A high-efficiency and energy-saving distillation and purification system and method for raspberry liquor

Abstract

This invention provides a highly efficient and energy-saving distillation and purification system and method for raspberry baijiu, belonging to the field of baijiu distillation and purification technology. It includes: a data acquisition module that collects multi-dimensional operating condition data throughout the distillation and purification process; a feature extraction module that extracts distillation heat and mass transfer, liquor quality, mash fermentation, and energy consumption characteristics; a parameter pre-tuning module that constructs a hybrid kernel support vector machine evaluation model to identify distillation stability and other operating conditions and perform preliminary parameter pre-tuning; a correlation quantification module that constructs a full-element correlation network; a strategy solving module that establishes a multi-objective optimization model to determine the optimal purification scheme; and a control execution module that converts the scheme into instructions to achieve precise separation of the liquor and maximize the retention of fruit aroma. This invention achieves refined distillation control, improves liquor quality and stability, balances impurity removal and fruit aroma retention, reduces energy consumption, and realizes automated and intelligent operation, providing support for the large-scale, high-quality production of raspberry baijiu.

Description

Technical Field

[0001] This invention belongs to the field of liquor distillation and purification technology, specifically a high-efficiency and energy-saving distillation and purification system and method for raspberry liquor. Background Technology

[0002] Raspberry baijiu, a distinctive beverage that blends the aroma of raspberries with the mellow fragrance of baijiu, relies on its distillation and purification process to directly determine product quality and production efficiency.

[0003] Currently, traditional raspberry baijiu distillation and purification methods largely rely on conventional distillation equipment and experience-based operation, resulting in numerous technical bottlenecks. Traditional processes struggle to achieve precise control over the entire distillation process, leading to instability in distillation conditions and errors in judging the timing of fractional distillation. This results in inaccurate separation of the heads, hearts, and tails, affecting not only the quality of the final product but also wasting valuable components. Furthermore, traditional processes are less effective at preserving the characteristic aroma components of raspberries; the high temperatures during distillation easily cause the loss of fruity aroma compounds, resulting in a less pronounced raspberry flavor in the finished baijiu. In addition, traditional distillation processes lack precise energy consumption control, with significant energy losses in heating and cooling stages, failing to meet the demands of energy-efficient industrial development. Moreover, the removal of impurities is limited, making it difficult to balance impurity removal with fruit aroma preservation. This hinders the market's demand for high-quality, low-energy raspberry baijiu, thus restricting the large-scale, high-quality development of the raspberry baijiu industry. Summary of the Invention

[0004] This invention provides a highly efficient and energy-saving distillation and purification system and method for raspberry liquor, in order to overcome the deficiencies in the prior art.

[0005] On one hand, the present invention provides a high-efficiency and energy-saving distillation and purification system for raspberry liquor, comprising: The data acquisition module is used to collect multi-dimensional operating data of the entire distillation process in real time when the raspberry fermentation mash is started for distillation and purification. The feature extraction module is used to extract key features from multi-dimensional operating data during the distillation heating process by employing improved translation-invariant wavelet threshold denoising and combining it with kernel principal component analysis. The parameter pre-tuning module is used to construct a mixed kernel support vector machine condition evaluation model during the distillation steady-state control process, identify distillation stability, accuracy of the distillation node, risk of impurity exceeding the standard and risk of fruit aroma loss, and perform preliminary parameter pre-tuning according to preset threshold rules; The correlation quantification module is used to perform spatiotemporal fusion of multi-dimensional operating condition data using extended Kalman filtering during coupled analysis of the distillation process. It constructs a full-element correlation network of the coupling relationship between quantification parameters, quality, mash, and energy efficiency through Bi-GRU. The strategy solving module is used to establish a multi-objective optimization model based on the full-factor correlation network during the distillation strategy optimization stage. It is then used to obtain the best purification control scheme under the current working conditions through NSGA-III solving and entropy weight-fuzzy TOPSIS screening. The control execution module is used to translate the optimal purification control scheme into equipment instructions during the segmented wine extraction process, so as to achieve segmented separation of wine and maximize the preservation of raspberry fruit aroma.

[0006] This invention provides a high-efficiency, energy-saving distillation and purification system for raspberry liquor, comprising multi-dimensional operating data including distillation kettle operation data, fractional liquor output data, aroma component data, fermentation mash property data, and energy efficiency loss data. Distillation kettle operation data includes distillation temperature, distillation pressure, heating power, liquid level inside the kettle, and distillation time. Fractional liquor output data includes liquor flow rate, alcohol content, distillation time, and the amounts of heads, hearts, and tails. Aroma component data includes raspberry ketone content, ester content, alcohol content, and aldehyde content. Fermentation mash property data includes mash sugar content, mash acidity, mash density, and mash temperature. Energy efficiency loss data includes heating energy consumption, cooling water consumption, heat loss rate, and energy consumption per unit output.

[0007] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor is provided, wherein the process of extracting key features by the feature extraction module includes: Multi-dimensional operating condition data are constructed into an n×m dimensional data matrix, where n is the number of sampling times and m is the data dimension.

[0008] By selecting a preset wavelet basis function, the data matrix is ​​decomposed into translation-invariant wavelet decompositions at a preset number of levels to obtain approximation coefficients and detail coefficients at each scale.

[0009] An adaptive soft thresholding function is constructed, which dynamically adjusts the threshold based on the noise variance of detail coefficients at each scale, and performs thresholding on the detail coefficients to suppress Gaussian noise and impulse noise.

[0010] The processed detail coefficients and the original approximation coefficients are reconstructed by translation-invariant wavelet to obtain the denoised multi-dimensional working condition data matrix.

[0011] Selecting a preset kernel function maps the denoised multi-dimensional working condition data matrix to a high-dimensional feature space, thus constructing a kernel covariance matrix.

[0012] Eigenvalue decomposition is performed on the kernel covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors sorted by size.

[0013] The eigenvectors corresponding to the top k eigenvalues ​​whose cumulative contribution rate reaches a preset threshold are selected as principal components. The principal component scores are calculated to obtain key features, which include distillation heat and mass transfer features, liquor quality features, mash fermentation features, and energy consumption features.

[0014] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor includes a parameter pre-tuning module that constructs a distillation condition evaluation model based on a hybrid kernel support vector machine, comprising the following steps: Collect historical operating condition data of raspberry baijiu distillation over a preset time period, extract key historical features, and label the corresponding distillation stability label, segmented distillation node label, impurity exceedance risk label, and fruit aroma loss risk label. Construct a training set containing s samples, with each sample containing k key features.

[0015] A hybrid kernel function is constructed by merging the first kernel function and the second kernel function. The hybrid kernel function adjusts the contribution ratio of the two types of kernel functions by preset kernel weight coefficients.

[0016] By introducing slack variables and penalty parameters, a multi-class support vector machine optimization problem is constructed.

[0017] The multi-class support vector machine optimization problem is transformed into a dual problem, and the dual problem is solved by the sequence minimum optimization algorithm to obtain the optimal Lagrange multiplier vector.

[0018] Calculate the weight vector, select the support vector that meets the preset conditions to calculate the bias term, and obtain the distillation condition evaluation model.

