A sun-cured tobacco distillation processing method

Abstract

The application discloses a sun-cured tobacco distillation processing technology method, and relates to the technical field of tobacco processing, in particular to a sun-cured tobacco distillation processing technology method.The application constructs a mapping relationship between raw material dielectric fingerprints and thermal response characteristics, combines with a coplanar waveguide sensor array to collect data in real time, uses a neural network to predict the thermal release starting temperature of key aroma components, dynamically allocates radio frequency heating parameters, realizes a gradient heating process including low-frequency uniform preheating, medium-frequency selective excitation and high-frequency local focusing, and effectively improves the recovery rate of target aroma components and the retention rate of heat-sensitive components.The application has beneficial effects such as online self-adaptive regulation and control, high extraction efficiency and continuous model optimization.The application constructs a dielectric fingerprint-thermal response characteristic correlation model, realizes the regulation and control of a heating strategy driven by the inherent physical properties of raw materials, significantly improves the scientificity and accuracy of process control, automatically adapts an optimal thermal treatment path without manual intervention, and greatly improves batch consistency and target component recovery stability.

Description

Technical Field

[0001] This invention relates to the field of industrial process control and intelligent radio frequency distillation technology, and in particular to a process method for processing sun-dried tobacco distillation. Background Technology

[0002] Currently, in the field of temperature gradient control and heating strategy optimization, the mainstream technologies in tobacco distillation processing systems mainly rely on traditional heat conduction or programmed heating, supplemented by external temperature probes and basic PID algorithms. Existing distillation processes widely employ isothermal heating, inert gas carriers, solvent extraction post-treatment, molecular distillation, and a small number of after-effect optimization schemes based on predictive models. However, the overall temperature control accuracy is limited by the probe installation position, the thermal inertia of the material, and the failure to dynamically reflect the physicochemical properties of the raw materials, resulting in uneven heat field distribution, easy thermal degradation of aroma components, and insufficient extraction rate of volatile components.

[0003] Common high-end radio frequency heating distillation methods are mostly used for the separation of fine chemicals or high-value biological raw materials. Their technical characteristics focus on uniform heating, rapid temperature rise, and low heat loss. However, in actual distillation processes of the tobacco extract-propylene glycol-glycerol system, a single mechanical temperature probe and a fixed heating mode are insufficient to match the complex component differences and volatility sensitivity distributions between batches of raw materials. Some published literature proposes using machine learning to assist in the heating process or to collect material surface temperatures to predict the aroma release stage; however, these methods are mostly post-distillation process analyses with limited real-time process adjustment capabilities and unclear effectiveness in retaining key aromas. Although dielectric constant measurement has been implemented for material identification in the field of materials characterization, its application as a basis for real-time distillation decisions has not yet been engineered.

[0004] Specifically, the existing technology mainly has the following problems: Low temperature gradient control accuracy. This is mainly due to limitations in single-point temperature sensing and uniform heating logic, which fails to achieve dynamic zone heating based on the actual volatilization behavior of the material, resulting in insufficient distillation efficiency and recovery rate of target aroma components.

[0005] It is highly dependent on a fixed heating program. It is not responsive enough to the individual characteristics of raw material batches, lacks adaptive control capabilities, and is prone to quality losses such as overheating and thermal degradation of aroma.

[0006] External temperature detection is delayed. Due to thermal inertia, there is a delay in temperature measurement and control response, making it difficult to capture the optimal release window of volatile components.

[0007] Machine learning is limited to post-process analysis. Existing data-driven solutions fail to achieve real-time online closed-loop adjustment of raw material physical properties, and cannot support advance optimization of process parameters.

[0008] The impact of the material's microstructure on the heating strategy cannot be fully identified. A mechanistic bridge between the material's state and thermal response has not been established, leading to a mismatch between heating mode switching and distillation product distribution.

[0009] In summary, there is an urgent need in this field for a method that can sense the molecular state of raw materials in real time, accurately invert the heat release window of target aroma components, and intelligently and adaptively adjust the radio frequency heating strategy based on the actual material properties. Summary of the Invention

[0010] This invention provides a process for processing tobacco by distillation, which aims to solve the problems of the prior art mentioned in the background section.

[0011] The technical solution of this invention is: a process for processing sun-dried tobacco by distillation, comprising the following steps: S1: Collect GC-MS spectra of water content, total sugar content, nicotine concentration and aroma components of multiple batches of sun-dried tobacco extract-propylene glycol-glycerol mixture samples, and measure the real and imaginary parts of their complex permittivity in the 10MHz–2.45GHz frequency band under controlled temperature and humidity conditions; S2: Based on the spectral data of the real and imaginary parts of the complex dielectric constant, a multidimensional mapping model is established through principal component analysis and partial least squares regression to form a unique dielectric fingerprint-thermal response characteristic correlation matrix for each batch of raw materials; S3: An annular coplanar waveguide sensor array is embedded in the side wall of the distillation heating chamber to collect real-time data on the dynamic change of the complex permittivity of the current batch of mixture at the center frequencies of 45MHz, 433MHz and 2.45GHz; S4: Input the dynamic change data of complex dielectric constant into a lightweight neural network model trained based on the dielectric fingerprint-thermal response feature correlation matrix to obtain the predicted value of the heat release initiation temperature distribution of the current batch of raw materials; S5: Generate a segmented RF power-frequency coupling control strategy based on the predicted value of the heat release initiation temperature distribution; S6: Based on the control strategy, three-stage gradient radio frequency heating is performed; S7: Monitor the concentration of target aroma components in the steam condensate during the heating process, calculate the actual recovery rate and retention rate of heat-sensitive components by HPLC quantitative analysis, and generate a model bias correction factor; S8: Feedback the model bias correction factor to the dielectric fingerprint-thermal response feature correlation matrix to dynamically update the multidimensional mapping model parameters.

