Self-learning cut tobacco drying material temperature control method
Through the self-learning wire drying material temperature control method, a closed-loop control loop is built using offline historical data and real-time data self-learning strategy, combining proportion-integrated control and iterative self-optimization algorithm, the traditional PID control model is solved in the local area of the wire dryer. The optimal solution problem improves the stability and robustness of temperature control and is suitable for the refined needs of tobacco silk making process.
Patent Information
- Application Number
- CN202411094098.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-15
AI Technical Summary
The traditional PID control model relies too much on manual optimization parameters or system parameter setting in the material temperature control of the wire dryer, which is easy to fall into the dilemma of local optimal solutions, making it difficult to maintain actual results and difficult to deal with nonlinearity in the wire drying process. change and large lag.
Using self-learning wire drying material temperature control method, the off-line historical data is preprocessed and real-time data self-learning strategy is constructed, and a closed-loop control loop is built, combining proportion-integral control and iterative self-optimization algorithm to reduce overshoot and steady state Error, improve control stability and robustness.
It realizes that on the basis of identifying the actual controlled object model, it reduces the overshoot and steady-state error, shortens the adjustment time, and improves the stability, convergence and robustness of material temperature control of the wire making and dryer, which is suitable for tobacco. Fine control requirements in the silk making process.
Smart Images

Figure CN118947948A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of cigarette cutting technology, and more specifically, to a self-learning method for controlling the temperature of cutting material. Background Art
[0002] The working principle commonly used by tobacco cutting and drying machines is: a certain flow of steam passes through the inner wall of the cylinder thin plate, and the steam condenses in the steam channel to effectively transfer the heat to the thin plate and the tobacco. At the same time, the fan sends the ambient air into the hot air generated by the steam-heated heat exchanger. During the flow through the drum, the heat is transferred to the tobacco and the moisture is taken away by convection, so that the tobacco can be evenly dried, heated, and the filling force is evenly increased, as well as the outlet moisture and moisture. At the same time, a dehumidification system is set inside the cylinder to extract part of the water-containing and hot exhaust gas for discharge to keep the moisture content of the material constant and indirectly affect the moisture content of the material.
[0003] Since the drying process is highly nonlinear, uncertain and has a large hysteresis, coupled with the special properties of tobacco leaves, the temperature control of the material in the drying process becomes very complicated. In essence, the moisture and temperature control of the material in the drying machine belong to a nonlinear time-varying large hysteresis system, such as Figure 1 As shown in the figure, the mainstream equipment manufacturers currently use the cascade PID control method to control the moisture at the outlet of the tobacco drying machine, while the material temperature is manually adjusted by the open-loop method, which is a non-important quality feedback control. However, with the increasing quality requirements of tobacco production, the refined control of the temperature of tobacco drying materials has gradually begun to be included in the scope of production quality assessment.
[0004] Therefore, although the traditional cascade PID control model can partially solve the problems of export material temperature control lag, large fluctuations and instability, the traditional PID control model relies too much on manual optimization of parameters or system parameter adjustment, and is prone to fall into the dilemma of local optimal solution, making it difficult to maintain the actual effect of the traditional PID. Summary of the invention
[0005] One object of the present invention is to provide a self-learning method for controlling the temperature of silk drying materials to solve the aforementioned problem of over-reliance on manual optimization of parameters or system parameter adjustment, which is prone to falling into the dilemma of local optimal solutions. The control method of this method can reduce overshoot and steady-state error, shorten the adjustment time, and thus improve the stability, convergence and robustness of the silk drying machine material temperature control model based on the identification of the actual controlled object model.
[0006] According to a first aspect of the present invention, a self-learning method for controlling the temperature of dried tofu material is provided, comprising:
[0007] Step 101, preprocessing the acquired offline historical data, wherein the offline historical data includes time series, film valve opening data, and outlet material moisture data, and fitting the processed data to obtain a mathematical model for drum wall heating and drying the material;
[0008] Step 102, using proportional-integral control to set the moisture control loop of the front stage, and building a cylinder wall temperature control model of the back stage based on the real-time data self-learning strategy;
[0009] Step 103, a closed-loop control loop is constructed through a mathematical model of the drum wall heating and drying material, a front-stage moisture control loop and a back-stage drum wall temperature control model, and the temperature of the dried silk material is controlled in combination with real-time data.