[0019] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor includes a parameter pre-adjustment module that performs preliminary parameter pre-adjustment according to a preset threshold rule, comprising the following steps: Pre-set the normal threshold, warning threshold, and over-limit threshold for distillation temperature, alcohol content, raspberry ketone content, methanol content, and fusel oil content, and clarify the operating condition judgment criteria.

[0020] The key features are input into the distillation condition evaluation model, which outputs the corresponding distillation stability status, segmented distillation node status, impurity exceedance risk level, and fruit aroma loss risk level.

[0021] Based on the output of the distillation condition evaluation model, adjust the corresponding parameters according to the preset rules.

[0022] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor is provided. The correlation quantization module uses an extended Kalman filter algorithm to perform spatiotemporal fusion of multi-dimensional operating condition data, including the following process: Core operating parameters are extracted from multi-dimensional operating data and used as state variables to construct a state vector with preset dimensions. The core operating parameters include distillation temperature, distillation pressure, heating power, alcohol output flow rate, mash sugar content, and mash temperature.

[0023] A nonlinear state transition equation for the distillation process is established to describe the relationship between the state vector and time.

[0024] Observation vectors are constructed based on multi-dimensional operating condition data, and observation equations are established to describe the mapping relationship between observation vectors and state vectors.

[0025] Initialize the state estimate and error covariance matrix, and calculate the current state prediction and prediction error covariance matrix based on the state transition equation.

[0026] Calculate the Kalman gain, update the state estimate by combining it with the observation vector at the current time, update the error covariance matrix, and output the unified state estimate after spatiotemporal fusion.

[0027] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor is provided. The process by which the correlation quantification module constructs a correlation network of all elements of raspberry distillation through a bidirectional gated circulation unit Bi-GRU includes: The unified state estimate is constructed as a time-series input sequence and fed into the Bi-GRU network.

[0028] The Bi-GRU network is configured to contain a preset number of hidden layers, each containing a preset number of GRU units. The forward GRU layer captures the forward dependencies of the time-series input sequence, and the backward GRU layer captures the backward dependencies of the time-series input sequence.

[0029] The outputs of the forward and reverse GRU layers are concatenated to obtain the fused hidden states at each time step.

[0030] The Bi-GRU network is trained using a preset loss function and preset training parameters until the loss function converges.

[0031] Using the fused hidden states corresponding to the state variables of each dimension as network nodes and the connection weights between nodes as the association strength, a network of association relationships of all elements is obtained.

[0032] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor is provided. The process of establishing a multi-objective optimization model by the strategy solving module includes: Distillation temperature, distillation pressure, reflux ratio, start time of core distillation, end time of core distillation, and heating power are selected as distillation control parameters to construct a decision vector.

[0033] Sub-objective functions were constructed, including maximizing the wine yield, maximizing the raspberry ketone retention rate, maximizing the methanol removal rate, maximizing the fusel oil removal rate, and minimizing the energy consumption per unit of wine produced.

[0034] Maximizing the yield of the core liquor is indicated by the percentage of the core liquor produced per unit mass of fermented mash.

[0035] The retention rate of raspberry ketone was maximized by using the ratio of the concentration of raspberry ketone in the wine center to the initial concentration of raspberry ketone in the mash.

[0036] The methanol removal rate is maximized by using the ratio of the difference between the initial methanol concentration in the mash and the methanol concentration in the center of the liquor to the initial methanol concentration in the mash.

[0037] The maximum removal rate of fusel oil is determined by the ratio of the initial concentration of fusel oil in the mash to the concentration of fusel oil in the center of the liquor, and the ratio of the initial concentration of fusel oil in the mash.

[0038] Minimizing energy consumption per unit of distillation output is indicated by the ratio of total energy consumption in the distillation process to the quality of the resulting spirit.

[0039] By combining the coupling constraints of the full-element correlation network, the equipment operation safety constraints, and the liquor quality standard constraints, the constraints of the multi-objective optimization model are constructed.

[0040] According to the present invention, a high-efficiency and energy-saving distillation and purification system for raspberry liquor is provided. The process by which the strategy solving module determines the optimal purification control scheme under the current operating conditions includes: Individuals of a predetermined population size are randomly generated as the initial population for the NSGA-III algorithm, with each individual representing a set of distillation control parameters.

[0041] Calculate the sub-objective function value for each individual, perform non-dominated sorting based on the reference point method, and calculate the crowding distance of each individual.

[0042] Excellent individuals are selected using a tournament selection method, and simulated binary crossover and polynomial mutation operations are performed to generate offspring populations.

[0043] The parent and offspring populations are merged, and non-dominated sorting and reference point association operations are performed again. Individuals of the preset population size are selected to form the next generation population.

[0044] The process iterates and evolves a preset number of times. Once the termination condition is met, the Pareto optimal solution set is output, which is the optimal solution set for the distillation control strategy.

[0045] The objective weights of the sub-objective function are calculated using the entropy weight method, resulting in a weight vector.

[0046] Construct a fuzzy decision matrix to determine the positive and negative ideal solutions. Calculate the distance from each solution in the optimal distillation control strategy solution set to the positive and negative ideal solutions to obtain the relative proximity.

[0047] The solution with the highest relative approximation is selected as the optimal purification control scheme under the current operating conditions.

[0048] On the other hand, the present invention also provides a highly efficient and energy-saving distillation and purification method for raspberry liquor, comprising: Real-time acquisition of multi-dimensional operating data throughout the entire process of raspberry liquor distillation and purification.

[0049] An improved translation-invariant wavelet thresholding algorithm was used to denoise the multi-dimensional operating condition data. Key features were extracted by combining kernel principal component analysis. These key features included distillation heat and mass transfer characteristics, wine quality characteristics, mash fermentation characteristics, and energy consumption characteristics.

[0050] A distillation condition evaluation model based on a hybrid kernel support vector machine is constructed. Based on key features, the model can determine the distillation stability, the accuracy of the segmented distillation node, the risk of excessive impurities, and the risk of loss of fruit aroma in real time, and perform preliminary parameter pre-adjustment according to preset threshold rules.

[0051] Extended Kalman filter algorithm is used to perform spatiotemporal fusion of multi-dimensional working condition data. A bidirectional gated cyclic unit (Bi-GRU) is used to construct a full-element correlation network for raspberry distillation. The full-element correlation network quantitatively represents the coupling strength and causal relationship between distillation control parameters, liquor quality, mash characteristics and energy efficiency.

[0052] A multi-objective optimization model was established based on the comprehensive element correlation network to calculate the overall wine yield, raspberry ketone retention rate, methanol removal rate, fusel oil removal rate, and unit wine output energy consumption. The NSGA-III algorithm was used to solve for the optimal distillation control strategy solution set, and the optimal purification control scheme under the current operating conditions was determined by combining the entropy weight-fuzzy TOPSIS method.

[0053] The optimal purification control scheme is translated into execution instructions to achieve precise separation of the heads, hearts, and tails of the wine and maximize the preservation of raspberry characteristic aromas.

