Molecular distillation process parameter value determination method, device, equipment, medium and product

Abstract

The invention provides a molecular distillation process parameter value determination method and device, equipment, a medium and a product. The method comprises the steps that multiple process parameters in the molecular distillation process and distillation results of the multiple process parameters under a preset parameter value combination are obtained; screening out at least two key process parameters from the plurality of process parameters based on the distillation result; constructing a prediction model comprising at least two sub-models in a neural network, a random forest and a support vector machine based on distillation results of the at least two key process parameters under a preset parameter value combination; wherein each sub-model is used for predicting an initial parameter value combination of the key process parameters based on an expected distillation result, and the prediction model is used for determining a final parameter value combination based on each initial parameter value combination. According to the method and the device, the molecular distillation process parameter values are automatically determined, the problems of low control precision and poor production efficiency caused by parameter coupling are effectively solved, and the purity and the yield of molecular distillation products can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method, apparatus, equipment, medium, and product for determining molecular distillation process parameter values. Background Technology

[0002] Molecular distillation, also known as short-path distillation, is characterized by low operating temperature, low distillation pressure, and short heating time. It is suitable for the extraction, separation, and purification of high-boiling-point, heat-sensitive, and high-viscosity substances, offering unparalleled advantages over conventional distillation techniques in the deodorization and purification of spices. However, due to the numerous, coupled, nonlinear, and hysteresis-related process parameters inherent in molecular distillation systems, traditional non-automated methods for determining these parameters cannot meet the high-efficiency and energy-saving requirements of modern industrial production, thus limiting the maximum capacity of molecular distillation.

[0003] Therefore, how to automatically determine the process parameters of molecular distillation has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, medium, and product for determining molecular distillation process parameters, in order to solve the technical problem of how to automatically determine molecular distillation process parameters in related technologies.

[0005] In a first aspect, this application provides a method for determining molecular distillation process parameters, including: Obtain multiple process parameters during molecular distillation, as well as the distillation results of multiple process parameters under preset parameter value combinations; Based on the distillation results, at least two key process parameters are selected from a plurality of process parameters; Based on the distillation results of at least two of the key process parameters under preset parameter value combinations, a prediction model is constructed that includes at least two sub-models among neural networks, random forests, and support vector machines. Each of the sub-models is used to predict the initial parameter value combination of the key process parameters based on the expected distillation results, and the prediction model is used to determine the final parameter value combination based on each of the initial parameter value combinations.

[0006] In some embodiments, the step of selecting at least two key process parameters from a plurality of process parameters based on the distillation results includes: Construct a covariance matrix for multiple process parameters under preset parameter value combinations; The eigenvalues ​​and eigenvectors of the covariance matrix are obtained based on the coupling relationship between multiple process parameters. Process parameters whose cumulative variance contribution rate is greater than or equal to a preset contribution rate threshold are selected as candidate key process parameters. The key process parameters among the candidate key process parameters are determined based on the eigenvalues ​​and eigenvectors.

[0007] In some embodiments, the construction of a prediction model comprising at least two sub-models among neural networks, random forests, and support vector machines, based on the distillation results of at least two of the key process parameters under preset parameter value combinations, includes: Based on the distillation results of at least two key process parameters under preset parameter value combinations, the hyperparameter value combinations of the at least two sub-models are determined within a preset hyperparameter search space; the prediction model is constructed based on the sub-model hyperparameter value combinations. Wherein, when the prediction model includes the neural network, the combination of hyperparameter values ​​of the sub-model includes the number of hidden layer neurons and the learning rate; When the prediction model includes the random forest, the combination of hyperparameter values ​​for the sub-model includes the number of decision trees and the maximum depth; When the prediction model includes the support vector machine, the combination of hyperparameter values ​​for the sub-model includes a penalty coefficient and a kernel parameter.

[0008] In some embodiments, the prediction model is used to determine a final combination of parameter values ​​based on each of the initial parameter value combinations, including: The optimization function value for each combination of initial parameter values ​​is determined based on a preset optimization function, and each optimization function value is mapped to multiple excellence levels; Based on the trust level of the at least two sub-models at each of the excellence levels, determine the overall trust level for each of the excellence levels; The final parameter value combination is obtained by weighted summing of each initial parameter value combination corresponding to the highest overall trust level.

[0009] In some embodiments, after constructing the prediction model including at least two sub-models among neural networks, random forests, and support vector machines, the method further includes: The prediction model is integrated into the control system of the molecular distillation equipment; The control system controls the operation of the molecular distillation equipment based on the final parameter value combination output by the prediction model, and sends the distillation result corresponding to the final parameter value combination to the prediction model.

[0010] In some embodiments, the method further includes: If the difference between the distillation result corresponding to the final parameter value combination and the expected distillation result is greater than a preset deviation, the prediction model is updated; and if the latest cumulative variance contribution rate of the key process parameter is less than or equal to a preset contribution rate threshold, the key process parameter is reselected from among the multiple process parameters.

[0011] Secondly, this application provides a molecular distillation process parameter determination apparatus, comprising: The acquisition module is used to acquire multiple process parameters in the molecular distillation process, as well as the distillation results of the multiple process parameters under a preset parameter value combination; A screening module is used to screen at least two key process parameters from a plurality of process parameters based on the distillation results; A construction module is used to construct a prediction model, including at least two sub-models among neural networks, random forests and support vector machines, based on the distillation results of at least two of the key process parameters under preset parameter value combinations. Each of the sub-models is used to predict the initial parameter value combination of the key process parameters based on the expected distillation results, and the prediction model is used to determine the final parameter value combination based on each of the initial parameter value combinations.

[0012] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to implement the above-described method when executing the program through the computer program.

[0013] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0014] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0015] The molecular distillation process parameter value determination method, apparatus, equipment, medium and product provided in this application embodiment screens key process parameters from multiple process parameters in the molecular distillation process and predicts the parameter value combination of key process parameters through multi-sub-model integration, thereby realizing the automated and intelligent determination of molecular distillation process parameter values. It effectively overcomes the problems of low control accuracy and poor production efficiency caused by parameter coupling in traditional methods, and can improve the purity and yield of molecular distillation products. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts illustrating the method for determining molecular distillation process parameters provided in this application.