[0010] Optionally, in step 101, preprocessing the offline historical data includes: denoising the offline historical data using an arithmetic mean filtering method with a fixed time window length.
[0011] Optionally, the formula for denoising offline historical data is as follows: (Formula 1); (Formula 2);
[0012] in, is the arithmetic mean filtering value of the cylinder wall temperature regulating film valve opening value, is the length of the fixed time window, The length of the data collection period to obtain offline historical data; The actual value of the film valve opening is adjusted for the i-th cylinder wall temperature within a fixed time window; is the arithmetic mean filtered value of the moisture content of the export material. is the actual moisture value of the i-th outlet material within a fixed time window.
[0013] Optionally, based on the preprocessed offline historical data, the transfer function is fitted into a second-order inertia lag model using system identification technology, where As the actuator output value, As the object acquisition value, the fitting data of the transfer function of the controlled object is performed. The transfer function expression of the controlled object fitting into the second-order lag model is: (Formula 3);
[0014] in, is the transfer function, s is the Laplace operator, K is the static amplification factor, T1 and T2 are the second-order inertia link coefficients, and the constant τ is the lag time. and The parameters K, T1, T2 and τ of the transfer function in Formula 3 are fitted by offline data.
[0015] Optionally, in step 102, the moisture control loop of the previous stage adopts proportional-integral control, and its control formula is: (Formula 4);
[0016] Among them, K p , K i are proportional and integral parameters; SP1(k) is the set value of outlet material moisture; PV1(k) is the actual value of outlet material moisture; u1(k) is the output value of the previous stage moisture control; e1(k) is the deviation between the set value SP1(k) of the previous stage moisture control and the process value PV1(k).
[0017] Optionally, the step 102 of constructing a subsequent stage cylinder wall temperature control model based on a real-time data self-learning strategy specifically includes:
[0018] Step S102-1, transform the transfer function of equation 3 into Perform inverse Laplace transform and convert to time domain differential equation. The state equation obtained by fitting is: (Formula 5);
[0019] In formula 5, is the system state at the next moment, is the system status at the current moment, is the control input, is the interference input, is the system output, h is the state input Mapped to the output in the form of transfer function G(s) ;
[0020] Step S102-2, in each data acquisition time period for obtaining offline historical data, a control input is generated according to the current system state and the parameters of the real-time data self-learning controller, wherein the control input is the output of the real-time data self-learning controller, and its expression is: (Formula 6);
[0021] is the output value of the real-time data self-learning controller, is the adjustment parameter of the control input at the current k moment, is the initial control output value, is the control increment adjusted by the real-time data self-learning strategy. The k and k-1 in brackets represent different moments.
[0022] Then the control output value Acting on the controlled object representing the system transfer function of equation 3, the actual output value z(k) is obtained, and finally the actual output z(k) and the predicted output z are calculated. ref The error calculation formula is: (Formula 7);
[0023] According to the deviation value, the parameters of the real-time data self-learning controller are automatically and continuously updated, and the automatic optimization update formula is: (Formula 8);
[0024] In formula 8, Control input adjustment parameters for the current k moment, is the control input adjustment parameter at time k-1, is the learning rate of the self-learning strategy based on real-time data at the current k moment, is the deviation between the expected output and the actual output at the current k moment, that is, , It is the control increment adjusted based on the real-time data self-learning strategy; is the control increment adjustment parameter at the current k moment; is the control increment adjustment parameter at time k-1.
