Stevioside production system based on dynamic weight adaptive MPC algorithm

Abstract

The invention relates to the technical field of stevioside preparation, and discloses a stevioside production system based on a dynamic weight adaptive MPC algorithm, which comprises a raw material pretreatment module, a solution extraction module, a purification module, a decolorization and post-treatment module, a crystallization and drying module, a central control module and an acquisition unit, the raw material pretreatment module is used for performing crushing and enzymolysis on stevia rebaudian and removing impurities to obtain enzymatic hydrolysate, the solution extraction module is used for extracting glucoside in the enzymatic hydrolysate, the purification module is used for separating and purifying the glucoside, the decolorization and post-treatment module is used for decolorizing the purified glucoside, the crystallization and drying module is used for crystallizing and drying the decolorized glucoside, and the finished product is obtained. The acquisition unit acquires production data of all the modules, and the central control module performs data processing and production control of all the modules by adopting STM32; real-time data acquisition and precise algorithm control are carried out, the automation level of stevioside production is improved, STM32 is adopted for control, and therefore full-process automation is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of stevioside production, and more particularly to a stevioside production system based on a dynamic weight adaptive MPC algorithm. Background Art

[0002] Stevioside is a glycoside extracted from the leaves of the Compositae plant Stevia rebaudiana. Stevia rebaudiana is native to Paraguay and Brazil. It has the characteristics of high sweetness and low calorie. Its sweetness is 200-300 times that of sucrose, and its calorific value is only 1 / 300 of sucrose. Stevioside is a new type of natural sweetener refined from the Compositae herb Stevia rebaudiana. Using Stevia rebaudiana as a herb and a sugar substitute has a history of hundreds of years. Stevioside has been widely used in the production of food, beverages, and seasonings;

[0003] Traditional stevioside extraction technologies have problems such as low raw material processing efficiency and loss of active ingredients. For example, the impurity content is high, and it is difficult to efficiently remove impurities by crushing and using acids and alkalis. Because stevioside has high thermosensitivity, high-temperature pretreatment will cause decomposition. Therefore, the control of temperature is a difficulty. In addition, there are also five major problems in traditional processes: First, if physical crushing is uneven, it is easy to affect the penetration efficiency; Second, it is difficult to control the extraction time and temperature of water, resulting in low production efficiency; Third, the purification process is complex. Traditional purification relies on precipitation centrifugation resin adsorption, and it is difficult to control the processing time; Fourth, in the metal salt precipitation method, problems such as excessive heavy metals may occur; Fifth, in the decolorization treatment, the decolorizing agent has low adsorption efficiency for glycosides. Sixth, the traditional problem in crystallization is that the crystal particles are uneven and the product is easy to absorb moisture. Therefore, a new automated stevioside production system is urgently needed. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a stevioside production system based on a dynamic weight adaptive MPC algorithm to solve the technical problems proposed in the background art.

[0005] To achieve the above object, the present invention provides the following technical solution: A stevioside production system based on a dynamic weight adaptive MPC algorithm, including a raw material pretreatment module, a solution extraction module, a purification module, a decolorization and post-treatment module, a crystallization and drying module, a central control module, and a collection unit. The raw material pretreatment module crushes and enzymatically hydrolyzes stevia leaves and removes impurities to obtain an enzymolysis solution. The solution extraction module extracts the glycosides in the enzymolysis solution. The purification module separates and purifies the glycosides. The decolorization and post-treatment module decolorizes the purified glycosides, and the decolorization and post-treatment module regenerates the adsorbent after decolorization. The crystallization and drying module crystallizes and dries the decolorized glycosides to complete the production of stevioside. The collection unit collects the production data of all modules. The central control module uses STM32 for data processing and production control of all modules;

[0006] The acquisition unit acquires the pressure data information YL during the crushing of stevia leaves in the raw material pretreatment module, as well as the temperature data information WD and the pH data information PH during enzymatic hydrolysis. The acquisition unit sends the acquired data information to the central control module. The central control module receives the data sent by the acquisition unit and calculates the temperature prediction value WY and the pH prediction value SY. The calculation formula for the temperature prediction value WY is where n is the data acquisition by the acquisition unit at the current latest nth second, WD N is the temperature data acquired at the current latest nth second, WD N+1 is the temperature data acquired at the current latest (n + 1)th second, WD is the current latest temperature data in the enzymatic hydrolysis tank, and the calculated temperature prediction value WY is the temperature data in the enzymatic hydrolysis tank five minutes later.