[0018] Figure 2 This is one of the schematic diagrams illustrating the importance of various process parameters provided in the embodiments of this application.

[0019] Figure 3 This is the second flowchart illustrating the method for determining molecular distillation process parameters provided in the embodiments of this application.

[0020] Figure 4 This is the second schematic diagram illustrating the importance of various process parameters provided in the embodiments of this application.

[0021] Figure 5 A schematic diagram of the molecular distillation process parameter determination device provided in the embodiments of this application.

[0022] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0025] The method for determining molecular distillation process parameters provided in this application is applicable to terminals, which can be various electronic devices with displays and web browsing capabilities, including servers, smartphones, tablets, laptops, and desktop computers.

[0026] Figure 1 This is one of the flowcharts illustrating the method for determining molecular distillation process parameters provided in the embodiments of this application, such as... Figure 1 As shown, the method includes steps 110, 120, and 130. These method steps are merely one possible implementation of this application.

[0027] Step 110: Obtain multiple process parameters during molecular distillation, as well as the distillation results under preset parameter value combinations.

[0028] Specifically, the execution subject of the molecular distillation process parameter determination method provided in this application embodiment is a molecular distillation process parameter determination device. This device can be a hardware device independently set in the terminal, or it can be a software program running in the terminal.

[0029] Process parameters refer to the controllable operational variables that affect the separation and purification effect during molecular distillation, such as distillation temperature, feed rate, scraper rotation speed, vacuum level, and condensation temperature. Preset parameter combinations are a set of specific process parameter values ​​set for each experiment. Distillation results refer to the quantitative indicators obtained by detecting the molecular distillation products after each experiment, such as the content, purity, or yield of the target component.

[0030] Multiple process parameters affecting molecular distillation were collected. Through single-factor experimental design, different preset parameter value combinations were set for these process parameters and experiments were conducted to obtain the corresponding distillation results.

[0031] For example, when purifying spices by molecular distillation, due to their prominent natural properties, complex composition, and the presence of various heat-sensitive substances, their stability is significantly affected by factors such as light, heat, and pH value. Therefore, the range of selected process parameters and preset parameter values ​​can include at least the following: distillation temperature 30-150℃, feed rate 0.5-5mL / min, scraper rotation speed 100-500rpm, system vacuum degree 0.1-10Pa, and condensation temperature -20 to 50℃.

[0032] The embodiments of this application are applicable to the refining process of heat-sensitive plant raw materials such as rosemary extract and Sichuan pepper oil. These raw materials typically contain volatile oils, alkaloids, flavonoids, polyphenols, and terpenes.

[0033] Step 120: Based on the distillation results, select at least two key process parameters from multiple process parameters.

[0034] Specifically, since there are many process parameters and they influence each other, in order to improve the efficiency and accuracy of predicting process parameter values, it is necessary to screen out the parameters that have a significant impact on the distillation results from all process parameters. These screened parameters are called key process parameters, which are process parameters that play a decisive or dominant role in the distillation results.

[0035] This step uses industrial data modeling software to standardize or normalize the data collected in step 110 to eliminate the influence of different parameter dimensions. Then, it combines Principal Component Analysis (PCA) as a dimensionality reduction technique to screen out key process parameters.

[0036] Step 130: Based on the distillation results of at least two key process parameters under preset parameter value combinations, construct a prediction model including at least two sub-models among neural networks, random forests, and support vector machines; Each sub-model is used to predict the initial parameter value combination of key process parameters based on the expected distillation results, and the prediction model is used to determine the final parameter value combination based on each initial parameter value combination.

[0037] Specifically, the desired distillation result refers to the pre-set, desired product quality indicators, such as a target component content greater than or equal to 95%.

[0038] By automatically optimizing the model using the Bayesian optimization algorithm, a predictive model is constructed that integrates at least two sub-models from Neural Network (NN), Random Forest (RF), and Support Vector Machine (SVM).

[0039] Neural networks (NNs) are mathematical or computational models that mimic the structure and function of biological neural networks, used to estimate or approximate functions. Randomized Randomized Forests (RFs) are classifiers that consist of multiple decision trees, and the class of their output is determined by the mode of the classes output by the individual trees. Supervised Virtual Machines (SVMs) are supervised learning models and related learning algorithms used in classification and regression analysis to analyze data.

[0040] Each seed model uses selected key process parameters as independent variables and distillation results such as target component content as dependent variables. It learns the relationship between key process parameters and distillation results, and can predict the initial parameter value combination required to achieve the desired distillation result. The prediction model then fuses these initial parameter value combinations, for example, through weighted summation, to obtain the optimal key process parameter value combination, i.e., the final parameter value combination. The final parameter value combination can include only key process parameter values ​​or other process parameter values, and can be configured according to actual conditions.

[0041] This method breaks through the limitations of traditional methods that rely on experience and repeated trial and error. It can efficiently and accurately complete the transfer and optimization of process parameters from laboratory pilot tests to industrial production, and is particularly suitable for the efficient refining of complex spices such as Sichuan pepper oil and rosemary oil.

[0042] The molecular distillation process parameter determination method provided in this application realizes the automated and intelligent determination of molecular distillation process parameter values ​​by screening key process parameters from multiple process parameters in the molecular distillation process and predicting the parameter value combination of key process parameters through multi-sub-model integration. It effectively overcomes the problems of low control accuracy and poor production efficiency caused by parameter coupling in traditional methods, and can improve the purity and yield of molecular distillation products.

[0043] It should be noted that each implementation method of this application can be freely combined, rearranged, or executed individually, and does not need to rely on or depend on a fixed execution order.

[0044] In some embodiments, step 120 includes: Construct the covariance matrix of multiple process parameters under preset parameter value combinations; The eigenvalues ​​and eigenvectors of the covariance matrix are obtained based on the coupling relationship between multiple process parameters; Process parameters whose cumulative variance contribution rate is greater than or equal to a preset contribution rate threshold are selected as candidate key process parameters. The key process parameters among the candidate key process parameters are determined based on eigenvalues ​​and eigenvectors.

[0045] Specifically, if the latest cumulative variance contribution rate of a key process parameter is less than or equal to a preset contribution rate threshold, the key process parameter is reselected from among multiple process parameters.