[0025] Optionally, the step 102 of constructing a subsequent stage drum wall temperature control model based on a real-time data self-learning strategy further includes:
[0026] Step S102-3, construct an exponential adaptive learning rate strategy, and the learning rate update formula is as follows: (Formula 9);
[0027] In formula 9, is the learning rate of the self-learning strategy based on real-time data at the current k moment, is the learning rate of the real-time data self-learning strategy at the k-1 moment, is the deviation between the expected output and the actual output at the current k moment, To adaptively control the learning rate adjustment parameter, it is designed to be a fixed value and serves as a hyper parameter of the entire algorithm.
[0028] The self-learning drying material temperature control method disclosed in the present invention has the following technical effects:
[0029] The present invention uses an arithmetic mean filtering method with a fixed time window length to denoise offline historical data, which can reduce noise and unnecessary fluctuations in the data and fit the corresponding mathematical model for subsequent control of the control system. The present invention fits the controlled object model for the moisture control of the tofu drying machine material into a second-order lag model, which is more in line with the complex mechanism of the controlled object in the steady state, and reduces the complexity of the model and the amount of self-optimization calculation, while ensuring a certain control accuracy and reducing the hardware computing resources and time overhead for finding the optimal solution.
[0030] The present invention adopts proportional-integral control to set the moisture control loop of the front stage, and constructs the drum wall temperature control model of the back stage based on the real-time data self-learning strategy. Compared with the traditional two-stage cascade PID control of the tofu drying machine, it is more capable of handling fuzzy, complex and uncertain large hysteresis systems, and has better control stability and accuracy. At the same time, it has good dynamic performance, and can better balance timely response and non-mutation control when the system changes rapidly, and has good response characteristics, robustness and adaptability.
[0031] The present invention utilizes data iteration technology to combine the control model with iterative calculation, realizes moisture control of the dryer outlet material under non-fixed model control, realizes fixed compensation and self-optimizing compensation on the feedback channel, is more in line with actual control scenarios, and has the dual control characteristics of non-fixed model self-optimization and fixed model rapid temperature.
[0032] The iterative self-optimization algorithm of the present invention, due to the use of the adaptive learning rate strategy in the exponential form of Formula 9, can comprehensively evaluate dynamic performance and steady-state accuracy, and basically eliminates overshoot in terms of characteristics. It is particularly suitable for the quality requirement of minimizing unqualified "material heads" in the tobacco silk drying machine process, improving the tobacco process quality while also increasing the economic benefit value of the present invention.
[0033] The parameter self-optimization method of the present invention adopts the form of multiple equations in series of formula 8, and realizes the function of the algorithm approaching the optimal solution at a faster rate. As the control process enters the stable optimization stage, the area near the optimal solution is accurately searched. Since the change of the self-optimization learning rate is affected by the system error, the error of the present invention is reduced at a faster rate in the early stage of self-optimization and then gradually tends to a stable change, indicating that the strategy can reduce the system error at a faster rate, speed up the adjustment speed, and reduce the time required to modify the control input to make the system reach stability again due to external interference input in actual engineering.
[0034] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
[0036] Figure 1 A schematic flow chart of a self-learning method for controlling the temperature of dried tofu material provided by an embodiment of the present invention;
[0037] Figure 2 A schematic diagram of a mathematical model for establishing a drum wall heating and drying material provided by an embodiment of the present invention;
[0038] Figure 3 A schematic diagram of a cylinder wall temperature control model based on real-time data self-learning provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0039] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.
[0040] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.
[0041] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.
[0042] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0043] Due to the strong nonlinearity, uncertainty and large hysteresis of the silk drying process, coupled with the special properties of tobacco leaves themselves, the material temperature control of the silk drying process becomes very complicated. At present, with the increasing quality requirements of tobacco production, the refined control of the temperature of silk drying materials has gradually begun to be included in the scope of production quality assessment. Although the traditional cascade PID control model can partially solve the problems of hysteresis, large fluctuation and instability of the outlet material temperature control, the traditional PID control model relies too much on artificial optimization parameters or system parameter setting, which is easy to fall into the dilemma of local optimal solution, making it difficult to maintain the actual effect of traditional PID. This application is based on the combination of transfer function model and real-time data self-learning technology, and solves the nonlinear control problem of silk drying machine temperature based on data iteration algorithm, and continuously corrects and amends the model based on real-time data, which basically solves the problem of control parameters falling into local optimal. At the same time, this method can reduce overshoot and steady-state error and shorten the adjustment time on the basis of identifying the actual controlled object model, thereby improving the stability, convergence and robustness of the material temperature control model of the silk drying machine.