[0012] Preferably, step S4 specifically includes: The dynamic change data of the real part and imaginary part of the dielectric constant of the current batch are processed in a time-series synchronization manner to obtain the original dataset that matches the real-time operating conditions of the distillation heating chamber. Frequency domain adaptive filtering was performed on the original dataset to extract effective data of the characteristic frequency bands at the center frequencies of 45MHz, 433MHz and 2.45GHz; The effective data of the characteristic frequency band is input into the lightweight neural network model to perform feature matching and temperature prediction calculations in order to obtain the original heat release initiation temperature prediction value. The original predicted heat release initiation temperature is subjected to probability distribution fitting to generate the predicted heat release initiation temperature distribution of the current batch of raw materials; A confidence assessment was performed based on the predicted heat release initiation temperature distribution to verify the reliability of the prediction results within the process safety boundary.

[0013] Preferably, in step S5, based on the temperature threshold range and the dielectric fingerprint-thermal response feature correlation matrix, feature point detection processing is performed on the imaginary part spectrum data of the complex dielectric constant to identify the dielectric phase transition feature point with double peak splitting of the imaginary part spectrum of the complex dielectric constant at the 433 MHz frequency point; and phase transition triggering conditions are modeled according to the temperature sensitivity distribution corresponding to the dielectric phase transition feature point to generate the phase transition triggering conditions for the mid-frequency excitation stage. Based on the dielectric phase transition characteristic points, the first derivative gradient of the dynamic change data of the real part of the complex dielectric constant is calculated and processed to detect the trigger threshold of the steep negative gradient of the real part of the complex dielectric constant at the 2.45 GHz frequency point; the trigger condition threshold is optimized according to the temperature change rate corresponding to the steep negative gradient to generate the steep gradient trigger condition in the high-frequency focusing stage.

[0014] Preferably, in step S6, the three-stage gradient radio frequency heating is as follows: When the predicted value of the heat release initiation temperature distribution falls within the range of [45, 65]℃, 45 MHz low-frequency deep penetration preheating is initiated. When a double-peak split of the imaginary part spectral number of the complex permittivity at a frequency of 433 MHz is detected, switch to selective excitation at the 433 MHz intermediate frequency; When a steep negative gradient appears in the real part of the complex permittivity at the 2.45 GHz frequency, 2.45 GHz high-frequency localized focused pulse modulation is triggered.

[0015] As a preferred approach, the model bias correction factor is used in incremental principal component analysis and PLSR incremental learning mechanism to dynamically correct the parameter loads and regression coefficients in the multidimensional mapping model, thereby achieving adaptive optimization and iterative updates of the prediction accuracy of the heat release initiation temperature for subsequent batches.

[0016] The beneficial technical effects of this invention are as follows: 1) This invention constructs a dielectric fingerprint-thermal response feature correlation model to achieve heating strategy regulation driven by the intrinsic physical properties of raw materials, which significantly improves the scientificity and accuracy of process control; 2) This invention systematically calibrates the complex dielectric constant of the tobacco extract-propylene glycol-glycerol system in a wide frequency domain and establishes a PLSR mapping relationship in combination with the thermal release initiation temperature of key aroma components, so that each batch of raw materials has an identifiable dielectric fingerprint. This transforms the heating decision from following environmental parameters to responding to the material state, effectively overcoming the process drift problem caused by the heterogeneity of raw materials. It automatically adapts to the optimal heat treatment path without human intervention, greatly improving batch consistency and target component recovery stability. 3) This invention integrates a highly dynamic response online dielectric sensing module and a multi-band switchable radio frequency heating execution system, realizing precise time and air control of complex phase transitions and chemical transformation behaviors during distillation; 4) This invention utilizes a ring coplanar waveguide (CPW) sensor array to achieve dynamic sampling of three center frequencies (45 MHz, 433 MHz, and 2.45 GHz), corresponding to the overall polarization behavior, the relaxed state of the hydrogen bond network, and the dipole resonance characteristics of small molecules, respectively. Combined with FPGA accelerated processing and lightweight TinyML model inference, dielectric spectrum feature extraction and thermal release initiation temperature prediction can be completed in milliseconds. The controller then determines the current stage of material development in real time: the low-frequency mode ensures the uniformity of the initial thermal field, the mid-frequency mode accurately triggers the dehydration and condensation of high-boiling-point precursors, and the high-frequency pulse mode implements localized focused heating at the volatilization critical point, greatly shortening the residence time of the thermosensitive components. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the method of this invention; Figure 2 This is a schematic diagram of the process for collecting dynamic change data of complex permittivity in this invention; Figure 3 This is a schematic diagram of the process for obtaining the predicted value of the heat release initiation temperature distribution in an embodiment of the present invention. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0020] like Figure 1 As shown, this embodiment provides a method for processing sun-dried tobacco through distillation, which specifically includes the following steps: S1: Collect GC-MS spectra of moisture content, total sugar content, nicotine concentration, and aroma components from no fewer than 50 batches of sun-dried tobacco extract-propylene glycol-glycerol mixture samples, and measure the real part of its complex permittivity in the 10 MHz–2.45 GHz frequency band under controlled temperature and humidity conditions. With the imaginary part Simultaneously record the thermal release initiation temperature of key aroma components. ; S2: Real part of the complex permittivity acquired in step S1 With the imaginary part Spectral data are used to establish a multidimensional mapping model through principal component analysis and partial least squares regression, forming a unique dielectric fingerprint-thermal response characteristic correlation matrix for each batch of raw materials. S3: An annular coplanar waveguide sensor array is embedded in the side wall of the distillation heating chamber to collect real-time data on the dynamic change of the complex permittivity of the current batch of mixture at the center frequencies of 45 MHz, 433 MHz and 2.45 GHz, with a sampling frequency of not less than 10 Hz; S4: Input the dynamic change data of complex permittivity collected in step S3 into a lightweight neural network model trained based on the dielectric fingerprint-thermal response feature correlation matrix in step S2 to obtain the heat release onset temperature of the current batch of raw materials. Predicted distribution values.