[0054] This invention provides a high-efficiency and energy-saving distillation and purification system and method for raspberry baijiu (Chinese white liquor). Through precise acquisition and intelligent processing of data from the entire process, it achieves refined control of the distillation process, effectively improving the stability of the distillation state, accurately determining the segmented distillation points, and achieving precise separation of the heads, hearts, and tails of the liquor, reducing waste of effective components and improving the quality of the finished product. Through optimized feature extraction and correlation quantification techniques, it can accurately capture the coupling relationship between various distillation elements, achieving a balance between impurity removal and the retention of raspberry characteristic aromas, maximizing the preservation of raspberry fruit aroma substances, and highlighting the unique flavor of raspberry baijiu. Simultaneously, through a multi-objective optimization model and intelligent solution algorithm, it achieves precise control of energy consumption, effectively reducing energy loss during the distillation process, aligning with the industrial development trend of high efficiency and energy conservation, and reducing production costs. Furthermore, the system achieves automated and intelligent operation of the distillation process, reducing manual intervention, lowering operational difficulty, and improving production efficiency, while ensuring the stability and consistency of the production process. This provides strong support for the large-scale, high-quality production of raspberry baijiu, promoting the technological upgrading and sustainable development of the raspberry baijiu industry. Attached Figure Description

[0055] The invention will now be further described with reference to the accompanying drawings.

[0056] Figure 1 This is a schematic diagram of the structure of a high-efficiency and energy-saving distillation and purification system for raspberry liquor according to the present invention; Figure 2 This is a schematic flowchart of a highly efficient and energy-saving distillation and purification method for raspberry liquor according to the present invention. Detailed Implementation

[0057] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0058] like Figures 1 to 2 As shown in the figure, the present invention provides a high-efficiency and energy-saving distillation and purification system and method for raspberry liquor. The main body of the execution can be a high-efficiency and energy-saving distillation and purification system for raspberry liquor, including a data acquisition module, a feature extraction module, a parameter pre-tuning module, a correlation quantization module, a strategy solving module and a control execution module.

[0059] The data acquisition module is used to collect multi-dimensional operating data of the entire distillation process in real time when the raspberry fermentation mash is started for distillation and purification.

[0060] The fermentation mash start-up distillation purification stage is a preparatory process before distillation. In actual production, workers first check the sealing and integrity of the distillation kettle, rectification column, and cooling pipes. After confirming that there is no leakage, the fermented raspberry mash in the fermentation tank is pumped into the distillation kettle through a screw pump. The loading amount does not exceed 2 / 3 of the distillation kettle volume. The kettle lid is closed and locked, and the cooling water circulation system is started.

[0061] The data acquisition module initiates a full-process sensor self-test, collecting multi-dimensional operating data in real time, including distillation kettle operation data, segmented alcohol output data, aroma component data, fermentation mash physical property data, and energy efficiency loss data.

[0062] Multi-dimensional operating data includes distillation kettle operation data, fractional distillation data, aroma component data, fermentation mash property data, and energy efficiency loss data. Distillation kettle operation data includes distillation temperature, distillation pressure, heating power, liquid level in the kettle, and distillation time. Fractional distillation data includes distillation flow rate, alcohol content, distillation time, and the amount of heads, heart, and tails. Aroma component data includes raspberry ketone content, ester content, alcohol content, and aldehyde content. Fermentation mash property data includes mash sugar content, mash acidity, mash density, and mash temperature. Energy efficiency loss data includes heating energy consumption, cooling water consumption, heat loss rate, and energy consumption per unit of distillate.

[0063] In this embodiment, PT100 platinum resistance temperature sensors are used and installed at three locations: the upper, middle, and lower parts of the distillation vessel, the inlet and outlet of the heating jacket, and the gas and liquid sides of each tray in the distillation column. These sensors are used to collect temperature gradient data related to distillation temperature and liquid level in the vessel.

[0064] A diffused silicon pressure transmitter is used, installed at the top of the distillation kettle and the top of the distillation column, to collect distillation pressure data.

[0065] An intelligent power transmitter is used, connected in series with the electric heating system of the distillation vessel, to collect heating power data.

[0066] The timer is integrated into the central processing unit and is used to record distillation time data. The above data together constitute the distillation kettle operation data.

[0067] An electromagnetic flow meter is used and installed on the wine outlet pipe and cooling water pipe to collect data on wine flow rate and cooling water consumption.

[0068] A vibration-type alcohol meter with a measurement range of 0-100%vol is used. It is installed at the front end of the solenoid valve in the liquor outlet pipeline to collect the alcohol content data of the liquor in real time.

[0069] The solenoid valve is an explosion-proof type with a response time of no more than 0.1 seconds. It is installed at the branch of the wine outlet pipeline to control the flow direction of the head, center, and tail of the wine. By recording the switching time of the solenoid valve and the corresponding flow data, the extraction time and segment quantity of the head, center, and tail of the wine can be obtained. The above data together constitute the segmented wine output data.

[0070] The online gas chromatograph is equipped with a flame ionization detector. A polar capillary column is used, and the column temperature is programmed. The initial temperature is 50℃ and held for 2 minutes, then increased to 200℃ at a rate of 5℃ / min and held for 10 minutes. The injection port temperature is set to 200℃, the detector temperature is set to 250℃, the carrier gas is high-purity nitrogen, the flow rate is 1mL / min, the split ratio is 10:1, and the wine sample is automatically collected every 30 seconds for analysis to obtain data on raspberry ketone content, total ester content, higher alcohol content, and aldehyde content, which constitute the aroma component data.

[0071] The saccharimeter is a refractometer with a measurement range of 0-50°Bx; the pH meter is a glass electrode pH meter with a measurement range of 0-14 pH; the densitometer is a vibrating densitometer; and the temperature sensor is a PT100 platinum resistance thermometer. All instruments are installed on the pipeline at the mash inlet to collect data on mash saccharimeter, mash acidity, mash density, and mash temperature, which constitute the physical property data of the fermentation mash.

[0072] Smart energy meters are used to collect heating energy consumption data. Thermal resistance heat flow meters are used, attached to the outer surface of the distillation vessel and distillation column, to collect heat loss rate data.

[0073] Based on the collected data on heating energy consumption, cooling water consumption, and heat loss rate, combined with the quality of the produced spirit core, the energy consumption per unit output was calculated. The specific calculation process was as follows: The sources of each core data point were clearly defined; heating energy consumption was directly collected by a smart energy meter. Cooling water consumption was collected by an electromagnetic flow meter, and the conversion was performed according to the standard conversion for the baijiu distillation industry, with 1m³... 3 The energy consumption corresponding to cooling water is approximately 0.86 kWh. Heat loss energy consumption is calculated using the heat loss rate, the heat dissipation area of ​​the outer walls of the distillation vessel and rectification column (preset to a fixed value according to equipment specifications), and the total distillation time: Heat loss energy consumption = Heat loss rate × Heat dissipation area × Total distillation time. The core spirit quality is calculated by combining the output flow rate collected by the flow sensor in the output pipe, the distillation period, and the core spirit alcohol content collected by the online alcohol content analyzer. The total energy consumption of the distillation process is calculated as: Total energy consumption = Heating energy consumption + Cooling water energy consumption + Heat loss energy consumption. Finally, the energy consumption per unit output is calculated as: Unit output energy consumption = Total distillation energy consumption ÷ Core spirit quality, forming the energy efficiency loss data.