[0044] The present invention proposes an embodiment of a self-learning method for controlling the temperature of dried silk material. Specifically, Figure 1 shown, including:
[0045] Step 101, pre-process the acquired offline historical data, the offline historical data includes time series, film valve opening data, and outlet material moisture data, and obtain a mathematical model of drum wall heating and drying materials based on the processed data fitting; wherein, the time series records the state of each time point in the drying process, the film valve opening data reflects the adjustment of the heating system on the material heating intensity, and the outlet material moisture data is an important indicator for measuring the drying effect; after cleaning and sorting the offline historical data, a mathematical relationship model between drum wall heating and material drying effect is established, such as Figure 2 shown.
[0046] Step 102, using proportional-integral control to set the moisture control loop of the front stage, and constructing the drum wall temperature control model of the back stage based on the real-time data self-learning strategy; wherein, through the PID controller, according to the deviation between the moisture of the outlet material detected in real time and the set value, the opening of the membrane valve is adjusted to achieve direct control of the moisture; using the real-time collected data, combined with the self-learning algorithm, the drum wall temperature control strategy is continuously adjusted to adapt to the changes in different batches of materials or process conditions. The self-learning mechanism enables the control model to be continuously optimized with the accumulation of time and experience.
[0047] Step 103, a closed-loop control loop is constructed by using a mathematical model of the drum wall heating and drying material, a moisture control loop at the front stage, and a drum wall temperature control model at the back stage, such as Figure 3As shown in the figure, the temperature of the dried silk material is controlled in combination with real-time data. Specifically, the system detects key parameters such as the moisture content and drum wall temperature of the outlet material in real time, directly adjusts the film valve opening through the moisture control loop of the front stage, and adjusts the heating strategy through the drum wall temperature control model of the back stage to achieve precise control of the temperature of the dried silk material. Through the continuous adjustment and optimization of the closed-loop control loop, the system can automatically adapt to various changing factors to ensure the stability and reliability of the silk drying process.
[0048] In an embodiment of the present invention, in step 101, preprocessing the offline historical data includes: denoising the offline historical data using an arithmetic mean filtering method with a fixed time window length to reduce random noise and unnecessary fluctuations in the data while retaining the main features of the data, thereby obtaining a smoother data sequence.
[0049] In the embodiment of the present invention, the formula for denoising offline historical data is as follows: (Formula 1); (Formula 2);
[0050] in, is the arithmetic mean filtering value of the cylinder wall temperature regulating film valve opening value, is the length of the fixed time window, The length of the data collection period to obtain offline historical data; The actual value of the film valve opening is adjusted for the i-th cylinder wall temperature within a fixed time window; is the arithmetic mean filtered value of the moisture content of the export material. is the actual moisture value of the i-th outlet material within the fixed time window.
[0051] In the embodiment of the present invention, based on the preprocessed offline historical data, the transfer function is fitted into a second-order inertia lag model using system identification technology, wherein As the actuator output value, As the object acquisition value, the fitting data of the transfer function of the controlled object is performed. The transfer function expression of the controlled object fitting into the second-order lag model is: (Formula 3);
[0052] in, is the transfer function, s is the Laplace operator, K is the static amplification factor, T1 and T2 are the second-order inertia link coefficients, and the constant τ is the lag time. and The parameters K, T1, T2 and τ of the transfer function in Formula 3 are fitted by offline data. The transfer function parameters of the second-order inertia lag model of the controlled object are fitted by offline historical data, and the error between the model output and the actual data is minimized by using system identification technology.