[0046] Specifically, the covariance matrix is ​​a mathematical matrix used to quantify the degree of synchronous change between any two process parameters. Coupling relationship refers to the degree to which a change in one process parameter affects another. Eigenvalues ​​and eigenvectors are mathematical quantities describing the main direction and intensity of data variation. The variance contribution rate refers to the percentage of total data variation explained by the first few principal components.

[0047] A covariance matrix is ​​constructed with preset parameter value combinations as row samples and multiple process parameters as column variables. The covariance matrix is ​​calculated to quantify the parameter coupling relationship and the matrix is ​​updated periodically. The eigenvalues ​​and eigenvectors of the covariance matrix are solved using MATLAB R2023a. Based on the cumulative variance contribution rate being greater than or equal to a preset contribution rate threshold, a number of principal components are retained, i.e., process parameters with a cumulative variance contribution rate greater than or equal to the preset contribution rate threshold are retained. These first few process parameters are selected as candidate key process parameters. Alternatively, the number of parameters to be retained can be set; for example, the first two process parameters with a cumulative variance contribution rate greater than or equal to the preset contribution rate threshold can be selected as candidate key process parameters. The factor loadings of the candidate key process parameters are calculated. , , For the feature vector elements, i.e., candidate key process parameters, The characteristic value is used to filter key process parameters based on the mean absolute value of the load being greater than a preset load threshold. Since different key process parameters have different levels of importance, the mean absolute value of the load can be normalized to obtain an importance score for each key process parameter. , ), The sign is multiplication. If the latest cumulative variance contribution rate of the key process parameters is less than or equal to the preset contribution rate threshold, the covariance matrix, eigenvalues, and loading values ​​will be automatically recalculated to dynamically update the key process parameters and the prediction model simultaneously.

[0048] Taking the purification of rosemary extract as Example 1, the specific steps for determining the key process parameters are as follows: Materials and equipment were configured. Rosemary extract (containing heat-sensitive antioxidants) was used as the material.

[0049] The instrumentation and software system include: a short-path molecular distillation apparatus, model KDL5, manufactured by UIC GmbH, Germany, equipped with an online sensor array, such as a distillation temperature sensor, feed rate flow meter, scraper speed encoder, vacuum transmitter, and condensation temperature sensor. An online detection system, integrating an online injection module with a gas chromatography-mass spectrometry (GC-MS) and high-performance liquid chromatography (HPLC) system, monitors distillate composition in real time. An intelligent control system includes an UnscramblerX10.4 data preprocessing module, a MATLAB R2023a prediction model engine, a programmable logic controller (PLC) unit, and a human-machine interface (HMI).

[0050] The core process parameters and level design of this application embodiment are shown in Table 1 below, where level refers to parameter value.

[0051] Table 1. Multiple process parameters and their corresponding values.

[0052] Figure 2 One of the schematic diagrams illustrating the importance of various process parameters provided in the embodiments of this application is shown below. Figure 2 As shown, the importance of each process parameter is determined using the PCA algorithm.

[0053] The online control architecture is configured such that the sensor group collects parameter data in real time and transmits it to the intelligent control system via a 485 bus. After data standardization and PCA processing, the Bayesian optimization prediction model driven by key process parameters calculates the optimal combination of process parameter values ​​in real time and outputs them to the PLC unit to regulate the actuators, heating module, feed pump, scraper motor, vacuum pump, and condensation circulation device. At the same time, the cumulative variance contribution rate of the first two principal components is recalculated every 10 minutes. If it is <70% or the target component, such as rosmarinic acid or limonene, has a deviation >10%, the covariance matrix, eigenvalues, and loading values ​​are automatically recalculated, and the key process parameters are dynamically updated.

[0054] The automated production process includes: initial parameter setting, which starts with intermediate-level parameters by default, such as distillation temperature 90℃, feed rate 2.5mL / min, scraper rotation speed 300rpm, vacuum degree 4.0Pa, and condensation temperature 20℃. Target components, such as rosmarinic acid content, are preset and increased through HMI. Online single-factor optimization mode, which fixes other process parameters at intermediate levels and performs gradient adjustments on individual process parameters one by one, such as automatically running 3 batches at each level. HPLC is used to detect the rosmarinic acid content in the distillate online, GC-MS is used to simultaneously analyze the volatile composition, and the sensory evaluation module automatically collects odor data, constructing a raw dataset containing 50 sets of data and storing it in the database. Real-time closed-loop control mode, which involves data preprocessing, standardization, and using the PCA algorithm to obtain key process parameters. The specific screening process is as follows: The software performs Z-score standardization on the sensor data, using the following formula: ,in, x The original value, The processed value. For real-time moving average, To determine the real-time standard deviation, an n-row × 5-column covariance matrix was constructed, with online experimental data as the row sample and five core process parameters (distillation temperature, feed rate, scraper rotation speed, vacuum degree, and condensation temperature) as the column variables. Outliers were removed using the 3σ principle.

[0055] The 5×5 covariance matrix is ​​calculated to quantify the coupling relationship of parameters. For example, when distillation temperature and vacuum degree are strongly positively correlated, the covariance is close to 0.85. The covariance matrix is ​​updated every 10 minutes. Then, the eigenvalues ​​and eigenvectors of the covariance matrix are solved using MATLAB R2023a. The first two principal components are retained based on a cumulative variance contribution rate of ≥70%, resulting in two candidate key process parameters: distillation temperature and vacuum degree. Subsequently, factor loadings are calculated, and key process parameters are screened based on the mean absolute value of the loadings >0.5. Distillation temperature has a loading of 0.92 in the first principal component and vacuum degree has a loading of 0.89 in the second principal component, both of which meet the threshold. Finally, distillation temperature and vacuum degree are extracted as key principal components. Finally, the mean absolute value of the loadings is normalized to obtain the importance scores of distillation temperature and vacuum degree.

[0056] The molecular distillation process parameter determination method provided in this application combines covariance analysis with principal component loading screening, which can more accurately and physically interpretably identify core influencing factors from numerous coupled process parameters, thereby improving the accuracy of key process parameter screening and the effectiveness of subsequent model construction.