[0074] The sampling frequency of all sensors is uniformly set to 1Hz. The raw data collected is transmitted to the central processing unit via RS485 bus. The central processing unit first performs format conversion on the raw data, converting the analog signals output by different sensors into a unified digital signal format. Then, it fills in the missing values ​​in the data by using the arithmetic mean of the effective values ​​of the three adjacent sampling times of the data point. After the filling is completed, the data is stored in a local industrial-grade solid-state drive database and simultaneously synchronized to a cloud server for off-site backup via 4G / 5G network.

[0075] The feature extraction module is used to extract key features from multi-dimensional operating data during the distillation heating process by employing improved translation-invariant wavelet thresholding for noise reduction and combining it with kernel principal component analysis.

[0076] The distillation heating stage is the preheating process for distillation. In actual production, workers start the heating system and adopt a gradient heating strategy. First, the mash is heated at a low power of 30kW for 30 minutes to uniformly raise the temperature from room temperature to 70°C. This avoids rapid heating that could cause the mash to burn or premature thermal decomposition of fruit flavorings. No alcohol flows out during this stage; it only completes the preheating of the mash and the prevaporization of light components.

[0077] The feature extraction module is activated to perform noise reduction and feature extraction on the time-series data of temperature, pressure, and power collected during the heating process.

[0078] The multi-dimensional operating condition data is constructed into an n×m dimensional data matrix, where n is the number of sampling times and m is the data dimension. In this embodiment, m is 22, corresponding to 5-dimensional distillation kettle operation data, 4-dimensional segmented alcohol output data, 4-dimensional aroma component data, 4-dimensional fermentation mash physical property data, and 5-dimensional energy efficiency loss data.

[0079] The db4 wavelet basis function is selected as the preset wavelet basis function. The data matrix is ​​decomposed into five levels of translation-invariant wavelet decomposition to obtain five scales of approximation coefficients and five scales of detail coefficients. The approximation coefficients contain the low-frequency effective components of the signal, and the detail coefficients contain the high-frequency noise components of the signal.

[0080] An adaptive soft threshold function is constructed, specifically in the form that when the absolute value of the detail coefficient is greater than or equal to the dynamic threshold, the detail coefficient is subtracted from the product of the threshold and the sign function of the detail coefficient; when the absolute value of the detail coefficient is less than the dynamic threshold, it is set to 0.

[0081] The dynamic threshold is dynamically adjusted based on the noise variance of the detail coefficients at each scale. The noise variance is calculated using the median absolute deviation method, with the formula being: noise variance equals the median absolute deviation of the detail coefficients at that scale divided by 0.6745. The threshold is equal to the noise variance multiplied by the square root of the natural logarithm. ,in The length of the detail coefficient at this scale.

[0082] By using this adaptive soft thresholding function to threshold the detail coefficients at each scale, Gaussian noise and impulse noise generated during the distillation process can be effectively suppressed simultaneously.

[0083] The processed detail coefficients and the original approximation coefficients are reconstructed by translation-invariant wavelet to obtain the denoised multi-dimensional working condition data matrix.

[0084] The Gaussian radial basis function is selected as the preset kernel function, and the width parameter of the kernel function is set to 1.0. The denoised multi-dimensional working condition data matrix is ​​mapped to a high-dimensional linearly separable feature space to construct the kernel covariance matrix.

[0085] Eigenvalue decomposition is performed on the kernel covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors arranged in descending order of size.

[0086] The eigenvectors corresponding to the top k eigenvalues ​​with a cumulative contribution rate of 95% are selected as principal components. Principal component scores are calculated to obtain distillation heat and mass transfer characteristics, liquor quality characteristics, mash fermentation characteristics, and energy consumption characteristics. In this embodiment, k is set to 8. The distillation heat and mass transfer characteristics include three principal components, reflecting the temperature distribution, pressure changes, and heat transfer efficiency during the distillation process. The liquor quality characteristics include two principal components, reflecting the alcohol content, aroma component content, and impurity content. The mash fermentation characteristics include two principal components, reflecting the degree of fermentation and changes in physical properties of the mash. The energy consumption characteristics include one principal component, reflecting the energy utilization efficiency of the distillation process.

[0087] The parameter pre-tuning module is used to construct a mixed kernel support vector machine condition evaluation model during the steady-state control of distillation, identify distillation stability, accuracy of distillation node, risk of impurity exceeding the standard and risk of fruit aroma loss, and perform preliminary parameter pre-tuning according to preset threshold rules.

[0088] The steady-state control stage of distillation is the initial distillation head collection process. In actual production, when the temperature inside the distillation vessel rises to about 78°C, low-boiling-point methanol and aldehydes begin to vaporize and distill out. Workers begin to collect the heads. The amount of heads collected is usually 1%-2% of the total output. At this stage, the liquid has a high content of impurities and a low content of fruity aroma substances, so it needs to be collected separately for subsequent centralized processing.

[0089] The parameter pre-tuning module is activated to construct a distillation condition evaluation model based on a hybrid kernel support vector machine. Based on the key features extracted during the heating stage, the accuracy of the head distillation node and the risk of impurities exceeding the standard are determined in real time.

[0090] Historical operating condition data of raspberry baijiu distillation over the past 12 months were collected. Each set of data contains complete multi-dimensional operating condition data and corresponding quality test results. Key features were extracted from the historical data. Technical personnel with more than 5 years of experience in raspberry baijiu distillation labeled the corresponding distillation stability labels, segmented distillation node labels, impurity exceedance risk labels, and fruit aroma loss risk labels according to unified standards. A training set containing s samples was constructed, with each sample containing 8 key features. The remaining data was used as a test set to verify the generalization ability of the model.

[0091] A hybrid kernel function is constructed by fusing the Gaussian radial basis function (first kernel function) and the polynomial kernel function (second kernel function). The expression for the hybrid kernel function is: the hybrid kernel function equals the kernel weight coefficient multiplied by the Gaussian radial basis function plus 1 minus the kernel weight coefficient multiplied by the polynomial kernel function. The kernel weight coefficient is set to 0.7, which can balance the local fitting ability of the Gaussian radial basis function and the global generalization ability of the polynomial kernel function. The degree of the polynomial kernel function is set to 3.

[0092] By introducing slack variables and penalty parameters, a multi-class support vector machine optimization problem is constructed. The penalty parameter is set to 10.0 to control the balance between model complexity and classification error. The upper limit of the slack variable is set to 1.0 to allow a small number of samples to have classification error in order to improve the model's generalization ability.

[0093] The multi-class support vector machine optimization problem is transformed into a dual problem. The dual problem is solved using the sequential minimum optimization algorithm, which decomposes the original problem into multiple minimal quadratic programming subproblems. The optimal Lagrange multiplier vector is obtained by iteratively solving the subproblems.

[0094] Calculate the weight vector, select support vectors with Lagrange multipliers greater than 0 and less than the penalty parameter to calculate the bias term, and obtain the distillation condition assessment model. The model has a classification accuracy of no less than 98% on the test set, which can meet the accuracy requirements of real-time condition assessment.