[0007] Furthermore, the calculation formula for the pH prediction value SY is PH N is the pH data acquired at the current latest nth second, PH N+1 is the pH data acquired at the current latest (n + 1)th second, PH is the current latest pH data in the enzymatic hydrolysis tank, and the calculated pH prediction value SY is the pH data in the enzymatic hydrolysis tank five minutes later. The central control module receives the temperature prediction value WY and the pH prediction value SY and controls the flow rate of the enzyme solution.

[0008] Furthermore, when the raw material pretreatment module crushes stevia leaves, it includes two steps: coarse crushing and fine crushing. During coarse crushing, the target function J1 is used to control the crushing cavity to perform coarse crushing work. The calculation formula for the target function J1 is J1 = W grain ·||D 50 - 100|| 2 + 0.2·||T - 40|| 2 , where W grain is the particle size uniformity weight, with a value of 0.8, D 50 is the median particle size of the powder average particle size, T is the temperature in the crushing cavity. During fine crushing, the target function J2 is used to control the crushing cavity to perform fine crushing work. The calculation formula for the target function J2 is where W temp is the temperature control weight, with a value of 0.6.

[0009] Furthermore, the acquisition unit uses a conductivity sensor and a near-infrared spectrometer to monitor the glycoside concentration in the extract in the solution extraction module in real time. When the solution extraction module performs solution extraction work, it includes a heating-up stage and a steady-state stage. The heating-up stage lasts for ten minutes, and the target function J3 is used to control the solution heating, J3 = 0.9·||T - 80|| 2 + 0.1·||PH - 6.0||2 , where T is the temperature of the solution, PH is the acidity and alkalinity of the solution during extraction, and in the steady state stage, the objective function J4 is used to control the solution to maintain the temperature and acidity and alkalinity for solution extraction. The calculation formula of the objective function J4 is J4 = 0.6·||T - 85|| 2 + 0.7·||PH - 5.5|| 2 .

[0010] Furthermore, when the purification module performs purification processing, it is controlled by the central control module. The central control module uses simulated moving bed chromatography to control the purification module, and the simulated moving bed chromatography uses a dynamic weight adaptive model predictive control algorithm. The dynamic weight adaptive model predictive control algorithm first uses the system state equation C k+1 to define the dynamic model of the component concentration in the chromatographic column. The system state equation C k+1 has the calculation formula of C k+1 = A·C k + B·uk + D·wk, where C k is the glycoside concentration vector at the current moment, uk is the control input value, wk is the external disturbance value, and A, B, and D are all system matrices identified by offline experimental data.

[0011] Furthermore, after the dynamic weight adaptive model predictive control algorithm in the central control module is defined through the system state equation C k+1 , a time-varying weight is introduced through the dynamic weight objective function J. The calculation formula of the objective function J is where α i (k) is the weight for dynamically adjusting the prediction step according to the concentration gradient, β j (k) is the adaptive adjustment value based on the smoothness of the historical control input. i represents the current i-th time of dynamically adjusting the concentration gradient. N P is the total number of times of dynamically adjusting the concentration gradient. C k+i / k is the dynamic change value of the component concentration. C target is the target value of the component concentration. j represents the current j-th time of adjusting the smoothness of the historical control input. Ne is the total number of times of adjusting the smoothness of the historical control input. Δuk is the target value of the adaptive adjustment. j / k is the adjustment value of the adaptive adjustment.

[0012] Furthermore, the steps for dynamically adjusting the dynamic weight are as follows: The first step is real-time feedback optimization. According to the glycoside concentration deviation detected by the UV-Vis spectrometer, the separation precision weight is dynamically increased. The second step is historical data learning. The historical process data stored in STM32 is used to generate a weight adjustment rule library through offline training and online match the optimal weight combination. The third step is disturbance compensation. When a sudden disturbance is detected by the pressure or pH sensor, the weight of the control quantity change is automatically reduced.