[0021] S5: Heat release initiation temperature obtained from S4 Distributed predicted values ​​are used to generate a segmented RF power-frequency coupling control strategy. S6: Based on the control strategy generated in step S5, execute three-stage gradient radio frequency heating; S7: Monitor the concentration of the target aroma components in the vapor condensate during the execution of step S6, calculate the actual recovery rate and the retention rate of heat-sensitive components by HPLC quantitative analysis, and generate a model bias correction factor. S8: Feedback the model bias correction factor generated in step S7 to the dielectric fingerprint-thermal response feature correlation matrix in step S2, and dynamically update the multidimensional mapping model parameters to optimize the heat release onset temperature of subsequent batches. Prediction accuracy.

[0022] In this embodiment, in step S1, GC-MS spectra of the water content, total sugar content, nicotine concentration, and aroma components of no less than 50 batches of sun-dried tobacco extract-propylene glycol-glycerol mixture samples are collected, and the real part of its complex permittivity in the 10MHz-2.45GHz frequency band is measured under temperature and humidity controlled conditions. With the imaginary part Simultaneously record the thermal release initiation temperature of key aroma components. Specifically, it includes the following steps: S1.1: Standardize the mixing process of tobacco extract, propylene glycol and glycerin raw materials to prepare no less than 50 batches of mixed liquid samples; based on the distribution characteristics of typical industry water content range of 12%–28%, total sugar content of 30–42% and nicotine concentration of 1.0–2.5 mg / g, generate a set of sample batches covering all working conditions to obtain a representative mixed liquid sample library; S1.2: Perform chemical composition parameter measurement processing on each sample in the batch set of mixed liquid samples prepared in step S1.1; calculate the chemical composition parameter dataset of each batch of samples based on the water content determination by drying method, the total sugar content determination by anthrone colorimetric method, and the nicotine concentration determination by high performance liquid chromatography, and extract the standardized chemical composition feature vector. S1.3: The mixed sample corresponding to the chemical component feature vector obtained in step S1.2 is subjected to GC-MS spectrum acquisition and processing for aroma components; volatile component separation and mass spectrometry identification are performed using gas chromatography-mass spectrometry to generate key aroma components including furfural, damascone, and... - Qualitative and quantitative spectral dataset of dihydroionone was used to obtain the aroma component distribution feature matrix; S1.4: Perform complex dielectric constant spectrum measurement on the mixed liquid sample after acquiring the GC-MS spectrum in step S1.3 within a temperature- and humidity-controlled environment. Set dynamic control parameters of temperature gradient 25-90℃ and humidity 30-75% RH, and use a broadband coaxial probe dielectric spectrometer to scan in the 10 MHz-2.45 GHz frequency band to obtain the real part of the complex dielectric constant. With the imaginary part The frequency domain response dataset is used to form a temperature-frequency coupled dielectric property database; S1.5: During the measurement of the complex permittivity in step S1.4, the onset temperature of the thermal release of key aroma components is simultaneously measured. Record processing; based on online thermogravimetric analysis and mass spectrometry monitoring technology, identify the volatilization initiation points of target components such as furfural and damascone, and calculate the volatilization initiation points at each temperature. Numerical sets are used to generate time series data of thermal response characteristics.

[0023] In this embodiment, in step S2, the real part of the complex permittivity acquired in step S1 is used as the basis for the calculation. With the imaginary part Spectral data is used to establish a multidimensional mapping model through principal component analysis and partial least squares regression, forming a unique dielectric fingerprint-thermal response characteristic correlation matrix for each batch of raw materials. The specific steps include the following: S2.1: Perform data cleaning and standardization on the real part spectrum data and imaginary part spectrum data of the complex permittivity collected in step S1 to eliminate measurement noise and unify the dimensions, and obtain the standardized real part and imaginary part spectrum dataset of the complex permittivity. S2.2: Based on the standardized spectrum dataset of the real and imaginary parts of the complex permittivity, the principal component analysis (PCA) algorithm is applied to extract key principal components and generate a principal component score matrix. S2.3: Input the principal component score matrix and the heat release initiation temperature data recorded synchronously in step S1 into the partial least squares regression algorithm PLSR to establish the mapping relationship between complex dielectric properties and thermal response behavior, and calculate the parameters of the partial least squares regression model. S2.4: Using the parameters of the partial least squares regression model, a multidimensional mapping model between the spectrum of complex dielectric constant and the heat release initiation temperature is constructed to form a unique dielectric fingerprint-thermal response feature correlation matrix for each batch of raw materials; S2.5: Perform k-fold cross-validation on the dielectric fingerprint-thermal response feature correlation matrix to evaluate the model's prediction accuracy for the heat release onset temperature, and output the model prediction accuracy index to verify the model's reliability.