[0053] In the embodiment of the present invention, in step 102, the moisture control loop of the previous stage adopts proportional-integral control, and its control formula is: (Formula 4);
[0054] Among them, K p , K i are proportional and integral parameters; SP1(k) is the set value of the outlet material moisture; PV1(k) is the actual value of the outlet material moisture; u1(k) is the output value of the previous stage moisture control; e1(k) is the deviation between the set value SP1(k) of the previous stage moisture control and the process value PV1(k). Among them, the proportional and integral control work together to enable the system to respond to deviations quickly and ultimately eliminate steady-state errors. By adjusting K p , K i The value of can optimize the control performance and make the system achieve the desired response speed and stability. By real-time monitoring the actual value PV1(k) of the moisture content of the outlet material and comparing it with the set value SP1(k), the deviation e1(k) is calculated, and then the control output u1(k) is calculated using the PID control algorithm to adjust the control actuator (such as adjusting the opening of the valve), thereby achieving precise control of the moisture content of the outlet material.
[0055] In the embodiment of the present invention, the construction of the subsequent stage cylinder wall temperature control model based on the real-time data self-learning strategy in step 102 specifically includes:
[0056] Step S102-1, transform the transfer function of equation 3 into Perform inverse Laplace transform and convert to time domain differential equation for analysis. The state equation obtained by fitting is: (Formula 5);
[0057] In formula 5, is the system state at the next moment, is the system status at the current moment, is the control input, is the interference input, is the system output, h is the state input Mapped to the output in the form of transfer function G(s) ;
[0058] Step S102-2, in each data acquisition time period for obtaining offline historical data, a control input is generated according to the current system state and the parameters of the real-time data self-learning controller, wherein the control input is the output of the real-time data self-learning controller, and its expression is: (Formula 6);
[0059] is the output value of the real-time data self-learning controller, is the adjustment parameter of the control input at the current k moment, is the initial control output value, is the control increment adjusted by the real-time data self-learning strategy, and k and k-1 in brackets represent different moments; wherein, the real-time data self-learning controller is a controller that can automatically adjust its parameters and behaviors according to real-time data. This controller combines a control algorithm (such as PID control) with a machine learning or optimization algorithm. In each data acquisition cycle, the controller uses the latest data to update its internal model or parameters in order to more accurately predict system behavior and generate more appropriate control inputs. By combining the initial control output value, the adjustment parameters and the control increment adjusted by the real-time data self-learning strategy, the system can more flexibly respond to changing environments and conditions, thereby achieving more accurate and efficient control. The present invention combines the front-stage PID control with the back-stage self-learning technology based on real-time data to realize fixed compensation and self-optimization compensation on the feedback channel, which is more in line with the actual control reality scene, and at the same time has the dual control characteristics of non-fixed model self-optimization and fixed model fast temperature.
[0060] Then the control output value Acting on the controlled object representing the system transfer function of equation 3, the actual output value z(k) is obtained, and finally the actual output z(k) and the predicted output z are calculated. ref The error calculation formula is: (Formula 7);
[0061] According to the deviation value, the parameters of the real-time data self-learning controller are automatically and continuously updated, and the automatic optimization update formula is: (Formula 8);
[0062] In formula 8, Control input adjustment parameters for the current k moment, is the control input adjustment parameter at time k-1, is the learning rate of the self-learning strategy based on real-time data at the current k moment, is the deviation between the expected output and the actual output at the current k moment, that is, , It is the control increment adjusted based on the real-time data self-learning strategy; is the control increment adjustment parameter at the current k moment; is the control increment adjustment parameter at time k-1. The algorithm is implemented to approach the optimal solution at a faster rate by adopting the form of multiple equations in series of formula 8. As the control process enters the stable optimization stage, the area near the optimal solution is accurately searched. Since the change of the self-optimization learning rate is affected by the system error, the error of the present invention decreases at a faster rate in the early stage of self-optimization and then gradually tends to a stable change, indicating that the strategy can reduce the system error at a faster rate, speed up the adjustment speed, and reduce the time required to modify the control input to make the system reach stability again due to external interference input in actual engineering.