[0013] Furthermore, after introducing time-varying weights through the dynamic weight objective function J, the central control module finally uses constraint conditions for constraint. The constraint conditions include mobile phase ratio constraint, solenoid valve switching frequency constraint, and concentration stability constraint. The constraint condition for the mobile phase ratio constraint is that each adjustment of the water / ethanol ratio is between 0.3% and 0.7%. The constraint condition for the solenoid valve switching frequency constraint is that the solenoid valve switching frequency is greater than or equal to once every ten seconds. The constraint condition for the concentration stability constraint is that the concentration change of glycoside per five minutes is within 0.5%.

[0014] Furthermore, the dynamic weights in the central control module use the reward function r for reinforcement learning. The calculation formula of the reward function r is where NH is the current energy consumption, 500 is the standard energy consumption, and the energy consumption unit is kWh m.

[0015] Furthermore, the crystallization and drying module includes a constant rate drying stage and a falling rate drying stage. The constant rate drying stage is controlled by the dynamic function J5. The calculation formula of the dynamic function J5 is where W rate is the drying rate weight, k target is the target drying efficiency, M is the water content, t is the time. The falling rate drying stage is controlled by the dynamic function J6. The calculation formula of the dynamic function J6 is J6 = W uniform ·||Tmax - Tmin|| 2 + 0.4·||M - 0.04|| 2 where W uniform is the temperature uniformity weight, Tmax is the highest temperature, Tmin is the lowest temperature, and M is the water content.

[0016] The technical effects and advantages of the present invention:

[0017] 1. The present invention performs real-time data collection and precise algorithm control to improve the automation level of sweet polysaccharide production. It uses STM32 for control to achieve full-process automation, reduce manual intervention, significantly shorten the production cycle, and can adjust process parameters in real time to maximize production efficiency. The real-time control ability of STM32 ensures the stability of key parameters and higher product purity;

[0018] 2. By calculating the temperature prediction value WY and the pH prediction value SY, the present invention is beneficial to understanding the temperature and pH in the enzymatic hydrolysis tank after five minutes. Therefore, the STM32 in the central control module of the present application will predict the change of enzyme activity according to the change trends of temperature and pH value, and adjust the enzyme solution flow valve five minutes in advance to keep the enzyme activity in the best state all the time;

[0019] 3. Through the dynamic weight mechanism, the present invention can adaptively adjust the weights of the objective function according to the concentration gradient, giving priority to ensuring the separation accuracy in the rapid component change stage and reducing the control energy consumption in the steady state stage. By combining the modeling of temperature and flow rate disturbances, the anti-interference ability of the algorithm is improved, and the matrix operation and iterative algorithm are optimized to ensure real-time operation on the central control module, solving the problems of the lagging response of traditional PID control and the insufficient adaptability of the fixed-weight MPC. Brief Description of the Drawings

[0020] Figure 1 It is a schematic diagram of the overall system composition of the present invention. Detailed Embodiments

[0021] Next, the technical solutions in the present invention will be clearly and completely described in conjunction with the drawings in the present invention. In addition, the forms of the various structures described in the following embodiments are merely examples. The stevioside production system based on the dynamic weight adaptive MPC algorithm involved in the present invention is not limited to the various structures described in the following embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0022] Referring to Figure 1 , the present invention provides a stevioside production system based on the dynamic weight adaptive MPC algorithm, including a raw material pretreatment module, a solution extraction module, a purification module, a decolorization and post-treatment module, a crystallization and drying module, a central control module, and a collection unit. The raw material pretreatment module crushes and enzymatically hydrolyzes stevia leaves and removes impurities to obtain an enzymolysis solution. The solution extraction module extracts glycosides from the enzymolysis solution. The purification module separates and purifies the glycosides. The decolorization and post-treatment module decolorizes the purified glycosides, and the adsorbent is regenerated after decolorization in the decolorization and post-treatment module. The crystallization and drying module crystallizes and dries the decolorized glycosides to complete the production of stevioside. The collection unit collects the production data of all modules, and the central control module uses STM32 for data processing and production control of all modules.