[0024] In this embodiment, in step S3, a ring-shaped coplanar waveguide sensor array is embedded in the side wall of the distillation heating chamber to collect real-time data on the dynamic changes of the complex permittivity of the current batch of mixture at the center frequencies of 45 MHz, 433 MHz, and 2.45 GHz, with a sampling frequency of not less than 10 Hz. Figure 2 As shown, the specific steps include the following: S3.1: Obtain the layout parameters of the ring coplanar waveguide sensor array based on the RF circuit design specifications, and generate a sensor hardware structure with three adjustable frequencies to provide the ability to measure the complex permittivity at the center frequencies of 45 MHz, 433 MHz and 2.45 GHz. S3.2: The completed annular coplanar waveguide sensor array is embedded and installed on the side wall of the distillation heating chamber. The reliability of the installation position and material contact is ensured by vacuum sealing process, so as to realize non-destructive online monitoring of the dielectric properties of the mixture. S3.3: Configure the sensor operating frequency according to the preset center frequency parameters of 45 MHz, 433 MHz and 2.45 GHz, and adjust the excitation signal spectrum through the radio frequency signal generator to generate dedicated measurement frequency points for different polarization mechanisms; S3.4: Perform real-time data acquisition on the mixed liquid material in the heating chamber, and continuously measure the dynamic changes of the real and imaginary parts of the complex dielectric constant at a sampling frequency of not less than 10 Hz based on an embedded microwave sensor to obtain the raw dielectric response data stream reflecting the state of the volatile components of the raw material; S3.5: Input the original dielectric response data stream into the digital signal processor, use a finite impulse response filter for noise suppression and perform clock synchronization processing to output dynamic change data of complex dielectric constant that meets the input requirements of the subsequent neural network model.

[0025] like Figure 3 As shown, in this embodiment, in step S4, the dynamic change data of the complex permittivity collected in step S3 is input into a lightweight neural network model trained based on the dielectric fingerprint-thermal response feature correlation matrix of S2 to obtain the heat release start temperature of the current batch of raw materials. The distribution prediction includes the following steps: Based on historical sample data in the dielectric fingerprint-thermal response feature correlation matrix of step S2, a lightweight neural network architecture is used for training and processing to generate a spectrum of complex dielectric constant characteristics and heat release onset temperature. A lightweight neural network model for mapping relationships; S4.1: Perform time-series synchronization processing on the dynamic change data of the real part and imaginary part of the dielectric constant of the current batch collected in step S3 to obtain the original dataset that matches the real-time operating conditions of the distillation heating chamber. S4.2: Perform frequency domain adaptive filtering on the raw dataset obtained in step S4.1 to eliminate environmental electromagnetic noise interference and extract effective data of characteristic frequency bands at the center frequencies of 45 MHz, 433 MHz and 2.45 GHz; S4.3: Input the effective data of the feature frequency band extracted in S4.2 into the lightweight neural network model, perform feature matching and temperature prediction calculation to obtain the original heat release initiation temperature prediction value; S4.4: Perform probability distribution fitting on the original predicted heat release initiation temperature values ​​obtained in S4.3 to generate predicted heat release initiation temperature distribution values ​​for the current batch of raw materials; S4.5: Perform a confidence assessment based on the predicted heat release initiation temperature distribution generated in S4.4 to verify the reliability of the prediction results within the process safety boundary.

[0026] In a batch of raw materials with a moisture content of 18.5%, a total sugar content of 36.2%, and a nicotine concentration of 1.6 mg / g, the initial temperature of heat release generated in step S4.4 is... The predicted values ​​were 62.7℃ with a standard deviation of 3.2℃. Using the Bayesian confidence interval calculation method, the interval [61.3, 64.1]℃ was obtained. The significance level was... Let it be 0.05, and the calculation formula is as follows: ; in, To predict the mean, The standard normal distribution is at a significance level. The critical value below, To predict the standard deviation, For sample size; As a comprehensive confidence index; Preferably, in step S5, the heat release initiation temperature obtained in step S4 is used. The distributed predicted values ​​are used to generate a segmented RF power-frequency coupling control strategy, which includes the following steps: S5.1: The initial temperature of heat release output in step S4 The predicted distribution values ​​are subjected to temperature sensitivity statistical distribution analysis. The dielectric fingerprint-thermal response feature correlation matrix is ​​used to perform the mapping calculation of the heat release onset temperature of key aroma components, generating the heat release onset temperature. The confidence intervals are divided into sections; based on the industry safety operation boundary, the confidence intervals are optimized by boundary constraints, and the temperature threshold interval [45, 65]℃ of the low-frequency preheating stage is calculated; this temperature threshold interval serves as the start-up trigger condition for the 45MHz radio frequency mode, ensuring that the mixture is uniformly preheated to the target temperature and avoiding uneven thermal field. Furthermore, the mapping calculation of the thermal release initiation temperature of key aroma components is realized through the dielectric fingerprint-thermal response feature correlation matrix mapping algorithm, and the thermal release initiation temperature is obtained. Distribute the information intervals.