[0063] In the embodiment of the present invention, the step 102 of constructing a subsequent stage cylinder wall temperature control model based on the real-time data self-learning strategy also includes:
[0064] Step S102-3, construct an exponential adaptive learning rate strategy, and the learning rate update formula is as follows: (Formula 9);
[0065] In formula 9, is the learning rate of the self-learning strategy based on real-time data at the current k moment, is the learning rate of the real-time data self-learning strategy at the k-1 moment, is the deviation between the expected output and the actual output at the current k moment, The learning rate adjustment parameter is designed to be a fixed value for adaptive control and is used as a super parameter of the entire algorithm. The adaptive learning rate strategy in the exponential form of formula 9 can comprehensively evaluate the dynamic performance and steady-state accuracy, and basically eliminates the overshoot in terms of characteristics. It is particularly suitable for the quality requirements of reducing unqualified "material heads" as much as possible in the tobacco silk drying machine process, improving the tobacco process quality while also increasing the economic value of the invention. The learning rate will be dynamically adjusted according to the error at the current moment (i.e., the deviation between the expected output and the actual output). When the error is large, the learning rate can be appropriately increased to speed up the learning speed; when the error is small, the learning rate is reduced to adjust the parameters more finely to avoid overfitting.
[0066] Specifically, Figure 3 The figure shows a schematic diagram of building a cylinder wall temperature control model based on real-time data self-learning, where: is a transfer function, which indicates how the drum wall temperature should change with time and system dynamics. It serves as the input of the drum wall temperature controller and guides the controller on how to adjust the drum wall temperature to achieve the ideal drying effect. The moisture controller adopts the front-stage PID control strategy. According to the deviation between the set outlet moisture value SP1 and the actual outlet moisture value PV1, it calculates and outputs the control signal u1, which can effectively reduce the error and respond quickly to system changes. The drum wall temperature controller not only accepts the drum wall temperature set value Φ(s) as input, but also adjusts the output u1 in combination with the real-time data self-learning model. This control method based on real-time data self-learning enables the system to adapt to environmental changes and improve the accuracy and robustness of control. The drying process is the core link, in which the material is dried under the heating of the drum wall. The actual moisture content of the outlet material is the target of system control. By comparing the actual outlet moisture value PV1 with the set outlet moisture value SP1, the control effect can be evaluated and the controller parameters can be adjusted as a feedback signal. CV1 is an analog signal used to monitor and feedback the moisture content of the outlet material in real time. CV1 is the feedback of the actual outlet and is compared with the set value to form a closed-loop control. Through the combination of front-stage PID control and back-stage real-time data self-learning model, external disturbances can be effectively responded to, ensuring that the moisture content of the outlet material during the drying process remains at a predetermined level.
[0067] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.
Claims
1. A self-learning method for controlling the temperature of dried silk material, characterized in that: include: Step 101, preprocessing the acquired offline historical data, wherein the offline historical data includes time series, film valve opening data, and outlet material moisture data, and fitting the processed data to obtain a mathematical model for drum wall heating and drying the material; Step 102, using proportional-integral control to set the moisture control loop of the front stage, and building a cylinder wall temperature control model of the back stage based on the real-time data self-learning strategy; Step 103, a closed-loop control loop is constructed through a mathematical model of the drum wall heating and drying material, a front-stage moisture control loop and a back-stage drum wall temperature control model, and the temperature of the dried silk material is controlled in combination with real-time data.
2. The self-learning method for controlling the temperature of dried tofu material according to claim 1, characterized in that: In step 101, preprocessing the offline historical data includes: using an arithmetic mean filtering method with a fixed time window length to denoise the offline historical data.
3. The self-learning method for controlling the temperature of dried tofu material according to claim 2, characterized in that: The formula for denoising offline historical data is as follows: (Formula 1); (Formula 2); in, is the arithmetic mean filtering value of the cylinder wall temperature regulating film valve opening value, is the length of the fixed time window, The length of the data collection period to obtain offline historical data; The actual value of the film valve opening is adjusted for the i-th cylinder wall temperature within a fixed time window; is the arithmetic mean filtered value of the moisture content of the export material. is the actual moisture value of the i-th outlet material within the fixed time window.