[0057] In some embodiments, step 130 includes: Based on the distillation results of at least two key process parameters under preset parameter value combinations, determine at least two sub-model hyperparameter value combinations within a preset hyperparameter search space; construct a prediction model based on the sub-model hyperparameter value combinations. In the case where the prediction model includes a neural network, the combination of hyperparameter values ​​for the sub-model includes the number of hidden layer neurons and the learning rate. When the prediction model includes random forest, the combination of sub-model hyperparameter values ​​includes the number of decision trees and the maximum depth; When the prediction model includes a support vector machine, the combination of sub-model hyperparameter values ​​includes penalty coefficients and kernel parameters.

[0058] Specifically, the MATLAB engine is used to call at least two sub-models from the NN, RF, and SVM after Bayesian optimization of hyperparameter values, using the test set determination coefficient R0. 2The system optimizes in real time with the goal of maximizing performance, outputting the optimal combination of hyperparameter values ​​under the current operating conditions. After the predictive model is built and optimized, the PLC unit automatically adjusts various actuators, such as the heating module, vacuum pump, and feed pump, according to the control commands of the optimal process parameter combination output by the predictive model. Simultaneously, it corrects sensor deviations through an error compensation algorithm to ensure parameter control accuracy, for example, temperature ±1℃, vacuum ±0.5Pa, and feed rate ±0.1mL / min, thereby achieving closed-loop control from intelligent decision-making to precise execution. Table 2 shows a set of optimal hyperparameter value combinations and related data of the corresponding predictive model.

[0059] Table 2 shows the optimal combination of hyperparameter values ​​and related data of the corresponding prediction model.

[0060] The embodiments of this application will be illustrated below.

[0061] Figure 3 The second schematic flowchart illustrates the method for determining molecular distillation process parameters provided in this application. Figure 3 Key indicators, namely key process parameters, such as Figure 3 As shown, real-time data acquisition and preprocessing are performed first. Distillation temperature is measured using a Pt100 resistance thermometer (accuracy ±0.5℃), vacuum level using a capacitive transmitter (accuracy ±1%FS), feed rate using an electromagnetic flowmeter (accuracy ±0.5%), scraper rotation speed using an incremental encoder (resolution ±1rpm), and condensation temperature using a thermocouple (accuracy ±1℃). All sensor signals are converted to 4-20mA standard signals via an isolation transmitter and uploaded to the control system in real-time via industrial Ethernet, with a sampling frequency of 10Hz. A sliding window filter is used, with a window length of 5s to remove noise. The Z-score normalization formula is: Where μ is the real-time mean and σ is the real-time standard deviation, which are dynamically updated through Exponentially Weighted Moving Average (EWMA).

[0062] PCA is used to calculate the cumulative variance contribution rate in real time. When new data causes the cumulative variance contribution rate of the first two principal components to fall below 70%, the model update program is automatically triggered to re-extract key process parameters.

[0063] Bayesian optimization determines the optimal combination of hyperparameter values ​​for each sub-model through a closed-loop logic of "preliminary configuration - periodic iteration - post-iteration selection". The specific process is as follows: Preliminary Configuration and Hyperparameter Space Construction: First, complete the basic configuration and construct the hyperparameter search space based on the preset hyperparameter value range. For example, the hyperparameter value combinations for NN include the number of hidden layer neurons (10-30), learning rate (0.001-0.1), and number of iterations (500-2000); the hyperparameter value combinations for RF include the number of decision trees (100-500) and maximum depth (5-20); and the hyperparameter value combinations for SVM include the penalty coefficient C (1-100) and kernel parameter γ (0.001-0.1). Input data uses key process parameters after Unscrambler standardization and PCA dimensionality reduction, such as distillation temperature and vacuum degree. Gaussian Process Regression (GPR) is selected as the probabilistic model, and the objective function is defined as the online detected distillation result, such as rosmarinic acid content, and the coefficient of determination R between the model prediction and the model prediction. 2 The optimization objective is to maximize R. 2 .in, , HPLC measured values For the model number i One predicted value, is the mean of the measured values, and m is the total number of predicted values.

[0064] Periodic iterative optimization: 50 iterations of fine-tuning are initiated in 10-minute cycles. The first 2 minutes are based on historical hyperparameter value combinations - R. 2 "Data updates the GPR model. Within 2-3 minutes, the expected improvement (EI) acquisition function is used to screen the combinations of hyperparameters to be tested, balancing predictive performance with exploration uncertainty. Within 3-9 minutes, the hyperparameter values ​​are deployed to the MATLAB model engine, which, in conjunction with HPLC, acquires data every 30 seconds to calculate R." 2 If R 2 If the value is less than 0.90, sensor error compensation will be triggered, and the "hyperparameter value combination - R" will be effectively applied within 9-10 minutes. 2 "Feedback is fed back to the historical dataset for the next iteration. After 50 iterations, models that maintain stable performance, such as achieving three consecutive R-values, are selected." 2 A cluster of hyperparameter values ​​with a value ≥0.95 and a fluctuation ≤2% was established. Then, secondary verification was conducted by combining the economic indicators of industrial production, such as a reduction of raw material waste ≥30% and a reduction in production cycle ≥25%, to finally determine the optimal combination of hyperparameter values ​​for each sub-model and integrate it into the PLC control system.

[0065] Multi-model fusion strategy: By fusing the initial parameter value combinations output by each sub-model through DS evidence theory, the initial parameter value combination with the highest confidence can be selected as the final parameter value combination as the control command, thereby improving the anti-interference capability.

[0066] The control system operates the molecular distillation equipment according to the final parameter values, using a proportional-integral-derivative (PID) algorithm to control the electric heating power of the distillation vessel, with a response time ≤30s. A variable frequency vacuum pump regulates the vacuum level, and a pneumatic baffle valve enables rapid pressure compensation. The feed pump uses a combination of a screw pump and a frequency converter (accuracy ±0.05mL / min), and the scraper motor is equipped with a servo driver (speed fluctuation ≤0.5%). The ethylene glycol refrigeration unit, in conjunction with a plate heat exchanger, uses a proportional valve to regulate the coolant flow rate, achieving precise control of the condensation temperature.