[0027] Furthermore, the boundary constraint optimization method is used to truncate and correct the confidence interval within the industry safety boundary, thereby generating the temperature threshold interval for the low-frequency preheating stage.

[0028] The temperature threshold range for the low-frequency preheating stage is calculated using the interval determination formula. :

[0029] in, The temperature threshold range for the low-frequency preheating stage is defined as follows: the lower limit of 45°C is the minimum temperature for uniform preheating, and the upper limit of 65°C is the critical temperature to prevent uneven heat field.

[0030] S5.2: Based on the temperature threshold range and dielectric fingerprint-thermal response characteristic correlation matrix generated in S5.1, the imaginary part of the complex dielectric constant is... Feature point detection processing was performed on the spectrum data to identify the 433 MHz frequency point. Bimodal phase transition characteristic points; based on the temperature sensitivity distribution corresponding to the dielectric phase transition characteristic points, phase transition triggering conditions are modeled to generate phase transition triggering conditions in the mid-frequency excitation stage; these phase transition triggering conditions serve as the basis for switching the 433 MHz radio frequency mode, precisely exciting the dehydration condensation reaction of high-boiling-point aroma precursors. S5.3: Based on the dielectric phase transition feature points identified in step S5.2, the first derivative gradient of the dynamic change data of the real part ε′ of the complex dielectric constant is calculated to detect the trigger threshold of the steep negative gradient of ε′ at the 2.45 GHz frequency point; the trigger condition threshold is optimized according to the temperature change rate corresponding to the steep negative gradient to generate the steep gradient trigger condition in the high-frequency focusing stage; the steep gradient trigger condition serves as the basis for pulse modulation triggering in the 2.45 GHz radio frequency mode to ensure that a millimeter-level local high temperature zone is formed at the vapor generation interface; The real part of the complex permittivity output for the dielectric phase transition characteristic points identified in step S5.2 For dynamically changing data, the first-order derivative gradient calculation method is used to calculate the gradient at the 2.45 GHz center frequency. Continuous estimation of the rate of change.

[0031] S5.4: Integrates the temperature threshold range generated in S5.1, the phase transition triggering condition generated in S5.2, and the steep gradient triggering condition generated in S5.3, performs RF power-frequency coupling parameter mapping processing, calculates the three-stage power ramp curve and frequency switching logic; dynamically coordinates and optimizes the coupling parameters based on preset process safety constraints, and generates a segmented RF power-frequency coupling control strategy table; this strategy table defines the complete execution sequence of low-frequency deep penetration preheating, mid-frequency selective excitation, and high-frequency local focusing, realizing dynamic adaptive control of the gradient heating process; Based on the temperature threshold range generated by S5.1, the phase transition triggering condition generated by S5.2, and the steep gradient triggering condition generated by S5.3, a multi-dimensional parameter mapping algorithm is used to realize the correlation mapping processing of three-stage RF power and frequency. Furthermore, a frequency switching logic generation algorithm is adopted to realize a bidirectional index structure between triggering conditions and frequency modes at different stages, and to generate a frequency switching logic table that can be called by the controller in real time.

[0032] Furthermore, dynamic coordination optimization of coupling parameters is performed through a process safety constraint optimization algorithm to ensure that power and frequency switching does not cause excessive degradation of thermosensitive components.

[0033] Furthermore, by employing a method for generating an RF power-frequency coupling strategy table, the optimized power ramp curve and frequency switching logic are integrated into a unified table structure, forming a segmented RF power-frequency coupling control strategy table that includes complete execution sequences for each stage: low-frequency deep penetration preheating, mid-frequency selective excitation, and high-frequency local focusing.

[0034] In this embodiment, in step S6, based on the control strategy generated in S5, three-stage gradient radio frequency heating is performed: When heat release begins temperature When the predicted distribution value falls within the range of [45, 65]℃, 45 MHz low-frequency deep penetration preheating is initiated; When a frequency of 433 MHz is detected Selective excitation at 433 MHz intermediate frequency during bimodal splitting; When a steep negative gradient appears at the 2.45 GHz frequency point ε′, 2.45 GHz high-frequency localized focused pulse modulation is triggered; specifically, the following steps are included: S6.1: The initial temperature of heat release obtained from S4 The predicted distribution value is subjected to threshold judgment processing in the range of [45, 65]℃ to identify the start conditions of the low-frequency preheating stage; a 45 MHz low-frequency deep penetration preheating start command is generated based on the judgment result; the radio frequency generator is controlled to output a 45 MHz radio frequency signal at a constant power of 1.2 kW; low-frequency preheating treatment is performed on the mixture; the heat field is homogenized to 60℃ to ensure the consistency of temperature distribution in the preheating stage; S6.2: The imaginary part of the complex permittivity at the 433 MHz frequency point collected by S3. Dynamic data is processed for bimodal splitting feature identification to detect the initial phase separation point of the glycerol-propylene glycol co-solubilized system; based on the identification results, a 433 MHz mid-frequency selective excitation switching command is generated; the radio frequency generator is controlled to switch the output frequency to 433 MHz and execute a power ramp-up to 2.8 kW; mid-frequency selective excitation is performed on the mixture; the dehydration condensation reaction of high-boiling-point aroma precursors is precisely excited to avoid thermal degradation loss; Furthermore, a second-order difference sequence of the imaginary part spectrum is calculated using a difference spectrum detection algorithm. Two local extrema are located in the detection sequence, and the following extrema detection function is used. Verify the feature points:

[0035] in, and These are the temperatures of the first and second peaks of the bimodal split, used to determine the degree of splitting between peaks; The center frequency.