[0023] In the embodiments of the present application, the present application adopts STM32 for control, so as to achieve full-process automation, reduce manual intervention, significantly shorten the production cycle, and can adjust process parameters in real time, maximize production efficiency, improve production capacity elasticity, and can operate continuously for 24 hours, greatly increasing the output. The real-time control ability of STM32 ensures the stability of key parameters and higher product purity. Through sensor feedback and algorithm control, the glycoside retention rate is increased, glycoside loss is reduced, the crystallization process is precisely regulated, and uniform crystals are generated to improve product quality. By optimizing the crushing, enzymatic hydrolysis, and extraction processes, raw material waste is reduced, the ion concentration of the waste liquid is monitored in real time, the neutralization reaction is automatically started, and the risk of environmental protection penalties is reduced. Through STM32 control, solvent recovery and recycling are carried out, solvent emissions are reduced, and environmental pollution is decreased.

[0024] Refer to Figure 1 , the acquisition unit acquires the pressure data information YL during the crushing of stevia leaves and the temperature data information WD and pH data information PH during enzymatic hydrolysis in the raw material pretreatment module. The acquisition unit sends the acquired data information to the central control module. The central control module receives the data sent by the acquisition unit and calculates the temperature prediction value WY and the pH prediction value SY. The calculation formula for the temperature prediction value WY is In the formula, n is the data acquisition by the acquisition unit at the current latest nth second, WD N is the temperature data acquired at the current latest nth second, WD N+1 is the temperature data acquired at the current latest (n + 1)th second, WD is the current latest temperature data in the enzymatic hydrolysis tank, and the calculated temperature prediction value WY is the temperature data in the enzymatic hydrolysis tank five minutes later. The calculation formula for the pH prediction value SY is PH N is the pH data acquired at the current latest nth second, PH N+1 is the pH data acquired at the current latest (n + 1)th second, PH is the current latest pH data in the enzymatic hydrolysis tank, and the calculated pH prediction value SY is the pH data in the enzymatic hydrolysis tank five minutes later. The central control module receives the temperature prediction value WY and the pH prediction value SY and controls the flow rate of the enzyme solution.

[0025] In the embodiment of the present application, during the enzymatic hydrolysis process after the stevia leaves are crushed, the temperature data information WD and the pH data information PH during enzymatic hydrolysis are collected, and the temperature prediction value WY and the pH prediction value SY are calculated. The temperature prediction value WY and the pH prediction value SY are the temperature and pH in the enzymatic hydrolysis tank after five minutes. Therefore, the STM32 in the central control module of the present application will predict the change of enzyme activity according to the change trend of temperature and pH value, and adjust the enzyme solution flow valve 5 minutes in advance to keep the enzyme activity always in the best state. When calculating the temperature prediction value WY of the present application, first calculate the difference in temperature change per second, and accumulate and calculate the total data of the change in one minute, that is, sixty seconds. The cumulative calculation method is adopted instead of directly collecting the data of the first second and the sixtieth second, so as to avoid the problem of inaccurate predicted temperature caused by sudden changes within the sixtieth second during enzymatic hydrolysis. And the total temperature change within five minutes is used for temperature prediction after five minutes to ensure the accuracy of temperature prediction, that is, the calculated temperature prediction value WY of the present application is accurate enough, and the pH prediction value SY is calculated in the same way as the temperature prediction value WY, and the calculation results are both relatively accurate.

[0026] Refer to Figure 1 , the central control module receives the pressure data information YL, and the central control module monitors the particle distribution state in the stevia leaf crushing cavity through the pressure data information YL. The central control module adjusts the ultrasonic power and frequency, and the crushing particle size during stevia leaf crushing is 50μm - 100μm;

[0027] When the raw material pretreatment module crushes stevia leaves, it includes two steps: coarse crushing and fine crushing. During coarse crushing, the target function J1 is used to control the crushing cavity to perform coarse crushing work, and the crushing cavity is controlled to coarsely crush stevia leaves to within 100μm. The calculation formula of the target function J1 is J1 = W grain ·||D 50 - 100|| 2 + 0.2·||T - 40|| 2 , where W grain is the particle size uniformity weight, with a value of 0.8, D 50 is the median particle size of the powder average particle size, T is the temperature in the crushing cavity. During fine crushing, the target function J2 is used to control the crushing cavity to perform fine crushing work, and the target particle size after fine crushing is within 50μm. The calculation formula of the target function J2 is J2 = 0.5·||D 50 - 50|| 2 + W temp ·||T - 35|| 2 , where W temp is the temperature control weight, with a value of 0.6. During coarse crushing and fine crushing, the ultrasonic crushing power model is used to predict the powder crushing degree. The ultrasonic crushing power model is where k is the raw material characteristic coefficient, t is the grinding time, and P is the pressure in the grinding chamber during grinding.