[0067] Production conditions included: processing scale of 1 kg / batch, target component content of raw material of 2.0%; the final parameter combination predicted by the model was: distillation temperature 110℃, vacuum degree 1.5 Pa, feed rate 2.0 mL / min, scraper rotation speed 350 rpm, and condensation temperature 0℃. Online monitoring was conducted by collecting 100 sets of real-time data every 30 seconds per batch, and HPLC / GC-MS was performed on each batch with 3 automatic injections.

[0068] Comparative Case 1 to Example 1: The standard production process for purifying rosemary extract involves operating at room temperature to protect heat-sensitive components and combining membrane separation and column chromatography for efficient purification. Specific steps include: pre-treating dried rosemary leaves by pulverizing them to 60-80 mesh and removing impurities to ensure the impurity content is below 1%. Extraction is then performed with 75% ethanol at 40-50°C for 90 minutes each time, repeated twice. The extract is filtered through a plate and frame filter press, then ultrafiltration is used to remove large molecular impurities such as proteins and polysaccharides. The solution is then concentrated 3-5 times using an organic solvent-resistant nanofiltration membrane, recovering ethanol while improving purity. Adsorption is then performed using a polar macroporous resin such as AB-8, followed by elution with 50-70% ethanol. The eluent containing the target component is collected. For higher purity, further separation can be achieved using silica gel column chromatography. The refined eluent is concentrated under reduced pressure at below 50°C to a paste-like state, diluted with water, and then spray-dried while controlling the inlet air temperature at 130-150°C to obtain a pale yellow powder with a water content of ≤5%, thus achieving the purification effect.

[0069] Comparative Case 2 to Example 1: The molecular distillation process for purifying rosemary extract utilizes a low-temperature, high-vacuum environment to protect heat-sensitive components and achieve efficient separation. The specific process includes: pulverizing dried rosemary leaves to 40-60 mesh to remove metallic impurities and debris, ensuring a raw material cleanliness of over 98%. Then, using 70% ethanol as a solvent, extraction is performed at 50-60℃ with stirring for 2-3 hours, repeated twice. The extract is then filtered while hot, centrifuged to remove fine particles, and concentrated to a paste state using a rotary evaporator under vacuum ≤100Pa and temperature ≤50℃, recovering the solvent while retaining the target components. In the core refining process, the concentrate is diluted with n-hexane and injected into a molecular distillation apparatus. Under parameters of distillation temperature 95℃, vacuum 0.8Pa, feed rate 1.8mL / min, scraper rotation speed 320rpm, and condensation temperature 5℃, low-boiling-point volatile oil impurities and medium-to-high-boiling-point polyphenolic components are separated, and the colorless to pale yellow target extract is collected in real time. The refined liquid is further concentrated in a vacuum concentrator to a density of 1.3-1.4 g / mL, and then made into powder by spray drying (inlet air temperature 120-140℃) or vacuum freeze drying, with the moisture content controlled at ≤5% and the solvent residue at ≤0.1%.

[0070] The comparison results between the implementation case and the control case are shown in Table 3.

[0071] Table 3 Comparison Results of Implementation Cases and Comparative Cases

[0072] The method for determining molecular distillation process parameters provided in this application significantly improves the fitting accuracy and generalization ability of the prediction model by automatically searching and determining the optimal hyperparameters of the model using a Bayesian optimization algorithm.

[0073] In some embodiments, the prediction model is used to determine the final combination of parameter values ​​based on each initial combination of parameter values, including: The optimization function value for each combination of initial parameter values ​​is determined based on a preset optimization function, and each optimization function value is mapped to multiple excellence levels. Based on the trust levels of at least two sub-models at each goodness level, determine the overall trust level for each goodness level; The final parameter value combination is obtained by weighted summing of each initial parameter value combination corresponding to the highest overall trust level.

[0074] Specifically, the optimization function is a function used to quantify the performance of a combination of given parameter values, such as a weighted function that comprehensively considers the retention rate of the target component and the odor removal rate. Confidence level is a numerical value that quantifies the model's degree of certainty about its prediction results.

[0075] For example, a multi-objective optimization function is used, where the target component retention rate × 0.6 + odor removal rate × 0.4. The target components are limonene and linalool, and the odor components are terpinene-4-ol, etc. Performance thresholds are preset, with limonene retention rate ≥ 95% and odor removal rate ≥ 85%. Based on 100 sets of historical spectral data, the continuous optimization function values ​​of SVM and the out-of-bag (OOB) sample prediction distribution of RF are uniformly divided into 3 excellence levels, with the identification framework = {A: Excellent, optimization function value ≥ 90; B: Good, 80 ≤ optimization function value < 90; C: Average, optimization function value < 80}, forming the initial prediction level set of the model.

[0076] The Basic Probability Assignment (BPA) is calculated by combining model weights and prediction confidence scores. BPA is a confidence metric with a value between 0 and 1. The SVM confidence score is derived from the decision function value. , The threshold value corresponds to the lower limit of the grade, such as grade A = 90. x is a combination of key process parameter values ​​input, f(x) is the output value of the model, and RF confidence score. Calculated using OOB error The OOB error is 5%-8%; for example, the initial parameter value combination predicted by SVM is of level A and =0.925, then ), For trust level, unassigned trust level is categorized as global uncertainty. =0.3525, RF is calculated similarly with a 30% weight to obtain the confidence level of each sub-model for each goodness level.

[0077] The conflict coefficient k is calculated to quantify the consistency of predictions between models. , i and j The indexes representing different levels of excellence are modeled based on key PCA process parameters, such as distillation temperature, vacuum level, and feed rate, with a cumulative contribution rate of 87.5% and k < 0.3. Then, a comprehensive BPA (Balanced Aptitude Test) is synthesized through orthogonal and rule-based methods, representing the overall confidence level. m(X) Prioritize selecting the A-level parameter value combination with the highest overall confidence level. If both sub-models predict A-level, the overall confidence level after synthesis is m(A)≈0.7469.