4. The self-learning method for controlling the temperature of dried tofu material according to claim 3, characterized in that: Based on the preprocessed offline historical data, the transfer function is fitted into a second-order inertia lag model using system identification technology, where As the actuator output value, As the object acquisition value, the fitting data of the transfer function of the controlled object is performed. The transfer function expression of the controlled object fitting into the second-order lag model is: (Formula 3); in, is the transfer function, s is the Laplace operator, K is the static amplification factor, T1 and T2 are the second-order inertia link coefficients, and the constant τ is the lag time. and The parameters K, T1, T2 and τ of the transfer function in Formula 3 are fitted by offline data.
5. The self-learning method for controlling the temperature of dried tofu material according to claim 1, characterized in that: In step 102, the moisture control loop of the previous stage adopts proportional-integral control, and its control formula is: (Formula 4); Among them, K p , K i are proportional and integral parameters; SP1(k) is the set value of outlet material moisture; PV1(k) is the actual value of outlet material moisture; u1(k) is the output value of the previous stage moisture control; e1(k) is the deviation between the set value SP1(k) of the previous stage moisture control and the process value PV1(k).
6. The self-learning method for controlling the temperature of dried tofu material according to claim 4, characterized in that: The step 102 of constructing a subsequent stage cylinder wall temperature control model based on the real-time data self-learning strategy specifically includes: Step S102-1, transform the transfer function of equation 3 into Perform inverse Laplace transform and convert to time domain differential equation. The state equation obtained by fitting is: (Formula 5); In formula 5, is the system state at the next moment, is the system status at the current moment, is the control input, is the interference input, is the system output, h is the state input Mapped to the output in the form of transfer function G(s) ; Step S102-2, in each data acquisition time period for obtaining offline historical data, a control input is generated according to the current system state and the parameters of the real-time data self-learning controller, wherein the control input is the output of the real-time data self-learning controller, and its expression is: (Formula 6); is the output value of the real-time data self-learning controller, is the adjustment parameter of the control input at the current k moment, is the initial control output value, is the control increment adjusted by the real-time data self-learning strategy. The k and k-1 in brackets represent different moments. Then the control output value Acting on the controlled object representing the system transfer function of equation 3, the actual output value z(k) is obtained, and finally the actual output z(k) and the predicted output z are calculated. ref The error calculation formula is: (Formula 7); According to the deviation value, the parameters of the real-time data self-learning controller are automatically and continuously updated, and the automatic optimization update formula is: (Formula 8); In formula 8, Control input adjustment parameters for the current k moment, is the control input adjustment parameter at time k-1, is the learning rate of the self-learning strategy based on real-time data at the current k moment, is the deviation between the expected output and the actual output at the current k moment, that is, , It is the control increment adjusted based on the real-time data self-learning strategy; is the control increment adjustment parameter at the current k moment; is the control increment adjustment parameter at time k-1.
7. The self-learning method for controlling the temperature of dried tofu material according to claim 6, characterized in that: The step 102 of constructing a subsequent stage cylinder wall temperature control model based on the real-time data self-learning strategy also includes: Step S102-3, construct an exponential adaptive learning rate strategy, and the learning rate update formula is as follows: (Formula 9); In formula 9, is the learning rate of the self-learning strategy based on real-time data at the current k moment, is the learning rate of the real-time data self-learning strategy at the k-1 moment, is the deviation between the expected output and the actual output at the current k moment, To adaptively control the learning rate adjustment parameter, it is designed to be a fixed value and serves as a hyper parameter of the entire algorithm.
Citation Information
Patent Citations
Reaction kettle control method based on robust adaptive iterative learning
CN117234087A
Fuzzy algorithm-based cut tobacco dryer outlet moisture control method and device
CN118203142A
State based adaptive feedback feed forward PID controller
CN1540461A
Cited By
Cigarette heating production line glycerol applying method based on viscosity control
CN119344499A