[0028] In the embodiments of the present application, the particle distribution in the stevia leaf grinding chamber is monitored in real time through a pressure sensor, and the ultrasonic power and frequency are dynamically adjusted to ensure uniform grinding particle size. The target particle size is 50μm - 100μm, reducing the oxidation loss of glycosides caused by excessive grinding. The grinding of stevia leaves in this application includes two steps: coarse grinding and fine grinding. The objective function J1 is used for coarse grinding to quickly reduce the particle diameter and facilitate subsequent fine grinding. When performing fine grinding, the particle diameter of the stevia leaves after grinding is precisely controlled. During both coarse grinding and fine grinding, the ultrasonic grinding dynamic model is used to predict the powder grinding degree to ensure the accuracy of the prediction.

[0029] Refer to Figure 1 , the acquisition unit uses a conductivity sensor and a near-infrared spectrometer to monitor the glycoside concentration in the extract in the solution extraction module in real time, and the acquisition unit transmits the collected glycoside concentration data to the central control module. The central control module receives the collected concentration data, calculates the concentration change rate K per minute, and the central control module compares the calculated concentration change rate K with a threshold of 1% / min. When the concentration change rate K is less than the threshold, the central control module sends an instruction to terminate heating and start the cooling process. When the concentration change rate K is greater than or equal to the threshold, work continues;

[0030] The acquisition unit uses a conductivity sensor and a near-infrared spectrometer to monitor the glycoside concentration in the extract in the solution extraction module in real time. When the solution extraction module performs solution extraction work, it includes a heating-up stage and a steady-state stage. The heating-up stage lasts for ten minutes, and the objective function J3 is used to control the solution temperature rise. J3 = 0.9·||T - 80|| 2 + 0.1·||PH - 6.0|| 2 , where T is the solution temperature and PH is the acidity and alkalinity during solution extraction. In the steady-state stage, the objective function J4 is used to control the solution to maintain the temperature and acidity and alkalinity for solution extraction. The calculation formula of the objective function J4 is J4 = 0.6·||T - 85|| 2 + 0.7·||PH - 5.5|| 2 , and the thermodynamic equilibrium equation is used for prediction during solution extraction. The thermodynamic equilibrium equation is where C glycoside is the current concentration of the solution, Ea is the activation energy, k is the dissolution rate constant, and C solid is the initial concentration of steviol glycoside.

[0031] In the embodiments of the present application, by integrating a conductivity sensor, a near-infrared spectroscopy module, a temperature sensor, and a liquid level sensor, the extract data is collected and transmitted to the central control module in real time. The central control module detects the change rate of glycoside concentration, controls the heating and cooling processes. Once the temperature exceeds the limit or the liquid level is abnormal, the heating is immediately cut off and an alarm is given to avoid over-extraction and potential safety hazards, and improve batch consistency. When extracting the glycoside concentration in the extract, different methods are used to control the temperature and pH value in the heating-up stage and the steady state stage, so as to ensure a higher glycoside concentration in the final extraction.

[0032] Refer to Figure 1 , when the purification module performs purification processing, it is controlled by the central control module. The central control module uses simulated moving bed chromatography to control the purification module, and the simulated moving bed chromatography uses a dynamic weight adaptive model predictive control algorithm. The dynamic weight adaptive model predictive control algorithm first uses the system state equation C k+1 to define the dynamic model of the component concentration in the chromatographic column. The calculation formula of the system state equation C k+1 is C k+1 =A·C k +B·uk+D·wk, where C k is the glycoside concentration vector at the current moment, uk is the control input value, wk is the external disturbance value, and A, B, and D are all system matrices identified by offline experimental data. Then, a time-varying weight is introduced through the dynamic weight objective function J. The calculation formula of the objective function J is where α i (k) is the weight for dynamically adjusting the prediction step length according to the concentration gradient, β j (k) is the adaptive adjustment value based on the smoothness of the historical control input. i represents the current i-th time of dynamically adjusting the concentration gradient, N P is the total number of times of dynamically adjusting the concentration gradient, C k+i / k is the dynamic change value of the component concentration, C target is the target value of the component concentration. j represents the current j-th time of adjusting the smoothness of the historical control input, Ne is the total number of times of adjusting the smoothness of the historical control input, Δuk is the target value of the adaptive adjustment, j / k is the adjustment value of the adaptive adjustment. Finally, constraints are used for constraint. The constraints include mobile phase ratio constraint, solenoid valve switching frequency constraint, and concentration stability constraint. The constraint condition of the mobile phase ratio constraint is that each time the water / ethanol ratio is adjusted between 0.3% and 0.7%. The constraint condition of the solenoid valve switching frequency constraint is that the solenoid valve switching frequency is greater than or equal to once every ten seconds. The constraint condition of the concentration stability constraint is that the concentration change of the glycoside concentration within every five minutes is within 0.5%. The dynamic weight in the central control module uses a reward function r for reinforcement learning. The calculation formula of the reward function r is Where NH is the current energy consumption, 500 is the standard energy consumption, and the energy consumption unit is kilowatt-hour.