[0078] Extract the optimal parameters of SVM and RF at level A, and calculate the final parameter value combination by weighting. For example, in this embodiment, the final parameter value combination = initial parameter value combination output by SVM × 0.7 + initial parameter value combination output by RF × 0.3, ensuring compliance with equipment control accuracy (distillation temperature ±0.5℃, vacuum degree ±0.1Pa); for example, if SVM predicts a distillation temperature of 75℃ and RF predicts 72℃, then the final distillation temperature = 75 × 0.7 + 72 × 0.3 = 74.1℃ (rounded to 74℃); transmit the final parameters to the actuators, such as the heating module, vacuum unit, and feed pump, through a Siemens S7-1500 PLC, and adapt to a 15-minute Bayesian optimization cycle; simultaneously, calibrate the excellence level threshold based on measured data every 10 batches, and automatically switch to single-model BPA correction mode when a sub-model fails, such as when SVM fails. To ensure the continuity of integration.

[0079] Example 2 uses the deodorization and purification of Sichuan pepper essential oil as an example to determine the final parameter combination. The production materials and intelligent equipment system include: raw material is crude Sichuan pepper essential oil (from Hanyuan, Sichuan, with limonene content of 35%, linalool content of 12%, and main off-flavor components: terpinene-4-ol (8.5%), α-pinene (6.2%), and myrcene (5.8%)). Reagents include anhydrous sodium sulfate (analytical grade, used for dehydration) and n-hexane (chromatographic grade, used for GC-MS analysis). A short-path molecular distillation apparatus (model: KDL5, UIC GmbH, Germany, with a custom-designed condenser module resistant to low temperatures down to -50°C) is used; a distillation temperature sensor (Pt100 RTD, accuracy ±0.3°C); a vacuum transmitter (capacitive, accuracy ±0.05%FS); a feed rate flow meter (gear type, accuracy ±0.3%); a scraper speed encoder (incremental type, resolution ±0.5 rpm); and a condensation temperature thermocouple (accuracy ±1°C, equipped with a liquid nitrogen circulation system). A GC-MS system (Agilent 7890B-5977E, equipped with an HP-5MS capillary column, for real-time qualitative and quantitative analysis of odor components) was used; an electronic nose (PEN3 type, 10-sensor array, for real-time monitoring of odor intensity values ​​(0-100)) was employed. The intelligent control platform included a data preprocessing unit (UnscramblerX10.4, containing an algorithm for extracting characteristic peaks of odor components); a model computation engine (MATLAB R2023a, integrating PCA dimensionality reduction and SVM regression models); and a PLC control terminal (Siemens S7-1500, linking servo motors and vacuum units).

[0080] The automated raw material pretreatment process includes: crude Sichuan pepper oil is mixed with anhydrous sodium sulfate (5% w / w) via a pipeline mixer, and solid impurities are separated by a horizontal centrifuge (3000 rpm, 5 min). The filtrate enters the raw material storage tank. The raw material is preheated to 40°C by a plate heat exchanger and then subjected to a high-pressure homogenizer (10 MPa) to crush any potentially slightly soluble impurities, ensuring uniform feeding.

[0081] Table 4 summarizes the molecular distillation process parameter system of the embodiments of this application, as shown in Table 4: Table 4 Molecular distillation process parameter system

[0082] PCA analysis results show that distillation temperature (contribution rate 45%), vacuum degree (32%), and feed rate (10.5%) are the key process parameters, with a cumulative contribution rate of 87.5%. Figure 4 This is the second schematic diagram illustrating the importance of various process parameters provided in the embodiments of this application, such as... Figure 4 As shown, the importance of each process parameter is determined using the PCA algorithm.

[0083] The online control core architecture collects GC-MS peak area data (odor components: terpinene-4-ol, α-pinene, myrcene) and electronic nose odor intensity values ​​in real time through an intelligent analysis layer. After Z-score standardization, the data is input into the PCA module. The SVM model uses "target component retention rate × 0.6 + odor removal rate × 0.4" as a multi-objective optimization function, with a Bayesian optimization cycle of 15 minutes, outputting the optimal parameter combination. The feeding system uses a screw pump + servo drive (response time ≤ 10s), supporting switching between pulse feeding and continuous feeding modes. The vacuum system uses a Roots pump + rotary vane pump combination (ultimate vacuum 0.01Pa), coupled with a pneumatic angle valve to achieve rapid vacuum stabilization (adjustment time ≤ 15s). The distillation kettle is electrically heated (PID algorithm, overshoot ≤ 1℃), and an automatic liquid nitrogen flow rate regulating valve at the condenser end (accuracy ±5%) is used.

[0084] Automated production implementation. After system startup, three batches of nitrogen purging were run to detect vacuum system leakage rate (≤0.05Pa / min). Historical production data (same batch raw material model) was loaded, and HMI preset targets were set: limonene retention rate ≥95%, linalool retention rate ≥90%, and odor removal rate ≥85%. With a fixed scraping speed of 300rpm and condensation temperature of -30℃, gradient tests were performed on the following parameters one by one: distillation temperature (50 / 60 / 70 / 80 / 90℃), three batches were run at each level, and GC-MS was used to quantitatively analyze the content of target components and odor components; vacuum degree (0.5 / 1.0 / 1.2 / 1.5 / 2.0Pa), and the electronic nose recorded the odor intensity change curve in real time, automatically generating a sensory evaluation radar chart.

[0085] Real-time closed-loop control mode. Data preprocessing is performed, using a sliding window filter (window length 15s) to remove GC-MS baseline drift noise. Dynamic normalization formula: ( This is the real-time average over the previous 60 minutes. (The standard deviation is the real-time value for the first 60 minutes).

[0086] When the odor removal rate deviates from the model prediction by more than 10%, the raw material batch identification algorithm is automatically triggered (calling nearly 100 sets of spectral data); a fusion architecture of SVM (70% weight) + Random Forest (30% weight) is adopted, and the optimal combination of process parameter values ​​adapted to the purification requirements of Sichuan pepper essential oil is output through DS evidence theory, which effectively reduces the risk of overfitting of a single model.

[0087] Comparative Case 1 of Example 2: Steam distillation is used. The principle is to heat to the boiling point of water under normal pressure, and the steam carries the essential oil components, which are then condensed and separated. A secondary solvent extraction is required to remove water. Disadvantages: High temperature (100℃) causes the degradation of heat-sensitive components (such as linalool), the removal rate of off-odor components (terpinen-4-ol) is only 60-65%, the yield is 70-75%, and the solvent residue (ethanol) is ≥5%.