[0036] Furthermore, the state-space trajectory of the imaginary part of the complex permittivity is generated by the phase space reconstruction method, and the dynamic instability of phase separation corresponding to the bimodal split is evaluated by combining the Lyapunov exponent calculation, thereby confirming the initial phase separation temperature of the glycerol-propylene glycol co-solubilized system.

[0037] By using the trigger condition determination function, the phase separation initial point temperature is matched with the phase transition trigger condition of the intermediate frequency excitation stage obtained by S5.2 modeling, so as to realize the automatic generation of the 433 MHz intermediate frequency selective excitation switching command. The command is sent to the frequency switching module of the RF generator via the control bus to execute the parameter setting of switching the output frequency from 45 MHz to 433 MHz and rising to 2.8 kW according to the power ramp curve.

[0038] By controlling the dynamic power loading, the medium-frequency selective excitation heating treatment of the mixture is achieved, which triggers the dehydration condensation reaction of high-boiling-point aroma precursors at the optimal temperature gradient and inhibits the occurrence of thermal degradation pathways in the process.

[0039] S6.3: Perform steep negative gradient detection processing on the dynamic change data of the real part ε′ of the complex permittivity at the 2.45 GHz frequency point collected in step S3 to capture the starting point of violent volatilization of low molecular weight aldehydes and ketones; generate a 2.45 GHz high-frequency local focusing pulse modulation trigger command based on the detection results; control the RF generator to switch to the 2.45 GHz high-frequency mode and execute a 30% duty cycle pulse modulation to output a peak RF signal of 3.5 kW; perform high-frequency local focusing processing on the mixture; accelerate the vaporization of the vapor generation interface and suppress the residence time >2s, reducing the degradation of heat-sensitive components.

[0040] The instantaneous rate of change of the real part ε′ of the complex dielectric constant at the center frequency of 2.45 GHz was extracted from the data collected by S3 using the first derivative gradient calculation method.

[0041] S6.4: Based on the segmented RF power-frequency coupling control strategy input generated by S5, the power allocation ratio of each frequency band is calculated in real time through the PID-Fuzzy composite algorithm; the power ramp or pulse modulation parameters of the current heating stage are optimized; RF power dynamic coordination commands are generated; closed-loop power allocation control is implemented for the RF generator; and the recovery rate of target aroma components and the retention rate of heat-sensitive components in the steam condensate are maintained within the preset process range. Furthermore, by employing a combined PID and fuzzy control algorithm, the instantaneous error suppression result is weighted and fused with the nonlinear adjustment coefficient to calculate the power allocation ratio vector for each frequency band. : ; in, These correspond to the real-time power ratios for the low-frequency, mid-frequency, and high-frequency stages, respectively.

[0042] S6.5: Perform heating cavity field distribution optimization processing on the RF power-frequency coupling command sequence generated in S6.1 to S6.4 to form a gradient RF field spatial distribution; coordinate the output timing of the RF generator based on the optimization results; implement three-stage seamless switching control; complete the gradient RF heating process; ensure that the temperature gradient control accuracy of the distillation process meets the process requirements and improves the extraction efficiency of aroma components.

[0043] In this embodiment, step S7 involves monitoring the concentration of the target aroma component in the vapor condensate during step S6, calculating the actual recovery rate and the retention rate of heat-sensitive components using HPLC quantitative analysis, and generating a model bias correction factor. Specifically, this includes: S7.1: Heat release initiation temperature obtained based on S4 The predicted values ​​are distributed, and the model bias correction factor is defined as the predicted heat release onset temperature. The normalized deviation coefficient between the distribution and the actual target aroma component recovery rate is used to characterize the systematic error of the prediction model; S7.2: Obtain the vapor condensate generated during step S6 as the monitoring object, and collect representative samples through an automated sampling device to ensure the reliability of subsequent HPLC quantitative analysis; S7.3: Perform high performance liquid chromatography (HPLC) quantitative analysis on the representative sample of vapor condensate obtained in step S7.2 to obtain the concentration data of the target aroma components; S7.4: Based on the target aroma component concentration data obtained in step S7.3, calculate the actual target aroma component recovery rate and the heat-sensitive component retention rate. The recovery rate is calculated by comparing the target component concentrations in the input raw material and the condensate, and the retention rate is determined by the ratio of the heat-sensitive component concentration to the benchmark value. S7.5: Based on the actual target aroma component recovery rate value, heat-sensitive component retention rate value calculated in step S7.4, and the model deviation correction factor defined in S7.1, generate a specific model deviation correction factor value for updating the model parameters. This value is calculated by weighted average actual recovery rate deviation and retention rate deviation.