[0033] In the embodiment of the present application, the dynamic weight mechanism can adaptively adjust the weight of the objective function through the concentration gradient, give priority to ensuring the separation accuracy in the stage of rapid component change, reduce the control energy consumption in the steady state stage, combine the modeling of disturbances such as temperature and flow rate, improve the anti-interference ability of the algorithm, optimize the matrix operation and iterative algorithm, and ensure real-time operation on the central control module. This formula and algorithm solve the problems of lagging response of traditional PID control and insufficient adaptability of fixed-weight MPC by integrating model predictive control, dynamic weight optimization and machine learning compensation mechanism.

[0034] Refer to Figure 1 , the decolorization and post-treatment module uses a magnetic nano-adsorbent for dosing control, and the decolorization and post-treatment module uses an online spectrometer to detect the pigment content and send it to the central control module. The central control module uses a neural network algorithm to calculate the optimal adsorbent dosing amount. The neural network algorithm first uses the historical pigment content, adsorbent dosing amount and glycoside loss rate data as the training set to train the neural network, and the value corresponding to the lowest glycoside loss rate obtained by training is the optimal adsorbent dosing amount.

[0035] In the embodiment of the present application, after the online spectrometer detects the pigment content, STM32 in the central control module uses a neural network algorithm to calculate the optimal adsorbent dosing amount. First, the historical pigment content, adsorbent dosing amount and glycoside loss rate data are used as the training set to train the neural network. During actual operation, the current pigment content data is input, and the optimal dosing amount is output by the neural network. After each addition of the adsorbent, wait for 5 minutes and then detect the pigment content again. If the target decolorization effect is not achieved, the adsorbent is supplemented proportionally according to the deviation value (20% of the deviation). At the same time, monitor the glycoside concentration. If the glycoside loss rate is close to 3%, immediately stop supplementing. The online spectrometer detects the pigment content and transmits it to STM32. STM32 uses the least squares method to calculate the optimal adsorbent dosing amount, and after dosing, detects the pigment content again, supplements according to the deviation proportionally, and monitors the glycoside concentration at the same time. If the loss rate is close to the threshold, stop supplementing. After the production of each batch is completed, optimize the dosing amount calculation model according to the pigment removal rate and glycoside loss rate;

[0036] In addition, during post-treatment, i.e., magnetic separation and adsorbent regeneration, when the central control module controls the on-off cycle of the electromagnet, the first separation uses a frequency of 1 second on and 1 second off for 3 consecutive times. If the separation effect is not good, it is adjusted to 2 seconds on and 1 second off until the adsorbent is quickly separated. During the adsorbent regeneration process, the flow rate of the regeneration liquid is dynamically adjusted according to the number of times the adsorbent is used and the degree of contamination. When the new adsorbent is regenerated for the first time, the flow rate of dilute hydrochloric acid is 0.5 L / min; every time the number of uses increases by 5 times, the flow rate increases by 0.1 L / min to ensure the regeneration effect of the adsorbent and extend its reuse times.