[0088] Comparative Case 2 of Example 2: Solvent extraction is used, which involves selectively dissolving essential oils with solvents such as hexane, followed by separation and concentration to remove impurities. Disadvantages include high solvent consumption (solvent:raw material = 10:1), difficulty in completely removing residual solvent (detection limit ≥20ppm), odor component removal rate of 70-75%, yield of 80-85%, and a production cycle of up to 24 hours.

[0089] Table 5 Comparison of Implementation Case 2 and Comparative Cases

[0090] As can be seen from Table 5, molecular distillation using multi-model fusion prediction based on this application performs better than other purification methods.

[0091] Once the predictive model is built, triple validation can be performed using leave-one-out cross-validation and pilot-scale experiments (≥1kg / batch) to achieve goodness-of-fit R-value. 2 (≥0.95, prediction error ≤5%, industrialization deviation ≤8%) to verify the effectiveness of the prediction model.

[0092] The coefficient of determination R can be used. 2Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) are used as metrics to evaluate model fit. The model's generalization ability can be assessed using Leave One Out Cross Validation (LOOCV) or K-fold cross-validation (K≥5). The model's prediction accuracy is validated using test set data; the relative error between predicted and measured values ​​should not exceed ±5%.

[0093] The molecular distillation process parameter determination method provided in this application combines the advantages of different algorithms through a multi-sub-model fusion strategy, such as DS evidence theory, reducing the risk of overfitting or prediction bias that may exist in a single sub-model, thereby making the final output process parameter value combination more reliable and robust.

[0094] In some embodiments, after step 130, the method further includes: Integrate the predictive model into the control system of the molecular distillation equipment; The control system operates the molecular distillation equipment based on the final parameter value combination output by the prediction model, and sends the distillation results corresponding to the final parameter value combination to the prediction model. The prediction model is updated if the difference between the distillation result corresponding to the final parameter value combination and the expected distillation result is greater than a preset deviation.

[0095] Specifically, once the prediction model can accurately predict the final parameter value combination, the prediction model can be integrated into the control system of the molecular distillation equipment.

[0096] The control system includes a parameter sensor module for real-time acquisition of process parameters such as distillation temperature and vacuum; a data processing module for data analysis based on data preprocessing and model algorithms; and an execution control module for automatically adjusting equipment parameters based on model prediction results.

[0097] Both during the training phase and after deployment of the predictive model, molecular distillation experiments can be conducted using the final parameter values ​​predicted by the model to verify its accuracy in industrial settings. Real-time monitoring can be used to calculate the relative deviation between the predicted and measured values, requiring this deviation to be less than or equal to a preset industrial deviation value (e.g., ≤8%). The difference between the distillation result corresponding to the final parameter value combination and the expected distillation result can be compared with a preset deviation to confirm whether this difference is less than or equal to the preset deviation value. After successful verification, the control system integrating the predictive model can automatically adjust process parameters such as distillation temperature and feed rate based on real-time feedback of the target component content, achieving closed-loop process control. Finally, the verified predictive model is applied to the industrial production of spices, achieving significant results compared to traditional processes, including a reduction of raw material waste of ≥30% and a production cycle shortening of ≥25%.

[0098] The molecular distillation process parameter determination method provided in this application optimizes process parameters by constructing a predictive model. By collecting multidimensional process parameters, performing principal component analysis for dimensionality reduction and Bayesian optimization, and integrating neural networks, random forests, and support vector machines to construct a predictive model, and verifying it in pilot-scale trials (processing volume ≥ 1 kg / batch, deviation ≤ 8%), it can automatically generate the final parameter value combination, reducing raw material waste by 30% and shortening the production cycle by 25%. It can protect heat-sensitive components with low-temperature parameters, and the synergistic effect of multiple models improves predictive performance, making it suitable for the refining of artemisinin, ginkgo flavonoids, etc., and promoting the intelligent upgrading of traditional Chinese medicine extraction.

[0099] The molecular distillation process parameter determination apparatus provided in the embodiments of this application is described below. The molecular distillation process parameter determination apparatus described below and the molecular distillation process parameter determination method described above can be referred to in correspondence.

[0100] Figure 5 A schematic diagram of the molecular distillation process parameter determination device provided in the embodiments of this application is shown below. Figure 5 As shown, the device includes an acquisition module 510, a filtering module 520, and a construction module 530.

[0101] The acquisition module is used to acquire multiple process parameters during the molecular distillation process, as well as the distillation results under preset parameter value combinations. The screening module is used to screen out at least two key process parameters from multiple process parameters based on the distillation results. A building module is used to construct a prediction model, including at least two sub-models among neural networks, random forests and support vector machines, based on the distillation results of at least two key process parameters under preset parameter value combinations. Each sub-model is used to predict the initial parameter value combination of key process parameters based on the expected distillation results, and the prediction model is used to determine the final parameter value combination based on each initial parameter value combination.

[0102] Specifically, according to embodiments of this application, any number of modules among the acquisition module, filtering module, and construction module can be merged into one module, or any one of these modules can be split into multiple modules.

[0103] Alternatively, at least some of the functionality of one or more of these modules can be combined with at least some of the functionality of other modules and implemented in a single module.

[0104] According to embodiments of this application, at least one of the acquisition module, screening module, and construction module can be at least partially implemented as hardware circuitry, such as a Field Programmable Gate Array (FPGA), Programmable Logic Array (PLA), System-on-a-Chip, System-on-a-Substrate, System-on-Package, Application Specific Integrated Circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in hardware or firmware, or in any one of the three implementation methods of software, hardware, and firmware, or in a suitable combination of any of them.

[0105] Alternatively, at least one of the acquisition module, filtering module, and construction module can be implemented at least partially as a computer program module that can perform corresponding functions when the computer program module is run.

[0106] In some embodiments, the filtering module is specifically used for: Construct the covariance matrix of multiple process parameters under preset parameter value combinations; The eigenvalues ​​and eigenvectors of the covariance matrix are obtained based on the coupling relationship between multiple process parameters; Process parameters whose cumulative variance contribution rate is greater than or equal to a preset contribution rate threshold are selected as candidate key process parameters. The key process parameters among the candidate key process parameters are determined based on eigenvalues ​​and eigenvectors.