[0044] Based on the actual target aroma component recovery rate value and heat-sensitive component retention rate value calculated in step S7.4, and the initial value of the model deviation correction factor defined in step S7.1 as input conditions, a normalized deviation calculation method is used to realize the recovery rate deviation. and retention rate deviation Dimensionless processing.

[0045] Furthermore, the model bias correction factor value is calculated using a weighted average method, and a target correction value for updating model parameters is generated. The weighted average formula is expressed as follows:

[0046] in, This is the final model bias correction factor value. , These are the weighting coefficients for recovery deviation and retention deviation, respectively. , These are the normalized deviation values.

[0047] In this embodiment, in step S8, the model bias correction factor generated in step S7 is fed back to the dielectric fingerprint-thermal response feature correlation matrix in step S2, and the multidimensional mapping model parameters are dynamically updated to optimize the heat release initiation temperature of subsequent batches. Prediction accuracy, specifically including: S8.1: Extract data from the model bias correction factor output in step S7. Based on the deviation between the actual recovery rate and the retention rate of the heat-sensitive component obtained by HPLC quantitative analysis, calculate the quantitative characterization of the prediction error of the current batch and use the generated parameter deviation vector as the input for model update.

[0048] S8.2: Based on the parameter deviation vector generated in step S8.1, feature space reconstruction is performed through incremental principal component analysis algorithm to calculate the correction amount of each principal component loading vector in the dielectric fingerprint-thermal response feature correlation matrix, so as to quantify the influence of the deviation between model parameters and actual distillation performance. S8.3: Based on the correction calculated in S8.2, the model parameter update is generated using a partial least squares regression incremental learning mechanism. The mapping coefficients of the dielectric fingerprint-thermal response feature correlation matrix are dynamically adjusted to optimize the multidimensional mapping model for the heat release initiation temperature. The predictive power of the distribution; S8.4: Inject the model parameter update amount generated in step S8.3 into the dielectric fingerprint-thermal response feature correlation matrix in step S2, and perform dynamic correction of matrix parameters through a weighted fusion algorithm to update the correlation function of the multidimensional mapping model; S8.5: Based on the dielectric fingerprint-thermal response characteristic correlation matrix corrected in S8.4, the thermal release initiation temperature of the new batch of raw materials. The predicted values ​​were cross-validated, and the effectiveness of the model update was evaluated by calculating the standard deviation of the prediction error, in order to verify the heat release onset temperature. The prediction accuracy is improved and the next round of optimization trigger signal is generated.

[0049] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0050] Unless otherwise defined, the technical or scientific terms used herein should be understood in their ordinary sense by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” involved in the embodiments of this invention refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0051] The above description is merely an exemplary embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for processing sun-dried tobacco through distillation, characterized in that, Includes the following steps: S1: Collect GC-MS spectra of water content, total sugar content, nicotine concentration and aroma components of multiple batches of sun-dried tobacco extract-propylene glycol-glycerol mixture samples, and measure the real and imaginary parts of their complex permittivity in the 10MHz–2.45GHz frequency band under controlled temperature and humidity conditions; S2: Based on the spectral data of the real and imaginary parts of the complex dielectric constant, a multidimensional mapping model is established through principal component analysis and partial least squares regression to form a unique dielectric fingerprint-thermal response characteristic correlation matrix for each batch of raw materials; S3: An annular coplanar waveguide sensor array is embedded in the side wall of the distillation heating chamber to collect real-time data on the dynamic change of the complex permittivity of the current batch of mixture at the center frequencies of 45MHz, 433MHz and 2.45GHz; S4: Input the dynamic change data of complex dielectric constant into a lightweight neural network model trained based on the dielectric fingerprint-thermal response feature correlation matrix to obtain the predicted value of the heat release initiation temperature distribution of the current batch of raw materials; S5: Generate a segmented RF power-frequency coupling control strategy based on the predicted value of the heat release initiation temperature distribution; S6: Based on the control strategy, three-stage gradient radio frequency heating is performed; S7: Monitor the concentration of target aroma components in the steam condensate during the heating process, calculate the actual recovery rate and retention rate of heat-sensitive components by HPLC quantitative analysis, and generate a model bias correction factor; S8: Feedback the model bias correction factor to the dielectric fingerprint-thermal response feature correlation matrix to dynamically update the multidimensional mapping model parameters.

2. The method for processing sun-dried tobacco by distillation according to claim 1, characterized in that, In step S2, the spectrum data of the real part and the spectrum data of the imaginary part of the complex permittivity are cleaned and standardized to obtain a standardized spectrum dataset of the real and imaginary parts of the complex permittivity; and the principal component analysis algorithm (PCA) is applied to extract key principal components and generate a principal component score matrix.

3. The method for processing sun-dried tobacco by distillation according to claim 1, characterized in that, In step S2, the principal component score matrix and the heat release initiation temperature data are input into the partial least squares regression algorithm PLSR to establish the mapping relationship between complex dielectric properties and thermal response behavior, and to calculate the parameters of the partial least squares regression model. Using the parameters of the partial least squares regression model, a multidimensional mapping model between the spectrum of complex dielectric constant and the heat release initiation temperature is constructed to form a unique dielectric fingerprint-thermal response feature correlation matrix for each batch of raw materials. We performed k-fold cross-validation on the dielectric fingerprint-thermal response feature correlation matrix to evaluate the prediction accuracy of the multidimensional mapping model for the heat release initiation temperature.