[0037] Referring to Figure 1 , the crystallization and drying module includes a crystallization unit and a drying unit. The crystallization unit adjusts the temperature gradient and stirring rate in the crystallizer through the central control module. The central control module uses the PID algorithm for adjustment. When the central control module uses the PID algorithm to adjust the temperature gradient and stirring rate in the crystallizer, a temperature integral separation coefficient is set. When the temperature deviation is less than ±0.05°C, the integral action is cancelled. When the deviation is greater than ±0.05°C, the integral action is restored. The stirring rate adjustment uses variable speed control. The first 30 minutes is the initial crystallization stage, and the stirring rate rises from 100 rpm to 150 rpm. The 30 - 90 minutes is the middle crystallization stage, and the stirring rate remains at 200 rpm. After 90 minutes is the late crystallization stage, and the stirring rate drops to 150 rpm.

[0038] In the embodiment of the present application, when the central control module uses the PID algorithm to adjust the temperature gradient and stirring rate in the crystallizer, a temperature integral separation coefficient is set. When the temperature deviation is less than ±0.05°C, the integral action is cancelled to avoid overshoot caused by integral saturation. When the deviation is greater than this range, the integral action is restored. The stirring rate adjustment uses variable speed control. The first 30 minutes is the initial crystallization stage, and the stirring rate rises from 100 rpm to 150 rpm to promote the formation of crystal nuclei; the 30 - 90 minutes is the middle crystallization stage, and the stirring rate remains at 200 rpm to facilitate crystal growth; after 90 minutes is the late crystallization stage, and it drops to 150 rpm to prevent crystal breakage. The crystal particle size distribution is detected every 15 minutes, and the temperature gradient and stirring rate are finely adjusted according to the D90 value. The central control module collects data through temperature and stirring rate sensors and uses an adaptive PID algorithm to adjust the crystallizer parameters. In different crystallization stages, the PID parameters are automatically adjusted. At the same time, according to the feedback of the crystal particle size distribution, the temperature gradient and stirring rate are finely adjusted. The crystal growth data is analyzed every 30 minutes to optimize the PID parameters and control strategy.

[0039] Referring to Figure 1 , the crystallization and drying module includes a constant - rate drying stage and a falling - rate drying stage. The constant - rate drying stage is controlled by the dynamic function J5. The calculation formula of the dynamic function J5 is where W rate is the drying rate weight, k target is the target drying efficiency, M is the moisture content, t is the time. The falling - rate drying stage is controlled by the dynamic function J6. The calculation formula of the dynamic function J6 is J6 = W uniform ·||Tmax - Tmin|| 2 +0.4·||M - 0.04|| 2 where W uniformis the temperature uniformity weight, Tmax is the highest temperature, Tmin is the lowest temperature, M is the water content, and the crystallization and drying module uses Fick's diffusion law and heat transfer equation for prediction. Fick's diffusion law and heat transfer equation are where D is the moisture diffusion coefficient, ▽ 2 M is the second spatial derivative of the moisture content, h is the convective heat transfer coefficient, ρ is the material density, C P is the specific heat capacity of the material, T air is the air temperature, T material is the material temperature.

[0040] In the embodiment of the present application, after the fluidized bed drying is optimized and the humidity sensor and hot air flowmeter are integrated, the central control module adjusts the drying temperature and wind speed according to the humidity change rate. When the humidity change rate is greater than 0.1% / min, the drying temperature is increased by 2°C, and at the same time the wind speed is increased by 0.1 m / s; if the humidity change rate is less than 0.1% / min, the temperature is decreased by 1°C and the wind speed is decreased by 0.05 m / s. When the moisture content of the finished product is 0.3%, it enters the fine drying stage, and the moisture content is detected every 10 minutes. Each time the temperature is adjusted by 0.5°C and the wind speed is adjusted by 0.02 m / s until the moisture content meets the standard and then the nitrogen filling and packaging are automatically triggered. Therefore, the crystallization and drying module of the present application includes a constant rate drying stage and a falling rate drying stage. The constant rate drying stage can quickly reduce the water content, and then through the falling rate drying stage, the water content is precisely controlled, and finally the water content of the final product meets the requirements more.

