control.

7. The moisture control method according to claim 1, characterized in that, Step S5 involves allocating the optimal solution of the control quantity to the execution component of the leaf-wetting machine and executing the optimal solution of the control quantity through the execution component. Following this, the process includes: Step S100: Send the optimal solution of the control quantity to the external monitoring terminal.

8. A moisture control device for a leaf-conditioning machine, wherein the moisture control device is applied to the moisture control method as described in any one of claims 1 to 7, characterized in that, The moisture control device includes: The moisture dynamic model construction module is used to collect key parameter data of tobacco leaves in real time through sensors at the inlet and outlet, and construct the moisture dynamic model of the leaf humidifier based on the key parameter data and the physical characteristic parameters of the tobacco leaves. The moisture content estimation module is used to construct an improved volumetric Kalman filter estimation model based on the moisture dynamic model, and obtain the moisture content estimate of tobacco leaves during the leaf wetting process through the improved volumetric Kalman filter estimation model. The feedforward compensation signal generation module is used to perform perturbation prediction on the moisture content estimate through a neural network based on an attention mechanism, and generate a feedforward compensation signal based on a preset target moisture content. The optimal solution calculation module for control quantity is used to construct a leaf-wetting machine control model based on the moisture dynamic model, the moisture content estimate, and the feedforward compensation signal, and to calculate the optimal solution for control quantity of the leaf-wetting machine control model. The optimal solution allocation module for control quantities is used to allocate the optimal solution for control quantities to the execution component of the leaf-wetting machine and execute the optimal solution for control quantities through the execution component.

9. An electronic device, characterized in that, The method includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the moisture control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed by a processor, enable the moisture control method as described in any one of claims 1 to 7.

Images