4. The method for processing sun-dried tobacco by distillation according to claim 1, characterized in that, In step S3, the sensor operating frequency is configured according to the preset center frequency parameters of 45MHz, 433MHz and 2.45GHz, and the excitation signal spectrum is adjusted by the radio frequency signal generator to generate dedicated measurement frequency points for different polarization mechanisms. Real-time data acquisition is performed on the mixed liquid material in the heating chamber. Based on the embedded microwave sensor, the dynamic changes of the real and imaginary parts of the complex dielectric constant are continuously measured at a sampling frequency of not less than 10Hz to obtain the raw dielectric response data stream reflecting the state of the volatile components of the raw material. The raw dielectric response data stream is input into a digital signal processor, where a finite impulse response filter is used for noise suppression and clock synchronization.

5. The method for processing sun-dried tobacco by distillation according to claim 1, characterized in that, Step S4 specifically includes: The dynamic change data of the real part and imaginary part of the dielectric constant of the current batch are processed in a time-series synchronization manner to obtain the original dataset that matches the real-time operating conditions of the distillation heating chamber. Frequency domain adaptive filtering was performed on the original dataset to extract effective data of the characteristic frequency bands at the center frequencies of 45MHz, 433MHz and 2.45GHz; The effective data of the characteristic frequency band is input into the lightweight neural network model to perform feature matching and temperature prediction calculations in order to obtain the original heat release initiation temperature prediction value. The original predicted heat release initiation temperature is subjected to probability distribution fitting to generate the predicted heat release initiation temperature distribution of the current batch of raw materials; A confidence assessment was performed based on the predicted heat release initiation temperature distribution to verify the reliability of the prediction results within the process safety boundary.

6. The method for processing sun-dried tobacco by distillation according to claim 1, characterized in that, In step S5, the thermal release initiation temperature mapping calculation of key aroma components is performed using the dielectric fingerprint-thermal response feature correlation matrix to generate a confidence interval for thermal release initiation temperature distribution; based on the industry safety operation boundary, boundary constraint optimization processing is performed on the confidence interval to calculate the temperature threshold interval [45, 65]℃ for the low-frequency preheating stage.

7. The method for processing sun-dried tobacco by distillation according to claim 6, characterized in that, In step S5, based on the temperature threshold range and the dielectric fingerprint-thermal response feature correlation matrix, feature point detection processing is performed on the imaginary part spectrum data of the complex dielectric constant to identify the dielectric phase transition feature point with double peak splitting of the imaginary part spectrum of the complex dielectric constant at the 433 MHz frequency point; the phase transition triggering condition is modeled according to the temperature sensitivity distribution corresponding to the dielectric phase transition feature point to generate the phase transition triggering condition of the mid-frequency excitation stage. Based on the dielectric phase transition characteristic points, the first derivative gradient of the dynamic change data of the real part of the complex dielectric constant is calculated, and the trigger threshold of the steep negative gradient of the real part of the complex dielectric constant at the 2.45 GHz frequency point is detected. The trigger condition threshold is optimized according to the temperature change rate corresponding to the steep negative gradient, and the steep gradient trigger condition in the high-frequency focusing stage is generated. Perform RF power-frequency coupling parameter mapping processing to calculate the three-stage power ramp curve and frequency switching logic.

8. The method for processing sun-dried tobacco by distillation according to claim 7, characterized in that, In step S6, the three-stage gradient radio frequency heating is as follows: When the predicted value of the heat release initiation temperature distribution falls within the range of [45, 65]℃, 45 MHz low-frequency deep penetration preheating is initiated. When a double-peak split of the imaginary part spectral number of the complex permittivity at a frequency of 433 MHz is detected, switch to selective excitation at the 433 MHz intermediate frequency; When a steep negative gradient appears in the real part of the complex permittivity at the 2.45 GHz frequency, 2.45 GHz high-frequency localized focused pulse modulation is triggered.

9. The method for processing sun-dried tobacco by distillation according to claim 8, characterized in that, Step S7 specifically includes the following steps: Based on the predicted value of the heat release onset temperature distribution, the model bias correction factor is defined as the normalized deviation coefficient between the predicted heat release onset temperature distribution and the actual target aroma component recovery rate, in order to characterize the systematic error of the prediction model. The vapor condensate generated during the execution process is used as the monitoring object. Representative samples are collected through an automated sampling device and subjected to high performance liquid chromatography (HPLC) quantitative analysis to obtain the concentration data of the target aroma components. Based on the target aroma component concentration data, the actual target aroma component recovery rate and the heat-sensitive component retention rate are calculated; Based on the actual target aroma component recovery rate, heat-sensitive component retention rate, and model bias correction factor, a specific model bias correction factor value is generated for updating the model parameters.

10. The method for processing sun-dried tobacco by distillation according to claim 9, characterized in that, The model bias correction factor is used in incremental principal component analysis and PLSR incremental learning mechanism to dynamically correct the parameter loads and regression coefficients in the multidimensional mapping model, so as to achieve adaptive optimization and iterative update of the prediction accuracy of the heat release initiation temperature of subsequent batches.

Images