[0041] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A stevioside production system based on a dynamic weight adaptive MPC algorithm, characterized in that: It includes a raw material pretreatment module, a solution extraction module, a purification module, a decolorization and post-treatment module, a crystallization and drying module, a central control module, and a data acquisition unit. The raw material pretreatment module crushes and enzymatically hydrolyzes stevia leaves and removes impurities to obtain an enzymatically hydrolyzed solution. The solution extraction module extracts glycosides in the enzymatically hydrolyzed solution. The purification module separates and purifies the glycosides. The decolorization and post-treatment module decolorizes the purified glycosides. The crystallization and drying module crystallizes and dries the decolorized glycosides to complete the production of stevioside. The data acquisition unit collects production data of all modules. The central control module uses STM32 for data processing and production control of all modules; The central control module receives the data sent by the acquisition unit and calculates the temperature prediction value WY and the pH prediction value SY. The calculation formula for the temperature prediction value WY is where n is the data acquisition by the acquisition unit at the current latest nth second, WD N is the temperature data collected at the current latest nth second, WD N+1 is the temperature data collected at the current latest (n + 1)th second, WD is the current latest temperature data in the enzymatic hydrolysis tank, and the calculated temperature prediction value WY is the temperature data in the enzymatic hydrolysis tank five minutes later.

2. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, wherein: The calculation formula for the predicted value SY of acidity and alkalinity is PH N is the acidity and alkalinity data collected at the current latest nth second, and PH N+1 is the acidity and alkalinity data collected at the current latest (n + 1)th second. PH is the current latest acidity and alkalinity data in the enzymatic hydrolysis tank, and the calculated predicted value SY of acidity and alkalinity is the acidity and alkalinity data in the enzymatic hydrolysis tank five minutes later. The central control module receives the predicted value WY of temperature and the predicted value SY of acidity and alkalinity and controls the flow rate of the enzyme solution.

3. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, characterized in that: When the raw material pretreatment module crushes stevia leaves, it includes two steps: coarse crushing and fine crushing. During coarse crushing, the crushing chamber is controlled to coarsely crush stevia leaves to within 100 μm. During fine crushing, the crushing chamber is controlled to perform fine crushing work, and the target particle size after fine crushing is within 50 μm.

4. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, wherein: The data acquisition unit uses a conductivity sensor and a near-infrared spectrometer to real-time monitor the glycoside concentration in the extract in the solution extraction module. When the solution extraction module performs solution extraction work, it includes a heating-up stage and a steady-state stage, and the heating-up stage lasts for ten minutes.

5. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, characterized in that: When the purification module performs purification processing, it is controlled by the central control module. The central control module uses simulated moving bed chromatography to control the purification module, and the simulated moving bed chromatography uses a dynamic weight adaptive model predictive control algorithm. The dynamic weight adaptive model predictive control algorithm first uses the system state equation C k+1 Define the dynamic model of component concentration in the chromatographic column.

6. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, wherein: The dynamic weight adaptive model predictive control algorithm in the central control module is defined by the system state equation C k+1 After definition, time-varying weights are introduced through the dynamic weight objective function J.

7. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 6, wherein: The steps for dynamically adjusting the dynamic weight are as follows: The first step is real-time feedback optimization. According to the glycoside concentration deviation detected by the UV-Vis spectrometer, the separation precision weight is dynamically increased. The second step is historical data learning. Using the historical process data stored in STM32, a weight adjustment rule library is generated through offline training, and the optimal weight combination is matched online. The third step is disturbance compensation. When a sudden disturbance is detected by a pressure or pH sensor, the weight of the control quantity change is automatically reduced.

8. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 6, wherein: After the central control module introduces time-varying weights through the dynamic weight objective function J, finally, constraint conditions are used for constraint. The constraint conditions include mobile phase ratio constraint, solenoid valve switching frequency constraint, and concentration stability constraint. The constraint condition for the mobile phase ratio constraint is that each time the water / ethanol ratio is adjusted between 0.3% and 0.7%. The constraint condition for the solenoid valve switching frequency constraint is that the solenoid valve switching frequency is greater than or equal to ten seconds each time. The constraint condition for the concentration stability constraint is that the concentration change of glycosides within every five minutes is within 0.5%.

9. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, wherein: The dynamic weight in the central control module uses a reward function r for reinforcement learning.

10. The stevioside production system based on the dynamic weight adaptive MPC algorithm according to claim 1, wherein: The crystallization and drying module includes a constant-rate drying stage and a falling-rate drying stage. The constant-rate drying stage is controlled by a dynamic function J5, and the falling-rate drying stage is controlled by a dynamic function J6.

Images