[0095] The process of performing preliminary parameter pre-tuning according to preset threshold rules includes: The pre-set normal threshold for distillation temperature is 85℃ to 95℃, the warning threshold is 82℃ to 85℃ and 95℃ to 98℃, and the exceeding threshold is below 82℃ and above 98℃. The normal threshold for alcohol content is 60% vol to 65% vol, the warning threshold is 55% vol to 60% vol and 65% vol to 70% vol, and the exceeding threshold is below 55% vol and above 70% vol. The normal threshold for raspberry ketone content is not less than 80 mg / L, the warning threshold is 60 mg / L to 80 mg / L, and the exceeding threshold is below 60 mg / L. The normal threshold for methanol content is not more than 0.4 g / L, the warning threshold is 0.4 g / L to 0.6 g / L, and the exceeding threshold is above 0.6 g / L. The normal threshold for fusel oil content is not more than 2.0 g / L, the warning threshold is 2.0 g / L to 2.5 g / L, and the exceeding threshold is above 2.5 g / L.

[0096] The operating condition judgment criteria are clearly defined. Distillation stability is divided into two categories: stable and unstable. Unstable means that the distillation state output by the distillation condition assessment model deviates from the normal threshold range. Segmented distillation nodes are divided into three categories: accurate, early, and late. Early means that the model determines the distillation node earlier than the standard distillation node, and late means that the model determines the distillation node later than the standard distillation node. The risks of excessive impurities and loss of fruit aroma are both divided into four levels: none, low, medium, and high. The medium and high levels refer to the risk level being in the preset medium level or above.

[0097] The key features extracted in real time are input into the distillation condition evaluation model, which outputs the corresponding distillation stability status, segmented distillation node status, impurity exceedance risk level, and fruit aroma loss risk level.

[0098] When the distillation stability assessment model outputs "unstable," the heating power, distillation column vent valve opening, and mash feed pump flow rate are adjusted according to the deviation of distillation temperature, distillation pressure, and in-bottle liquid level data, using preset adjustment step sizes. The adjustment step size for heating power is 5kW, for distillation column vent valve opening is 5%, and for mash feed pump flow rate is 0.5m. 3 / h, wait 30 seconds after each adjustment, and re-evaluate after the operating conditions stabilize until the distillation stability returns to stability.

[0099] When the distillation condition evaluation model outputs an "advanced" segmented distillation node, the switching time from the heads to the center of the spirit is delayed by 30 seconds. When the output is "delayed," the switching time from the heads to the center of the spirit is advanced by 30 seconds.

[0100] When the distillation condition assessment model outputs a medium-to-high level risk of impurity exceeding the standard, the reflux ratio of the rectification section is increased by 10%, and the rectification time is extended by 5 minutes.

[0101] When the distillation condition assessment model outputs a medium-to-high risk of fruit aroma loss, the distillation temperature is reduced by an adjustment step of 1℃, the heating power is reduced by an adjustment step of 5kW, and the high-temperature distillation time is shortened by 3 minutes.

[0102] The strategy solving module is used to establish a multi-objective optimization model based on the full-factor correlation network during the distillation strategy optimization stage. The model is then solved by NSGA-III and screened by entropy weight-fuzzy TOPSIS to obtain the optimal purification control scheme under the current operating conditions.

[0103] The distillation strategy optimization stage is the core main distillation process. In actual production, after the heads are collected and the distillation temperature is stable at 85-95℃, the core spirit begins to be distilled. At this stage, the alcohol content is stable at 60-65% vol, and the content of fruit aroma substances such as raspberry ketones is the highest. This is the core product fraction of raspberry baijiu. Workers need to stabilize the heating power and control the reflux ratio to ensure stable distillation. This stage accounts for more than 70% of the total distillation time.

[0104] The process of spatiotemporal fusion of multi-dimensional working condition data using the extended Kalman filter algorithm includes: Core operating parameters are extracted from multi-dimensional operating data and used as state variables to construct a 6-dimensional state vector. The core operating parameters include distillation temperature, distillation pressure, heating power, alcohol output flow rate, mash sugar content, and mash temperature.

[0105] A nonlinear state transition equation for the distillation process is established. The state transition equation adopts the form of a first-order nonlinear difference equation to describe the relationship between the state vector and time. The nonlinear terms in the equation are obtained by fitting historical operating data using the least squares method, which can accurately reflect the dynamic relationship between various parameters during the distillation process.

[0106] A 6-dimensional observation vector is constructed based on multi-dimensional operating condition data, and an observation equation is established. The observation equation is a linear equation that describes the mapping relationship between the observation vector and the state vector. The covariance matrix of the observation noise is determined by the accuracy parameters of each sensor, and the diagonal elements are the squares of the accuracy of the corresponding sensor.

[0107] The initial state estimate is the observation value at the first sampling time after startup. The initial error covariance matrix is ​​a diagonal matrix, and the diagonal elements are the squares of the corresponding sensor accuracy.

[0108] State prediction based on nonlinear state transition equation: Based on the established nonlinear state transition equation of the distillation process, the state prediction value at the current moment is calculated. This state prediction value is the predicted value of the core operating parameters (distillation temperature, distillation pressure, heating power, alcohol output flow rate, mash sugar content, mash temperature). At the same time, the prediction error covariance matrix corresponding to this state prediction value is calculated. This matrix is ​​used to quantify the error range and uncertainty of the predicted value.

[0109] The state update is combined with the observation vector: the Kalman gain is calculated, which reflects the degree of correction of the state estimate by the observation value. Combined with the observation vector constructed at the current moment (multi-dimensional measured core parameter data of the actual working condition), the state prediction value obtained in the previous state prediction steps is corrected to obtain the updated state estimate. The error covariance matrix is ​​updated synchronously to correct the prediction error range, further reducing the deviation between the updated state estimate and the actual working condition data.

[0110] Iteratively perform prediction and update, and output the unified state estimate after spatiotemporal fusion.

[0111] The process of constructing a comprehensive correlation network for raspberry distillation using a bidirectional gated cyclic unit (Bi-GRU) includes: The unified state estimate is constructed as a time-series input sequence of length 60, which contains the state estimates of the past 60 seconds, and then input into the Bi-GRU network.

[0112] The Bi-GRU network consists of two hidden layers, each containing 64 GRU units. The activation function is the tanh function, and the output layer uses a linear activation function. The forward GRU layer captures the forward dependencies of the time-series input sequence, and the backward GRU layer captures the backward dependencies of the time-series input sequence. This allows for the simultaneous use of past and future time-series information, improving the accuracy of association mining.

[0113] The outputs of the forward and reverse GRU layers are concatenated to obtain the fused hidden states at each time step.

[0114] The mean squared error loss function is used as the preset loss function, and the Adam optimizer is used for training. The learning rate is set to 0.001, the batch size is set to 32, and the number of iterations is set to 100. When the loss function value decreases by less than 0.0001 in 10 consecutive iterations, the training is terminated early to prevent the model from overfitting.

[0115] Using the fused hidden states corresponding to the state variables of each dimension as network nodes, and the connection weights between nodes as the correlation strength, the correlation strength ranges from [-1, 1]. Positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the larger the absolute value, the stronger the correlation. This yields a full-element correlation network. The correlation strength between any two nodes in the network is obtained by calculating the Pearson correlation coefficient between the corresponding hidden states, which can accurately quantify the coupling strength and causal relationship between each element.