[0107] In some embodiments, the construction module is specifically used for: Based on the distillation results of at least two key process parameters under preset parameter value combinations, determine at least two sub-model hyperparameter value combinations within a preset hyperparameter search space; construct a prediction model based on the sub-model hyperparameter value combinations. In the case where the prediction model includes a neural network, the combination of hyperparameter values ​​for the sub-model includes the number of hidden layer neurons and the learning rate. When the prediction model includes random forest, the combination of sub-model hyperparameter values ​​includes the number of decision trees and the maximum depth; When the prediction model includes a support vector machine, the combination of sub-model hyperparameter values ​​includes penalty coefficients and kernel parameters.

[0108] In some embodiments, the apparatus further includes a determining module, which is specifically used for: The optimization function value for each combination of initial parameter values ​​is determined based on a preset optimization function, and each optimization function value is mapped to multiple excellence levels. Based on the trust levels of at least two sub-models at each goodness level, determine the overall trust level for each goodness level; The final parameter value combination is obtained by weighted summing of each initial parameter value combination corresponding to the highest overall trust level.

[0109] In some embodiments, the device further includes an integration module, which is specifically used for: Integrate the predictive model into the control system of the molecular distillation equipment; The control system controls the operation of the molecular distillation equipment based on the final parameter value combination output by the prediction model, and sends the distillation results corresponding to the final parameter value combination to the prediction model.

[0110] In some embodiments, the device further includes an update module, which is specifically used for: If the difference between the distillation result corresponding to the final parameter value combination and the expected distillation result is greater than the preset deviation, the prediction model is updated. Also, if the latest cumulative variance contribution rate of the key process parameter is less than or equal to the preset contribution rate threshold, the key process parameter is reselected from multiple process parameters.

[0111] It should be noted that the molecular distillation process parameter determination device provided in this application embodiment can realize all the method steps implemented in the above-mentioned molecular distillation process parameter determination method embodiment, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0112] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 6As shown, the electronic device may include a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the computer program in the memory 630 to execute the above-described method.

[0113] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional modules and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0114] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the methods provided in the above embodiments.

[0115] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing the processor to execute the methods provided in the above embodiments.

[0116] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0117] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for determining molecular distillation process parameters, characterized in that, include: Obtain multiple process parameters during molecular distillation, as well as the distillation results of multiple process parameters under preset parameter value combinations; Based on the distillation results, at least two key process parameters are selected from a plurality of process parameters; Based on the distillation results of at least two of the key process parameters under preset parameter value combinations, a prediction model is constructed that includes at least two sub-models among neural networks, random forests, and support vector machines. Each of the sub-models is used to predict the initial parameter value combination of the key process parameters based on the expected distillation results, and the prediction model is used to determine the final parameter value combination based on each of the initial parameter value combinations.

2. The method for determining molecular distillation process parameters according to claim 1, characterized in that, The process of selecting at least two key process parameters from a plurality of process parameters based on the distillation results includes: Construct a covariance matrix for multiple process parameters under preset parameter value combinations; The eigenvalues ​​and eigenvectors of the covariance matrix are obtained based on the coupling relationship between multiple process parameters. Process parameters whose cumulative variance contribution rate is greater than or equal to a preset contribution rate threshold are selected as candidate key process parameters. The key process parameters among the candidate key process parameters are determined based on the eigenvalues ​​and eigenvectors.

3. The method for determining molecular distillation process parameters according to claim 1, characterized in that, The prediction model, constructed based on the distillation results of at least two of the key process parameters under preset parameter value combinations, includes at least two sub-models among neural networks, random forests, and support vector machines, comprising: Based on the distillation results of at least two key process parameters under preset parameter value combinations, the hyperparameter value combinations of the at least two sub-models are determined within a preset hyperparameter search space; the prediction model is constructed based on the sub-model hyperparameter value combinations. Wherein, when the prediction model includes the neural network, the combination of hyperparameter values ​​of the sub-model includes the number of hidden layer neurons and the learning rate; When the prediction model includes the random forest, the combination of hyperparameter values ​​for the sub-model includes the number of decision trees and the maximum depth; When the prediction model includes the support vector machine, the combination of hyperparameter values ​​for the sub-model includes a penalty coefficient and a kernel parameter.

4. The method for determining molecular distillation process parameters according to claim 1, characterized in that, The prediction model is used to determine the final parameter value combination based on each of the initial parameter value combinations, including: The optimization function value for each combination of initial parameter values ​​is determined based on a preset optimization function, and each optimization function value is mapped to multiple excellence levels; Based on the trust level of the at least two sub-models at each of the excellence levels, determine the overall trust level for each of the excellence levels; The final parameter value combination is obtained by weighted summing of each initial parameter value combination corresponding to the highest overall trust level.

5. The method for determining molecular distillation process parameters according to claim 1, characterized in that, After constructing the prediction model, which includes at least two sub-models from neural networks, random forests, and support vector machines, the method further includes: The prediction model is integrated into the control system of the molecular distillation equipment; The control system controls the operation of the molecular distillation equipment based on the final parameter value combination output by the prediction model, and sends the distillation result corresponding to the final parameter value combination to the prediction model.

6. The method for determining molecular distillation process parameters according to claim 5, characterized in that, Also includes: If the difference between the distillation result corresponding to the final parameter value combination and the expected distillation result is greater than a preset deviation, the prediction model is updated; and if the latest cumulative variance contribution rate of the key process parameter is less than or equal to a preset contribution rate threshold, the key process parameter is reselected from among the multiple process parameters.

7. A device for determining molecular distillation process parameters, characterized in that, include: The acquisition module is used to acquire multiple process parameters in the molecular distillation process, as well as the distillation results of the multiple process parameters under a preset parameter value combination; A screening module is used to screen at least two key process parameters from a plurality of process parameters based on the distillation results; A construction module is used to construct a prediction model, including at least two sub-models among neural networks, random forests and support vector machines, based on the distillation results of at least two of the key process parameters under preset parameter value combinations. Each of the sub-models is used to predict the initial parameter value combination of the key process parameters based on the expected distillation results, and the prediction model is used to determine the final parameter value combination based on each of the initial parameter value combinations.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the molecular distillation process parameter determination method according to any one of claims 1 to 6 through the computer program.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for determining molecular distillation process parameters as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for determining molecular distillation process parameters as described in any one of claims 1 to 6.

Images