[0116] The strategy solving module is used to establish a multi-objective optimization model based on the comprehensive wine yield, raspberry ketone retention rate, methanol removal rate, fusel oil removal rate, and unit wine output energy consumption based on the full-element correlation network. The NSGA-III algorithm is used to solve the optimal distillation control strategy solution set, and the entropy weight-fuzzy TOPSIS method is combined to determine the best purification control scheme under the current working conditions.

[0117] The process of establishing a multi-objective optimization model includes: Distillation temperature, distillation pressure, reflux ratio, start time of core distillation, end time of core distillation, and heating power were selected as distillation control parameters to construct a 6-dimensional decision vector.

[0118] Sub-objective functions are constructed as follows: The sub-objective function for maximizing the core yield is: core yield equals the core yield per unit mass of fermented mash multiplied by 100%. The sub-objective function for maximizing raspberry ketone retention is: raspberry ketone retention equals the ratio of the raspberry ketone concentration in the core to the initial raspberry ketone concentration in the mash multiplied by 100%. The sub-objective function for maximizing methanol removal is: methanol removal rate equals the ratio of the difference between the initial methanol concentration in the mash and the methanol concentration in the core to the initial methanol concentration in the mash multiplied by 100%. The sub-objective function for maximizing fusel oil removal is: fusel oil removal rate equals the ratio of the difference between the initial fusel oil concentration in the mash and the fusel oil concentration in the core to the initial fusel oil concentration in the mash multiplied by 100%. The sub-objective function for minimizing energy consumption per unit of distillation is: energy consumption per unit of distillation equals the ratio of the total energy consumption of the distillation process to the core yield mass.

[0119] Combining the coupling constraints of the full-element correlation network, equipment operation safety constraints, and liquor quality standard constraints, a multi-objective optimization model is constructed with the following constraints: distillation temperature ranges from 80℃ to 100℃, distillation pressure ranges from 0.08MPa to 0.12MPa, reflux ratio ranges from 1 to 5, the start time of core distillation ranges from 5 to 15 minutes after distillation begins, the end time of core distillation ranges from 45 to 60 minutes after distillation begins, heating power ranges from 50kW to 150kW, methanol content ranges from no more than 0.4g / L, fusel oil content ranges from no more than 2.0g / L, and raspberry ketone content ranges from no less than 80mg / L.

[0120] The process of determining the optimal purification control scheme under the current operating conditions includes: A population of 100 individuals is randomly generated as the initial population for the NSGA-III algorithm. Each individual represents a set of distillation control parameters, and the parameter values ​​are randomly generated within the corresponding constraint range.

[0121] The sub-objective function value for each individual is calculated. Non-dominated sorting is performed based on the reference point method to divide individuals in the population into different non-dominated levels. The crowding distance of individuals is calculated to measure the degree of dispersion among individuals.

[0122] The tournament selection method is used to select outstanding individuals. The tournament size is set to 3. Three individuals are randomly selected from the population, and the individual with the highest non-dominant level and the largest crowding distance is selected as the parent individual.

[0123] Perform simulated binary crossover and polynomial mutation operations, with the crossover probability set to 0.9 and the mutation probability set to 0.1, to generate the offspring population.

[0124] The parent and offspring populations are merged, and non-dominated sorting and reference point association operations are performed again. Individuals with a population size of 100 are selected to form the next generation population.

[0125] After iterating 200 times and satisfying the termination condition, the Pareto optimal solution set is output. The size of the Pareto optimal solution set is not less than 20. Each solution in the solution set is a non-dominated solution, and it is impossible to improve the performance of a certain sub-objective without reducing the performance of other sub-objectives.

[0126] The objective weights of the sub-objective functions are calculated using the entropy weight method. First, the values ​​of each sub-objective function are standardized to eliminate the influence of different dimensions. Then, the information entropy of each sub-objective is calculated. The smaller the information entropy, the greater the amount of information provided by the sub-objective, and the higher its weight. Finally, the weight vector is calculated based on the information entropy. In this embodiment, the weight of the wine yield is 0.25, the weight of the raspberry ketone retention rate is 0.3, the weight of the methanol removal rate is 0.15, the weight of the fusel oil removal rate is 0.15, and the weight of the energy consumption per unit of wine production is 0.15.

[0127] Construct a fuzzy decision matrix to determine the positive ideal solution and the negative ideal solution. The positive ideal solution is the optimal value of each sub-objective function, and the negative ideal solution is the worst value of each sub-objective function. Calculate the Euclidean distance from each solution in the optimal distillation control strategy solution set to the positive ideal solution and the negative ideal solution to obtain the relative proximity. The relative proximity is equal to the distance from the solution to the negative ideal solution divided by the sum of the distances from the solution to the positive ideal solution and the distances from the solution to the negative ideal solution.

[0128] The solution with the highest relative approximation is selected as the optimal purification control scheme under the current operating conditions. This scheme can achieve the optimal balance among multiple sub-objectives.

[0129] The control execution module is used to translate the optimal purification control scheme into equipment instructions during the segmented wine extraction process, so as to achieve segmented separation of wine and maximize the preservation of raspberry fruit aroma.

[0130] The fractional distillation stage is the tail distillation process. In actual production, when the distillation has been going on for 45-60 minutes and the alcohol content of the distillate has dropped below 55% vol, high-boiling-point fusel oils and higher alcohols begin to distill out, and the flavor of the liquor deteriorates. Workers then begin to collect the tail distillate, which can be returned to the next distillation for further purification. Heating is stopped after this stage is completed.

[0131] The parameters such as distillation temperature, distillation pressure, reflux ratio, start time of core distillation, end time of core distillation, and heating power in the optimal purification control scheme are converted into digital execution instructions and sent to the corresponding actuators via the Modbus-RTU bus.

[0132] The heating controller adjusts the output power of the electric heating tube according to the received heating power command to ensure that the temperature inside the distillation vessel remains stable at the set value.

[0133] The distillation column reflux valve is an electrically controlled regulating valve. It adjusts the valve opening according to the received reflux ratio command to control the ratio of reflux flow rate to output flow rate of the distillation column.

[0134] The three explosion-proof solenoid valves in the dispensing pipeline correspond to the head, center, and tail storage tanks, respectively. The valves are controlled to open and close according to the received dispensing time command, so as to realize the automatic diversion of the head, center, and tail of the liquor. The response time of the solenoid valves does not exceed 0.1 seconds, ensuring the accuracy of segmented dispensing.

[0135] The mash feed pump is a variable frequency speed control pump, which adjusts the feed flow rate according to the received flow command to maintain a stable liquid level in the distillation vessel.

[0136] During the distillation process, data acquisition, feature extraction, operating condition evaluation, and parameter pre-tuning are performed every 10 seconds to promptly correct operating condition deviations and ensure the stability of the distillation process. Correlation quantization and strategy solving are performed every minute to update the optimal purification control scheme and ensure the distillation process remains in an optimal state at all times.

[0137] After distillation is completed, an operation report for this distillation is generated. The report includes the entire operating condition curve of the distillation process, the quality test data of the distillate, the energy consumption statistics, the control parameter adjustment records, etc. It is stored in the local database and uploaded to the cloud server for subsequent model iteration optimization and continuous improvement of the distillation process.

[0138] In summary, this embodiment provides a high-efficiency and energy-saving distillation and purification system for raspberry baijiu. Through precise acquisition and intelligent processing of data from the entire process, it achieves refined control of the distillation process, effectively improving the stability of the distillation state, accurately determining the segmented distillation points, and achieving precise separation of the heads, hearts, and tails of the spirit, reducing waste of effective components and improving the quality of the distillate. Optimized feature extraction and correlation quantification techniques accurately capture the coupling relationships between various distillation elements, achieving a balance between impurity removal and the retention of raspberry characteristic aromas, maximizing the preservation of raspberry fruit aroma substances, and highlighting the unique flavor of raspberry baijiu. Simultaneously, through a multi-objective optimization model and intelligent solution algorithm, it achieves precise control of energy consumption, effectively reducing energy loss during the distillation process, aligning with the industrial development trend of high efficiency and energy conservation, and reducing production costs. Furthermore, the system achieves automated and intelligent operation of the distillation process, reducing manual intervention, lowering operational difficulty, improving production efficiency, and ensuring the stability and consistency of the production process. This provides strong support for the large-scale, high-quality production of raspberry baijiu, promoting the technological upgrading and sustainable development of the raspberry baijiu industry.

[0139] Based on the same general inventive concept, this invention also protects an efficient and energy-saving distillation purification method for raspberry liquor. The efficient and energy-saving distillation purification method for raspberry liquor provided by this invention is described below. The efficient and energy-saving distillation purification method for raspberry liquor described below can be referred to in correspondence with the efficient and energy-saving distillation purification system and method for raspberry liquor described above.

[0140] A highly efficient and energy-saving distillation purification method for raspberry liquor includes: Real-time acquisition of multi-dimensional operating data throughout the entire process of raspberry liquor distillation and purification.

[0141] An improved translation-invariant wavelet thresholding algorithm was used to denoise the multi-dimensional operating condition data. Key features were extracted by combining kernel principal component analysis. These key features included distillation heat and mass transfer characteristics, wine quality characteristics, mash fermentation characteristics, and energy consumption characteristics.

[0142] A distillation condition evaluation model based on a hybrid kernel support vector machine is constructed. Based on key features, the model can determine the distillation stability, the accuracy of the segmented distillation node, the risk of excessive impurities, and the risk of loss of fruit aroma in real time, and perform preliminary parameter pre-adjustment according to preset threshold rules.

[0143] Extended Kalman filter algorithm is used to perform spatiotemporal fusion of multi-dimensional working condition data. A bidirectional gated cyclic unit (Bi-GRU) is used to construct a full-element correlation network for raspberry distillation. The full-element correlation network quantitatively represents the coupling strength and causal relationship between distillation control parameters, liquor quality, mash characteristics and energy efficiency.

[0144] A multi-objective optimization model was established based on the comprehensive element correlation network to calculate the overall wine yield, raspberry ketone retention rate, methanol removal rate, fusel oil removal rate, and unit wine output energy consumption. The NSGA-III algorithm was used to solve for the optimal distillation control strategy solution set, and the optimal purification control scheme under the current operating conditions was determined by combining the entropy weight-fuzzy TOPSIS method.

[0145] The optimal purification control scheme is translated into execution instructions to achieve precise separation of the heads, hearts, and tails of the wine and maximize the preservation of raspberry characteristic aromas.

[0146] 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 part that contributes to the prior art, can be embodied in the form of a software product. 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.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these 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 the present invention.

Claims

1. A high-efficiency and energy-saving distillation and purification system for raspberry liquor, characterized in that, include: The data acquisition module is used to collect multi-dimensional operating data of the entire distillation process in real time when the raspberry fermentation mash is started for distillation and purification. The feature extraction module is used to extract key features from the multi-dimensional working condition data during the distillation heating process by using improved translation-invariant wavelet threshold noise reduction and combining it with kernel principal component analysis. The parameter pre-tuning module is used to construct a mixed kernel support vector machine condition evaluation model during the distillation steady-state control process, identify distillation stability, accuracy of the distillation node, risk of impurity exceeding the standard and risk of fruit aroma loss, and perform preliminary parameter pre-tuning according to preset threshold rules; The correlation quantification module is used to perform spatiotemporal fusion of the multi-dimensional operating condition data using extended Kalman filtering during coupled analysis of the distillation process, and to construct a full-element correlation network of the coupling relationship between quantification parameters, quality, mash and energy efficiency through Bi-GRU. The strategy solving module is used to establish a multi-objective optimization model based on the full-factor correlation network during the distillation strategy optimization stage. The model is then solved by NSGA-III and screened by entropy weight-fuzzy TOPSIS to obtain the best purification control scheme under the current working conditions. The control execution module is used to translate the optimal purification control scheme into equipment instructions during the segmented wine extraction process, so as to achieve segmented separation of wine and maximize the preservation of raspberry fruit aroma.

2. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 1, characterized in that, The multi-dimensional operating data includes distillation kettle operation data, fractional distillation data, aroma component data, fermentation mash physical property data, and energy efficiency loss data. The distillation kettle operation data includes distillation temperature, distillation pressure, heating power, liquid level in the kettle, and distillation time. The fractional distillation data includes distillation flow rate, alcohol content, distillation time, and the amount of heads, hearts, and tails. The aroma component data includes raspberry ketone content, ester content, alcohol content, and aldehyde content. The fermentation mash physical property data includes mash sugar content, mash acidity, mash density, and mash temperature. The energy efficiency loss data includes heating energy consumption, cooling water consumption, heat loss rate, and energy consumption per unit of distillation output.

3. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 1, characterized in that, The process of extracting key features by the feature extraction module includes: The multi-dimensional working condition data is constructed into an n×m dimensional data matrix, where n is the number of sampling times and m is the data dimension; Select a preset wavelet basis function and perform a preset number of translation-invariant wavelet decompositions on the data matrix to obtain approximation coefficients and detail coefficients at each scale; An adaptive soft thresholding function is constructed, and the threshold is dynamically adjusted according to the noise variance of the detail coefficients at each scale. The detail coefficients are then subjected to thresholding to suppress Gaussian noise and impulse noise. The processed detail coefficients and the original approximation coefficients are reconstructed by translation-invariant wavelet to obtain a denoised multi-dimensional working condition data matrix. Select a preset kernel function to map the denoised multi-dimensional working condition data matrix to a high-dimensional feature space and construct a kernel covariance matrix; The kernel covariance matrix is ​​subjected to eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors sorted by size; The feature vectors corresponding to the top k feature values ​​whose cumulative contribution rate reaches a preset threshold are selected as principal components. The principal component scores are calculated to obtain the key features, which include distillation heat and mass transfer features, alcohol quality features, mash fermentation features, and energy consumption features.

4. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 1, characterized in that, The process by which the parameter pre-tuning module constructs a distillation condition evaluation model based on a hybrid kernel support vector machine includes: Collect historical operating condition data of raspberry baijiu distillation over a preset time period, extract key historical features, and label the corresponding distillation stability label, segmented distillation node label, impurity exceedance risk label, and fruit aroma loss risk label. Construct a training set containing s samples, with each sample containing k of the aforementioned key features. A hybrid kernel function is constructed by merging the first kernel function and the second kernel function. The contribution ratio of the two types of kernel functions is adjusted by a preset kernel weight coefficient. By introducing slack variables and penalty parameters, a multi-class support vector machine optimization problem is constructed. The multi-class support vector machine optimization problem is transformed into a dual problem, and the dual problem is solved using the sequence minimum optimization algorithm to obtain the optimal Lagrange multiplier vector. Calculate the weight vector, select the support vector that meets the preset conditions to calculate the bias term, and obtain the distillation condition evaluation model.

5. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 2, characterized in that, The parameter pre-adjustment module performs the initial parameter pre-adjustment process according to preset threshold rules, including: Pre-set the normal threshold, warning threshold, and over-limit threshold for distillation temperature, alcohol content, raspberry ketone content, methanol content, and fusel oil content, and clarify the operating condition judgment criteria. The key features are input into the distillation condition evaluation model, which outputs the corresponding distillation stability status, segmented distillation node status, impurity exceedance risk level, and fruit aroma loss risk level. Based on the output of the distillation condition evaluation model, adjust the corresponding parameters according to preset rules.

6. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 1, characterized in that, The process of spatiotemporal fusion of multi-dimensional working condition data using the extended Kalman filter algorithm in the correlation quantization module includes: Core operating parameters are extracted from the multi-dimensional operating data and used as state variables to construct a state vector of preset dimensions. The core operating parameters include distillation temperature, distillation pressure, heating power, alcohol output flow rate, mash sugar content, and mash temperature. A nonlinear state transition equation for the distillation process is established to describe the relationship between the state vector and time. Based on the multi-dimensional working condition data, an observation vector is constructed, an observation equation is established, and the mapping relationship between the observation vector and the state vector is described. Initialize the state estimate and error covariance matrix, and calculate the current state prediction and prediction error covariance matrix according to the state transition equation; Calculate the Kalman gain, update the state estimate by combining it with the observation vector at the current time, update the error covariance matrix, and output the unified state estimate after spatiotemporal fusion.

7. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 6, characterized in that, The process by which the correlation quantification module constructs a correlation network of all elements of raspberry distillation using Bi-GRU includes: The unified state estimate is constructed as a time-series input sequence and input into the Bi-GRU network; The Bi-GRU network is configured to include a preset number of hidden layers, each containing a preset number of GRU units. The forward GRU layer captures the forward dependencies of the time-series input sequence, and the backward GRU layer captures the backward dependencies of the time-series input sequence. The outputs of the forward and reverse GRU layers are concatenated to obtain the fused hidden states at each time step. The Bi-GRU network is trained using a preset loss function and preset training parameters until the loss function converges. Using the fused hidden states corresponding to the state variables of each dimension as network nodes, and the connection weights between nodes as the association strength, the full-element association network is obtained.

8. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 1, characterized in that, The process of establishing a multi-objective optimization model by the strategy solving module includes: Distillation temperature, distillation pressure, reflux ratio, start time of core distillation, end time of core distillation, and heating power are selected as distillation control parameters to construct a decision vector; Sub-objective functions were constructed, including maximizing the wine yield, maximizing the raspberry ketone retention rate, maximizing the methanol removal rate, maximizing the fusel oil removal rate, and minimizing the energy consumption per unit of wine output. The maximum yield of the spirit is defined as the percentage of spirit mass produced per unit mass of fermentation mash. The maximum retention rate of raspberry ketone is measured by the ratio of the concentration of raspberry ketone in the wine core to the initial concentration of raspberry ketone in the mash. The methanol removal rate is maximized by using the difference between the initial methanol concentration in the mash and the methanol concentration in the center of the liquor as the indicator. The maximum removal rate of fusel oil is measured by the difference between the initial concentration of fusel oil in the mash and the concentration of fusel oil in the center of the liquor, and the ratio of the initial concentration of fusel oil in the mash to the concentration of fusel oil in the liquor. The minimum energy consumption per unit of distillation is indicated by the ratio of total energy consumption in the distillation process to the quality of the produced spirit. By combining the coupling constraints of the aforementioned all-element correlation network, equipment operation safety constraints, and liquor quality standard constraints, the constraints of the multi-objective optimization model are constructed.

9. The high-efficiency and energy-saving distillation and purification system for raspberry liquor according to claim 1, characterized in that, The process by which the strategy solving module determines the optimal purification control scheme under the current operating conditions includes: Individuals of a predetermined population size are randomly generated as the initial population for the NSGA-III algorithm, with each individual representing a set of distillation control parameters. Calculate the sub-objective function value for each individual, perform non-dominated sorting based on the reference point method, and calculate the crowding distance of the individuals; Excellent individuals are selected using a tournament selection method, and simulated binary crossover and polynomial mutation operations are performed to generate offspring populations. The parent population and the offspring population are merged, and non-dominated sorting and reference point association operations are performed again to select individuals of the preset population size to form the next generation population. The process iterates and evolves a preset number of times. Once the termination condition is met, the Pareto optimal solution set is output, which is the solution set of the optimal distillation control strategy. The objective weights of the sub-objective function are calculated using the entropy weight method to obtain the weight vector; Construct a fuzzy decision matrix to determine the positive ideal solution and the negative ideal solution. Calculate the distance from each solution in the optimal distillation control strategy solution set to the positive ideal solution and the negative ideal solution to obtain the relative proximity. The solution with the highest relative approximation is selected as the optimal purification control scheme under the current operating conditions.

10. A highly efficient and energy-saving distillation purification method for raspberry liquor, wherein the highly efficient and energy-saving distillation purification method implements the highly efficient and energy-saving distillation purification system for raspberry liquor as described in any one of claims 1 to 9, characterized in that the method... include: Real-time collection of multi-dimensional operating condition data throughout the entire process of raspberry liquor distillation and purification; An improved translation-invariant wavelet thresholding algorithm is used to denoise multi-dimensional operating condition data. Key features are extracted by combining kernel principal component analysis. These key features include distillation heat and mass transfer characteristics, alcohol quality characteristics, mash fermentation characteristics, and energy consumption characteristics. A distillation condition evaluation model based on a hybrid kernel support vector machine is constructed. Based on the key features, the model can determine the distillation stability, the accuracy of the segmented distillation node, the risk of impurity exceeding the standard, and the risk of fruit aroma loss in real time, and perform preliminary parameter pre-adjustment according to the preset threshold rules. An extended Kalman filter algorithm is used to perform spatiotemporal fusion of multi-dimensional operating condition data. A bidirectional gated cyclic unit (Bi-GRU) is used to construct a full-element correlation network for raspberry distillation. The full-element correlation network quantitatively represents the coupling strength and causal relationship between distillation control parameters, liquor quality, mash characteristics and energy efficiency. Based on the aforementioned full-element correlation network, a multi-objective optimization model is established to measure the overall wine yield, raspberry ketone retention rate, methanol removal rate, fusel oil removal rate, and unit wine output energy consumption. The NSGA-III algorithm is used to solve for the optimal distillation control strategy solution set, and the entropy weight-fuzzy TOPSIS method is combined to determine the optimal purification control scheme under the current working conditions. The optimal purification control scheme is converted into execution instructions to achieve precise separation of the heads, hearts, and tails of the wine and maximize the preservation of raspberry characteristic aromas.

Images